Spie Press: The Wonder of Nanotechnology: Quantum Optoelectronic Devices and Applications
Spie Press: The Wonder of Nanotechnology: Quantum Optoelectronic Devices and Applications
Spie Press: The Wonder of Nanotechnology: Quantum Optoelectronic Devices and Applications
Electrons, photons, and even thermal properties can all be engineered at the
nanolevel. Possibly the simplest aspect of nanotechnology, the 2D quantum well has
dramatically enhanced the efficiency and versatility of electronic and optoelectronic
devices. Nanotechnology has now progressed to 1D (quantum wire) and 0D
(quantum dot) systems that exhibit remarkable and sometimes unexpected
behaviors. With these components serving as the modern engineer's building
blocks, it is remarkable to consider the endless possibilities that nanotechnology
holds in store.
P.O. Box 10
Bellingham, WA 98227-0010
ISBN: 9780819495969
SPIE Vol. No.: PM238
Library of Congress Cataloging-in-Publication Data
Published by
SPIE
P.O. Box 10
Bellingham, Washington 98227-0010 USA
Phone: + 1 360.676.3290
Fax: + 1 360.647.1445
Email: [email protected]
Web: http://spie.org
The content of this book reflects the work and thought of the author(s). Every effort has
been made to publish reliable and accurate information herein, but the publisher is not
responsible for the validity of the information or for any outcomes resulting from
reliance thereon.
I Historic Overview 1
1 Role of Symmetry in Conductance, Capacitance, and Doping
of Quantum Dots 3
Raphael Tsu and Tim LaFave, Jr.
1.1 Introduction 3
1.2 Birth of the Superlattice 5
1.2.1 Response of a time-dependent electric field and
Bloch oscillation 7
1.3 Resonant Tunneling in Manmade Quantum Wells 11
1.3.1 Time-dependent resonant tunneling 15
1.3.2 Quantum cascade laser with superlattice components 17
1.3.3 Type-II superlattice 17
1.3.4 Terahertz sound in Stark ladder superlattices 18
1.3.5 Cold cathode 18
1.4 Size-Dependent Dielectric Constant ´(a) 19
1.5 Role of Symmetry in Capacitance of Few-Electron Quantum Dots 20
1.5.1 A classical correspondence between quantum dots
and atomic structure 23
1.5.2 Toward a general solution of the Thomson problem
and atomic structure 26
1.5.3 The dielectric function and atomic dimension 28
1.6 Symmetry: Key in Interaction with Nanotechnology 29
1.7 A Few Important Considerations 31
References 33
v
vi Contents
II Materials 39
2 Electrical, Optical, and Structural Studies of InAs/InGaSb VLWIR
Superlattices 41
Gail J. Brown, Said Elhamri, William C. Mitchel, Heather J. Haugan,
Krishnamurthy Mahalingam, Mu J. Kim, and Frank Szmulowicz
2.1 Introduction 41
2.2 Sample Fabrication/Design 42
2.3 Structural Characterization 43
2.4 Optical Characterization 46
2.5 Electronic Transport Measurements 48
2.6 Electronic Transport Modeling 51
2.7 Summary 52
References 53
3 InAs/InAs1 xSbx Superlattices on GaSb Substrates: A Promising
Material System for Mid- and Long-Wavelength Infrared Detectors 59
Elizabeth H. Steenbergen, Oray Orkun Cellek, Hua Li, Shi Liu,
Xiaomeng Shen, David J. Smith, and Yong-Hang Zhang
3.1 Introduction 60
3.2 Design 61
3.3 Growth and Structural Characteristics 68
3.4 Optical Characteristics 70
3.5 Infrared Detectors 77
3.6 Summary 77
References 78
4 Thermal Conductivity and Thermal Distribution in Superlattice
Structures 85
Chuanle Zhou and Matthew A. Grayson
4.1 Introduction 85
4.2 Thermal Conductivity Tensor 86
4.2.1 Cross-plane thermal conductivity 88
4.2.2 In-plane thermal conductivity 89
4.2.3 Error analysis 90
4.3 Thermal Conductivity of T2SL 94
4.4 Thermal Distribution 95
4.5 Conclusion 98
Acknowledgments 99
References 99
5 Superlinear Luminescence and Enhancement of Optical Power
in GaSb-based Heterostructures with High Conduction-Band
Offsets and Nanostructures with Deep Quantum Wells 105
Maya P. Mikhailova, Leonid V. Danilov, Karina V. Kalinina, Edward V. Ivanov,
Nikolay D. Stoyanov, Georgy G. Zegrya, Yury P. Yakovlev, Alice Hospodková,
Jirí Pangrác, Markéta Zíková, and Eduard Hulicius
Contents vii
V Applications 595
23 Current Status of Mid-infrared Semiconductor-Laser-based
Sensor Technologies for Trace-Gas Sensing Applications 597
Rafal Lewicki, Mohammad Jahjah, Yufei Ma, Przemysl aw Stefanski,
Jan Tarka, Manijeh Razeghi, and Frank K. Tittel
23.1 Introduction 598
23.2 Tunable Diode Laser Absorption Spectroscopy (TDLAS)
for Ethane Detection 602
23.2.1 Laser characterization 602
23.2.2 Optical sensor architecture 603
23.2.3 Experiments and results 604
23.3 Environmental Detection of Ammonia using an EC-QCL-based
C-PAS Sensor Platform 605
23.3.1 Sensor configuration and results 606
23.4 Quartz-Enhanced Photoacoustic Spectroscopy (QEPAS) 609
23.4.1 Methane and nitrous oxide detection 609
23.4.1.1 7.83-mm DFB-QCL for methane and nitrous oxide
detection 609
23.4.1.2 Experiments and results 609
23.4.2 Environmental detection of nitric oxide 611
23.4.3 QEPAS-based ppb-level detection of carbon monoxide
and nitrous oxide 614
23.4.3.1 CW DFB-QCL-based QEPAS sensor system
for CO and N2O 615
23.4.3.2 Experimental results and discussion 616
23.4.4 Sulfur dioxide experiments 620
23.4.4.1 SO2 QEPAS sensor architecture and performance 620
23.5 Conclusions 622
Acknowledgments 624
References 624
24 Application of Quantum Cascade Lasers for Safety and Security 633
Ulrike Willer, Mario Mordmüller, and Wolfgang Schade
24.1 Introduction 633
24.2 Pulsed Laser Fragmentation 635
24.3 Experimental Setup 635
24.4 Results 637
24.5 Discussion 639
24.6 Conclusions 641
References 641
xvi Contents
Index 879
Foreword
Twenty years ago, I met with Prof. Klaus von Klitzing and Prof. Manijeh
Razeghi and other top researchers from around the world for the inauguration
of a new Center for Quantum Devices at Northwestern University. A full two
decades of research later, we have chosen the occasion of the International
Conference on Infrared Optoelectronics (MIOMD-XI) to join together again
at Northwestern to celebrate all of the accomplishments of the intervening
years. This conference not only marks the latest progress in new materials and
devices that followed from my own work in this field, it also highlighted the
richest accomplishments of a full spectrum of prominent world-class scientists.
With the success of this conference, it was decided that a more permanent
volume should commemorate the achievements presented there, so as cochair
of MIOMD-XI, I am happy to announce the occasion in this foreword. This
book collects the best and highest-impact talks from that conference, develops
them into chapters, and collects them into a single condensed volume
representing the current state-of-the-art in infrared materials and devices.
The chapters in this book bear witness to how far we have come since the
invention of manmade semiconductor superlattices in 1969. What started with
the new physics of the Esaki tunnel diode has matured into nanoscale
engineering of semiconductor superlattices to create whole synthetic band
structures. After years of considerable effort to bring this technology to
maturity, we now see the results of this formidable new science in almost every
electronic and photonic device that we encounter. We see it in the electronics
that flood the consumer market, the communication infrastructure that is
rapidly shrinking our world, and in the specialized components such as
quantum cascade lasers or type-II superlattice cameras used for defense and
security—this is truly the age of nanotechnology. I look back with wonder at
all of the exciting developments of the last 44 years and can only imagine
where the future will take this technology and what exciting discoveries await.
Leo Esaki
University of Tokyo
Komaba, Meguro, Japan
xxi
xxii Foreword
xxv
xxvi Introduction
culminates in an infrared camera that sees the heat signal given off by the
same hands that crafted the device from the atom up.
In a broader scope, this volume collects the latest world-class research
breakthroughs that have brought quantum engineering to an unprecedented
level, creating light detectors and emitters over an extremely wide spectral
range from 0.2 to 300 mm. Devices include light-emitting diodes in the deep-
ultraviolet to visible wavelengths. In the infrared, compounds can be
nanoengineered to create quantum cascade lasers and focal plane arrays
based on quantum dots or repeated layers of one material inside another.
These are fast becoming the choice of technology in crucial applications such
as environmental monitoring and space exploration. Last but not least, on the
far-infrared end of the electromagnetic spectrum, also known as the terahertz
region, new nanotechnology allows emission of terahertz waves in a compact
device at room temperature. Continued effort is being devoted to all of the
abovementioned areas, with the intention to develop smart technologies that
meet the current challenges in environment, health, security, and energy. This
volume documents the latest contributions to the world of semiconductor
nanoscale optoelectronics.
The research efforts represented here share a common genesis in the
MIOMD-XI conference at Northwestern University, hosted by the Center for
Quantum Devices in September 2012. The novelty and quality of the work
presented at that conference inspired their collection into this special volume,
representing both the state-of-the-art and the future trends of nanotechnology.
It is a privilege to be able to introduce these works here for posterity so
that they might mark our remarkable progress in the past decades and usher
in the wonders of what nanotechnology holds in store for our future.
Manijeh Razeghi
Center for Quantum Devices
Electrical Engineering & Computer Science Department
Evanston, Illinois, USA
xxix
xxx An Imaging Perspective from the Nanometer Scale
The optics is designed as multiple lenses made from bulk materials and aligned
in a barrel. However, traditional approaches are unlikely to yield large
improvements in infrared camera development. Specific limitations are large
format, multiple colors, and wide-band detector design with high resolution,
which require incompatible materials for different colors, scaling of pixel size,
and wide-band optics, to list a few. As a consequence, infrared cameras are
large and expensive, and generally limited to military applications. A
paradigm shift in the way components of cameras and other optoelectronic
devices are made is needed to fulfill the future requirements. This shift in
approach will make smaller and lower-cost infrared cameras, lasers, and many
other optoelectronic products available for both civilian and military markets.
Nanotechnology is paving the way for a new dimension involving more
versatile material designs that enable large format, multicolor, and wide-band
infrared focal plane arrays. One example is the type-II superlattice approach
that uses a set of different compound semiconductor materials to design
multiple band detectors on a single substrate. The type-II superlattice
technique takes advantage of nanometer-scale stacking of different exotic
materials to tailor the bandgap. The nanometer-scale manipulation of
different exotic materials, therefore, allows for a new material design whose
optical properties can be modified from the individual bulk material. Thus, an
artificially created new “lattice structure” can be formed in mixed
semiconductor crystals, allowing for bandgap engineering. Another example
is the nanometer-scale structuring of a thin compound semiconductor material
to fabricate a photonic crystal. Subwavelength-sized semiconductor pillar
arrays within a single detector can be designed and structured as an ensemble
of photon trapping units to significantly increase absorption and quantum
efficiency for a wide band of wavelengths. Each sub-element in each pixel can
be a 3D photonic structure fabricated using either a top-down or bottom-up
process. The sub-element architecture can be of different shapes such as
pyramidal, sinusoidal, or rectangular. Additionally, the sub-elements
themselves can have p–n junctions. The motivation for this design is to
significantly increase photon trapping of a wide range of wavelengths, and
their subsequent absorption and generation of electron–hole pairs in the
absorber material. Such a design also leads to a reduction in the material
volume and, thus, a decrease in the dark current. The subwavelength photonic
trap allows for high absorption and increases the signal-to-noise ratio.
Metamaterials to manipulate light is yet another technique leveraged by
nanotechnology and can be used to develop monolithic filters directly on
wide-band detectors. Such an arrangement offers a real shift in the way
infrared focal plane arrays are designed. Nanometer-scale structuring also has
merit in solar cells, lasers, and light-emitting diodes. Bandgap engineering and
nanometer-scale structuring both modify the fundamental building block of
the materials.
An Imaging Perspective from the Nanometer Scale xxxi
architectures in focal plane array designs for wide-band and multi-color, pixel
sizes at subwavelengths, wafer scale optics, wafer scale IR cameras, novel
system architectures for gigapixel-class cameras, and bio-inspired nanometer-
scale sensor technologies. These efforts have culminated into a new set of
infrared camera technologies and tools that are revolutionizing the way focal
plane arrays, optics, and cameras are produced. Dr. Dhar has authored
numerous papers and chapters on infrared technology, served as chairperson
on numerous conferences and committees, and served as coeditor of several
conference proceedings. He mentored and served on eight doctoral thesis
advisory committees on various subjects. He is also Fellow of SPIE.
List of Contributors
1.1 Introduction
1.2 Birth of the Superlattice
1.2.1 Response of a time dependent electric field and Bloch oscillation
1.3 Resonant Tunneling in Manmade Quantum Wells
1.3.1 Time dependent resonant tunneling
1.3.2 Quantum cascade laser with superlattice components
1.3.3 Type II superlattice
1.3.4 Terahertz sound in Stark ladder superlattices
1.3.5 Cold cathode
1.4 Size Dependent Dielectric Constant «(a)
1.5 Role of Symmetry in Capacitance of Few Electron Quantum Dots
1.5.1 A classical correspondence between quantum dots and atomic structure
1.5.2 Toward a general solution of the Thomson problem and atomic structure
1.5.3 The dielectric function and atomic dimension
1.6 Symmetry: Key in Interaction with Nanotechnology
1.7 A Few Important Considerations
References
1.1 Introduction
A significant difference between classical and quantum systems involving
interactions is generally accepted. Classical systems are characterized by
potentials for interactions with physical parameters like mass and charge
3
4 Chapter 1
close encounters. Asked about his legacy of the famous Wigner–Seitz cell,
Frederick Seitz explained to Tsu, “The coordinates I introduced, ,100>,
,111>, and so on will outlast the validity of the Wigner–Seitz cell!” In other
words, determination of boundary conditions in quantum mechanics is a
fundamental issue. In our view, the validity of boundary conditions depends
on determining the degree of accuracy required of a given interaction. The
boundary condition for a large interfacial plane separating two regions is very
different from that of a curved common interface of nanometer dimensions.
The issue is whether averaging is to be applied to the boundary conditions
determining the wavefunctions or to the wavefunctions to be used at the common
boundary. In essence, we need to revisit the issue and applicability of RPA—
the difference between averaging probability densities versus wavefunctions as
commonly used in quantum mechanics. Learning to appreciate a new concept
needs a vehicle. This is why some of the background on the development of
manmade materials must be revisited—particularly those concerning the role
of coherence.
Starting with the distribution function for constant relaxation time t given by
Eq. (1.1) and with
1 @E
vx ¼ ,
ħ @kx
then
Z Z Zt0
1 1 1 @Eðt0 Þ @EðtÞ dt
hvx i ¼ 3 f vx dk ¼ 3 f0 dk exp½ðt t0 Þ=t :
4p 4p ħ @kx @kx t
k 1
ð1:2Þ
8 Chapter 1
where Jn are Bessel functions. Then, the expectation value of velocity becomes
X
1
vB1 vB1 sinðm nÞv1 t þ ðvB þ nv1 Þt cosðm nÞv1 t
hvx i ¼ v0 H Jm Jn ,
m, n¼ 1
v1 v1 ðvB þ nv1 Þ2 t2 þ 1
ð1:7Þ
with vB eFd/ħ, and vB1 eF1d/ħ. For H ¼ 1, hvx i in Eq. (1.7) was first
obtained by Tsu in 1990.10
For vB1=v1 ,, 1 (small F1),
hvx i ¼ hvx i0 þ Rehvx i1 cos v1 t þ Imhvx i1 sin v1 t, ð1:8Þ
where
vB t
hvx i ¼ v0 H , ð1:9Þ
ðvB tÞ2 þ 1
which is identical, as it should be, to the previous results. For the case of
H ¼ 1, then v0H ¼ E1d/ħ, and the maximum extent hxi m ¼ hvx i mt ¼ E1=2eF0.
Since length is measured by nd, with n as the number of periods, the maximum
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 9
Figure 1.1 The in-phase Rehvi1 and out-of-phase Imhvi1 components of the linear
response function of a SL with an applied electric field F ¼ F0 þ 2F1 cos vt, vB eF0 d=ħ,
and vB1 eF1 d=ħ (reprinted from Ref. 10).
10 Chapter 1
maximum gain are never at the Bloch frequency. Only as vBt ! 1 does the
maximum gain coincide with the Bloch frequency. For both vBt >> 1 and vt >> 1,
it is seen that Rehvi3 can have a substantial negative region, indicating that in the
region of nonlinear optics, an intense optical field is needed for gain.
Experimentally, one needs to arrange the polarization with a component of the
electric field in the SL direction. For t ¼ 0.5 ps and vBt ¼ 3, eF0d 4 meV,
corresponding to F0 4 103 V cm 1 for d ¼ 10 nm. Therefore, the condition
for self-oscillation at v ¼ 2vB/3, or v 4 1012 Hz, should be quite accessible.
In fact, not only can a Bloch electron go into oscillation, it can also serve as an
amplifier. Similar to the traveling wave amplifier, Bloch oscillators have larger
bandwidth and higher power-handling capacity.
Also, higher-energy photons cause transitions between minibands, providing
additional nonlinear response. This is because k is conserved to within multiples
of the reciprocal lattice vector, as in umklapprozesse. In the usual solids, optical
nonlinearity arises from small nonparabolicity of the E–k relation as treated by
Jha and Bloembergen,11 as well as from optical phonons in a multilayer
dielectric medium treated by Bloembergen and Sievers.12 However, in
manmade SLs, nonparabolicity is enormous, leading to substantial second
and third harmonics, shown in Fig. 1.2, from Tsu and Esaki,13 where the
nonlinear effect is 20 greater than that calculated by Wolf and Pearson.14
Figure 1.2 Calculated |Jn| / J0 versus v for two values of t with J ¼ en0 hvx i, where n0 is the
usual sum over the density of states (DOS) (reprinted from Ref. 13).
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 11
Note that there is a term ðk0‘ =k‘ Þ in Eq. (1.12) that is left out in the
original Tsu–Esaki. 15 With separation of variables due to the
planar geometry, the transverse momentum k t is conserved so that T *T
depends only on the longitudinal direction. After integration over the
transverse d k t,
Z1
em kT 1 þ exp½ðEf E‘ Þ=kb T
J¼ dE‘ T T ln : ð1:13Þ
2p2 ħ3 1 þ exp½ðEf E‘ eV Þ=kb T
0
For kB T ! 0,
ZEf
em
J¼ ðEf E‘ ÞT TdE‘ , when eV Ef : ð1:14Þ
2p2 ħ3
0
2 3
EZf eV ZEf
em 6 7
J¼ 4 ðeV ÞT TdE‘ þ ðEf E‘ ÞT TdE‘ 5, when eV
Ef :
2p2 ħ3
0 Ef eV
ð1:15Þ
Let us discuss an important issue. Although T T reaches unity at V ¼ 0,
*
T*T may be very much reduced at V 6¼ 0, due to the loss of symmetry, a point
not highly appreciated at first. To achieve large NDC, one needs to design the
structure such that it is nearly symmetrical at the operating point V. We shall
go into detail on this point as well as some other important points later. But
first we offer a physical picture of why NDC appears. Figure 1.3(a) shows an
electron at energy E, incident from the left and transmitted to the right for: (a)
12 Chapter 1
Figure 1.3 (a) Electron transmission at an energy E incident from the left and transmitted
to the right through a single barrier. (b) The same barrier with a ‘cut,’ representing a QW
between two barriers, or double-barrier resonant tunneling (DBRT). (c) A resonant
transmission peak at energy E1 appears and is superposed onto the original barrier without
the ‘cut.’
through a single barrier of height Vb, and (b) a quantum well (QW) with a
state denoted E1, an “eigenstate” (resonant state to be precise) created by the
QW as in typical elementary quantum mechanics, and (c) transmission |T |2.
For a symmetrical structure, constructive interference gives rise to resonance
with T*T ¼ 1, centered about E1, similar to the optical Fabry–Pérot
interferometer.
Consider the structure in Fig. 1.3 but with an applied voltage between
the left and right contacts. A current peak appears at V 2E1. The
approximation sign is due to a loss of symmetry with an applied voltage. If
symmetry can be preserved, the resonant tunneling peak appears exactly at
V ¼ 2E1. The width of the current peak is about the width of the Fermi level EF.
As the voltage is such that the state E1 moves into the forbidden gap, the
supply of electrons from the left contact vanishes, resulting in a rapid decrease
in the transmitted current. Thus, even without the use of a formula for the
resonant tunneling current, the mechanism for NDC is easily understood.
0
Before we deal with the ðk‘ =k‘ Þ term left out from Eq. (1.12) in the
original Tsu–Esaki formula, we shall deal with some fairly elementary, but
important, principles. For example, most quantum mechanics texts do
not emphasize that 1 |R|2 6¼ |T|2. Let an electron incident on a barrier with
a potential V ¼ Vu(x), with the Heaviside function u(x), so that the wavefunction
for x , 0 is
c L ¼ eik x þ R e ik x
, ð1:16Þ
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 13
Z1 Z1
em 2 «‘ þ eV 1=2
jRL ¼ 2 3 jTLR j d«‘ d«t ½ f ð« þ eV Þ f ð«Þf ð« þ eV Þ,
2p ħ «‘
0 0
ð1:20Þ
and the net current from the left contact to the right contact becomes
Z1 1=2 Z1
em 2 «‘ þ eV
j ¼ jLR jRL ¼ jTLR j d«‘ d«t ½ f ð«Þ f ð« þ eV Þ:
2p2 ħ3 «‘
0 0
ð1:21Þ
The original Tsu–Esaki expression13 left out the (k 0‘ =k‘ ) ¼ ½ð«‘ þ eV Þ=«‘ 1=2
term, referred to as the ‘kinematic term’ by Coon and Liu.16 Theories dealing
with resonant tunneling appeared before and even after Coon and Liu
used the Tsu–Esaki expression instead of having the kinematic factor.
p
The magnitude of this factor is 3 because near resonance, eV 2E‘ .
Although this term generally does not affect the overall physical picture, let us
look at what happens to the transmitted electron at an energy eV above the
Fermi energy at the right contact. An electron jk 0‘ > resonantly appearing as a
hot electron at the right contact at an energy «‘ þ eV, readily scatters to the
equilibrium state jk‘ > at energy «‘ inside the right contact within perhaps a
14 Chapter 1
few angstroms. Early on, Tsu struggled with this difficulty and tried to create
a transitional layer inside the contact without much positive result. The
visualization was to create a transitional region in the contact somewhat
similar to what has been discussed by Datta.17 But any attempt to create two
planes, 1 and 2, inside the contacts presents a formidable task with dubious
consequences. From the intuitive pictures, we know that a large current in
solids almost certainly originates from a large carrier density such as in
metals, or via multiplication mechanisms.
Thus, we may view the process in the following sequence of events: (1)
Electrons gain energy under an applied voltage and exit into a contact defined
by the definition of a contact being an equal potential. (2) The extra energy is
readily lost via scattering toward equilibrium with a final product: creation of
heat. (3) Such a process may be described using Fermi’s golden rule in a rate
equation rather than a transport equation, allowing the establishment of
equilibrium. Sometimes it is necessary to define a hot electron temperature if
elastic scattering dominates inelastic events. Within the crude assumption that
the Fermi levels between the two contacts are similar, electrons enter
the collector at E þ eV, and after relaxation, return to the same Fermi
distribution of the contact. Inclusion of a self-consistent potential such as in
methods by Cahay et al.18 and Bandara and Coon,19 is in the right direction
but insufficient without taking scattering inside the contact into account.
Nevertheless, the net result may very well be the creation of a term that
cancels the kinematic term.
There is another nontechnical issue. The main reason for going to the
DBRT was to model NDC without the domain oscillations pointed out by
Gunn. Note that the very title of the paper, “Tunneling in a finite
superlattice,”15 indicated the frame of mind: producing the basis of extending
the tunneling results as a model for transport in a SL showing NDC, where
clearly equilibrium distribution should dominate every few tunneling events,
or at least within a length limited by the mean free path. At first there was the
book by C. B. Duke,20 which treated everything but the needed feature,
allowing resonance to occur as in a Fabry–Pérot optical filter. The double-
barrier structure is simply an electron filter that allows an electron
into the state approximated by k‘ ¼ np/w. Without an elaborate scattering
formulation, it is clear that transmitted electrons simply lose the extra energy
while cascading down. Note that by taking the velocity term as ħk‘ =m instead
of ħk‘0 =m accomplishes just that. By doing so, a complicated scattering
formulation inside the right contact is circumvented. Whatever one does to
take scattering in the contact into account, the end result will involve giving
back the extra energy. From this perspective, the Tsu–Esaki formulism
represents a crude approximation. However, the events showed that the
original work of Tsu–Esaki has led to an enormous amount of improvement,
including many-body and hot electron effects. We need to recognize that the
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 15
Figure 1.4 Transmitted current J versus t (fs) for the case of GaAs well width w ¼ 6 nm,
and GaAlAs barrier b ¼ 2 nm. The energy of resonance is E4 ¼ 0.0682 eV, E3 ¼ 0.079 eV,
E2 ¼ 0.05 eV, and E1 ¼ 0.03 eV [reprinted from Ref. 26, taken from Sen’s unpublished thesis
(Ref. 27)].
Figure 1.5 Delay time t ¼ df=dk, for the first two resonant peaks. Solid line: t0 ¼ dm*/ħk.
Dashed line and circle: t0Q (reprinted from Ref. 30).
Figure 1.6 Nanostructured layer of the cold cathode (reprinted from Ref. 39).
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 19
Figure 1.7 Solid state field-emission cathodes with TiO2 on Pt with a field of 50–140 V/mm
for increased tunneling (reprinted Ref. 40).
Figure 1.8 Size-dependent static dielectric constant «(a) for Si. Solid line: modified Penn
model with «B ¼ 12; dash–dot line: modified Penn model with «B ¼ 11.3; line with crosses:
converted from Walter–Cohen42 with q ¼ p/a; long-dash line: from Lannoo et al.43; short-
dash line: from Wang and Zunger44 (adapted from Ref. 41).
Using a modified Penn model, Fig. 1.8 shows the reduction of the size-
dependent dielectric constant of a silicon nanosphere. Basically, the reduction
of dielectric constant in the nanoscale regime is caused by quantum
confinement restricting the movement of electrons. However, strictly, the
dielectric function cannot be rigorously defined in quantum dots.
In essence, there is no basis set to define «(a), except in cases in which the
medium is either uniform, or at least periodic, unless some sort of average
may be defined. The best we can do is to either apply the RPA or some sort of
averaging procedure. In fact, this problem arises in all constitutive equations,
particularly so for physics in the nanoregime, which shall be treated in a
special way at the end of this chapter.
discrete charge dielectric (DCD) model, was developed.45,46 The DCD model
is fundamentally rooted in a Green’s function approach to solutions of the
classical Poisson equation. Electrostatic distributions of N electrons are
obtained by global minimization of the total Coulomb repulsion and
polarization interaction energy. The total potential energy stored within a
given N-electron quantum dot may be expressed in terms of the summation of
three unique types of electrostatic interactions:
" #
N X N X
1X N X N
1X N
UðNÞ ¼ UC ðqi , qj Þ þ Udir ðqi , si Þ þ Uind ðqi , qj Þ : ð1:25Þ
2 i 1 j6 i i 1
2 i 1 j6 i
The first term in Eq. (1.25) is a summation over all Coulomb repulsion
interaction energies UC among all N electrons. The bracketed terms are
summations over two types of polarization interaction energies: direct
polarization energies Udir, involving each electron qi and its associated
polarized surface charge element si, and indirect polarization energies Uind,
involving interactions between electrons qj and surface charge elements
formed by other electrons. Of interest here is the treatment of free electrons in
spherical quantum dots. The fundamental physics involves changes of spatial
charge symmetry resulting from the introduction of individual electrons. For
example, the global minimum energy of a spherical quantum dot containing a
single electron is obtained when the electron resides at the origin of the sphere
(Fig. 1.9). As the second electron is introduced, the first electron is displaced,
and each electron takes up location on either side of the origin to minimize the
global electrostatic energy [Eq. (1.25)]. When the dielectric constant inside is
greater than outside, the polarized dielectric consists of a net-polarized surface
charge that repels electrons to the interior—the fundamental cause of
dielectric confinement. As N increases, the distance of each electron from the
origin increases due to Coulomb repulsion, as shown schematically in
Fig. 1.9. Indeed, to a good approximation, a global minimum energy is
obtained for each N-electron system when all electrons are at the same
distance b from the origin. As the symmetric phase of the system changes with
the addition of each new electron, parameters may be characterized as
interphasic. Parameters characteristic of a particular N-electron phase may
then be characterized as monophasic. Figure 1.9 shows monophasic solutions
for up to N ¼ 12 electrons in a spherical quantum dot of dielectric constant
« ¼ 20«0.
From within the framework of the DCD model, the monophasic
capacitance45 48 emerged as a rigorously derived general expression of classical
electrostatic capacitance. The expression C(N) ¼ (Ne)2/U(N) is dependent on
the average electrostatic potential experienced by each electron in the quantum
dot in order to circumvent a thorough evaluation of complex 3D equipotential
surfaces from which capacitance must be defined due to the unique spatial
22 Chapter 1
(xz) and (yz),56 while the solution for N ¼ 12, sandwiched between them, is
one of only five highly symmetric geometries of the Platonic solids. If there is a
classical electrostatic correspondence with the atomic structure, it would be
due to abrupt differences in the spatial symmetry of neighboring solutions in
the Thomson problem. Energy solutions of the Thomson problem are well fit
by the smooth function57,58
N2
UðNÞ ¼ aN 3=2 : ð1:26Þ
2
The solutions tend to vary slightly from this function. Indeed, there are no
noticeable variations in the vicinity of 11
N
13 to reflect the disparate
changes in spatial symmetry noted by Edmundson. The problem is that the
energies U(N) depend on changes in both N and symmetry. In February 2006,
LaFave devised a procedure to evaluate the energy associated strictly with
changes in charge symmetry, that is, interphasic energy DUþ(N), by moving a
single electron from any given N-electron configuration [Fig. 1.10(a)] to the
origin. The remaining (N 1) electrons relax to form a corresponding N 1
configuration about q0 [Fig. 1.10(b)], a transformation that leaves the net charge
Q ¼ Ne unchanged, while changing only the spatial symmetry of the N electron.
Moreover, it is noted that as q0 moves to the origin of a dielectric sphere,
its mathematically equivalent image charge q00 moves away from the sphere—a
transformation similar to the ionization of a single electron from an atom.
Figure 1.10 Discrete symmetry changes between neighboring (a) N-electron systems
(N ¼ 5 shown as example) are obtained by displacing (b) one electron q0 to the origin,
resulting in an (N 1)-electron solution of the Thomson problem. The potential energy
distribution associated with this transition exhibits (c) disparities that correspond remarkably
well58 with shell-filling features in natural atomic structure. (See color plate section.)
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 25
83-charge system, nor an 85th charge that may appear within a pair of
protons, for instance, nor an 86th or 87th charge, for that matter.
Interestingly, Glasser and Avery58 chose not to report any error
evaluation with respect to the fit function, Eq. (1.26), noting that solutions
of the Thomson problem are not random but are in fact “systematic.” As
demonstrated in Fig. 1.10(c), LaFave shows that interphasic symmetry-
dependent energy corresponding to the periodic table of elements is the
underlying physical nature of this “systematic” behavior. Moreover, noting a
regular, periodic occurrence of the largest disparities in Fig. 1.10(c) at N ¼
(11, 12), (31, 32), (47, 48), and (71, 72), one may conjecture that spatial
limitations imposed on discrete point charges constrained to a spherical region
of space (e.g., dielectric sphere or central field potential) is the underlying
physical mechanism responsible for the Periodic law—the very foundation of
the entire field of chemistry. Indeed, regular occurrence of these disparities
vanishes for N > 100! This may be verified by the interested reader using data
found in an interactive database hosted by Syracuse University.60
These results should not be entirely surprising since the quantum
mechanical Schrödinger equation treats the kinetic energy of particles quite
well, whereas the classical Poisson equation usefully treats the potential
energy of charged particles. Moreover, as noted earlier, energy associated with
a discrete symmetry change due to the displacement of q0 to the origin is
related to the ionization of a neutral atom.
Figure 1.11 Open circles denote numerical N-electron solutions of the Thomson problem.
General solutions require local minima at integer values of N. A few possibilities are
shown.61
28 Chapter 1
Figure 1.12 (a) Interphasic energy versus electron number N using dielectric constant « ¼
20«0 adjusted with empirical atomic radii remp61 and (b) ionization energy of neutral atoms.62
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 29
Figure 1.13 Conductance G (ms) versus V in volts showing that the linewidth is kBT at RT
(reprinted from Ref. 65).
Role of Symmetry in Conductance, Capacitance, and Doping of Quantum Dots 31
Figure 1.14 Conductance oscillation measured at 1 MHz showing slowing down near the
end of a 900-s trace, with switching speed changing from 2 s to more than 10 s, ending in
the high conducting state (from one of many unpublished traces taken in 1993).
may be caused by interactions leading to 1/f noise, long range, and having the
most mysterious part being nearly all different from hundreds of traces! For
these reasons, Tsu was led to think in terms of wave impedance of electrons
(Fig. 1.13) having integer nG0 in empty space but integers as well as fractions
in solids with periodicity. See Chapter 11 of Ref. 29.
Tsu spent a large portion of his life looking into superlattices and resonant
tunneling with well-developed theories and experiments until running into this
kind of phenomena, which prevented him from exploring the subject in full,
partially due to running into problems in publications too often to be seriously
accepted by him as well as others! He now thinks that the phenomena may be
related to polar catastrophe69 taking place between nanoscaled interfaces or
that it may be involved with some undiscovered mechanism of resistive
switching. These oscillations may have periods anywhere in the range from
hertz to megahertz! In fact, we have also observed light emission in these
current oscillations. As mysterious as it may seem, yet something of which we
are sure, such data did not appear in clusters much larger than nanometers in
size! An educated guess involves a nanometer-sized system, a quantum dot or
coupled quantum dots, having a complicated interface other than planar that
involves collective excitations and de-excitations with the most obvious
appearance of oscillations and hysteresis!
References
1. D. Pines and D. Bohm, “A collective description of electron interactions
II: Collective vs individual particle aspects of the interactions,” Phys. Rev.
85(2), 338 (1952).
2. D. Bohm and D. Pines, “A Collective description of electron interactions
III: Coulomb interactions in a degenerate electron gas,” Phys. Rev. 92(3),
609 (1953).
3. P. Nozières and D. Pines, “Correlation energy of a free-electron gas,”
Phys. Rev. 111(2), 442–454 (1958).
4. A. B. Pippard, Dynamics of Conduction Electrons, Gordon & Breach, New
York (1965).
5. P. A. Lebwohl and R. Tsu, “Electrical properties in a superlattice,”
J. Appl. Phys. 41(6), 2664 (1970).
6. The interested reader is directed to online lecture notes for the Rensselaer
Polytechnic Institute course on Semiconductor Devices and Models: E. F.
Schubert, “Band Diagrams of Heterostructures,” Rensselaer Polytechnic
Institute (2003).
7. L. L. Chang, L. Esaki, and R. Tsu, “Resonant tunneling in semiconductor
double barriers,” Appl. Phys. Lett. 24(12), 593 (1974).
8. R. Tsu, “Superlattices: Problems and new opportunities, nanosolids,”
Nanoscale Res. Lett. 6, 127–136 (2011).
9. J. B. Krieger and G. J. Iafrate, “Time evolution of Bloch electrons in a
homogeneous electric field,” Phys. Rev. B 33(8), 5494 (1986).
10. R. Tsu, Q.-Y. Ye, and E. H. Nicollian, “Resonant tunneling in
microcrystalline silicon quantum box diode,” Proc. SPIE 1361, 232–237
(1990) [doi: 10.1117/12.24358].
11. S. S. Jha and N. Bloembergen, “Nonlinear optical susceptibilities in
group-IV and III-V semiconductors,” Phys. Rev. 171(3), 891–898 (1968).
12. N. Bloembergen and A. J. Sievers, “Nonlinear optical properties of
periodic laminar structures,” Appl. Phys. Lett. 17(11), 483–486 (1970).
13. R. Tsu and L. Esaki, “Nonlinear optical response of conduction electrons
in a superlattice,” Appl. Phys. Lett. 19(7), 246–248 (1971).
14. P. A. Wolf and G. A. Pearson, “Theory of optical mixing by mobile
carriers in semiconductors,” Phys. Rev. Lett. 17(19), 1015–1017 (1966).
15. R. Tsu and L. Esaki, “Tunneling in a finite superlattice,” Appl. Phys. Lett.
22(11), 562–565 (1973).
16. D. D. Coon and H. C. Liu, “Tunneling currents and two-body effects in
quantum well and superlattice structures,” Appl. Phys. Lett. 47(2), 172–
174 (1985).
34 Chapter 1
Said Elhamri
Department of Physics, University of Dayton, Dayton, Ohio, USA
2.1 Introduction
2.2 Sample Fabrication/Design
2.3 Structural Characterization
2.4 Optical Characterization
2.5 Electronic Transport Measurements
2.6 Electronic Transport Modeling
2.7 Summary
References
2.1 Introduction
InAs/InGaSb superlattice (SL) materials are an excellent candidate for
infrared photodiodes with cutoff wavelengths beyond 15 mm, i.e., in the very
long-wavelength infrared (VLWIR) range. There are relatively few options
for high-performance infrared detectors to cover wavelengths longer than
15 mm, especially for operating temperatures above 15 K. A variety of
41
42 Chapter 2
possible SL designs cover the VLWIR range, including designs with and
without indium alloying of the GaSb layers.1 For homogeneous InGaSb
alloys, transport modeling found that alloy scattering should be negligible
for electrons.2 In addition, there can be benefits for incorporating InGaSb
into the VLWIR SL design; these benefits include a higher molecular beam
epitaxy (MBE) growth temperature, which should reduce point defects in the
InGaSb, simpler interfaces with the continuous indium flux, suppressed
Auger recombination rates, and larger absorption coefficients due to thinner
periods in these designs.
Our focus is on designs with 25% indium in the gallium antimonide to
achieve energy bandgaps less than 50 meV with a SL period on the order of 68
Å. Similar to the work reported on InAs/GaSb LWIR and VLWIR SLs,3 5
our designs employ InGaSb layers less than seven monolayers in width. While
the SL designs are strain balanced to the GaSb substrate, care was also taken
to minimize strain spikes in the interfacial regions. High-resolution transmis-
sion electron microscope images were analyzed to create strain mapping
profiles of the SL layers and interfaces. By focusing on a narrow set of
VLWIR SL designs, the deposition parameters for the MBE SL growth could
be carefully optimized.
The electrical and optical properties of the VLWIR SLs were
characterized by variable-temperature Hall effect measurements and by
infrared photoresponse spectra. The photoresponse spectra consistently
showed a bandgap energy of 47 3 meV for the samples studied and a 50%
cutoff wavelength at 19 mm. The repeatability of these very narrow-
bandgap SLs over multiple sample depositions even while some growth
parameters were being adjusted shows the tight control obtainable with
MBE. Narrow-bandgap designs are very sensitive to small changes in layer
widths.6 Variable-temperature Hall effect measurements found that the
mobility of electrons in the SL was 10,000 cm2/Vs below 80 K and was
relatively constant in magnitude.
Table 2.1 Sample designs, SL period measured by x-ray diffraction, and the optical
bandgap energy determined by the photoresponse spectra.
Sample InAs (Å) GaSb (Å) In (%) IF (Å) Period (Å) Eg (meV)
A 47 21.5 25 0 68 46
B 47 21.5 25 0 68 44
C 46.5 18 0 1.5/2.5 69 63
D 47 21.5 25 0 68.5 45
the InAs and InGaSb layers was not intentionally controlled by shutter
sequence to be either InSb-like or GaAs-like. The residual strain of the two
InGaSb-based SLs remained small and slightly compressive at þ0.2%
(samples A and B). For the GaSb-based VLWIR SL, we used a design
based on our previous studies,8 of 48 Å InAs / 20.5 Å GaSb, including the
controlled InSb-like interfaces. The controlled interfaces are necessary to
strain balance the lattice-mismatched InAs. With the controlled InSb-like
interfaces, the net strain of sample C was close to zero. The structural
parameters of the samples, such as SL period, residual strain, and individual
layer thickness, were confirmed by high-resolution transmission electron
microscopy (HRTEM) and high-resolution x-ray diffraction (HRXRD)
measurements. A summary of the intended SL designs, measured HRXRD
SL period, and measured bandgap energy determined from the onset of the
photoresponse spectra is given in Table 2.1.
Figure 2.1 A HRTEM image of sample A showing the first few layers in the InGaSb/InAs
SL near the GaSb substrate.
appear dark and bright, respectively. From these images the average SL
period was determined to be 67.6 0.3 Å, with an average layer thickness of
24.1 1.6 Å for InAs and 44.1 1.2 Å for In0.25Ga0.75Sb. The results from
strain analysis of the HRTEM images are shown in Fig. 2.2, where Fig. 2.2(a)
is a map of the strain tensor «yy, and Fig. 2.2(b) is a profile of its distribution
across individual layers, averaged parallel to the interface within the area
indicated in the figure. It is observed that the InGaSb layers (bright yellow
regions) are in compressive strain, which is consistent with the In content
(x ¼ 0.25) in these layers, and that the InAs layers (bright green) are in tensile
strain of about 0.01, which is in agreement with theoretical calculations based
on published values of elastic constants. The InGaSb-on-InAs and InAs-on-
InGaSb interfaces are seen to exhibit strain inversion so that the overall strain
in these regions is negligible.
While, for the most part, the observations for sample A were similar to
other SLs reported in recent studies,11,12,14 it is important to emphasize
that the strain distribution profiles at the InGaSb-on-InAs interfaces for
this sample were distinctly different from those observed in earlier studies.
To elaborate further, we present results from a recent study14 that pertain
to sample D in Table 2.1. Although samples A and D are similar in design,
the HRXRD profiles for the two samples showed that sample A was in
compressive strain with respect to the GaSb substrate (« ¼ þ0.17%),
Electrical, Optical, and Structural Studies of InAs/InGaSb VLWIR Superlattices 45
Figure 2.2 Strain analysis of sample A showing (a) the map of the strain tensor «yy parallel
to the growth direction and (b) a plot of the strain profile averaged parallel to the interface,
over the boxed region in (a). (See color plate section.)
Figure 2.3 Strain analysis of sample D showing (a) the map of the strain tensor «yy parallel
to the growth direction and (b) a plot of the strain profile averaged parallel to the interface,
over the boxed region in (a). (See color plate section.)
avoids the band-tail issue. The measured bandgap energies are shown in
Table 2.1. For a much larger sample set than these few samples, the average
bandgap energy for the same SL design was 47 3 meV. The measured
bandgap energies are lower than the theoretically predicted value of 80 meV
by Grein et al.,7 and the cutoff wavelength of 19 mm, measured at the point
where the photoresponse intensity has dropped by 50%, is longer. The
HRXRD and cross-sectional TEM measurements confirm that the SL
samples grown were very close to the intended design so that the discrepancy
between theory and experiment must originate in the modeling.
A good photoresponse signal was measured despite the fact that the SL
total thickness was only 3% of the VLWIR photon wavelength. For a
maximum absorption coefficient of 2500 cm 1 at 5.6 mm, the absorption
quantum efficiency would only be 7% in the mid-infrared in a single pass,
and about half this value by l > 12.5 mm. The heavy fringing on the long-
wavelength portion of the spectrum is not noise but a result of multiple
internal reflections through the substrate/buffer/SL stack. The lightly n-type-
doped GaSb wafers do have significant infrared transparency.15 As shown in
Fig. 2.4, the SL has its maximum response over a wide wavelength range, up
to 15 mm.
The impact of reducing the interface strain on the SL photoresponse was
studied. For this study we compared sample A and sample D used in the
HRTEM study. These two samples were grown about a week apart such that
the overall MBE conditions were nearly the same except for the arsenic beam
equivalent pressure, which was set 30% higher for sample D. A comparison of
the two photoresponse spectra measured at 8 K is shown in Fig. 2.5. Overall,
48 Chapter 2
Figure 2.5 Comparison of the photoresponse spectra of samples A and D taken under the
same experimental conditions at 8 K.
the two spectra are very similar, having the same onset energy and 50% cutoff
wavelength. There is a difference in the magnitude of the photoresponse such
that sample D produced about 80% of the signal of sample A. Our run-to-run
variation of photoresponse intensity when remounting and retesting the same
sample is on the order of 10% or less. So, the intensity difference between these
two samples is larger than experimental variation. Still, the intensity difference
is not very large, so the impact of a negative strain spike at the InGaSb-on-InAs
interface may be relatively minor.
Figure 2.8 Mobility as a function of temperature for a lightly n-type GaSb substrate.
Figure 2.9 Mobility versus temperature (linear T scale) for alloyed and unalloyed SLs.
Sample B (o) and sample C (x).
Electrical, Optical, and Structural Studies of InAs/InGaSb VLWIR Superlattices 51
Figure 2.10 Calculated horizontal and vertical mobilities, using L ¼ 200 Å and D ¼ 1.65 Å,
and the measured horizontal mobility as a function of temperature for the 48.1-Å InAs/20.4-Å
GaSb SL.
52 Chapter 2
2.7 Summary
There are many advantages to using InGaSb alloys in the design and growth
of VLWIR InAs/(In)GaSb SLs: higher MBE growth temperature, simpler
interface growth and mitigation of strain spikes, reduced Auger rates,7 and
higher quantum efficiency. Alloy scattering of charge carriers was shown not
to be a factor in electronic transport, so there are no disadvantages to
incorporating indium into the GaSb layers. Variable-temperature Hall effect
measurements found that the in-plane mobility of electrons in the SL was
10,000 cm2/Vs below 80 K and was relatively constant in magnitude.
Theoretical modeling of the vertical transport in the IRS limit found that the
electron mobility perpendicular to the SL layers was nearly equivalent to the
in-plane mobility below 20 K but then had a stronger temperature dependence
and dropped to 6,000 cm2/Vs by 60 K. Vertical transport is an important
parameter in photodiode performance.
Electrical, Optical, and Structural Studies of InAs/InGaSb VLWIR Superlattices 53
References
1. E. R. Heller, K. Fischer, F. Szmulowicz, and F. L. Madarasz,
“Superlattice parameters for optimum absorption in InAs/InxGa1 xSb
superlattice infrared detectors,” J. Appl. Phys. 77, 5739 (1995).
2. J. J. Tietjen and L. R. Weisberg, “Electron mobility in GaAs1 xPx alloys,”
Appl. Phys. Lett. 7(10), 261–263 (1965).
3. P.-Y. Delaunay, B.-M. Nguyen, D. Hoffman, and M. Razeghi, “320256
infrared focal plane array based on type-II InAs/GaSb superlattice with a
12-m m cutoff wavelength,” Proc. SPIE 6542, 654204 (2007) [doi: 10.1117/
12.723832].
4. Y. Wei, A. Gin, M. Razeghi, and G. J. Brown, “Advanced InAs/GaSb
superlattice photovoltaic detectors for very long wavelength infrared
applications,” Appl. Phys. Lett. 80(18), 3262 (2002).
5. Y. Wei, A. Gin, M. Razeghi, and G. J. Brown, “Type-II InAs/GaSb
superlattice photovoltaic detectors with cutoff wavelength approaching
32 mm,” Appl. Phys. Lett. 81(19), 3675 (2002).
6. F. Szmulowicz, E.R. Heller, K. Fischer, and F. L. Madarasz,
“Optimization of absorption in InAs/InxGa1 xSb superlattices for long-
wavelength infrared detection,” Superlattices and Microstructures 17(4),
373–379 (1995).
7. C. H. Grein, W. H. Lau, T. L. Harbert, and M. E. Flatte, “Modeling of
very long infrared wavelength InAs/GaInSb strained layer superlattice
detectors,” Proc. SPIE 4795, 39–43 (2002) [doi: 10.1117/12.452265].
8. H. J. Haugan, G. J. Brown, and L. Grazulis, “Effect of interfacial
formation on the properties of very long wavelength InAs/GaSb super-
lattics,” J. Vac. Sci. Technol. B 29(3), 03C101 (2011).
54 Chapter 2
Said Elhamri graduated from Kenyon College and received his Ph.D. in
physics from the University of Cincinnati in 1995. Following a two-year
postdoctoral position, he joined the University of Dayton in 1997 where he is a
professor of physics. In recognition of his achievement in the classroom and
the lab, he has received both teaching and research awards. He has served as a
visiting scientist in the Nanoelectronic Materials Branch at Wright Patterson
Air Force Base on several occasions. His research interests are in transport
studies of quantum semiconductor structures, including nitride heterostruc-
tures, Sb-based type-II superlattices, and graphene.
Heather J. Haugan received her Doctoral degree in physics from the State
University of New York at Buffalo, and is currently a research scientist in
the Materials and Manufacturing Directorate at the Air Force Research
Laboratory. Her research has been focused on the molecular beam
epitaxial growth of novel detector materials such as InAs/Ga(In)Sb
superlattice materials. She has authored/coauthored approximately 60
peer-reviewed journal articles and a book chapter during the last 10 years,
and her papers have been cited over 500 times in leading journals. As an
acknowledged expert in this field, one of her proceeding papers was
highlighted in a review article in the prestigious industry-trade III-V Review
publication in 2006. Her article described significant new accomplishments
in detector materials.
3.1 Introduction
3.2 Design
3.3 Growth and Structural Characteristics
3.4 Optical Characteristics
3.5 Infrared Detectors
3.6 Summary
References
59
60 Chapter 3
3.1 Introduction
Semiconductor superlattices (SLs) have intrigued researchers for more than
40 years1,2 due to their electronic properties that reveal the quantum effects
occurring in the periodic structure of nanometer-thin layers of differing
materials. Mid-infrared SLs have advanced from a material of purely
scientific interest to one being practically implemented in devices for
detectors and lasers intended for military target recognition, chemical
detection, and night vision systems because of the advantages over devices
composed of bulk materials. SLs promise larger effective masses, leading to
lower tunneling currents,3 less Auger recombination arising from proper
bandgap engineering,4 longer wavelengths (smaller bandgaps) available than
in the individual materials provided from type-II band alignments, and
varying bandgaps designed by choosing compositions and layer thicknesses.
The advantages of type-II SLs (T2SLs) are expected to lead to lower dark
currents, higher operating temperatures, and greater quantum efficiencies
for infrared detectors.
As one of the many promising antimonide T2SLs, InAs/InAs1 xSbx
T2SLs were proposed for mid-wavelength infrared (MWIR) and long-
wavelength infrared (LWIR) light-emitting diodes, lasers, and photodetec-
tors.5 Although MWIR and LWIR InAs/InAs1 xSbx T2SL structures grown
on InAs substrates6 and GaSb substrates7 were successfully demonstrated in
the 1990s, they were set aside as potential infrared detector materials in favor
of the InAs/InxGa1 xSb SL.4 However, the short minority carrier lifetime, 30
ns,8 of LWIR InAs/InxGa1 xSb T2SLs currently limits the detector
performance as evidenced by higher-than-predicted dark currents9 and
reduced quantum efficiencies.10
Recent measurements using time-resolved photoluminescence (PL) on an
8-mm LWIR InAs/InAs1 xSbx T2SL show that the minority carrier lifetime is
greater than 412 ns at 77 K, more than one order of magnitude greater than
that of LWIR InAs/InxGa1 xSb SLs.11 Shortly after this result was reported,
another order of magnitude improvement in the carrier lifetime was
demonstrated: 9 ms at 77 K for a 5.4-mm MWIR InAs/InAs1 xSbx T2SL.12
The significant improvement in lifetime is mainly due to less Shockley–Read–
Hall (SRH) nonradiative recombination, possibly due to the lack of mid-
bandgap defect energy states related to Ga in the T2SL, and is expected to
improve the detector dark current, quantum efficiency, and operating
temperature. The detailed strain-balanced design, molecular beam epitaxy
(MBE) growth, structural properties from high-resolution x-ray diffraction
(HRXRD) and transmission electron microscopy (TEM), and optical
properties from PL and time-resolved PL are presented in this chapter. The
excellent quality of these T2SL samples exemplifies the promise of the InAs/
InAs1 xSbx SLs for infrared device applications.
InAs/InAs1xSbx Superlattices on GaSb Substrates… 61
3.2 Design
The design of the SL bandgap and the experimental results correlate well for
the InAs/GaSb SL as it has been studied for over 30 years. The InAs/
InAs1 xSbx SL, however, is less studied, and a few critical material parameters
for the SL design have been recently debated. The type-II band alignment
(Fig. 3.1) with the InAs conduction band higher in energy than the InAsSb
conduction band13 16 and the type-II band alignment with the InAsSb
conduction band higher in energy than the InAs conduction band5,6,16,17 have
both been reported. The magnitude of the band offsets for InAs and InAs1 xSbx
must be known, in addition to the band alignment, in order to design the SL
bandgap. The fractional valence-band offsets are defined as
DEc Ec InAs Ec InAsSb
Qc ¼ ¼ , ð3:1Þ
DEg Eg InAs Eg InAsSb
Qc þ Qv ¼ 1, ð3:3Þ
and the signs of the offsets (shown in Fig. 3.1) are based on the band
alignment type. The bandgap bowing of InAs1 xSbx adds an additional
variable that is not present for the InAs/GaSb SL. The InAs1 xSbx bandgap is
written as18
Eg InAsSb ¼ ð1 xÞEg InAs þ xEg InSb xð1 xÞCEg InAsSb , ð3:4Þ
Figure 3.1 Three possible band alignments between InAs and InAs1–xSbx (reprinted from
Ref. 20).
62 Chapter 3
where CEg InAsSb is the bandgap bowing factor. The recommended value for
CEg InAsSb is 0.67 eV,18 but a more recent value of 0.8 eV was found for
relaxed InAs1 xSbx alloys grown on compositionally graded buffer layers.19
The allocation of the bowing between the conduction and valence bands has
been assigned several different values in the literature.20 Figure 3.2 shows the
InAs1 xSbx conduction and valence bands versus x for different bowing
percentages in the valence band CEv bowing when the valence-band energy is
written as20
Ev InAsSb ¼ ð1 xÞEv InAs þ xEv InSb xð1 xÞCEv bowing : ð3:5Þ
Another important aspect of the InAs/InAs1 xSbx SL design is strain
balance, which is achieved in a slightly different manner than that used for the
InAs/GaSb SL. InAs/InAs1 xSbx SLs are strain balanced on GaSb by
choosing appropriate combinations of layer thicknesses and InAs1 xSbx alloy
compositions as calculated with the zero-stress method,21 which takes the
elastic constants of the layers into account. In order to achieve high-quality
material with low misfit dislocation densities, the critical thicknesses22 of InAs
and InAs1 xSbx on GaSb are used as the limits for the layer thicknesses in the
strain-balanced SL designs. GaSb is the ideal substrate for strain balancing
InAs/InAs1 xSbx SLs because its lattice constant is between that of the two
layers, eliminating the need for complicated, compositionally graded buffer
Figure 3.2 InAs1–xSbx conduction and valence bands calculated at 0 K with an InAs/
InSb valence-band offset of 0.59 eV, CEg_InAsSb of 0.67 eV, and different scenarios for the
InAs1–xSbx bandgap bowing distribution between the conduction and valence bands, which
can result in different band-edge alignments of InAs/InAs1–xSbx heterojunctions (reprinted
from Ref. 20). (See color plate section.)
InAs/InAs1xSbx Superlattices on GaSb Substrates… 63
layers and thus simplifying the growth process. Unlike the InAs/GaSb SL, it is
not necessary to control the interface between the two layers during the
growth of the InAs/InAs1 xSbx SL to achieve high-quality strain-balanced
materials.
To study the valence-band offsets and valence-band bowing at higher
Sb mole fractions of x ¼ 0.28–0.40, a set of InAs/InAs1 xSbx SL samples
were designed, starting with the type-II (Fig. 3.1) alignment with Ev InAs ¼
–0.59 eV,18 Ev InSb ¼ 0 eV,18 and CEv bowing ¼ –0.65 eV.23 The InAs, InSb,
and InAs1 xSbx material parameters were primarily taken from Ref. 18.
A three-band envelope-function approximation model,24 which includes coupling
between the electrons, light holes, and spin-orbit split-off holes, is used to
calculate the SL effective bandgap, defined as the electron-to-heavy-hole
transition energy.
The first set of InAs/InAs1 xSbx SL samples (see Table 3.1) was grown by
MBE on undoped 2-in GaSb substrates with 500-nm GaSb buffer layers at
IQE, Inc. The SLs consist of the number of periods necessary for a total
thickness of 500 nm, are unintentionally doped n-type, and are capped with
10 nm of pþ InAs. AlSb barriers,10 nm in thickness, above and below the SL
minimize surface recombination and therefore increase the PL signal intensity.
PL measurements were made with a Fourier-transform infrared (FTIR)
spectroscopy system using the double-modulation technique. The samples
were measured at 12 K in a cryostat with a ZnSe window and optically excited
with a 780-nm laser diode modulated at 50 kHz with an average power density
on the order of 10 W/cm2. The PL signal was collected with gold-coated
parabolic mirrors and analyzed with an FTIR system equipped with a 15-mm
HgCdTe detector. The PL results are shown in Fig. 3.3.
The same model used to design the samples was employed to fit the
measured PL peak positions using various valence-band offsets CEg InAsSb and
CEv bowing values as fitting parameters. Over the range of x ¼ 0.28–0.40,
CEv bowing was found to vary linearly with x, as shown in Fig. 3.4, for both
Table 3.1 Summary of the InAs/InAs1–xSbx SL samples grown by MBE at IQE, Inc. and
their peak PL emission wavelengths at 12 K (from Ref. 20).
Figure 3.4 The calculated InAs1–xSbx bandgap bowing attributed to the valence band for
CEg_InAsSb ¼ 0.67 eV (solid symbols) and for CEg_InAsSb ¼ 0.80 eV (open symbols) for the
IQE samples studied here and two sets of samples from Refs. 25 and 26. The model used
Ev_InAs ¼ –0.59 eV and Ev_InSb ¼ 0 eV (reprinted from Ref. 20).
InAs/InAs1xSbx Superlattices on GaSb Substrates… 65
CEg InAsSb ¼ 0.67 eV (solid symbols) and 0.80 eV (open symbols), with the
magnitude of CEv bowing decreasing as x increased. The CEv bowing values that
were fit using CEg InAsSb ¼ 0.80 eV (open symbols) are slightly larger than
those fit using CEg InAsSb ¼ 0.67 eV (solid symbols) with the discrepancy
increasing as x increases. The linear fits for CEv bowing for the sample set
studied here are shown extended to x ¼ 0.05 in Fig. 3.4 for CEg InAsSb ¼ 0.67 eV
(CEv bowing ¼ 1.58x – 0.62 eV as a dash-dot line) and for CEg InAsSb ¼ 0.80 eV
(CEv bowing ¼ 1.76x – 0.65 eV as a dotted line). Also shown are calculated
CEv bowing values for two sets of sample structures from the literature: MBE-
grown structures on InAs substrates25 and MOCVD-grown structures on GaSb
substrates.26 Although there is more spread in data from the literature,25,26 the
CEv bowing values lie reasonably along the line calculated for the data presented
here.20
The results are shown in terms of the fractional valence-band offset
Qv [Eq. (3.2)] for InAs and InAs1 xSbx strained on the substrate in Fig. 3.5 for
the present set of samples and for Liu25 and Lackner26 samples for the cases of
CEg InAsSb ¼ 0.67 eV (solid symbols) and CEg InAsSb ¼ 0.80 eV (open
symbols). Together, all three sets of samples cover the InAs1 xSbx
composition range of x ¼ 0.07–0.40. The best fitting Qv values for the three
sets are shown in Table 3.2 for the unstrained- and strained-on-the-substrate
cases. The results agree with each other within experimental error, with the
IQE set covering x ¼ 0.28–0.40 giving the lowest value and the least variance.
Figure 3.5 The fractional valence-band offset Qv versus x for InAs/InAs1–xSbx, strained on
the substrate, for the IQE samples studied here and two sets of samples from Refs. 25 and
26 (reprinted from Ref. 20).
66 Chapter 3
Table 3.2 Summary of the InAs/InAs1–xSbx fractional valence-band offset Qv for three sets
of SL structures (from Ref. 20).
Liu 25
1.26 0.06 1.09 0.06
Lackner26 1.25 0.09 1.08 0.08
IQE grown20 1.03 0.03 0.87 0.03
The Qv values for CEg InAsSb ¼ 0.80 eV (open symbols) are 0.30 less than
those for CEg InAsSb ¼ 0.67 eV (solid symbols). However, the calculated
fractional valence-band offset is observed to be fairly constant versus x for
both values of InAs1 xSbx bandgap bowing using the current model for the
range of samples studied. Therefore, designing InAs/InAs1 xSbx SLs using the
type-II alignment and Qv parameter is simpler than using the CEv bowing
parameter. The calculated bowing in the InAs1 xSbx valence band and the
fractional valence-band offset parameter seem to depend on the growth
conditions, resulting in the variation in the data shown in Figs. 3.4 and 3.5 for
the samples grown by three different machines. It should be noted that the
model used here assumes abrupt interfaces between the two SL layers. In
reality, As and Sb intermixing occurs during the SL growth and results in
compositional grading at the interfaces. The present calculations do not take
into account any interface grading or layers.
The experimental unstrained Qv parameter, 1.03, along with CEg InAsSb ¼
0.67 eV, was used to examine the trend of the SL bandgap versus the SL
layers’ thicknesses and the InAs1 xSbx composition, as shown in Fig. 3.6.
Within the composition range of x ¼ 0.30–0.45, InAs/InAs1 xSbx SL designs
are shown to cover the entire MWIR and LWIR ranges. For a given x,
smaller bandgaps (or longer wavelengths) result from thicker layers, but the
overlap between the electron and heavy-hole wavefunctions decreases with
increasing layer thicknesses. The absorption coefficient is proportional to the
square of the wavefunction overlap (shown in Fig. 3.7) and the density of
states, so larger wavefunction overlaps are desirable. To optimize the
wavefunction overlap for a particular bandgap, the layers should be thin with
larger x, but material growth becomes more challenging as x increases due to
the competition between arsenic and antimony for incorporation into the film.
To optimize the absorption coefficient, a tradeoff must occur between thin
InAs/InAs1xSbx Superlattices on GaSb Substrates… 67
Figure 3.6 Calculated effective bandgaps lg covering the MWIR and LWIR, for strain-
balanced type-II InAs/InAs1–xSbx SLs on GaSb substrates for four different InAs1–xSbx
compositions.
Figure 3.7 Calculated square of the electron and heavy-hole wavefunction overlap for
different strain-balanced type-II InAs/InAs1–xSbx SL designs having bandgaps equivalent to
8, 10, and 12 mm.
layers for large wavefunction overlaps and thicker layers for increased density
of states. More detailed calculations incorporating the density of states and the
wavefunction overlap are necessary to properly maximize the SL absorption at
a particular bandgap. The wavefunction overlap is inherently lower for the
longer-wavelength designs due to the larger x resulting in larger valence-band
68 Chapter 3
offsets and increased heavy-hole confinement. Figures 3.6 and 3.7 show the SL
bandgap and overlap saturating for very thin layers. This is due to the material
approaching an alloy for such thin layers.
Table 3.3 The structural and composition parameters of SL samples grown by MBE at
ASU and their peak PL emission wavelengths at 12 K (from Ref. 28).
Figure 3.8 Epilayer structure of the PL samples grown on GaSb substrates by MBE at
ASU and detailed in Table 3.3 (reprinted from Ref. 28).
Figure 3.9 HRXRD patterns of the ASU-grown InAs/InAs1–xSbx SLs with x 0.36 and
periods of 12, 20, and 24 nm (reprinted from Ref. 28).
Figure 3.10 Improvement in the SL zero-order peak FWHM to 20 arcsec during an ASU
growth campaign. The FWHM was measured with HRXRD (reprinted from Ref. 28).
Figure 3.11 HRXRD (004 and 113) patterns for ASU-grown InAs/InAs1–xSbx SL Sample 5
showing the SL0 peak overlapping the substrate peak, indicating perfect strain balance of
the SL on the GaSb substrate.
Figure 3.12 Electron micrographs showing typical cross sections of the ASU-grown InAs/
InAs1–xSbx SLs with x 0.36 and periods of (a) 12 nm, (b) 20 nm, and (c) 24 nm.
72 Chapter 3
Figure 3.13 PL spectra of the ASU T2SL structures covering the MWIR and LWIR bands
at 12 K. The measurements were carried out under identical conditions, and the same
order of magnitude PL intensity is conserved in all wavelength ranges, showing the
excellent optical properties of the samples and reproducibility of the MBE growths
(reprinted from Ref. 28). (See color plate section.)
at 50–60 kHz, and the PL signal was collected and collimated using parabolic
gold-coated mirrors and then detected by an mercury cadmium telluride
(MCT) detector.
Figure 3.13 shows the PL spectra covering 4–12 mm for the samples
listed in Table 3.3 measured under identical conditions at 12 K. The
samples with similar Sb compositions but increasing periods (Samples 5, 3, 8)
have longer peak wavelengths and smaller wavefunction overlaps, as
expected. The samples with similar periods but increasing Sb compositions
(Samples 1, 4, 5 and Samples 6, 8; the two groups have similar periods
within the group but different periods between the groups) also have longer
peak wavelengths. The PL spectra of Samples 6 and 7 illustrate that a larger
Sb composition with a smaller period can increase the wavefunction overlap
for designs of similar peak wavelengths, or bandgaps, agreeing with the
modeling results.
For thermally activated nonradiative recombination mechanisms, smaller
bandgaps result in higher nonradiative recombination rates, which lower the
PL intensity. Decreasing PL peak intensity with decreasing SL bandgap was
reported for InAs/InxGa1 xSb SLs with an x-axis intercept of 11.0 mm at
77 K.30 Figure 3.14 also shows a decrease in the total number of integrated PL
photons with increasing wavelength with an estimated x-axis intercept of
14.6 mm at 12 K. This qualitative discussion suggests that the optical quality
of these samples is conserved over the range of 4–12 mm. However, the
InAs/InAs1xSbx Superlattices on GaSb Substrates… 73
Figure 3.15 Temperature-dependent spectral PL data for (a) sample F and (b) sample B
from Table 3.1 (reprinted from Ref. 31). (See color plate section.)
Figure 3.17 Temperature-dependent lifetime data for samples (a) F and H and (b) B and G
from Table 3.1. The data points are connected with a spline to guide the eye (reprinted from
Ref. 31).
samples B and G show similar behavior in the lifetime data [see Fig. 3.17(b)]
for temperatures above 60 K: there is a local maximum around 200 K, from
which the lifetime decreases toward high temperature due to nonradiative
recombination and decreases toward low temperature due to radiative
recombination. The temperature of that local maximum varies with the SL
period and the calculated wavefunction overlap, as an increased overlap
(decreased SL period) results in a stronger contribution from radiative
recombination and therefore observation of radiative-dominant behavior at
higher temperatures. Samples F and H have periods of 245 Å and 180 Å,
InAs/InAs1xSbx Superlattices on GaSb Substrates… 77
3.6 Summary
InAs/InAs1 xSbx SLs being promising for longer carrier lifetimes but less
studied than InAs/GaSb SLs required an investigation of the material
parameters to properly design the strain-balanced SL bandgap. A model was
78 Chapter 3
developed to predict the InAs/InAs1 xSbx SL bandgap and refined with the
experimental results. For the range of x ¼ 0.28–0.40, the fractional valence-
band offset between InAs and InAs1 xSbx strained on GaSb was 1.75 0.03,
and the amount of bowing in the InAs1 xSbx valence band varied linearly with
x. To optimize the wavefunction overlap for a particular bandgap, the layers
should be thin with larger x; however, the wavefunction overlap is inherently
lower for longer-wavelength designs due to larger x resulting in larger valence-
band offsets and increased heavy-hole confinement.
The InAs/InAs1 xSbx SLs grown by MBE on GaSb substrates achieved
high structural quality, as indicated by HRXRD, in relatively few growth
runs compared to the InAs/Ga Sb SL. PL confirmed that SL bandgaps from
4 to 12 mm are easily attained with the InAs/InAs1 xSbx material system.
Temperature-dependent PL and TRPL indicated an order of magnitude
increase in the carrier lifetime and that radiative recombination dominated the
carrier lifetime at low temperatures. Some infrared detectors composed of
InAs/InAs1 xSbx SLs have been reported, but more work is required for the
infrared detector properties to improve so that the advantages offered by this
material are realized. Better understanding of the cause of the longer lifetime
of InAs/InAs1 xSbx SLs will enable further improvements in the material and
thus the detectors.
References
1. L. Esaki and R. Tsu, “Superlattice and negative differential conductivity
in semiconductors,” IBM J. Res. Dev. 14, 61–5 (1970).
2. G. Sai-Halasz, R. Tsu, and L. Esaki, “A new semiconductor superlattice,”
Appl. Phys. Lett. 30, 651–3 (1977).
3. D. L. Smith, T. C. McGill, and J. N. Schulman, “Advantages of the
HgTe-CdTe superlattice as an infrared detector material,” Appl. Phys.
Lett. 43, 180–2 (1983).
4. D. H. Chow, R. H. Miles, J. N. Schulman, D. A. Collins, and T. C. McGill,
“Type II superlattices for infrared detectors and devices,” Semicond. Sci.
Technol. 6, C47–C51 (1991).
5. Y.-H. Zhang, “InAs/InAsxSb1 x Type-II Superlattice Midwave Infrared
Lasers,” in Optoelectronic Properties of Semiconductors and Superlattices,
Vol. 3: Antimonide-Related Strained-Layer Heterostructures, M. O. Manasreh,
Ed., Gordon and Breach Science Publishers, Amsterdam, 461–500 (1997).
6. Y.-H. Zhang, “Continuous wave operation of InAs/InAsxSb1 x mid-
infrared lasers,” Appl. Phys. Lett. 66(2), 118 (1995).
7. A. Y. Lew, E. T. Yu, and Y. H. Zhang, “Atomic-scale structure of InAs/
InAs1 xSbx superlattices grown by modulated molecular beam epitaxy,”
J. Vac. Sci. Technol. B 14(4), 2940 (1996).
InAs/InAs1xSbx Superlattices on GaSb Substrates… 79
4.1 Introduction
4.2 Thermal Conductivity Tensor
4.2.1 Cross plane thermal conductivity
4.2.2 In plane thermal conductivity
4.2.3 Error analysis
4.3 Thermal Conductivity of T2SL
4.4 Thermal Distribution
4.5 Conclusion
Acknowledgments
References
4.1 Introduction
Optimization of high-power semiconductor light emitters requires knowledge
of the thermal conductivity tensor of its various functional semiconductor
layers from room temperature down to cryogenic temperatures. Optical
devices such as long-wavelength (LWIR) and midwave infrared (MWIR)
high-output light-emitting diodes (LEDs), quantum cascade lasers (QCLs),
and interband cascade lasers (ICLs) perform best at low temperatures in
continuous wave (CW) or are commonly operated not too much higher than
room temperature in pulsed mode,1 6 and the lifetime of an optical device
decreases exponentially with increasing temperature. Therefore, it is impor-
tant to know the thermal conductivity of the active region and cladding layer
85
86 Chapter 4
in such structures to model the heat flow, and it is preferable to use a material
with high thermal conductivity to deliver the heat out of the active region of
the device and reduce the operating temperature. Furthermore, most of the
optical devices nowadays are made of nanostructured materials, such as
semiconductor superlattices, which can have anisotropic thermal conductivi-
ties, either by virtue of their anisotropic structure or their anisotropic shape
with length scales below the phonon mean free path. The heat conduction
model can be more accurate if both the in-plane and cross-plane thermal
conductivity are calibrated.
We use the two-wire 3v method7 to characterize thin film thermal
conductivities for both in-plane and cross-plane directions.8 Previous
publications on the 3v method did not perform detailed error analysis on
different parameters or explicitly address the most important parameter. To
calibrate our experiment and analysis, we measured a 2.5-mm GaAs film from
20 K to 300 K, and analyzed different parameters of the fitting to determine
which experimental parameters are important for an accurate fitting.
Temperature-dependent thermal conductivities of the GaAs film show a
significantly smaller thermal conductivity than in the bulk GaAs substrate,
with only weak temperature dependence and no evidence of the low-
temperature phonon peak seen in the bulk material. We interpret this as
evidence that the phonon mean free path is limited by the GaAs layer
thickness of 2.5 mm.
InAs/GaSb type-II broken-gap superlattices (T2SLs) have been successfully
developed for use in LWIR and MWIR detectors9 13 with band-engineered
cutoff wavelengths and a reduction in dark current compared to direct-gap bulk
semiconductors. High-power LWIR and MWIR ICLs and photodiodes (PDs)
based on T2SL-band-structure engineering1 3,14 18 also offer tunable infrared
light emission. In this work, we measure the cross-plane thermal conductivity of
T2SLs19 from 13 K to 300 K using the 3v method.20 22 Using the measured
thermal conductivity of T2SL, we set up a power-law approximation for
modeling thermal conductivity.23 We calculate the temperature profile in the
active region using a simple analytical calculation. Adding the temperature
dependence to the thermal conductivity, we find that the maximum temperature
in the active region of an emitter will increase or decrease by a significant
amount with negative or positive power-law exponents in the thermal
conductivity, respectively. Therefore, the actual operating temperature of the
device can be easily estimated with these analytical tools.
through the metal filament and can heat it up by DTð2vÞ, so the resistance of
the filament will be changed by DRð2vÞ. By measuring the third harmonic
voltage across the filament V3v ¼ I ðvÞDRð2vÞ as a function of frequency, one
can deduce the thermal conductivity of the substrate.
A GaAs thin film on GaAs substrate was first used to calibrate our
measurement. The sample structure is shown in Figs. 4.2 (b) and (d). A d AlAs,f ¼
0.5-mm AlAs layer, composed of a digital alloy of 75 monolayers (MLs) of
AlAs for every 3 MLs of GaAs, was first grown on a (100) semi-insulating (SI)
GaAs substrate, followed by a d GaAs, f ¼ 2.5-mm GaAs film. For the 3v method
sample preparation, we etched the GaAs film from half the sample area,
sputtered a d SiO2 ,f ¼ 150-nm thick SiO2 layer as an insulating layer, and
deposited 200-nm-thick Au filaments as both heaters and thermometers. One
wide filament of 2b ¼ 30 mm and one narrow filament of 2b ¼ 2 mm were
deposited on both the film and the substrate, as shown in Fig. 4.1.
We measured the sample in an Oxford variable-temperature insert (VTI)
helium gas flow cryostat from 300 K down to 20 K, using standard lock-in
techniques. Since there are background 3v voltages from the lock-in power
source and nonlinear components in the measurement circuit and in the lock-
in A-B input channels, we also measured a reference background 3v signal
with a low-thermal-coefficient resistor of equal resistance to the Au filament.
We defined the important parameters in the 3v method as follows: k
represents the thermal conductivity; b is the heater half-width; d is the layer
thickness; v is the angular frequency of the current; a is the thermal
diffusivity; r is the material density; Cp is the specific heat capacity; the
Figure 4.1 (a) Mask design for two-wire 3v method. There are four pairs of 30-mm-wide
and 2-mm-wide filaments. Each filament has two current contacts and two voltage contacts.
(b) Photograph of the GaAs film sample. GaAs film is wet etched from the right half of the
sample, and the left half is film with substrate.
88 Chapter 4
Figure 4.2 (a) and (c) Frequency-dependent DT of the Au filaments on the thin film (solid
circles) and on the substrate (open circles) measured at 300 K. (b) and (d) Cross-sectional
layer structure of the sample. (a) DT values of the 30-mm Au filaments are used to calculate
the cross-plane thermal conductivity of the GaAs film kGaAs yy
,f
and the thermal conductivity of
the GaAs substrate k GaAs, b
. (c) DT values of the 2-mm Au filaments are used to fit the in-
plane thermal conductivity of the GaAs film kGaAsxx
,f
. The solid lines and the dashed lines are
the fitted curves for the film and substrate data within 100–2000 Hz. Extended data in 30–
6000 Hz are shown. (Adapted from Ref. 8.)
Thermal Conductivity and Thermal Distribution in Superlattice Structures 89
P df
kfyy ¼ , ð4:3Þ
2b l DT f
where the superscript s and f stands for the substrate and film, the subscript y
stands for the cross-plane direction, P is the power applied to the heater, and l
is the length of the Au filament. DT f ¼ DT fþs DT s , where DT fþs and DT s
are the temperature oscillation amplitude of the Au filament on the film-plus-
substrate and on the substrate alone, respectively.
The second step is to measure the narrow filament. We then used the two
2-mm-wide filaments to determine the in-plane thermal conductivity of the thin
film. In the multilayer-on-semi-infinite-substrate model for heat conduction, if
the sample has n layers, which may have different thermal conductivities, with
the index i layer starting from the top, the heater temperature oscillation
amplitude can be calculated with the iterative equation
Z1
P 1 sin2 ðblÞ
DT ¼ dl, ð4:4Þ
plkyy1 A1 B1 ðb2 l2 Þ
0
where
k B
Ai kyyiyyi1 Bii tanhðfi 1 Þ
Ai 1 ¼ k B
1
, i ¼ 2...n ð4:5Þ
1 Ai kyyiyyi1 Bii 1 tanhðfi 1 Þ
is an iterative parameter, the value of which for the upper layer is
calculated from that of the lower layers, and An ¼ –1 for the semi-infinite
substrate:
kxx 2 i2v 1=2
Bi ¼ l þ , ð4:6Þ
kyy ayyi
with the unit of wave number being related to the anisotropic thermal
diffusivity in the ith layer. l is an integration variable with the unit of wave
number; the dimensionless variable fi ¼ Bi di is related to the anisotropic heat
flow within a finite layer thickness; a ¼ k=rCp is the thermal diffusivity. We
estimated all of the fit parameters in Eq. (4.4) based on the literature values (k, r,
and Cp ) or measured values (b, d SiO2 , kGaAs, b , and kyy
GaAs, f
) and use the iterative
equation to calculate the heater temperature oscillation amplitude DT. Each fit
parameter is modified to calculate DT fit [solid and dashed lines in Fig. 4.2(c)]
that best matches the measured DT [solid and open circles in Fig. 4.2(c)].
(a) (b)
(c) (d)
Figure 4.3 Plot of the fitting error « for the 2-mm filament data calculated from Eq. (4.7) when
the value of each fit parameter varies by 20%. The parameters for the layer of interest are
plotted in black, and incidental fit parameters are plotted in gray. (a) We first fitted the data of
the 2-mm Au filament on the substrate at 300 K to determine the best-fit values for the effective
width of the Au filament on the substrate 2bs , the thickness of the SiO2 film d SiO2 ,f , the
cross-plane and in-plane thermal conductivity of SiO2 film kSiO yy
2 , f and kSiO2 , f , respectively,
xx
the cross-plane and in-plane thermal conductivity of GaAs substrate kGaAs yy
,b
and kGaAs
xx
,b
,
respectively, and the cross-plane and in-plane thermal conductivity of AlAs film kAlAs yy
,f
and kAlAs
xx
,f
, respectively (gray lines). (b) We then fitted the the data of the 2-mm Au filament on
the film at 300 K to determine the best-fit values for the effective width of the Au filament on the
film 2bf , the thickness of the GaAs film d GaAs,f (gray lines), and the in-plane and cross-plane
thermal conductivity of the GaAs film kGaAs yy
,f
and kGaAs
xx
,f
, respectively (black lines). (c) Fitting
error for the data of the 2-mm Au filaments on the substrate at 24 K. (d) Fitting error for the data
of the 2-mm Au filaments on the film at 24 K. (Adapted from Ref. 8.)
where DT fit ðvj Þ and DTðvj Þ are the fitted and measured DT at the angular
frequency vj , respectively. The best-fit value for each parameter gives the
smallest total error « in units of K2 .
We can now qualify which parameter is more important, and quantify the
uncertainty of the parameters of interest. Figure 4.3 shows the total error « as
a function of D, the percentage change from the best-fit value, for different fit
parameters. We first fitted the data of the 2-mm Au filament on the substrate
92 Chapter 4
to obtain the best-fit values for the effective width of the Au filament on the
substrate 2bs , the thickness of the SiO2 film d SiO2 , f , the cross-plane and in-
plane thermal conductivity of SiO2 film kSiO yy
2 ,f
and kSiO xx
2 ,f
, respectively, the
cross-plane and in-plane thermal conductivity of GaAs substrate kyy GaAs, b
and
kxx
GaAs, b
, respectively, and the cross-plane and in-plane thermal conductivity of
AlAs film kAlAs yy
,f
and kAlAs
xx
,f
, respectively [grey lines, Fig. 4.3(a) and (c)]. The
specific heat capacities Cp for all of the layers always give « values of less than
10 3 K2 , so we used the values from the literature.24,25 We then used these
parameter values to fit the the data of the 2-mm Au filament on the film and
determine the best-fit values for the effective width of the Au filament on the
film 2bf , the thickness of the GaAs film d GaAs, f [grey lines, Fig. 4.3(b) and (d)],
and the in-plane and cross-plane thermal conductivity of the GaAs film
kGaAs
xx
,f
and kyyGaAs, f
, respectively [black lines, Fig. 4.3(b) and (d)]. In addition,
the processing parameters d GaAs, f , d SiO2 , f , bs , and bf are the same for one
sample at all temperatures, so we only fitted their values at 300 K.
We found the most important fit parameters and the error bars from the
error analysis. The heater half-width b, the SiO2 layer thickness d SiO2 , and the
cross-plane thermal conductivity of the SiO2 film kSiO yy
2,f
are the most sensitive
parameters in the fitting. So the measurement uncertainty of these three
parameters is the major factor that determines the fitting resolution, i.e., the
error bars for the parameters of interest. For example, at 300 K, the maximum
uncertainty of these three parameters can be up to 4% with all of the other
fitted material parameters (kxx SiO2 , f
, kGaAs
yy
,b
, kxx
GaAs, b
, kyy
AlAs, f
, and kAlAs
xx
,f
) with
2
reasonable values, including the error of 0.01 K in Fig. 4.3(a). This error
value determines the estimation resolution of the in-plane thermal conductiv-
ity of the GaAs film kxx GaAs, f
in Fig. 4.3(b) to be 20% of the best-fit value, which
is the error bar in Fig. 4.4. To fit the data and estimate the thermal
conductivities more accurately, it is important to control and measure b and
d SiO2 accurately during the sample processing. For example, the lift-off resist
(LOR) can improve the lift-off process of the Au filament and the width b
uniformity. The cross-plane thermal conductivity of the SiO2 film kyy SiO2 , f
plays
an even more important role at low temperatures. 4% change of kyy SiO2 , f
may
induce 100% uncertainty of the in-plane thermal conductivity of the GaAs
film at 24 K [Figs. 4.3(b) and (d)], so the measured kxx GaAs, f
values are far less
accurate below 100 K.
We first compare the measured thermal conductivity of GaAs substrate to
the published value. The solid circles in Fig. 4.4 show the measured GaAs
bulk thermal conductivity as a function of temperature from 20 K to 300 K.
The measured thermal conductivity of the bulk GaAs substrate kGaAs, b is
isotropic and matches the published value (solid line) above 50 K but has a
Thermal Conductivity and Thermal Distribution in Superlattice Structures 93
Figure 4.4 Measured thermal conductivity of the GaAs substrate kGaAs,b (solid square)
matches the literature data (solid line26 3 1018 Zn curve). The difference between the
cross-plane kGaAs
yy
,f
and the in-plane thermal conductivity kxx
GaAs, f
of the GaAs film is less than
5% for all temperatures. Because the phonon mean free path of the GaAs film is much
shorter than that of the bulk GaAs substrate at low temperatures, the average values of
thermal conductivity of the film are 1.5 orders of magnitude lower than those of the bulk at
temperatures lower than 100 K. (Adapted from Ref. 8.)
Figure 4.5 Cross-plane thermal conductivity for T2SL samples (solid triangles). Measured
GaSb substrate thermal conductivity is shown in solid circles. Published data for bulk
thermal conductivity of InAs28 and GaSb29 are shown for comparison in solid lines. The
dashed line is a fitting polynomial for the average value of T2SL thermal conductivity
described in the text. (Adapted from Ref. 19.)
Thermal Conductivity and Thermal Distribution in Superlattice Structures 95
local power-law assumption, the exact solution for the temperature profile in
the active T2SL region with the same boundary conditions becomes
1
DTðxÞ sþ1
TðxÞ ¼ T0 1 þ ðs þ 1Þ ; ð4:14Þ
T0
where DTðxÞ is the same expression from Eq. (4.13), and s is the power-law
exponent. In real devices, the change in temperature DTðxÞ will normally not
exceed the absolute base temperature T0 , so this expression can be expanded
for small DTðxÞ=T0 < 1:
1 DT 2 ðxÞ
TðxÞ ’ T0 þ DTðxÞ s : ð4:15Þ
2 T0
The first two terms describe the thermal distribution for constant baseline
thermal conductivity k0 , and the third term accounts for the local power-law
assumption, proportional to the exponent s. Note that the position of
maximum temperature does not change between Eqs. (4.13) and (4.15), nor
does the thermal derivative boundary condition at x ¼ 0.
To compare how different exponent values might affect thermal
distribution in different temperature ranges, the thermal distribution is
plotted with the power-law exponent s of 1=2, 0, and 1 when the heat sink
temperature T0 is at 77 K and 300 K in Fig. 4.6, using Eq. (4.15) under high-
power dissipation P. Note that for the 300-K data, the realistic trace for the
T2SL is s ¼ 1=2, and for the 77-K data, the realistic trace is s ¼ 0.
The typical power density dissipated in PDs1,16 is 10–250 kW/cm2 , and
the threshold power density in ICLs3,17 is 0.3–2 kW/cm2 when pulsed and
1–5 kW/cm2 at CW. The dissipated power will be even larger when the device is
operated at high output powers. For these calculations, a voltage bias of 5 V is
applied to a 120 120 mm2 mesa, and the applied currents are 1000 mA, 500 mA,
and 250 mA, corresponding to powers of 5 W, 2.5 W, and 1.25 W, respectively.
The peak temperature is shown to be reduced when the exponent is
positive, while the peak temperature is increased when the exponent is
negative. For heat sink temperature T0 ¼ 300 K [Fig. 4.6(a)], the highest
power, 5 W, causes a temperature rise of 100 K in the active area, and this
value is increased by 9 K with a negative power exponent s ¼ 1=2, and
decreased by 18 K with a positive power exponent s ¼ 1. However for a lower
power of 2.5 W, the power-law corrections to the temperature rise are less
than 5 K, and for 1.25 W, the power-law corrections are less than 1 K. The
temperature rise is more significant in high-power devices with low thermal
conductivity, a result that is made clear in Eq. (4.13), where DT is
proportional to 1=k0 . At low temperature T0 ¼ 77 K, the power-law
correction is more than 20% of the temperature rise with the power exponent
s ¼ 1 at the highest power 5 W. The power-law exponent s has more effect at
98 Chapter 4
Figure 4.6 Calculated temperature profile in the active region normalized to the cold-sink
substrate temperature T0 with different power-law exponents s at three typical power
dissipations. A voltage bias of 5 V is applied to a 120 120 mm2 mesa, and the applied
currents are 1000 mA, 500 mA, and 250 mA, corresponding to powers of 5 W, 2.5 W,
and 1.25 W, respectively. (a) The substrate (heat sink) is at T0 ¼ 77 K. (b) The substrate is at
T0 ¼ 300 K. (Adapted from Ref. 23.)
4.5 Conclusion
In the two-wire 3v method, by the error analysis, the heater half-width b, the
insulating layer thickness d SiO2 , and the cross-plane thermal conductivity of
the insulating film kSiO
yy
2 ,f
are found to be the most sensitive parameters in the
Thermal Conductivity and Thermal Distribution in Superlattice Structures 99
Acknowledgments
This work is supported by AFOSR grants FA-9550-09-1-0237 and
FA-9550-12-1-0169, Initiative for Sustainability and Energy at Northwest-
ern (ISEN) and NSF MRSEC grants No. DMR-0520513 and No.
DMR-0748856 through both instrumentation grants and an NSF MRSEC
Fellowship.
References
1. E. J. Koerperick, J. T. Olesberg, J. L. Hicks, J. P. Prineas, and
T. F. Boggess, Jr., “High-power MWIR cascaded InAs-GaSb super-
lattice LEDs,” IEEE J. Quantum Elect. 45(7), 849–853 (2009).
2. E. J. Koerperick, D. T. Norton, J. T. Olesberg, B. V. Olson, J. P. Prineas,
and T. F. Boggess, Jr., “Cascaded superlattice InAs/GaSb light-emitting
diodes for operation in the long-wave infrared,” IEEE J. Quantum Elect.
47(1), 50–54 (2011).
3. I. Vurgaftman, C. L Canedy, C. S. Kim, M. Kim, W. W. Bewley, J. R.
Lindle, J. Abell, and J. R. Meyer, “Mid-infrared interband cascade
lasers operating at ambient temperatures,” New J. Phys. 11, 125015
(2009).
100 Chapter 4
5.1 Introduction
5.2 Superlinear Electroluminescence in GaSb based Narrow Gap Heterostructures with High
Conduction Band Offsets
5.3 Superlinear Electroluminescence in GaSb based Nanostructures with a Deep Al(As)Sb/
InAsSb/Al(As)Sb QW
5.4 Theoretical Consideration of Radiative and Auger Recombination in Deep QWs
5.5 Conclusions
5.6 Acknowledgment
References
105
106 Chapter 5
5.1 Introduction
Heterostructures and nanostructures with quantum wells (QWs) based on the
GaSb/InAs/AlSb system are promising for developing optoelectronic devices
(light-emitting diodes, lasers, photodetectors)1 6 because they cover the mid-
infrared spectral range (1.6–5.0 mm), where absorption bands of many natural
and industrial gases and other substances exist (CO2, CO, CH4, H2O, H2S,
NH3, and many others). However, the optical power and quantum efficiency
of light-emitting diodes (LEDs) based on the narrow-gap semiconductor
alloys (InAsSb, InGaAsSb) are not sufficiently high and are limited by
nonradiative Auger recombination. It was shown in Refs. 7 and 8 that Auger
recombination can be suppressed at the type-II abrupt heterointerface and in
nanostructures with deep QWs.
A system of InGaAsSb/GaSb alloys is also of interest due to the fact
that, by varying the composition of the layers, one can obtain type-II
heterojunctions with both staggered and broken-gap alignment.1 A specific
feature of type-II heterojunctions is that self-aligned QWs on both sides of the
boundary in which charge carriers are localized maintain the electrons and
holes spatially separated. In such structures it is possible to obtain emission at
longer wavelengths because of tunneling transitions of charge carriers through
the heteroboundary.9 Due to the interaction of carriers with the hetero-
boundary in type-II heterostructures, the Auger recombination process is
thresholdless, since the conservation law for the momentum component
perpendicular to the heteroboundary is not fulfilled. In this case, the Auger
recombination rate is a power function of temperature.7
The possibility of creating a laser for the mid-infrared spectral range based
on a semiconductor heterostructure with deep AlSb/InAs0.84Sb0.16 QWs was
theoretically studied,10 and parameters of such a QW where nonradiative
Auger recombination can be essentially suppressed were defined. Earlier it
was proposed to use large conduction-band offsets DEC (or DEV) in an
abrupt heterointerface with a narrow-gap active region for creation of hot
electrons that can cause impact ionization beginning from their zero
kinetic energy.11 An increase in the quantum yield of the internal
photoeffect in InSb due to impact ionization by high-energy photons was
considered for the first time by Tauc and Abrahám.12 Capasso et al.13
proposed the use of a large conduction-band offset at the heterointerface
in order to raise the electron ionization coefficient in comparison with the
hole ionization coefficient in a GaAs/AlGaAs superlattice avalanche
photodiode (APD) (both ionization coefficients can be approximately the
same in bulk material). However, this approach has not yet been applied
to light-emitting devices.
Thus, it is important to find a way to increase the quantum efficiency and
optical power of light-emitting devices based on narrow-gap semiconductors.
Superlinear Luminescence and Enhancement of Optical Power… 107
Figure 5.10 Emission optical power dependence on the drive current for n-GaSb/Al(As)Sb/
InAsSb/Al(As)Sb/p-GaSb heterostructure at 77 and 300 K. Points represent experimental
data; solid curve represents approximation according to P ¼ A IB (reprinted from Ref. 20).
Kane model.8 In this case, mixing of the hole states of p-symmetry with
electron s-states should be considered. Following this model, the hole energy
spectrum consists of three branches: heavy, light, and split-off holes.
Nevertheless, use of the ternary solution as an active region of the
heterostructure limits split-off hole participation in the carrier recombination
processes.21 This is achieved by decreasing the energy bandgap and
simultaneous increase of spin-splitting parameter Eg DSO. Additionally,
in the case of ml mh, the density of the light-hole energy states can be
neglected for the heavy-hole states. Thus, we will consider only electrons and
the heavy-hole spectrum.
116 Chapter 5
levels.10 Moreover, in deep QWs only the energy conservation law must be
completed.23
Radiative recombination in the structure under study is realized between
electron and hole quantum levels E1 and h1. The time of energy relaxation of
hot electrons to the first level and the time of impact ionization are much
shorter (<10 11 to 10 12 s) than the radiative recombination time (<10 8 to
10 9 s). Due to this fact, additional electrons produced by the impact
ionization process will be accumulated on the Ee1 quantum-sized level and
radiatively recombine with holes. Owing to the stronger overlapping of the
electron and hole wavefunctions, the impact ionization rate in QWs is higher
than that observed in bulk semiconductor structures.
The evaluation presented above shows that at QW width d ¼ 5-nm
potential drop (DEC – Ee1) exceeds the energy distance between the first
electron and the first hole level [Ee1 þ Eg (InAsSb) þ Eh1], so that the energy
conservation law presumes creation of one electron–hole pair:
ðDEC Ee1 Þ ¼ 0:876 eV > ðEe1 þ Eg ðInAsSbÞ þ Eh1 Þ at T ¼ 300 K;
ðDEC Ee1 Þ ¼ 0:877 eV > ðEe1 þ Eg ðInAsSbÞ þ Eh1 Þ at T ¼ 77 K,
where impact ionization threshold energies «ie for electrons are 0.669 eV
at 300 K and 0.717 eV at 77 K, due to the increased bandgap energy of
InAsSb by 48 meV at 77K. Since the threshold ionization energy for electrons
at 77 K is increased while the value (DEC Ee1) remains approximately the
same, the contribution of impact ionization in radiative recombination
reduces. That is why the exponent B at 300 K is higher than at low
temperatures, and we observed it in the experiment (see Fig. 5.10).
We have also considered dependence of the impact ionization intensity on
the width d of the QW. For RT, three situations are possible:
1. 4.2 nm < d < 6.6 nm: The energy conservation law allows one electron–
hole pair to be created because the following condition is fulfilled:
DEC – Ee1 > [Ee1 þ Eg (InAsSb) þ Eh1].
2. 6.6 nm < d < 10 nm: Creation of two electron–hole pairs is possible
because in this case the value (DEC – Ee1) is two times higher than the
distance between the first electron and the first hole level: DEC – Ee1 >
(2[Ee1 þ Eg (InAsSb)] þ Eh1).
3. d < 4.2 nm: Impact ionization is not possible because the energy threshold
condition is not met.
It is remarkable that if the energy of the incident electron is equal to the
energy distance between the first electron level Ee1 and the first hole level Eh1,
the “resonance” impact ionization can be observed, as in bulk semiconductors
of InAs and GaSb types when the energy bandgap is close to the spin–orbit
splitting value DSO.15
118 Chapter 5
Figure 5.11 Band diagram of a heterostructure with a deep QW; solid arrows show the
CHCC AR process, and dashed arrows show the CHHS AR process.
0 1
kc sin kc xh lc q cos kc x
iħg B C
Cc ðq, xÞ ¼ A1 @ iq cos kc xh þ ilc kc sin kc x A
Z
lc kc sin kc x þ lc q cos kc xh
0 1
kc cos kc x lc q sin kc xh
iħg @
þ A2 ilc kc cos kc xh iq sin kc x A, ð5:4Þ
Z
lc q sin kc x lc kc cos kc xh
120 Chapter 5
where
ð5:7Þ
Figure 5.12 Dependence of the energy spectrum of the (a) CHCC and (b) CHHS process
on the QW width.
where
e2
Mfi ¼
Cf ðr1 , r2 Þ C i ðr1 , r2 Þ ð5:11Þ
k0 j r1 r2 j
is the matrix element of the electron–electron interaction, r1 and r2 are the
carrier coordinates, e is the elementary charge, and k0 is the static relative
permittivity of the semiconductor.
In Eq. (5.10), f1 and f2 are the Fermi distribution functions of carriers in
the initial state, and f3 and f4 those in the final state. For highly excited states,
f4 can be taken to be zero. It should be noted that 1 f3 is the hole
distribution function.
Superlinear Luminescence and Enhancement of Optical Power… 123
where G is the optical confinement factor for a single QW, g0max is the
maximum gain, aint is the internal loss at the lasing threshold, aext ¼
ð1=LÞlnð1=RÞ is the mirror loss, and NQW is the number of QWs.
The relation for the maximum gain per QW, expressed in terms of the
carrier concentration within the well, has the form26
2 p a M mc Eg ðEg þ DSO Þ 1
g0 ¼ Icv p
max
1
a «1 þ M m0 2 E
Eg þ DSO g
3
" ! !#
n2D
QW 1 p2D
QW
1 exp exp , ð5:14Þ
Nc2D M Nc2D
Figure 5.14 (a) Threshold current density versus the number of QWs. (b) Internal quantum
efficiency versus the number of QWs.
5.5 Conclusions
This chapter reports on the observation of superlinear luminescence in the
spectral range of 0.2–0.8 eV and the enhancement of the optical power in bulk
GaSb-based type-II heterostructures with high potential barriers and
InGaAsSb narrow-gap active layer and in quantum-sized heterostructures
with deep Al(As)Sb/InAs0.84Sb0.16/Al(As)Sb QW grown by MOVPE on a
n-GaSb substrate.
It was established that in both types of structures under study the
obtained effects of superlinear luminescence can be explained by a similar
mechanism: the contribution to radiative recombination of extra electron–
hole pairs due to impact ionization by hot electrons that were heated via
high conduction-band offset at the interface of narrow-gap heterostructures
or in a deep QW (inverse Auger process).
Optical power increase in both types of structures is described by the same
power law P ¼ A IB, where A is the fitting parameter, I is the drive current,
and B is the exponent. B reaches 1.5–2.2 in bulk heterostructures and 2–3
in nanostructures at 77 and 300 K due to the stronger localization of carriers
in a deep QW. Theoretical estimations are in good agreement with the
experimental data.
Obtained results of the increase in optical power can be also applied for
improving the parameters of light-emitting devices as well as solar cells and
thermophotovoltaic cells. Theoretical estimations of radiative and Auger
recombination were considered depending on QW width, and design of a new
high-efficiency laser for the mid-infrared range operating at RT was proposed
as a further task.
126 Chapter 5
Acknowledgment
This work was supported by CSF Project 13-15286S, by grant of RBRF RAS
#12-00597, and by Program #24 of RAS Presidium.
References
1. M. P. Mikhailova and A. N. Titkov, “Type II heterojunctions in the
GaInAsSb/GaSb system,” Semicond. Sci. Technol. 9, 1279 (1994).
2. M. Mikhailova, N. Stoyanov, I. Andreev, B. Zhurtanov, S. Kizhaev,
E. Kunitsyna, K. Salikhov, and Yu. P. Yakovlev, “Optoelectronic
sensors on GaSb- and InAs-based heterostructures for ecological
monitoring and medical diagnostics,” Proc. SPIE 6285, 658526 (2007)
[doi: 10.1117/12/723507].
3. A. N. Baranov, B. E. Dzhurtanov, A. N. Imenkov, A. A. Rogachev,
Y. M. Shernyakov, and Yu. P. Yakovlev, “Generation of coherent
radiation in a quantum-well structure with one heterojunction,” Sov.
Phys. Semicond. 20, 1385–1387 (1986).
4. D. Barate, R. Teissier, Y. Wang, and A. N. Baranov, “Short wavelength
intersubband emission from InAs/AlSb quantum cascade structures,”
Appl. Phys. Lett. 87, 051103 (2005).
5. R. Teissier, D. Barate, Y. Devinson, A. N. Baranov, X. Marcadet, C.
Bernard, and C. Sirtory, “Antomonide quantum cascade lasers for the
3-5 mm wavelength spectral range,” Inst. Phys. Conf. Ser. 187, p. 307 (2005).
6. M. P. Mikhailova, I. A. Andreev, K. D. Moiseev, E. V. Ivanov, N. D.
Stoyanov, Yu. P. Yakovlev, E. Hulicius, A. Hospodková, J. Pangrác,
K. Melichar, and T. Šimeček, “Photovoltaic detector based on type II
p-InAs/AlSb/InAsSb/AlSb/p-GaSb heterostructures with a single quan-
tum well for mid-infrared spectral range,” Proc. SPIE 7138, 713813 (2008)
[doi: 10.1117/12.818007].
7. G. G. Zegrya and A. D. Andreev, “Theory of the recombination of
nonequilibrium carriers in type-II heterostructures,” Zh. Eksp. Teor. Fiz.
109, 615–638 (1996).
8. L. V. Danilov and G. G. Zegrya, “Theoretical study of Auger recombina-
tion processes in deep quantum wells,” Semicond. 42, 550–556 (2008).
9. H. Kroemer and G. Griffiths, “Staggered-lineup heterojunctions as
sources of tunable below-gap radiation: Operating principle and
semiconductor selection,” IEEE Electron Device Lett. 4, 20–22 (1983).
10. L. V. Danilov and G. G. Zegrya, “Threshold characteristics of an IR laser
based on deep InAsSb/AlSb quantum wells,” Semicond. 42, 557–562 (2008).
11. M. Z. Zhingarev, V. I. Korolkov, M. P. Mikhailova, V. V. Sazonov, and
D. N. Tretyakov, “Avalanche multiplication and coefficients of impact
Superlinear Luminescence and Enhancement of Optical Power… 127
6.1 Introduction
6.2 Molecular Beam Epitaxy Growth of InSb Quantum Dots
6.3 Characterization of InSb Quantum Dots
6.4 MBE Growth of GaSb Quantum Dots
6.5 Solar Cells Containing Stacks of GaSb Quantum Rings
6.6 Summary
Acknowledgments
References
6.1 Introduction
The 3D confinement of carriers in quantum dot (QD) structures offers an
attractive alternative compared with bulk or quantum well (QW) structures
for optoelectronic devices because of the improved (d-like) density of states
(DOS) leading to higher radiative transition rates, narrower spectral
linewidth, and the possibility to minimize Auger recombination. This chapter
133
134 Chapter 6
is concerned with the molecular beam epitaxial growth and optical properties
of self-assembled InSb and GaSb QDs in InAs and GaAs, respectively.
Section 6.2 describes investigations into the growth of InSb on InAs using
conventional Stranski–Krastanov growth. In Section 6.3 the structural and
optical properties of InSb submonolayer QDs grown using an Sb–As
exchange reaction are described, and their performance within a p-i-n LED
is evaluated. The growth of GaSb QDs in GaAs is reported in Section 6.4
together with the formation of quantum rings (QRs), and the application of
stacks of such GaSb QRs within a solar cell is described in Section 6.5.
Figure 6.1 (a) PL spectra (80 K) of the InSb QD structures grown at different temperatures.
The numbers on the left correspond to the InSb nominal thickness calculated from XRD
simulation. (b) Power-dependent PL spectra at 4 K for sample A0150 (reprinted from Ref. 6).
(See color plate section.)
using either (Sb2,As2) or (Sb4,As4). We found that for dimers the growth
temperature should be reduced by 50–100 8C in order to obtain the same PL
wavelength when using tetramers. XRD measurements revealed that for the
structures grown using (Sb2,As2), the amount of InSb deposited is less than for
(Sb4,As4) at the same growth temperature. This causes the PL to shift toward
shorter wavelengths since the reversed As-to-Sb exchange interaction is easier
under As2 flux, which is more aggressive than As4. This shift indicates that
there exists a dynamic equilibrium between Sb-to-As and As-to-Sb exchange
that depends on the growth temperature and As and Sb fluxes. However, no
difference was found in the behavior of Sb2 and Sb4 species for the growth
temperatures used.
When using Sb-to-As exchange with Sb2 and As2 cracker cells, low growth
temperatures (<350 8C) are needed to form QDs with sufficient localization
energy. However, EL from LEDs containing such QDs was weak at RT. At such
low growth temperatures, the material quality is poor and there is strong surface
segregation of Sb atoms in the surrounding InAs matrix that forms rough
interfaces and that alters the band structure considerably. This problem could be
overcome by using Sb4/As4, but then long growth interruptions are needed to
change the flux when growing laser structures, and the material quality is poor.
To overcome these difficulties and to grow larger QDs at higher growth
temperatures using cracker cells for LEDs, a short InSb (0.7 ML) deposition by
MEE was used following the Sb exchange at 430 8C. PL spectra from a sample
grown using this technique are shown in Fig. 6.2, where the QD peak dominates
most of the temperature range and is comparable in intensity to InAs at RT. LED
Antimonide Quantum Dot Nanostructures for Novel Photonic Device Applications 137
Figure 6.2 (a) Temperature-dependent PL emission spectra for 10 layers of InSb QDs
grown using a short InSb deposition following the exchange at 430 8C. (b) Dependence of
the PL intensity for the 10 layers of InSb QD and the surrounding bulk InAs matrix. (See
color plate section.)
Figure 6.3 RT emission spectra at various injection currents from a type-II InSb/InAs QD
LED structure with a lattice-matched AlGaAsSb barrier (reprinted from Ref. 7). (See color
plate section.)
138 Chapter 6
shows the RT EL spectra for one of the QD LEDs with a blocking barrier where
a single QD-related peak is observed at 3.8 mm. Note that the InAs-related
emission at 3.4 mm is hardly observable in these spectra. LED output power
under quasi-continuous drive conditions (1 kHz at 50% duty cycle) was
measured at RT using an integrating sphere and calibrated PbSe photodetector.
The corresponding internal quantum efficiency and output power were
estimated to be 0.4% and 6 mW, respectively. This value can be significantly
improved, and further device optimization is anticipated to lead to mid-infrared
sources that could be used in a variety of applications.
Figure 6.4 (a) HRTEM and (b) filled-state STM images illustrating excellent material
contrast between GaAs- and GaSb-rich regions and showing no crystal defects. (c) Low-
magnification dark-field 002 TEM showing several GaSb QRs (light) in the GaAs (dark)
matrix. (d) A local-electron DOS image of a similar ring taken using an STM reveals a
maximum toward the center of the ring. The dashed ellipses mark GaSb-rich lobes, as a
guide. The image in Fig. 6.4(d) was taken with a tip bias of þ0.44 V. As the latter technique
requires a tunnel current to facilitate local DOS measurements, a low level of background
doping (n-type 1017cm3 using GaTe) was introduced to the growth of all layers in the
sample (reprinted from Ref. 12). (See color plate section.)
140 Chapter 6
evidence of crystal defects. When the rings are measured in cross section, two
GaSb-rich lobes are typically seen, with separations varying between 0 and
25 nm, dependent on the position of the section with respect to the center of
the ring (Fig. 6.4). The major radius of the rings is unlikely to influence the
heavy-hole wavefunction in this system, as it is expected to be localized in a
relatively small section of the ring. Despite this, the geometry does strongly
influence the excitonic properties of the system. STM measurements indicate
that the GaAs in the center of the ring is quite pure, i.e., contains much less Sb
than that surrounding it. The local electron DOS measurements confirm the
presence of a maximum toward the center of the ring, as shown in Fig. 6.4(d).
Figure 6.5 shows the high-resolution XRD (HRXRD) spectra obtained
from a stack of GaSb/GaAs QRs where several pronounced high-order
satellite peaks are observed. The full width at half maximum (FWHM) of the
satellite peaks is approximately 50 arcsec and indicates the high crystalline
perfection of the nanostructured layers. Figure 6.5 (b) shows the correspond-
ing PL spectra measured at 4 K for the 5 and 10 layer where three
characteristic peaks can be clearly identified: one at 1.48 eV from GaAs, a
wetting layer peak near 1.3 eV, and a peak at 1.1 eV, corresponding to
recombination of holes localized in the GaSb QRs with electrons in the
surrounding GaAs. Despite the type-II band alignment, the intense peak from
the GaSb QRs indicates the strong exciton oscillator strength necessary for
efficient light absorption. This is due to the reduced strain around the QR that
allows the electron to sit near or inside the QR in close proximity with the
confined holes.
Recently, we developed a new growth technique that does not use a cold
cap to produce (ten) stacked layers of high-purity GaSb/GaAs QRs.18 The
resulting samples exhibit excellent crystalline quality and long-wavelength PL
peaking around 1.31 mm at 1.4 K. Analysis of the PL spectra reveals that a
GaAs cold cap increases the number of carbon acceptors, providing extra
holes to the QRs that then induce a blue-shift of the emission energy due to
capacitive charging. Figure 6.6(a) shows the calculated band structure along
with the first confined heavy-hole state for one such sample. The transition
and localization energies are E ¼ 0.925 and 0.595 eV, respectively, in excellent
agreement with those calculated previously.19 Figure 6.6(b) plots the change
in emission energy as a function of QR height (the range is chosen to be
roughly equal to the uncertainty in the TEM measurements). The emission
energy shows only a small variation, 20 meV, since confinement energies for
holes are less sensitive to size effects than are electrons. This is smaller than the
difference observed between the PL peak energies, 40 meV, and therefore
cannot explain the observed redshift. The size and shape of the QRs can also
influence the PL energy; however, this can be discounted since in these
samples the dimensions of the QRs are nearly identical. One possible
explanation is that the higher Sb composition in the GaAs matrix would
Antimonide Quantum Dot Nanostructures for Novel Photonic Device Applications 141
Figure 6.5 (a) HRXRD spectra obtained from the GaSb/GaAs QR solar cells where several
pronounced high-order satellite peaks are observed. The FWHM of the satellite peaks is
approximately 50 arcsec. (b) PL spectra measured at 4 K for the 5- and 10-layer samples
where the transitions corresponding to the GaSb/GaAs QRs, wetting layer, and GaAs are
identified. The intense emission from the QRs indicates the high exciton oscillator strength.
(See color plate section.)
reduce the emission energy, but the GaAs content inside and near the top of
the QRs where the electrons reside is very pure in both samples. Detailed
calculations regarding the effect of the Sb segregation on the strain and
emission energy can be found in Ref. 18. A likely explanation for the large
142 Chapter 6
Figure 6.6 (a) Calculated band structure of the GaSb/GaAs QR structure showing the
first confined heavy-hole (hh) state. (b) QR peak energy as a function of the QR height.
(c) Schematic illustrating the mechanism for charging the rings at zero laser power, in which
holes are supplied from carbon acceptors present in the GaAs cap. The additional holes in
sample B are due to the higher level of carbon acceptors present in the cold cap (reprinted
from Ref. 20).
Although the introduction of QDs into the active region of a solar cell can
extend the absorption edge into the infrared spectral range, thus increasing the
short-circuit current, it is necessary to stack multiple layers to increase the
light absorption. However, the large mismatch (7.8%) makes stacking GaSb
QDs difficult, and threading dislocations may be generated due to buildup of
internal strain. Previous work on GaSb/GaAs QD solar cells has focused on
the use of GaSb QDs grown using interfacial misfit (IFM) growth, which
relieves the strain energy associated with the lattice mismatch by the
formation of lateral-propagating dislocations.8 The resulting solar cells
exhibited extended infrared spectral response with short-circuit current and
open-circuit voltage of 1.29 mA/cm2 and 0.37 V, respectively. However, an
alternative strategy for stacking multiple layers is possible through the use of a
GaSb QR, which reduces the net strain without generating lateral or threading
dislocations.32 A schematic of the GaSb QR solar cell structure is shown in
Fig. 6.7(a). A 3-mm-thick n-type GaAs:Te base layer was grown first at 570 8C
with a doping density of 1017cm 3. This was followed by an intrinsic region
containing either 5 or 10 GaSb QR sheets, a 0.5-mm p-type GaAs:Be emitter
layer with doping density of 2 1018 cm 3, followed by a 30-nm Al0.8Ga0.2As
window layer and a 40-nm GaAs cap. A p-i-n GaAs control cell was also
grown with a 400-nm intrinsic region that contained no QRs. (The QRs were
grown using the abovementioned procedures). Figure 6.7(b) shows an image
of the fabricated (3.5-mm diameter) solar cell employing a circular metal grid
design. The corresponding energy band diagram of the structure is shown in
Fig. 6.7(c). Absorption of photons with energy below the bandgap of GaAs
excites electrons from the discrete states in the GaSb QRs into the GaAs
matrix. The photogenerated electron–hole pairs are then separated under the
influence of the built-in electric field.
The spectral response of the solar cells was measured using a 100-W
tungsten-halogen light source through a 0.25-m monochromator. Current
density (J)–voltage (V) curves were obtained under 1-sun illumination
(AM 1.5) using a 150-W Oriel® solar simulator. Figure 6.8(a) shows the
external quantum efficiency (EQE) for the solar cells [without antireflection
(AR) coatings]. The QR solar cells show extended long-wavelength
photoresponse [Fig. 6.8(b)] compared to the GaAs control cell, due to
absorption of low-energy photons. The increased response at 950 nm is due to
absorption within the wetting layer, whereas the longer-wavelength absorp-
tion up to 1400 nm is associated with the QRs. The long-wavelength EQE is
increased by increasing the number of QR stacks. Although the QR solar cells
show enhanced absorption below the GaAs bandgap, the EQE is reduced
above the bandgap. The EQE above the GaAs bandgap can be understood
using a 1D drift-diffusion model that divides the cell into emitter, depletion
region, and base regions and where the minority-carrier diffusion length is
dependent on the doping concentration.34,35 Figure 6.9 shows the fit obtained
Antimonide Quantum Dot Nanostructures for Novel Photonic Device Applications 145
Figure 6.7 (a) Schematic of the GaSb/GaAs QR solar cell structure grown by MBE
(reprinted from Ref. 33). (b) (top) Image of the fabricated (3.5-mm diameter) solar cell
employing a (bottom) circular metal grid design. (c) Schematic energy band diagram of the
structure under illumination. Absorption of photons with energy below the bandgap of GaAs
excites electrons from the discrete states in the GaSb QRs into the GaAs matrix. The
photogenerated electron hole pairs are then separated under the influence of the built-in
electric field. Also shown is the capture of holes and the recombination loss associated with
the n-Base current.
for both the QR solar cell and the GaAs control cell. The model shows that
the decrease in the EQE is mainly due to a reduction in the base current.
Minority holes that are photogenerated in the base region and undergo drift in
the depletion region are captured by the QRs, reducing the short-circuit
current. These trapped holes then act as recombination centers, thereby
decreasing the open-circuit voltage.
146 Chapter 6
Figure 6.8 (a) Spectral response for the GaSb/GaAs QR and GaAs control solar cells.
(b) Magnified view of the spectral response at longer wavelengths showing the enhanced
response for the QR solar cells due to the absorption of low-energy photons (adapted
from Ref. 32). (See color plate section.)
Figure 6.10 shows the illuminated 1-sun (AM 1.5) I-V characteristics for
the QRs and GaAs control solar cells, and Table 6.1 lists the corresponding
sample details, open-circuit voltage Voc, short-circuit current Jsc, fill factor
(FF), and efficiency. An enhancement of Jsc in all of the GaSb QR samples
compared to the GaAs control cell is observed due to the absorption of long-
wavelength photons by the QRs. Our best results to date were obtained for the
5-layer 1.5-ML solar cell that showed an open-circuit voltage of 0.62 V and a
short-circuit current of 20.7 mA/cm2, which are higher than those reported for
GaSb/GaAs QD solar cells grown using the IFM growth method. The Jsc and
Voc decrease as the size of the GaSb QRs and the number of stacks increase.
The lower Voc reduces the FF and efficiency compared to the GaAs control
cell, despite the increased Jsc.
Antimonide Quantum Dot Nanostructures for Novel Photonic Device Applications 147
Figure 6.9 Spectral response and curve fitting for (a) GaAs control cell, (b) 10-layer GaSb/
GaAs QR solar cell, and (c) 5-layer GaSb/GaAs QR solar cell. The contributions from the
emitter (electron current), base (hole current), and depletion region are labeled along with
the total contribution (&) and experimental data (!). It can be observed that the introduction
of QRs reduces the cell base current (reprinted from Ref. 32). (See color plate section.)
6.6 Summary
The MBE growth and characteristics of InSb and GaSb QD nanostructures
has been investigated with a view toward their incorporation into photonic
devices. Self-assembled InSb submonolayer QDs in an InAs matrix that have
excellent structural quality and exhibit bright PL up to RT have been grown
by MBE using Sb2 and As2 fluxes. Light-emitting diodes containing 10 layers
148 Chapter 6
Figure 6.10 Current density voltage curves for the GaAs control and GaSb/GaAs QR
solar cells (SCs) obtained using 1-sun AM-1.5 illumination. The short-circuit current is
enhanced by 5.9% in the solar cell containing 10 layers of QRs (reprinted from Ref. 32). (See
color plate section.)
Table 6.1 Open-circuit voltage (Voc), short-circuit current (Jsc), fill factor (FF) and efficiency
for each of the solar cells studied.
Acknowledgments
This work was supported by the Engineering and Physical Sciences Research
Council (grant numbers EP/H006419, EP/E028209/1, and EP/G070334/1) and
the EU Marie-Curie Training network (PROPHET) PITN-GA-2010-264687,
also in the framework of the QD2D project, and the Royal Society-Brian
Mercer Feasibility Award. The authors are grateful to A. Sanchez and
R. Beanland (University of Warwick) and K T. Lai and S. K. Haywood
(University of Hull) for TEM and FTIR PL measurements, respectively, as
well as V. Solov’ev (Ioffe Institute) for assistance with the exchange growth of
InSb QDs. P. J. Carrington acknowledges support from the Royal Academy
of Engineering through a Research Fellowship.
References
1. A. Krier, X. L. Huang, and A. Hammiche, “Midinfrared photolumines-
cence of InAsSb quantum dots grown by liquid phase epitaxy,” Appl.
Phys. Lett. 77(23), 3791 (2000).
2. A. G. Norman, N. J. Mason, M. J. Fisher, J. Richardson, A. Krier, P. J.
Walker, and G. R. Booker, “Structural and optical characterisation of
MOVPE self-assmebled InSb wuantum dots in InAs and GaSb matrices,”
Inst. Phys. Conf. Ser. 137, 353 (1997).
3. S. V. Ivanov, A. N. Semenov, V. A. Solov’ev, O. G. Lyublinskaya, Ya. V.
Terent’ev, B. Ya. Meltser, L. A. Prokopova, A. A. Sitnikova, A. A.
Usikova, A. A. Toropov, and P. S. Kop’ev, “Molecular beam epitaxy of
type II InSb/InAs nanostructures with InSb submonolayers,” J. Cryst.
Growth 278, 72–77 (2005).
4. O. G. Lyublinskaya, V. A. Solov’ev, A. N. Semenov, B. Ya. Meltser,
Terent’ev Ya. V. Terent’ev, L. A. Prokopova, A. A. Toropov, A. A.
Sitnikova, O. V. Rykhova, S. V. Ivanov, K. Thonke, and R. Sauer,
“Temperature-dependent photoluminescence from type-II InSb/InAs
quantum dots,” J. Appl. Phys. 99, 093517 (2006).
5. V. A. Solov’ev, I. V. Sedova, O. G. Lyublinskaya, A. N. Semenov,
B. Ya. Meltser, S. V. Sorokin, Ya. V. Terent’ev, and S. V. Ivanov,
“Midinfrared injection-pumped laser based on a III-V/II-VI hybrid
heterostructure with submonolayer InSb insets,” Tech. Phys. Lett. 31,
235–237 (2005).
6. P. J. Carrington, V. A. Solov’ev, Q. Zhuang, S. V. Ivanov, and A. Krier,
“InSb quantum dot LEDs grown by molecular beam epitaxy for mid-
infrared applications,” Microelectron. J. 40(3), 469–472 (2009).
150 Chapter 6
7.1 Introduction
7.2 Review of Te Doped GaSb Growth
7.3 Dopant Impurities
7.4 Growth of Te Doped GaSb
7.4.1 Metalorganic sources
7.4.2 Growth condition
7.5 Characterization
7.6 Results and Discussion
7.6.1 Surface morphology and growth rate
7.6.2 Electrical properties
7.7 Conclusions
References
7.1 Introduction
Antimony-based III-V semiconductors grown either lattice matched or
slightly strained on GaSb substrates have received much attention both for
their potential applications as optical devices in the wavelength of 1–4 mm
and for their potential use in tunnelling structures, exploiting the heterojunc-
tion offset with the InAs, GaSb, and AlSb material systems. Although
investigations in laser diodes have been carried out by many groups,1 4
problems still remain with the growth of a device-quality GaSb layer.5
157
158 Chapter 7
Figure 7.1 Impurity levels of GaSb (reprinted from Ref. 21 with permission; © 1997, AIP
Publishing LLC).
n-Type Doping in GaSb using Dimethyltellurium (DMTe)… 161
Formula GaðCH3 Þ3
Formula weight 114.83
Melting point, 8C 15.8
Boiling point, 8C 55.7
Density, grams/cc 1.10
Appearance Colorless liquid
Storage Stable under most ambient conditions when packaged in leak tight cylinders
Elemental analysis Si detected
NMR* results No organic impurities detected
No oxygen detected
*nuclear magnetic resonance
Formula SbðCH3 Þ3
Formula weight 166.86
Melting point, 8C 87.6
Boiling point, 8C 80.6
Density, grams/cc 1.528
Appearance Colorless liquid
Storage stable under most ambient conditions when packaged in leak tight cylinders
Elemental analysis Te detected
NMR results No organic impurities detected
Formula ðCH3 Þ2 Te
Formula weight 157.68
Melting point, 8C 10
Boiling point, 8C 91 92
Density, grams/cc
Appearance Light yellow to orange liquid
Storage Stable under most ambient conditions when packaged in leak tight cylinders
and bubblers
Elemental analysis
NMR results No organic impurities detected
No oxygen detected
to 7.1 10 6, a growth temperature of 540 8C, and a total flow rate of 8 l/min.
Values of DMTe flow with corresponding mole fractions of DMTe in the
input gas stream for various samples are given in Table 7.2. Substrates used
were SI GaAs(100) from Freiberger Compound Materials GmbH (Germany).
n-Type Doping in GaSb using Dimethyltellurium (DMTe)… 163
Table 7.2 Values of DMTe flows with corresponding mole fractions of DMTe for n-type
GaSb growth.
#37 10 7.08 10 6
#36 5 3.54 10 6
#44 3 2.12 10 6
#53 2 1.42 10 6
#55 1 7.10 10 7
Prior to the growth, substrates were immersed for 5 min in each of the
following: trichloroethylene at a temperature of 100 8C, acetone, and methanol,
and then etched in H2 SO4 : H2 O2 : H2 O ¼ 1 : 1 : 8 solution for 30 s followed by
a DI water rinse. The substrates were then blown dry by N2, before being
loaded into the reactor.
7.5 Characterization
Van der Pauw Hall measurements were performed in the temperature range of
20–300 K on the 3-mm thick epilayers grown on the SI GaAs substrates.
Photographs of sample morphology were taken using a Nomarski phase
contrast microscope.
Figure 7.3 Nomarski photograph of Te-doped n-GaSb grown on SI GaAs (reprinted from
Ref. 29 with permission; © 2011 Carl Hanser Verlag, Munich). (See color plate section.)
Figure 7.4 The growth rate of Te-doped GaSb as a function of DMTe flow.
Table 7.3 Hall measurements on Te-doped GaSb layers as a function of DMTe vapor
pressure.
Ehsani et al.31 for the growth of InGaSb with DETe doping: the electron
concentration was found to increase linearly as the DETe increased, reaching
a saturated concentration of 1.5 1018 cm 3. When the DETe mole fraction
was increased further, the electron concentration decreased.
To analyze our Hall measurement results, a two-conduction-band model
is required. According to Sagar,32 the conduction-band L valleys of GaSb,
which have a high density of states (DOS) and a low mobility, lie very close in
energy to the central G minimum. The measured Hall concentration is a
weighted average of the G-valley and L-valley mobility. Therefore, the Hall
measurement will underestimate the total carrier concentration, particularly
at high temperatures because more carriers transfer to the upper band.33 The
total electron concentration is closer to the measured Hall concentration at
lower temperatures. Therefore, the total donor concentration was estimated
by adding the measured Hall concentration at 77 K to the estimated acceptor
background concentration (1016 cm 3). The observed donor level is
proportional to the arrival rate of the molecular dopant species up to about
n < 1.4 1018 cm 3, beyond which, the carrier concentration saturated. This
dopant saturation level is much higher than 4 1016 cm 3, the highest level
achievable for n-type GaSb using H2S.30 A comparable doping saturation
level (1.2 1018 cm 3) in n-type GaSb was observed previously when using
DETe.34
The measured Hall mobilities as a function of Hall concentrations for
n-type GaSb layers, both at RT (300 K) and 77 K, are shown in Figs. 7.6
and 7.7, respectively. Room-temperature mobilities as high as 2543 cm2/Vs
were measured for an n-type GaSb layer with nH ¼ 3.5 1017 cm 3. The
n-Type Doping in GaSb using Dimethyltellurium (DMTe)… 167
Figure 7.6 Hall mobilities versus carrier concentrations for n-type GaSb layers at 300 K
(data from Ref. 16 used for comparison), where a is the compensation ratio (a ¼ NA =ND ),
and NA and ND are the acceptor and donor density, respectively.
mobilities were reduced to 997 and 956 cm2/V s for nH ¼ 1.0 1018 and
nH ¼ 1.4 1018 cm 3, respectively. Also included in Figs. 7.6 and 7.7 are the
best previous results for MBE-grown GaSb films using an SnTe dopant
source.16 Our measured mobilities compare favorably to those from previous
results, indicating good quality layers. Similar to the previous reports for
n-type GaSb, the mobilities at 77 K are only slightly higher than those at RT,
suggesting that ionized impurity scattering dominates throughout the
temperature range measured.
A comparison with the calculated Hall mH data given by Chin35 is also
shown in Figs. 7.6 and 7.7. In order to account for electrons present in
Figure 7.7 Hall mobilities versus carrier concentrations for n-type GaSb layers at 77 K
(data from Ref. 16 used for comparison).
168 Chapter 7
Figure 7.8 Mobility of Te-doped n-GaSb with different DMTe flows as a function of
temperature GaAs (reprinted from Ref. 29 with permission; © 2011 Carl Hanser Verlag,
Munich).
170 Chapter 7
Figure 7.9 Carrier concentration of Te-doped n-GaSb with different DMTe flow as a
function of temperature GaAs (reprinted from Ref. 29 with permission; © 2011 Carl Hanser
Verlag, Munich).
7.7 Conclusions
It has been observed that the electron concentration n is proportional to the
Te partial pressure in the vapor phase, until n saturates at high Te partial
pressure. Electron concentrations as high as 1.4 1018 cm 3 have been
measured with imperfect morphology, and as high as 1.24 1018 cm 3 with
excellent mirror-like morphology. These appear to be the highest electron
concentrations reported to date for any MOCVD-grown epitaxial n-type
GaSb doped with DMTe and grown at 540 8C with a V/III ratio of 1.4. With
this source, we have observed that a high growth rate significantly improves
the morphology until the DMTe flow rate reaches 5 sccm.
The results of calculation into the transport characteristic of electrons in
this material are reported. Matthiessen’s rule is invoked to determine the total
mobility from the individual mobilities arising from the major scattering
mechanism. At high temperature, both nonpolar and polar phonon and
acoustic deformation potential scattering give a significant contribution to the
total mobility, while the ionized impurity scattering becomes a limiting
mobility scattering process at low temperature. The experimental data for the
n-GaSb samples grown with DMTe flow rates of 2, 3, 5 and 10 sccm shows
good agreement with the theoretical calculation for the density of ionized
impurities, those being 4.5 1016, 1 1017, 1.1 1017, and 1.2 1017 cm 3,
respectively.
n-Type Doping in GaSb using Dimethyltellurium (DMTe)… 171
References
1. R. J. Menna, D. R. Capewell, R. U. Martinelli, P. K. York, and R. E.
Enstron, “3.06 mm InGaAsSb/InPSb diode lasers grown by organometal-
lic vapor-phase epitaxy,” Appl. Phys. Lett. 59, 2127–2129 (1991).
2. H. K. Choi and S. J. Eglash, “High-efficiency high-power GaInAsSb-
AlGaAsSb double-heterostructure lasers emitting at 2.3 mm,” IEEE J.
Quantum Electron. 27(6), 1555–1559 (1991).
3. R. U. Martinelli and T. J. Zamerowski, “InGaAs/InAsPSb diode lasers with
output wavelengths at 2.52 mm,” Appl. Phys. Lett. 56, 125–127 (1990).
4. A. K. Srivastava, J. C. DeWinter, C. Caneau, M. A. Pollack, and J. L.
Zyskind, “High performance GaInAsSb/GaSb p-n photodiodes for the
1.8–2.3 mm wavelength range,” Appl. Phys. Lett. 48, 903–904 (1986).
5. A. H. Ramelan and P. Arifin, “Surface morphology, electrical and optical
properties n-type doped MOCVD grown GaSb using dimethyltellurium,”
Int. J. Mat. Res. 102(11), 1403–1407, (2011).
6. C. Anayama, T. Tanahashi, H. Kuwatsuka, S. Nishiyama, S. Isozumi,
and K. Nakajima, “High-purity GaSb epitaxial layers grown from Sb-rich
solutions,” Appl. Phys. Lett. 56(3), 239 (1990).
7. K. F. Longenbach, S. Xin, and W. I. Wang, “p-type doping of GaSb by
Ge and Sn grown by molecular beam epitaxy,” J. Appl. Phys. 69, 3393–
3395 (1991).
8. A. H. Ramelan and E. M. Goldys, “Elemental analysis of antimony
ternary compound semiconductor grown by MOVPE,” International
Conference on Materials Science and Technology (ICMST) and Asian
Workshop on Solid State Ionic, Serpong 19–23 October 2010.
9. F. Pascal, F. Delannoy, J. Bougnot, L. Gouskov, G. Bougnout, P. Grosse,
and J. Kaoukab, “Growth and characterization of undoped and n-type
(Te) doped MOVPE grown gallium antimonide,” J. Electron. Mater.
19(2), 187–195 (1990).
10. T. H. Chiu, J. A. Ditzenberger, H. S. Luftman, W. T. Tsang, and N. T.
Ha, “Te doping study in molecular beam epitaxial growth of GaSb using
Sb2Te3,” Appl. Phys. Lett. 56(17), 1688–1690 (1990).
11. A. H. Ramelan, E. M. Goldys, and P. Arifin, “Electrical properties of p-n
junction GaSb fabricated from spin coating using Zn diffusion method,”
Conf. on Optoelectronic and Microelectronic Materials and Devices
(COMMAD), Canberra, Australia, 12–15 December, 2010, pp. 183–
184, IEEE (2010).
12. A. H. Ramelan, P. Arifin, H. Harjana, and E. M. Goldys, “Analysis of Al
incorporation in GaSb brown by MOCVD,” Proceeding of Symposium:
The International Conference on Materials for Advanced Technology
172 Chapter 7
Manijeh Razeghi
Center for Quantum Devices, Department of EECS, Northwestern University,
Evanston, Illinois, USA
175
176 Chapter 8
Figure 8.1 GaN thickness as a function of GaN deposition time for three different growth
temperatures: 900, 970, and 1035 8C (reprinted from Ref. 18). (See color plate section.)
indicating that the interface quality of the SLs may be better (mainly due to
reduced thinning of GaN wells during lower-temperature growth). Better-
defined interfaces correlate with higher strain in the well, consistent with the
observed photoluminescence (PL) energy/intensity and ISB absorption energy
behavior previously addressed.
>1200 8C for AlN growth and a lower V/III ratio than Ga. In contrast,
high-structural-quality GaN can be grown under higher pressure (>100
mbar) and high V/III ratios (>3000) at 1000 8C. One way to deposit GaN
with a high V/III ratio while decreasing parasitic pre-reactions during AlN
deposition is to use temporal separation of TMAl and NH3 (or pulsing).
This enhances the surface adatom migration of both the GaN and AlN and
maximizes the growth efficiency.24 Two growth steps are used for each AlN
and GaN layer (Fig. 8.2): Steps (I) and (II) result in AlN deposition with
enhanced Al adatom mobility and may be repeated several times to obtain
thicker AlN layers. Step (III) deposits GaN using conventional bulk
deposition to ensure a high N/Ga ratio. Step (IV) nitridizes the surface and
removes excess Ga to avoid the formation of AlGaN prior to deposition of
the next AlN layer. Thicker GaN layers may be realized by either varying
the length of the bulk-like growth step [step (III)] or by repeating steps (III)
and (IV). By using the pulsed SL growth technique, the V/III ratio for AlN is
decreased, whereas that of GaN is maximized. The pulsed growth results in a
much higher growth rate for AlN and Al adatom mobility than what would
otherwise be achieved with the conventional growth. A typical sketch of the
SL is given in the inset of Fig. 8.3(a).
Figure 8.2 The growth sequence of an AlN/GaN SL. One period consists of two main
deposition phases: (1) AlN deposition, which is realized via pulsing TMAl and NH3 to
enhance Al adatom mobility [steps (I) and (II)], and (2) conventional GaN deposition [step
(III)], which is realized by conventionally supplying TMGa and NH3 simultaneously. Step IV
nitridizes the surface (reprinted from Ref. 25).
AlGaN-based Intersubband Device Technology 181
Figure 8.3 (a) AlN thickness versus number of AlN pulses for TMAl duration of 2 s. Inset
shows a cross-sectional diagram of an SL. (b) GaN thickness versus number of GaN pulses
for TMGa duration of 2 s. Inset shows the (1 1 mm) atomic force microscopy (AFM) SL
surface without TMIn. (c) GaN thickness versus TMGa duration for a single GaN pulse. Inset
shows the (1 1 mm) AFM SL surface with TMIn (reprinted from Ref. 25).
with the ordinate is above abscissa. However, a linear deposition growth rate
with respect to AlN (or GaN) pulses and GaN deposition time allows for
controllable AlN/GaN SL growth.
It should be realized that different ISB devices require tunability of the
individual well (GaN) layer for a constant barrier (AlN) thickness. Keeping
the AlN barrier thickness at 3.1 0.2 nm, we have varied the GaN thickness
from 0.8 to 3.5 0.2 nm by changing the GaN deposition time from 4 to 11 s,
respectively. GaN and AlN SL thicknesses were calculated from periodic
fringes in the XRD spectra.
Figure 8.4 (002) omega/2theta XRD of 50 period of {1.9 nm-thick GaN with 3.1-nm thick
AlN} for different well n-doping levels (CPS ¼ counts per second) (reprinted from Ref. 25).
AlGaN-based Intersubband Device Technology 183
Figure 8.5 Room-temperature (RT) PL of 50 period {1.8 nm-thick GaN / 3.1 nm-thick AlN}
SL with different AlN capping thicknesses (uncapped, 30 nm, and 100 nm). Inset shows (002)
omega/2theta XRD of the SL with the different capping thicknesses (reprinted from Ref. 25).
(See color plate section.)
Figure 8.6 Omega/2theta (002) XRD of SLs with four different well thicknesses: from top
to bottom these are 1.2-nm, 2.6-nm, 3.7-nm, and 5.1-nm GaN. Wells are Si doped to
1 1018 cm3. The barriers for all samples are 2.9-nm-thick Al0.2Ga0.8N. Inset: SL period
(from XRD analysis) as a function of well growth time is used to extract the GaN growth rate
and determine the well thicknesses (reprinted from Ref. 32). (See color plate section.)
Figure 8.7 PL intensity as a function of energy for four different GaN well thicknesses:
1.2 nm, 2.6 nm, 3.7 nm, and 5.1 nm, all with a constant 2.9-nm Al0.2Ga0.8N barrier
thickness. The SL consists of 50 periods in which each well is Si doped to 1 1018 cm3.
Inset: Comparison theory/experiment PL energy as a function of well width (reprinted from
Ref. 32). (See color plate section.)
the 1.2-nm, 2.6-nm, 3.7-nm, and 5.1-nm GaN wells, respectively. The absence
of yellow emission (N vacancies)35 in addition to the low FWHM values imply
an SL of good structural quality.
on well thickness, devices with well thicknesses of 1.2 nm, 2.6 nm, 3.7 nm, and
5.1 nm were fabricated with a constant barrier thickness of 2.9 nm. All
samples were capped with a 30-nm Al0.2Ga0.8N layer. The same structures
were also grown with doped wells (Si: 1 1018 cm 3) to study the effects of
doping.
Figure 8.9(a) shows both p- (solid line) and s-polarization (dashed line)
transmission data for the sample with doped 2.6-nm wells and 2.9-nm
barriers. For both polarizations, we observe 100% absorption above l ¼ 6.1
mm, corresponding to absorption in the sapphire substrate. This absorption
is a significant limiting factor in developing longer-wavelength ISB devices.
An absorption edge is also observed in the p polarization at a wavelength of
4.5 mm, whereas no corresponding absorption is observed for s polarization,
verifying that this feature corresponds to ISB absorption. For the well
thickness of 1.2 nm, no ISB absorption feature is observed; this agrees with
the theory, as only one bound state may be present in this well; thus, no ISB
absorption should occur. For the remaining three well thicknesses,
normalized p- and s-polarization ratios (corresponding to the ISB absor-
bance) are plotted in Fig. 8.9(b). The ISB absorbance peaks at 4.5 mm,
4.9 mm, and 5.3 mm for the 2.6-nm, 3.7-nm, and 5.1-nm wells, respectively.
The increase in the well thickness causes a red shift in the ISB absorption
peak as the first two confined states of the well become closer in energy.
Figure 8.9 (a) Solid line is the p-polarization transmission for 2.6-nm well width and 2.9-nm
barrier width. Wells are Si doped to 1 1018 cm3, and the dashed line corresponds to
s-polarization transmission for the same sample. Inset: ISB absorption wavelength as a
function of well width: experimental results (squares) and simulation (circles) for a fully
strained structure. (b) From left to right, ISB absorbance for 2.6 nm (red), 3.7 nm (blue),
and 5.1 nm (magenta) well widths. Lorentzian fits are shown (dashed lines) for the 3.7- and
5.1-nm well widths (reprinted from Ref. 32). (See color plate section.)
188 Chapter 8
For the 2.6-nm well, we observe a broader absorption peak compared to that
in two thicker wells (3.7 and 5.1 nm). For the 2.6-nm well, some oscillations
appear around 6 mm in the absorbance. These are attributed to the
normalization with s-polarized spectra close to the sapphire cutoff. For each
of the two thicker wells, we observe one prominent absorption feature that
can be fitted with Lorentzian functions (dashed lines), a characteristic of ISB
transitions. Lorentzian functions do not perfectly fit the experimental curves
around 6 mm due to the normalization. Furthermore, the decrease of ISB
absorption intensity with the increase in well width can be explained by
reduced quantum confinement for the second subband compared to the
fundamental state. With increasing well thickness, we have increased the
wavelengths as long as 5.3 mm, with further wavelength increase limited by
the sapphire substrate reabsorption.
Simulations of ISB absorption wavelength as a function of well
thickness were performed and compared with the experimental data of
Fig. 8.9(a) (inset). The best agreement between the simulated and
experimental data is obtained for the thinnest well (2.6 nm), but the
discrepancy occurred with increased well thickness. The discrepancy may be
due to uncertainties in internal electric field and conduction-band offset, as
the two subbands are located in the triangular part of the well for thicker
wells (error 5%). Further studies of these two parameters are required to
minimize this discrepancy.
In summary, we have studied Al0.2Ga0.8N/GaN SLs grown by MOCVD
with various well thicknesses via XRD, PL, and polarization-dependant ISB
absorption measurements. A theoretical model was developed, taking into
account strain in binary/ternary SLs, and was correlated with both
experimental interband emission and ISB absorption results. We have
demonstrated tunability of ISB absorption from 4.5 to 5.3 mm by tuning
the well thickness. This should lead to application of GaN/AlGaN SLs in
longer-wavelength ISB devices.
Figure 8.10 GaN-, AlN-, and SiC-substrate transmission in the infrared terahertz regime.
(See color plate section.)
8.3.4 Conclusion
Large conduction-band offset and fast transition speeds in III-nitrides are
promising constituents for ISB devices. The first demonstration of ISB
transitions in nitrides was shown in the MBE-grown layers operating at a
wavelength of 1–2 mm. Early attempts to fabricate ISBs using MOCVD-
grown material failed. We have overcome earlier shortcomings of MOCVD-
grown III-nitride devices by developing a novel pulsed deposition technique.
In addition, ISB design engineering with AlGaN barriers has resulted in
devices operating successfully at 1.0–5.3 mm (near- to mid-infrared).
agrees well with recent theoretical studies that predict that dislocation
densities of <106 cm 2 are required for a reliable RTD.52 Nevertheless,
improved reliability of NDR dictated that further material engineering in the
elimination of piezoelectric fields to minimize band bending at DB
heterointerfaces53 was required. In addition, smaller device mesas were used
to minimize sidewall effects on NDR and improve NDR performance.
Higher structural quality of nitride growth on FS GaN was correlated
with improved response of ultraviolet detectors.38,54 It should be noted,
however, that in addition to the structural quality, the effect of the polar
(c-plane) and nonpolar (m-plane) nature of FS substrates is equally important,
as the piezoelectric field in the FS substrates directly impacts the optical
properties of AlGaN/GaN SLs.55 This field degrades the device response,
which becomes bias dependent. We have therefore studied GaN-based RTD
structures on both polar and nonpolar FS GaN substrates.
In summary, successful operation of RTDs (fabricated on FS GaN) under
both reverse and forward biases at RT (and below) has been demonstrated for
the first time.37,52 An alternative nanostructure-based approach56 has also
been successfully applied to reduce dislocation density in the grown nitride
films and has yielded operational DB RTDs.
Figure 8.11 Schematic, crystallographic directions, and energy band diagram of a polar
RTD active layer. (a) Relative conduction-band discontinuities DEC of the RTD active layer
and the distribution of electrons. (b) Schematic of the RTD active layer with layer
thicknesses shown, along with the crystallographic directions (reprinted from Ref. 57). (See
color plate section.)
192 Chapter 8
over to the GaN, resulting in a discrete electronic state confined in GaN below
the AlGaN conduction band. Basically, the GaN acts as a quantum well, and
the AlGaN layers act as barriers to the electron transport [Fig. 8.11(b)]. By
reducing the thickness of the AlGaN barrier, tunneling of electrons can be
achieved for the RTD operation.
Simulated band structure for double barriers is obtained by solving
Poisson’s and Schrodinger’s equations. Polarization fields are also calculated
by the simulations. Carrier concentrations in the quantum well are calculated
using the Boltzmann approximation, and for degenerate regions modified
Sommerfeld approximation is employed. Simulation results for active layers
in polar and nonpolar RTDs are shown in Figs. 8.11 and 8.12, respectively.
Conduction and valance bands, Fermi level and electron concentrations are
schematically represented in the figures.
GaN and (2) high-quality lateral epitaxial overgrowth (LEO) GaN templates
(grown on sapphire).58 First, 750-nm-thick n-GaN was grown for the bottom
contact, followed by the active layer of GaN. For polar RTDs, a 1.5-nm
Al0.20Ga0.80N barrier layer was grown to prevent lattice relaxation via
dislocations. The RTD active layer stack consists of 2.0-nm i-GaN, 1.5-nm
i-Al0.20Ga0.80N, 1.25-nm i-GaN, 1.5-nm i-Al0.20Ga0.80N, and 2.0-nm i-GaN.
This structure enables a conduction-band offset of 0.42 eV in the AlGaN
barrier, resulting in a single and discrete electronic level (E) at 0.32 eV in the
GaN well. To minimize impurity scattering due to dopant interdiffusion into
the barrier regions, thin (a couple nanometers) undoped or low-n-doped GaN
spacer layers are grown on either side of the barriers. Finally, a top contact
layer of 250-nm-thick nþ-GaN was grown to finish the RTD structure. Figure
8.11 shows the polar RTD energy diagram near the active GaN layer.
8.4.3.2 Nonpolar devices
The growth sequence of nonpolar RTDs is similar to that used for polar
devices. It starts with baking of m-GaN substrates followed by 3-mm i-GaN
and 400-nm n-GaN for the bottom contact. The active layer stack consists
of 2.6-nm i-GaN, 1.6-nm i-Al0.10Ga0.90N, 1.6-nm i-GaN, 1.6-nm
i-Al0.10Ga0.90N, and 2.6-nm i-GaN. This structure enables a conduction-
band offset of 0.16 eV in the AlGaN barrier, resulting in a discrete electronic
level at 0.10 eV in the nonpolar GaN well. Similar to what is described earlier,
impurity scattering due to dopant interdiffusion in the barriers is minimized
by growing a thin (a couple nanometers) undoped or low-n-doped GaN spacer
layers on either side of the barriers. Finally, a top contact layer of 250-nm
nþ-GaN is grown to finish the nonpolar RTD structure. Figure 8.12 shows
the nonpolar RTD energy diagram near the active GaN layer.
Figure 8.13 Typical side-view schematic and top-view images of an RTD. (a) Side-view
schematic of a fabricated RTD device. (b) Top-view optical and (c) scanning electron
micrograph of a fabricated RTD device. Inset of (b) shows the scanning electron microscopy
(SEM) bird’s eye view of the mesa. Points A through G correspond to side (a) and top (b)
and (c) views of the indicated RTD locations (reprinted from Ref. 57).
Figure 8.14 Electrical characterization of polar RTDs. (a) I-V curve of an RTD at RT for
different scan directions. (b) I-V curve of an RTD showing NDR for both polarities (reprinted
from Ref. 57).
Figure 8.15 Electrical characterization of nonpolar RTDs. (a) I-V curve of the RTD at RT.
The 1st, 30th, and 50th I-V curve is plotted to demonstrate NDR reliability and reproducibility.
(b) I-V curve of the RTD at RT is plotted with low-temperature (77 K) I-V curve (reprinted
from Ref. 57).
196 Chapter 8
(Figs. 8.11 and 8.12) is pulled down in energy with respect to the more
negative electrode. Electrons flowing from outside the well will be largely
deflected by the barriers, except those residing within a very narrow
range of discrete energy levels of the well. The well therefore acts as an
electron energy filter. Thus, the tunneling current through the well depends
on the density of occupied states in the well and alignment of the discrete
states with Fermi levels of the contact layers. In resonance, the electron
states in the electrode align with the discrete energy states of the well, and
the peak current IP corresponds to the applied peak voltage VP.
With further increase in bias, the emitter electron states fall below the
conduction-band edge, minimizing the tunnel current. The current and
voltage in this regime are defined as the valley current IV and valley
voltage VV.
8.5 Summary
III-V-nitride-based ISB devices have successfully been demonstrated recently
because of pronounced improvements in material and device technologies.
For example, GaN-based RTDs based on quantum transport among ISB
energy levels with negative resistance of 67 V has been achieved, as shown in
Fig. 8.16(b). Our results suggest that with proper device design and energy gap
engineering, the (sub)bands of AlGaInN can be engineered to operate over a
wide optical range from ultraviolet to terahertz, as shown in Fig. 8.16(a).
An RTD is an electronic device that displays NDR due to a unique
quantum-interference phenomenon. Unlike other negative resistance
devices such as (Esaki) tunnel and transferred-electron devices, RTDs
have higher performance and can operate at higher temperatures. In
particular, AlGaInN alloys have a wide bandgap (6.2 eV), high carrier
mobility (1000–1500 cm2/Vs), and thermal stability (800 8C), making
them ideal for high-power (>20 mW), high-frequency (>500 GHz) RTDs.
Moreover, a larger conduction-band discontinuity in AlGaN (1.8 eV)
results in higher NDR than in other III-Vs (such as GaAs) at RT. It
should be noted that earlier efforts on GaN-based RTD structures have
failed to achieve a reliable and reproducible NDR. Recently, we have
demonstrated for the first time that minimization of dislocation density
198 Chapter 8
Figure 8.16 (a) Demonstration of ISB transitions in AlN/GaN and AlGaN/GaN SLs from
near-infrared to mid-infrared, respectively, and (b) RT (300 K) and low-temperature (77 K)
operation of a GaN-based resonant tunneling diode. The inset shows the reliability and
reproducibility of NDR at RT (reprinted from Ref. 39).
8.6 Conclusions
In conclusion, we demonstrated a reliable and reproducible NDR (R of
67 V, CPVR of 1.08, and PMAX of 0.52 mW) in MOCVD-grown III-nitride
RTDs for the first time, as shown in Fig. 8.16(b). These RTDs operate both in
reverse- and forward-bias conditions with improved behavior at low
temperatures. We further demonstrated that very low dislocation density
(<105 cm 2) and minimization of the polarization charges at DB
heterointerfaces are a prerequisite for reliable and reproducible NDR in III-
nitride RTDs. These findings are expected to encourage further research in
material and device engineering on GaN-based quantum devices, including
RTD-based electronic oscillators and tunneling-based quantum cascade
photonic devices operating at terahertz.
AlGaN-based Intersubband Device Technology 199
References
1. E. R. Mueller, “Terahertz radiation: Applications and sources,” Physics
Today 9, 27–29 (2003).
2. S. M. Kim, F. Hatami, J. S. Harris, A. W. Kurian, J. Ford, D. King,
G. Scalari, M. Giovannini, N. Hoyler, J. Faist, and G. Harris, “Biomedical
terahertz imaging with a quantum cascade laser,” Appl. Phys. Lett. 88,
153903 (2006).
3. R. M. Woodward, V. P. Wallace, R. J. Pye, B. E. Cole, D. D. Arnone,
E. H. Linfield, and M. Pepper, “Terahertz pulse imaging of ex vivo basal
cell carcinoma,” J. Invest. Dermatol. 120, 72–78 (2003).
4. P. C. Upadhya, K. L. Nguyen, Y. C. Shen, J. Obradovic, K. Fukushige,
R. Griffiths, L. F. Gladden, A. G. Davies, and E. H. Linfield,
“Characterisation of crystalline phase transformations in theophylline by
time domain terahertz spectroscopy,” Spectrosc. Lett. 39, 215–224 (2006).
5. M. Hangyo, M. Tani, and T. Nagashima, “Terahertz time-domain
spectroscopy of solids: A review,” Int. J. Infrared Millimeter Waves 26,
1661–1689 (2005).
6. L. L. Nguyen, T. Friscic, G. M. Day, L. F. Gladden, and W. Jones,
“Terahertz time-domain spectroscopy and the quantitative monitoring of
mechanochemical cocrystal formation,” Nature Mater. 6, 206–209 (2007).
7. Y. C. Shen, T. Lo, P. F. Taday, B. E. Cole, W. R. Tribe, and
M. C. Kemp, “Detection and identification of explosives using terahertz
pulsed spectroscopic imaging,” Appl. Phys. Lett. 86, 241116 (2005).
8. H. Li, J. C. Cao, Y. J. Han, X. G. Guo, Z. Y. Tan, J. T. Lu, H. Luo,
S. R. Laframboise, and H. C. Liu, “A study of terahertz quantum cascade
lasers: Experiment versus simulation,” J. Appl. Phys. 104, 043101 (2008).
9. B. S. Williams, S. Kumar, H. Callebaut, Q. Hu, and J. L. Reno,
“Terahertz quantum cascade laser operating up to 137 K,” Appl. Phys.
Lett. 83, 5142 (2003).
10. O. P. Marshall, V. Apostolopoulos, J. R. Freeman, R. Rungsawang,
H. E. Beere, and D. A. Ritchie, “Surface-emitting photonic crystal
terahertz quantum cascade lasers,” Appl. Phys. Lett. 93, 171112 (2008).
11. M. A. Belkin, F. Capasso, A. Belyani, D. L. Sivco, A. Y. Cho,
D. C. Oakley, C. J. Vineis, and G. W. Turner, “Terahertz quantum-
cascade-laser source based on intracavity difference-frequency gener-
ation,” Nature Photonics 1, 288–292 (2007).
12. L. Nevou, M. Tchernycheva, F. H. Julien, F. Guillot, and E. Monroy,
“Short wavelength (l ¼ 2.13 mm) intersubband luminescence from GaN/
AlN quantum wells at room temperature,” Appl. Phys. Lett. 90, 121106
(2007).
200 Chapter 8
updated version of Volumes 1 and 2. She holds 50 U.S. patents and has given
more than 1000 invited and plenary talks. Her current research interest is in
nanoscale optoelectronic quantum devices.
Dr. Razeghi is a Fellow of MRS, IOP, IEEE, APS, SPIE, OSA, a Fellow
and Life Member of the Society of Women Engineers (SWE), a Fellow of
the International Engineering Consortium (IEC), and a member of the
Electrochemical Society, ACS, AAAS, and the French Academy of Sciences
and Technology. She received the IBM Europe Science and Technology Prize
in 1987, the Achievement Award from the SWE in 1995, the R. F. Bunshah
Award in 2004, and many best-paper awards.
Part III: Lasers
Chapter 9
Advances in High-Power
Quantum Cascade Lasers
and Applications
Arkadiy Lyakh, Richard Maulini, Alexei Tsekoun, and Boris Tadjikov
Pranalytica, Inc., Santa Monica, California, USA
C. Kumar N. Patel
Pranalytica, Inc., Santa Monica, California, USA and Department of Physics &
Astronomy, University of California, Los Angeles, California
9.1 Introduction
9.2 MWIR Laser Design
9.3 Tapered Waveguide Geometry
9.4 Silicon Carbide Submounts
9.5 MWIR QCL Experimental Data
9.6 LWIR QCL Design
9.7 LWIR QCL Experimental Data
9.8 Conclusion
References
9.1 Introduction
Quantum cascade lasers (QCLs) are important infrared light sources with
numerous applications in defense and civilian fields. Low optical absorption
in the atmospheric windows spanning from 3.5 to 4.8 mm and 8 to 12 mm has
given rise to a number of applications. While QCL performance in the 5- to
7-mm spectral region has improved significantly,1 culminating in continuous-
wave (CW) room-temperature wallplug efficiency (WPE) exceeding 20% and
optical power of 5 W at 5 mm,2 progress in laser performance in the
technologically important 3.5- to 4.8-mm and 8- to 12-mm spectral regions has
significantly slowed down.
209
210 Chapter 9
Injection efficiency for the upper laser level is usually considered to be the
main factor limiting laser performance of the short-wavelength QCLs.
Thermal carrier leakage from the upper laser level 4 to continuum states
located above the barriers reduces laser population inversion at a given
pumping current density and therefore increases laser threshold current
density (Fig. 9.1). Another carrier leakage path is due to carrier excitation to
level 5 and subsequent scattering to states other than level 4.3 Carrier escape
to continuum states C and level 5 can be suppressed by increasing energy
spacings EC4 and E54. A straightforward approach to increase EC4 is to
increase potential barrier heights by employing higher-strain InxGa1 xAs/
AlyIn1 yAs composition. In addition to larger EC4, employment of higher
barriers also leads to larger E54 for the same laser transition energy.
Therefore, use of higher strain composition is a very attractive path for
suppressing carrier leakage from the upper laser level.
Another advantage to using a high-strain composition is that the
position of indirect X and L valleys relative to direct G-valley minimum
increases as strain increases.4 Carrier leakage from the upper laser level
through indirect states progressively increases as laser emission wavelength
reduces below 5 mm. Therefore, short-wavelength QCLs with high- strain
composition should benefit from suppressed carrier leakage through indirect
states. Increase in barriers height, however, also increases linewidth G43 of
the laser transition.
Linewidth is proportional5 to
X
g43 EC 2 D2 L2 ½C4 2 ðzk ÞC3 2 ðzk Þ2 , ð9:1Þ
where EC is conduction-band offset (barrier height), D is interface roughness,
L is roughness correlation length, C is electron wavefunction, zk is interface
location, and summation is carried out over all interfaces k where either the
upper or lower laser level wavefunction is nonzero. Laser threshold current
density is inversely proportional to linewidth. Therefore, an increase in
linewidth for a composition with higher barriers can offset laser performance
improvements due to suppressed carrier leakage from the upper laser level.
Typical thicknesses for the first and the second active-region barriers are
in the range of two to five monolayers (6–15 Å). Depending on epigrowth
quality, a barrier thickness of two to five monolayers is comparable to, or
even less than, total roughness at both interfaces of a barrier 2D. In other
words, the whole barrier thickness d may be irregularly graded. In this case,
Eqs. (9.1) may not be directly applicable for calculating laser transition
linewidth since it is derived assuming that d > 2D. Using electroluminescence
(EL) data, we show in this work that very good epiquality can be achieved for
high-strain material both in MWIR and LWIR spectral regions, leading to
improved laser characteristics in both spectral regions.
application, on the other hand, the taper was part of the laser cavity, as in
our application. However, the motivation for tapering the waveguide in this
application was also very different from ours. In Refs. 11 and 12, the main
purpose of the taper was to increase the device area in order to increase its
peak power in low-duty-cycle pulsed operation while maintaining a good
beam quality. In such devices, the length of the tapered section was typically
more than half of the entire length of the device, and its area typically
constituted most of the area of the entire device. Since self-heating is not
significant in low-duty-cycle pulsed operation, the output facet width was
typically 50–200 mm. While adequate for these operating conditions, such
large waveguide widths are not compatible with high-duty-cycle and/or CW
operation because of the high active-region self-heating resulting from the
significant waste heat generated in QCLs. In our work, the purpose of the
taper was to reduce optical intensity at the output facet, while keeping self-
heating to a minimum for high-performance CW operation. Therefore, the
taper width and length are kept to a minimum. As a result, the taper length is
only a small fraction of the length of the entire device, and the taper area is a
small fraction of the entire device area.
Figure 9.4 Beam image of the hermetically packaged 10-mm-long QCL with a 7.5-mm-
wide central section tapered up to 20 mm at laser facets mounted on AlN/SiC submounts.
The laser current was set to 0.75 A (see Fig. 9.3 for LIV). The laser temperature was set to
283 K (reprinted from Ref. 8). (See color plate section.)
Figure 9.5 Reliability data with periodic hourly laser turns on/off that subject the QCL to a
large thermal stress (reprinted from Ref. 8).
Advances in High-Power Quantum Cascade Lasers and Applications 217
with a TEC assembly is crucial for practical applications with limited cooling
capability.
Figure 9.6 (a) EL spectra of a round mesa at threshold and roll-over voltages. (b)
Dependence of EL linewidth on voltage (reprinted from Ref 15). (See color plate section.)
Advances in High-Power Quantum Cascade Lasers and Applications 219
Figure 9.7 Comparison between pulsed and CW optical power versus current and voltage
versus current characteristics measured at 293 K for an uncoated 3-mm by 10-mm laser-
mounted epi-side down on an AlN/SiC composite submount. Inset shows pulsed laser
spectrum taken at maximum current (reprinted from Ref. 15). (See color plate section.)
of two.16 Figure 9.7 inset shows that the pulsed laser spectrum taken at
maximum current was centered close to 9.2 mm.
An important aspect of the LIV curves shown in Fig. 9.7 is their
behavior at bias values above the LI curve roll over. The pulsed LI curve
shows a very abrupt decrease in optical power, while the pulsed IV curve
shows signs of negative differential resistance (NDR). NDR was more
pronounced for some other chips processed from the same wafer. This
behavior in the vicinity of the roll-over condition demonstrates that carrier
tunneling from the injector to the active-region states other than the upper
laser level is suppressed. In other words, these results indicate improved
injection efficiency for the upper laser level.
9.8 Conclusion
In conclusion, we have presented experimental data on tapered 4.7-mm QCLs
based on the nonresonant extraction approach. EL FWHM of only 22 meV
showed that excellent epiquality was achieved despite using highly strained
active-region QWs and barriers. WPE of 16.3% and optical power of over
4.5 W have been demonstrated for a 10-mm by 7.5-mm high-reflection-coated
laser mounted on an AlN/SiC submount. Reliable CW operation was
demonstrated for a laser system for over 2,000 h at a 3-W level with overall
system efficiency (including power input to the TEC) exceeding 10%. We have
also presented experimental data on high-strain 9-mm QCLs. Record-high
220 Chapter 9
pulsed and CW WPE of 16 and 10% and optical power of 4.4 and 2.0 W,
respectively, were demonstrated at 293 K for an uncoated 3-mm by 10-mm
laser mounted epi-side down on AlN/SiC composite submounts. Laser
characteristics indicate improved injection efficiency for the LWIR structure.
References
1. R. Maulini, A. Lyakh, A. Tsekoun, and C. Kumar N. Patel, “l7.1 mm
quantum cascade lasers with 19% wall-plug efficiency at room
temperature,” Opt. Express 18, 17203 (2011).
2. Y. Bai, N. Bandyopadhayay, S. Tsao, S. Slivken, and M. Razeghi,
“Room temperature quantum cascade lasers with 27% wall plug
efficiency,” Appl. Phys. Lett. 98, 181102 (2011).
3. D. Botez, S. Kumar, J. C. Shin, L. J. Mawst, I. Vurgaftman, and J. R.
Meyer, “Temperature dependence of key electro-optical characteristics for
midinfrared emitting quantum cascade lasers,” Appl. Phys. Lett. 97,
071101 (2010).
4. W. Masselink, M. Semtsiv, S. Dressler, M. Ziegler, and M. Wienold,
“Physics, growth, and performance of (In, Ga)As-AlP/InP quantum-
cascade lasers emitting at <4 mm,” Phys. Stat. Sol. B 244, 2906 (2007) .
5. A. Wittmann, Y. Bonetti, J. Faist, E. Gini, and M. Giovannini,
“Intersubband linewidth in quantum cascade laser design,” Appl. Phys.
Lett. 93, 141103 (2008).
6. A. Lyakh, R. Maulini, A. Tsekoun, R. Go, C. Pflügl, L. Diehl, Q. J. Wang,
F. Capasso, and C. K. N. Patel, “3 W continuous-wave room temperature
single-facet emission from quantum cascade lasers based on nonresonant
extraction design approach,” Appl. Phys. Lett. 95, 141113 (2009).
7. C. K. N. Patel, A. Tsekoun, R. Maulini, A. Lyakh, C. Pflugl, L. Diehl,
Q. Wang, and F. Capasso, “Quantum cascade laser,” U.S. Patent
#8,014,430 (September 6, 2011).
8. A. Lyakh, R. Maulini, A. Tsekoun, R. Go, and C. K. N. Patel, “Tapered
4.7 mm quantum cascade lasers with highly strained active region
composition delivering over 4.5 watts of continuous wave optical power,”
Opt. Express 20(4), 4382–4388 (2012).
9. M. Troccoli, C. Gmachl, F. Capasso, D. L. Sivco, and A. Y. Cho, “Mid-
infrared (l 7.4 mm) quantum cascade laser amplifier for high power single-
mode emission and improved beam quality,” Appl. Phys. Lett. 80, 4103 (2002).
10. S. Menzel, L. Diehl, C. Pflügl, A. Goyal, C. Wang, A. Sanchez,
G. Turner, and F. Capasso, “Quantum cascade laser master-oscillator
power-amplifier with 1.5 W ouput power at 300 K,” Opt. Express 19,
16229 (2011).
Advances in High-Power Quantum Cascade Lasers and Applications 221
10.1 Introduction
10.2 Manufacturing of High Performance QC Lasers
10.2.1 Design
10.2.2 Growth
10.2.3 Fabrication
10.3 Results
10.3.1 High power multimode devices
10.3.2 Low power consumption distributed feedback laser devices
10.3.3 Power scaling: arrays
10.4 Conclusions
Acknowledgments
References
10.1. Introduction
Quantum cascade lasers (QCLs) are semiconductor lasers based on
intersubband transitions between energy states created by quantum confine-
ment in the conduction band of semiconductor multilayers.1 The laser core is
composed of hundreds of layers of quantum wells and barrier materials
(typically InGaAs/InAlAs) that together define most of the laser properties
such as wavelength, bias voltage, and output power, to a degree that is
unprecedented among semiconductor devices and that led the inventors of this
type of lasers to talk about “materials by design” and “bandgap
engineering.”2 The great advantage of QCLs over other laser types in the
225
226 Chapter 10
came at first by modifying the wavefunction of the upper laser level in order to
maximize the electron probability distribution away from the conduction-
band offset (so called “diagonal design”), thus, minimizing the probability of
escape (“leakage”) to higher energy levels or to the conduction-band
continuum. Nevertheless, at short wavelengths in the 3–5 mm range, the
upper laser level would still be too close to the conduction-band offset of the
lattice-matched or moderately strained GaInAs/AlInAs material to reason-
ably avoid this leakage from significantly affecting the laser operation.
Highly strained material can offer a larger band offset, and improve-
ments in the growth technologies, first in molecular beam epitaxy (MBE)7
and subsequently also in metalorganic chemical vapor deposition
(MOCVD)8 performed at AdTech Optics, Inc. (ATO), have enabled the
design and growth of materials with very high strain and large band offsets
capable of preventing leakage currents during high-temperature operation at
short wavelengths. The combination of these design features, among others,
has paved the way to high-power, highly efficient devices. State-of-the-art
MBE-grown devices have demonstrated powers as high as 5.1 W from a
single emitter, and efficiencies of more than 20% at room temperature.4
However, RT operation requires buried-heterostructure (BH) fabrication
techniques that can only be performed by MOCVD. These hybrid MBE/
MOCVD QCLs have been, so far, the most successful research-grade devices
in terms of performance, but the use of different technologies at different
stages of the laser manufacturing is significantly more expensive for
industrial-grade devices and would not enable volume production of QCLs
at reasonable cost. In this paper, we present the results on MOCVD-grown
highly strained QCL devices that are capable of emitting close to 3 W of
power in continuous wave (CW) at RT, with electro-optical conversion
efficiencies of 11.5%. These performances are obtained by using the strain-
compensated laser design illustrated in Fig. 10.1, where both the uniform
strain and the selective strain design approaches have been explored. The
material composition is Ga0.28In0.72As/Al0.64In0.36As. Figure 10.1 illustrates
one period of the active material, which is repeated 40 times in current laser
designs for high-power emission. Both uniformly strained and selectively
strained designs are used in the lasers manufactured at ATO.
10.2.2 Growth
QCL material growth, both of the active region and the waveguide, was
performed by low-pressure MOCVD. The growth system is designed in a
showerhead gas injection arrangement, in close proximity with the wafers,
allowing for a very short transient gas switching time between layers. In
addition, the MOCVD is equipped with ultrasonic gas concentration
monitoring systems that allow automatic modification of the flow rate, and
consequently the growth rate and composition, to compensate for fluctuations
228 Chapter 10
Figure 10.1 Examples of (a) selectively strained and (b) uniformly strained designs for high-
power, high-temperature emission at MWIR wavelengths. The wavy horizontal lines represent
the electronic wave function probabilities within the conduction band of the GaInAs/AlInAs
multilayer structure, represented by the step-like black curve. (See color plate section.)
process. For every wafer, we measure the x-ray scattering pattern, polaron C-V,
Hall conductivity, near-IR photoluminescence map, among others. For strain-
compensated materials, a small amount of residual strain in the complete
structure needs to be achieved and can be verified by x-ray measurements.
Polaron C-V testing is used to monitor the doping in the structure, while high-
resolution scanning electron microscope (SEM) and Nomarski microscope
techniques are used to ensure the grown wafer surface quality.
10.2.3 Fabrication
All QCLs were fabricated in BH waveguides. The laser stripe width varied from
4 mm for low-power-consumption devices to about 12 mm for high-power
devices, while their cavity length was between 2 mm for distributed-feedback
(DFB) lasers and up to 5 mm for high-power devices. After completion of the
MOCVD growth, ridge waveguides were defined by photolithography and
etched by a combination of dry and wet etching, and insulating InP:Fe was
regrown on the sides of the laser waveguide. The top and bottom metal contact
depositions by electron beam evaporation completed the device fabrication.
Figure 10.2 shows the finished device. In the case of DFB lasers, the MOCVD
growth was interrupted just after completion of the active core, including an
InP buffer and an InGaAs sacrificial grating layer. The grating pattern for
Figure 10.2 (a) SEM image of the front edge of a finished BH device and (b) close-up of
the same laser indicating the different regions of the device. The downward arrow indicates
the direction of electron injection through the structure.
230 Chapter 10
10.3 Results
10.3.1 High-power multimode devices
MWIR devices (designed according to the uniformly strained and selectively
strained approaches illustrated in Section 10.2.1) have been processed in BH-
type devices for optimized CW operation at room temperature (RT), leading
to high-power emission at wavelengths in the range of l ¼ 4–5 mm. In an
effort to minimize current leakage and optimize temperature sensitivity of the
device characteristics (represented by the characteristic temperatures T0 and
T1), the selectively strained devices have been proven to yield the highest-
power outputs so far. Our devices are showing performances comparable to
the best-in-class and are fabricated using industrial-process manufacturing
rather than research-grade equipment and procedures.
Figure 10.3 illustrates the performances of a selectively strained device at
RT. The total emitted power from a 5-mm-long device installed on a CS-type
mount, with HR coating on the back facet and uncoated front facet, rises to
operated with 200-ns pulses at 2% duty cycle, while the average output
power was measured with a thermopile detector and divided by the duty
cycle in order to estimate the peak power during the pulse.
To the best of our knowledge, the MWIR and LWIR results presented
here are the best industrial-grade result obtained with MOCVD-grown QCLs
in terms of power and efficiency. MBE-grown devices have demonstrated
higher efficiencies in the MWIR range, but MBE technology does not allow
for BH fabrication, which still needs to be performed by MOCVD and
constitutes the most critical step of the fabrication procedure. For industrial
purposes, and in order to have the lasers grown and fabricated at the same
facility, use of MOCVD for both the growth and the fabrication portion of the
laser manufacturing will greatly simplify the production and make it more
affordable in volume quantities.
When installed in high heat load (HHL) packages, Adtech’s QCL devices
produce up to 1 W, at wavelengths in the strategically important ranges of 4.1,
4.6, and 9 mm. The results are shown in Fig. 10.5. The HHL packages include
a thermoelectric cooler (TEC), temperature sensor, and collimating optics so
that the device can be simply connected to a laser driver and TEC controller
for RT operation.
Figure 10.5 High-power laser characteristics after installation in HHL packages for RT
operation. All devices at various wavelengths across the mid-IR show continuous power
output close to 1 W. (a) The HHL package. The emission wavelengths shown in (b), (c), and
(d) are 4.6, 4.1, and 9 mm, respectively. (See color plate section.)
Figure 10.6 (a) Optical power versus current and (b) spectra of a low-power-consumption
DFB laser at l ¼ 4.59 mm. The total power required to operate the laser progresses from
1.2 W at threshold to about 2.5 W at maximum power. (See color plate section.)
solutions, such as high-Q cavities, would affect the total power output and
limit the potential for application to those that require only a few milliwatts of
laser power, such as certain gas detection systems or some close-range
detection of high concentrations for substances where selectivity is important
but sensitivity and access to the point of detection are not an issue.
In this chapter, we present the results on low power consumption of
4.59-mm DFB QC lasers. Figure 10.6 shows the optical power-versus-
current characteristics of a 2-mm laser processed in BH waveguides of
about 5-mm width. The threshold current is only 0.1 A, leading to a power
consumption at a threshold of about 1.2 W. The maximum power from
these devices is more than 50 mW, which is still quite large and suitable for
most sensing applications, including those based on photoacoustic
spectroscopy or similar techniques where optical output power makes a
difference in the sensitivity that can be achieved. Another upside of the low-
consumption devices is that they can be installed into smaller packages
needing lower heat dissipation. Figure 10.7 illustrates the optical power
output from a l ¼ 4.55-mm DFB device operated in CW mode at RT after
installation in a TO-3 package. The total power output is 60 mW, and the
total power consumption at threshold power is 0.96 W, while it rises to
more than 2 W at maximum power.
processing, surgery, defense, among others, may benefit from even higher levels
of power. In order to scale up the power of a single emitter, several approaches
have been proposed, either related to the device design and operation, such as
increasing the number of active stages and working at low temperatures, or
using alternative devices schemes such as optical amplifiers, unstable
resonators, broad area devices, and arrays. All solutions have advantages and
disadvantages. For example, amplifiers, first demonstrated in 200111 and
recently improved,12 have the advantage of giving high-power output in a good
optical mode with narrow far-field angular dispersion. On the other hand,
master oscillator power amplifier (MOPA) devices are more easily damaged due
to the uneven current distribution along the device cavity, and their fabrication
is more challenging than regular BH waveguides. Also, given the small gain of
intersubband transitions, the amplification effect brings about only a modest
improvement in terms of output power with respect to an optimized high-power
device of comparable length. Still, this could be a solution for high-power single-
mode lasers where the lateral mode quality is critical.
One potential solution we explore in this chapter is fabricating arrays of
QCL emitters that can be subsequently beam combined in order to achieve a
single higher-power beam. The beam-combining step goes beyond the scope
of this publication and will not be discussed here. Figure 10.8 shows the
separate operation of several lasers in the same bar as cleaved from the
processed wafer. The power-versus-current characteristics of the various
devices overlap with each other, thus demonstrating the optimized uniformity
achieved in the growth and processing that is necessary in order to realize a
236 Chapter 10
working laser bar with no dark emitters and uniform operating conditions
across the different devices. Figure 10.8(b) shows the simultaneous biasing of
all emitters, as compared to the single emitter, and demonstrates that in pulsed
operation the total power from four emitters in the same bar mounted on a
CS-type mount is four times the power from a single emitter. The total power
was only limited by our pulse-generator capability to achieve high currents.
We estimate that without this limitation the four-emitter array could achieve a
peak power of 10 W.
In order to test continuous operation, the multi-emitter bar was operated at
various duty cycles and different pulse widths. The results are shown in Fig. 10.9.
The array was tested with up to 10-ms pulse width and showed some power increase
up to 50% duty cycle. However, the average intrapulse power stayed constant only
up to about 10% duty cycle and started decreasing for duty cycles above 10% for
pulse widths of 0.5–10 ms, eventually shutting off the laser for duty cycles above
50%. Improving the heat sinking and device multi-emitter geometry will allow
achievement of full CW operation at RT for multi-emitter configurations.
10.4 Conclusions
In conclusion, we have shown high-power operation in both single-emitter and
multiple-emitter configurations for MOCVD-grown and -processed devices and
have demonstrated that MOCVD-grown QCLs can be used in low-power
regimes as well, when limited to smaller cavity sizes and optimized packages.
All of the results presented here were obtained from industrial-grade QCL
High-Performance Quantum Cascade Lasers for Industrial Applications 237
Acknowledgments
We acknowledge support from the following funding agencies: the Office for
Naval Research under contract no. N00014-11-C-0252, and the Missile
Defence Agency under contract no. HQ0147-11-C-7701. We would like to
acknowledge the following people for their help and partial contributions:
Hien Quach, Rommel Ceballos, Armando Marquez, and Ulisses Gamboa.
238 Chapter 10
References
1. J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, and
A. Y. Cho, “Quantum cascade laser,” Science 264, 553–556 (1994).
2. F. Capasso, C. Gmachl, D. L. Sivco, and A. Y. Cho, “Quantum cascade
lasers,” Physics Today 55, 34–38 (2002).
3. A. Lyakh, R. Maulini, A. Tsekoun, R. Go, C. Pflügl, L. Diehl, Q. J. Wang,
F. Capasso, and C. K. N. Patel, “3W continuous-wave room temperature
single-facet emission from quantum cascade lasers based on nonresonant
extraction design approach,” Appl. Phys. Lett. 95, 141113–141115 (2009).
4. Y. Bai, N. Bandyopadhyay, S. Tsao, S. Slivken, and M. Razeghi, “Room
temperature quantum cascade lasers with 27% wall plug efficiency,” Appl.
Phys. Lett. 98, 181102–181105 (2011).
5. M. Troccoli, J. Fan, and X. Wang, “Quantum cascade lasers: high-power
emission and single-mode operation in the long-wave infrared (l > 6 mm),”
Opt. Engin. 49, 111106 (2010) [doi: 10.1117/1.3498778].
6. A. Tredicucci, F. Capasso, C. Gmachl, D L. Sivco, A. L. Hutchinson, and
A Y. Cho, “High performance interminiband quantum cascade lasers
with graded superlattices,” Appl. Phys. Lett. 73, 2101–2103 (1998).
7. Y. Bai, N. Bandyopadhyay, S. Tsao, E. Selcuk, S. Slivken, and
M. Razeghi, “Highly temperature insensitive quantum cascade lasers,”
Appl. Phys. Lett. 97, 251104–251106 (2010).
8. A. Lyakh, C. Pflügl, L. Diehl, Q. J. Wang, F. Capasso, X. J. Wang, J. Y. Fan,
T. Tanbun-Ek, R. Maulini, A. Tsekoun, R. Go, C. Kumar, and N. Patel,
“1.6 W high wall plug efficiency, continuous-wave room temperature quantum
cascade laser emitting at 4.6 mm,” Appl. Phys. Lett. 92, 111110 (2008).
9. A. Lyakh, R. Maulini, A. Tsekoun, R. Go, and C. K. N. Patel, “Multiwatt
long wavelength quantum cascade lasers based on high strain composition
with 70% injection efficiency” Opt. Express 20(22), 2472–9 (2012).
10. M. Troccoli, L. Diehl, D. P. Bour, S. W. Corzine, N. Yu, C. Y. Wang,
M. A. Belkin, G. Höfler, R. Lewicki, G. Wysocki, F. K. Tittel, and
F. Capasso, “High performance quantum cascade lasers grown by metal-
organic vapor phase epitaxy and their applications to trace gas sensing,”
IEEE J. Lightwave Technol. 26, 3534–3555 (2008).
11. M. Troccoli, C. Gmachl, F. Capasso, D. L. Sivco, and A. Y. Cho,
“Mid-infrared l ¼ 7.4 mm quantum cascade laser amplifier for high power
single-mode emission and improved beam quality,” Appl. Phys Lett. 80,
4103–4105 (2002).
12. P. Rauter, S. Menzel, A. K. Goyal, B. Geokden, C. A. Wang, A. Sanchez,
G. W. Turner, and F. Capasso, “Master-oscillator power-amplifier quantum
cascade laser array,” Appl. Phys. Lett. 101, 261117–261119 (2012).
High-Performance Quantum Cascade Lasers for Industrial Applications 239
11.1 Introduction
11.2 Some Properties of Narrow Gap Lead Chalcogenides (IV VI Compound Semiconductors)
11.2.1 Structure, binary compositions, alloying
11.2.2 Band structure and Auger recombination
11.2.3 Permittivities
11.2.4 Defects and non lattice matched growth
11.2.5 Growth on Si(111) and thermal mismatch dislocation glide
11.3 Applications
11.3.1 Broadband photovoltaic IV VI mid infrared detectors
11.3.2 Resonant cavity enhanced detectors
11.3.3 Edge emitting laser diodes
11.3.4 Monolithic vertical cavity surface emitting lasers (VCSELs) and microdisk lasers
11.4 VECSELs
11.4.1 Principle and structure of the long cavity
11.4.2 Optical and electronic simulation
11.4.3 Short cavity and end pumping
11.5 Conclusions
References
243
244 Chapter 11
11.1 Introduction
We describe the development of a tunable semiconductor disk laser, also
called a vertical external-cavity surface-emitting laser (VECSEL) for the mid-
infrared range (wavelength 2.5 mm to 10 mm). The intended applications
are mainly gas spectroscopy for trace gas analysis, air pollution monitoring,
or medical use.
The mid-infrared range is especially suited for gas spectroscopy since the
fundamental modes of many common gases lie in this range. This includes gases
such as CO2, CO, nitric oxides (NOx), many organic compounds (including
C–H vibration lines), SO2, or H2O. Gas spectroscopy in this range is therefore
especially interesting since the fundamental modes yield to much higher
sensitivity than when using shorter wavelengths in the near-infrared or even
visible range, where only higher-order lines can be analyzed.1
For the VECSEL, lead chalcogenide (IV-VI) narrow-gap semiconductor
layers are employed,2 as they are perfectly suited for the mid-infrared
wavelength range, while VECSELs fabricated with the well established III-V
technologies are restricted to wavelengths below 2.8 mm.3 6
Generally, VECSELs offer several advantages compared to conven-
tional edge-emitting laser diodes:3 VECSELs emit perpendicular to the
surface, so batch processing is possible leading to extremely low costs. The
emitted beam is circular with a cone angle of only a few degrees. Large
continuous tuning without any mode hops is easily attained by simply
changing the length of the cavity.
Note that quantum cascade lasers (QCLs)7 as well as interband cascade
lasers (ICLs)8 fabricated with III-V technology emit in the mid-infrared, but
both are edge-emitting lasers. They therefore emit a strongly astigmatic beam
with an aperture angle in the fast axis of up to 60 deg. This requires elaborate
optics for beam conditioning. Edge-emitting laser diodes additionally need
individual delineation of the edge mirrors. Tuning can be done with an external
cavity or by using QCL arrays;9 however, the design of the cavity is quite
elaborate in order to obtain a mode-hop-free tuning. An additional advantage
of the lead chalcogenide VECSEL is that the layers are epitaxially grown by
molecular beam epitaxy (MBE) on easily available and robust Si substrates.
The chapter is organized as follows: Since lead chalcogenides (IV-VI
semiconductors) are presently not widely employed, we first summarize some of
their properties. The reasons the IV-VIs can be grown by MBE on Si substrate
material, despite the huge lattice–thermal-expansion mismatch, are outlined
in Section 11.2. Some infrared applications are reviewed in Section 11.3. In
Section 11.4, the design and realization of IV-VI VECSELs on Si is
described, followed by their experimental and theoretically feasible properties.
True monomode emission is achieved with short cavities and end-pumping
(Section 4.3). Finally, a spectroscopic application is presented as an example.
Mid-infrared Tunable Surface-Emitting Lasers for Gas Spectroscopy 245
Figure 11.1 Change in cutoff wavelength when alloying PbTe with Sn, Eu, or Sr at two
distinct temperatures (reprinted from Ref. 30).
246 Chapter 11
semiconductors, where the Auger effect for wavelengths toward the mid-
infrared range increases so strongly that sophisticated structures for lasers must
be developed in order to arrive at acceptable low recombination.8
11.2.3 Permittivities
The permittivity of IV-VI semiconductors is extremely high (102 to >103 at
low temperatures). This makes infrared detectors slower than their
competitors, Hg1 xCdxTe detectors. (This is one reason that development
of IV-VI semiconductors for scanning infrared cameras for thermal imaging
was discontinued in the late 1970s in favor of HgCdTe.11 However, for
staring infrared focal plane arrays, this is not an issue). On the other hand,
the high permittivity efficiently shields charged defects. Therefore, even with
quite a high number of defects, mobilities are high (>106 cm2/Vs at low
temperatures), and device-quality layers result.
11.3 Applications
11.3.1 Broadband photovoltaic IV-VI mid-infrared detectors
Using these techniques, in 1985 one of the authors successfully demonstrated
IV-VI infrared sensor arrays on Si substrates.13 Only homogeneous layers of a
few microns in thickness were grown on CaF2-covered Si(111) substrate
wafers. Cutoff wavelengths ranged from 3 mm up to >15 mm.10 In one
application, a whole 2D monolithic infrared focal plane array was fabricated
on a Si substrate that contained integrated multiplexing electronics in a true
monolithic fashion.15 The individual infrared sensors, in this case photovoltaic
PbTe, were delineated on one part of the individual pixel with a bare Si(111)
surface, while the prefabricated multiplexing transistors were located on the
remaining part of the pixel.
thickness of the active layer in the cavity, a low noise volume results that can
lead to higher sensitivity than in a similar bulk detector with a broadband
response. It is interesting to note that IV-VI materials are currently the only
materials with which RCEDs can be realized in the mid-infrared range.
11.4 VECSELs
11.4.1 Principle and structure of the long cavity
A semiconductor VECSEL, also called a disk laser, consists of two high-
reflectivity mirrors that form a cavity. The length l of the cavity is typically a
few centimeters. A stack of thin active semiconductor gain layer(s) is placed
inside the cavity, but most of the cavity is free space. Emission is
perpendicular to the gain layers and the mirror surfaces. One of the mirrors
is flat, and, in most cases, the active-layer stack is monolithically grown onto
this mirror. The second mirror is curved with radius r (Fig. 11.2). The ratio l/r
determines the mode diameter on the active layer. This leads to much more
design freedom compared to the (monolithic) VCSEL.
Mid-infrared Tunable Surface-Emitting Lasers for Gas Spectroscopy 249
Since the active layers are thin, they must exhibit a high gain, and the
reflectivities of the two mirrors must be very high in order to reach
threshold. For a direct-gap semiconductor and especially for lead
chalcogenides, gain is typically very high, and threshold is easily reached.
In addition, Bragg mirrors with very high reflectivity can easily be grown
using epitaxial lead-chalcogenide layers owing to the high index contrast in
the mid-infrared range. As for the RCED, a well-suited high (H) index
material is PbSrTe (n > 5), while a low (L) index material is EuTe (n ¼ 2.3).
A few pairs of such H/L quarter-wavelength l0/4 layers (l0 is the design
emission wavelength) suffice to obtain a reflectivity R 99.9 %, even over a
broad wavelength range.
As with most III-V-based VECSELs, our mid-infrared IV-VI VECSELs
are optically pumped from the side (Fig. 11.2) using a separate pump laser
diode. This is very convenient since the focus of this diode can be
appropriately designed to fit to the mode diameter, and optical pumping
leads to a very homogeneous and scalable pump spot. No doping of the active
layers is needed, which ensures the lowest electronic loss. A disadvantage is
the quantum deficit (ratio of the photon energy of the pump wavelength with
respect to that of the mid-infrared emission wavelength), which leads to lower
wall-plug efficiency. We use commercial laser diodes emitting at 1.5- or 2-mm
wavelength as pump lasers.
The active-gain-layer stack has an optical thickness of a multiple of half-
wavelengths l0/2. This leads to resonance at the design wavelength l0 and
therefore the lowest threshold. However, due to the strong absorption of the
pump beam in this layer stack, its thickness is restricted to about 2l0.
250 Chapter 11
Figure 11.3 Normalized spectra of a VECSEL based on (a) a PbS QW and (b) “bulk”
PbSnSe (reprinted from Ref. 30).
Figure 11.4 Calculated gain g (left scale) and calculated threshold power Pthr (right scale)
for a VECSEL with a 830-nm-thick active PbTe gain layer, corresponding to a design
wavelength of l0 ¼ 5.3 mm (130 K), in comparison to the experimental Pthr (points) (reprinted
from Ref. 25).
252 Chapter 11
Figure 11.5 Calculated Pthr of a VECSEL with 7 PbSe QWs, each 9.5-nm thick and
separated by Pb0.93Sr0.07Se barrier layers for Shockley Read lifetimes of 0.1 and 1 ns
(dashed lines), in comparison to the experimental values (points) (reprinted from Ref. 29).
Mid-infrared Tunable Surface-Emitting Lasers for Gas Spectroscopy 253
much shorter than those achieved for homogeneous gain layers, and nearly
three orders of magnitude above the theoretical limit when t sr can be
neglected. With PbS-based QW structures, Shockley–Read lifetimes around
1 ns were obtained, again still much shorter than for “bulk” homogeneous
layers.24
The reasons for this discrepancy remain unclear. One may argue that:
• SR recombination is stronger in 2D (QW) than in 3D (“bulk”)
structures.
• Asymmetrical band offsets lead to carrier leakage at the higher
temperatures.
• 2D Auger recombination is much stronger than in 3D. Such behavior
was found for certain conditions in III-V QW structures.31 For IV-VI
materials, no such calculations are known.
• The design of QW lasers is still far from optimized.
At the time of writing this overview, threshold pump powers have been
significantly decreased to 1 Wp near RT. This is due to improved growth
procedures, design, and fabrication. Details will be described in a forthcoming
article.32
Figure 11.6 Layout of a tunable monomode VECSEL with end pumping (reprinted from
Ref. 34).
Figure 11.7 Normalized spectra for different piezovoltages of a tunable monomode PbSe
QW VECSEL (reprinted from Ref. 2).
pump beam onto the surface of the gain layer, or by limiting the lateral
extension of the excited region by, e.g., an aperture.
In Fig. 11.7, some results for a tunable PbSe QW VECSEL designed to
probe C–H absorption lines around 3.3-mm wavelength are shown. The cavity
length is around 50 mm, and operation is near RT. The normalized spectra at
Mid-infrared Tunable Surface-Emitting Lasers for Gas Spectroscopy 255
Figure 11.8 Measured transversal intensity distribution of the output beam of the
monomode VECSEL. The observed single TEM00 mode is circular with an angle of the
output cone of a few degrees (reprinted from Ref. 35). (See color plate section.)
11.5 Conclusions
Lead chalcogenide (IV-VI) narrow-bandgap semiconductors are the only
material system suited to fabricate VECSELs (also called disk lasers) for the
mid-infrared range (3 to >10 mm wavelength), while all III-V-based lasers
such as QCLs or ICLs for this range are edge emitters. VECSELs offer several
advantages compared to edge emitters as they can be easily tuned, are
scalable, and exhibit a high beam quality.
256 Chapter 11
Figure 11.9 Absorption spectrum of H2O vapor around 1844 cm1 (5.42 mm) recorded
with a PbTe-based tunable monomode VECSEL (reprinted from Ref. 34).
References
1. A. W. Mantz, A. I. Nadezhdinskii, M. W. Sigrist, and F. K. Tittel, Articles
from Special Issue: Selected papers presented at the 7th International
Conference on Tunable Diode Laser Spectroscopy,” Appl. Phys. B 100(2),
231–232 (2009).
2. M. Fill, F. Felder, M. Rahim, A. Khiar, R. Rodriguez, H. Zogg, and
A. Ishida, “IV-VI mid-infrared VECSEL on Si-substrate,” Proc. SPIE
8242, 82420H (2012) [doi 10.1117/ 12.905643].
3. M. Kuznetsov, F. Hakimi, R. Sprague, and A. Mooradian, “High-power
(>0.5 W) CW diode pumped vertical-external-cavity surface emitting
semiconductor lasers with circular TEM00 beams,” IEEE Photon.
Technol. Lett. 9, 1063 (1997).
Mid-infrared Tunable Surface-Emitting Lasers for Gas Spectroscopy 257
external cavity surface emitting laser on Si,” J. Appl. Phys. 110, 023101
(2011).
29. M. Fill, A. Khiar, M. Rahim, F. Felder, and H. Zogg, “PbSe quantum
well mid-infrared vertical external cavity surface emitting laser on
Si-substrates,” J. Appl. Phys. 77, 277 (1995).
30. M. Fill, “PbSe quantum well based mid-infrared vertical surface emitting
lasers on silicon,” Ph.D. thesis: Diss. ETH No. 19980 (2011).
31. R. G. Bedford, G. Triplett, D. H. Tomich, S.W. Koch, J. Moloney, and
J. Hader, “Reduced Auger recombination in mid-infrared semiconductor
lasers,” J. Appl. Phys. 110, 073108 (2011).
32. M. Fill and F. Felder, article in preparation for journal submission.
33. A. Khiar, M. Rahim, M. Fill, F. Felder, F. Hobrecker, and H. Zogg,
“Continuously tunable monomode mid-infrared vertical external cavity
surface emitting laser on Si,” Appl. Phys. Lett. 97, 151104 (2010).
34. M. Rahim, A. Khiar, M. Fill, F. Felder, and H. Zogg, “Continuously
tunable singlemode VECSEL at 3.3 mm wavelength for spectroscopy,”
Electron. Lett. 47, 1037–1039 & 1008 (2011).
35. H. Zogg, “Lead chalcogenide widely tunable monomode vertical external
cavity surface emitting layer (VECSEL) near room temperature,”
(abstract) http://miomd-11.northwestern.edu/technical/speakers.php.
12.1 Introduction
12.2 Frequency Noise and Laser Linewidth in QCLs: Experimental Methods
12.2.1 Relation between frequency noise and laser linewidth
12.2.2 Frequency noise measurement methods
12.3 Intrinsic Linewidth in QCLs
12.4 Impact of Technical Noise on the QCL Experimental Linewidth
12.5 Overview of Reported Frequency Noise Spectra in 4 to 5 mm QCLs
12.5.1 Free running QCLs
12.5.2 Frequency stabilized QCLs
12.6 Temperature Dependence of the Frequency Noise in a QCL
12.7 The Origin of Frequency Noise in QCLs
12.8 Conclusion and Outlook
References
12.1 Introduction
Since their first demonstration in 1994,1 quantum cascade lasers (QCLs) have
shown remarkable improvements in terms of technology and overall
performance. The first devices were operated in pulsed mode only, at
cryogenic temperatures, with a multimode emission and a tiny average output
power. In less than one decade, advancements in the design, fabrication, and
261
262 Chapter 12
with a reference CW QCL, since the beat linewidth was dominated by the
contribution of the CW laser.
Therefore, many applications would greatly benefit from QCLs with
narrower linewidth, and the development of such lasers is of significant
interest for high-resolution spectroscopy. External cavity (EC) QCLs have
been developed, achieving a large tuning range of over 15% of their central
wavelength,28 but so far without significant improvement in terms of
linewidth as compared to DFB QCLs.29 In order to develop QCLs with
narrower linewidth, the mechanisms contributing to the linewidth need to be
better understood. Generally speaking, fluctuations of the laser emission
frequency occurring at different timescales are responsible for the broadening
of the linewidth. These fluctuations are characterized by the laser frequency
noise power spectral density (PSD), expressed in units of Hz2/Hz. Until
recently, this quantity has been studied and optimized very little in QCLs.
This is in sharp contrast to the other parameters that are more commonly
considered, such as the output power or the wavelength tuning, to name a few.
However, there has been a growing interest during the last couple of years for
this topic, ranging from basic studies of the frequency noise in QCLs at either
cryogenic19 or room temperature,20,21,24 to studies of its dependence as a
function of the laser temperature,22 and investigations of its possible
origin,22,23 as well as studies in relation to the frequency stabilization of
mid-infrared QCLs.30 33
In this chapter, we present an overview of experimental results obtained in
recent years on the frequency noise of QCLs. The overview is based on a
compilation of both our own work and studies from other laboratories. We
also briefly discuss the frequency noise reduction obtained by different active
stabilization methods. First of all, we begin in Section 12.2 with a short
reminder about the relation between frequency noise and linewidth in a laser
and show how the frequency noise spectrum of a QCL can be experimentally
measured. Then we discuss the intrinsic linewidth of QCLs in Section 12.3 and
the impact of technical noise on the experimentally observed linewidth in
Section 12.4. Subsequently, we revisit the experimental results on the
frequency noise measured in different QCLs reported worldwide (Section
12.5), and we present our study of the temperature dependence of the
frequency noise in a 4.6-mm QCL (Section 12.6). Finally, Section 12.7
mentions some aspects of the origin of frequency noise in QCLs.
be directly measured from the heterodyne beat between the laser under test
and a reference laser or from self-homodyning using a long-delay-line
interferometer.34 This method is straightforwardly applicable to near-infrared
lasers but not to QCLs due to the lack of low-loss and affordable optical fibers
in the mid-infrared spectral region.
The linewidth is a single value that is adequate for a rough comparison of
the spectral properties of different lasers, but it gives a very incomplete view of
the spectral distribution of the laser frequency noise. On the other hand,
complete information is given by the laser frequency noise PSD Sdn ðf Þ, which
represents the spectral distribution of the laser frequency fluctuations.
Knowing the frequency noise PSD of a laser, its exact lineshape and the
corresponding linewidth can be calculated with the two-step integration
introduced by Elliott et al.35 and discussed later on by other authors:36 38
2 0 13
Z1 Z1 2
sin ðpf tÞ
SE ðnÞ ¼ 2 e i2pnt 4E02 ei2pn0 t exp@ 2 Sdn ðf Þ df A5dt: ð12:1Þ
f2
1 0
Figure 12.1 Measured frequency noise spectrum of a 4.6-mm DFB QCL composed of
flicker (1/f ) noise. An approximate linewidth is obtained from the surface A (colored area in
the plot) of the slow-modulation area for which the frequency noise PSD exceeds the
b-separation line (dashed red line). A low-frequency cutoff fc ¼ 200 Hz (to ¼ 5 ms) is
introduced in the calculation of the surface A to prevent the divergence of the surface and of
the retrieved linewidth for infinite observation times (adapted from Ref. 24).
The surface A is the geometrical area under the frequency noise PSD
obtained for all Fourier frequencies for which Sdn( f ) exceeds the
b-separation line.40 In the case of a typical frequency noise spectrum
dominated by flicker (1/f ) noise, as displayed in Fig. 12.1 for a DFB QCL
at 4.6 mm, a cutoff frequency fc (inverse of the observation time to in which
the linewidth is measured, fc ¼ 1/to) needs to be introduced to prevent the
divergence of the surface A at low frequency. An experimental validation of
the approximated linewidth given by Eq. (12.2) proved its accuracy (within
the experimental uncertainties) over more than three decades of linewidth
values.39 In the example of Fig. 12.1, the linewidth of the DFB QCL
approximated using Eq. (12.2) is around 900 kHz for a cutoff frequency of
1 Hz (1-s observation time) and 560 kHz for a cutoff frequency of 200 Hz
(5-ms observation time).
manner, but this has been rarely implemented with QCLs in the mid-
infrared, in contrast to the near-infrared where the method is quite common
owing to the much easier availability of Fabry–Pérot cavities. N2O at
8.5 mm,41 CO2 at 4.3 mm,19,21,23 or CO at 4.6 mm20,22,24 are some examples
of reported molecular frequency discriminators. This method has also been
implemented recently with a terahertz QCL using a Doppler-broadened
methanol molecular transition.42 The laser under test is generally tuned to
the side of the absorption line, to a linear region where frequency
fluctuations are linearly converted into intensity fluctuations that are
subsequently measured with a photodiode. Reduced cell pressure operation
is required to achieve a narrow-linewidth absorption feature, providing a
high conversion factor of frequency noise into intensity noise (the so-called
discriminator slope). The discriminator slope must be sufficiently high that
the contribution of the frequency noise converted into intensity noise is
much larger than the direct contribution of the laser intensity noise, which
also impacts the detector signal. The parameters of the gas cell (path length,
gas pressure) must therefore be properly selected in order to optimize the
discriminator slope.24
The experimental setup implemented in our studies is shown in Fig. 12.2.
The strongly divergent QCL emission is collimated with an aspheric ZnSe
Figure 12.2 Scheme of the experimental setup used to measure the frequency noise of a
QCL at 4.6 mm. The R(14) CO absorption profile measured in the gas cell is shown in the
lower part of the figure. The linear side of the absorption profile is used as a frequency
discriminator to convert frequency noise (FM) into intensity noise (IM). The discriminator
slope D is indicated by the dashed line.
Frequency Noise and Linewidth of Mid-infrared Continuous-Wave QCLs… 267
lens, then passes through the gas-filled absorption cell and is finally focused
onto a low-noise HgCdTe photodiode. The cell is only 1-cm long and is
filled with pure CO at a pressure of about 20 mbar, which is sufficient to
produce a high absorption, as a result of the intense linestrength of the CO
lines in the fundamental vibrational band. The voltage fluctuations of the
photodiode output signal are measured with a fast Fourier transform (FFT)
spectrum analyzer for Fourier frequencies of up to 100 kHz and with an
electrical spectrum analyzer (ESA) at higher frequencies. The PSD of the
photodiode output voltage is converted into frequency noise PSD of the
laser by scaling with the squared value of the measured discriminator slope
D, which is determined from the recorded cell transmission spectrum
obtained when scanning the laser through the absorption line. A linear fit of
the side of the molecular transmission profile provides the discriminator
slope, as shown in Fig. 12.2 for the R(14) rovibrational transition of CO at
2196.6 cm 1.
A second method to measure the frequency noise spectrum of a laser is
by beating the laser under test with a reference laser and by analyzing the
frequency noise of the radio-frequency (RF) beat with an RF discriminator.
The method is particularly attractive at wavelengths where a proper optical
discriminator is not available, but the price to pay is the need for a second
laser. This reference laser is either similar to the first one or has a much
smaller frequency noise. In the first case, the two lasers are considered to
contribute equally to the noise of the RF beat, and the frequency noise PSD
of the beat signal is twice that of a single laser. In the second case, the
frequency noise PSD of the laser under test is directly obtained since the
noise of the reference laser is negligible. An RF discriminator has been
successfully implemented to characterize near-infrared lasers43,44 but, to the
best of our knowledge, has not been used so far with QCLs.
Finally, a third method to measure the frequency noise spectrum of a
QCL was recently proposed by Knabe et al.,29 based on the characteriza-
tion of the instantaneous optical frequency of an EC QCL by comparison
to a near-infrared optical frequency comb. The near-infrared fiber comb
centered at 1550 nm was combined with the 4.5-mm QCL in a nonlinear
crystal to generate a new frequency comb at 1150 nm by sum-frequency
generation. This new comb combines the frequency properties of the QCL
and of the near-infrared comb. The frequency fluctuations of the QCL
were obtained by a subsequent heterodyning of the 1150-nm comb with a
supercontinuum comb generated from the spectrally broadened near-
infrared comb. The resulting beat signal contains mainly the frequency
fluctuations of the QCL, the contribution of the comb repetition rate being
negligible for a stabilized comb. The beat signal was digitized and
processed to extract the QCL instantaneous frequency and its frequency
noise spectrum.
268 Chapter 12
upper limit for the intrinsic linewidth of 300 Hz, a value that is interesting to
compare with the theoretical linewidth given by Eq. (12.3). As already
mentioned, the linewidth enhancement factor ae is close to zero in a QCL, and
one can reasonably consider ae ¼ 0 in Eq. (12.3). A delicate aspect in properly
evaluating the Schawlow–Townes linewidth for a DFB QCL lies in the
difficulty in determining both the total cavity losses atot ¼ am þ awg (mirror
and waveguide) and the mirror losses am to compute the intracavity optical
power from the output power P0. It is not correct to consider the losses of a
Fabry–Pérot laser, which can be straightforwardly determined. This may lead
to an overestimation of the theoretical linewidth because the losses of the
grating (aDFB) need to be accounted for in a DFB laser, and these losses are
generally lower than the mirror losses in a Fabry–Pérot laser. The total losses
in our DFB QCL were estimated by comparing its threshold current with
those of a similar Fabry–Pérot device (same manufacturer and close design,
but with slightly different dimensions), assuming that the ratio of the total
losses coincides with the ratio of the threshold currents. Considering similar
waveguide losses awg 4.5 cm 1 in both cases (as measured by the
manufacturer on Fabry–Pérot devices), the grating losses are estimated to
aDFB ¼ 1.47 cm 1, which is in relatively good agreement with data reported
for a 9-mm DFB QCL (awg ¼ 6.7 cm 1, aDFB ¼ 0.7 cm 1).50 For comparison,
the mirror losses in the Fabry–Pérot laser were higher: am ¼ 2.7 cm 1. With a
spontaneous emission coefficient nsp ¼ 1, an optical power P0 ¼ 6 mW, and
using typical values of the laser parameters provided by the manufacturer, an
intrinsic linewidth DnST 380 Hz is obtained. This value is very slightly
affected by the only partial knowledge of the waveguide losses in the
Fabry–Pérot laser, as changing these losses in the range of 4 to 5 cm 1 only
changes the computed intrinsic linewidth by a couple of hertz. The calculated
theoretical linewidth is thus in good agreement with the value of 300 Hz
assessed from the frequency noise spectrum.
laser, Dn/DT and Dn/DI, respectively. These are typically on the order of
5 GHz/K and 0.3–1 GHz/mA in DFB QCLs at 4- to 5-mm wavelength. An
average temperature stability of the laser of 0.01 K is usually sufficient and
is achievable with a standard regulation loop that includes a thermoelectrical
cooler and a negative thermal coefficient resistor as a temperature sensor.
Temperature variations arise on a long timescale and only induce a slow drift
of the laser emission frequency.
Much more attention must be paid to the current source that drives the
laser because the current noise of the driver can degrade the laser frequency
noise and therefore broadens the linewidth to a significant extent. A low-noise
driver is required to observe the frequency noise originating from the QCL
itself. We refer to this frequency noise as the noise inherent to the laser (not to
be confused with the intrinsic noise that is responsible for the intrinsic
linewidth discussed in Section 12.3), without limitation from the current noise
of the driver. In Ref. 24, we experimentally compared the impact of two
different current sources on the frequency noise and linewidth of a 4.6-mm
DFB QCL. The current noise of these two drivers was measured on a resistive
load (30 V) at the same voltage and current used to drive the QCL. The
current noise spectral density is shown in Fig. 12.3(a). The first driver (labelled
Driver-1) has a white current noise of about 2 nA/Hz1/2, whereas the second
driver has an average noise floor of 350 pA/Hz1/2 (and even 200 pA/Hz1/2
in the range of 1–10 kHz). The impact on the laser frequency noise is clearly
visible in Fig. 12.3(b): Driver-1 induces an excess frequency noise in the range
of 1 kHz to 3 MHz. The contribution of the current source to the laser
frequency noise, calculated from the white current noise and combined with
the dynamic frequency response of the laser, is also displayed in Fig. 12.3(b).
This current noise fully explains the observed excess frequency noise, which
leads to a laser linewidth of about 1.5 MHz (at 5-ms observation time).
On the other hand, the laser frequency noise spectrum obtained with
Driver-2 corresponds to the noise inherent to the laser, without any
degradation induced by the driver, apart from the small bumps visible
around 400 kHz and at higher frequencies. The corresponding linewidth is
550 kHz (5-ms observation time), a factor of 3 narrower than with the use
of Driver-1. The importance of having a low-noise current source to reach the
frequency noise inherent to the QCL and avoid any excess noise clearly
appears in these results. They show that a current noise of roughly 2 nA/Hz1/2
already broadens the laser linewidth by a factor 3.
The maximum tolerable driver-current noise that enables the inherent
noise of the laser to be reached without degradation from technical noise was
simulated in Ref. 24 for a 4.6-mm DFB QCL. For this purpose, the impact of
different levels of white-current noise on the QCL linewidth was calculated
based on the formalism of Section 12.2.1. White current noise was considered
because it is the most representative type of noise in typical laser drivers.
Frequency Noise and Linewidth of Mid-infrared Continuous-Wave QCLs… 271
Figure 12.3 (a) Current noise spectral density of two QCL drivers. The noise peaks
in the range of 10 100 kHz are experimental artifacts due to the measurement setup.
The noise of Driver-2 below 100 Hz is limited by the 1/f instrumental noise floor.
(b) Frequency noise of a 4.6-mm DFB QCL obtained with the use of two current drivers.
The contribution of the drivers to the frequency noise is displayed as thin lines obtained
by combining the current noise with the measured laser transfer function (shown by
green circles on the right vertical scale). (See color plate section.)
The contribution of the current noise to the laser frequency noise was
calculated taking into account the dynamic frequency response of the laser.
Figure 12.4(a) shows the calculated frequency noise corresponding to three
different levels of current noise: 350 pA/Hz1/2, 2 nA/Hz1/2, and 10 nA/Hz1/2.
The first two values correspond to the noise of the two aforementioned laser
drivers, whereas the highest current noise of 10 nA/Hz1/2 typically corresponds
to some commercial QCL drivers. The calculated impact of the driver current
noise on the QCL linewidth is shown in Fig. 12.4(b). One notices that
the linewidth is unaffected for current noise densities <0.5 nA/Hz1/2 but
drastically increases at a rate of 1.6 MHz/(nA/Hz1/2) for current noise
>1 nA/Hz1/2.
272 Chapter 12
1010
108 2 [nA/ Hz ]
106
104
line
ion
ep arat 2 ) f 350 [pA/ Hz ]
102 -s )
n(2
(8L
100
101 102 103 104 105 106 107 108
Frequency [Hz]
(a)
2
10
FWHM linewidth [MHz]
101
100
10-1
10-1 100 101 102
Current noise [nA/ Hz ]
(b)
Figure 12.4 (a) Simulation of the laser frequency noise induced by drivers with different
current noise densities. (b) Calculated laser linewidth (FWHM) as a function of the driver
current noise density (white current noise is considered). The dashed line represents the
linear broadening rate [1.6 MHz/(nA/Hz1/2)] of the laser linewidth as a function of the
current noise (reprinted from Ref. 24).
Table 12.1 Comparison of different studies performed during recent years on the
frequency noise of 4- to 5-mm QCLs (DFB or EC). The reported linewidth is usually
determined from the frequency noise spectrum using either the exact numerical integration
of Eq. (12.1) or the approximation given by Eq. (12.2) for an observation time in the range of
1 to 10 ms. [RT: room temperature (unspecified value).]
Laser type l (mm) Top (K) Manufacturer Discriminator FWHM (MHz) Group Ref.
Figure 12.5 Frequency noise spectra of two similar DFB QCLs from the same fabrication
run, operated in slightly different conditions (QCL-1: T ¼ 5 8C, Iop ¼ 350 mA; QCL-2: T ¼ 5 8C,
Iop ¼ 400 mA). The excess noise at high frequency (around 400 kHz and above 1 MHz) is
due to technical noise induced by the laser driver. Inset: heterodyne beat-note signal between
the two QCLs, showing a beat linewidth of <1 MHz. (See color plate section.)
Figure 12.6 Comparison of the frequency noise PSD reported for different DFB QCLs in
the 4.3- to 4.6-mm range, operated at cryogenic temperature [blue curve (from Ref. 19)] and
room temperature [red and orange curves (from Refs. 20 and 21, respectively, courtesy of
S. Bartalini, LENS)]. (See color plate section.)
Figure 12.6 shows a comparison of the frequency noise spectra obtained for
the three aforementioned DFB QCLs in the Fourier frequency range from
20 Hz to 50 MHz. The two lasers operated at room temperature show a rather
similar 1/f noise spectrum, whereas the cryogenic QCL has a two-orders-of-
magnitude-higher noise at low frequency, namely up to 100 kHz. At higher
frequencies, the noise dependence turns into a 1=f 2 regime in the cryogenic
QCL, and the three lasers seem to reach a similar white frequency noise floor
at high frequency. These results may indicate a general influence of
Frequency Noise and Linewidth of Mid-infrared Continuous-Wave QCLs… 275
reducing the QCL linewidth. In contrast, optical injection locking has been
recently demonstrated for the first time to link a 4.67-mm QCL to a frequency
comb.31 In this case, a significant reduction of the QCL linewidth to 40 kHz
was achieved. A linewidth of 25 kHz (at 1-ms observation time) was also
achieved by phase locking a 9-mm QCL to a thulium comb.32 However the
comb was free-running in this case, so an absolute frequency accuracy of the
QCL was not achieved.
Finally, a subkilohertz linewidth was reported for a 4.3-mm DFB QCL
stabilized to a sub-Doppler CO2 transition by polarization spectroscopy.33
The linewidth and frequency accuracy of the stabilized QCL was assessed
from the beat signal with a frequency comb.
Figure 12.7 Frequency noise PSD of a 4.5-mm DFB QCL measured at different
temperatures ranging from 128 to 283 K. The laser operating current was adjusted to
I0/Ith 1.5 at each temperature (reprinted from Ref. 22). (See color plate section.)
Figure 12.8 Temperature dependence of the frequency noise PSD in a DFB QCL
measured at 3-kHz Fourier frequency (red diamonds). The gray lines result from a fit of
the experimental data on both sides of the transition occurring at T 200 K,
corresponding to S3kHz ¼ 7 106 Hz2/Hz for T > 200 K and S3kHz(T) 2 1012 exp
( 0.06 T) for T < 200 K. The black crosses represent the noise measured on the voltage
across the laser, converted into an equivalent frequency noise using the laser differential
resistance and the current-tuning coefficient (see discussion in Section 12.7). The yellow
markers represent published values of QCL frequency noise obtained at different
temperatures19–21 (reprinted from Ref. 22). (See color plate section.)
This internal electrical noise showed the same 1/f nature as the observed laser
frequency noise. More importantly, it showed the same abrupt increase at
temperatures below 200 K.
We evaluated the impact of this internal electrical noise by determining its
contribution to the laser frequency noise. This was achieved by multiplying
the electrical noise by the current-tuning rate of the laser measured at the
corresponding temperature. In Fig. 12.8, the 3-kHz component of the
frequency noise assessed from the measured electrical noise is displayed as
black crosses. It is found to be well correlated with the frequency noise
optically measured using the CO lines. These results tend to show that the
laser frequency noise is induced by internal electrical noise in the QCL
structure and that the strong increase in frequency noise at low temperatures
results from the strong increase in the electrical noise. However, the
mechanism that produces the electrical noise in the QCL is not yet well
understood. 1/f noise is quite typical in semiconductor devices and has been
studied for a long time in various experiments,55 but its origin is generally
difficult to identify. Among possible effects, additional current noise in QCLs
may originate from the contacts, from lattice scattering,56 from carriers
trapped by material defects,57 or from fluctuations in electron tunneling
through the multiple-barrier QCL structure.23 Additional experiments, such
as the characterization of frequency-noise temperature dependence in other
QCLs, possibly with different parameters, will be needed to better understand
the origin of frequency noise.
affect the laser frequency noise. In terms of numbers, we showed that a white-
current noise of 1 nA/Hz1/2 is a typical upper limit tolerable for a laser
controller to reach the linewidth inherent to the QCL at the megahertz level.
The requirement is even more stringent for the narrowest linewidth free-
running QCLs observed to date (500-kHz linewidth at about 10 ms, which
requires a current noise inferior to 500 pA/Hz1/2). These values can slightly
vary between different QCLs depending on their current-tuning coefficient,
but they represent a good approximation for any QCL. The main mechanism
of frequency tuning via the drive current is through thermal effects induced by
thermal dissipation in the laser structure. The current noise of the laser
controller is thus transformed into frequency noise via the laser dynamic
tuning response.
Even in the absence of external frequency noise, the linewidth of QCLs
that is experimentally observed is much broader than that given by the
Schawlow–Townes formula. Different experimental results tend to show that
this frequency noise is induced by internal current noise produced in the laser
structure. So far, the origin of the internal current noise in QCLs has not been
clearly identified, even if different options have been proposed, such as noise
induced in the electrical contacts, noise resulting from lattice scattering, noise
from carriers trapped by material defaults, or from fluctuations of electron
tunneling through the multiple-barrier QCL structure. Further experimental
investigations will likely provide new insights on and, hopefully, soon, clarify
this question.
References
1. J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, and
A. Y. Cho, “Quantum cascade lasers,” Science 264, 553–556 (1994).
2. J. Faist, F. Capasso, C. Sirtori, D. L. Sivco, J. N. Baillargeon,
A. L. Hutchinson, S. N. Chu, and A. Y. Cho, “High power mid-infrared
(l 5 mm) quantum cascade lasers operating above room temperature,”
Appl. Phys. Lett. 68(26), 3680–3682 (1996).
3. J. Faist, C. Gmachl, F. Capasso, C. Sirtori, D. L. Silvco, J. N. Baillargeon,
and A. Y. Cho, “Distributed feedback quantum cascade lasers,” Appl.
Phys. Lett. 70(20), 2670–2672 (1997).
4. M. Beck, D. Hofstetter, T. Aellen, J. Faist, U. Oesterle, M. Illegems,
E. Gini, and H. Melchior, “Continuous wave operation of a mid-infrared
semiconductor laser at room temperature,” Science 295, 301–303 (2002).
5. P. Corrigan, R. Martini, E. A. Whittaker, and C. Bethea, “Quantum
cascade lasers and the Kruse model in free space optical communication,”
Opt. Express 17, 4355–4359 (2009).
Frequency Noise and Linewidth of Mid-infrared Continuous-Wave QCLs… 281
John Zeller
Magnolia Optical Technologies, Inc., Woburn, Massachussetts, USA
13.1 Introduction
13.1.1 Motivation
13.1.2 Terahertz QCLs: background and recent developments
13.1.3 Terahertz QCLs: challenges
13.2 Terahertz QCLs: Structure and Design
13.2.1 Lasing in terahertz QCLs
13.2.2 Rate equations of a three level QCL
13.2.3 Electron transmission in QCLs
13.3 Simulation and Analysis
13.3.1 Absorption and optical gain
13.3.2 Terahertz output power and wall plug efficiency
13.3.3 Polar versus nonpolar cases
13.4 Conclusion
References
289
290 Chapter 13
13.1 Introduction
13.1.1 Motivation
Terahertz lasers spanning 100 GHz to 10 THz of the electromagnetic
spectrum offer a technology platform with anticipated applications ranging
from detection/imaging of chemico-biological systems to integrated circuits.
Terahertz is a non-ionizing radiation that has also found wide application in
noninvasive and contactless imaging of obscure or hidden objects with
potential use in security communities. For such applications, high-power
terahertz sources operating at room temperature (RT) are often preferred but
become bulky if implemented using standard optical techniques. Optically
pumped terahertz lasers provide power in the range of hundreds of milliwatts,
while output is restricted to a few microwatts if terahertz radiation is
generated using optical heterodyning.
Solid state terahertz lasers provide compactness; however, high-power RT
operation still remains a challenge. Shur et al.1 pioneered terahertz generation
using plasma waves formed in 2D electron fluid in the channel of advanced
heterostructure field-effect transistors (HFETs). Emission of terahertz
radiation has been reported in a GaN HFET with a channel length of
1.5 mm at cryogenic temperatures. Knapp et al.2 report terahertz generation
using InGaAs high-electron-mobility transistors with 60-nm channels.
Recently, Fathololoumi et al.3 were able to generate terahertz radiation at
temperatures as high as 199.5 K using the resonant phonon method, and
Wade et al.4 achieved the same at a temperature of 225 K in a high magnetic
field. Resonant tunneling diodes (RTDs) and quantum cascade lasers (QCLs)
operating at terahertz frequencies offer another technology platform for the
generation of terahertz radiation. While RTDs have limited output power,
QCLs are suitable candidates for high-power, continuous-wave (CW)
terahertz sources operating at RT.
while maintaining the value of the optical dipole matrix element. Since the
eigenstates are separated only by a small amount of energy, it becomes more
arduous to create population inversion by using resonant longitudinal optical
phonon scattering to adjust the lifetimes of each level. As was pointed out by
Tonouchi,10 the present research goals are to reduce the lasing frequency,
operate the terahertz QCL system at RT, and increase the range of tunable
frequencies. Of the different approaches to realize QCLs, RP has been the
choice due to its ability to enable high-power operation at elevated
temperatures. Using III-V semiconductor heterostructures with modified
barrier design, Bai et al.11 reported a shallow-well design demonstrating InP-
based QCLs emitting around 4.9-mm wavelength with 27% and 21% wall-plug
efficiencies (WPEs) at RT (298 K) pulsed and CW operations, respectively,
providing maximum single-facet CW power at RT of 5.1 W. Maulini et al.12
achieved a maximum WPE of 19% in pulsed mode at 293 K, with CW output
power of 1.4 W and WPE of 10%, and 1.2 W of average power in uncooled
operation.
Table 13.1 lists the recently demonstrated performance of terahertz QCLs.
The low longitudinal optical (LO) phonon energy restricts operation at
elevated temperatures and can be avoided by using SiGe/Si heterostructures
that have a dominant phonon mode at 64 meV. III-nitrides also offer an
interesting alternative with an LO phonon energy of 90 meV and favoring RT
Figure 13.1 Schematic of a Zn0.95Mg0.05O terahertz QCL device along with conduction-
band profile and associated wavefunctions for the ZnO/Zn0.95Mg0.05O MQW structure,
biased at 100 kV/cm. The layer sequence for one period of the structure beginning from the
left in the injector region (in nanometers) is 4/1.5/3.2/2.5/2.3/3.5/2.2/3. Lasing is expected to
occur between the upper state (n ¼ 3) and lower state (n ¼ 2), and rapidly depopulate with
LO phonon scattering out of the lower state (n ¼ 2) into the ground state (n ¼ 1). The black
script denotes the walls and the barriers, respectively (reprinted from Ref. 22).
where the diagonal matrix elements are dj ¼ ðħ2 =mħ20 Þ þ Vj , Vj is the potential
at point j, ħ0 is the interval between adjacent sampling points, and the
adjacent off-diagonal matrix elements are uj ¼ ðħ2 =2mħ20 Þ . The eigenenergies
and eigenfunctions may be calculated by using Eq. (13.3).
Figure 13.1 shows two periods of the conduction-band energy profile with
each period consisting of injector, active, and collector layers. The energy
profile and potential distribution were calculated by solving the Schrödinger
equation. The overall QC structure consists of eight sets of the periodic
structures sandwiched between metal electrodes and biased externally with an
electric field F ¼ 100 kV/cm. This figure also shows the calculated eigenstates
obtained by solving the one-electron Schrödinger equation under the effective
mass approximation. The material layer sequence of the structure shown in
Fig 13.1 is 4/1.5/3.2/2.5/2.3/3.5/2.2/3 (in nanometers), beginning from the left
at the injector region. Parameters dependent on the Mg mole fraction of the
barrier material (xMg ¼ 0.05) of the ternary compound such as conduction-
band offset and barrier electron effective mass are estimated by using linear
interpolation of values for ZnO and MgO: [Parameter (Zn1 xMgxO) ¼ x
Parameter (MgxO) þ (1 x) Parameter (ZnO)].
The values of the important binary material parameters are listed in
Table 13.2. The indices in Fig. 13.1 indicate individual eigenstate levels that
contribute to electron scattering. In this process, electrons are injected
from the upper state (n ¼ 3) into the lower state (n ¼ 2) and make vertical
transitions into the ground state (n ¼ 1). The expected emission frequency is
Table 13.2 Comparison of the material parameters of AlN, GaN, ZnO, and MgO.
1
c2 ¼ p : ð13:10Þ
ðP2 ÞðCeiP2 x þ De iP2 x Þ
The carrier population difference DN between the initial and final states at
finite temperature is expressed as44
2 3
ðEi EF Þ
1 þ exp
4pnr 6 kB T 7
DN ¼ Ni Nf ¼ 2 kB T ln4 6 7, ð13:25Þ
h L ðEf EF Þ 5
1 þ exp
kB T
where L is the width of the QW, Ei is the intial energy, and Ef is the final
energy:44
Ec þ Ev 3 m
EF ¼ þ kB T ln h ; ð13:26aÞ
2 4 me
2 2mr
fif ¼ P2
¼ Eg h f jZjii2 ð13:26bÞ
m ħv if ħ
is the oscillator strength, Pif is the momentum matrix element, and Eg ¼ Ef
Ei is the transition energy. Once the wavefunction has been decided, the
operator is solely the excited distance.
A comparison of the absorption coefficients also favors ZnO/Zn0.05Mg0.95O
heterostructures, as can be seen from Fig. 13.2. The absorption coefficient
of ZnO/Zn0.15Mg0.85O is 2.8 105 cm 1, compared to 1.5 105 cm 1 for the
AlGaN/GaN-material system. The larger absorption coefficient in ZnO/
Zn0.85Mg0.15O is attributed to the comparatively smaller average effective
mass m* along with larger oscillator strength fif.
Figure 13.2 Optical absorption coefficient spectra for the Zn1 xMgxO heterostructure. The
line shape of the absorption coefficient can be changed by using different Mg concentrations.
Shift of the absorption coefficient to higher energy values with increasing Mg mole fraction is
observed. (See color plate section.)
Wide-Bandgap Semiconductor Quantum Cascade Lasers… 303
Figure 13.3 RT optical gain for polar and nonpolar orientations for Zn0.95Mg0.05O and
Al0.05Ga0.95N MQW structures consisting of eight periods with total lengths of 133.2 nm.
The optical gain for phonon emission is defined by Sushil et al.8 as follows:
v
gopt ðħvÞ ¼ e p
2
jZfi j2 dðEf Ei þ ħvÞ, ð13:27Þ
câ0 nr
where nr is refractive index, Zfi is the matrix element, and Ei and Ef represent
the energy levels of the initial and final states, respectively.
Figure 13.3 shows the RT optical gain for polar and nonpolar
orientations of Zn0.95Mg0.05O and Al0.05Ga0.95N MQW structures consist-
ing of eight periods with total lengths of 133.2 nm. For a carrier density of
2 1018 cm 3 the optical gain increases near the bandgap energies
[Eg(Zn0.95Mg0.05O) ¼ 3.57 eV, and Eg(Al0.95Ga0.05N) ¼ 3.41 eV]. As shown in
Fig. 13.3, the peak optical gain of the ZnO/Zn0.95Mg0.05O MQW
with nonpolar orientation (308 cm 1) is larger than that of the ZnO/
Zn0.95Mg0.05O MQW with polar orientation (273 cm 1). Likewise, the peak
optical gain for the GaN/Al0.05Ga0.95N MQW with nonpolar orientation is
105.95 cm 1, compared to 94.2 cm 1 for that with polar orientation. This
may be attributed to the larger matrix element Z21 and calculated radiation
frequency, the latter of which is directly proportional to the optical gain, as
can be seen from Eq. (13.27). For an electric field of 100 kV/cm, the
calculated lasing frequencies are 5.756 THz and 5.57 THz in ZnO/
Zn0.95Mg0.05O, and 5.22 THz and 5.18 THz in GaN/Al0.05Ga0.95N, for
nonpolar and polar orientations, respectively. Comparing the results for the
Zn0.95Mg0.05O and Al0.05Ga0.95N MQW structures, the larger optical gain
determined for Zn0.95Mg0.05O with nonpolar orientation is attributed to the
larger optical matrix element of the zinc material compared to that of the
nitride-based structure.
304 Chapter 13
Figure 13.4 (a) Optical output power as a function of number of periods for the
Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW and the Al0.05Ga0.95N /GaN/Al0.05Ga0.95N MQW,
each biased at 100 kV/cm. (b) WPE as a function of number of periods for the
Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW and the Al0.05Ga0.95N/GaN/Al0.05Ga0.95N MQW.
higher output optical powers. It is to be noted that the peak output power is
weakly dependent on the Al or Mg mole fraction. The higher terahertz optical
emission frequency of ZnMgO compared to AlGaN can be attributed the
lower polarization field of the former, which is dependent on its mechanical
properties; decreasing the polarization field will cause a shift in the
eigenstates, thereby increasing the radiation frequency and affecting output
power. The total polarization, calculated as the sum of the spontaneous and
piezoelectric polarizations, is 0.0785 C/m2 for the ZnMgO-based material and
0.0336 C/m2 for the AlGaN-based system, assuming the same mole fraction,
x ¼ 0.05, of Mg and Al. In addition, it can be seen that the output power
increases only slightly, while the WPE shows a significant drop, with
increasing number of periods.
The corresponding radiation frequency as a function of number of periods
is shown in Fig. 13.4(c). The change in radiation frequency with increasing
total device length due to the increasing number of periods is the result of
variation in the eigenstates. From the simulation results, a decrease in
eigenstate 3 and eigenstate 2 was observed as a function of number of periods.
These changes will affect the differential between these two states, DE32, which
is proportional to the radiation frequency, as shown in the inset of Fig. 13.4(c).
Figure 13.4(d) shows the WPE curve, which is proportional to DE32, and can
be found by dividing the radiation frequency by the confinement factor. The
inset of Fig. 13.4(d) shows the corresponding optical confinement factor of
ionic and nitride-based materials, which is observed to decrease with
increasing number of periods due to its frequency dependence according to
Eq. (13.31). Changing the number of periods alters the total length of the
Figure 13.4 (cont’d.) (e) Voltage, internal, optical, and electrical efficiencies versus the
number of periods for the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW. (f) Voltage, internal,
optical, and electrical efficiencies versus the number of periods for the Al0.05Ga0.95N/GaN/
Al0.05Ga0.95N MQW.
Figure 13.5 (a) Terahertz output power versus radiation frequency for the Zn0.95Mg0.05O/
ZnO/Zn0.95Mg0.05O MQW and the Al0.05Ga0.95N/GaN/Al0.05Ga0.95N MQW. (b) WPE versus
radiation frequency for the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW and the Al0.05Ga0.95N/
GaN/Al0.05Ga0.95N MQW. (c) Output power, which is proportional to the quotient of the
radiation frequency and confinement factor. The inset shows the details of optical
confinement factor as a function of radiation frequency. (d) The WPE curve follows that of
the stage number multiplied by the radiation frequency. The inset shows the details of stage
number as a function of radiation frequency.
stage number multiplied by the radiation frequency; these results are provided
in Figs. 13.5(c) and (d), respectively. The confinement factor is shown in the
inset of Fig. 13.5(c). The corresponding number of stages is shown in the inset
of Fig. 13.5(d); different numbers of stages are necessary to achieve
frequencies of 5, 5.5 and 6 THz.
Figure 13.5(e) shows voltage, internal, optical, and electrical efficiencies
versus radiation frequency for the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW
Wide-Bandgap Semiconductor Quantum Cascade Lasers… 309
Figure 13.5 (cont’d.) (e) Voltage, internal, optical, and electrical efficiencies versus radiation
frequency for the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW with a fixed injection current of
20 mA. (f) Voltage, internal, optical, and electrical efficiencies versus radiation frequency for the
Al0.05Ga0.95N/GaN/Al0.05Ga0.95N MQW. Compared to the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O
MQW, voltage efficiency has a similar shape and a peak value of 36% at 5.5 THz.
with a fixed injection current of 20 mA. Slight and moderate changes are
observed in electrical and voltage efficiencies, respectively, the latter of which
has the greatest impact on the shape of the WPE curve. The radiation
frequencies (approximately 5, 5.5, and 6 THz) were chosen to represent values
practical for ionic and nitride-based materials. Voltage efficiency was affected
only by the number of stages N and the radiation frequency v. Radiation
frequencies of 5.5 and 6 THz may be realized by using 52 stages and 44
stages, respectively.
Figure 13.5(f) shows voltage, internal, optical, and electrical efficiencies
versus radiation frequency for the Al0.05Ga0.95N/GaN/Al0.05Ga0.95N MQW.
The voltage efficiency has a similar shape compared to that of the
Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW, with a peak value of 36% at
5.5 THz. As can be seen in this figure, the shape of the WPE curve is most
affected by the voltage efficiency. Here, 42 and 38 stages were used to achieve
radiation frequencies of 5.5 and 6 THz, respectively.
Figures 13.6(a) and (b) show the optical output power and the WPE
versus the Mg and Al mole fraction for the ZnMgO/ZnO/ZnMgO MQW and
the AlGaN/GaN/AlGaN MQW. The simulations incorporated fixed total
number of photons, charge, threshold current, threshold voltage, and the
injection current set at 20 mA.
The radiation frequency as shown in Fig. 13.6(c) increases as the Al and
Mg mole fractions are increased. This phenomenon could be attributed to the
variation of eigenstates, as discussed in Ref. 46. Increasing the mole fraction
310 Chapter 13
Figure 13.6 (a) Optical output power versus Mg and Al mole fraction for the ZnMgO/ZnO/
ZnMgO MQW and the AlGaN/GaN/AlGaN MQW. (b) WPE versus Mg and Al mole fraction
for the ZnMgO/ZnO/ZnMgO MQW and the AlGaN/GaN/AlGaN MQW. (c) shows the
radiation frequency as the Ag and Mg mole fraction. (d) Frequency divided by the optical
confinement factor, which is proportional to the terahertz power.
will also increase the total polarization of the material; hence, the structure
properties will be affected. In the simulation results, eigenstate 3 and
eigenstate 2 both decrease with increasing Al and Mg mole fraction. DE32,
which is plotted in the inset of Fig. 13.6(c), is seen to be proportional to the
radiation frequency.
The terahertz power is proportional to the radiation frequency divided by
the optical confinement factor, as shown in Fig. 13.6(d). The inset of Fig. 13.6(d)
shows the optical confinement factor as a function of the Mg and Al mole
fraction for ionic and nitride-based materials, where an increase in the
confinement factor with mole fraction is observed. Here the optical confinement
factor is frequency dependent, as shown in Eq. (13.29). The radiation frequency
Wide-Bandgap Semiconductor Quantum Cascade Lasers… 311
Figure 13.6 (cont’d.) (e) Voltage, internal, optical, and electrical efficiencies versus Mg
mole fraction for the ZnMgO/ZnO/ZnMgO MQW. (f) Voltage, internal, optical, and electrical
efficiencies versus Al mole fraction for the AlGaN/GaN/AlGaN MQW. As the Al mole fraction
increases, the radiation frequency likewise increases.
increases with the mole fraction, as seen from Eq. (13.29), resulting in an
increasing optical confinement factor as the mole fraction increases.
Figure 13.6(e) shows voltage, internal, optical, and electrical efficiencies
versus Mg mole fraction for the ZnMgO/ZnO/ZnMgO MQW. Here, we
observe that the WPE varies with the voltage efficiency due to the dependence
of radiation frequency on increasing Mg mole fraction. Figure 13.6(f) shows
voltage, internal, optical, and electrical efficiencies versus Al mole fraction for
the AlGaN/GaN/AlGaN MQW. As the Al mole fraction increases, the
radiation frequency likewise increases. Similar behavior for WPE with
changing radiation frequency is observed, where the peak value is found at a
mole fraction equivalent to 0.15.
In this section the theoretical calculated output power and WPE of polar
ZnO/ZnMgO heterostructures were thus compared, taking into account the
presence of spontaneous and piezoelectric polarizations as well as nonpolar
heterostructures based on fundamental device physics.
Figure 13.7 (a) RT terahertz QCL output power versus injection current for Al0.05Ga0.95N/
GaN/Al0.05Ga0.95N and Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW structures having eight
periods (reprinted from Ref. 22). (b) Optical output power as a function of number of periods
for the Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O MQW and the Al0.05Ga0.95N/GaN/Al0.05Ga0.95N
MQW for polar and nonpolar planes, each biased at 100 kV/cm (reprinted from Ref. 22).
(c) Corresponding radiation frequency as a function of number of periods of the ionic and
nitride-based materials with polar and nonpolar orientations (reprinted from Ref. 22).
(d) Corresponding optical confinement factor of the ionic and nitride-based materials with
polar and nonpolar orientations (reprinted from Ref. 22). (See color plate section.)
and 2.41 mW (polar)]. The lower calculated terahertz frequency for the
Al0.05Ga0.95N/GaN/Al0.05Ga0.95N MQW structure [vZnMgO ¼ 5.756 THz (non-
polar), 5.57 THz (polar), vAlGaN ¼ 5.22 THz (nonpolar) and 5.18 THz (polar)]
is due to its larger calculated refractive index, while the higher terahertz output
power for Zn0.95Mg0.05O/ZnO/Zn0.95Mg0.05O is attributed to its lower
refractive index nZnMgO
r ¼ 1.65 compared to nAlGaN
r ¼ 2.49, since, as is evident
from Eq. (13.29), a lower refractive index results in higher peak optical output
power.
Wide-Bandgap Semiconductor Quantum Cascade Lasers… 313
13.4 Conclusion
In conclusion, we described the calculation of the subband structure,
absorption coefficient, optical gain, optical output power, and WPE of
GaN/AlxGa1 xN and ZnO/Zn1 xMgxO quantum structures using an envelope
function framework that takes into account the polarization fields in the
quantum wells. The parameters of the potential barriers of the two-barrier
structure were determined, and the transmission coefficients of this structure
314 Chapter 13
were calculated both with and without the presence of electric fields. The
number of levels in the quantum well decreases under high-electric-field action
due to the Stark effect. The transmission coefficient maxima shift to lower
energy, and their amplitudes change under exposure to external bias voltage,
which is due to an increase in the potential asymmetry under the action of the
external field. The modification of the band energies due to polarization and
its effect on the generated terahertz radiation has been explored. Decreasing
the polarization field, which is higher in ZnMgO compared to AlGaN
resulting in a potentially higher THz optical emission frequency for the
former, causes a shift in the eigenstates, thereby affecting the radiation
frequency and output power. In addition, the relatively high photon energy of
Zn1 xMgxO results in higher voltage efficiency and thus greater overall WPE
than that for AlxGa1 xN systems. The sensitivity of generated terahertz
radiation to varying structural dimensions and Al and Mg mole fraction
allows the design of QCLs operating at specific frequencies in the range of 2 to
15 THz. Tunability of the generated terahertz radiation can also be achieved
by controlling the applied bias.
References
1. M. S. Shur and M. Asif Khan, “GaN and AlGaN Ultraviolet Detectors,” in
Semiconductors and Semimetals, Vol. 57, J. I. Pankove and T. D. Moustakas,
Eds., Academic Press, San Diego, pp. 407–439 (1998).
2. W. Knapp, F. Teppe, Y. Meziani, N. Dyakonova, J. Lusakowski, F. Boeuf,
T. Skotnicki, D. Maude, S. Rumyantsev, and M. S. Shur, “Plasma wave
detection of sub-terahertz and terahertz radiation by silicon field-effect
transistors,” Appl. Phys. Lett. 85, 675–677 (2004).
3. S. Fathololoumi, E. Dupont, C. W. I. Chan, Z. R. Wasilewski, S. R. Laframboise,
D. Ban, A. Mátyás, C. Jirauschek, Q. Hu, and H. C. Liu, “Terahertz quantum
cascade lasers operating up to 200 K with optimized oscillator strength and
improved injection tunneling,” Opt. Express 20, 3866 (2012).
4. A. Wade, G. Fedorov, D. Smirnov, S. Kumar, B.S Williams, Q. Hu, and
J. L. Reno, “Magnetic field controlled THz quantum cascade laser
operating up to 225K,” Nature Photon. 3, 41–45 (2009).
5. R. Kazarinov and R. Suris, “Possibility of the amplification of
electromagnetic waves in a semiconductor with a superlattice,” Sov.
Phys. Semicond. 5, 707–709 (1971).
6. J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, and Y. Cho,
“Quantum cascade laser,” Science 264, 553–556 (1994).
7. R. Köhler, A. Tredicucci, F. Beltram, H. E. Beere, E. H. Linfield, A. G. Davies,
D. A. Ritchie, R. C. Iotti, and F. Rossi, “Terahertz semiconductor-
heterostructure laser,” Nature 417, 156–159 (2002).
Wide-Bandgap Semiconductor Quantum Cascade Lasers… 315
14.1 Introduction
14.2 The HgCdTe Era
14.3 Alternative Material Systems
14.3.1 PbSnTe
14.3.2 InSb and InGaAs
14.3.3 GaAs/AlGaAs QW SLs
14.3.4 InAs/GaInSb strained layer SLs
14.3.5 Hg based alternatives to HgCdTe
14.4 Readiness Level of LWIR Detector Technologies
14.5 Summary
References
14.1 Introduction
Observing the history of the development of the infrared (IR) detector technology
after World War II, many materials have been investigated. A simple theorem,
after Norton,1 can be stated: “All physical phenomena in the range of
about 0.1–1 eV will be proposed for IR detectors.” Among these effects are:
thermoelectric power (thermocouples), change in electrical conductivity
(bolometers), gas expansion (Golay cell), pyroelectricity (pyroelectric detectors),
photon drag, Josephson effect [Josephson junctions, superconducting quantum
interference devices (SQUIDs)], internal emission (PtSi Schottky barriers),
fundamental absorption (intrinsic photodetectors), impurity absorption (extrin-
sic photodetectors), low-dimensional solids [superlattice (SL), quantum well
(QW), and quantum dot (QD) detectors], different type of phase transitions, etc.
323
324 Chapter 14
Figure 14.1 History of the development of IR detectors and systems. Three generations of
systems can be considered for principal military and civilian applications: 1st Gen. (scanning
systems), 2nd Gen. (electronically scanned staring systems) and 3rd Gen. (multicolor
functionality and other on-chip functions). Material systems marked in bold indicate the main
HgCdTe competitors.
Figure 14.2 The discoverers of HgCdTe ternary alloy (reprinted from Ref. 4).
ternary alloy in a 1957 patent.4 They were joined by E. H. Putley in the first
publication.3
Discovery of the variable-bandgap HgCdTe alloy has provided an
unprecedented degree of freedom in IR detector design. The fundamental
properties of HgCdTe (high optical-absorption coefficient, high electron
mobility, and low thermal-generation rate), together with the capability for
bandgap engineering, make this alloy system almost ideal for a wide range of
IR detectors. The bandgap energy tunability results in IR detector
applications that span the short wavelength (SWIR: 1–3 mm), middle
wavelength (MWIR: 3–5 mm), long wavelength (LWIR: 8–14 mm), and very
long wavelength (VLWIR: 14–30 mm) ranges.
The first paper by Lawson et al.3 reported both photoconductive and
photovoltaic HgCdTe response at wavelengths extending out to 12 mm. Soon
thereafter, working under a U.S. Air Force contract with the objective of
devising an 8- to 12-mm background-limited semiconductor IR detector that
would operate at temperatures as high as 77 K, the group lead by Kruse at the
Honeywell Corporate Research Center in Hopkins, Minnesota developed a
modified Bridgman crystal growth technique for HgCdTe. They soon
reported both photoconductive and photovoltaic detection in rudimentary
HgCdTe devices.5 The parallel programs were carried out at Texas
Instruments and Santa Barbara Research Center.
Figure 14.4. A time line of the evolution of HgCdTe IR detectors and key developments in
the process technologies that made them possible [reprinted from Ref. 6 with permission
from Association of Polish Electrical Engineers (SEP); © 2002 COSiW SEP].
Figure 14.5 A 180-element common module FPA mounted on a dewar stem (reprinted
from Ref. 12).
328 Chapter 14
Figure 14.7 Progression of ROIC format at Raytheon Vision Systems over time (reprinted
from Ref. 16).
A general trend has been to reduce pixel sizes, and this trend is expected to
continue. Pixel sizes as small as 15 mm have been demonstrated in hybrid
systems. Systems operating at shorter wavelengths are more likely to benefit from
small pixel sizes because of the smaller diffraction-limited spot size. Diffraction-
limited optics with low f-numbers (e.g., f/1) could benefit from pixels on the order
of one wavelength across, about 10 mm in the LWIR. Oversampling the
diffractive spot may provide some additional resolution for smaller pixels, but
this saturates quickly as the pixel size is decreased. Pixel reduction is also
mandatory for cost reduction of a system (reduction of the optics diameter,
dewar size and weight, and power, together an increase in reliability).
SWIR, MWIR, and LWIR electronically scanned HgCdTe arrays with a
CMOS multiplexer are commercially available from several manufactures.
Most manufactures produce their own multiplexer designs because these often
need to be tailored to the applications. Figure 14.8 shows an example of large
HgCdTe FPAs.17,18 While the size of individual arrays continues to grow, the
very large FPAs required for many space missions are fabricated by mosaic
assembly of a large number of individual arrays. An example of a large
mosaic developed by Teledyne Imaging Sensors is a 147-megapixel FPA that
comprises 35 arrays, each with 2048 2048 pixels.18
Third-generation HgCdTe systems are now being developed. These
systems provide enhanced capabilities such as larger number of pixels, higher
frame rates, better thermal resolution, as well as multicolor functionality and
HgCdTe versus Other Material Systems: A Historical Look 331
Figure 14.8 Large HgCdTe FPAs: (a) A mosaic of four Hawaii-2RG-18s (4096 4096
pixels, 18-mm pitch). (b) Hawaii-4RG-10 (4096 4096 pixels, 10-mm pitch) as is being used
for astronomy observations (reprinted from Ref. 18). (c) 16 2048 2048 HgCdTe arrays
assembled for the VISTA telescope (reprinted from Ref. 17 with permission).
14.3.1 PbSnTe
One of these was PbSnTe, which was vigorously pursued in parallel with HgCdTe
in the late 60s and early 70s.19 PbSnTe was comparatively easy to grow, and good-
quality LWIR photodiodes and lasers were readily demonstrated.
332 Chapter 14
Figure 14.9 Liquidus and solidus lines in HgTe-CdTe, HgTe-ZnTe, and PbTe-SnTe
pseudo-binary systems. (See color plate section.)
Figure 14.9 shows the liquidus and solidus lines in three pseudo-binary
systems. In comparison with PbTe-SnTe, the wide separation between the
HgCdTe liquidus and solidus leads to marked segregation between CdTe and
HgTe that is instrumental in the development of the bulk growth techniques of
this system. In addition to solidus–liquidus separation, high-Hg partial pressure
is also influential both during growth and postgrowth heat treatments.
In their review paper published in 1974,20 Harman and Melngailis, both
involved in studies of HgCdTe and PbSnTe ternary alloys in Massachusetts
Institute of Technology, wrote:
“In comparing the two materials we anticipate that Pb1 xSnxTe will
be more widely used in the future for detection of blackbody radiation
in the 8- to 14-mm region because crystal growth techniques for this
alloy are potentially cheaper and adaptable to mass production. In
addition, Pb1 xSnxTe appears to be more stable and less likely to
degrade at elevated temperatures than Hg1 xCdxTe. However, for
heterodyne detection and other high-speed applications, Hg1 xCdxTe
can be expected to be more useful at frequencies in the gigahertz range
because of the inherent advantage of a lower dielectric constant.”
Several years later, this opinion was completely changed. In the late
1970s, development of IV-VI alloy photodiodes was discontinued because
the chalcogenides suffered two significant drawbacks. The first was a high
dielectric constant that resulted in high diode capacitance and therefore
HgCdTe versus Other Material Systems: A Historical Look 333
Figure 14.10 Linear CTE of PbTe, InSb, HgTe, and Si versus temperature (after Ref. 22).
limited frequency response (for PbSnTe the observed values of the static
dielectric constant have been widely distributed from 400 to 5800, and at the
same temperature these values have been scattered in the range up to one order
of magnitude21). For scanning systems under development at that time, this was
a serious limitation. However, for staring imaging systems under development
today using 2D arrays, this would not be as significant of an issue.
The second drawback to IV-VI compounds is their very high coefficients
of thermal expansion (CTEs).22 This limited their applicability in hybrid
configurations with silicon multiplexers. Differences in CTE between the
readout and detector structure can lead to failure of the indium bonds after
repeated thermal cycling from room temperature to the cryogenic temperature
of operation.23 Figure 14.10 shows the dependence of the CTE of PbTe, InSb,
HgTe, and Si on temperature. At room temperature, the CTE of HgTe
and CdTe is about 5 10 6 K 1, while that of PbSnTe is in the range of 20
10 6 K 1. This results in much greater CTE mismatch with silicon (CTE of
about 3 10 6 K 1).
spectral range. Having lower dark current and noise than indirect-bandgap
germanium—the competing near-IR material—the material is addressing both
entrenched applications including lightwave communication systems, low-light-
level night vision, and new applications such as remote sensing, eye-safe range
finding, and process control. Due to the similar band structures of InGaAs and
HgCdTe ternary alloys, the ultimate fundamental performance of both types of
photodiodes is similar in the wavelength range 1.5 < l < 3.7 mm.25 InGaAs
photodiodes have shown high device performance close to theoretical limits for
material whose composition is nearly matched to that of InP (ffi1.7-mm cutoff
wavelength) and InAs (ffi3.6-mm cutoff wavelength). However, the performance
of InGaAs photodiodes decreases rapidly at intermediate wavelengths due to
substrate-lattice-mismatch-induced defects.
Figure 14.12 Thermal generation current versus temperature for GaAs/AlGaAs MQWs
and HgCdTe alloy detectors at lc ¼ 8.3 mm and 10 mm. The assumed effective quantum
efficiencies are h ¼ 12.5% and 70% for GaAs/AlGaAs and HgCdTe detectors, respectively
(after Ref. 26).
Table 14.1 Essential properties of LWIR HgCdTe type-II SL photodiodes and QWIPs at 77
K (R0A product is the resistance-area product).
FOV 0)
HgCdTe versus Other Material Systems: A Historical Look 337
antimonide alloys.31 The best results have been obtained using inductively
coupled plasma dry etching and polyimide passivation.32
Optimization of the SL photodiode architectures is still an open area for
research. Additional design modifications (e.g., in so-called W-structure and
M-structure photodiodes) dramatically improve the photodiode performance.
The modification of the structures confines the electron wavefunctions,
reducing the tunneling probability, increasing the electron effective mass, and
effectively decreasing the dark current. The introducion of unipolar barriers in
various designs based on type-II SLs further drastically changes the
architecture of IR detectors.30 Unipolar barriers are used to impede the flow
of majority-carrier dark current (electrons or holes). However, a key feature of
the devices is a pair of complementary barriers, namely, an electron barrier
and a hole barrier formed at different depths in the growth sequence. Such a
structure is known as a complementary-barrier IR device (CBIRD) and was
invented by Ting and others at Jet Propulsion Lab (JPL).33 Figure 14.13(a)
shows a bandgap diagram at equilibrium and a schematic device structure of
the CBIRD. An electron barrier (eB) appears in the conduction band, and a
hole barrier (hB) in the valence band toward the left. The two barriers
complement one another to impede the flow of dark current. The substrate is
toward the right. The absorber region, where the bandgap is smallest, is
p-type, and the top contact region at the left is n-type, making an n-on-p
polarity for the detector element. In sequence from the top (left), the first three
regions are composed of SL material: the n-type cap, the p-type absorber,
and the p-type eB. The highly doped InAsSb layer below the eB is an alloy.
Further to the right (not shown) are a GaSb buffer layer and the GaSb
substrate.
Figure 14.13(b) plots the RAeff values for a set of selected antimonide SL
devices made at JPL. For reference, the HgCdTe “Rule 07” trend line is also
Figure 14.13 The complementary-barrier IR detector (CBIRD). (a) Zero-bias band diagram
and a schematic device structure. (b) The 77 K effective resistance-area product plotted against
the detector cutoff wavelength for a number of LWIR InAs/GaSb SL detectors (reprinted from
Ref. 33). (See color plate section.)
HgCdTe versus Other Material Systems: A Historical Look 339
shown. Of note is the steady performance gains achieved in the past few
years. In particular, the introduction of device designs containing
unipolar barriers has taken the LWIR SL detector performance close to
the Rule 07 trend line (Rule 07 is a simple empirical relationship that
conveniently estimates state-of-the-art HgCdTe photodiode dark current
performance, introduced by Tennant et al. in 2007 34). The barriers prove
to be very effective in suppressing the dark current. In the case of a
detector having a cutoff wavelength of 9.24 mm, the value of R0A > 105
Vcm2 at 78 K was measured, as compared with about 100 Vcm2 for an
InAs/GaSb homojunction of the same cutoff. For good photoresponse,
the device must be biased to typically 200 mV; the estimated internal
quantum efficiency is greater than 50%, while the RA eff remains above
104 Vcm2.
The predicted thermally limited detectivities of type-II SLSs are larger
than those for HgCdTe. Figure 14.14 compares the calculated detectivity of
type-II SLSs and p-on-n HgCdTe photodiodes as a function of wavelength
and temperature of operation, with the experimental data of type-II
detectors operated at 78 K. The solid lines are theoretical thermal limited
detectivities for HgCdTe photodiodes, calculated using a 1D model that
assumes that the diffusion current from the narrower bandgap n-side is
dominant, and minority carrier recombination is determined via Auger and
radiative processes.
Type-II SLS is promising material for high-operation-temperature (HOT)
IR detectors.36 Recently, high-performance HOT detectors have been
fabricated using the interband cascade design of the MWIR detector.
It should be mentioned that stacked photovoltaic cell detectors connected in
Figure 14.14 The predicted detectivity of type-II and p-on-n HgCdTe photodiodes as
functions of wavelength and temperature. Experimental data at 78 K are marked for
SLSs (reprinted from Ref. 35). (See color plate section.)
340 Chapter 14
series were fabricated with HgCdTe by VIGO System S.A. (Poland), but with
low producibility due to interface instability (weak Hg chemical bonding in
the materials).
In the last two years the first megapixel MW and LW type-II superlattice
FPAs have been demonstrated with excellent imaging.37 39 At about 78 K, an
NEDT (noise-equivalent differential temperature) value below 20 mK was
presented. Type-II InAs/GaSb SLs have also emerged as a material for third-
generation IR detectors. Both dual-band MW/MW40,41 and MW/LW42 FPAs
have been demonstrated.
The above-described very promising results confirm that the antimo-
nide SL technology is now a direct competitor to HgCdTe IR detector
technology. In order to establish U.S. industry capability of producing
high-quality type-II SL FPAs, a new four-year program was established
in 2009.43
Table 14.2 Comparison of existing state-of-the-art device systems for LWIR detectors
(from Ref. 45).
14.5 Summary
Over the past fifty years, HgCdTe ternary alloy has successfully fought off
major challenges from extrinsic silicon and lead-tin telluride devices, but
despite this, it has more competitors today than ever before. These include
Schottky barriers on silicon, AlGaAs multiple QWs, InAs/GaSb strained-
layer SLs, high-temperature superconductors, and, especially, two types of
thermal detectors: pyroelectric detectors and silicon bolometers. However, it is
interesting that none of these competitors can compete in terms of
fundamental properties. They may promise to be more manufacturable but
can never provide higher performance or, with the exception of thermal
detectors, operate at higher or even comparable temperatures.
It is well known that the InAs/GaSb family of the type-II SLs is the only
one known IR detector material having a theoretically predicted higher
performance than HgCdTe. However, the theoretical promise of even higher
detectivity than that of HgCdTe has not been realized because of the presence
of the Shockley–Read–Hall recombination characterized by a relatively short
carrier lifetime.
342 Chapter 14
References
1. P. R. Norton, “Infrared detectors in the next millennium,” Proc. SPIE
3698, 652–665 (1999) [doi: 10.1117/12.354568].
2. A. Rogalski, Infrared Detectors, Second ed., CRC Press, Boca Raton
(2010).
3. W. D. Lawson, S. Nielson, E. H. Putley, and A. S. Young, “Preparation
and properties of HgTe and mixed crystals of HgTe-CdTe,” J. Phys.
Chem. Solids 9, 325–329 (1959).
4. T. Elliot, “Recollections of MCT work in the UK at Malvern and
Southampton,” Proc. SPIE 7298, 72982M (2009) [doi: 10.1117/
12.820214].
5. P. W. Kruse, M. D. Blue, J. H. Garfunkel, and W. D. Saur, “Long
wavelength photoeffects in mercury selenide, mercury telluride and
mercury telluride-cadmium telluride, Infrared Phys. 2, 53–60 (1962).
6. P. Norton, “HgCdTe infrared detectors, Opto-Electron. Rev. 10, 159–174
(2002).
7. C. Verie and R. Granger, “Propriétés de junctions p-n d’alliages
CdxHg1 xTe,” C. T. Acad. Sc. 261, 3349–3352 (1965).
8. G. C. Verie and M. Sirieix, “Gigahertz cutoff frequency capabilities of
CdHgTe photovoltaic detectors at 10.6 mm,” IEEE J. Quant. Electr. 8,
180–184 (1972).
9. B. E. Bartlett, D. E. Charlton, W. E. Dunn, P. C. Ellen, M. D. Jenner,
and M. H. Jervis, “Background limited photoconductive detectors for use
in the 8–14 micron atmospheric window,” Infrared Phys. 9, 35–36 (1969).
10. M. A. Kinch, S. R. Borrello, and A. Simmons, “0.1 eV HgCdTe
photoconductive detector performance,” Infrared Phys. 17, 127–135
(1977).
11. A. Rogalski, “History of infrared detectors,” Opto-Electron. Rev. 20, 279–308
(2012).
HgCdTe versus Other Material Systems: A Historical Look 343
new ternary alloy detectors such as lead salts, InAsSb, HgZnTe, and
HgMnTe. In 2004, he was elected as a corresponding member of the Polish
Academy of Sciences.
Professor Rogalski’s scientific achievements include determining the
fundamental physical parameters of InAsSb, HgZnTe, HgMnTe, and lead
salts; estimating the ultimate performance of ternary alloy detectors; elaborating
on studies of high-quality PbSnTe, HgZnTe, and HgCdTe photodiodes operated
in the 3- to 5-mm and 8- to 12-mm spectral ranges; and conducting comparative
studies of the performance limitation of HgCdTe photodiodes versus other types
of photon detectors (especially QWIP and QDIP IR detectors).
Professor Rogalski has given more than 50 invited plenary talks at
international conferences. He is author and coauthor of about 230 scientific
papers, 12 books, and 26 monographic papers (book chapters). He is a fellow
of SPIE, vice president of the Polish Optoelectronic Committee, vice president
of the Electronic and Telecommunication Division at the Polish Academy of
Sciences, Editor-in-Chief of the journal Opto-Electronics Review, Deputy
Editor-in-Chief of the Bulletin of the Polish Academy of Sciences: Technical
Sciences, and a member of the editorial boards of Journal of Infrared and
Millimeter Waves, International Review of Physics, and Photonics Letters of
Poland. He is an active member of the international technical community and
is a chair and cochair, organizer, and member of scientific committees of
many national and international conferences on optoelectronic devices and
material sciences.
Chapter 15
Type-II Superlattices:
Status and Trends
Elena A. Plis and Sanjay Krishna
Center for High Technology Materials, Department of Electrical and Computing
Engineering, University of New Mexico, Albuquerque, New Mexico, USA
15.1 Introduction
15.2 Limitations of T2SLS Technology
15.2.1 Short carrier lifetime
15.2.2 Passivation
15.2.3 Heterostructure engineering
15.2.4 Nonuniformity and reproducibility issues
15.2.5 Spectral crosstalk in multicolor T2SLS imagers
15.3 Proposed Solutions
15.3.1 Ga free type II InAs/InAsSb superlattice detectors
15.3.2 Interband cascade infrared photodetector (ICIP) architecture
15.3.3 InAs/GaSb T2SLS MWIR detectors grown on (111) GaSb substrates
15.4 Summary
Acknowledgments
References
15.1 Introduction
InAs/(In,Ga)Sb type-II strained-layer superlattices (T2SLSs) were first
proposed as an infrared (IR) sensing material in the 1980s by Sai-Halasz,
Tsu, and Esaki.1,2 IR detectors based on InAs/(In,Ga)Sb T2SLSs have been
under investigation ever since they were suggested several decades ago by
Smith and Mailhiot.3 In 1990, Chow and coworkers reported the first
Ga1 x InxSb/InAs T2SL material with high structural quality, longwave IR
(LWIR) photoresponse, and LWIR photoluminescence.4 While theoretical
predictions of detector performance seem to favor the InAs/InGaSb system
347
348 Chapter 15
due to the additional strain provided by the InGaSb layer, the majority of the
research in the past five years has been focused on the binary InAs/GaSb
system. This is attributed to the complexity of structures grown with the large
mole fraction of In.
The InAs/GaSb T2SLS consists of alternating layers of nanoscale
materials whose thicknesses vary from 4 to 20 monolayers (MLs). The
overlap of electron (hole) wave functions between adjacent InAs (GaSb)
layers results in the formation of electron (hole) minibands in the conduction
(valence) band. Optical transitions between holes localized in GaSb layers and
electrons confined in InAs layers are employed in the IR detection process.
Thus, using two mid-bandgap semiconductors, devices can be fabricated with
operating wavelengths anywhere between 3 and 32 mm.5 7
Tunneling currents in T2SLS are lower than in mercury-cadmium-telluride
(MCT) detectors of the same bandgap due to larger electron effective mass.3
Auger recombination is supressed due to the large splitting between heavy-hole
and light-hole valence subbands.8,9 Moreover, the T2SLSs are less sensitive to
the bandgap variations due to compositional nonuniformities than the MCT
alloys with the same bandgap.10 In contrast with quantum well infrared
photodetectors (QWIPs), normal incidence absorption is permitted in T2SLSs,
contributing to high conversion quantum efficiency. Two factors provide
T2SLSs with technological advantages over competing technologies: commer-
cially available low defect density substrates and a high degree of growth and
processing uniformity over a large area. Focal plane arrays (FPAs) of sizes up to
1024 1024 have been demonstrated using the T2SLS material system,
highlighting the potential of this technology.11 Thorough comparisons between
MCT, InSb, QWIP, and T2SLS technologies can be found in the literature.12 14
Some very exciting work on everything from computational modeling of
superlattices to fabrication of large-area FPAs is being conducted at various
institutions. For example, Grein and Flatte have undertaken extensive
theoretical modeling of the bandstructure of superlattices.15 17 Bandara
et al.18 have modeled the effect of doping on the Shockley–Read–Hall (SRH)
lifetime and the dark current; DeWames and Pellegrino19 have performed
extensive modeling to extract the SRH lifetime from dark-current measurements.
Belenky20 and Connelly21 have measured time-resolved photoluminescence (PL)
on T2SLS structures. Low-dark-current architectures with unipolar barriers
such as M-structure,22 complementary-barrier infrared detector (CBIRD),23
W-structure,24,25 N-structure,26 nBn,27,28 pBiBn,6 etc. have been designed and
fabricated into single-pixel detectors and FPAs at university laboratories
[Northwestern University, Arizona State University, University of Oklahoma,
University of Illinois, Georgia Tech University, Bilkent University (Turkey),
University of New Mexico], federal laboratories [JPL, NRL, ARL, NVESD,
SNL), and industrial laboratories (Raytheon, Teledyne Imaging Systems,
Hughes Research Laboratories, QmagiQ LLC, etc.).
Type-II Superlattices: Status and Trends 349
Figure 15.1 Dark-current density of T2SLS detectors compared with Rule 07. Abbreviations
for the different institution working on T2SL detectors: Fraunhofer-Institut (IAF), JPL, Naval
Research Laboratory (NRL), Northwestern University (NWU), Raytheon Vision Systems (RVS),
University of California, Santa Barbara (UCSB), Columbia University (Columbia), University of
Illinois, Urbana-Champaign (UIUC), and University of New Mexico (UNM). (Adapted with
modifications and permission from Ref. 29; © 2012, Springer.) (See color plate section.)
350 Chapter 15
approaches midgap. Thus, the most effective SRH centers are those located
near the middle of the bandgap.31 Analysis of the defect formation energy of
native defects dependent on the location of the Fermi level stabilization
energy has been performed by Walukiewicz,32 who reported that, in bulk
GaAs and GaSb, the stabilized Fermi level is located near either the valence
band or the midgap, whereas in bulk InAs the stabilized Fermi level is located
above the conduction-band edge. From this observation, the midgap trap
levels in GaAs and GaSb are available for SRH recombination, whereas in
InAs they are inactive for the SRH process, suggesting a longer carrier lifetime
in bulk InAs than in bulk GaSb and GaAs materials. Experimentally
measured values of carrier lifetimes yielded 325 ns for bulk InAs and
100 ns for bulk GaSb, thereby confirming the initial observation.33 It may
then be hypothesized that native defects associated with GaSb are responsible
for the SRH-limited minority-carrier lifetimes observed in InAs/GaSb T2SLS.
15.2.2 Passivation
Scaling of the lateral dimensions of a T2SLS detector (e.g., typical mesa
dimensions of individual FPA pixels are 20 mm 20 mm) makes FPA perfor-
mance strongly dependent on surface effects due to a large pixel surface/
volume ratio. Thus methods for elimination of surface currents need to be
developed. Passivation is also very important because surface and interface
recombination is much faster than bulk carrier recombination. Therefore,
lifetime is affected by passivation. Concentrated efforts of numerous research
groups have been devoted to the development of the surface passivation for
T2SLS material system.
Passivation methods of T2SLS detectors may be categorized into two
directions: (1) encapsulation of etched detector sidewalls with thick layers of
dielectrics, organic materials (polyimide and various photoresists), or wider-
band-gap III-V material, and (2) chalcogenide passivation, that is, saturation
of unsatisfied bonds on semiconductor surfaces by sulphur atoms. A review of
passivation techniques for T2SLS detectors operating in MWIR, LWIR, and
VLWIR spectral regions may be found in the work of Plis et al.34 Here we
briefly describe available passivation techniques and their limitations.
Dielectric passivation, though shown to be effective,35 37 presents the
challenge of developing high-quality dielectrics with low fixed and interfacial
charge densities at process temperatures substantially lower than the InAs/
GaSb T2SLS growth temperature to prevent T2SLS period intermixing.
Moreover, dielectric passivation layers alter the band bending presented at the
mesa sidewalls caused by the abrupt termination of the periodic crystal
structure. This band bending induces accumulation or type inversion of
charge, which results in surface tunneling currents along sidewalls. As was
shown by Delaunay et al.,38 narrow-bandgap devices (with bandgap of
120 meV or lower) are more susceptible to the formation of charge conduction
Type-II Superlattices: Status and Trends 351
channels along the sidewalls. Native fixed charges present in the dielectric
passivation layer (e.g., SiO2 ) can either improve or deteriorate the device
performance.39 Consequently, the dielectric passivation may not passivate
low-bandgap materials as effectively as high-bandgap materials. Despite all of
the potential drawbacks, the ability to establish the flat-band condition at the
T2SLS–dielectric interface along with compatibility with the T2SLS FPA
fabrication procedure makes dielectric passivation very attractive for
passivation of LWIR T2SLS detectors.
A method of band-bending control at the T2SLS–SiO2 interface through
the application of a voltage along the device sidewalls has been proposed by
Chen et al.39 By applying a negative bias voltage along MWIR (l50%cut off ¼
4.7 mm at 110 K) T2SLS device sidewalls, the accumulated electrons repel
from the insulator–T2SLS interface, thus establishing flat-band conditions
that suppress the leakage current.
Passivation with organic materials (polyimide or various photoresists) is
attractive due to the simplicity of its integration into the T2SLS detector
fabrication procedure. Usually, organic passivants are spin-coated onto a
detector at room temperature and thicknesses varying from 0.2 to 100 mm,
based on the spin speed. Passivation of MWIR and LWIR T2SLS detectors
with SU-8 photoresist,40 42 polyimide,43 and AZ-1518 photoresist44 has been
reported.
Surface currents can be suppressed by reducing the exposure of narrow-
gap materials to the environment, e.g., as a result of encapsulation of etched
sidewalls with a wide-bandgap material or a “shallow etch” technique that
isolates the neighboring devices and terminates within a wider-bandgap layer.
The buried architecture approach,25,45 also referred to as the shallow-etch
mesa isolation (SEMI) technique, restricts the exposed surfaces to the wide-
bandgap layers, thus suppressing surface recombination and the ability of
surface electric fields to produce conducting channels that would bypass the
junction. In addition, it reduces the junction area to the smallest area of the
pixel defined by the diffusion length of the minority carriers sufficient to
collect photogenerated carriers, thus lowering the bulk junction current.
However, if the lateral diffusion lengths are larger than the distance between
neighboring pixels in the FPA (which is typically on the order of several
microns), crosstalk between the FPA elements can be encountered, degrading
the FPA image resolution.46 Since no crosstalk has been registered for LWIR
FPA mini-arrays (256 256) fabricated with the SEMI technique, this
approach may provide a solution to passivation difficulties in T2SLS detectors
operating in the LWIR region.
Molecular beam epitaxy (MBE) re-growth of a lattice-matched wide-
bandgap III-V semiconductor layer on top of the exposed mesa sidewalls
effectively passivates narrow-bandgap T2SLS detectors in the LWIR
wavelength range.47 Mesa sidewall encapsulation with GaSb has been
352 Chapter 15
Figure 15.4 Calculated energy band diagram of a cascade detector demonstrating flow of
photogenerated carriers and band alignments of different segments. The band offsets of
different layers have been obtained by an empirical pseudo-potential method. (Reprinted from
Ref. 76 with permission; © 2012, American Institute of Physics.) (See color plate section.)
the right, while the holes move to the left. Electrons then relax in the graded SL
transport region 3, while holes tunnel through region 2. The separation between
the quantized energy level in the GaSb quantum well of region 2 and the valence
band in region 1 was designed to be equal to the longitudinal optical (LO)
phonon energy in AlSb to make the tunneling of holes a phonon-assisted
process. The relaxation region 3 and the interband tunneling region 2 also act as
hole and electron barriers, respectively, to block the flow of dark carriers from
one cascade stage into the other. The junction between regions 2 and 3 acts as a
recombination site for electrons and holes that leads to photoconductive (PC)
gain being equal to 1/N, where N is the number of cascade stages. Reduction in
photoconductive gain and zero-bias operation are necessary for FPAs to obtain
a low noise-equivalent differential temperature (NEDT) at a high operating
temperature as these limit the rate at which the charge capacitor is filled. In the
current design, the total thickness of the absorber is 1 mm, and the absorption QE
can be increased by increasing the number of stages. However, the conversion
QE is lower than that of the absorption QE by a factor of N.
Several key features in the ICIP structure ensure improved device
performance. In particular, the interband tunneling region is designed so that
the difference in hole energies is close to the LO phonon energy in AlSb to
facilitate phonon-assisted tunneling. Carrier transport through the graded
InAs/AlSb T2SLS relaxation region is very efficient. Finally, the bottom
N-contact layer has a wider-bandgap 12.5 ML InAs/5 ML AlSb SL, which
can suppress dark current by reducing thermionic generation and tunneling
currents. It should be noted, however, that the cascade physics in T2SLS
detectors is still not very well understood; it’s definitely advantageous in
358 Chapter 15
Figure 15.5 Measured photocurrent signal of MWIR T2SLS ICIP detector as a function of
temperature. (Reprinted from Ref. 76 with permission; © 2012, American Institute of Physics.)
(See color plate section.)
Table 15.1 Performance of T2SLS MWIR detectors grown on GaSb (111)B and
(100) substrates.
15.4 Summary
The presented chapter provides a review of the current status of IR detectors
based on an InAs/GaSb T2SLS. Despite the numerous advantages that
T2SLSs offer over current detection technologies, such as MCT, InSb, and
QWIP-based detectors, the promise of superior performance of T2SLS
detectors has not been yet realized. We point out the essential scientific and
technological challenges that need to be overcome in order for T2SLS to be
the technology of choice for high-performance imaging systems. These are
short carrier lifetime, passivation, nonuniformity/reproducibility issues, and
spectral crosstalk in multicolor T2SLS imagers. Further, we discussed a few
possible approaches investigated by our group to address some of these
limitations. In particular, we suggest the use of type-II Ga-free SL to
surmount carrier lifetime issues imposed by the GaSb layer in InAs/GaSb
T2SLS and the ICIP architecture and the growth of T2SLS structures on (111)
GaSb substrates for the realization of MWIR HOT devices.
Acknowledgments
Support from AFOSR FA9550-10-1-0113 and AFRL FA9453-12-1-0336 is
acknowledged.
References
1. G. A. Sai-Halasz, R. Tsu, and L. Esaki, “A new semiconductor
superlattice,” Appl. Phys. Lett. 30, 651–653 (1977).
2. L. Esaki, “InAs-GaSb superlattices-synthesized semiconductors and
semimetals,” J. Cryst. Growth 52, 227–240 (1981).
3. D. L. Smith and C. Mailhiot, “Proposal for strained type-II superlattice
infrared detectors,” J. Appl. Phys. 62, 2545–2548 (1987).
4. R. H. Miles, D. H. Chow, J. N. Schulman, and T. C. McGill, “Infrared
optical characterization of InAs/Ga1 xInxSb superlattices,” Appl. Phys.
Lett. 57, 801–803 (1990).
5. C. Cervera, I. Ribet-Mohamed, R. Taalat, J. P. Perez, P. Christol, and
J. B. Rodriguez, “Dark current and noise measurements of an InAs/
GaSb superlattice photodiode operating in the midwave infrared
domain,” J. Electron. Mater. 41, 2714–2718 (2012).
6. N. Gautam, H. S. Kim, M. N. Kutty, E. Plis, L. R. Dawson, and
S. Krishna, “Performance improvement of longwave infrared photo-
detector based on type-II InAs/GaSb superlattices using unipolar current
blocking layers,” Appl. Phys. Lett. 96, 231107–231109 (2010).
Type-II Superlattices: Status and Trends 361
Figure 16.1 HgCdTe diode cross section showing key features for analysis (adapted from
Ref. 2).
substrate. Light of a wavelength shorter than the active layer cutoff (that is,
having energy greater than the active layer bandgap) enters through the
transparent substrate, passes through the wide-bandgap HgCdTe buffer
layer used to enhance collection of the return electron current to a remote
ground, and is absorbed in the n-type-as-grown active layer, generating
an electron–hole pair. The hole diffuses to the depletion region of the planar
p-junction, formed by arsenic implantation and activation, and is passivated
by a wide-bandgap cap layer and an overcoating insulator. From the
depletion region, the hole is swept out to the p-side and to the metal
contact, which is connected to the outside circuit (not shown). Because
the active layer extends laterally beyond the junction area, diffusion can
extend the effective optical collection area for holes by about one diffusion
length, creating a diode whose optical size is typically larger than the
junction size.
Competing with the photocurrent are dark currents from other sources
within a diffusion length of the junction. These dark currents give no signal
but add noise to the measurement and provide unwanted charge that must
be dealt with. The Auger generation mechanism is the most common dark-
current source in the best devices. This mechanism is due to thermally
excited free majority carriers (electrons, in this case) having sufficient
kinetic energy to create an electron–hole pair by impact ionization. The
Auger generation rate (equilibrium minority carrier concentration density/
Auger lifetime) is proportional to the majority carrier concentration. Thus,
typically, decreasing doping reduces Auger-generated dark currents. Since
these currents are generated by free carriers, they are diffusion currents
originating in the charge-neutral region of the diode (outside the depletion
372 Chapter 16
Figure 16.2 SLS diode cross section showing key features for analysis.
MWIR Detectors: A Comparison of Strained-Layer Superlattice Photodiodes… 373
Figure 16.3 Spectral QE of diodes from two MWIR HgCdTe wafers reported by Tennant.2
Fit points are taken from measured data, with fits being the result of the model fit to the
measured points. (See color plate section.)
from the substrate. That the peak QE is above 70% indicates that the internal
QE is near 90% for sample D and higher for sample C. The points in the curve
are a subsample of data taken from a high-resolution spectrophotometer for
the purpose of fitting to a model described in the paper. To obtain the fit, only
the composition and thickness of the layer were varied. The model fits are
given by the smooth lines. Figure 16.4 shows the reported absolute value of
current density plotted against voltage for these diodes. The flat reverse-bias
characteristic beyond a few times the thermal voltage indicates that these
devices are diffusion limited. Table 16.1 gives the measured and inferred
parameters for the wafers from Tennant’s paper.
Figure 16.4 Current voltage characteristic of HgCdTe MWIR diodes from wafers C and D.
(See color plate section.)
MWIR Detectors: A Comparison of Strained-Layer Superlattice Photodiodes… 375
Figure 16.5 Current voltage characteristic of gated SLS MWIR diode from Razeghi’s
group. Solid line is with a 40 V potential applied to the gate. Flat reverse bias characteristic
indicated diffusion-limited performance (adapted from Ref. 3).
376 Chapter 16
SLS 6.60 10 27
MWIR Detectors: A Comparison of Strained-Layer Superlattice Photodiodes… 377
assumed a value of 1.7 1013/cm3, the geometric mean of the two HgCdTe
values. In addition to not knowing the precise density of states, it is also noted
that the doping level was given only approximately as 1017/cm3, so the SLS
Auger coefficient value must be seen as an approximation. Note, finally, that
any other source of diffusion, such as traps that reduce lifetimes, or generating
interfaces remote from the junction, will add to the dark-current density and
make the inferred Auger coefficient larger than the actual value. That being
said, the SLS device appears to have an Auger coefficient 5 or more times lower
than HgCdTe.
The author was not able to find literature values of the Auger
coefficient at 150 K; however, 300-K data from Bewley et al.4 show that
Auger coefficients of SLS devices are 5–20 lower than those of HgCdTe,
consistent with the above inferences in this comparison. It should be noted
that Bewley et al.4 show that some further improvement in performance
can be obtained by modifying the growth structure to obtain a W
configuration by using a wide-bandgap ternary in the InAs sublayer.
Grein’s5 theoretical modeling of these W structures (albeit at 200 K) shows
Auger coefficients that compare to the 300-K-measured data of Bewley et
al.4 The inferred values at 150 K are also near the 300-K-measured values
of Bewley et al. and the 300-K-calculated values of Grein.6 For HgCdTe,
at least, the model described by Tennant2 gives a 150-K value that is about
3 larger than the 300-K value, so it does appear that these devices are
near the Auger limit.
References
1. A. Rogalski, “Superlattice Detectors” Chapter 17 in Infrared Detectors,
Second Ed., CRC Press, Boca Raton, FL, pp. 601–622 (2011) [with
references provides a comprehensive review of type-II SLS and other
related structures.]
2. W. E. Tennant, “Interpreting mid-wave infrared MWIR HgCdTe
photodetectors,” J. Prog. Quantum Electron. 36, 273 (2012).
3. G. Chen, B.-M. Nguyen, A. M. Hoang, E. K. Huang, S. R. Darvish, and
M. Razeghi, “Elimination of surface leakage in gate controlled type-II InAs/
GaSb mid-infrared photodetectors,” Appl. Phys. Lett. 99, 183503 (2011).
4. W. W. Bewley, J. R. Lindle, C. S. Kim, M. Kim, C. L. Canedy, I. Vurgaftman,
and J. R. Meyer, “Lifetimes and Auger coefficients in type-II W interband
cascade lasers,” Appl. Phys. Lett. 93, 041118 (2008).
5. C. Grein and M. E. Flatté, “Optimization of MWIR type-II superlattices
for infrared detection,” Proc. SPIE 7660, 76601L (2010) [doi: 10.1117/
12.851862].
6. C. H. Grein, M. E. Flatté, J. T. Olesberg, S. A. Anson, L. Zhang, and
T. F. Boggess, “Auger recombination in narrow-gap semiconductor
superlattices incorporating antimony,” J. Appl. Phys. 92, 7311 (2002).
7. B.-M. Nguyen, D. Hoffman, P.-Y. Delaunay, and M. Razeghi, “Dark current
suppression in type II InAs/GaSb superlattice long-wavelength infrared
photodiodes with M-structure barrier,” Appl. Phys. Lett. 91, 163551 (2007).
17.1 Introduction
17.2 The Complementary Barrier Infrared Detector (CBIRD)
17.2.1 CBIRD structure and characterization
17.2.2 CBIRD contact designs
17.2.3 Turn on and dark current characteristics
17.2.4 CBIRD focal plane arrays
17.3 Quantum Dot Barrier Infrared Detector (QD BIRD)
17.4 Summary
Acknowledgment
References
17.1 Introduction
The antimonide material system consisting of the nearly lattice-matched
semiconductors of InAs, GaSb, and AlSb (and their alloys with InSb, GaAs,
and AlAs) has recently emerged as a highly effective platform for the
development of sophisticated heterostructure-based mid-wavelength infrared
(MWIR) and long-wavelength infrared (LWIR) detectors, as exemplified by
the high-performance double heterostructure (DH),1 nBn,2 4 XBn,5 8 and
type-II superlattice infrared detectors.9 17 A key enabling device design
element is the unipolar barrier,17 which is used to implement the barrier
infrared detector (BIRD) architecture for increasing the collection efficiency
of photogenerated carriers, and reducing dark-current generation without
379
380 Chapter 17
other,17,34,35 the concept of the unipolar barrier has been in existence since
1963,36,37 in connection with the DH laser. In fact, many heterojunction
superlattice infrared detectors are based on variations of the DH design,
including the superlattice DH structure,38,39 the p-p-M-n detector,40 and the
PbIbN structure.41 We have been working with the CBIRD structure,17 which
is also a variation of the DH design.
Figure 17.1 (a) Calculated zero-bias energy band diagram of a CBIRD structure, where a
LWIR InAs/GaSb superlattice absorber is surrounded by an InAs/AlSb superlattice hole-
blocking (hB) unipolar barrier and a shorter period InAs/GaSb superlattice electron-blocking
(eB) unipolar barrier (reprinted from Ref. 17). (b) The schematic device structure layer
diagram.
382 Chapter 17
An Arrhenius plot shows that at 0.2 V, the dark current is diffusion limited for
a device temperature above 77 K. Spectral response measured under 0.2 V
applied bias at 77 K shows that the device has a 10-mm cutoff (defined by 50%
peak responsivity), with a peak responsivity of 1.5 A/W. We calculated the
shot-noise-limited blackbody D , where the noise spectrum is determined by
the measured dark current and photocurrent integrated over the 8- to 10-mm
spectral range (the overlap between the atmospheric window and the detector
cutoff). Under 0.2 V, the detector reaches BLIP operation at 86 K with a
blackbody BLIP D value of 1.1 1011 cm-Hz1/2/W for 300-K background
with f/2 optics. For 300-K background with a 2p field of view, the device
shows a BLIP temperature of 101 K with a blackbody BLIP D value of
2.6 1010 cm-Hz1/2/W. The device has a zero-bias dynamic resistance-area
product of R0A ¼ 14,000 V-cm2 at 77 K. However, since the detector is
expected to operate at a higher bias (0.2 V), a more relevant quantity is the
effective resistance-area product, given by RAeff ¼ kT=qJd. Under a 0.2-V
bias, the RAeff for this device is 670 V-cm2 at 77 K.
Optical characterization tools are invaluable in the study of the material
properties of CBIRD detectors. We have been investigating CBIRD devices
using photoluminescence (PL) and transmission spectroscopy techniques. We
find that the absorption quantum efficiency (QE), deduced from the
transmission measurements, serves as a good estimate of the upper limit of
the external QE, and the PL peak position is shown to correlate well with the
detector cutoff wavelength. In a comparison between the PL intensity and the
dark-current characteristics, a good correlation between a high PL intensity
and low dark current was observed, showing that the PL intensity well reflects
the material quality. Also, Shockley–Read–Hall (SRH) processes were
identified as the limiting factor of the minority-carrier lifetime of the CBIRD
material studied. We have also investigated the relation between the PL
intensity and the PL peak wavelength.42 A linear decrease of the PL intensity
with increasing cutoff wavelength of LWIR CBIRDs was observed at 77 K
(see Fig. 17.2). The trend remained unchanged in the temperature range of
10–77 K. This relation between the PL intensity and the peak wavelength can
be used favorably for comparison of the optical quality of samples with
different PL peak wavelengths. A strong increase in the width of the PL
spectrum in the studied temperature interval was observed and was attributed
to thermal broadening. We have also used the optical modulation response
technique to investigate the minority-carrier lifetimes in LWIR InAs/GaSb
SLs used for CBIRD absorbers. Minority-carrier lifetimes in the range of
33–38 ns were observed, in good agreement with other published results on
similar material.25 27 Furthermore, when comparing SLs with essentially the
same PL peak wavelength, correlation between the minority-carrier lifetime
and the PL intensity was observed. This shows that the PL intensity serves as a
Mid- and Long-Wavelength Barrier Infrared Detectors 383
Figure 17.2 Wavelength dependence of the PL peak intensity from CBIRD absorbers. The
data is collected from CBIRD samples with various thicknesses (ranging from 300 to 1,000
absorber superlattice periods). A fitted linear trend line is also shown. (See color plate section.)
that can suppress the leakage current and prevent the onset of frequency-
dependent noise. More details of these noise and gain studies can be found in
Ref. 44.
Figure 17.3 Zero-bias energy band diagram of a CBIRD structure, as conceived originally.
The n-type and p-type contacts are made from the same SL as the absorber.
Mid- and Long-Wavelength Barrier Infrared Detectors 385
Figure 17.4 The energy band diagram near the bottom contact region of a modified CBIRD
structure, with a double broken-gap tunnel junction design (adapted from Ref. 46).
InAs/GaSb SL. Note that although the n-type InAsSb bottom contact has
replaced the originally conceived p-type T2SL bottom contact, its function is
still the same, i.e., to allow the easy removal of excess holes from the absorber
layer. It does so effectively via a low-resistance broken-gap junction at the
eB SL/InAsSb interface, as depicted in Fig. 17.1, and as described in Ref. 17.
In an effort to further reduce growth demands, we have made another
modification to the bottom contact. The design described in Ref. 17 uses a
rather thick (1 mm) InAsSb bottom contact layer for the purpose of providing
a large processing window for etching to the bottom contact. In the modified
CBIRD design, the hB, absorber, and eB composing the active region are
the same as before, but the bottom contact is modified. As depicted in
Fig. 17.4, the width of the InAsSb layer is decreased from 1 mm to 0.2 mm
(to reduce growth demands) and is doped at n ¼ 1 1018 cm 3. It is grown on
a thick p-type GaSb buffer layer, which also serves as the new bottom contact.
In this new structure, a second low-resistance broken-gap tunnel junction is
formed between the n-InAsSb and the (new) p-GaSb bottom contact. The
doping levels are high around both of the interfaces to ensure low junction
resistance. The two broken-gap junctions provide a low-resistance electrical
connection between the active region (consisting of the hB SL, absorber SL,
and eB SL) and the p-GaSb bottom contact layer, enabling the transport of
excess majority carriers (holes) from the absorber to the bottom contact under
operating bias conditions. Further description of the double tunnel junction
contact is reported elsewhere.46
Figure 17.5 The peak responsivity (blue open squares), and the activation energy (red
open circles) extracted from Arrhenius analysis of the temperature dependence of I-V
characteristics, plotted as a functions of applied bias for a CBIRD device.
requires more than 200 mV of applied bias before it becomes fully responsive.
Arrhenius analysis of the dark current–voltage (I-V) characteristics of the
device also shows that, while under sufficiently high bias, the activation
energy reaches a value of 0.125 eV, which is very close to the energy
bandgap of the LWIR absorber (indicating diffusion-limited behavior), and at
lower bias the activation energy is higher than the bandgap (see Fig. 17.5).
The bias dependences of both the peak response and the activation energy
shown in Fig. 17.5 suggest the presence of an unintended small energy barrier
that needs to be overcome by applied bias before the device fully turns on.
This small energy barrier most likely resides in the conduction band, near the
hB/absorber interface, and acts to block the flow of minority carriers from the
p-type SL absorber to the top contact. One possible mechanism is conduction-
band offset between the hB and the absorber, as illustrated in Fig. 17.6(a).
Simulations using a DEc ¼ 80 meV conduction-band offset results in a
conduction-band barrier of 38 meV after band bending. Another possibility
is that the electrical (doping) junction is moved into the wide-gap hB
region, instead of coinciding with the metallurgical junction, as illustrated in
Fig. 17.6(b). Either case results in an unwanted conduction-band energy barrier.
In order to gain some insight into the nature of the mechanism responsible
for the observed turn-on behavior, we modeled the CBIRD active region using
drift-diffusion simulations.47 As it turns out, the observed CBIRD dark I-V
characteristics are less compatible with the situation depicted in Fig. 17.6(a),
where the electrical and metallurgical junctions coincide. In that case, the dark
I-V characteristics would have strong G-R dark-current characteristics due
to depletion in the (narrow gap) absorber adjacent the junction. The
actual observed I-V characteristics show nearly diffusion-limited behavior.
Mid- and Long-Wavelength Barrier Infrared Detectors 387
Figure 17.6 (a) Energy band diagram of an N p heterojunction with an 80-meV conduction-
band offset. (b) Energy band diagram of a zero conduction-band offset heterojunction with the
doping junction in the wide-gap region.
Simulations show that this can only be produced by moving the electrical
junction into the wide-gap region, regardless of whether there is a nonzero
conduction-band offset. The details of these calculations are reported in Ref. 48.
Both the conduction-band-offset-induced barrier, and the doping-profile-
induced barrier (or combinations of the two) can be removed by making small
adjustments to the device structure, through changes in the doping profile or
the design of the hB SL. Indeed, newer CBIRD structures have demonstrated
zero-bias turn on, while maintaining the same levels of dark-current density
and QE as those reported earlier in Ref. 17.46
LWIR InAs/GaSb SLs studied thus far appear to have substantially
shorter lifetimes compared to MCT;25 28 therefore, one might expect
correspondingly inferior dark-current performance. Yet, as shown in
Fig. 17.7, the measured CBIRD dark-current density levels are surprisingly
low. The explanation of this is two-fold. The first is related to the suppressed
tunneling in SLs.32 Recall that the diffusion dark-current density from the
p side of a p–n diode is given by Je,diff ¼ qn0 L=t n , where n0 is the minority-
carrier density in the diffusion layer, L is the diffusion layer width (or absorber
layer width), and tn is the minority-carrier (electron) lifetime. Assuming
(quasi-) equilibrium conditions, the expression can also be written in the form
of Je, diff ¼ qn2i LN =ðNa tn Þ, where Na is the acceptor dopant density. In a
388 Chapter 17
Figure 17.8 A set of images taken with a 320 256 format LWIR CBIRD FPA operating at
78 K.
Figure 17.9 (a) Schematic layer diagram of the QD-BIRD structure. (b) (Top) Schematic
energy band diagram of the active region of the QD-BIRD device; (bottom) expanded view of
the schematic energy band diagram for the QD-BIRD absorber near an InSb quantum-dot
layer.
The schematic layer diagram in Fig. 17.9 shows the growth sequence of
the QD-BIRD. The energy band diagram illustrates the structure of the QD-
BIRD, which is very similar to the standard nBn device structure as originally
described by Maimon and Wicks,3 consisting of an AlSbAs barrier
sandwiched between the InAsSb top contact layer and absorber layer. The
slight modification we introduced is the periodic insertion of 2.8 monolayers
(MLs) of InSb, which form self-assembled InSb QD layers in the InAsSb
absorber matrix, as illustrated in Fig. 17.9. The alloy composition of the
InAsSb matrix was adjusted slightly to reduce the Sb content. Details of the
structure have been reported earlier.18
Figure 17.10 shows the normalized PL spectrum of the QD-BIRD with two
distinct peaks at 4.0 mm and 5.5 mm. The origins of the two peaks are illustrated
in the bottom right panel of Fig. 17.9, which shows the schematic energy band
diagram in the vicinity of an InSb quantum-dot insertion layer in the InAsSb
matrix. The 4.0-mm peak is easily identified with the bandgap Eg of the InAsSb
matrix. The 5.5-mm peak is related to the QD. The band diagram shows that the
strained InSb forms a type-II broken-gap band alignment with the InAsSb
matrix, with both the conduction- and valence-band edge of InSb being
substantially higher than the conduction-band edge of InAsSb. The InSb
392 Chapter 17
result, since after initially passing through the absorber, light could re-enter the
absorber region after reflecting off the bottom of the substrate (the test devices
are mounted on a chip carrier with a metallic surface). Like the PL spectrum, the
spectral QE also shows a distinct bimodal behavior. The photoresponse
associated with the direct band-to-band transition in the InAsSb matrix is seen at
the shorter wavelengths, with approximate plateau QE values of 0.055, 0.33,
0.55, and 0.61 at T ¼ 77, 125, 175, and 225 K, respectively. For this InAsSb
matrix response, the 50% QE cutoff wavelengths are approximately 3.97, 4.15,
4.30, and 4.46 mm at T ¼ 125, 175, and 225 K, respectively.
Beyond the cutoff wavelength associated with the bulk InAsSb matrix, we
also observe an extended response that drops off approximately linearly; this is
attributed to the type-II transition between the QD valence-band state and the
InAsSb matrix conduction-band state. The extended response associated with
the QDs is noticeably weaker than the bulk InAsSb response. The external
quantum efficiencies at 5 mm for T ¼ 77, 125, 175, and 225 K are respectively
0.011, 0.086, 0.16, and 0.175, which are only approximately 20–30% of the
corresponding values found for the InAsSb matrix in the 3- to 4-mm range, even
though the PL intensities of the QD-to-matrix and the bulk transitions are
comparable. One possible reason for the weaker response of the QD-to-matrix
transition is that the QD hole state is confined by the InAsSb matrix. The
confinement energy DE is given by the difference between the InAsSb bandgap
Eg and the QD-to-matrix transition energy Emd (see Fig. 17.9). The photoexcited
hole (minority carrier) in the QD has to overcome this additional energy barrier
DE in order to escape and be collected. There is likely a distribution of QD sizes,
with a corresponding distribution Emd and DE. Smaller dots with smaller Emd
that lead to the longer extended wavelength would need a larger activation
energy DE for the photogenerated hole to escape; this is associated with a lower
escape probability. This would explain why the extended cutoff response
decreases as the wavelength increases.
Figure 17.11 also shows that the photoresponse increases with tempera-
ture; the responsivity at 225 K is approximately twice that at 125 K. Similar
behavior is found in Fig. 17.12, which shows that the T ¼ 175 K spectral QE
under 0.1V, 0.2 V, and 0.5 V. It appears that 200-mV reverse bias is
required for the photoresponse to fully turn on. In examining Figs. 17.11 and
17.12, we note that the ratio of the bulk InAsSb matrix response to the dot-to-
matrix response stays approximately the same as the temperature or the
applied bias increases. This indicates a common mechanism that blocks both
types of photocurrent. One possible mechanism responsible for this behavior
is the presence of a small unintended hB resulting from the valence-band
mismatch between the absorber and the AlSbAs barrier (the valence-band
edges of the absorber and the barrier should be aligned in an ideal nBn
structure). This barrier would block photocurrents generated from both the
InAsSb matrix absorption and the dot-to-matrix absorption.
394 Chapter 17
Figure 17.13 (a) Schematic illustration of a QD-BIRD structure with an unintended hole
barrier. (b) Calculated energy band diagrams under various biasing conditions for a QD-
BIRD structure with a 30-meV valence-band offset DEv between the eB and the absorber.
(See color plate section.)
Mid- and Long-Wavelength Barrier Infrared Detectors 395
Figure 17.14 Dark-current density as a function of applied bias of a QD-BIRD taken at 125,
175, and 240 K. (See color plate section.)
Higher temperature could aid thermionic emission processes for climbing over
the barrier, and applied bias could lower the barrier. We note that not all
of the applied bias goes toward lowering the barrier; simulation results in
Fig. 17.13 show that a barrier formed by a valence-band offset of DEv ¼
30 meV would require 200-mV reverse bias to overcome.
Figure 17.14 shows the measured dark-current density for a QD-BIRD as a
function of applied bias at 125, 175, and 240 K. The reverse-bias (negative top
contact bias) current–voltage characteristics appear diffusion limited at 175 and
240 K. Under 200-mV bias, the dark-current density levels are 1.52 10 7 A/
cm2 and 3.77 10 4 A/cm2, respectively at 125 and 175 K. We computed the
blackbody specific detectivity D for f/2 optics, in 300-K background conditions.
The photocurrent is determined from the integrated photoresponse in the 3- to 6-
mm spectral range. For a detector temperature of T ¼ 175 K, under 200 mV
bias, the blackbody D is dark-current limited and has a value of 1.07 1011 cm-
Hz1/2/W. At T ¼ 125 K, the blackbody D becomes background limited and has
a value of 3.76 1012 cm-Hz1/2/W. Overall, the performance of the QD-BIRD is
quite good. Optimization of the valence-band alignment between the absorber
and the barrier should improve device characteristics. The bimodal photo-
response seems to be a basic property of this device architecture.
17.4 Summary
The antimonide material system is relatively robust and has the potential for
good manufacturability. The versatility of the material system, with the
availability of three different types of band offsets, provides great flexibility in
device design. In the MWIR, the use of unipolar barriers in the nBn design has
396 Chapter 17
already seen success. In the LWIR, type-II InAs/Ga(In)Sb SLs have been shown
theoretically to have reduced Auger recombination and suppressed band-to-
band tunneling. Suppressed tunneling allows for higher doping in the absorber,
resulting in reduced diffusion dark current. Heterostructures such as those based
on the CBIRD design have been effectively used to suppress G-R dark current.
As a result, the dark-current performance of antimonide SL-based single-
element LWIR detectors is now approaching that of the state-of-the-art MCT
detector, with sufficient performance for tactical applications and potential for
strategic applications.56 Reliable surface leakage current suppression methods,
such as that based on robust surface passivation, would be needed to achieve
high performance in FPAs. Recent improvement in the CBIRD device has
reduced turn-on bias from over 200 mV to nearly zero, without affecting the level
of dark current and quantum efficiency. FPA results based on the new CBIRD
material have demonstrated excellent performance.
The nBn or XBn BIRD has the advantage of reduced dark current resulting
from suppressed Shockley–Read–Hall recombination and surface leakage.
High-performance detectors and FPAs based on InAsSb absorber lattice
matched to GaSb substrate, with a matching AlAsSb unipolar eB, have been
demonstrated. The bandgap of lattice-matched InAsSb yields a detector cutoff
wavelength of approximately 4.2 mm when operating at 150 K. We report
results on the QD-BIRD, which is a simple modification of the standard MWIR
nBn detector with an InAsSb absorber lattice matched to the GaSb substrate.
We show that by incorporating self-assembled InSb QDs into the InAsSb
absorber matrix, we can extend the detector cutoff wavelength from 4.2 mm to
6 mm, thus providing better spectral coverage of the MWIR transmission
window. The QD-BIRD has been observed to show infrared response at 225 K.
Acknowledgment
The authors thank S. Bandara, E. S. Daniel, E. R. Blazejewski, D. R. Rhiger,
and J. N. Schulman for helpful discussions. The authors also thank R. Liang,
M. Herman, E. Kolawa, S. Khanna, T. Cwik, and P. Dimotakis of JPL,
R. Stephan of the Game Changing Technology Division at the NASA Office of
the Chief Technologist, and M. Tidrow of the U.S. Army Night Vision Electronics
Sensor Directorate for encouragement and support. The research described in this
publication was carried out at the Jet Propulsion Laboratory, California Institute
of Technology, under a contract with the National Aeronautics and Space
Administration. Government sponsorship is acknowledged.
References
1. M. Carras, J. L. Reverchon, G. Marre, C. Renard, B. Vinter,
X. Marcadet, and V. Berger, “Interface band gap engineering in InAsSb
photodiodes,” Appl. Phys. Lett. 87(10), 102103 (2005).
Mid- and Long-Wavelength Barrier Infrared Detectors 397
18.1 Introduction
18.2 Mid wavelength Infrared QWIP Device
18.3 MTF of Megapixel MWIR QWIP FPA
18.4 Long Wavelength Infrared QWIP Device
18.5 MTF of Megapixel LWIR QWIP FPA
18.6 Dual Band QWIP Device Structure
18.7 Testing and Characterization of Multiband QWIP FPA
18.8 NEDT and MTF of Megapixel Multiband QWIP FPA
18.9 Complementary Barrier Infrared Detector (CBIRD) Device Structure
18.10 Testing and Characterization of CBIRD FPA
18.11 MRDT and MTF of CBIRD FPA
18.12 Conclusion
Acknowledgment
References
18.1 Introduction
Fast Fourier transformation of a spatial impulse response of an electro-optical
imaging system provides the optical transfer function or the modulation
transfer function (MTF) of the system in the spatial frequency domain. The
MTFs of the subsystems in the spatial frequency domain can be multiplied to
obtain the overall MTF of an imaging system. This is a much more
407
408 Chapter 18
Figure 18.1 Schematic depiction of the decrease in modulation depth from object to
image. Modulation depth also decreases with increasing spatial frequency (reprinted from
Ref. 2).
Modulation Transfer Function Measurements of Infrared Focal Plane Arrays 409
Emax Emin
M¼ , ð18:1Þ
Emax þ Emin
where E is the irradiance. Modulation M goes to zero when (Emax Emin) goes
to zero, which means that there is no detectable signal above the noise floor of
the system. On the other hand, modulation depth approaches its maximum
value of unity when Emin goes to zero, which represents an ideal imaging
system. Once the modulation of an image is measured experimentally, the MTF
of the imaging system can be calculated for that spatial frequency, using
Mimage
MTF ¼ : ð18:2Þ
Mobject
Generally, MTF is measured over a range of spatial frequencies using a
series of bar pattern targets. It is also customary to work in the
frequency domain rather than the spatial domain.3 This is done using a fast
Fourier transform (FFT) of the digitally recorded image. The absolute
value of the FFT of the point spread function (PSF) is then squared to yield
the power spectral density Simage of the image. The MTF can be calculated
using
s
Simage
MTF ¼ : ð18:3Þ
Sobject
The approach we have taken to measure the MTF of an electro-optical
system is by imaging a knife-edge target along the horizontal and vertical
orientations. Figure 18.2 shows how a slightly tilted knife edge can be imaged
at the pixelated focal plane arrays (FPAs).
The edge spread function (ESF) is constructed by selecting a region of
interest (ROI) that intersects the knife-edge image. The ROI on Fig. 18.2 has
10 columns and 8 rows. The ESF can be constructed by plotting the signal
strength of pixels starting from the lower right and continuing in the direction
indicated by the arrow. When the top of column 9 is reached, continue the
plotting process from row 8 and column 9, ending at row 1 and column 1. The
advantage of this approach is that it preserves the correlation of the data
points, except for the end points, where it can create discontinuity.4 By
properly choosing the number of rows and columns, one can construct
roughly a continuous (smooth) ESF.5 6 One can equally start from the upper
left-hand corner (i.e., row 1 and column 1) and end at the right bottom (i.e.,
row 8 and column 10). This will also give an ESF but starts with high signal
pixels on the left and ends up with low signal pixels on the right (i.e., inverse
of the right side of Fig. 18.2). Figure 18.3(a) shows a ROI of an actual
image of a knife edge. Figure 18.3(b) shows the ESF constructed from this
image using 735 data points.
The ESF is numerically differentiated to obtain the line-spread function
(LSF). Figure 18.4(a) shows the LSF corresponding to the image shown in
Fig. 18.3(a). The zero-frequency-normalized absolute value of the Fourier
transform of the LSF is the 1D MTF of the system. Figure 18.4(b) shows the
MTF as a function of the special frequency that corresponds to the image in
Fig. 18.3(a).
The MTF of the optical assembly can be removed by dividing the
measured total MTF of the aggregate system by the MTF of optical assembly.
This provides the upper limit for the FPA MTF, assuming no MTF loss due
to the electronics, analog-to-digital conversion process, display, connecting
cables, etc.
Figure 18.3 (a) ROI of an image of slightly tilted knife edge. (b) ESF constructed from the
data in (a) (reprinted from Ref. 2). (See color plate section.)
Modulation Transfer Function Measurements of Infrared Focal Plane Arrays 411
Figure 18.4 (a) The LSF associated with Fig. 18.3(b). (b) MTF of the FPA that produced
the image shown in Fig. 18.3(a) (reprinted from Ref. 2).
into the 0.5-mm-thick doped GaAs bottom contact layer. The pitch of the
FPA is 19.5 mm, and the actual pixel size is 17.5 17.5 mm2. The 2D gratings
on top of the detectors were then covered with Au/Ge and Au for ohmic
contact and high reflectivity. A few QWIP FPAs were chosen and hybridized
(via an indium bump-bonding process) to a 1024 1024 silicon complemen-
tary metal-oxide semiconductor (CMOS) ROIC at bias VB ¼ 1 V. At
temperatures below 90 K, the signal-to-noise ratio of the system is limited by
array nonuniformity, ROIC readout noise, and photocurrent (photon flux)
noise. At temperatures above 90 K, temporal noise due to the QWIP’s higher
dark current becomes the limitation.
We used the following equation to calculate the noise-equivalent
differential temperature (NEDT) of the FPA:
p
AB
NEDT ¼ , ð18:4Þ
DB ðdPB =dTÞsin2 ðu=2Þ
where DB is the blackbody detectivity, dPB/dT is the derivative of the
integrated blackbody power with respect to temperature, and u is the field-of-
view angle [i.e., sin2(u/2) ¼ (4f 2 þ 1) 1, where f is the f-number of the optical
system]. Figure 18.7 shows the NEDT of the FPA estimated from test
structure data as a function of temperature for bias voltages of VB ¼ 1 V.
The background temperature TB ¼ 300 K, the area of the pixel A ¼ (17.5
17.5 mm2), the f-number of the optical system is 2.5, and the frame rate is
10 Hz. Figure 18.7 shows the measured NEDT of the imaging system at an
operating temperature of T ¼ 90 K, 60-ms integration time, bias VB ¼ 1 V
for a 300-K background with f/2.5 optics and a mean value of 23 mK.
A 1024 1024 QWIP FPA hybrid was mounted onto a 5-W integral Sterling
414 Chapter 18
Figure 18.7 NEDT histogram of the 1,048,576 pixels of the 1024 1024 array showing
high uniformity of the FPA (reprinted from Ref. 8).
Figure 18.8 Image from a 1024 1024 pixel QWIP FPA mounted on a 84-pin leadless
chip carrier (reprinted from Ref. 8).
Figure 18.9 Horizontal and vertical MTFs of the MWIR imaging system based on a
1024 1024 pixel QWIP MWIR camera (reprinted from Ref. 8).
because this FPA was back illuminated through the flat thinned substrate
membrane (thickness 800 Å). This substrate thinning (or removal) should
completely eliminate the pixel-to-pixel optical crosstalk of the FPA. In
addition, this thinned GaAs FPA membrane has completely eliminated the
thermal mismatch between the silicon CMOS ROIC and the GaAs-based
QWIP FPA. Basically, the thinned GaAs-based QWIP FPA membrane
416 Chapter 18
Figure 18.11 Detectivity as a function of temperature at bias 1 V (reprinted from Ref. 8).
Figure 18.12 NEDT histogram of the megapixel LWIR QWIP FPA showing high uniformity
of the FPA. The uncorrected nonuniformity after single-point correction nonuniformity
reduced to 0.8% (reprinted from Ref. 8).
Figure 18.13 Horizontal and vertical MTFs of the MWIR imaging system based on a
1024 1024 pixel QWIP MWIR camera (reprinted from Ref. 8).
with f/2.5 optics and a mean value of 16 mK. Figure 18.13 shows the MTF of
the imaging system as a function of spatial frequency. The MTF of the spot
scanner optics at Nyquist frequency is 0.2; therefore, the MTF of the FPA
should be >0.5 at the Nyquist frequency Ny ¼ 25.6 cy/mm. As mentioned
earlier, the MTF of an ideal FPA (i.e., no pixel-to-pixel crosstalk) is 64% at
Nyquist frequency. Thus, the pixel-to-pixel optical and electrical crosstalk of
this LWIR megapixel FPA is negligibly small. We have observed oscillations
in many of our MTF measurements that may be due to the unfiltered
Modulation Transfer Function Measurements of Infrared Focal Plane Arrays 419
Figure 18.14 Responsivity spectrum of the LWIR1 and LWIR2 bands in the dual-band
QWIP. This data was taken using a 45-deg polished wedge with large-area test devices.
detector are Dl/l ¼ 32% and lc ¼ 12.2 mm, respectively. Figure 18.14 shows
the responsivity spectrums of both LWIR1 and LWIR2 detectors. The dark-
current density of the LWIR1 and LWIR2 detectors at operating bias
VB ¼ 1V and temperature T ¼ 40 K are 2.4 10 7 and 2.5 10 6 A/cm2,
respectively. Thus, the limiting dark-current density is 2.5 10 6 A/cm2,
which was determined by the LWIR2 pixels.
Figure 18.15 Architecture of 1024 1024 pixel QWIP FPA. (a) The left half of the FPA
contains LWIR1 active pixels, and the right half contains the LWIR2 active pixels. From left
to right, the light-coupling grating parameters progressively change with peak light-coupling
efficiencies at 7.75 11.75 mm. (b) The two adjacent pixels (one on each side) at the LWIR1-
to-LWIR2 transition line at the middle of the FPA.
MQW layers into the 0.8-mm-thick doped GaAs intermediate contact layer.
The LWIR2 detector pixels (on the right half) of the detector arrays were
fabricated by dry etching both of the MQW stacks into the 1.0-mm-thick
heavily doped GaAs bottom contact layer. The pitch of the detector array is
19.5 mm, and the actual LWIR1 and LWIR2 pixel sizes are 17.5 17.5 mm2.
The 2D grating reflectors on top of the detectors were then covered with Au/Ge
and Au for ohmic contact and reflection. Twelve 1024 1024 format detector
arrays were processed on a four-in. GaAs wafer. Indium bumps were then
evaporated on top of the detectors and silicon ROICs for hybridization. Several
multiband detector arrays were chosen and hybridized (via an indium bump-
bonding process) to 1024 1024 pixel ROICs.
Voids between the pixels of the detector array and the ROIC were backfilled
with epoxy. The original GaAs substrate was completely removed by lapping
and dry etching. One selected multiband FPA hybrid was mounted onto the cold
finger of a liquid helium cooled laboratory test dewar and biased at VB ¼ 1 V.
This selected FPA was tested at a temperature of 40 K. The detector dark
currents at T ¼ 40 K of LWIR1 and LWIR2 detectors are 2.6 10 15 and
2.5 10 12 A, respectively. The experimentally measured responsivity
spectrums of the 1024 1024 nine-band QWIP FPA are shown in Fig. 18.16.
Figure 18.16 (a) Responsivity spectrums of the LWIR1 portion of the 1024 1024 pixel
multiband QWIPFPA, which contains five spectral bands. (b) Responsivity spectrum of the
LWIR2 section of the FPA, which contains four spectral bands.
Figure 18.17 Histograms showing the NEDT of each segment of the 1024 1024 pixel
FPA.
the lens MTF. This leaves the MTF of the FPA and electronics. Figure 18.18
shows MTF(f)/MTF(f ¼ 0) of the QWIP FPA at different spectral bands.
Assuming ideal electronic MTF, this MTF represents the MTF of the FPA.
The horizontal MTFs at Nyquist frequency based on pixel pitch a (¼1/2a,
a ¼ 19.5mm) 25.64 cy/mm at 9.75 and 10.75 mm are 0.30 and 0.10,
respectively. Unfortunately, the vertical MTF shows a great loss around
10–15 cy/mm, and the MTF at Nyquist could not be measured due to poor
signal-to-noise ratio. The ideal MTF value at Nyquist frequency is 0.64, which
can be estimated from the sinc function sin(paf )/paf,9 where a is the pixel
pitch, and f is the spatial frequency. The loss of MTF can be attributed to
pixel-pixel crosstalk, diffraction-limited point spread, spherical aberration,
chromatic aberration, defocusing, electronics crosstalk, ROIC crosstalk, and
poor signal. It is clear from both Figs. 18.18(a) and (b) that the horizontal and
vertical LWIR1 (i.e., 9.75 mm) MTF values are lower LWIR2 (i.e., 10.75-mm)
424 Chapter 18
Figure 18.18 Experimentally measured (a) horizontal and (b) vertical MTFs as a function
of spatial frequency at 9.75-mm (LWIR1) and 10.75-mm (LWIR2) spectral bands.
MTF values. The lower MTF of LWIR1 band could be attributed to pixel-to-
pixel optical crosstalk in LWIR1 band due to the underlying nonreticulated
LWIR2 MWQ layers. The loss in overall vertical MTF could be due to some
sort of enhanced optical crosstalk between pixels having the same grating
parameters (i.e., vertical bands) and vertical pixel-to-pixel electronic crosstalk
due to the ROIC column amplifiers.
FPAs. The detector arrays and ROICs were hybridized using the SET FC-300
flip-chip bonder. After hybridization, the FPAs were backfilled with epoxy
and cured overnight. The substrate was completely removed by mechanical
lapping followed by a selective dry-etching process all the way down to the
etch-stop layer.
Figure 18.20 Dark current of CBIRD at bias 128 mV and 78-K operating temperature
(reprinted from Ref. 13).
Figure 18.21 Measured CBIRD NEDT histogram operating at 78 K, bias 128 mV, and
integration time of 370 ms. The mean NEDT is 18.6 mK (reprinted from Ref. 13).
Figure 18.22 (a) MRDT and (b) MTF as a function of spatial frequency measured from a
320 256 CBIRD FPA operating at 78 K, at bias 128 mV, and integration time of 370 ms
(reprinted from Ref. 13).
Nyquist frequency 16.67 cy/mm) in both the vertical and horizontal direction,
even after moving the target slightly to compensate for the phasing effect and
raising the temperature of the background.4 It is observed that only three bars
were apparent instead of four, and two of the bars merge into one at a
frequency close to Nyquist.
MTF technically provides a measure of image resolution or spatial
frequency response of the infrared imaging system. It is a measure of the way
in which the contrast is transferred from object space to image space as a
function of spatial frequency. MTF is inversely related to MRDT.9 The ESF is
then constructed as previously described. The ESF is numerically differenti-
ated to obtain the LSF. The zero-frequency-normalized absolute value of
the Fourier transform of the LSF is the 1D MTF of the system. The lens
MTF is removed by dividing the measured MTF by the lens MTF. The plot in
Fig. 18.22(b) is MTF( f )/MTF( f ¼ 0) of the FPA and electronics in horizontal
and vertical orientation.
The higher MTF at low frequency produces better contrast (see
Fig. 18.23); therefore, better images are observed at low spatial frequency.
Higher MTF values at high frequency produce good-quality images at higher
frequency. The horizontal and vertical MTFs at Nyquist frequency based on
pixel pitch a (¼1/2a, a ¼ 30 mm) 16.67 cy/mm are about 0.49 and 0.52,
respectively. The Nyquist frequency is well below the optical cutoff frequency
of 56.8 cy/mm based on the 8.8-mm detector cutoff wavelength. The loss of
MTF can be due to defocusing,5 6,9 an effect that is eliminated by acquiring
data at the best focus and then collecting data by moving the FPA by 50 mm
along the optical axis from the best focus location. This-50 mm move is
roughly the size of the Airy disk.
Figure 18.23 Image taken with the long-wavelength infrared CBIRD SL FPA, showing
good-quality reproduction of low and high spatial frequency. The FPA is operated at 78 K
with NEDT of 18.6 mK with f/2 optics at 300-K background (reprinted from Ref. 13).
Modulation Transfer Function Measurements of Infrared Focal Plane Arrays 429
The FPA MTF can be separated into the product of two components. The
geometric aperture MTF is related to the pixel size, and the diffusion MTF is
related to electro-optical properties.1,3 6 The diffusion MTF depends on the
diffusion length and geometry. The carrier diffusion degrades high-frequency
MTF and manifests as crosstalk (or MTF loss). However, the CBIRD pixels
are delineated down to the bottom contact, and it is expected that no lateral
carrier diffusion into the next neighbor can occur. The advantage of
delineation is the reduction of crosstalk. The disadvantage (in nonplanar
device structures) is that the fill factor is less than 100%. Shorter wavelengths
on the other hand can be absorbed near the top surface and can diffuse to the
next neighbor. In CBIRD FPAs, the only channel left through which the
charge carriers can diffuse to an adjacent pixel is the thin VDET COM layer.
The geometric aperture MTF can be estimated using a sinc function. Since
the pixel is square, the aperture MTF is the same in the horizontal and vertical
orientations. For a pixel pitch of 30 mm (the CBIRD FPA pixel size) a sinc
function describing an aperture MTF is plotted in Fig. 18.22(b). Smaller pixel size
actually improves high-frequency MTF since, at Nyquist (sampling using FPA
pitch), its value is greater than 0.64. The difference between aperture MTF and
the measured MTF is the upper limit on the diffusion MTF (crosstalk) since other
MTF components such as electronic and other effects, including surface
recombination, are not completely understood. The ROIC crosstalk is small,
0.1% by design. At Nyquist frequency, the difference between measured
horizontal/vertical and the ideal MTF is 0.14, but at low frequencies, the
difference is small. The MTF loss is basically an effective increase of the pixel size.
The geometric aperture MTF function decreases with increasing pixel size and
frequency. Thus, detectors can be viewed as an overlapping Gaussian-like array.
For example, for horizontal and vertical MTF data in Fig. 18.22, the pixel size
that will closely match the MTF data is 36 mm, which is larger than the pitch.
Imagery was performed at 78-K FPA operating temperature; Fig. 18.23 shows
the outside natural scenery. The image quality of the natural scene attests to the
very good MTF behavior at low and high frequencies. This FPA gave good
images, with more than 97% of the pixels being operable. Video images were
taken at a frame rate of 30 Hz and integration time of 0.37 ms.
18.12 Conclusion
The MTFs of megapixel single-band and multiband QWIP FPAs were
experimentally measured. MTFs of the single-band fully pixelated MWIR
and LWIR QWIP FPAs were approximately 50%, including the MTF
degradation due to electronics and cooler. A significant degradation of the
MTF was observed in the shorter spectral bands of the nine-band QWIP FPA
due to the thick underlying undelineated materials. A 320 256 format LWIR
CBIRD FPA has been demonstrated with 18.6-mK NEDT for a 300-K
430 Chapter 18
background with f/2 cold stop at 78-K FPA operating temperature. The
horizontal and vertical MTFs of this fully pixel-delineated CBIRD FPA at
Nyquist frequency are 49% and 52%, respectively. In conclusion, MTF
measurement is a powerful and simple technique that can easily provide a
measure of pixel-to-pixel crosstalk of FPAs.
Acknowledgment
The authors thank R. Cox, R. Liang, M. Herman, and E. Kolawa of Jet
Propulsion Lab, and S. Bandara and M. Tidrow of the U.S. Army Night Vision
Electronics Sensor Directorate. The research described in this chapter was carried
out at the Jet Propulsion Laboratory, California Institute of Technology, through
an agreement with the National Aeronautics and Space Administration.
References
1. G. C. Holst, “Infrared Imaging System Testing,” Chapter 4 in The
Infrared Electro-Optical Systems Handbook, Vol. 4: Electro-Optical
Systems Design, Analysis and Testing, M. C. Dudzik, Ed., ERIM, Ann
Arbor, MI and SPIE Press, Bellingham, WA (1993).
2. S. D. Gunapala, S. B. Rafol, D. Z. Ting, A. Soibel, J. K. Liu,
A. Khoshakhlagh, S. A. Keo, J. M. Mumolo, and J. Nguyen,
“Modulation transfer function of QWIP and superlattice focal plane
arrays,” Infrared Phys. Technol. [In Press] (2013).
3. W. J. Smith, Modern Optical Engineering the Design of Optical Systems,
Second ed., McGraw-Hill, New York (1990).
4. G. D. Boreman, Modulation Transfer Function in Optical and Electro-Optical
Systems, SPIE Press, Bellingham, WA (2001) [doi: 10.1117/3.419857].
5. M. Davis, M. E. Greiner, J. G. Sanders, and J. T. Wimmers, “Resolution
issues in InSb focal plane array system design,” Proc. SPIE 3379, 288–299
(1998) [doi: 10.1117/12.317596].
6. H. Holloway, “Collection efficiency and crosstalk in closely spaced
photodiode arrays,” J. Appl. Phys. 60, 1091 (1998).
7. S. D. Gunapala, S. V. Bandara, S. B. Rafol, and D. Z. Ting, “Quantum
Well Infrared Photodetectors,” in Semiconductors and Semimetals, Vol.
84, Academic Press, San Diego, pp. 59–151 (2011).
8. S. D. Gunapala, S. V. Bandara, J. K. Liu, C. J. Hill, S. B. Rafol,
J. M. Mumolo, J. T. Trinh, M. Z. Tidrow, and P. D. LeVan, “1024 1024
pixel mid-wavelength and long-wavelength infrared QWIP focal plane array
for imaging applications,” Semicond. Sci. Technol. 20, 473–480 (2005).
9. S. B. Rafol and E. Cho, “Modulation transfer function measurement on
QWIP focal plane array,” Proc. SPIE, 6941, 69410X (2008) [doi: 10.1117/
12.780050].
Modulation Transfer Function Measurements of Infrared Focal Plane Arrays 431
19.1 Introduction
19.1.1 Infrared detection basics
19.1.1.1 Photocurrent
19.1.1.2 Dark current
19.1.1.3 Detector metrics
19.2 QDs for Infrared Detection
19.2.1 Benefits of QDs for ISB detectors
19.2.1.1 High gain and the phonon bottleneck
19.2.1.2 Low dark current
19.2.1.3 Normal incidence absorption
19.2.1.4 Versatility
19.2.2 The potential of QDIPs
19.3 QD Growth
19.3.1 The formation of QDs in the SK growth mode
19.3.2 Properties of SK grown dots and their effect on QDIP performance
19.3.2.1 QD size
19.3.2.2 QD shape
19.3.2.3 QD density
19.3.2.4 QD uniformity
19.4 Device Fabrication and Measurement Procedures
19.5 Gallium Arsenide based QD Detectors
19.5.1 InGaAs/InGaP QDIP
19.5.2 First QDIP FPA
19.5.3 Two temperature barrier growth for morphology improvement
19.6 Indium Phosphide based QD Detectors
19.6.1 InAs/InP QDIP
19.6.2 Detection wavelength tuning using QD engineering
435
436 Chapter 19
19.1 Introduction
The strong interest in low-dimensional semiconductor structures originates
from their exciting electronic properties, which can have an important
impact on the performance of electronic and photonic devices. Quantum
dots (QDs), also known as quantum boxes, are nanometer-scale islands in
which electrons and holes are confined in 3D potential boxes. QDs are
expected to show a 0D, d-function density of states (DOS) and are able to
quantize an electron’s free motion by trapping it in a quasi-0D potential
confinement. As a result of the strong confinement imposed in all three
spatial dimensions, QDs are similar to atoms and are, in fact, frequently
referred to as “artificial atoms.” Due to this confinement, novel physical
properties will emerge that can lead to new semiconductor devices as well as
drastically improved device performance.
As the particles are confined in all three dimensions, there is no dispersion
curve, and the DOS is dependent only on the number of confined levels. For a
single dot, only two (spin-degenerate) states exit at each energy level, and the
plot of the DOS versus energy will be a series of d functions. Figure 19.1
shows the change of the DOS from a bulk system to the low-dimensional
systems of quantum wells (shown in the figure as QWL), quantum wires
(shown in the figure as QWR) and QDs. The calculation of these DOSs can be
found in an introductory solid state text.1
In QDs, the width of the electron energy distribution is zero in an ideal
case. This means that electrons in those structures are distributed in certain
discrete energy levels, and the energy distribution width is fundamentally
independent of temperature. In real semiconductor structures, due to many
interacting processes such as electron–electron and electron–phonon scatter-
ing (which can also be reduced by QDs due to the lack of phonons to satisfy
the energy conservation—the so-called phonon bottleneck3), a certain width
in the electron energy distribution exists. However, this width is expected to be
much smaller compared to that in bulk and quantum well (QW) systems.
The condition under which novel and interesting electronic properties
occur in a QD-based device is when the lateral size of the QD is smaller than
the coherence length and the elastic scattering length of the carriers.
Additional quantum-size effects require the structural features to be reduced
to the range of the de Broglie wavelength. The advantages in operation
depend not only on the absolute size of the nanostructures in the active region,
but also on the uniformity of size and shape. A large distribution of sizes
would “smear” the DOS of QDs, thus making the DOS more like that of bulk
material. Therefore, the repeatable fabrication of these nanometer-size 3D
quantum structures requires methods with atomic-scale accuracy, which
presents a major challenge for current nanostructure material fabrication
technologies.
The fabrication techniques of QDs can be categorized into top-down
methods that use lithography and etching, and bottom-up methods that use
self-assembly. Examples of top-down methods usually include electron-beam
lithography, dry etching, and, sometimes, patterned substrate regrowth. QDs
can be etched from QW structures via low-energy electron-beam lithography.
Another method of creating QDs is realized by applying voltage to
nanoelectrodes. The spatially modulated electric field created by the voltage
localizes electrons in a small area. QDs can also be created through the
selective growth of a narrow-gap semiconductor material on a patterned wide-
gap substrate. One problem with such top-down methods is the low optical
efficiency of the resultant dots; the high surface-to-volume ratios of these
nanostructures and the associated high surface-recombination rates, plus the
damage introduced during fabrication together impede the successful
formation of high-quality QD devices.
A breakthrough in QD fabrication techniques was self-assembly based on
the Stranski–Krastanov growth mode.4 In this method, the lattice constants of
the substrate and the crystallized material differ greatly, so only the first
deposited monolayer crystallizes in the form of epitaxial strained layers, where
the lattice constant is equal to that of the substrate. When the critical thickness
of the epitaxial layer is exceeded, the significant strain that occurs in the
438 Chapter 19
epitaxial layer leads to the breakdown of this ordered structure and to the
spontaneous creation of randomly distributed islets (e.g., QDs) of regular
shapes and similar sizes. The small sizes of the self-assembled QDs, the
homogeneity of their shapes and sizes in a macroscopic scale, the perfect
crystal structure, and the fairly convenient growth process, without the
necessity to precisely deposit electrodes or etching, are among this method’s
advantages.
QDs are expected to lead to novel semiconductor devices and improve
existing devices’ performance. One successful example is the QD laser. The
main advantages of QD lasers over conventional QW lasers are lower
threshold current density, high gain, weak temperature dependence (high
characteristic temperature T0), and low chirp.5,6 Another application of QDs
is in QD infrared detectors (QDIPs), which is the main focus of this chapter.
Figure 19.2 Schematic device structure of (a) a QWIP and (b) a QDIP (reprinted from
Ref. 9).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 439
Figure 19.3 Bandstructure of a QWIP or QDIP under bias (adapted from Ref. 10).
19.1.1.1 Photocurrent
We will start by looking at where the signal comes from in a QWIP or QDIP.
Photon detection occurs when incident infrared light excites an electron in the
ground state out of the QW or QD and into the continuum. There are two
possible paths (depicted in Fig. 19.4) by which the photoexcited electron can
escape from the confining potential. In the first path, the light directly excites
the electron from the ground state to the continuum. In the second path, the
light first excites the electron from the ground state to an excited state. Then,
from the excited state the photoexcited carrier can thermally escape to or
tunnel to the continuum. While under an applied bias, once an electron has
been excited to the continuum, it will be swept toward the contacts and
contribute to the photocurrent.
Also depicted in Fig. 19.4 are the relaxation or recapture paths, which
occur when a photocurrent electron does not exit the well or dot, or does not
reach a contact. While the photoexcitation process just described is caused by
electrons, these relaxation processes are caused by phonons. The time it takes
for a carrier traveling in the continuum to be recaptured into a well or dot is
the carrier lifetime. The ratio of the carrier lifetime to the time it takes for an
electron to travel across the entire device from contact to contact is defined as
the gain g:
t lifetime carrier lifetime
g¼ ¼ ð19:1Þ
t transit carrier transit time:
A gain greater than 1 means that a photogenerated electron can travel
through the device more than once, creating a greater signal per photoexcita-
tion event, and likewise, increasing the responsivity.
Figure 19.5 Dark-current mechanisms in QWIPs and QDIPs: (A) thermionic emission,
(B) thermally assisted tunneling, and (C) sequential tunneling (reprinted from Ref. 7).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 441
where A is the area, In is the noise current, R is the responsivity, and Df is the
noise bandwidth. D is intended to account for variations in detector
operation parameters (size and noise bandwidth) to allow for easy comparison
of the sensitivity of different devices and device technologies.
The final metric to be discussed here is noise-equivalent difference
temperature (NEDT). NEDT is often used as a measure of focal plane array
(FPA) performance to compare different FPAs. It gives the minimum
resolvable temperature difference when viewing a scene and can be defined in
terms of D by the expression
p p
1 ADf 1 ADf
NEDT ¼
¼Z , ð19:6Þ
Cd ðDlÞ D dR D
Dl dl
dl
where a thermal variation in an object of DT gives a change in the blackbody
emittance of Cd ðDlÞDT over a spectral range Dl, and dR=dl is the emittance.
A more practical definition in terms of measuring the NEDT of an actual
FPA uses the following two relations:
signalT2 signalT1
SNR ¼ , ð19:7Þ
noise
DT
NEDT ¼ ð19:8Þ
SNR,
where SNR is the signal-to-noise ratio, signalT1 and signalT2 are the
FPA signal levels corresponding to the FPA viewing a blackbody source
at the two different temperatures T1 and T2, noise is the temporal noise
of the FPA signal, and DT is the difference in temperature between T1
and T2.
have also been used successfully for commercial FPAs. They utilize ISB
transitions for detection and can take advantage of mature III-V compound
growth and fabrication technology. However, one major limitation of
QWIPs is that, due to the transition selection rules, the most widely used
n-type QWIPs are not sensitive to normally incident light and typically have
a narrow response range in the infrared. p-type QWIPs are able to detect
normal incidence light due to band mixing; however, their low detectivity
limits their practical use. Also, QWIPs require cryogenic cooling to
eliminate their high intrinsic dark current.14
QDIPs extend the 1D confinement in QWIPs to 3D confinement. QDIPs
utilize ISB absorption between bound states in the conduction or valence
band in QDs. With high-uniformity and high-density QD layers, QDIPs are
actually predicted to outperform QWIPs15,16 due to their (1) intrinsic
sensitivity to normal incidence light (2) longer photoexcited electron lifetime
due to the phonon bottleneck, and (3) lower dark and noise currents.17 These
benefits will ultimately allow for higher operating temperatures that will
reduce the cost and complexity of detector and imaging systems by reducing
the cooling requirements normally associated with cryogenically cooled
detector systems.
The potential of QD-based detectors has spurred a great deal of research
activity in the area.18 23 QDIPs can be used in FPA-based infrared imaging
systems, which have been widely investigated for MWIR and LWIR
applications.24 27 So far, most QDIPs reported have showed inferior or, at
best, comparable performance to that of QWIPs with similar parameters,
although high-temperature demonstrations with moderate performance are
becoming more common. The major challenge facing QDIPs is the QD
fabrication. To achieve their potential advantages, QDIPs need to have
uniform and high-density QD layers. New device designs for QDIPs are also
required to further improve the technology’s performance as a competitive
infrared photodetector platform.
The Center for Quantum Devices (CQD) at Northwestern University has
many years of experience in QD-based detector and FPA development. Some
of this work is presented later in this chapter. After a brief introduction to
some of the experimental methods, GaAs-based QD devices and FPAs are
discussed, followed by InP-based devices and FPAs. Within each section, the
self-assembled QD growth by metalorganic chemical vapor deposition
(MOCVD) is discussed along with device design and results; where applicable,
FPA imaging results are also presented.
QDIPs have the potential for excellent infrared detector performance. The
interest in QDIPs comes primarily from their predicted ability to achieve high
performance at high operating temperatures (near room temperature). High-
temperature operation is an important but difficult technological hurdle for
current photon detectors, which typically require some level of cryogenic
cooling. The high-operating-temperature capability of QDIPs comes from
two QD-related effects: low dark current and high photoelectric gain. QDIPs
also have two other important technological benefits, which will also be
discussed.
ð19:9Þ
where F1 and F2 are the electron envelope functions, « is the polarization
vector for the incident infrared light, and the p is the momentum operator.
For normally incident light, «x and «y are nonzero, and «z is zero. In the QW
case, F1 and F2 are only functions of z (the confined dimension). For the
oscillator strength, in the x and y directions the partial derivative goes to
zero, and in the z direction «z is zero; therefore, f12 is zero. In the QD case, F1
and F2 are functions of x, y, and z; therefore, even though «z is zero, the first
446 Chapter 19
Figure 19.7 Schematic of the polarization sensitivity of (a) QD and (b) QW absorption
(reprinted from Ref. 7).
two terms are nonzero and contribute to the oscillator strength for normal
incidence.
In QWIPs the normal incidence limitation is overcome by placing a grating
on the detector surface that redirects the light, changing the polarization
orientation. In QDIPs with normal-incidence absorption, this extra, nontrivial
processing step could be removed, simplifying the device processing. Also, the
polarization sensitivity limits the ultimate QE of a QWIP to 50%, whereas an
ideal QDIP would not have this limit on the QE.
19.2.1.4 Versatility
This last advantage of QDIPs is not a performance advantage but a design or
technology advantage. Compared with bulk and QW detectors, QDIPs have
more numerous adjustable design parameters and thus greater potential
versatility. For example, to tune the wavelength, the main adjustment
parameter for bulk systems is material composition. In a QW system the
parameters are material composition and well thickness, with strain being a
secondary parameter. Finally, in a QD system, material composition, strain,
and QD size and shape can all be used to tune the wavelength; there is larger
number of available “knobs to turn” to achieve the desired outcome.
However, this can be a double-edged sword if the parameters are difficult to
control, which is, in fact, the case for QDs, as is discussed in Section 19.3 on
QD growth.
and for LWIR detection where the D for QDIPs was predicted to be more than
10 higher than for HgCdTe. However, they also showed that the
demonstrated QDIPs in the literature are all still achieving D values at least
an order of magnitude lower than this ideal, predicted value.
19.3 QD Growth
The biggest hurdle in the study of low-dimensional semiconductor systems has
always been the physical realization of such systems. Since the relevant feature
sizes for these systems are at the nanometer level, even slight imperfections in the
material can mask or eliminate any quantum size effects. For example, in early
experimental investigations of QW systems, low-quality well material and rough
well-barrier interfaces prevented the observation of quantum-sized effects. These
problems were overcome with the advent of molecular beam epitaxy (MBE) and
metalorganic chemical vapor deposition (MOCVD) technologies. After much
progress was made in QWs, attention turned toward structures with lower
dimensionality, but QDs faced similar fabrication challenges. For detector
applications, QD systems should have the following qualities:
• QDs of appropriate size
• uniform QDs in high-density arrays
• defect-free QDs
Creating QDs with these qualities proved difficult by methods such as
direct patterning using electron-beam lithography. A key innovation in QD
formation was the rediscovery and application of the Stranski–Krastanov
(SK) growth mode, which is a 3D crystal growth mode that can occur in the
epitaxy of III-V materials. Of the available methods for creating QDs, the use
of the SK growth mode comes closest to meeting the aforementioned three
main requirements for QDs in detectors. This section looks more closely at the
SK growth mode, which is also referred to as QD fabrication by self-assembly.
Figure 19.8 Illustration of the Stranski Krastanow growth mode (reprinted from Ref. 7).
materials is In(Ga)As on GaAs. Other systems that have been studied include
InAs on InP, Sb-based III-Vs, and III-nitrides.
The SK growth process can be understood by considering the interplay
between the surface, interface, and strain energies in the configuration.
Initially, the sum of the epilayer surface energy and the interface energy is
lower than the surface energy of the matrix; therefore, it is favorable to have
layer-by-layer growth of the QD material. After several monolayers of
deposition, the buildup of strain energy makes it no longer favorable to have
flat, layer-by-layer growth. With this built-up energy, the material needs to
relieve the strain and does so by forming 3D dot structures.
The key factor here for optoelectronic devices is that the relaxation
process in SK growth can be controlled to produce defect-free (also called
coherent) QDs. This coherent relaxation was first noted in the mid-1980s.4
Because the dot material experiences a compressive strain, when the coherent
relaxation takes place, the lattice constant expands toward the dot edges, as
illustrated in Fig. 19.9. This coherent relaxation means that the QDs grown
by SK mode do not have the defects that reduce the 3D confinement in other
QD fabrication methods.
Figure 19.9 Schematics illustrating (a) an SK dot that has relaxed via defect formation and
(b) an SK dot that has experienced coherent relaxation and thus has no defects (reprinted
from Ref. 7). (See color plate section.)
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 449
19.3.2.1 QD size
In order for QDs to be useful, they must be of an appropriate size. The
absolute minimum size for a dot is the smallest size for which there will still be
one electron energy level in the dot. For a spherical dot, this size is31
pħ
Dmin ¼ q : ð19:10Þ
ð2mq DEc Þ
In practice, the dot must actually be larger than this minimum size in
order to provide strong carrier localization at nonzero temperature. This is
easily achieved in SK growth. The challenge, then, is in not surpassing the
maximum useful dot size. The maximum size has been surpassed when the
QDs have more than 1 or 2 energy levels within the dot and no longer show
strong 3D confinement effects. When there are more than two energy levels
in a dot, these additional levels can break the phonon bottleneck and dark-
current reduction. Also, if the dot is too large, the energy levels become
more widely spread, further reducing the phonon bottleneck. QDIP device
models showing good performance have been based on dot sizes around
15 nm.28,29 In the literature, QDs grown by SK self-assembly vary from
about 10 to 50 nm for coherent dots.
Size control of the QDs is also important for controlling the position of
the excited-state energy level of the QD relative to the continuum. If the
excited state is too deep in the dot, a high bias may be required to extract a
photoexcited electron from a QD. This is the case for the calculated dot
configurations shown in Fig. 19.10.
19.3.2.2 QD shape
Along with the size, the shape of the QD is important in determining the
energy level structure. “Ideal” QDs are spherical or at least symmetrically
shaped, but often this is not the case for SK grown dots, which tend to have
pyramidal or lens-like shapes that are flattened in the growth plane (i.e.,
wider than they are tall). If large enough, this asymmetry can result in a
polarization-dependent absorption because the confinement becomes more
like that in QWIPs. The shape of the QD is not as easily controlled as the
QD size.
19.3.2.3 QD density
The QD density in SK growth is determined by the QD nucleation
process and subsequent ripening. During SK growth, the adatom
mobility on the growth surface strongly determines how closely together
the QDs will nucleate. For example, most growth experiments have
shown that decreasing the growth temperature can increase the QD
density. Since the adatom mobility is reduced at lower temperature, a
450 Chapter 19
Figure 19.10 Example calculations of energy levels and oscillator strengths for InGaAs
dots in InGaP barriers for dots with (a) a tall and narrow shape and (b) a flat and wide shape
(reprinted from Ref. 7). (See color plate section.)
equilibrium size and spacing. Under conditions that favor ripening, the
dots will migrate and merge with one another, reducing the density and
creating dots that either are too large for 3D confinement effects or relax
incoherently via defects.
For detector applications, a high density of QDs above 5 1010 dots/cm2
is desired. For SK growth of In(Ga)As dots on GaAs substrate, densities of up
to 1011 dots/cm2 have been demonstrated, though densities in the low or mid-
1010 range are more typical.
The QD density affects both the absorption and transport characteristics
of a QD detector. Phillips formulated the absorption coefficient a in a QDIP
as follows:29
" #
n1 s QD ðE EG Þ2
aðEÞ ¼ A exp ½cm 1 , ð19:11Þ
D s ens s2ens
19.3.2.4 QD uniformity
In a QDIP device, the uniformity of the QDs has a very strong correlation
with device performance. The nonuniformity of the dots can occur in any of
the material parameters such as strain, composition, shape, or size, but size is
usually considered the most dominant and is also the most easily quantifiable
via methods such as atomic force microscopy (AFM) or transmission electron
microscopy (TEM). One source of size nonuniformity is the presence of
452 Chapter 19
incoherent, relaxed dots in an array. These are usually of much lower density
than coherent dots but are significant due to their very large size and potential
to create defects in subsequent layers. These defect dots can be almost
completely avoided under good growth conditions. The main source of
nonuniformity is the Gaussian distribution of the size of the coherent dots.
There are two possible “dimensions” of nonuniformity to consider; we first
discuss layer-to-layer nonuniformity, then we look at nonuniformity within a
single layer.
In a real QDIP, in order to obtain a sufficiently high volume density of dots, it
is necessary to stack QD layers on top of one another in the same way QWIPs
consist of stacks of QWs. The layer grown on top of the QDs to cover them is
called the cap layer. Fortunately, for device applications, the strain profile of the
SK QDs results in preferential growth such that the cap layer returns to a flat
surface once the QDs have been completely covered, facilitating the stacking of
more dot layers on top. This flattening process is illustrated in Fig. 19.11.
For the stacking of dots, the layer-to-layer spacing is important because
below a certain thickness the strain field of the buried dot layer will cause the
QDs in subsequent layers to vertically align with the dots below. For general
detector applications, this is not desired because aligned dots tend to
increase in size with stack number, have lower spatial coverage due to
nonrandom positioning, and may increase the probability of dot-to-dot
tunneling, increasing the dark current. For sufficiently thick spacings, and if
there are neither excess strain nor defects, each dot layer grows more or less
independently of the another, with each layer having similar in-layer dot
uniformity.
In the literature, the standard deviation in size for SK-grown QDs is
around 10%. Optical, electrical, and structural measurements on QD arrays
suggest that the variation is a Gaussian distribution. One possible cause for
this nonuniformity is asynchronous nucleation of islands due to local
variations in the thickness of the QD layer.33 This amount of nonuniformity
can significantly affect the detector performance. Recall from Eq. (19.11) that
the relevant measure of nonuniformity is the ratio of the linewidth of a single
dot to that of the entire array s QD=sens. Increased nonuniformity in the dot
array will broaden the spectral response, resulting in a lower peak responsivity
(although the integrated responsivity will be similar to a more uniform array).
Figure 19.11 Schematic illustrating the mechanism for flat cap layer growth (reprinted from
Ref. 7).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 453
Values of 0.01 for sQD=s ens are typical for current fabrication technology and,
according to Phillips’ modeling results,29 cause an order of magnitude
decrease in detectivity from the ideal case of s QD=sens ¼ 1.
cleaning process, the dies were flip-chip bonded and underfilled with epoxy.
Then, the FPA substrate was thinned using mechanical lapping and polishing.
Finally, the hybridized die was mounted in, and wire bonded to, a leadless
ceramic chip carrier. A more detailed description of the FPA fabrication can be
found in Refs. 19 and 24.
The FPAs were tested on a CamIRaTM infrared FPA evaluation system
made by SE-IR Corp. The FPA hybrid was mounted on the cold finger in liquid
nitrogen cryostat. The cryostat window was a 3- to 12-mm broadband Ge filter.
A Janos Technology Asio MWIR lens with f/2.3 was used. In the following tests,
the background temperature was 300 K. Varying operating temperatures, biases,
and frame rates were used. For the GaAs work, a Litton ROIC was used, whereas
in our InP FPA work, an Indigo 9705 ROIC was used. The FPA response was
characterized using an extended-area blackbody from CI Systems. The current
injection efficiency was estimated from the single-detector measurements as
described in Ref. 24. The temporal noise of the FPA was measured by taking the
standard deviation of the FPA signal. A common metric for FPA performance is
the noise-equivalent temperature difference (NEDT). In order to measure the
NEDT, the FPA signal is measured for two different temperature targets
provided by the extended-area blackbody (25 and 35 8C). The differential signal
for the two temperatures divided by the temporal noise gives the SNR, and the
target temperature difference divided by the SNR gives the NEDT.
Figure 19.12 A planar AFM scan of closely packed InGaAs QDs on an InGaP matrix
(1 mm 1 mm) showing spherical dots (reprinted from Ref. 35).
measured responsivity increases linearly with bias at low bias voltage and
saturates at around 4.5 V, as shown in Fig. 19.15. This may be due to the
saturation velocity of the generated carriers in a large electric field.
except for the active region, which was grown at 440 8C. The major improvements
over our previous InGaAs/InGaP QDIP structure32 were that the active region
was grown at lower growth temperature and the doping level of the QD layer was
optimized to give the maximum photoresponse.39 Also, the InGaP barrier was
grown with a slower growth rate (1.3 Å/sec) to improve the morphology and
reduce the occurrence of defects.
The dark current Id of the QDIP mesa was measured as a function of bias
Vb at different temperatures (see Fig. 19.16). Also shown in Fig. 19.16 is the
300-K background (BG) photocurrent with a 150-deg field of view (FOV).
The background-limited performance (BLIP) temperature was measured to be
140 K for the 2.6 V < Vb < 3.7 V range. For 0 V < Vb < 2 V, the dark
current did not show any increase from T ¼ 30 K to T ¼ 140 K. At Vb ¼ 1
V, dark current increased less than two orders of magnitude from T ¼ 30 K to
120 K. Positive threshold voltages seemed to exist. For T ¼ 30, 50, 77, 95, 120,
and 140 K, the threshold voltages were 4.8, 4.2, 3.2, 2.5, 2.1, and 1.7 V,
respectively. Above threshold voltages, dark current increased exponentially.
This behavior is similar to the dark current shown in bound-to-bound QWIPs
and may be due to the complex tunneling process caused by the high-field
domain formation.15 Overall, very low dark current was observed for this
QDIP. At a bias of 0.1 V and T ¼ 120 K, dark current of 0.98 pA was
observed, which corresponded to a current density of 6 10 10 A/cm2. A very
asymmetric I-V relationship was also observed in this QDIP, especially at
lower temperature ranges. At T ¼ 77 K and bias of 2 V, the Id was 5.1 pA,
while the Id was five orders of magnitude higher at bias of 2 V (0.43 mA). The
properties observed in our QDIP were very different from the dark-current
Figure 19.16 Dark current measured as a function of bias for an InGaAs/InGaP QDIP at
different temperatures (reprinted from Ref. 19).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 459
Figure 19.17 Relative spectral response of the InGaAs/InGaP QDIP structure (reprinted
from Ref. 19).
460 Chapter 19
and bias of 5 V, very high peak responsivity of 3.1 A/W was observed for our
QDIP. Due to the low dark current, photoresponse with a peak responsivity of
6.7mA/W was observed at a temperature of T ¼ 200 K with Vb ¼ 0.7 V.
After measuring the noise current, the detectivity of our QDIP as a function
of bias was measured at both T ¼ 77 K and T ¼ 95 K and is shown in Fig. 19.19.
The highest detectivities of our QDIP were 3.67 1010 cmHz1/2/W at bias 1.6 V
and 3.62 1010 cmHz1/2/W at bias 1.4 V for T ¼ 77 K and T ¼ 95 K,
respectively. Figure 19.19 shows that our QDIP performance did not degrade
from 77 K to 95 K. At T ¼ 95 K and bias 1.6 V, g was found to be 823, and h was
found to be 2.0 10 4. The very high gain value can be attributed to the very long
lifetime of the photocarriers inside QDs.41 The very low QE for our QDIP was
due to the smaller fill factor of QDs across a single epitaxial layer. In addition, the
electrons are better confined in dots than in QWs.
After taking the single-element QDIP detector measurement described
above, a 256 256 detector array was fabricated from the same wafer. Our
first InGaAs/InGaP QDIP FPA gave acceptable imaging with 90% of the
pixels being operational. The measured NEDT was 509 mK at an operating
temperature of 77 K and bias of 1.6 V. The higher NEDT shown in the
actual FPA might be due to nonuniformity of the array pixels.15 Since only
2-mm-tall indium bumps were used for the QDIP array pixels, some pixels
did not have a good connection with the readout units during the flip-chip
bonding (with a bonding pressure of 12 kg). This resulted in about 10%
nonoperational pixels. In addition, since only mechanical polishing was used
for the FPA substrate thinning, some pits and defects on the surface might
create scattering centers that reduce the photoresponse of QDIP pixels. Like
QWIPs, our QDIPs are also high-impedance devices. They yield a very high
current-injection efficiency; based on the QDIP test mesa, the current
injection efficiency was calculated to be 90% at a bias of 1.6 V, temperature
of 77 K, and frame rate of 35 Hz.
Due to the high-temperature performance of our QDIPs, thermal imaging
can be achieved at temperatures as high as 120 K. Figure 19.20 shows a
Figure 19.20 Thermal image of a hot soldering iron head taken by a QDIP FPA camera
operating at T ¼ 120 K (reprinted from Ref. 42).
462 Chapter 19
thermal image of a hot soldering iron head that was taken at T ¼ 120 K
without bad pixel replacement or two-point correction.
It should be noted that our first GaInAs/GaInP QDIP FPA has not been
fully optimized. First, the Litton ROIC with direct-injection readout was
designed for MWIR InSb photodiode FPAs, so this ROIC is definitely not the
best choice for our QDIP FPA. Second, no antireflection coating was used on
the backside of FPA substrate (GaAs), resulting in about a 30% reflection
loss. Third, no passivation layer was used for the QDIP detector pixels.
Figure 19.21 (a) 11 mm2 AFM of a single layer of uncapped InGaAs QDs on an InGaP
matrix, showing a dot density of about 3 1010 cm2. (b) and (c) 5 5 mm2 AFM images of
the surface in (a) when the InGaAs QDs were capped (b) with a 35-nm InGaP layer grown at
440 8C and (c) with a 6-nm-thick InGaP layer grown at 440 8C and then 29 nm of InGaP layer
grown at 480 8C. The root-mean-square (RMS) roughness of the surfaces of (b) and (c) are
6.0 nm and 1.9 nm, respectively (reprinted from Ref. 42). (See color plate section.)
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 463
6.0 nm (when the entire barrier was grown at low temperature) to 1.9 nm [see
Figs. 19.21(b) and (c)].
Photoluminescence (PL) measurements can provide useful information on
the interband transitions in the QDs, although it does not reveal any
information about the ISB transitions, which are directly related to the QDIP
operation. However, the PL data can be used to check the optical quality of
the buried QDs. Room-temperature measurements were performed on the
capped samples described in Figs. 19.21(b) and (c). The peak luminescence
wavelength was the same for both samples, meaning that the two-step barrier
growth technique does not have a dramatic effect on the size or In
composition of the QDs, possibly resulting from an enhancement of In–Ga
intermixing. However, when the InGaP barrier was grown at two different
temperatures, the intensity of the PL spectrum was two times higher than for
the single, low-temperature barrier sample, confirming that the growth of the
upper part of the barrier at a higher temperature improved the material
quality (see Fig. 19.22).
To test the effect of this growth technique on the device performance, two
QDIP devices were grown and based on the two dot and barrier growth
conditions described in Fig. 19.21(b) and (c). A 0.5-mm-thick GaAs:Si (n ¼
5 1017 cm 3) bottom contact layer was initially deposited at 480 8C on a
semi-insulating GaAs (100) substrate, followed by a 100-nm-thick In0.49
Ga0.51P layer lattice matched to the GaAs bottom contact layer. The active
region consisted of ten In0.68Ga0.32As QD layers separated by 35-nm-thick
In0.49Ga0.51P barriers. Finally, an InGaP layer followed by a top GaAs:Si
contact layer identical to the bottom ones were deposited at 480 8C to
complete the structure. The InGaP barriers of device A [see Fig. 19.23(a)]
Figure 19.22 PL spectra of the two samples described in Figs. 19.21(b) and (c), whose
upper InGaP barrier was grown (a) at 400 8C (dashed line) and (b) at 2 different
temperatures: 440 and 480 8C (continuous line) (reprinted from Ref. 42).
464 Chapter 19
Figure 19.23 Schematic diagram of QDIP structures with an active region grown by
(a) single-step barrier growth and (b) two-step barrier growth. AFM images of the surfaces of
the first period of a QD layer capped with the InGaP barrier and top surface of the devices
are shown to the right of the diagrams (reprinted from Ref. 42).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 465
were grown at the same temperature as the QDs themselves (440 8C), by what
is called ‘single-step barrier growth.’ In device B [see Fig. 19.23(b)], 6 nm of
InGaP material was deposited at 440 8C, followed by 29 nm of InGaP grown
at 480 8C, by ‘two-step barrier growth.’ The morphology of the samples was
investigated with a Dimension 3100 AFM operating in ambient conditions
and tapping mode. Figure 19.23 shows the schematics of the device structures
and AFM images of the surfaces of the first period of a QD layer capped with
the InGaP barrier and top surface of the two devices.
The spectral response and peak responsivity measured at 120 K are
shown in Fig. 19.24 for the two devices. Both devices had a response peak
around 4.7 mm, which was consistent with the fact that the QDs
were always directly covered by an InGaP cap layer deposited at low
temperature. The spectral width Dl/lpeak was 7% and 19% for devices A
and B, respectively. This increase was most probably related to the higher
temperatures used for the second part of the InGaP barrier of device B. The
peak responsivities at 120 K are also shown in Fig. 19.24. At low bias, the
signal varies by several orders of magnitude, consistent with a bound-to-
bound transition and with the extraction of the photoexcited carriers out of
the QDs by voltage-assisted tunneling. Device B exhibited a much higher
responsivity than device A at positive bias. Under such experimental
conditions, the photoexcited electrons escape from the QDs into the upper
barriers that are of better quality in devices B as a consequence of the
higher growth temperature.
The noise of the dark current was measured at 120 K and is shown in
Fig. 19.25. A floor around 4 10 4 A was observed in all of the curves and
was due to the instrumental limitations of the experimental setup. At negative
bias, the dark current and noise of device B were significantly lower than those
Figure 19.24 (a) Normalized spectral response of QDIP devices A and B at 120 K.
(b) Peak responsivity of the QDIPs as a function of bias voltage at 120 K (reprinted from
Ref. 42).
466 Chapter 19
Figure 19.25 (a) Noise of the dark current as a function of bias voltage at 120 K. At
low bias, the signal is limited by the experimental setup around 4 1014 A. (b) Specific
peak detectivity of the QDIPs as a function of bias voltage at 77 K and 120 K (reprinted from
Ref. 42). (See color plate section.)
of device A, at both 77 K (not shown here) and 120 K. The noise of device A
takes off from the floor around 0.5 V, whereas the noise of device B starts to
increase only below 1.5 V.
The noise and detectivity are shown in Fig. 19.25 for devices A and B at
77 K and 120 K. It is worth noting that, in this kind of device, the maximum
peak detectivity value always corresponds to the onset of the noise floor
because it was at that point that the ratio of the peak responsivity over the
noise was maximized. The highest peak detectivity of sample B at 77 K was
3.35 1012 cmHz1/2/W with responsivity of 3.4 A/W at 1.9 V, which was
three times higher than 1.13 1012 cmHz1/2/W with responsivity of 1.15 A/W at
0.9 V of sample A. The improvements can be seen more clearly in higher-
temperature tests because we can leave the instrument-limited regime. The
highest peak detectivity at 120 K is still 4.5 1011 cmHz1/2/W with responsivity
of 0.53 A/W at 1.5 V of sample B. This was more than one order of
magnitude higher than 5.75 1010 cmHz1/2/W, 186 mA/W at 0.6 V of sample
A. The high detectivity was attributed to the lower noise in high-bias regions
brought by the double-barrier structure.
The BLIP temperature of the QDIP was found to be 220 K under a FOV
of 45 deg by comparing temperature-dependent dark-current measurements19
with the photocurrent generated by a 300-K background, as shown in
Fig. 19.26.
The QE h was determined to be 1.0% at 77 K. The QE of QDIPs is known
to be low, but it is generally compensated by a very high photoconductive gain.
Such low QE is generally attributed to the small fill factor and to the low
oscillator strength of the principal ISB transition involved in the optical process.
However, it could potentially be considerably improved by incorporating the
QDIP structure into a resonant cavity.
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 467
Figure 19.26 Dark current at various temperatures and 300-K background photocurrent
(reprinted from Ref. 21). (See color plate section.)
of the InAs/InP QDs will expand our understanding of the basic mechanisms
of dot formation, enrich the existing QD systems such as InAs/GaAs and Si/
Ge,50 and will enable devices with new applications or better performances.
The lattice mismatch between the QD material and the substrate—a critical
parameter in determining the formation of the QDs—is only roughly half
(3.2%) that of the InAs/GaAs system (7%). Different matrix materials such as
InP, InGaAs, and AlGaAs can be used to modify the dot formation and
engineer the bandgap.51 In the case of QWIPs, high-performance devices have
been demonstrated with the InGaAs/InP system grown on InP substrates
using MOCVD.52,53 The high mobilities and low effective masses of the
InGaAs/InP system give rise to high responsivity and long-wavelength
devices. Additionally, when compared to MBE, MOCVD has advantages
such as relative simplicity, high adaptability to industrial fabrication, and
lower cost.54
Figure 19.27 AFM image (1 mm 1 mm) of a single layer of InAs QDs on GaAs/InP
matrix. The scan shows a dot density of about 4 1010 cm2 (reprinted from Ref. 21).
a dark current below the pico-ampere range was observed between 0.8 V
and 1.8 V. An asymmetric I-V relationship was also observed in this QDIP,
especially at lower temperatures. The asymmetry of the dark current for
our QDIP can be attributed to several factors. First, it could be due to
the asymmetry of the device structure, as every GaAs/InAs/AlInAs/InP period
of the active region is asymmetric. Secondly, the InAs QD has a lens
shape that is not symmetric in the growth direction. Thirdly, dopant diffusion
into the QD layer during the material growth might create a small built-in
electric field.
The spectral response of the QDIP peaked at 6.4 mm with a cutoff at
6.6 mm (see the inset in Fig. 19.29). The spectral width Dl/lpeak was 12%,
which indicated bound-to-bound ISB absorption. The shape, peak, and cutoff
of this QDIP showed negligible change with varying temperature (from T ¼
77 K to T ¼ 160 K) and bias (from Vb ¼ 1 V to Vb ¼ 3 V). The absolute
magnitude of the blackbody responsivity Rbb was determined by measuring
the photocurrent Ip with a calibrated blackbody source at 800 8C. The test
mesa was illuminated from the top of the mesa with normally incident
infrared radiation. The peak-responsivity Rp results at both 77 K and 100 K
are shown in Fig. 19.29. An asymmetry was also observed for the peak
responsivity, at both 77 K and 100 K. The asymmetry of the responsivity was
caused by the asymmetry of the potential in the QD itself. At T ¼ 77 K and
bias of 5 V, a peak responsivity of 1.0 A/W was observed for our QDIP.
The noise current In was measured at both T ¼ 77 K and T ¼ 100 K. As
shown in Fig. 19.30(a), very low noise was observed and was almost constant
Figure 19.29 Peak responsivity as a function of bias for the QDIP at 77 K and 100 K. The
inset shows the relative spectral response measured at bias of 1 V and temperature of 77 K
(reprinted from Ref. 21).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 471
Figure 19.30 (a) Noise current as a function of bias at 77 K and 100 K. (b) Peak detectivity
as a function of bias at 77 K and 100 K (reprinted from Ref. 57).
for biases 1.0 V < Vb < 2.5 V at 77 K and 0.8 V < Vb < 2.3 V at 100 K.
Beyond these ranges, the noise was dominated by G-R noise of the dark
current.34 Based on the measured noise current and dark current Id, we
extracted the gain g of the device, which strongly depends on the bias and
increases from 5 to 5000 when the bias changes from 1.2 V to 3.5 V. The
high value of the gain in a given voltage range was a unique feature of this
QDIP technology. The gain in our QDIP devices was much higher than the
gain measured in typical QW infrared detectors. The detectivities of our QDIP
as a function of bias at both T ¼ 77 K and T ¼ 100 K are shown in Fig. 19.30(b).
The highest detectivities of our QDIP were 1.0 1010 cmHz1/2/W at a bias of
1.1 V and 2.3 109 cmHz1/2/W at a bias of 0.9 V for T ¼ 77 K and T ¼ 100 K,
respectively. The internal QE h of our QDIP was less than 0.1%. The very low
472 Chapter 19
QE was due to the low oscillator strength for s-polarized light and was less-than-
unity fill factor of the QD layer. A low oscillator strength must be expected for
s-polarized light in a flat (50 nm by 5 nm) lens-like QD at this wavelength. The QE
can be improved by increasing dot density, optimizing dot shape, size, and
uniformity,29 and by using a resonant cavity.56
conditions the same. 50, 100, and 150 sccm flow rates of TMIn were used. The
dot density was checked with AFM. Although 150 sccm gave slightly higher
dot density, PL showed that 100 sccm has better quality. Thus 100sccm was
used in the following optimization steps.
Next, the V/III ratio was optimized by varying the flow rate of dilute
arsine. Flow rates of 25, 50, 75, 100 and 200 sccm were used. 100 sccm was
used for TMIn, as previously determined. 50 sccm of dilute arsine gave the
highest dot density and was chosen for the next growth.
The ripening time is the wait time after the deposition of the dot material
but before capping is carried out. It was used to allow the dot to fully form, or
to “ripen.” 60-, 30-, 10-, and 0-s ripening times were used to study the effect on
dot formation. For this optimization, the above single-layer QD growth
cannot be used because during the cooling down after growth, the QD on the
surface may still change. As a result, AFM is no longer a feasible
characterization technique anymore. Instead, the QDs were capped and then
studied using PL. From the room-temperature PL of these samples, as shown
in Fig. 19.31, it was seen that the ripening time had a dramatic effect on the
QDs. With decreased ripening time, there was a blue shift of the peak
wavelength (indicating smaller dots), reduced FWHM (indicating improved
uniformity), and higher intensity (indicating higher dot density.)
While PL provides useful information on the effect of ripening time on
the interband transitions in the QDs, it does not reveal information about the
ISB transitions, which are directly related to QDIP operation. Instead, the
measurement of absorption was done by an FTIR spectrometer. QDIPs with
peak detection wavelengths from 4 to 6.4 mm were obtained by choosing the
Figure 19.31 Room-temperature (RT) PL from a single layer of InAs QDs grown and capped
with InP at 440 8C with different ripening times (reprinted from Ref. 57). (See color plate section.)
474 Chapter 19
growth temperature and ripening time. At 500 8C, the minimum wavelength
that could be reached was around 6 mm, even with a 0-s ripening time. At
440 8C, 4 and 5 mm QDIPs were obtained with 10 and 60 s ripening times,
respectively. Therefore, to obtain MWIR QDIPs, at least the QD and its
adjacent layers had to be grown at 440 8C.
Finally, with all of the growth parameters of QDs optimized, a single-
layer InAs QDs on top of the GaAs/InP was obtained with a dot density of
about 3–5 1010 cm2. The InAs QDs had a lens shape with a typical base
diameter of 50 nm and a height of 5 nm for the QDs grown at 500 8C, and a
height of 4 nm for the QDs grown at 440 8C.
A QDIP device structure will consist of multiple layers of QDs. In order to
guarantee the uniformity of the QDs in the different layers as well as improve
the quality of the barrier material, a two-step growth of the InP barrier after
each QD layer was used: first the InAs QDs were covered with 10 nm of InP
grown at the same temperature as the QDs, then the temperature was increased
to grow the rest of the barrier. AFM showed that the InP barrier surface was
much improved, with clear atomic steps, while PL showed no change of peak
wavelength or FWHM, which indicates that this process did not affect the QDs.
We compared three devices (here denoted as A, B, and C) with different
QD growth conditions and different structures. The typical device structures
were as follows. First, a 0.5-mm-thick undoped InP buffer layer was grown at
590 oC, followed by a 0.5-mm bottom InP contact layer doped with dilute
SiH4 to n ¼ 1 1018 cm 3, followed by the active region with 10 QD
layers separated by InP barriers, and finishing with a 0.2-mm-thick top
Ga0.47In0.53As contact layer doped with dilute SiH4 to n ¼ 5 1017 cm 3. The
difference between these three devices was only in the active region, as shown
in Fig. 19.32. Device A was based on our previous 6.4-mm QDIP and acts as a
reference. Its entire active region was grown at 500 8C. The detailed structure
of device A is described in Ref. 21. The active regions of devices B and C were
Figure 19.32 Schematic diagram of the three QDIP structures with different active regions
(reprinted from Ref. 57).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 475
Figure 19.33 Normalized spectral response of QDIP devices A, B, and C (reprinted from
Ref. 57).
476 Chapter 19
Figure 19.34 Peak responsivities as a function of bias at 77 K (reprinted from Ref. 60).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 477
Figure 19.35 Schematic of the detector device structure (reprinted from Ref. 61).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 479
without any optical coupling structures. In this device structure, both the
InAs QD layers and GaInAs QW layers were involved in the infrared
absorption process. The coupling of QDs and QWs has been used in other
QDIP device structures, such as dot-in-a-well (DWELL),20,62 where the
ISB transition occurred between the hybrid states of the QD and the QW.
The device here differed from the typical DWELL design in that the QW
was only on the top of the QDs, as opposed to the dot being surrounded by
the QW. Figure 19.36 shows the photocurrent spectra of the device at
various operating temperatures. The spectrum was clearly measurable at
280 K. For applied bias of smaller than 2 V, there were two peaks in the
detection spectrum, around 3.2 and 4.0 mm. Both peaks were visible in the
two high-temperature spectra in Fig. 19.36. The two peaks were present at
lower temperatures as well, but in the spectra in Fig. 19.36 the more-
intense longer-wavelength peak drowned out the shorter-wavelength peak.
Also, measurements showed that the intensity of the peak near 3.2 mm did
not increase significantly as the temperature increases. The peak around
3.2 mm came from a bound-to-continuum transition where the electrons
are photoexcited from the ground state to a continuum state. That was
the reason that the increase in temperature does not improve the
photoresponse around 3.2 mm. On the other hand, the photoresponse
around 4.0 mm increased significantly with the temperature because it
came from a bound-to-bound transition in the InAs QD/GaInAs QW
hybrid states; thus, the temperature can help the photoexcited electrons
escape to the continuum. Both the temperature dependence and the bias
sensitivity of the peak near 4.0 mm indicated that the transition of the
photoexcited electrons took place between bound states of the QD/QW
hybrid. Looking more closely at the 4.0-mm peak, it showed a red shift and
broadening with increased operating temperature. This behavior was similar to
that observed in QWIP structures, as documented in Ref. 63. A model of the
QD/QW hybrid system is required to gain an understanding of where this
effect comes from and whether the causes are similar to those in QWIP
structures.
The peak responsivity Rp was measured as a function of bias and
temperature and is shown in Fig. 19.37. The responsivity increased with
temperature from 77 to 200 K and started decreasing above 200 K. The
highest measured peak responsivity was 778 mA/W at 150 K and 5 V.
The peak responsivity at 250 K was 54 mA/W at 2 V. In QDIPs or QWIPs,
the photocurrent can increase or decrease with the temperature, depending
on the balance of the competition between phonon-induced relaxation of the
photoexcited carrier and temperature-aided escape of the photoexcited carrier
to the continuum state. Above a certain temperature, the adverse thermal
increase of the relaxation of the photoexcited electrons back to the lower
state dominates any improvement in escape created by the temperature.58
In the device, the data showed that turnover took place around 200 K,
above which the responsivity started decreasing with increasing temperature.
Figure 19.38 Dark current as a function of bias at several temperatures (reprinted from
Ref. 61).
482 Chapter 19
Figure 19.40 FPA imaging at an operating temperature of (a) 130 K and (b) 200 K
(reprinted from Ref. 61).
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 483
imaging tests were carried out at a fixed frame rate of 32.64 Hz and
depending on the operating temperature with the bias varying from 1 V to
around 3 V and integration times from 0.34 ms up to 30.41 ms. Two-point
nonuniformity correction was used for the imaging. Low-contrast imaging
of human targets was possible up to around 150 K, and a soldering iron
could be imaged up to 200 K. Above 200 K, the dark current of the detector
became too high for any imaging. While the single-detector measurements
showed that the responsivity was still appreciable at temperatures up to
room temperature, the dark current set the upper limit on the operating
temperature because at temperatures above 200 K the dark current saturated
the ROIC for any integration times long enough to collect any appreciable
photocurrent signal. The capacity of the ROIC used was 18 million electrons
at its lowest gain setting. In order to operate the FPA at even higher
temperatures, the detector dark current should be further reduced or an
ROIC with a larger electron capacity could be used.
The FPA photocurrent was measured using an extended-area blackbody
source from CI Systems. The photocurrent of the FPA was measured with the
extended-area blackbody set at 35 8C. The responsivity and conversion
efficiency could then be calculated by dividing the measured photocurrent by
the appropriate optical input from the system radiometry. The mean
responsivity and mean conversion efficiency of the FPA were 34 mA/W and
1.1%, respectively. The temporal noise of the FPA was measured by taking
the standard deviation of the FPA signal. At 120 K, the average NEDT was
344 mK and the percentage of connected pixels was greater than 99%. The
histogram of the NEDT is shown in Fig. 19.41. The number of pixels with
NEDT within 2 standard deviations of the mean NEDT was 91%. It should be
noted that the NEDT measurements were taken without two-point non-
uniformity correction. The injection efficiency was 99% at 120 K. At 200 K, the
injection efficiency was still 98% due to the high differential resistance, even at
high temperature.
The imaging quality in terms of contrast and uniformity for this FPA is
not yet comparable to that of high-quality QWIPs. From the single-device
measurement, the optimum bias condition for this detector was at (or beyond)
the limit of the test setup’s biasing capability: 5 V. However, since the ROIC
cannot apply more than 3 V nominally, the detector had to be operated
under conditions where the photocurrent was at such a low level that it was
barely detectable by the camera system. The same low photocurrent limitation
also meant that operating the FPA at 77 K actually gave lower performance
compared to 120 K because the detector’s responsivity decreases with
temperatures below 180 K. This low responsivity was the reason that the
NEDT was 344 mK even though the corresponding single-detector D was
greater than 1010 cmHz1/2/W under similar operating conditions. The detector
properties were no longer dominant, and any noise and nonuniformity in the
484 Chapter 19
Figure 19.41 Histogram of the FPA NEDT at an operating temperature of 120 K (reprinted
from Ref. 61).
ROIC and the camera system itself will be more significant at these low signal
levels. Of course, any nonuniformities in the detector material and processing
will also still factor in. As a result, one important step for the application of
this detector structure to FPAs will be optimizing the device to run at a lower
bias without sacrificing its high SNR and QE. The new single-device results
reported here still require a high operating bias; thus, the device is not suitable
for our particular imaging application.
We have demonstrated a high-performance InAs QD mid-infrared
photodetector grown on InP substrate that operates up to room temperature.
The peak detection wavelength was observed at 4.0 mm. The peak responsivity
and the specific detectivity at 150 K were 778 mA/W and 6 1010 cmHz1/2/W,
respectively. Low dark-current density and a high QE of 48% were obtained in
this device. A high-temperature FPA was also demonstrated. Imaging was
achieved at temperatures up to 200 K, with human body imaging up to 150 K.
The very low dark current of the device enabled the high-temperature imaging
capability.
19.7 Conclusion
QDs are extremely interesting nanostructures to study both from a basic
physics standpoint and a device technology standpoint. In infrared detector
applications in particular, QDs show great potential to enable the next
generation of devices with high QE and high operating temperature.
Quantum Dots for Infrared Focal Plane Arrays Grown by MOCVD 485
References
1. M. Razeghi, Fundamentals of Solid State Engineering, Springer, New York
(2010).
2. M. Razeghi, W. Zhang, H. Lim, and S. Tsao, “Quantum Dot Infrared
Photodetectors by Metal-Organic Chemical Vapour Deposition,” in Hand-
book of Self-Assembled Semiconductor Nanostructures for Novel Devices in
Photonics and Electronics, M. Henini, Ed., Elsevier, Amsterdam, p. 621 (2011).
3. K. Mukai and M. Sugawara, Semiconductors and Semimetals 60,
Academic Press, San Francisco, Chapter 5 (1999).
4. L. Goldstein, F. Glas, J. Y. Marzin, et al., “Growth by molecular beam
epitaxy and characterization of InAs/GaAs strained-layer superlattices,”
Appl. Phys. Lett. 47(10), 1099–1101 (1985).
5. Y. Arakawa and H. Sakaki, “Multidimensional quantum well laser and
temperature dependence of its threshold current,” Appl. Phys. Lett. 40,
939–941 (1982).
6. M. Asada, Y. Miyamoto, and Y. Suematsu, “Gain and the threshold of
three-dimensional quantum-box lasers,” IEEE J. Quantum Electron. 22,
1915–1921 (1986).
7. M. Razeghi, Technology of Quantum Devices, Springer, New York (2009).
8. A. Rogalski, Infrared Detectors, CRC Press, Boca Raton, FL (2010).
9. P. Martyniuk and A. Rogalski, “Quantum-dot infrared photodetectors:
Status and outlook,” Prog. Quantum Electron. 32, 89–120 (2008).
10. H. Schneider and H. C. Liu, Quantum Well Infrared Photodetectors:
Physics and Applications, Springer Series in Optical Sciences 126,
Springer-Verlag, Berlin Heidelberg, p. 46 (2007).
11. E. L. Dereniak and G. D. Boreman, Infrared Detectors and Systems,
Wiley-Interscience, New York (1996).
12. A. Rogalski, “Infrared detectors: status and trends,” Prog. Quantum
Electron. 27(2–3), 59–210 (2003).
13. A. Piotrowski, P. Madejczyk, W. Gawron, et al., “MOCVD HgCdTe
heterostructures for uncooled infrared photodetectors,” Proc. SPIE 5732,
273–284 (2005) [doi: 10.1117/12.593337].
14. A. Rogalski, “Quantum well photoconductors,” J. Appl. Phys 93(8), 4355
(2003).
15. B. F. Levine, “Quantum-well infrared photodetectors,” J. Appl. Phys.
74(8), R1–R81 (1993).
16. S. D. Gunapala, S. V. Bandara, J. K. Liu, et al., “1024 1024 pixel mid-
wavelength and long-wavelength infrared QWIP focal plane arrays for
imaging applications,” Semicon. Sci. Technol. 20, 473–480 (2005).
486 Chapter 19
20.1 Introduction
20.2 Background Theory: How SiAPDs Work
20.3 Design Challenges of NIR Detectors
20.3.1 Modeling and simulation
20.3.2 Fabrication: standard and dedicated CMOS process
20.3.3 Premature edge breakdown (PEB) effects
20.3.4 APD structure
20.4 SiAPDs Circuitry Design
20.4.1 Circuitry required for SiAPD based front ends
20.4.2 Linear mode SiAPD front end
20.4.3 Geiger mode SiAPD front end
20.5 Optimally Adaptive Control for Low Noise, Low Power, and Fast Photodetection
20.6 Conclusion
Acknowledgment
References
20.1 Introduction
Infrared sensors have been available since the 1940s to detect, measure, and
monitor the thermal radiation emitted by objects.1,2 Silicon avalanche
photodiodes (SiAPDs) are a potential candidate for low-level light detection,
especially in the visible and near-infrared (NIR) regions due to their bias-
dependent internal gain and their ability to amplify the photogenerated
signal by avalanche multiplication. SiAPDs became popular for several
491
492 Chapter 20
Table 20.1 Performance comparison of current solid state photon detectors [HV – high
voltage, FPN – fixed pattern noise (defined as variation of the pixel response at one moment
of an image sensor in the dark), EMI – electromagnetic interference, PDE – photon detection
efficiency, ph.e. – number of photoelectrons].
Figure 20.1 Structure of (a) a p–i–n diode and (b) a reach-through SiAPD. Their charge
densities and electric field distributions are included below each diagram.
496 Chapter 20
absorption depth, thus improving the absorption in the red and IR regions of
light. It also decreases the junction capacitance, thereby reducing the
resistor-capacitor (RC) delay constant.
The transit time of carriers drifting across the depletion region along with
the RC time response are the main sources of the response time of photodiode
detectors. The response time of p–i–n photodiodes can be as low as a few
hundred picoseconds.23 It should be noted that in the case of silicon, the
maximum responsivity occurs on wavelengths that are substantially shorter
than the bandgap wavelength. This is due to the indirect-gap nature of silicon,
where the photon absorption transitions typically take place from valence- to
conduction-band states that are above the conduction-band edge.24 The APD
can be considered as a modified p–i–n junction, as depicted in Fig. 20.1(b).
The basic structure of an APD contains two regions: the photon absorption
region and the multiplication region. The photon absorption region absorbs
incoming photons and converts them to EHPs. Then, the generated primary
charge carriers travel to the multiplication region, which is usually formed by
a p–n or a p–i–n junction.
The APD must be inversely biased. This creates a depletion region and
therefore a high electric field across the junction (due to the space charge). The
electric field in the multiplication region will accelerate any primary charge
carrier that enters this region and will increase its kinetic energy. If this energy
is high enough, the accelerated carriers (electron or hole) can excite new EHPs
by transferring a part of their energy into the lattice. The newly generated
EHPs also will be accelerated due to the electric field of the multiplication
region and will create more and more EHPs in this region. This impact
ionization process constructs the internal avalanche gain property and causes
a single photon to be multiplied, introducing an avalanche current.
The photogenerated EHPs can be generated due to two different processes:
(1) the band-to-band transitions, which are called intrinsic excitation, or
(2) the forbidden-gap energy-level transitions, which are called extrinsic
excitation. Intrinsic excitation is the desired process and uses well-controlled
growth processes that are the dominant mechanism contributing to the
photon absorption process.25 27
The incoming photons, whose energy is higher than the semiconductor
bandgap energy, can excite the electron to be transferred from the valence
band to the conduction band and introduce photogenerated electrical current.
The relation between the semiconductor bandgap energy (Eg) and the long
cutoff wavelength (lmax) can be defined as: Eg (eV) ¼ 1.24/ lmax (mm). For
APDs, internal avalanche gain is generated through the impact ionization
process or the so-called avalanche multiplication process. When a free charge
carrier enters the high field multiplication region, with a certain probability, it
initiates the avalanche process, during which the primary carrier can be
multiplied. Then, at a certain point, the ionization chain breaks as the last free
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 497
carrier exits the multiplication region without exciting subsequent new EHPs
due to statistical fluctuation of the impact ionization process. For an
avalanche initiated by a single carrier, multiplication gain M is defined as the
total number of EHPs generated during the multiplication process.
APDs have a bias-dependent internal gain, which makes them compatible
for low-level light detection in the visible and NIR regions. The bias of an
APD just near but below a breakdown voltage Vbr is referred to as being in
linear mode (LM) operation. At this bias voltage, the gain is high, and the
output signal is proportional to the amount of scintillation light interacting in
the APD. A schematic illustration of avalanche multiplication is depicted in
Fig. 20.2. In linear mode, only one type of carrier (electron or hole) with a
higher ionization coefficient takes part in the avalanche process. In Geiger
mode, both electrons and holes produce new EHPs that take part in the
avalanche process and in promoting a self-sustained avalanche.26 28
Figure 20.4 Device simulation of the SiAPD under reverse bias: (a) the electric distribution
without a ring guard. (b) Using a ring guard, the maximum electric field distribution is mainly
located at the active region of the device under reverse bias. (See color plate section.)
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 499
Table 20.2 Simulated characteristics of the proposed SiAPDs using TCAD and 3DTCAD.
Reference SiAPD
applied device simulation tool and the inflexibility of the standard CMOS
technology in precisely simulating and implementing the desired character-
istics. The inefficiency of the applied modeling techniques along with the
complexity of the fabrication process can lead to a considerable number of
nonfunctional APDs after fabrication (e.g., only 28% functional APDs
according to Ref. 24).
punch through of the p-well guard ring to the p substrate.11 A similar guard-
ring structure can also be created from the substrate doping by bringing rings
of n well into close proximity of each other.5 This device cannot be scaled
much below a 5-mm diameter10 because the p-well regions become so close
that the active area of the APD is almost fully depleted. The highest electric
field no longer occurs at the pþ/n-well interface, and the APD performs as a
planar p-well n-well diode. APDs with guard rings formed by either n-well
spacing in p substrate or p-well implants in high-voltage wells were successful
at providing low DCR.5
Another approach is to use triple-well process steps and shallow-trench
isolation (STI) as a guard ring to withstand the high electric fields between the
anode and cathode.12 Here, the edges of the drain implant are confined by the
oxide trench, and formation of the curved edges is prevented. As a result, a
uniform field is achieved more compactly than with a diffused p-well ring.
However, STI dramatically increases the density of deep-level carrier
generation centers at its interface.8 Since the active region of the SiAPD is
in direct contact with the isolated trench, the injection of free carriers into the
sensitive region of the detector results in a very high (1 MHz) DCR, due to
the traps at the Si–SiO2 interface, and degrades the performance of the
SiAPD. Even by considering the PEB effect and applying different PEBP
techniques, only a few percent of fabricated APDs using standard CMOS
technology are functional, even with proper design specifications interpreted
from theory and device simulation results obtained before fabrication.49,50
In Ref. 34 we studied the most commonly applied PEBP techniques as
well as techniques for APD fabrication using standard CMOS technology and
proposed a new, practical, efficient design technique for fabricating a
functional APD based on our simulation and fabrication experiences.34,51
Inefficiency of the applied PEBP techniques and insufficiency of device
simulation results can lead to a low performance or nonfunctional SiAPD and
a significant discrepancy between design simulation and measurement
results.34 In addition to the doping profile and electric field distribution in
the SiAPD structure, the punch-through, high-tunneling, and PEB effects
need to be considered to design a high-quality SiAPD using a standard CMOS
process. The sharpness of the transition between band-to-band tunneling and
the avalanche is reported as a critical design criterion.
New APD structures are proposed and implemented using the relatively
low-doped layers available in standard CMOS process technology. These
structures are modified to have a most efficient PEBP to reach maximum
efficiency for low-noise and low-intensity light detection in the NIR
wavelength region to be applied in medical imaging applications such as
functional near-infrared spectroscopy (fNIRS). In order to boost the quality
of the APDs, we need to apply a wavelength-specific SiAPD design
procedure. CMOS SiAPDs with wide depletion region (>10 mm) are
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 503
Figure 20.5 Cross section (first row), electric-field distribution (second row), and I-V
characteristics (third row) of the implemented (a) p-well, (b) p-sub, and (c) n-well SiAPDs
(second-row figures adapted from Ref. 34). (See color plate section.)
504 Chapter 20
and octagonal shapes. These provide a tradeoff between fill factor and
angularity and a feasible way to validate the efficiency of the applied
guard rings. The p-well APD is a pþ/n-well APD developed in square
and octagonal shape with a p-well guard ring to prevent PEB. The guard
ring is realized by low-doped p well around a pþ active area with 100-mm
diameter. In p-substrate (p-sub)-based guard-ring APDs, no explicit guard
ring is added. In other words, spacing between the active region and n well is
achieved by substrate intrinsic doping. The n well is cleaved into two n tubs
separated by a small interval (d 0. 9 mm) constituting the guard ring. This
structure is based on connecting n wells surrounding an island of substrate
as a guard ring. Deep n wells can connect two n wells and isolate the guard
ring between the n wells. By careful positioning of doping profiles, two
separate n wells can also be connected directly without using deep n wells.
We developed this APD with two different active areas of 100 mm 100 mm
and 400 mm 400 mm.
The device simulations are performed using TCAD and Spectra imaging
to find the best dimensional values that minimize PEB. In order to analyze
APD performance, we considered both the extension of p/nþ in the guard
ring (active diffusion in the guard ring) for PEBP and the polysilicon width
to control the separation between the active region and the guard ring.
Electric field distributions show the maximum electric field occurring in the
active regions, a necessity for GM operation. The electric field distribution
of the device under reverse bias was verified. The simulation results also
show that the maximum electric field (160 kV/cm) appeared in the active
pþ/deep-n-well junction and that the device is able to withstand the electric
field without breakdown. In order to prevent the high electric field at the
connection between n wells, an occurrence that can cause edge breakdown,
we modified the doping profile in these regions. These APDs are fabricated
by TSMC (Taiwan Semiconductor Manufacturing Co.) via CMC (Canadian
Microelectronics Corp.) microsystems using standard CMOS 0.35-mm
technology.
Here, with an n-well PEBP technique, we designed another new APD
that is an nþ/p-sub SiAPD. It uses the connection between the highly doped n
region and the substrate as the active region. Figure 20.5(c) shows the cross
section of an n-well-based guard-ring APD in 350-nm technology.
The electric field distribution strength in the cross section of the APD
(when Ve ¼ 1 V) at different depths in the absence of any radiation is
depicted in Fig. 20.5(c). The figure shows that the guard ring is most
effective near the surface of the APD, where most of the photon absorption
takes place, and the maximum electric field (750 kV/cm) is in the active
region, so the device is able to endure the electric field without breakdown.
In order to avoid high DCR, the excess bias Vex must be kept sufficiently
low, at the cost of a lower PDP. Furthermore, high tunneling is expected in
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 505
Figure 20.6 I-V characteristics of the proposed SiAPDs. The n well2 in (b) is gained with
higher doping in thicker epitaxial compared to the n well. (See color plate section.)
high Vex values due to the applied relatively highly doped substrate in the
APD. This APD suffers from a relatively high dark current (120 nA,
generated by tunneling and Shockley–Read–Hall processes), which may
impose an insignificant power consumption increase in high illuminations.
The sharp transition between the band-to-band tunneling region and the
avalanche region makes the n-well APD suitable for GM operation (Fig. 20.6).
Figure 20.7(a) depicts more clearly the quantified values of electric field
distribution of the n-well guard-ring-based APD. Figure 20.7(b) shows the
DCR versus excess bias voltage in different temperatures of the n-well APD.
The PDP versus excess voltage and in different wavelengths are compared for
all proposed APDs. Among the three families of structures (p-well, p-substrate,
and n-well guard ring), only the APD with an n-well-based guard-ring structure
could be biased in GM because the other structures suffer from drawbacks such
as punch through, high tunneling, and PEB. We also examined the effect of
doping and scale variations in order to optimize the APD performance.
Figure 20.7 (a) Electric field distribution and (b) DCR of the n-well SiAPD (reprinted from
Ref. 51).
506 Chapter 20
Figure 20.8 Comparison of (a) the PDP of SiAPDs and (b) PDP versus Vex in 650 nm
(reprinted from Ref. 51).
Figure 20.9 Block diagram of the integrated circuit developed for GM and LM SiAPD
operation.
Some other components may be added to amplify, filter, and enhance the
process, but, in general, the linear mode circuitry (the TIA front end) is not
complicated and can be simply a TIA. The circuitries required to operate the
SiAPDs in GM are more complicated and need more dedicated analog and
digital components.
Due to the self-sustained behavior of the avalanche process in GM
(which is also called single-photon counting or SPAD mode), a quench
circuit must be used to suppress the avalanche current. A reset circuit is
also required to prepare the APD for the subsequent avalanche and for
counting the subsequent photons. The quench-reset front end can include
additional active and passive timing control and enhancement components
for different applications. For example, in order to measure the arrival time
of photons, readout systems such as time-to-digital converters or time-to-
amplitude converters are needed. Different active and passive quench-reset
techniques have also been developed for this mode of operation. In active
mode, active circuitries are used to control the process; however, in passive
mode, the avalanche current passively controls the process by way of a
ballast resistance device. In order to achieve the best performance from our
design, the sensitivity of each component presented in the design needs to
be analyzed, and the best values for optimal sizing selected. A high-gain,
low-power, narrow-band, low-noise circuit in LM and a fast, low-noise,
high-efficiency circuit in GM are developed in Refs. 15, 43, 54, and 55. The
block diagram of this integrated circuit and its components are shown in
Fig. 20.9.
photocurrent into a voltage signal.14,54,58 Due to the ultralow level and usually
high source impedance of input signals, this amplifier should be established to
meet certain basic requirements and must cope with various challenges in
order to detect and extract the required signals. The challenges of designing
such a TIA for portable biomedical imaging systems are: high common mode
rejection ratio to reject interference,59 high-pass-filter characteristics for
filtering differential dc offset,60 low noise for high signal quality,61,62 ultralow-
power dissipation (<50 mW) for long-term power autonomy,63,64 configur-
able gain and filter characteristics that suit the needs of different biopotential
signals and different applications, high transimpedance gain (>1 kV),65,66
narrow bandwidth (around 100 kHz), high-output swing, wide dynamic
range, ambient light rejection, and low-voltage operation.67 By adopting a
different topology for the gain stage, the stability of the preamplifier can be
greatly enhanced and the preamplifier circuit significantly simplified.
Searching for a TIA that can be efficiently applied for LM-SiAPD operation,
one finds three prominent TIA structures reported in literature: common-gate
TIA (CG-TIA), resistive-feedback TIA (RF-TIA), and capacitive-feedback
TIA (CF-TIA), as shown in Fig. 20.10.
A new structure that combines a CG-TIA and an RF-TIA is introduced and
developed in Refs. 43 and 54. This new structure offers a low-noise, high-gain,
high-bandwidth (BW) TIA. Furthermore, using a dynamic-threshold voltage
MOSFET in this TIA structure enhances the differential input common-mode
range. The design consists of a current amplifier implemented in a transimpe-
dance configuration. In this circuit [Fig. 20.11(a)], the combination of two
transistors (M6 and M7) works as a feedback resistor to minimize the output
ripple and omit the extra drawn current. The dc transimpedance gain is given by
Vout gM5 Rf 1
¼ , ð20:1Þ
Iin gM4 þ gM5
where gM4 and gM5 are the transconductance of transistors M4 and M5,
respectively, and Rf is the feedback resistance implemented by M6 and M7
Figure 20.10 Different TIA structures: (a) common gate, (b) resistive feedback, and
(c) capacitive feedback (reprinted from Ref. 47).
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 509
Figure 20.11 Schematic of the proposed TIA front ends: (a) TIA1 and (b) TIA2 (OTA ¼
operational transconductance amplifier; LA ¼ limiting amplifier) (reprinted from Ref. 14).
biased in the linear region. The frequency where the loop gain of the TIA is
unity is
A gm1 R1
vt , ð20:2Þ
Rf CD Rf CD
where CD is the photodiode capacitance. The closed-loop bandwidth of the
TIA is approximately equal to the unity-gain frequency:
1 þ A gm1 R1
BW vt : ð20:3Þ
Rf CD Rf CD
The bandwidth of the TIA increases by decreasing CD. We used CD ¼
1 pF in our simulations, as this is the commonly reported value.47,68 In order
to boost the voltage swing and match the output impedance to drive the
photoreceiver output (usually a demultiplexer or demux), we design a
limiting amplifier (LA)69 and an operational transconductance amplifier
(OTA) to be added to the output of the TIA. The OTA used in the proposed
front-end amplifier design is a current-mirror OTA, which is modified from
the OTA reported in Ref. 70. The performance of this OTA largely depends
on the bias current Ibias and the sizing of the transistors. Therefore, we
considered these two parameters in order to attain the best performance.
510 Chapter 20
To increase the maximum output swing and improve the stability of the
circuit, we also used a filtering block followed by the TIA and the LA.
Because one of the main requirements of biosignal amplifiers is to have a
wide dynamic range, we considered the proposed photoreceiver circuit by
adding the ability of parameter tuning. Using the constant applied voltage
Vconst of 0 V < Vconst < 2 V, the gain, bandwidth, power consumption, and
dynamic range of the output could be changed in a desirable wide output
range. Figure 20.11(a) shows the layout of the currently fabricated circuit
using 0.35-mm CMOS technology. Measurements on this integrated circuit
show that the input noise is very low compared to that of the existing TIAs.
p
The measured output noise is 1.8 mV/ Hz. The power consumption of the
front-end circuit is also very low. The TIA has high transimpedance gain and
high bandwidth.
By varying the Vconst in the proposed variable-gain front end between 0
and 1.5 V, we reached the very high and fixed value of 45 109 for the gain–
bandwidth product (GBW). This value is tunable between 10 M and 45 G
for various applications. We can reach the transimpedance gain in the range
of 2–400 MV/A and bandwidth in the range of 1.7–5 MHz using this
configuration. The power consumption of this circuit is in the range of
0–3 mW, which is very convenient for biomedical wireless and portable
applications. The input noise of the proposed design is one of the lowest
reported compared to all similar photoreceiver amplifiers. The measured
p
output noise is also 1.8 mV/ Hz. The output pulse with the maximum swing
of 3.3 V is also one of the best indices to show the appropriateness of the
proposed design. In order to verify the effect of the supply voltage, we also
tested the tunable-gain configuration by increasing the supply voltage to 1 V.
Another TIA structure has been also proposed in Ref. 14 as a four-stage
TIA, as is depicted in Fig. 20.11(b). For cascade TIAs, in order to reduce the
relative noise contribution of the subsequent amplifier stages, the SNR of
the first stage should be tuned to be as large as possible so that the feedback
transistor Mf is placed in series with feedback resistor Rf to reduce the input
noise of the TIA by applying different control voltages via the constant
applied voltage Vconst. One of the main characteristics of biomedical
amplifiers is their ability to be used in an extended dynamic range. Using
Vconst at each stage, the gain, bandwidth, power consumption, and dynamic
range of the output can be changed in a desirable wide output range. The dc
gain of a single stage is
gm1 þ gm2
Ai , ð20:4Þ
gm3
and the bandwidth is calculated by
1 þ Ai
BW ¼ , ð20:5Þ
2pRC
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 511
where A is the gain, and C is the input capacitance of each stage. The
equivalent thermal noise of each stage is
4 KTA2
2
diin ¼ : ð20:6Þ
3 gm R2
In order to achieve better performance, we selected the width-to-
length ratio (W/L) for M2, M4, M5, M7, M8 high and for M3, M6,
and M9 low. The W2 /W1, W5/W4, and W8/W7 are also considered to be
high. The sensitivity of each component presented in this design was
analyzed, and the best values for optimal sizing were selected. This
TIA front end was fabricated using CMOS 0.35-mm technology.
The microphotograph of the fabricated integrated circuit is shown in
Fig. 20.11(b). The transimpedance gain of the proposed circuit is
tunable between 10 KV/A and 250 MV/A, and the bandwidth is also
tunable from dc to 100 KHz and 1 MHz by varying the control voltage
and based on the frequency of the p noise. The total power consumption is
4 mW. The input noise is 200 fA/ Hz at 1 kHz, and the maximum output
swing is 3 V.
Now we will introduce a state-of-the-art logarithmic TIA (LogTIA)
with automatic gain control (AGC) and ambient light rejection for low-
intensity-light detection applications. 71 The LogTIA is practical in
systems that need scale-invariant and wide-dynamic-range operation.
Its sensitivity to the contrast (ac/dc) of the input and its scale-invariant
fractional amplification are beneficial in several applications where
percentage changes rather than absolute changes carry information. This
photoreceptor was inspired by the operation of biological photoreceptors
in turtle cones and has many of the properties of these photoreceptors,
including higher ac gain than dc gain, a contrast-sensitive response, and a
relatively wide dynamic range of operation. Unfortunately, the merit
characteristics of this amplifier, especially for biomedical imaging and
optoelectronics circuits and systems, are not well introduced, and only a
limited application of LogTIA in photodetectors implementation is
reported. In this chapter we introduce the unique characteristics of the
LogTIA as a state-of-the-art front-end circuit for photodetection,
especially in the NIR region of the light spectrum. We have designed
and implemented a new LogTIA to be applied in the fNIRS photodetec-
tor front end. This is the first proposed and successfully designed and
implemented application of LogTIA in a NIR photodetector front end
and in a fNIRS application.
In this proposed circuit (see Fig. 20.12), M1 acts as the logarithmic
amplifier transistor, and M6-M7 provides the feedback resistor. M9 acts as
a feedback transistor placed directly across the input and output terminals
of the current mirror. This direct-feedback topology decreases the input
512 Chapter 20
Figure 20.12 Proposed LogTIA with ambient light rejection and AGC loops (reprinted from
Ref. 15).
impedance seen by the photodiode and improves the speed of the circuit, at
the cost of a lower output swing. Using the logarithmic amplifier also
makes the response to a fixed image contrast invariant to absolute light
intensity and improves the dynamic range of the photodetector. The N1
transistor at the output of the TIA causes the circuit to act as a cascade
current mirror, reducing the output voltage variation by boosting the
output impedance and reducing the VDS-mismatch effect. Using AGC and
dc-rejection feedback, we increased the sensitivity and bandwidth. The
transimpedance gain of the linear TIA increases the sensing speed by
decreasing the time constant such that the rapid changes in the input are not
filtered at the output.
The input voltage Vin is kept at a virtual reference value Vref by the
feedback loop such that it does not change with variation in the input current;
thus, the current variations due to the Vin variation (e.g., due to early effects and
other sensor effects) are minimized. By increasing the power supply, the
dynamic range of the output voltage can be maximized while maintaining the
APD breakdown voltage at the input. In contrast, the LogTIA shown in
Fig. 20.12 uses a subthreshold transistor as the feedback element with an
exponential parameter of ks (the sub-threshold exponential coefficient of Mf)
and a pre-exponential constant of IOS. The key motivation for using logarithmic
instead of linear sensing is that the logarithmic amplifier is inherently sensitive
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 513
to the contrast (ac/dc) of the input photocurrent signal. By assuming Vref and
considering that iin and vout are the small changes in the operating-point current
Iin and voltage Vout , respectively, the output voltage of the LogTIA is
0 1
A=ð1 þ AÞ
KT iin B C
vout ðsÞ ¼ @ sðC in =g f Þ A
q Iin 1þ
1þA
0 1
A=ð1 þ AÞ
KT DIin B C KT Iin
¼ @ sðC in =g f ÞA ¼ Vref þ ln :
q Iin 1þ qks IIos
1þA
dVout KT 1=IIos
vout ¼ iin ¼ iin
dIin qks Iin =IIos
KT iin KT iac, in
¼ ¼ : ð20:7Þ
qks Iin qks IDC, in
Table 20.3 General characteristics of the proposed LogTIA (AGC – automatic gain control,
PRBS – pseudo-random binary sequence).
Parameter Value
Table 20.4 General characteristics of the proposed TIAs compared to those in other
referenced studies [p is pico (1012), f is femto (1015)].
TIA1
TIA 0.35 3 3.5 250 G 0.151 112 f 3 10785714 Gain, BW
TIAþLA 0.35 2 2 250 k 11.5 44.86 p 2 1437.5 Gain, BW
Power
TIAþ 0.35 1 1.5 90 M 0.072 59 f 1 4320 Gain, BW,
OTA þ Power
TIAþLA 3 0.8 400 M 1000 1f 3 500 M Gain, BW,
Power
TIA2 0.35 3 4 250 M 1 200 f 3 62500 Gain, BW
Log TIA 0.18 1 0.04 300 M 1K 5G <10 f 1 7200G Gain, BW
Power, Noise
G BW
*FOM (figure of merit) ¼ [GHzV/mW].
ðpower dissipationÞ noise
Figure 20.13 Schematic of the passive quench circuit in (a) voltage-mode and (b) current-
mode configurations. (c) The SPAD cathode voltage and (d) current plotted versus time in
response to single-photon arrival (reprinted from Ref. 15).
High power dissipation can drift the breakdown voltage and change the
SPAD response to detection efficiency and noise. Delayed release of trapped
charges due to the large charge trapping in SPADs can retrigger the detector
and cause false ignitions, called after-pulsing, which causes a nonlinear
distortion in photon counting.31 Decreasing the avalanche time duration
reduces the power dissipation, charge trapping, and optical crosstalk due to
the minimization of the hot-carrier photon emission.31,88 Because of the
complementary action of the active quenching in MQCs, there is more
flexibility in choosing the passively quenched load (RL) in order to further
suppress the initially passively quenched avalanche. Therefore, by reducing
the load resistor RL, one can achieve quicker detection of the photon.89 By
increasing the light intensity received by the SPAD, the current flow through
the diode and the series-connected resistor RL will also increase. The resulting
increase in voltage drop across the RL decreases the bias voltage across the
SPAD so that the gain of the SPAD is reduced. Therefore, the dynamic range
of the SPAD optical input is increased for a fixed dynamic range of electrical
output.90
In order to operate the SiAPD in GM for single-photon counting, a new
controllable MQC with the capability of real-time control of the hold-off time
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 517
Figure 20.14 Schematic diagram of (a) the controlled MQC and (b) the SPAD cathode
voltage in response to single-photon arrival and (c) multiple-photon arrivals between 35 and
80 ns (reprinted from Ref. 47).
518 Chapter 20
quench transistors (Squench1 and Squench2) ON via Sfeedback. This starts the
active quenching action by quickly pulling the SPAD cathode down to
ground. This brings the reverse bias of the SPAD below breakdown, and the
avalanche current quickly dissipates. The quench transistors (Squench1 and
Squench2) are then turned OFF, and the three parallel reset transistors (Sreset1,
Sreset2, and Sreset3) are turned ON. The reset transistors are activated by an
output pulse from the reset monostable, which triggers the end of the hold-off
period. These reset transistors are equivalent of the three low-resistance
transistors and reset the quiescent bias of the SPAD, bringing the SPAD
cathode voltage back to detect the next photon. The short duration of the
reset-time decreases the dead-time between photon counts.93 Figures 20.14(b)
and (c) show the SPAD cathode voltage in response to the single- and
multiple-photon arrivals. It is important to note that in this circuit the excess
voltage Vex is given by
Vex ¼ Vdd þ jVop j Vbr , ð20:8Þ
where Vop is the operating voltage.
The performance of the SPAD is usually related to the excess electric field
above the breakdown voltage,94 so the excess bias voltage has a significant
influence on detector performance. Therefore, the MQC can also improve the
performance comparing to the PQC. Using this circuit, faster quenching
results in lower power loss and hence less heating of the SPAD.
Figure 20.15 Schematic diagram of the proposed controlled MQC (CMQR), which
includes an on-chip signal processing unit for automatic control and tuning of the hold-off.
520 Chapter 20
Figure 20.16 Schematic diagram of the proposed TIA with AGC, ABC, and ambient-light
rejection loops.
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 521
20.6 Conclusion
SiAPDs can be applied as efficient, low-power, compact, safe, inexpensive,
and sensitive tools for NIR light detection. Proficient implementation
techniques for these photodetectors will promote their incorporation into
different science and technology applications. By overcoming some of the
difficulties encountered in their fabrication using standard CMOS process,
miniaturized, reconfigurable, low-noise light detectors can be designed
especially for high sensitivity in the NIR region of the light spectrum.
Developing high-gain, low-noise, low-power circuitry for LM-operation
SiAPDs, and ultrafast, low-power, low-noise quench-reset and control
circuitry for GM operation are the main challenges in SiAPD light detector
design and fabrication. TIA front ends implemented for LM-SiAPDs offer
high GBW, low power consumption, and ultralow input current noises. These
characteristics make them a qualified candidate for LM-SiAPD-based light-
detection applications such as CW fNIRS for biomedical monitoring of
ultralow intensity biosignals. A high-speed photon-counting system that
includes an ultrafast quench-reset circuit, pulse generator, monostable, and
counter with small quenching time and controllable reset time is also
implemented for GM-SiAPDs. Its high rate detection, low power, and
compact characteristics make it a good candidate to be applied in SiAPDs for
GM-SiAPD and single-photon counting applications. Its integration can
significantly improve the gain, noise, and power characteristics of the
detectors. This issue has been proved with on-chip integration of the SiAPDs
522 Chapter 20
Acknowledgment
We gratefully acknowledge the financial support from the Canadian Institutes
of Health Research (CIHR), the Heart and Stroke Foundation of Canada,
and the Canada Research Chair in smart medical devices and CMC
Microsystems.
References
1. M. Bass, V. N. Mahajan, and E. Van Sryland, Eds., Handbook of Optics,
Third edition, Vol. II: Design, Fabrication, and Testing; Sources and
Detectors; Radiometry and Photometry, McGraw-Hill, New York (2010).
2. A. Rogalski, Fundamentals of Infrared Detector Technologies, CRC Press,
Boca Raton, FL (2010).
3. M. A. Itzler, M. Entwistle, M. Owens, K. Patel, X. Jiang, K. Slomkowski,
S. Rangwala, P. F. Zalud, T. Senko, and J. Tower, “Geiger-mode
avalanche photodiode focal plane arrays for three-dimensional imaging
LADAR,” Proc. SPIE 7808, 78080C (2010) [doi: 10.1117/12.861600].
4. W. L. Wolfe, Ed., Handbook of Military Infrared Technology, Office of
Naval Research, U.S. Govt. Printing Office (1965).
5. I. S. Glass, Handbook of Infrared Astronomy, Cambridge University Press,
Cambridge (1999).
Near-Infrared Light Detection using CMOS Silicon Avalanche Photodiodes (SiAPDs) 523
84. C.-K. Chien, H.-H. Hsieh, H.-S. Chen, and L.-H. Lu, “A transimpedance
amplifier with a tunable dandwidth in 0.18-mm CMOS,” IEEE Trans.
Microwave Theory Techniques 58, 498–505 (2010).
85. E. Hammoudi and A. Mokhtar, “2.75 GHz low noise 0.35 mm CMOS
transimpedance amplifier,” 2010 18th Mediterranean Conf. Control &
Automation (MED), 928–932 (2010).
86. S. H. Lee, J. Kim, Q. Le, M. Lee, H. Kim, and C. S. Park, “A single-chip
2.5-Gb/s burst-mode optical receiver with wide dynamic range,” IEEE
IEEE Photonics Technol. Lett. 23, 85–87 (2011).
87. H. M. Lavasani, W. Pan, B. Harrington, R. Abdolvand, and F. Ayazi, “A
76-dBV 1.7-GHz 0.18-mm CMOS tunable transimpedance amplifier using
broadband current pre-amplifier for high frequency lateral micromecha-
nical oscillators,” 2010 IEEE Int. Solid-State Circ. Conf. (ISSCC) Digest
of Tech. Papers, 318–319 (2010).
88. M. Ghioni, S. Cova, F. Zappa, and C. Samori, “Compact active
quenching circuit for fast photon counting with avalanche photodiodes,”
Rev. Scientific Instrum. 67, 3440–3448 (1996)
89. F. Zappa, M. Ghioni, S. Cova, C. Samori, and A. C. Giudice, “An
integrated active-quenching circuit for single-photon avalanche diodes,”
IEEE Trans. Instrument. Measur. 49, 1167–1175 (2000).
90. F. Guerrieri, S. Bellisai, A. Tosi, G. Padovini, F. Zappa, and S. Tisa,
“SPAD arrays for parallel photon counting and timing,” in 2010 23rd
Annual Meeting IEEE Photon. Soc., 355–356 (2010).
91. E. Kamrani, S. H. A. Yun, F. Lesage, and M. Sawan, “Optimal-adaptive
control system for low-noise, low-power and fast photodetection in
functional near infra-red spectroscopy,” presented at MIOMD-XI: 11th
International Conference on Infrared Optoelectronics: Materials and
Devices, Chicago, IL (2012).
92. F. Zappa, A. Giudice, M. Ghioni, and S. Cova, “Fully integrated active-
quenching circuit for single-photon detection,” Proc. 28th Eur. Solid-State
Circ. (ESSCIRC) Conf., 355–358 (2002).
93. D. Cronin, A. M. Moloney, and A. P. Morrison, “Simulated monolithi-
cally integrated single photon counter,” High Frequency Postgraduate
Student Colloquium 2004, 9–14 (2004).
94. G. F. Pulice, “Stabilization biasing circuit for avalanche photodiodes,”
U.S. Patent 5270533 (Dec. 14, 1993).
95. W. Becker, Ed. Advanced Photon Counting Techniques II, Proc. SPIE
6771(2007) [doi: 10.1117/12.774832].
530 Chapter 20
21.1 Introduction
21.2 Seebeck Effect Consideration
21.2.1 Seebeck coefficient diffusion component
21.2.2 Seebeck coefficient phonon drag component
21.3 Device Design Consideration
21.3.1 Performance of a thermoelectric sensor
21.3.2 AlGaAs/InGaAs thermopile design
21.3.2.1 H PILE structure
21.3.2.2 High sensitivity performance design (type A)
21.3.2.3 High speed performance design (type C)
21.3.3 Scaled down approach
21.4 Sensor Fabrication Technology
21.5 Measured Sensor Performance and Discussion
21.6 Conclusion and Future Prospects
Acknowledgments
References
533
534 Chapter 21
21.1 Introduction
Infrared image sensors are essential for automotive vehicle night vision, rescue
robot-eye vision, thermal imaging in biology and medicine, remote sensing in
security surveillance systems, and terahertz electronics applications. Two
different kinds of approaches are used to detect infrared radiation,
particularly for far-infrared radiation. These include very narrow-band
photon detectors and broad-band thermal detectors. Photon detectors are
used because of their high detectivity and high-speed performance. However,
they need cooling with liquid nitrogen and/or liquid helium and thus make the
system expensive and complex to handle. Based on the AlGaAs/GaAs
material system, quantum-well infrared photodetectors (QWIPs) at a
wavelength of 8.3 mm operating at 77 K,1 and multicolor QWIPs at
wavelengths of 8–9 mm and 10–15 mm operating at 40–70 K have been
realized.2 Conventional HgCdTe-based photodiodes can be used for the far-
infrared range up to 50 mm.3 On the other hand, uncooled broad-band
thermal detectors for detection of far-infrared radiation of around 10-mm
wavelength, either of pyroelectric, bolometric,4 or thermopile types,5 still
require high-speed operation as well as high-responsivity performances with
low cost. The thermoelectric approach offers a low-cost potential due to its
simple operation. This is because temperature stabilization takes place by the
Seebeck effect, and biasing with self-heating limitations such as those in the
bolometric type is not required. In addition, this approach has good
compatibility with integrated technology, unlike the pyroelectric discrete
approach. The polysilicon thermopile, which is compatible with complemen-
tary metal-oxide semiconductor (CMOS) technology, has served in low-cost
approaches up until now and led to the demonstration of an infrared image
sensor.5 However, it showed inferior performance compared to commercial
bolometric sensors.4 Heterostructure-based InGaAs/InP thermopiles6 and
modulation-doped heterostructure thermopiles (H-PILEs) were also
reported.7 11 However, as yet, their performance advantages have not been
demonstrated in detail.
In this chapter, the performance advantages of the modulation-doped
AlGaAs/InGaAs H-PILEs are presented based on demonstrated data. These
devices can achieve high performance due to the advantages of both superior
Seebeck coefficient and the exceedingly high mobility of 2D electron gas
(2DEG) and 2D hole gas (2DHG) at the heterojunction interface. This
chapter first describes the features of the Seebeck effect consideration and
device design considerations in Sections 21.2 and 21.3, respectively. The
fabrication technology for realizing infrared sensors and the measured
performances are presented in Sections 21.4 and 21.5, respectively. Finally,
the future prospects for uncooled infrared FPA image sensor applications is
discussed.
Modulation-Doped AlGaAs/InGaAs Thermopiles (H-PILEs)… 535
Figure 21.2 Schematic cross-sectional structure and energy band diagram of the AlGaAs/
InGaAs H-PILE (reprinted from Ref. 18).
538 Chapter 21
difference between the highly thermally isolated absorbing area and the cold
pad of heat sink. MEMS technology allows the fabrication of highly
thermally isolated absorbing areas.
Figure 21.3 Layout pattern of type-A H-PILE relating to pixel area for infrared detection.
Modulation-Doped AlGaAs/InGaAs Thermopiles (H-PILEs)… 539
The total electric resistance of the series pile couples (N ¼ 2) is Rel ¼ 240 kV,
where q is the elementary charge, m is the mobility of 8000 for electron and
320 cm2/Vs for hole, the sheet carrier concentration Ns ¼ 1.2 1012 cm 2 of
10-nm-thick 2DEG and/or 2DHG, and the electrical resistivities of the
InGaAs layer re ¼ 6.5 10 4 Vcm and rh ¼ 2.0 10 3 Vcm are calculated
for electron and hole, respectively. The calculated value D is 5.6 109
cmHz1/2/W, which is one order of magnitude higher than that of the
bolometer.4 Using Eq. (21.5), the response time t is calculated to be 8 ms, with
the heat capacitance of the sensor Cheat ¼ 4.2 10 9 J/K, and the total
thermal resistance of parallel pile couples (N ¼ 2) of Rth,total ¼ 1.8 106 K/W.
21.3.2.3 High-speed performance design (type-C)
Figure 21.4 shows the layout pattern and the size related to the pixel area for
infrared detection for the high-speed design approach (type-C).11,15 18 The
responsivity R is calculated using Eqs. (21.3), with the absorption coefficient
a ¼ 1, and the number of pile couples N ¼ 8. The thermal resistance Rthp,n of
each pile (p1, p2, n1, and n2) is calculated as Rthp1 ¼ 1.4 105, Rthp2 ¼ 2.1 105,
Rthn1 ¼ 5.4 105, and Rthn2 ¼ 3.6 105 K/W, respectively, where the thermal
resistivity W ¼ 9.2 for AlGaAs and 2.3 Kcm/W for GaAs.15 The device
parameters of the cross-sectional structure and the assumptions made for the
design calculations are the same as were presented in Section 21.3.2.2.
The calculated responsivity R is 4.900 V/W, almost comparable with polysilicon.5
The calculated detectivity D is 2.0 109 cmHz1/2/W, which is one order of
magnitude higher than that of a bolometer,4 where the total electric resistance of
series pile couples (N ¼ 8) is Rel ¼ 35 kV. The response time t is calculated to
be 110 ms [using Eq. (21.5)], with the heat capacitance of the sensor Cheat ¼
7.0 10 9 J/K, and the total thermal resistance of the parallel pile couples (N ¼ 8)
of Rth,total ¼ 1.5 104 K/W.
Figure 21.4 Layout pattern of type-C H-PILE relating to pixel area for infrared detection
(reprinted from Ref. 15).
540 Chapter 21
Figure 21.5 Performance territory for responsivity R versus response time t, compared
with alternative uncooled infrared sensors.
Figure 21.7 Pixel number and pixel pitch depending on design rule.
Figure 21.8 Specified epitaxial structure analyzed by SIMS measurement (reprinted from
Ref. 18). (See color plate section.)
Figure 21.9 Microphotographs of 32 32 matrix arrays: (a) type-A pixel pattern and
(b) type-C pixel pattern [part (b) reprinted from Ref. 16].
Figure 21.11 Waveform of thermovoltage measured under 1-Hz chopped radiation signal
at Tb ¼ 1200 K: (a) type-A (23-dB preamplifier) and (b) type-C (40-dB preamplifier).
Figure 21.12 Thermovoltage and lTb-integral rate versus blackbody radiation temperature
(reprinted from Ref. 15). (See color plate section.)
Figure 21.13 Spectrum of the bandpass filter used in the experiment (14 > l > 8 mm).
radiation intensity based on Planck’s equation for l > 3 mm, l > 5 mm (Ge-
window), and l > 8 mm, respectively. It is noted that the calibrated data (4) for
l > 8 mm are close to the measured data (s) with filter (14 > l > 8 mm). The
Tb dependence of Vout is also calculated from Eqs. (21.6), where R is assumed to
be from 1000 to 5000 V/W, as a parameter. The calibrated data for l > 5 mm
are close to R of 3000 V/W. The measured value for Rel is 2.8 MV, larger than
the designed value of 35 kV. This would be due to process-induced large
contact resistance of p and n electrodes. The detectivity D might be roughly
estimated to be 1.4 108 cmHz1/2/W, although D needs to be analyzed based
on the frequency dependence of the noise spectrum measurement.
Figure 21.14 Performance territory for responsivity R versus response time t for different
uncooled infrared sensors (reprinted from Ref. 15).
Figure 21.15 Comparison of detectivity spectra for different infrared detectors, including
p-i-n photodiodes. The 300-K blackbody-radiation-limited D*, 300-K BLIP (background-limited
infrared photoconductance) limit is also included as a reference (reprinted from Ref. 8).
Acknowledgments
The authors would like to thank Prof. Masafumi Kimata, Ritsumeikan
University for his kindly offering an infrared sensor measurement system
and useful suggestions, Prof. Dimitris Pavlidis and Mr. Chong Jin,
Technische University of Darmstadt, and Mr. Joo-Shik Park, Mr. Saint
Woo, and Mr. Satoru Okubo, Eye View Tech Co., Ltd. for their continuous
encouragement and support.
References
1. B. F. Levine, C. G. Bethea, G. Hasnain, J. Walker, and R. J. Malk,
“High-detectivity D ¼ 1.0 1010 cm Hz1/2/W GaAs/AlGaAs multi-
quantum well l ¼ 8.3 mm infrared detector,” Appl. Phys. Lett. 53,
296–298 (1988).
2. S. D. Gunapala, S. V. Bandara, J. K. Liu, S. B. Rafol, and J. M. Mumolo,
“640 512 pixel long-wavelength infrared narrowband, multiband, and
broadband QWIP focal plane arrays,” IEEE Trans. Electron Devices 50,
2353–2360 (2003).
3. C. Lucas, “Infrared detection, some recent developments and future
trends,” Sens. Actuators A 25–27, 147–153 (1991).
4. InfraVision LSI, Inc., “Uncooled Bolometer FPA SpecSheet,” Quebec
City, Quebec, Canada (1998).
5. M. Hirota, Y. Nakajima, Y. Hirose, M. Saito, F. Satou, and M.
Uchiyama, “120 90 element thermoelectric infrared focal plane array
with precisely patterned Au-black absorber,” Proc. of 19th Sensor
Symposium, 117–121 (2002).
6. A. Dehe, D. Pavlidis, K. Hong, and H. L. Hartnagel, “InGaAs/InP
thermoelectric infrared sensor utilizing surface bulk micromachining
technology,” IEEE Trans. Electron Devices 44, 1052–1059 (1997).
7. M. Abe, “AlGaAs/InGaAs thermopile-based infrared image sensor
utilizing HEMT and MEMS process technology,” 8th Topical Workshop
on Heterostructure Microelectronics (TWHM), WeC-3, 42–43 (2009).
8. M. Abe, “Modulation doped heterostructure thermopiles for uncooled
infrared image-sensor application,” IEICE Trans. Electron. E93-C,
1302–1308 (2010).
9. C. Jin, K. Boye, M. Abe, and D. Pavlidis, “GaAs-based infrared detector
for array application,” 34th Workshop on Compound Semiconductor
Devices and Integrated Circuits (WOCSDICE) W6, 117–118 (2010).
10. M. Abe, “Modulation-doped AlGaN/GaN thermopiles for uncooled
infrared image sensor,” 34th Workshop on Compound Semiconductor
Devices and Integrated Circuits (WOCSDICE) W6, 115–116 (2010).
548 Chapter 21
T. L. Hoai Nguyen
Institute of Physics, Vietnam Academy of Science and Technology,
Hanoi, Vietnam
Jean-Eric Wegrowe
École Polytechnique, LSI, CNRS and CEA/DSM/IRAMIS, Palaiseau, France
Guy Fishman
Université Paris-Sud, IEF, CNRS, Orsay, France
22.1 Introduction
22.2 General Definition of Current Operators
22.2.1 Current associated with a quantum mechanical operator
22.2.2 Symmetry properties of current operators
22.3 Probability Current Related to an Effective Hamiltonian
22.3.1 The general nth order Hamiltonian
22.3.2 Velocity operator in the presence of spin orbit interaction
22.3.3 Velocity and probability current operators in effective Hamiltonian formalism
22.4 Spin Current Operator
22.5 BenDaniel Duke like Formulation and Boundary Conditions
551
552 Chapter 22
22.1 Introduction
Spin transport and dynamics is one of the crucial issues of spintronics,
regardless of whether the system of interest is magnetic or nonmagnetic:1,2
semiconductors are of special interest because in such systems the charge
carrier dynamics can be accurately treated through single-particle approxi-
mation, making use of the precise knowledge of wavefunctions. Here the
single-electron, possibly spin-dependent wavefunction is well suited to provide
reliable fully quantum mechanical models, or to derive semiclassical
theoretical models, which eventually can be extended to other kinds of
systems. In this context transport operators play a key role and, in particular,
the probability current is a fundamental concept in quantum mechanics,
which connects the wave-like description of a quasi-particle to the notion of
transport current. When we consider a general Schrödinger problem where the
Hamiltonian is
b0 ¼ bp þ UðrÞ,
2
H ð22:1Þ
2m
where UðrÞ is the real periodic potential in a crystalline solid, and m is the free-
electron mass, we are led to the basic definition of the free-electron current
probability:3
b
p ħ
J ½c ¼ Re c
f
c ¼ Im½c rc: ð22:2Þ
m m
b¼H
H b0 þ H
bSO , ð22:3Þ
Spin–Orbit Engineering of Semiconductor Heterostructures 553
with
bSO ¼ ħ
H ðrU bp Þ bs : ð22:4Þ
4m2 c2
Following the arguments developed by Nguyen et al.,4 it is reasonable to
express the full Hamiltonian, involving SOI terms, as an effective
Hamiltonian that consists of momentum operator bp power series expansion:
indeed, besides the kinetic energy, which is quadratic in bp , the SOI provides
leading terms that are linear and cubic in bp , known respectively as Bychkov–
Rashba5,6 and Dresselhaus7 or D’yakonov–Perel’ (DP)8 terms. Then, since the
SOI potential is nonreal, a more general definition of the probability current
J½c should be taken into account.9 Considering interactions that include
higher-order polynomial terms in the Hamiltonian, we have to deal with an
effective Hamiltonian of order n.
Furthermore, an open question, strictly linked to the one above, concerns
the definition of spin current (SC). Indeed, in semiconductor physics that
provides paradigmatic systems for spintronics, it is known that the SC
standard definition, used by many authors,10 15 can be suitably applied to 2D
systems with Rashba SOI but fails to describe spin-dependent transport
phenomena in bulk cubic semiconductors, where SOI induces a DP term in
the conduction band. The existence of extra-current terms was pointed out in
Ref. 16, and Drouhin et al.17 have shown that a redefinition of SC is
mandatory to obtain a unified treatment, enlightening the fact that a properly
symmetrized spin-current operator bJ"ð#Þ ½c, where " ð#Þ refers to up- (down-)
spin channel, gives unexpected results when applied to tunneling through
evanescent states in GaAs barriers.
As pointed out by Rashba in Ref. 10, there are still concerns about the
fact that a consistent theory of spin transport current has not been
formulated yet. From a general point of view, this means that we cannot
immediately approach such a topic in terms of nonequilibrium thermody-
namics. In fact, a difficulty is related to the definition of system in order
to formulate relevant balance equations as well as the boundary terms
that should be possibly included in the effective Hamiltonian. Shi et al.18
have proposed an alternative spin-current operator, satisfying the continu-
ity equation, that allegedly supports important conclusions concerning
conservation of spin currents,19 21 but that appears to rely on nonexplicit
assumptions (see Section 22.2).
The inclusion of SOI in the Hamiltonian of a system has direct and
practical consequences in heterostructures, where a consistent analysis of
the tunneling phenomena is required. The pragmatic BenDaniel–Duke
(BDD) approach,22 which works when dealing with quadratic Hamilto-
nians under effective-mass approximation, cannot be straightforwardly
extended because it is not always possible to ensure both the continuity of
554 Chapter 22
As pointed out by Shi et al. in Ref. 18, the continuity of J can be ensured
by introducing a general source term G, as in Eq. (22.5), but the source term
is not uniquely defined, and this leads to possible confusion when
considering the conservation laws in terms of nonequilibrium thermody-
namic equations. In any case, we first need to clearly state the local
properties of a current operator, postponing the analysis of its global
properties. For this purpose, we consider a linear operator Ab that does not
explicitly depend on time and acts over a generic state c. In the following we
adopt the notation ð Ab Þc ¼ ðcj Ab cÞ ¼ c† Ab c used in Ref. 4. The general
Schrödinger equation reads
@ b c,
c¼H
iħ ð22:6Þ
@t
where H b denotes the Hamiltonian in general form. For example, H b may equal
b b
H [defined in Eq. (22.3)] or H eff [defined below in Eq. (22.36)]. The derivative
of ð Ab Þc with respect to time is
@ b @ † b @ † b @
ð A Þc ¼ c Ac ¼ c A c þ c† Ab c : ð22:7Þ
@t @t @t @t
Using Eq. (22.6) we obtain
@ b 1 b † b 1 † b b 1 † b b
bc
†
b :
ð A Þc ¼ Hc Acþ c A Hc ¼ c AHc H Ac
@t iħ iħ iħ
ð22:8Þ
If Ab is an Hermitian matrix (its elements are complex numbers but not
differential operators), i.e.,
†
Hb c Ab c ¼ c† Ab H
bc , ð22:9Þ
then we can rewrite Eq. (22.8) in a more suitable way that we refer to as the
local form of the Ehrenfest theorem:
@ b 2
ð A Þc ¼ Im c† Ab H bc : ð22:10Þ
@t ħ
A thermodynamic model only based on the continuity equation Eq. (22.5) cannot take into
account the exchange of energy e with the environment introduced by the current generator and
the SOI. Indeed, the consequence of the Gibbs equation, de=dt ¼ T@s=@t þ m@r=@t (where s is
the entropy, T the temperature, and m is the chemical potential), is that the continuity equation
of the density of carriers r is necessarily determined by the power dissipated at the boundaries.
At zero temperature, the source term is equal to the power dissipated divided by the chemical
potential @r=@t ¼ ð1=mÞ de=dt. This is the reason that the decomposition between the
divergence term and the source term G cannot be unique without specifying the boundary
conditions.
556 Chapter 22
The integration over the whole space leads to the well-known Ehrenfest
theorem, whose global form is valid for any (possibly differential) Hermitian
operator A: b
@ D
E 1 hD
b b
E D b
Ei 1 D
h b b i
E
cA c ¼ c AH c Hc A c ¼ c
A, H
c : ð22:11Þ
@t iħ iħ
We can write
@ b 1 n b bo 1 †h b bi
A ¼ Im c† A, H c þ Im c A, H c , ð22:12Þ
@t c ħ ħ
with fba ,bbg ¼ ba bb þ bbba , and, by integration over the whole space, we get
Z n o
bH
d 3 r Im c† A, b c ¼ 0: ð22:13Þ
when the integration is performed over the volume of the system ðVÞ. Then,
Z Z Z
d r G ¼ d r r JG ¼ JG ds ¼ 0,
3 3
ð22:18Þ
V V S
where the volume integral is changed into a surface integral through
Ostrogradski’s theorem (here S is the surface that limits V, and ds is the
surface element oriented along the normal to S). Such a relation is obviously
satisfied provided that JG d s ¼ 0, i.e., provided that JG is a tangential
vector to S, a condition that is “ physically” reasonable. Shi et al.18 further
assume that JG “is a material property that should vanish outside the
sample;” this is a more restrictive and questionable hypothesis. For instance,
in the case of a magnetic field, the effect of the associated vector potential
cannot a priori be overlooked outside the sample. In any case, let us assume
that JG ¼ 0 at the surface S. Following the calculation by Shi et al.,18 it is
straightforward to show, after partial integration where the boundary
contribution cancels, that
Z Z
@JG, x @JG, y @JG, z
dy dz dx x þ þ ¼ d 3 r JG, x , ð22:19Þ
@x @y @z
in which JG, x , JG, y , and JG, z are the Cartesian components of JG . Then,
Z Z Z
d 3 r JG ¼ d 3 r rr JG ¼ d 3 r rG
Z h i Z h i
1 † b b 1 bHb c : ð22:20Þ
¼ d r r Im c A, H c ¼
3
d 3 r Im c† r A,
ħ h iħ
It is easy to check that, provided that A, b r ¼ 0,
h i h i
r A, bH b ¼ Ar,
b H b b
iħ by A, ð22:21Þ
h i
where r, H b ¼ iħ by . Thus,
Z Z h i Z
1 b H b c þ d 3 r Re c† by Ac b
d 3 r JG ¼ d 3 r Im c† Ar,
ħ
1
Z h i Z
¼ d r Im c Ar, H c þ d 3 r J~A :
3 † b b
ð22:22Þ
ħ
Here, J~A is the canonical current defined as
!
b
y Ab þ b
Aby
J~A ¼ Re c† c : ð22:23Þ
2
According to Eq. (22.8), we can write
2 3
Z Z b
d Ar
6 c7
d 3 r JG ¼ d 3 r 4J~A 5: ð22:24Þ
dt
558 Chapter 22
ð22:29Þ
where Kb0 is the time-reversal Kramers operator for a spinless particle, which
consists of taking the complex conjugate in the r -representation. Let us check
Spin–Orbit Engineering of Semiconductor Heterostructures 559
Here, Kb ¼ Rb Kb0 is the Kramers operator for a particle with spin 1=2,
Rb ¼ is y ð Rb † ¼ Rb 1 Þ, and «A ¼ 1, depending on whether Ab verifies that
(this relates to a real or purely imaginary operator, which corresponds to the
cases of interest here)28
Kb Ab Kb ¼ «A A,
b i:e:; Rb Ab Rb ¼ «A Ab : ð22:32Þ
b Ab c,
Similarly, for the term c† H
b AbKbc ¼ «A Kb0 c
Rb† Kb H
Kbc
H b Ac
b
¼ «A Kb0 c
Kb0 H
b Ac
b ¼ «A c
H
b Ac
b : ð22:33Þ
Thus, we obtain
n o n o † n o
bH
2i Im c † A, b c ¼ c† A,
bHb c þ «A Kbc bH
A, b Kbc : ð22:34Þ
H b p þ V ðrÞ,
b eff ¼ H ð22:36Þ
560 Chapter 22
where bp lðkÞ is the momentum operator associated with the lðkÞ Cartesian
coordinate and where clð1Þ ; . . . ;lðnÞ are Hermitian matrices that are invariant
under permutation of the subscripts. The abstract form of Eq. (22.37) allows
us to perform easy calculations. In Section 22.3.2 we show how to handle such
a general expression to deal with concrete situations.
Formally, we perform the identification
cx . . . cx cy . . . cy cz . . . cz ¼ cx . . .x , y . . . y , z|fflffl.{zffl
. .fflz} ,
|fflfflffl
ffl{zfflfflfflffl} |fflfflfflffl{zfflfflfflffl} |fflfflfflffl{zfflfflffl} |fflffl{zfflffl} |fflffl{zfflffl} ð22:38Þ
a b g a b g
with
n!
cabg ¼ ca cb cg : ð22:41Þ
a!b!g! x y z
We are now in a position to tackle the problem of velocity, first, when the
Hamiltonian H b takes into account the SOI, and, second, when the Hamiltonian
b
Heff is an effective Hamiltonian.
If ði=ħÞh½ H,b b
b r ic ¼ h@ H=@pi c , then b
b bp because two linear
v ¼ @ H=@
operators that have the same mean values over all possible states are equal:
8c, h Ab ic ¼ h Bbic () Ab ¼ B. b 29 Then, it is enough to show that ði=ħÞ½ H,b
b r ¼
b b
@ H=@ bp to prove that bv ¼ @ H=@ bp .
Spin–Orbit Engineering of Semiconductor Heterostructures 561
or
ihb i ihb i @Hbeff
Hp , br ¼ Heff , br ¼ : ð22:46Þ
ħ ħ @bp
Using Eqs. (22.38) through (22.41), it is then easy to calculate the
ðnÞ b ðnÞ :
j component by j ðj ¼ x, y, zÞ of the velocity operator bv ðnÞ associated with H
ðnÞ
ðnÞ @Hb
by j ¼ ¼ ncj ðcx bp x þ cy bp y þ cz bp z Þn 1 : ð22:47Þ
@bp j
We introduce the scalar product between the momentum bp and the
velocity operator bv ðnÞ :
ðnÞ
bp x by ðnÞ
x þbp y by ðnÞ
y þbp z by ðnÞ
z ¼ nðcx b
b
p x þ cy bp y þ cz bp z Þn ¼ n H : ð22:48Þ
Equation (22.48) means that
b ðnÞ ¼ n H
bp bv ðnÞ ¼ bp rp H b ðnÞ , ð22:49Þ
which is nothing but Euler’s relation for a homogenous function. Eventually,
!
X 1 ðnÞ
Hbeff c ¼ bp bv c þ V c ¼ Ec: ð22:50Þ
n
n
As pointed out in Section 22.2, we are allowed to define current operators
in open systems provided that we properly take into account their boundary
conditions. In Appendix 22.A, we show that performing the proper
562 Chapter 22
ðnÞ X h
Jbj ðr0 Þ ¼ cj, lð1Þ, . . . , lðn 1Þ dr0 bp lð1Þ . . . bp lðn 1Þ þ bp lð1Þ dr0 . . . bp lðn 1Þ
lðkÞ2fx, y, zg
k 1, ... , n1
2 X X X
¼ Im cj , lð1Þ, . . . , lðn 1Þ ðcjb
p j bp lð1Þ . . . bp lðn 1Þ jcÞ:
ħ n j fx, y, zg lðkÞ2fx, y, zg
k 1, ... , n 1
ð22:53Þ
Then, we recover all of the terms of Eq. (22.14) so that Eq. (22.52)
indeed provides a correct definition of the current operator. Obviously,
adding a term proportional to the curl of any vector field would not affect
the result. Such a definition of bJ provides an unambiguous and general tool
for evaluating the probability current. Provided the Hamiltonian of the
whole system is known, this probability-current operator guarantees the
requirements of the continuity equation.
Now it is useful to introduce the Hermitian symmetrized velocity operator
bv ðr0 Þ so that the velocity vðr0 Þ at the point r0 is vðr0 Þ ¼ hcjbv ðnÞ ðr0 Þjci. We
ðnÞ
have
ðnÞ nX
bv j ðr0 Þ ¼ cj, lð1Þ, . . . , lðn 1Þ ½dr0 bp lð1Þ . . . bp lðn 1Þ þ bp lð1Þ . . . bp lðn 1Þ dr0 :
2
lðkÞ2fx, y, zg
k 1, ... , n1
ð22:54Þ
For example, for n 3, the comparison between Eqs. (22.52) and (22.54)
ðnÞ
clearly shows that bJ j ðr0 Þ contains n 2 extra terms, which are straight-
forwardly obtained from @ H beff =@bp . For instance, with H beff bp n , we have
@Hbeff =@bp nbp n 1
such that bv ðnÞ ðr0 Þ ðn=2Þðdr0 bp n 1
þ bp n 1 dr0 Þ, whereas
JbðnÞ ðr0 Þ ðdr0 bp n 1
þ bp dr0 bp n 2 n 1
þ . . . þ bp dr0 Þ. As shown in Ref. 4, extra
Spin–Orbit Engineering of Semiconductor Heterostructures 563
P P P
where H ð1Þ ¼ j aj bp j , H ð2Þ ¼ j, k bjk bp j bp k , and H ð3Þ ¼ j, k, l cjkl bp j bp k bp l . The
bp j , bp k , and bp l operators are the bp components; aj , bjk , and cjkl ( j, k, and l
refer to Cartesian coordinates) are ð2 2Þ Hermitian matrices operating in
the spin space, invariant under permutation of j, k, and l. Indeed, the
third-order expansion of the Hamiltonian covers almost all practical
situations in semiconductors, although higher-order terms could be
incorporated, if needed, without altering the physics discussed here. The
linear (first) terms describe the Rashba Hamiltonian, the quadratic
(second) terms correspond to the usual kinetic-energy contribution, and
the cubic (third) terms describe the DP Hamiltonian. Let us point out that
the Rashba and DP Hamiltonians are vastly different. Only with some
approximations or along special crystallographic directions does the DP
Hamiltonian reduce to a form that is a unitary equivalent to the Rashba
one.6 This arises in the situation considered in Ref. 34. Equation (22.57)
was introduced in Ref. 4 in order to analyze 1D tunneling through
evanescent states whose spin properties are determined by the DP field
(see also Section 22.5).
Hereafter, we detail a procedure to derive the SC expression relevant
to the Hamiltonian given in Eq. (22.57). We will show that the standard
expression of the SC tensor is correct up to the quadratic terms, which
includes Rashba/Klein Hamiltonians,10,35 but that, concerning the DP
Hamiltonian, a generalized expression has to be used instead.
First, let us consider the total probability current, associated with particle
conservation, J½c originating from the wavefunction c. It has been shown in
Ref. 4 that the j component of J½c is
X
Jj ½c ¼ ðcjaj cÞ þ ðcjbjk bp k cÞ þ cc
k
X
þ 3ðbp k cjcjkl bp l cÞ þbp k bp l ðcjcjkl cÞ , ð22:58Þ
k, l
where cc means complex conjugate. The structure of this current can be made
b 0 Þ jci, where the
more intuitive by writing, as in Eq. (22.52), J½cðr0 Þ ¼ hcj Jðr
j component of the ð2 2Þ Hermitian current operator Jðr b 0 Þ—acting on a
spinor—is defined as
Jbj ðr0 Þ ¼ dr0 aj þ ðdr0 bp þbp dr0 Þ b j
þ ðdr0bp bp t þ bp dr0 bp t þ bp bp t dr0 Þ : c j : ð22:59Þ
In this expression, the notation “ : ” refers to the generalized double-dot
product defined by M1 : M2 ¼ Trr ðM1 M2 Þ, where M1 and M2 are arbitrary
matrices, and P Trr is the partial trace calculated over only the space states, i.e.,
bp bp t : c j ¼ kl pk pl cjlk . In the space states, the ð2 2Þ Hermitian spin matrix aj
has to be viewed as a scalar operator, b j as a vectorial operator of components
Spin–Orbit Engineering of Semiconductor Heterostructures 565
construction leads to the widely used expressions of the probability and spin
currents up to the first order. The same goes for the quadratic and cubic terms.
Eventually, we obtain
X
2Jus , j ½c ¼ ðcjfps , aj gcÞ þ ½ðcjfps , bjk g bp k cÞ þ cc
k
X
þ 3ð bp k cjfps , cjkl g bp l cÞ þ bp k bp l ðcjfps , cjkl gcÞ , ð22:61Þ
k, l
1 h i
GuðnÞ
,s ½c ¼ Im c †
p s , H ðnÞ
c : ð22:62Þ
ħ
ðnÞ
Gu, s ½c, with n ¼ 1 or 3, refers to the source contribution originating from
the linear and cubic terms in the Hamiltonian. To comply with Kramers
symmetry, the even-order terms cannot induce any spin splitting: bjk are
ð2Þ ðnÞ ðnÞ
diagonal matrices, and Gu, s ½c ¼ 0. Now, we have Gu, " ½c þ Gu, # ½c ¼ 0, and
Ju ½c ¼ J"u ½c þ J#u ½c. The SC dJu, j ½c is the difference between the up- and
down-spin currents, dJu, j ½c ¼ Ju", j ½c Ju#, j ½c, and the corresponding source
P ðnÞ ðnÞ
terms are dGu ½c ¼ n ðGu, " ½c Gu, # ½cÞ. We obtain
X
2dJu, j ½c ¼ ðcjfs u , aj gcÞ þ ½ðcjfs u , bjk gbp k cÞ þ cc
k
X
þ 3ðbp k cjfs u , cjkl gbp l cÞ þ bp k bp l ðcjfs u , cjkl gcÞ, ð22:63Þ
k, l
and
1 h i
dGu ½c ¼ Im c† s u , H b c , ð22:64Þ
ħ
where we have used the relation s u ¼ p" p# . This constitutes a natural
extension of the standard definition. Starting from the expression of the j
component of the total probability current J½c [Eq. (22.58)], the SC is
0
straightforwardly obtained by the substitutions aj ! aj ¼ ð1=2Þ fsu , aj g,
0 0
bjk ! bjk ¼ ð1=2Þ fsu , bjk g, and cjkl ! cjkl ¼ ð1=2Þ fsu , cjkl g. These operators
are still Hermitian matrices, invariant under permutation of the subscripts.
Thus, the same calculation allows one to write the SC operator in a form
From the Ehrenfest theorem, the average value of the source term is ð1=iħÞhcj½su , Hb jci ¼
dhsu i=dt. In a previous paper [T. L. Hoai Nguyen, H. J. Drouhin, J. E. Wegrowe, and G.
Fishman, “Spin orbit enginering of semiconductor heterostructures: A spin sensitive quantum
phase shifter,” Appl. Phys. Lett. 95, 082108 (2009)], we have shown that a stack of GaAs tunnel
barriers acts as a spin rotator, so that angular momentum should be transferred from the free
electron beam to the crystal upon tunneling [also see: P. M. Haney and M. D. Stiles, “Current
induced torques in the presence of spin orbit coupling,” Phys. Rev. Lett. 105, 126602 (2010)].
Spin–Orbit Engineering of Semiconductor Heterostructures 567
similar to Eq. (22.59). In the case of Rashba splitting, it can easily be checked
that these SCs and source-term definitions reduce to the standard formulas.12
The situation is drastically different in the presence of a DP field.
We can extend this result for the more general case of the nth -order
Hamiltonian: the j th Cartesian component of the spin-current operator d dJu is
th
obtained from the j component of the probability-current operator after the
substitution
1
cj , lð1Þ;...;lðnÞ ! bs u , cj, lð1Þ;...;lðnÞ g, ð22:65Þ
2
and the source terms are
1 h i
G ¼ Im c† bs u , H b c : ð22:66Þ
ħ
This result is a consequence of Eq. (22.14) which shows that, for a general
operator Ab that verifies the relation ½ A, b bp ¼ 0, the current JbA can be deduced
from Eq. (22.52) after the substitution
1 b
cj, lð1Þ;...;lðnÞ ! f A, cj, lð1Þ;...;lðnÞ g: ð22:67Þ
2
by x ¼ a þ 2b bp , ð22:69Þ
must always be satisfied under stationary regime. Because this condition is not
linear versus c, we will use it to first order only. Obviously, we must relax the
continuity condition on the envelope function itself at the interface that we do
not use. Let us again emphasize that no information is obtained on the true
wavefunction near the origin. Equation (22.72) does not ensure either
the continuity of the envelope function or the existence of derivatives at
the interface.
To give some insight into Eq. (22.72), consider two regions ð1Þ and ð2Þ
made of materials with the same band parameters (only the potential V ðrÞ may
take a different constant value in each region). We consider a volume V, limited
by a surface S, that surrounds an interface portion. Similarly to the BDD
technique, we start from Eq. (22.50) and we integrate the Schrödinger equation
over V. Using Ostrogradski’s theorem, when V tends to zero, we obtain
Z !
X1
ðnÞ
lim ds v c ¼ 0, ð22:74Þ
V!0 S
n
n
where ds is normal to the surface S.
For a 1D case with the interface at the origin, Eq. (22.74) becomes
" #þ«
X1
ðnÞ
lim v c ¼ 0: ð22:75Þ
«!0
n
n
«
ðnÞ
When choosing a set of a parameters, we will take care that these
parameters are compatible with Eq. (22.75) when the two regions are made
from materials with the same band parameters.
We will check in Section 22.6.4 that this procedure is well suited to deal
with a tunneling problem involving a cubic DP term in the Hamiltonian.
Remarkably, the boundary conditions that we need to solve the problem
drastically change when moving to the case of a DP Hamiltonian with cubic
terms. The crucial point, which we illustrate in this model case, is that we
cannot make any hypothesis about the continuity of the wavefunction
because, if we need to ensure probability-current conservation at an interface,
we must accept an envelope function c that is no longer continuous. To give
insight into the expression of the current operator and into the conservation of
the probability current, let us again come back to an interface between two
semi-infinite 1D media ð1Þ and ð2Þ. In each bulk crystal, the relevant
Hamiltonian is
X ðnÞ
Hbr ¼ b r þVr ,
H ð22:76Þ
n
where
b ðnÞ
H ðnÞ n
r ¼ gr bp , ð22:77Þ
570 Chapter 22
where hjiðrÞ means summation over the bulk part of region ðrÞ. Because the
total probability in the domain 1, w1 [ ½w2 , þ1½ has to be conserved,
from the Ehrenfest theorem [Eq. (22.11)] with Ab ¼ Qð x w1 Þ þ Qðx w2 Þ]
we must have
hcj½Qð x w1 Þ þ Qðx w2 Þ, Hjci ¼ hcj HbjCi hCj Hbjci ¼ 0, ð22:81Þ
i.e.,
hCj Hbjci ¼ hcj HbjCi ¼ EhcjCi: ð22:82Þ
Observe that
b ðnÞ
H 1 ½Qð x
ðnÞ
w1 Þw1 ¼ g1 bp n Qð x w1 Þw1
ðnÞ
¼ Qð x w1 Þg1 bp n w1
ðnÞ
þ iħg1 ½dðx þ w1 Þbp n 1
þ bp dðx þ w1 Þbp n 2
þ . . . þ bp n 1 dðx þ w1 Þw1
ðnÞ ðnÞ
¼ Qð x w1 Þg1 bp n w1 þ iħ Jb1 ð w1 Þw1 , ð22:83Þ
Spin–Orbit Engineering of Semiconductor Heterostructures 571
so that
b 1 ½Qð x
H w1 Þw1 ¼ EQð x w1 Þw1 þ iħ Jb1 ð w1 Þw1 , ð22:84Þ
and, similarly,
b 2 ½Qðx
H w2 Þw2 ¼ EQðx w2 Þw2 iħ Jb2 ðw2 Þw2 : ð22:85Þ
Eventually,
HbC ¼ EC þ iħ½ Jb1 ð w1 Þw1 Jb2 ðw2 Þw2 : ð22:86Þ
Then, Eq. (22.82) is satisfied, provided that
Jb2 ½w2 ðw2 Þ ¼ Jb1 ½w1 ð w1 Þ: ð22:87Þ
The important point is not the conservation relation, which might appear
as physically obvious, but that, in Eqs. (22.83) and (22.85), the symmetrized
current operator is automatically generated in the form derived in Eq. (22.52),
providing physical insight into this mathematical expression. Because we only
deal with wavefunctions at some distance from the heterojunction, the
continuity of the true wavefunction at x ¼ 0 does not imply the continuity of
the envelope function C, which may be discontinuous. This is an illustration
of Harrison’s ideas on tunneling.27 Equations (22.75) and (22.87) generate a
set of boundary conditions relevant to the tunneling problem.
V( z)
I II III z
ξ ξ ξ
q Q iK q
kI = ξ + q kII = ξ + Q + iK kIII = ξ + q ez
Figure 22.1 Sketch of the tunnel geometry with definition of notations. The spin-orbit-split
barrier material of thickness a (medium II) is located between two free-electron-like materials
(media I and III). The tunnel axis, normal to the barrier, is the z axis. In the free-electron-like
materials, the real electron wave vector in the z direction is referred to as q. In the barrier
material, the evanescent wave vector along the z axis is referred to as Q þ iK, where Q and
K are real quantities. The transverse wave vector component j in the barrier plane is
conserved in the tunnel process. Then, the overall wave vectors in the three media are
kI ¼ kIII ¼ j þ q and kII ¼ j þ Q þ iK. (Reprinted from Ref. 4.)
bottom of the conduction band, and the top of the valence band lies at energy EG .
In this basis, the k p Hamiltonian is readily written as
0 Pk
b
H¼ : ð22:88Þ
Pk EG
In this expression, P ¼ ðħ=mÞhSjbp jV i (the phase of jV i has been chosen
so that P is real). The corresponding secular equation is
l 2 þ EG l ðPkÞ2 ¼ 0, ð22:89Þ
and the l eigenvalues are related to the energy through
ħ2 k 2
E ¼lþ : ð22:90Þ
2m
The evanescent states located in the bandgap correspond to k ¼ iK, where
K is real. Then the secular equation only admits a real solution if jKj
EG =2P
and
q
1
l¼ EG EG 4P K :
2 2 2 ð22:91Þ
2
For GaAs, the parameters are P ¼ 9:3 eV Å and EG ¼ 1.52 eV so that we
have jKj
0:08 Å 1 . Because the width of the Brillouin zone is about 1 Å 1 ,
the evanescent states are confined to a K extent on the order of 7% around the
Spin–Orbit Engineering of Semiconductor Heterostructures 573
2.5
2.0
1.5
Energy (eV)
1.0
0.5
0.0
0.5 iK k
zone center. The eigenvalues are plotted in Fig. 22.2. In a more realistic
situation, the energy dispersion curves result from the diagonalization of a
ðn nÞ k p Hamiltonian H b . Note that k is now a complex vector such that
the H b matrix is no longer Hermitian and the evanescent states are associated
only with real eigenvalues E, i.e., to the real-energy lines. In the Pidgeon–
Brown model,36 derived from the Kane model,37 the three upper valence
bands and the first conduction band are treated as a quasi-degenerate subset
where the ð8 8Þ k p Hamiltonian, including the SOI, is to be diagonalized.
Then, the valence-band structure includes the heavy-hole, the light-hole, and
the spin-orbit-split bands. In this framework, the bands remain spin degenerate,
the lack of inversion center not being taken into account. Both the lack of
inversion center and the SOI can be taken into account with n ¼ 14 or
n ¼ 30:38,39 The spin degeneracy is now lifted: the spin splitting, with the so-
called “k 3 dependence” given in Refs. 7,8,37, and 40–42, arises through third-
and fourth-order perturbation from remote bands.43 To find the energy
dispersion curves, we have to solve the secular equation det MðkÞ ¼
b E Ib ¼ 0, where Ib is the identity. Because the Hamiltonian is
det ½ H
Hermitian when k is a real vector, we have the relation MðkÞt ¼ Mðk Þ.
Thus, det Mðk Þ ¼ ½det MðkÞt ¼ ½det MðkÞ . It follows that En ðkÞ ¼ En0 ðk Þ,
where the band indices n and n0 may or may not refer to the same band.44,45
Moreover, Kramers conjugates correspond to the same energy so that the state
associated with ðk, jupiÞ and the state associated with ð k, jdowniÞ are
degenerate.28,46 Thus, in GaAs, the spin degeneracy is lifted, and we expect
574 Chapter 22
that the four states ½ðk, jsiÞ, ðk , js0 iÞ, ð k , j siÞ, andð k, j s0 iÞ will be
degenerate, jsi and js0 i being up-spin states in directions that, generally, are not
parallel. This results in a particular topology of the evanescent states.47 Hereafter,
we will see a concrete example where jsi and js0 i are quantized in the same
direction.
bDP ¼ gc k 2 1 gk3 ¼ gc bp 2 g bp 3 ,
H ð22:95Þ
2 ħ2 2ħ3
where the þ (resp. ) sign applies to the up (down) spin, quantized along
the DP field. In the barrier, a simple idea for finding the evanescent states
would be to take kII ¼ piK2 ½110, which leads to «ðkÞ ¼ gc K 2 i g2 K 3 .
This quantity is not real and cannot be an energy EðkÞ. Moreover,
Richard et al.,38 have shown that no wave vector of the form
! !
iK½1; 1; 0 þ , where is a real vector orthogonal to [110], provides
acceptable evanescent states; we are therefore led to consider a wave
vector such that kII ¼ p12 ðQ iKÞ½110. The calculation for the tunneling
states is given in Appendix B. The resulting band is plotted in Fig. 22.3,
over a very broad energy domain to reveal its general structure. We are
only interested in evanescent states located in the forbidden bandgap, i.e.,
states with a small negative energy. For our purpose, a key point is that, at
a given energy, we have exactly the four possible states, the wave vectors
Spin–Orbit Engineering of Semiconductor Heterostructures 575
Figure 22.3 Mathematical plot of the real-energy lines for k along [110] as a function of the
real part of the wave vector Q in the barrier. The calculation is performed for the ratio
g=gc ¼ 0:438 Å. We are only concerned with negative energies, which refer to evanescent
states. More precisely, the physical states are located within a very small energy domain
below the origin. The domain Q > 0 refers to up-spin states, whereas the domain Q < 0
refers to down-spin states. In each case, the imaginary component of the wave vector can
take the values K. Thus, at a given energy, we have exactly the four possible states
ðQ iKÞ " and ð Q iKÞ #. The down-spin states are Kramers conjugates of the up-spin
states. (Reprinted from Ref. 4.)
of which are ðQ iKÞ with spin " and ð Q iKÞ with spin #, the latter
being obtained from the former through Kb. In short,
E" ðkÞ ¼ E" ðk Þ ¼ E# ð kÞ ¼ E# ð k Þ: ð22:96Þ
Equation (22.96) provides a concrete example of the ideas developed by
Jones,45 who showed that EðkÞ ¼ Eðk Þ. The corresponding four plane waves
are eiðQiKÞ ", eið QiKÞ # or eKz eiQz ", eKz e iQz #.
If we consider the state ½ðiK þ QÞz, ", the total probability current
(charge current) equals the up-spin current, which is also the SC. This current
has to be conserved as neither charges nor particles can be created inside the
material. Let us check this basic conservation rule with the standard
expressions. The wavefunction is now
cþ ðzÞ ¼ eiðQþiKÞz " , ð22:97Þ
and the DP field
1
þ ¼ ðQ þ iKÞ3 : ð22:98Þ
2
576 Chapter 22
The probability current should be equal to the standard spin current [after
Eq. (22.55)], which yields
g
Ju, z ½c ¼ dJu, z ½c ¼ 4 c Q jc" j2 : ð22:99Þ
ħ
Then, we obtain a contradictory result: Ju, z ½c is nonzero and not even
divergence free. This would imply the existence of a nonzero source term that
is not physically acceptable and must be excluded. Even a constant current
would be nonphysical, as the same current would be associated with a
vanishingly small wavefunction for z ! þ1 and with a diverging wavefunc-
tion for z ! 1. In that case, the total probability current (and so the SC)
must be zero.
Indeed, the correct expression for the current yields a zero current for an
evanescent state along a real-energy line. For the DP Hamiltonian, the current
[Eq. (22.60)] becomes
2 !
g
@
@ 2
J ½c ¼ Jf ½c 3
c
jc j2 : ð22:100Þ
2ħ
@z
@z2
On a real-energy line [see Eq. (22.5)] concerning an evanescent wave
c ¼ eðKiQÞz , the continuity equation is satisfied by
@jcþ j2 2ħ g
¼ r Jþ ½cþ ¼ KQe 2Kz
þ Im ðQ þ iKÞ3 e 2Kz
@t m ħ
2K 1
¼ 2gc Q þ gð3Q2 K 2 Þ e 2Kz ¼ 0, ð22:101Þ
ħ 2
and it is also easy to check that J ½eðKiQÞz ¼ 0 on a real-energy line.
It has to be emphasized that an incorrect definition of the spin currents
implies an incorrect definition of the torques. In this example, if the free-
electron probability current were used instead of the correct expression,
spurious spin torques would be found. These spurious torques could be
“intuitively” related to electron spin precession around the complex DP
field.
@ 2 c" 1 @ 3 cð0Þ
gc þ ig ¼ ½E V ðzÞc" : ð22:108Þ
@z2 2 @z3
Using the properties of cð0Þ [Eqs. (22.106) and (22.107)], integrating this
equation from one side of the interface to the other and taking the
578 Chapter 22
approximation to the first order in q=K provides us the jump of the derivative
at the interfaces for the up-spin wavefunction:
B
@c" z0 1 g 1 g
¼ i ðK 2 þ q2 Þcð0Þ ðz0 Þ i K 2 cð0Þ ðz0 Þ ¼ 2iQ" cð0Þ ðz0 Þ:
@z zW 2 gc 2 gc
0
ð22:109Þ
For the down-spin wave, Q# ¼ Q" , and we have
B
@c# z0
¼ 2iQ# cð0Þ ðz0 Þ: ð22:110Þ
@z zW
0
dA2 ðq, K, QÞ dB2 ðq, K, QÞ
a2 ¼ ; b2 ¼ , ð22:114Þ
dQ Q 0 dQ Q 0
dA3 ðq, K, QÞ
a3 ¼ ’ iaa3 þ a03 : ð22:115Þ
dQ Q 0
Spin–Orbit Engineering of Semiconductor Heterostructures 579
continuity of the current at the interface [Eq. (22.122)], and we find the
equation
† þ 1 g þ
Re ðgc bp c0 Þ ½c ðgc bp c0 Þ ¼ 0: ð22:125Þ
4ħ g2c
Equation (22.125) is obviously satisfied when c verifies that
† þ 1 g þ
ðgc bp c0 Þ ½c ðgc bp c0 Þ ¼ 0, ð22:126Þ
4ħ g2c
and a fortiori when c verifies:
þ 1 g þ
½c ¼ ðgc bp c0 Þ: ð22:127Þ
4ħ g2c
When g=g2c is a constant, c is continuous, as observed in the case of constant
mass and constant g.
Observe that the boundary condition given by Eq. (22.127) is derived
similarly to the case of free electrons, when only assuming the continuity of
the velocity and the conservation of the probability current at the interface. In
that very case, the boundary conditions are the continuity of the velocity vð2Þ c
at the interface, i.e., gc bp c continuous, and the conservation of the free-
electron probability current at the interface, i.e., Reðc† gc bp cÞ continuous, so
that we obtain
½gc bp cþ ¼ 0, ð22:128Þ
and
þ
zþ 1 g z0 iz
½cz00 ¼ ae ðgc bp c0 Þ: ð22:134Þ
4ħ g2c z
0
We recover the important result that turning on the SOI does not change
the probability current but creates pure phase shifts between the two spin
channels. This is equivalent to a spin precession around the DP field
direction and would be intuitive if the DP field were a real vector, which is
not the case.
22.7 Conclusion
Tunnel heterojunctions, where III-V semiconductors are involved, consti-
tute fundamental systems to test the reliability of transport models; in fact,
they provide a DP field and/or Rashba field when considering quasi-2D
electron gas and give valuable insight into spin-split evanescent states. In
this context, we have proposed a systematic procedure to construct
properly symmetrized current operators that reduce to standard ones when
the DP field is not taken into account in the Hamiltonian but that ensure
consistent physical properties when applied to III-V heterostructures. We
have shown that, in these systems, the boundary conditions for solution to
the tunneling problems have to be consistently changed, adapting the
BenDaniel–Duke approach to deal with heterostructures where the SOI is
included and introducing generalized boundary conditions. We have
shown that, up to the second order, the usual matching conditions, i.e., the
continuity of the envelope function and of the velocity at an interface,
yield solutions that comply with the conservation of the probability
current. This no longer holds as soon as the cubic D’yakonov–Perel’ term
is included. We have illustrated our findings from the model case of a
½110-oriented GaAs barrier (previously addressed in Ref. 4) and have
recovered and generalized the results in a more simple and direct
treatment. A striking result is that the envelope function is discontinuous
at the interfaces. This discontinuity may also be taken as a boundary
condition. The tools we have developed can be applied to the holes in the
valence bands or to the electrons in the conduction band, making them
important for semiconductor-based spintronics, and especially for spin-
orbit engineering of heterostructures.
584 Chapter 22
Appendix
lð1Þ fx, y, zg cj , lð1Þ ½dr0 bp lð1Þ þ bp lð1Þ dr0 , and clð1Þ, lð2Þ ¼ ð1=2mÞdlð1Þ, lð2Þ . The
aim of this appendix is to show that, for a Hamiltonian H ðnÞ þ V ðrÞ, the
following form of the j th component of the probability current operator
ðnÞ
X
Jb j ðr0 Þ ¼ cj, lð1Þ, ..., lðn 1Þ ½dr0 bp lð1Þ bp lð2Þ . . . bp lðn 1Þ þ bp lð1Þ dr0 bp lð2Þ . . .bp lðn 1Þ
lðkÞ2fx, y, zg
k 1, ..., n1
¼ ½ bp lð1Þ cðr0 Þ† bp lð2Þ . . . bp lðn 1Þ cj , lð1Þ, ..., lðn 1Þ cðr0 Þ: ð22:A:2Þ
Spin–Orbit Engineering of Semiconductor Heterostructures 585
where c ¼ cðr0 Þ. From Eq. (22.3), we can find the generic divergence term
related to the derivative with respect to bp j :
ðnÞ
X
bp j Jj ½c ¼ c† bp j bp lð1Þ . . . bp lðn 1Þ c ðbp j cÞ† bp lð1Þ . . . bp lðn 1Þ cj, lð1Þ,..., lðn 1Þ c
lðkÞ2fx, y, zg
k 1, ..., n1
†
þ ðbp lð1Þ . . . bp lðk 1Þ cÞ bp j bp lðkÞ . . . bp lðn 1Þ cj , lð1Þ, ..., lðn 1Þ c
†
ðbp j bp lð1Þ . . . bp lðk1Þ cÞ b p lðkÞ . . . bp lðn 1Þ cj, lð1Þ,..., lðn 1Þ c
†
þ ðbp lð1Þ . . . bp lðkÞ cÞ bp j bp lðkþ1Þ . . . bp lðn 1Þ cj, lð1Þ,..., lðn 1Þ c
ðbp j bp lð1Þ . . . bp lðkÞ cÞ† bp lðkþ1Þ . . . bp lðn 1Þ cj, lð1Þ,..., lðn 1Þ c þ ...
þ ðbp lð1Þ . . . bp lðn 1Þ cÞ† bp j cj , lð1Þ,..., lðn 1Þ c
†
ðbp j bp lð1Þ . . . bp lðn 1Þ cÞ cj , lð1Þ, ..., lðn 1Þ c : ð22:A:4Þ
In Eq. (22.4) all of the terms that have the same order in bp (two consecutive
terms except for the first and last one) vanish after summation over j:
X X h
ðbp j bp lð1Þ . . . bp lðk 1Þ cÞ† bp lðkÞ . . . bp lðn 1Þ cj, lð1Þ,..., lðn 1Þ c
j fx, y, zg lðkÞ2fx, y, zg
k 1, ..., n1
þ ðbp lð1Þ . . . bp lðkÞ cÞ† bp j bp lðkþ1Þ . . . bp lðn 1Þ cj , lð1Þ, ..., lðn 1Þ c ¼ 0: ð22:A:5Þ
Now r JðnÞ ½c ¼ ði=ħÞbp JðnÞ ½c, Eq. (22.A.7) results in a collection of
pure imaginary terms, and the final expression for the divergence of the
probability current reads
2 X X
r JðnÞ ½c ¼ Im ðcjbp j bp lð1Þ . . . bp lðn 1Þ cj, lð1Þ;...;lðn 1Þ jcÞ:
ħ j fx, y, zg lðkÞ2fx, y, zg
k 1, ..., n1 ð22:A:7Þ
Eventually, X
r J½c ¼ r JðnÞ ½c: ð22:A:8Þ
n
This straightforwardly explains, for instance, how tunneling occurs for electrons located in
semiconductor side valleys (the SOI is not considered here). Indeed, the states associated with
the wave vectors z þ q, where z corresponds to the center of the considered valley, and to z q,
are energy degenerate (in this example, we do not necessarily have z q ¼ 0). To solve such a
problem, we only have to find usual formal solutions dealing with plane waves of vectors q
outside the barrier, and iK in the barrier, respectively. This provides us with the reflection and
transmission coefficients. Then, we obtain the true wavefunctions by multiplying the formal
solutions by eizr .
Spin–Orbit Engineering of Semiconductor Heterostructures 589
References
1. I. Žutić, J. Fabian, and S. Das Sarma, “Spintronics: Fundamentals and
applications,” Rev. Mod. Phys. 76, 323 (2004).
2. E. B. Sonin, “Spin currents and spin superfluidity,” Adv. Phys. 59, 181
(2010).
3. J.-L. Basdevant and J. Dalibard, Quantum Mechanics, Springer, Berlin,
p. 50 (2005).
4. T. L. Hoai Nguyen, H.-J. Drouhin, J.-E. Wegrowe, and G. Fishman,
“Spin rotation, spin filtering, and spin transfer in directional tunneling
through barriers in noncentrosymmetric semiconductors,” Phys. Rev. B
79, 165204 (2009).
5. Yu. A. Bychkov and E. I. Rashba, “Properties of a 2D electron gas with
lifted spectral degeneracy,” Pis’ma Zh. Eksp. Teor. Fiz. 39, 66 (1984)
[JETP Lett. 39, 78 (1984)].
6. E. I. Rashba and V. I. Sheka, in Landau Level Spectroscopy, G. Landwehr
and E. I. Rashba, Eds. Elsevier, Amsterdam, p. 178 (1991).
7. G. Dresselhaus, “Spin-orbit coupling effects in zinc blende structures,”
Phys. Rev. 100, 580 (1955).
8. M. I. D’yakonov and V. I. Perel’, “Spin orientation of electrons associated
with the interband absorption of light in semiconductors,” Zh. Eksp. Teor.
Fiz. 60, 1954 (1971); Sov. Phys. JETP 33, 1053 (1971). M. I. D’yakonov
and V. I. Perel’, “Spin relaxation of conduction electrons in noncentro-
symmetric semiconductors,” Fiz. Tverd. Tela, 13, 3581 (1971); Sov. Phys.
Solid State 13, 3023 (1972).
9. F. Bottegoni, H.-J. Drouhin, G. Fishman, and J.-E. Wegrowe,
“Probability- and spin-current operators for effective Hamiltonians,”
Phys. Rev. B 85, 235313 (2012).
10. E. I. Rashba, “Spin currents in thermodynamic equilibrium: The
challenge of discerning transport currents,” Phys. Rev. B 68, 241315(R)
(2003).
11. J. Sinova, D. Culcer, Q. Niu, N. A. Sinitsyn, T. Jungwirth, and
A. H. MacDonald, “Universal intrinsic spin hall effect,” Phys. Rev. Lett.
92, 126603 (2004).
12. E. B. Sonin, “Equilibrium spin currents in the Rashba medium,” Phys.
Rev. B 76, 033306 (2007) and Erratum: “Equilibrium spin currents in the
Rashba medium,” Phys. Rev. B 77, 039901(E) (2008).
13. V. A. Sablikov and Y. Ya. Tkach, “Evanescent states in two-dimensional
electron systems with spin-orbit interaction and spin-dependent trans-
mission through a barrier,” Phys. Rev. B 76, 245321 (2007).
590 Chapter 22
46. C. Kittel, Quantum Theory of Solids, Wiley & Sons, New York, p. 183
(1987).
47. T. L. Hoai Nguyen, H.-J. Drouhin, and G. Fishman, “Spin trajectory
along an evanescent loop in zinc-blende semiconductors,”Phys. Rev. B
80, 075207 (2009).
48. E. A. de Andrada e Silva, G. C. La Rocca, and F. Bassani, “Spin-orbit
splitting of electronic states in semiconductor asymmetric quantum
wells,” Phys. Rev. B 55, 16293 (1997).
49. A. Messiah, Quantum Mechanics, North-Holland, Amsterdam, Ch. X,
Sec. 4, p. 372 (1962).
Yufei Ma
Electrical & Computer Engineering Department, Rice University, Houston,
Texas, USA and National Key Laboratory of Science and Technology on
Tunable Lasers, Harbin Institute of Technology, Harbin, China
Manijeh Razeghi
Center for Quantum Devices, Department of EECS, Northwestern University,
Evanston, Illinois, USA
Frank K. Tittel
Electrical & Computer Engineering Department, Rice University, Houston,
Texas, USA
597
598 Chapter 23
23.1 Introduction
23.2 Tunable Diode Laser Absorption Spectroscopy (TDLAS) for Ethane Detection
23.2.1 Laser characterization
23.2.2 Optical sensor architecture
23.2.3 Experiments and results
23.3 Environmental Detection of Ammonia using an EC QCL based C PAS Sensor Platform
23.3.1 Sensor configuration and results
23.4 Quartz Enhanced Photoacoustic Spectroscopy (QEPAS)
23.4.1 Methane and nitrous oxide detection
23.4.1.1 7.83 mm DFB QCL for methane and nitrous oxide detection
23.4.1.2 Experiments and results
23.4.2 Environmental detection of nitric oxide
23.4.3 QEPAS based ppb level detection of carbon monoxide and nitrous oxide
23.4.3.1 CW DFB QCL based QEPAS sensor system for CO and N2O
23.4.3.2 Experimental results and discussion
23.4.4 Sulfur dioxide experiments
23.4.4.1 SO2 QEPAS sensor architecture and performance
23.5 Conclusions
Acknowledgments
References
23.1 Introduction
The development of highly sensitive and selective optical sensor systems using
tunable semiconductor-laser-based spectroscopic trace-gas detection techniques
is reported in this chapter. The quantitative detection and monitoring of trace-
gas molecules in real-world applications such as atmospheric chemistry,
pollution monitoring, and industrial process control in most cases require the
targeting of fundamental vibrational–rotational (V-T) molecular absorption
bands located between the 3- and 24-mm wavelengths. The mid-infrared
fundamental absorption bands of several small molecules of potential interest
for trace-gas monitoring are shown in Fig. 23.1 within two mid-infrared
atmospheric transmission windows. The upper panel shows absorption spectra
in the atmospheric window between the bending fundamental of water centered
at around 1600 cm 1 and the water OH stretches starting above 3200 cm 1.
The lower panel shows absorption spectra in the atmospheric window below the
water bending fundamental. The logarithmic ordinate scales are the integrated
intensities of the lines on a per-molecule basis. These spectral regions can be
covered by narrow-linewidth and high-performance semiconductor lasers, in
particular quantum cascade lasers (QCLs)1 2 and interband cascade lasers
(ICLs).3 4 Therefore, trace-gas optical spectroscopic sensors using a QCL or
ICL as an excitation source are responsible for improving the spectral
resolution of the measurements and achieving real time, continuous ultrasensi-
tive detection of trace-gas molecular species at the concentration levels from the
percent level down to parts per trillion (ppt).
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 599
Figure 23.1 High-resolution simulated molecular absorption spectra within two mid-infrared
atmospheric transmission windows (reprinted from Ref. 14). (See color plate section.)
For quantitative measurements of CH4 and N2O, SO2, NO, and CO,
different CW room-temperature (RT) DFB QCL sources emitting at 7.83,
7.24, 5.26, and 4.6 mm were employed, respectively. As a detection technique,
a novel approach to the photoacoustic detection of trace gases, utilizing a
quartz tuning fork (QTF) as an acoustic transducer, was employed. This
technique, named quartz-enhanced photoacoustic spectroscopy (QEPAS),
was first reported by our Rice University Laser Science Group in 2002.13,36
The key innovation of QEPAS is to invert the common CPAS approach and
accumulate the acoustic energy in a sharply resonant piezoelectric QTF with a
very high quality Q factor of >10,000, rather than in a broadband
microphone with a low-Q-factor (200) resonant CPAS gas cell. Due to the
very narrow QTF resonance curve, QEPAS possesses high immunity to
environmental acoustic noise. Moreover, small dimensions of the QTF allow
performance of sensitive trace-gas concentration measurements using an
ultrasmall acoustic detection module (ADM), where the total volume of the
analyzed gas sample is <4 mm3.37
A readily available QTF that is well suited for QEPAS applications is
commonly used in digital clocks and watches as frequency standards and has
a resonance frequency at 32,768 Hz in vacuum. The QTF is able to detect
weak acoustic waves generated when the modulated optical radiation interacts
with a trace gas. The mechanical deformation of the QTF due to interaction
with the acoustic waves results in the generation of electrical charges on its
electrode pairs. The electrical charges can be measured as either a current or
voltage by the electrodes that are connected to an ultralow-noise preamplifier
and lock-in detection electronics. An enhancement of the QEPAS signal can
be achieved when two metallic tubes acting as a microresonator (mR) are
added to the QTF sensor architecture. A typical configuration, used in most
reported QEPAS-based gas sensors where the QTF is positioned between the
mR tubes to probe the acoustic waves excited in the gas contained inside the
mR, can be found in Refs. 10, 16, 36, 38, and 39. Other QEPAS
configurations, such as off-beam QEPAS, are also possible.40,41 Furthermore,
two novel modifications of the QEPAS sensor architecture based on
interferometric photoacoustic spectroscopy42 and resonant opto-thermoacoustic
detection43 were recently reported.
A recent optimization study of the geometrical mR parameters showed
that the highest QEPAS signal-to-noise ratio (SNR) is achieved for two
4.4-mm-long and 0.5- to 0.6-mm-inner-diameter tubes.39 However, for a
typical QCL beam, short mR tubes with a larger inner diameter are
advantageous in facilitating the optical alignment of the QCL excitation beam
with respect to the mR and the QTF. Therefore, to simplify the optical
alignment process and eliminate any potential optical fringes, the 4-mm-long
tubes with 0.84-mm inner diameter are commonly used for QEPAS
experiments in the mid-infrared region. A direct, side-by side comparison of
602 Chapter 23
Figure 23.2 (a) Light intensity versus current and voltage (LIV) curves for a 3.36-mm CW,
TEC, DFB laser diode operating at 10 8C. (b) Temperature and current tuning characteristics
of the laser diode (reprinted from Ref. 5).
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 603
Figure 23.3 Schematic of a C2H6 gas sensor using a 3.36-mm DFB laser diode as an
excitation source. M – mirror, CL – collimating lens, DM – dichroic mirror, MC – multipass
cell, L – lens, SCB – sensor control board.
604 Chapter 23
Figure 23.4 (a) Photograph of an innovative long-path, small-volume multipass gas cell
(MC dimensions: 17 6.5 5.5 cm; distance between MC mirrors: 12.5 cm; effective optical
path length: 57.6 m with 459 passes) with novel compact surface-mounted control
electronics from Sentinel Photonics. (b) Spot pattern of a modified spherical Herriott MC
(from Sentinel Photonics website).
Figure 23.5 (a) 2f WMS signal for a C2H6 line at 2976.8 cm1 at 200 Torr obtained with
compact multipass gas cell. (b) 2f WMS amplitude signal as function of C2H6 concentration.
Figure 23.6 (a) Mid-infrared AM-PAS-based sensor platform for NH3 atmospheric
detection. (b) AM-PAS signal for reference mixture of 5-ppm NH3 in N2 after one, two,
and three laser beam passes through the photoacoustic cell.
photoacoustic cell, was 2 ppbv for a 5-s data acquisition time [see Fig. 23.6(b)].
After averaging the data for 100 s, a sub-ppbv NH3 concentration level of
0.7 ppbv was achieved.
The AM-PAS-based NH3 sensor platform was deployed at the University
of Houston North Moody Tower monitoring site during three different
seasons (winter 2010, summer/fall 2011, winter 2011) and at the continuous
ambient monitoring station (CAMS) 75 monitoring site in the Fort Worth/
Dallas area (between May 30, 2011 and June 30, 2011) as part of the summer
2011 TCEQ campaign. The 2010/2011 results of environmental ammonia
measurements are reported in Ref. 8. The NH3 sensor platform was again
deployed on top of the University of Houston North Moody Tower during
the 2012 summer months in order to investigate the correlations between NH3
and particle number concentrations (PNCs) [measured by a condensation
particle counter (CPC)] and to compare NH3 concentration levels over the
same period of last year. Monitoring PNC alongside NH3 levels and
incorporating meteorological parameters as well as other air pollutants data
measured by a TCEQ CAMS allows us to investigate the effect of NH3 on
local and regional air quality with respect to PM formation.
A time series of atmospheric NH3 concentration and particle number
concentration levels for city of Houston, TX, measured between July 19, 2012
and July 31, 2012 is depicted in Fig. 23.7. The emission sources that were
identified as being responsible for some of the elevated concentration levels in
the Greater Houston urban area are: the W. A. Parish electric power plant
and the Houston Ship Channel. In general, the high NH3 concentration levels
observed during this period were expected due to the more intense ammonia
evaporation from natural sources (soil, vegetation, decomposition of plants)
and agricultural (animal waste, mineral fertilizer, pesticides). In addition, high
ambient temperatures observed during the summer significantly improve the
conversion of ammonia from the particle phase to gas phase.
Figure 23.7 Time series of atmospheric NH3 concentration and particle number concentration
levels for Houston, TX, measured from July 20 to August 21, 2012. (See color plate section.)
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 609
Figure 23.8 (a) Measured optical power for Adtech Optics 7.83-mm CW DFB QCL
operating at 15, 25, 35, and 45 8C. (b) Temperature and current properties.
610 Chapter 23
optical modulation depth and the CH4 and N2O gas pressure, a detailed
evaluation of the QEPAS system was performed.
Figure 23.9 depicts the 2f QEPAS signals for N2O and CH4 in ambient
laboratory air (blue plot) and for a calibrated mixture of 1.8-ppmv N2O in N2
(red plot). Both measurements were recorded at a total gas pressure of 100
Torr.
Additional enhancement of the QEPAS signal amplitude was achieved by
the addition of water vapor, which is an efficient accelerator for V-T
relaxation processes in the gas phase. For the targeted N2O absorption line
located at 1275.5 cm 1, the laboratory air N2O concentration was calculated
to be 393 ppbv based on the two N2O QEPAS measurements shown in
Fig. 23.9. The minimum detectable concentration (MDC) of the currently
developed QEPAS system for CH4 and N2O detection is 7 ppbv and 20 ppbv,
respectively, for a 1-s data acquisition time. The obtained MDC values allow
performance of long-term and sensitive environmental measurements of CH4
and N2O because their natural abundance in the atmosphere is at a constant
level of 1.8 ppmv and 320 ppbv, respectively.
Our next goal is to implement a line-locking procedure for CH4 and N2O
detection. Moreover, we plan to transfer the current QEPAS optical platform
from a standard laboratory optical table to a custom compact optical
platform (see Fig. 23.10). During this procedure, overall dimensions of the
QEPAS sensor will be optimized in order to achieve a compact and
transportable sensor system for CH4 and N2O and detection at 1275.5 cm 1
and 1275.041 cm 1, respectively.
Figure 23.9 2f QEPAS signals for a moisturized 1.8-ppmv mixture of N2O in N2 (red plot) and
for CH4 and N2O in ambient laboratory air (blue plot); the dotted curve represents the optical
sensor baseline. Total gas pressure for both scans was P ¼ 100 Torr. (See color plate section.)
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 611
Figure 23.10 (a) Front view and (b) side view of the portable compact sensor for CH4 and
N2O detection.
Figure 23.11 (a) Schematic and (b) completed design of a compact, RT, CW, DFB-QCL-
based NO sensor platform. Pc L – plano-convex lens, Ph – pinhole, QTF – quartz tuning
fork, mR – acoustic microresonator, RC – reference cell.
(WM) technique was used. A DFB-QCL current and temperature were set and
controlled by a custom-made control electronics unit (CEU), which is also
employed to modulate the laser current, to lock the laser frequency to the
selected absorption line, and to measure the current generated by the QTF in
response to the photoacoustic signal.
In order to improve the quality of the QCL beam, two plano-convex CaF2
lenses (40-mm and 25-mm focal length) and a 300-mm pinhole are used as a
spatial filter. The second lens is used to direct the laser radiation through the
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 613
mR and between the prongs of the QTF mounted inside the ADM with a
transmission efficiency of 98%. The QCL beam exiting from the ADM is
directed to a 5-cm-long reference cell, filled with a 0.11% NO in N2 mixture at
175 Torr and a pyroelectric detector (InfraTec, LIE-332f-63). The 3f
pyroelectric detector signal with a zero crossing point at the maximum of
the 2f WM QEPAS signal was used as the reference signal for line locking of
the QCL frequency to the center of the selected NO absorption line. All of the
measurements were performed at a gas pressure of 250 Torr, which, according
to previous experimental results, is within the optimal pressure range for
QEPAS-based NO detection.10
The 2f QEPAS signal when the DFB-QCL frequency is tuned across
and locked to the H2O and CO2 interference-free NO doublet absorption line at
1900.08 cm 1 is depicted in Figs. 23.12(a) and (b), respectively. For a 95-ppb NO in
N2 calibrated mixture and 2.5% water vapor concentration, the calculated noise-
equivalent (1 s) concentration of NO with a 1-s averaging time is 3 ppbv at gas
pressure of 250 Torr. The corresponding absorption coefficient normalized to the
detection bandwidth and optical power is 6.2 10 9 cm 1 W/Hz1/2.
To investigate the long-term stability of the NO sensor platform, an Allan
variance analysis was performed while an ultrapure nitrogen was flushed
through the ADM cell. A graph of the Allan deviation, defined as the square
root of the Allan variance, is depicted in Fig. 23.13(a). This plot shows that the
measurement accuracy improves with averaging time, since random noise
sources can be effectively removed by the averaging process. For the completed
NO QEPAS sensor, the optimum averaging time is 200 s, which corresponds to
an improved NO MDC of 0.3 ppbv [red line of Fig. 23.13(a)]. For the
purpose of environmental monitoring, where sensor time response is not a
critical parameter, a 200-s averaging time can be normally utilized to allow
a detection limit of NO below 1 ppbv.
Figure 23.12 (a) 2f QEPAS signal amplitude when QCL frequency is tuned across and
(b) locked to the NO doublet absorption line at 1900.08 cm1.
614 Chapter 23
Figure 23.13 (a) Allan deviation calculated for a period during which ultra-high-purity
nitrogen was flushed through the NO QEPAS sensor system. Blue, green, and red lines
show the improved NO sensitivity after 25, 100, and 200 s of the averaging time. (b) NO
atmospheric data for different averaging times. (See color plate section.)
Figure 23.14 (a) LIV curve of the 4.61-mm, RT, CW, DFB-QCL from Center for Quantum
Devices, Northwestern University. (b) DFB-QCL current tuning at different DFB-QCL
operating temperatures (reprinted from Ref. 11). (See color plate section.)
frequency was set to the center of the absorption line and actively controlled
by the feedback signal of the internal CEU proportional controller based on
the 3f component from the pyroelectric detector output [see Fig. 23.11(a) as an
example]. The proportional controller signal compensates for any laser
frequency drift by generating a correction signal to maintain the frequency
always at the center of the targeted absorption line.
For the current CO sensor platform based on the NO design described in
Section 23.4.2, the maximum 2f QEPAS signal level of the R(6) CO absorption
line is obtained at 600 Torr and with a modulation depth of 50 mA. For the
same modulation depth, the QEPAS signal value at a pressure of 760 Torr is
only 10% lower as compared with the maximum CO signal. However, from a
practical point of view, the QEPAS sensor is operated at atmospheric pressure
because in this case the pressure controller and flow meter become redundant.
Similar to the NO QEPAS system, the CO signal amplitude is strongly
dependent on the V-T relaxation rate. Hence, the addition of water vapor to
the analyzed gas mixture helps to improve the energy transfer for the V-T
states of the excited CO molecules. Therefore, water vapor at different
concentration levels was added into the CO:N2 gas mixture by means of a
commercial permeation tube (Perma Pure model MH-110-24F-4), which was
immersed inside a water-circulating bath (LAUDA-Brinkmann, LP., RM6).
Figure 23.16 (a) QEPAS signal amplitude recorded in the line-locking mode as the CO
concentration is varied at atmospheric pressure and with a modulation depth of 40 mA.
(b) QEPAS signals amplitude averaged from (a) as a function of CO concentration. 1 cnt ¼
6.67 1016 A (reprinted from Ref. 11).
Figure 23.17 Continuous monitoring of atmospheric CO and N2O concentration levels from
an air sampled on Rice University campus, Houston, TX, USA (latitude and longitude are:
29 deg 430 N/95 deg 230 W). (a) CO concentration measurements. (b) N2O concentration
measurements (reprinted from Ref. 11).
Figure 23.18 (a) Single-frequency CW DFB-QCL radiation for different laser current values
at a QCL-operating temperature of 20 8C. (b) CW DFB-QCL optical power and current tuning
at four different operating temperatures (reprinted from Ref. 12). (See color plate section.)
Figure 23.19 2f WM QEPAS signals for different SO2 concentrations when the laser was
tuned across the 1380.9 cm1 line. a.a – ambient air. Inset: Dependence of the measured 2f
signals as a function of SO2 concentrations (reprinted from Ref. 12). (See color plate section.)
622 Chapter 23
its inset. The noise level was determined from the baseline recorded with the gas
cell filled with ambient air. The minimum detectable SO2 concentration level
(1 s) was calculated to be 100 ppb for the lock-in amplifier time constant set
to 1 s. In order to distinguish single absorption lines from the congested
spectrum of the SO2, the pressure inside the ADM was kept at 100 Torr.
Future tasks for SO2 detection will include line locking for continuous and
long-term measurements as well as using a high-power (>200 mW) QCL
targeting the strongest SO2 absorption line at 1348 cm 1. In this situation, the
SO2 detection limit can be improved to low ppb levels; therefore, QEPAS
sensors will have the potential to be used as a compact, portable, analytical
devices to perform real-time environmental and industrial emission measure-
ments of SO2.
23.5 Conclusions
This chapter described autonomously operated, compact, reliable, and real-
time gas sensors and demonstrated their potential as practical monitoring
devices in atmospheric chemistry, and urban and rural environmental
monitoring as well as in industrial emission and manufacturing control
processes. These sensors can also be used in biomedical and life sciences, such
as for noninvasive medical diagnostics that involve the detection and
monitoring of numerous exhaled breath biomarkers, and for sensing of
trace-gas species relevant to spacecraft habitat air quality and safety or
planetary atmospheric science. Several examples of ultra-sensitive and
selective sensors, with detection limits at ppb and sub-ppb levels, are reported
using mid-infrared CW DFB diode laser EC-QCL and DFB-QCL excitation
sources. Details of three spectroscopic techniques that included TDLAS,
CPAS, and QEPAS were discussed. The choice of a specific spectroscopic
measurement method was determined by the application as well as the readily
commercial availability of the appropriate mid-infrared laser source. For
example, to detect C2H6, a 3.36-mm CW TEC TO-5 packaged diode-laser-
based TDLAS sensor system with a 1 s C2H6 detection sensitivity of 130 pptv
(1-s sampling time) was employed. For environmental NH3 detection, a 1 s
NH3 detection sensitivity of 1 ppbv (200-s averaging time) was reported for
a 10.36-mm Daylight Solutions EC-QCL-based CPAS sensor. To monitor
CH4 and N2O, NO, CO, and SO2 concentration levels, four different QEPAS-
based sensors based on a DFB-QCL from AdtechOptics (7.28 mm), Maxion
Technologies (5.26 mm), Northwestern University (4.61 mm), and Hama-
matsu (7.24 mm) were employed, respectively. Detection sensitivities for these
five gas species were 20 ppbv, 7 ppbv, 3 ppbv, 2 ppbv, and 100 ppbv,
respectively for a 1-s sampling time. Improvements and innovations in
TDLAS-, CPAS-, and QEPAS-based sensor platforms (in particular more
stable, mass-produced optical/mechanical designs as well as data acquisition
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 623
Table 23.1 Near-IR laser diode and mid-IR QCL-based QEPAS performance for 17 trace-
gas species. RH – relative humidity.
Molecule (host) Frequency (cm 1) Pressure (Torr) NNEA Power (mW) NEC (ppmv)
(cm 1W/Hz1/2)
Acknowledgments
The Rice University group acknowledges financial support from a National
Science Foundation (NSF) grant EEC-0540832 entitled “Mid-Infrared
Technologies for Health and the Environment (MIRTHE),” an NSF-ANR
award for international collaboration in chemistry: “Next generation of
compact infrared laser-based sensor for environmental monitoring
(NexCILAS)” and grant C-0586 from the Robert Welch Foundation.
References
1. J. Faist, F. Capasso, D. L. Sivco, A. L. Hutchinson, and A. Y. Cho,
“Quantum cascade laser,” Science 264, 553–556 (1994); J. Faist, Quantum
Cascade Lasers, Oxford University Press, Oxford (2013).
2. F. Capasso, “High-performance mid-infrared quantum cascade lasers,”
Opt. Eng. 49(11), 111102 (2010) [doi: 10.1117/1.3505844].
3. R. Q. Yang, “Infrared laser based on intersubband transitions in quantum
wells,” Superlattices and Microstructures 17, 77–83 (1995).
4. J. R. Meyer, I. Vurgaftman, R. Q. Yang, and L. R. Ram-Mohan, “Type-
II and type-I interband cascade lasers,” Electron. Lett. 32, 45–46 (1996).
5. K. Krzempek, R. Lewicki, L. Nähle, M. Fischer, J. Koeth, S. Belahsene,
Y. Rouillard, L. Worschech, and F. K. Tittel, “Continuous wave,
distributed feedback diode laser based sensor for trace-gas detection of
ethane,” Appl. Phys. B 106, 251–255 (2012).
6. A. Kosterev, L. Dong, D. Thomazy, F. Tittel, and S. Overby, “QEPAS
for chemical analysis of multi-component gas mixtures,” Appl. Phys. B:
Lasers and Optics 101, 649–659 (2010).
7. M. Jahjah, S. Belahsene, L. Nähle, M. Fischer, J. Koeth, Y. Rouillard, and
A. Vicet, “Quartz enhanced photoacoustic spectroscopy with a 3.38 mm
antimonide distributed feedback laser,” Opt. Lett. 37, 2502–2504
(2012).
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 625
20. J. Ye, L.-S. Ma, and J. L. Hall, “Ultrasensitive detections in atomic and
molecular physics: Demonstration in molecular overtone spectroscopy,”
J. Opt. Soc. Am. B 15, 6–15 (1998).
21. A. Foltynowicz, W. Ma, and O. Axner, “Characterization of fiber-laser-
based sub-Doppler NICE-OHMS for quantitative trace gas detection,”
Opt. Express 16, 14689–14702 (2008).
22. G. Litfin, C. R. Pollock, R. F. Curl, and F. K. Tittel, “Sensitivity
enhancement of laser absorption spectroscopy by magnetic rotation
effect,” J. Chem. Phys. 72, 6602–6605 (1980).
23. H. Ganser, W. Urban, and A. M. Brown, “The sensitive detection of NO
by Faraday modulation spectroscopy with a quantum cascade laser,”
Molec. Phys. 101, 545–550 (2003).
24. R. Lewicki, J. H. Doty, R. F. Curl, F. K. Tittel, and G. Wysocki,
“Ultrasensitive detection of nitric oxide at 5.33 mm by using external
cavity quantum cascade laser-based Faraday rotation spectroscopy,”
Proc. Natl. Acad. Sci. USA 106, 12587–12592 (2009).
25. M. J. Thorpe, D. D. Hudson, K. D. Moll, J. Lasri, and J. Ye, “Cavity-
ringdown molecular spectroscopy based on an optical frequency comb at
1.45-1.65 mm,” Opt. Lett. 32, 307–309 (2007).
26. M. J. Thorpe, D. Balslev-Clausen, M. S. Kirchner, and J. Ye, “Cavity-
enhanced optical frequency comb spectroscopy: Application to human
breath analysis,” Opt. Express 16, 2387–2397 (2008).
27. B. Hirst, G. Gibson, S. Gillespie, I. Archibald, O. Podlaha, K. D. Skeldon,
J. Courtial, S. Monk, and M. Padgett, “Oil and gas prospecting by ultra-
sensitive optical gas detection with inverse gas dispersion modelling,”
Geophys. Res. Lett. 31, L12115 (2004).
28. C. Wang and P. Sahay, “Breath analysis using laser spectroscopic
techniques: Breath biomarkers, spectral fingerprints, and detection
limits,” Sensors 9, 8230–8262 (2009).
29. A. Bari, V. Ferraro, L. R. Wilson, D. Luttinger, and L. Husain,
“Measurements of gaseous HONO, HNO3, SO2, HCl, NH3, particulate
sulfate and PM2.5 in New York, NY,” Atmospheric Environment 37,
2825–2835 (2003).
30. A. Rossi, R. Buffa, M. Scotoni, D. Bassi, S. Iannotta, and A. Boschetti,
“Optical enhancement of diode laser-photoacoustic trace gas detection by
means of external Fabry-Perot cavity,” Appl. Phys. Lett. 87, 041110
(2005).
31. A. Miklos, P. Hess, and Z. Bozoki, “Application of acoustic resonators in
photoacoustic trace gas analysis and metrology,” Rev. Sci. Instrum. 72,
1937–1955 (2001).
Current Status of Mid-infrared Semiconductor-Laser-based Sensor Technologies… 627
Wolfgang Schade
Institute for Energy Research and Physical Technologies (IEPT) and Energy
Research Center of Lower Saxony (EFZN), Clausthal University of Technology,
Clausthal-Zellerfeld, Lower Saxony, Germany and Fraunhofer Heinrich Hertz
Institute (HHI), Goslar, Germany
24.1 Introduction
24.2 Pulsed Laser Fragmentation
24.3 Experimental Setup
24.4 Results
24.5 Discussion
24.6 Conclusions
References
24.1 Introduction
Spectroscopic techniques are widely used in a variety of applications due to
their pre-eminent properties: They allow in situ and online tracing of
compounds, require, in most cases, no sample preparation, and can be used to
set up rugged and easy-to-use sensor devices. In contrast to chemical sensing
methods, the results are available within seconds, making these techniques
feasible for the monitoring of distinctive species as well as the accurate
633
634 Chapter 24
Figure 24.1 Experimental setup for laboratory measurements (QCL – quantum cascade
laser, DAQ – data acquisition).
and 1600 cm 1 are used, respectively. They are operated at a repetition rate of
120 kHz and exhibit quasi-continuous-wave powers in the microwatt range.
The second setup shows a standoff configuration (Fig. 24.2), in which the
probing lasers are aimed at the gas volume where the generated nitrous oxides
are present, and the reflected and backscattered light is collected with a
telescope and imaged onto the detector. Here, for practical reasons, the
chemical reaction between nitric acid and copper has been used to generate
NO and NO2 instead of a fragmentation laser (3Cu þ 8HNO3 ! 2NO þ
3Cu(NO3)2 þ 4H2O and 2HNO3 þ NO ! H2O þ 3NO2). The setup was
realized outdoors to provide the distance of 20 m as well as for ventilation,
thus, omitting the need for exhaust aspiration.
24.4 Results
Pilot measurements using chemiluminescence to analyze the plumes generated
by photofragmentation revealed that energetic material and nonenergetic
material show different ratios of NO and NO2, making a discrimination of
harmful and harmless substances possible.12 However, chemiluminescence
measurements involve bulky equipment and, more importantly, the generated
gas is led via a hose to the analyzer, presenting the risk of chemical reaction
before analysis. Furthermore, a setup in standoff configuration is impractical.
In contrast, by using two suitable QCLs for the detection of the nitrous
oxides, an in situ measurement after fragmentation is possible, omitting these
disadvantages. The two QCLs were used for conventional MIR absorption
spectroscopy. While this typically ensures a quantitative, selective, and sensitive
analysis, here, only a qualitative evaluation is pursued because the dimension of
the generated plume and thus the optical path is unknown. Furthermore, a
homogeneous spreading within the plume cannot be expected; thus, Beer–
Lambert’s law is not easily applied. Figure 24.3 shows the extinction caused by
NO and NO2 generated by PLF of (a) RDX, (b) TNT, and (c) PETN measured
as the two probing lasers are tuned around their respective peak frequency by
application of a voltage ramp. These results have been achieved using the setup
given in Fig. 24.1 with 1-min fragmentation duration at l ¼ 532 nm and pulse
energy E ¼ 87 mJ and E ¼ 27 mJ, respectively.
A qualitative analysis of these measurements can rely either on the heights
of the extinction peaks or their area. Both methods were pursued, and both
are suitable to achieve a discrimination of the substances. As can be seen in
Fig. 24.4(a) for the area and (b) for the peak height, distinction of PETN from
the other explosives is possible by plotting the respective values of NO2 versus
those of NO. For the discrimination of RDX and TNT, some ambiguity
remains, as the correlation ellipses show.
With this procedure, a database can be generated for different explosives
and nonenergetic materials. This graphical data presentation provides the
opportunity to identify unknown substances by image recognition.
To demonstrate the appropriateness of the simultaneous detection of NO
and NO2 with QCLs in a standoff configuration, the setup described in
Fig. 24.2 was used. Here, the nitrous oxides were not generated by PLF but
were provided by the chemical reaction of nitrous acid with copper. The
instrument was placed at a distance of 20 m from the beaker, and the light that
was backscattered and reflected by a metal plate positioned behind the beaker
was collected with a telescope and focused onto the detector. Figures 24.5(a)
and (b) show the measured extinction for NO2 and NO, respectively. Since the
generated concentrations of the gases as well as their distribution in the gas
phase above the beaker and the dimension of the volume with increased
concentration are unknown (especially when taking the influence of wind into
638 Chapter 24
Figure 24.3 Extinction for cell measurements of NO and NO2 after fragmentation of RDX
(a), TNT (b) and PETN (c), respectively.
Application of Quantum Cascade Lasers for Safety and Security 639
Figure 24.4 Discrimination of different explosives by the content of generated NO and NO2
based on (a) peak area and (b) peak height.
Figure 24.5 Extinction of (a) NO and (b) NO2 produced by the chemical reaction of nitric
acid and copper measured in standoff configuration.
account for the outdoor setup), a quantitative analysis and comparison with
theoretical values is not possible. However, in the lower traces, a simulation is
given for an extinction caused by absorption on a path length of 10 cm at
atmospheric pressure. The concentration was a free parameter; similar peak
extinction values were reached for concentrations of 150-ppm NO2 and 3700-
ppm NO, respectively. From the chemical reaction, a higher NO concentra-
tion is expected, supporting the derived result.
24.5 Discussion
With the analysis of generated molecular fragments, in this case NO and NO2,
subsequent to PLF, it is possible to discriminate different explosives. The
640 Chapter 24
Figure 24.6 Measurement and simulation for the extinction caused by NO2 for a broad
spectral region and a resolution of 1 cm1. (See color plate section.)
Application of Quantum Cascade Lasers for Safety and Security 641
readily available ingredients; thus, there is a high demand for its sensing. In
contrast to nitrogen-based explosives, it possesses a high vapor pressure and is
present in the gas phase in a concentration that allows direct measurement.
Hence, not only the detection of nitrogen-based explosives subsequent to PLF
would be feasible but also the detection of TATP, with the same instrument.
24.6 Conclusions
Quantum cascade lasers are a valuable tool for the detection of gaseous
fragmentation products of explosives generated by pulsed laser fragmentation.
This method to circumvent the difficulties of their spectroscopic detection in
the gas phase caused by their extremely small vapor pressures enables the
discrimination of different harmful substances and their distinction from
harmless matter. As for all methods applied in security applications, the
demand for reliability and sensitivity as well as the prevention of false-positive
measurement results asks for a certain degree of redundancy and, thus, for the
combination of different sensing methods. An example of this combination
includes laser-induced breakdown spectroscopy, Raman spectroscopy, or
MIR absorption spectroscopy applied subsequently to PLF in one standoff
sensor device, as pursued within the European Union 7th framework project
OPTIX.17
References
1. B. C. Dionne, D. P Rounbehler, E. K. Achter, J. R. Hobbs, and
D. H. Fine, “Vapor pressure of explosives,” J. Energetic Mat. 4(1–4),
447–472 (1986).
2. J. Oxley, J. Smith, K. Shinde, and J. Moran, “Determination of the vapor
density of triacetone triperoxide (TATP) using a gas chromatography
headspace technique,” Propellants, Explosives, Pyrotechnics 30(2),
127–130 (2005).
3. A. Freedman, P. L. Kebabian, Z. Li, W. A. Robinson, and J. C.
Wormhoudt, “Apparatus for determination of vapor pressures at ambient
temperatures employing a Knudsen effusion cell and quartz crystal
microbalance,” Meas. Sci. Technol. 19(12), 125102 (2008).
4. V. Swayambunathan, G. Singh, and R. C. Sausa, “Photofragmentation:
Fragment detection and pyrolysis—Laser-induced fluorescence studies on
energetic materials,” Appl. Opt. 38(30), 6447 (1999).
5. J. Cabalo and R. C. Sausa, “Detection of hexahydro-1,3,5-trinitro-1,3,5-
triazine (RDX) by laser surface photofragmentation: Fragment detection
spectroscopy,” Appl. Spectroscopy Rev. 57(9), 1196–1199 (2003).
642 Chapter 24
25.1 Introduction
25.1.1 Standoff detection of explosives
25.1.2 In line spectroscopy of drinking water
25.2 Eye Safety in the Mid infrared Spectral Region
25.3 External Cavity Quantum Cascade Laser
25.3.1 Broadband tuning
25.3.2 Fast wavelength tuning
25.4 Standoff Detection of Explosives
25.4.1 Backscattering spectroscopy
25.4.2 Samples
25.5 Hyperspectral Data Analysis
25.5.1 Adaptive matched subspace detector
25.5.2 Background endmember extraction
25.5.3 Reference spectra
25.5.4 Experimental results
25.5.5 Larger distances
25.6 Spectroscopy of Hazardous Chemicals in Drinking Water
25.7 Conclusions
Acknowledgments
References
645
646 Chapter 25
25.1 Introduction
Broadband-tunable external-cavity quantum cascade lasers (EC-QCLs) are
established as attractive light sources for mid-infrared (MIR) ‘finger print’
molecular spectroscopy, e.g., for detection and identification of chemical
compounds. In this chapter we report on two prototypical examples of the use
of EC-QCLs for the spectroscopic detection of hazardous substances: standoff
detection of explosives and sensing of hazardous chemicals in water. In both
application scenarios, a multidisciplinary approach is applied. Our standoff
system allows the contactless identification of traces of various common
explosives over distances of several meters. Besides laser power and tuning
range, sophisticated hyperspectral image analysis is essential for providing high
sensitivity at a low false-alarm rate. For this purpose, we also address the
question of eye safety. Furthermore, because of the high spectral irradiance, the
QCL enables measurement on systems with extremely low light transmission. A
QCL-based infrared spectroscopic system allows extension of the optical path
length in water by a factor of ten. This feature enables an autonomous
measurement mode in a real-world environment using a simple transmission
cell for continuous water flow, avoiding the need for manual sampling.
QCLs are unipolar mid- to far-infrared light sources based on electrons
making intersubband transitions within the conduction band.1 In the past few
years, QCLs have experienced tremendous progress due to the improvement
in the design of both the active regions and waveguide structures. Advances
are also reported in the field of fabrication technology as well as in mounting
and encapsulating technology. The advances include room-temperature
continuous-wave (CW) operation, high output power, high wall-plug
efficiency, and wide spectral coverage.2 4 Exploiting the benefits of mature,
established materials, the emission wavelengths of QCLs can be engineered
from 3.3 mm5 to about 12 mm using InGaAs/AlInAs semiconductor
heterostructures, grown by lattice matching or strain compensation on InP
substrates.6 8 The high versatility and robustness of QCLs makes them ideal
radiation sources for real-world applications.
The MIR-wavelength range is of utmost importance for spectroscopic
applications, as the fundamental rotational–vibrational transitions of most
organic molecules are found in this energy range. Fourier transform infrared
spectroscopy (FTIR) has long been a standard method in analytical chemistry
for material analysis and identification. For certain types of applications,
spectroscopic methods based on laser sources offer superior possibilities in
comparison to the use of an FTIR setup. This is especially the case for high-
resolution spectroscopy in the gas phase, as the linewidth of a CW laser source
is much smaller than the resolution achievable even for FTIRs with very long
mirror paths. Other aspects include the much higher spectral power density
offered by a laser and the possibility of guiding a collimated laser beam over
large distances, a feature that is crucial for standoff measurement schemes.
Broadband-Tunable External-Cavity Quantum Cascade Lasers… 647
Figure 25.1 Reflectance spectra measured with FTIR spectroscopy of 4 explosives from
6.7 to 11 mm. Several strong, characteristic absorption bands suitable for identification of the
compounds lie in this spectral range. For wavenumbers <1350 cm1, the atmosphere is
sufficiently transparent for standoff detection using laser light at these frequencies. [ANFO –
ammonium nitrate fuel oil, PETN – pentaerythritol tetranitrate, RDX – Research Department
Explosive (cyclotrimethylenetrinitramine), TNT – trinitrotoluene.] (See color plate section.)
evaluate the water quality with very high sensitivity, but to provide an early
warning system for harmful concentrations of toxic substances in case of critical
incidents or terror attacks. Ideally, such systems should seamlessly fit into
existing infrastructures and run maintenance free over extended time periods.
The high spectral power density of a laser source allows IR spectroscopic
measurements on highly absorbing water films with thicknesses about 10
larger compared to traditional FTIR spectroscopy. Therefore, a continuous
flow of water through the optical cell is already possible at moderate
pressures, and the risk of frequent clogging of a measurement cell that has
been in use for longer times is greatly reduced. Consequently, the laser source
will enable a maintenance-free measurement setup with a small footprint at
reasonable costs. The spectroscopic nature of this approach allows not only
detection of the presence of dangerous substances, but also their identification
and, in addition, their concentration. This statement also holds for the case of
explosives detection that is outlined above.
Figure 25.2 Maximum permissible exposure (MPE) as power density versus exposure
time for various wavelengths according to IEC 6082517. (reprinted from Ref. 16).
Figure 25.3 (a) Normalized emission spectra of the external-cavity laser system made up of
a QCL chip with 3500-mm length. Also included are reflectance spectra recorded with the laser
setup in backscattering geometry of RDX, TNT, and PETN in the solid phase. It becomes clear
that the spectral width of the pulsed laser radiation is sufficiently narrow for spectroscopic
investigations on condensed matter. (b) EC-QCL module developed at Fraunhofer IAF.
652 Chapter 25
Figure 25.4 (a) Setup for a fast scanning external-cavity laser using a MEMS mirror;
below, a photograph of the MEMS chip is shown. (b) Emission lines at different times during
the scan. (c) Emission wavelength as a function of time, showing the expected sinusoidal
behavior due to the harmonic oscillation of the mirror. (d) Magnified view of one emission line
from (b). (See color plate section.)
Broadband-Tunable External-Cavity Quantum Cascade Lasers… 653
This mirror acts as a scanning folding mirror, while the diffractive grating is
fixed. To verify the fast wavelength tuning during oscillation of the mirror, the
laser emission is analyzed with an FTIR spectrometer equipped with a fast
mercury cadmium telluride (MCT) detector and a gated boxcar integrator.
The gate window of the boxcar and the driving current of the laser (both
having a width of 100 nm) are synchronized with the driving voltage of the
MEMS mirror. Therefore, the emission spectrum at arbitrary times during the
oscillation period of the mirror can be measured by varying the time
delay between the driving voltage of the mirror and the boxcar window.
Figure 25.4(b) shows the laser emission for different delay settings. In
Fig. 25.4(c), the spectral position and the corresponding angle of incidence on
the grating are displayed, following the expected.
Figure 25.6 Spectra of TNT (left) and PETN (right) recorded in different geometries.
(a) Standard ATR measurement with FTIR and (b) reflectance with microscope attached to an
FTIR with an angular distribution close to the specular direction. (c) Diffuse backscattering with
standoff setup. (See color plate section.)
656 Chapter 25
25.4.2 Samples
Samples were prepared by deposition of artificial fingerprints of explosives on
various substrates using a synthetic replica of a human thumb. Such fingerprints
do not leave a well-defined reproducible surface coverage on the substrate but—
in contrast to common laboratory methods of sample deposition—realistically
represent the morphology of actual residues left by a person handling explosives.
The total amount varies roughly from 20 mg to >1 mg. As substrates, we used
pieces of factory-painted sheets from the autobody of common midsized cars,
polyamide, and tissue, as well as different everyday objects (e.g., keypad of a
telephone or laptop, keys, computer mouse).
ti ¼ argfmaxr ½ðP? T ?
½t0 Ui 1 rÞ ðP½t0 Ui 1 rÞg, where Ui 1 ¼ ½t1 t2 . . . ti 1 :
If multiple target spectra are available, the initialization phase (steps one and
two) can be easily adjusted, accordingly.
The algorithm terminates either using a cutoff criterion based on the
magnitude of the projected target signature or, alternatively, using a fixed
number of endmembers (i.e., fixed dimensions of B). An appropriate choice
for the latter would be the number of spectrally distinct materials present in
the scene and corresponds to the notion of the intrinsic dimensionality of the
image.
Figure 25.7 (a) Photograph of a painted autobody sheet covered with fingerprints of different
explosives. (b) Background-corrected IR image with active illumination at 1150 cm1 of the
encircled area from (a), showing diffuse scattering from fingerprint 1 and a scratch in the car paint
above the fingerprint. (c) Fingerprint (HMTD) and the scratch can easily be distinguished by their
IR spectra. (d) Backscattering spectra of three other explosives on this sample, compared with our
reference spectra taken on a different type of autobody sheet. (See color plate section.)
shown in Fig. 25.7(c). Figure 25.7(d) also displays spectra (integrated over the
region in the camera image where the respective component resides) from
PETN, TNT, and RDX fingerprints on the same substrate. These spectra
closely match our reference spectra, recorded on fingerprints on another
painted autobody sheet from a different car manufacturer [Fig. 27.7(d)].
Figure 25.8 shows the visible image of a polyamide substrate with a series
of eight consecutive fingerprints of PETN. The green circle indicates the part
of the sample illuminated by the MIR laser, i.e., the seventh fingerprint, for
recording the backscattering image shown in Fig. 25.8(b). The amount of
deposited material was 20 mg, and the measurement distance was
approximately 1.5 m. In contrast to the autobody sheets shown above, the
polyamide surface acts as a source of diffusive scattering, as can be seen from
the background-corrected IR image with active illumination at 1200 cm 1
[Fig. 25.8(b)]. The backscattering image contains significant contributions
from the substrate. Applying the detection algorithm outlined above to the
recorded hyperspectral datacube, PETN was clearly detected, while no false
positives for TNT and RDX were obtained [Fig. 25.8(c)].
Broadband-Tunable External-Cavity Quantum Cascade Lasers… 661
Figure 25.8 (a) Photograph of the sample: repeated PETN fingerprints on polyamide;
green circle: illuminated area. (b) Background-corrected IR image with active laser
illumination; the substrate shows strong diffuse scattering. (c) Results of the detection
algorithm (AMS detector) applied to the hyperspectral datacube. The area colored red
indicates where PETN was detected with sufficient accuracy by setting a suitable threshold;
the light blue area is the part of the illuminated area where no PETN signatures in the
backscattering signal were found. Using the same threshold value as for PETN, no false
positives were found for TNT or RDX (that would have showed up as green/orange
areas). (Reprinted from Ref. 16.) (See color plate section.)
Figure 25.9 (a) Detection of PETN residues on the plastic handle of a key. The area where
PETN is detected (colored in red) is overlaid on the passive IR image of the scene. The
detection algorithm also recognized the PETN signature in the mirror image of the key on the
metallic table (lower red spot). (b) Photograph of the key (from a somewhat different angle).
The arrow indicates where the PETN fingerprint has been applied. (See color plate section.)
In Fig. 25.9 we show the detection results for PETN residues on the plastic
handle of a key. The detected area is overlaid as a color map over the passive
IR image. The PETN quantity is roughly estimated to be some tens of
micrograms, based on visual comparison to other samples with known
amounts of material. The fingerprint is clearly detected by the analysis
algorithm. Again, no false alarms appeared when the system searched for
662 Chapter 25
RDX and TNT. What seems to be a PETN false alarm just below the key is
actually the mirror image of the contaminated key on the metallic surface of
the table. Similar results have been obtained on several other everyday
objects, e.g., a computer mouse, the keypad of a telephone, or the touchscreen
of a smartphone.
Figure 25.11 Setup for transmission measurements on water. The spacer thickness was
100 mm for uncooled pyrodetectors and 200 mm for a CMT detector at 77 K.
664 Chapter 25
Figure 25.12 Absorption spectra of (a) a pesticide (100 mg/l) and (b) caffeine (200 mg/l) in
water. The spectra obtained with the QCL system agrees well with FTIR measurements on
the same substance.
Broadband-Tunable External-Cavity Quantum Cascade Lasers… 665
25.7 Conclusions
We reported on the application of broadband-tunable EC-QCLs for the
imaging standoff detection of explosives as well as for detection of
hazardous contaminations in drinking water. The EC-QCL-based standoff
detection system operates at fully eye-safe power levels. As an example,
contactless spectroscopy of fingerprints of PETN, TNT, and RDX on
different substrates has been demonstrated. A hyperspectral data analysis
algorithm was successfully employed to suppress background contributions.
As a second application for EC-QCL-based spectroscopy, detection of small
concentrations of hazardous contaminations in drinking water has been
demonstrated. Despite the strong IR absorbance of water, the high spectral
power density of the laser allows transmission spectroscopy for path lengths
exceeding 100 mm.
Acknowledgments
Most of our results have been obtained within the collaborative projects
IRLDEX, funded by the German Federal Ministry of Education and
Research (contract number FKZ 13N4543) and IRLSENS (contract number
FKZ13N11034), as well as the EU project EMPHASIS (grant agreement
number 261381). Helpful discussions with J. Beyerer (Fraunhofer IOSB) are
gratefully acknowledged. We thank K. Schwarz, U. Weinberg, O. Göhlich,
and A. Michailov for expert technical assistance, W. Schweikert and
F. Schnürer (Fraunhofer ICT) for sample preparation, and D. Richter
(TZW Karlsruhe) and M. Germer (Bruker Optics) for FTIR spectra on
drinking water samples.
666 Chapter 25
References
1. J. Faist, F. Capasso, D. L. Sivco, C. A. Sitori, A. L. Hutchinson, and
A. Y. Cho, “Quantum cascade laser,” Science 264, 553 (1994).
2. M. Razeghi, Y. Bai, S. Slivken, and S. R. Darvish, “High-performance
InP-based mid-infrared quantum cascade lasers at Northwestern Uni-
versity,” Opt. Eng. 49(11), 111103 (2010) [doi: 10.1117.1.3497623].
3. Y. Bai, N. Bandyopadhyay, S. Tsao, S. Slivken, and M. Razeghi, “Room
temperature quantum cascade lasers with 27% wall plug efficiency,” Appl.
Phys. Lett. 98(18), 181102 (2011).
4. R. Maulini, A. Lyakh, A. Tsekoun, and C. K. N. Patel, “l similar to
7.1 mm quantum cascade lasers with 19% wall-plug efficiency at room
temperature,” Opt. Express 19(18), 17203–17211(2011).
5. A. Bismuto, M. Beck, and J. Faist, “High power Sb-free quantum cascade
laser emitting at 3.3 mm above 350 K,” Appl. Phys. Lett. 98, 191104
(2011).
6. R. Maulini, A. Mohan, M. Giovannini, J. Faist, and E. Gini, “External
cavity quantum-cascade laser tunable from 8.2 mm to 10.4 mm using a
gain element with a heterogeneous cascade,” Appl. Phys. Lett. 88, 201113
(2006).
7. T. Dougakiuchi, K. Fujita, N. Akikusa, A. Sugiyama, T. Edamura, and
M. Yamanishi, “Broadband tuning of external cavity dual-upper-state
quantum-cascade lasers in continuous wave operation,” Appl. Phys.
Express 4(10) 102101 (2011).
8. A. Hugi, R. Maulini, and J. Faist, “External cavity quantum cascade
laser,” Semicond. Sci. Technol. 25(8), 083001 (2010).
9. S. Wallin, A. Petterson, H. Östmark, and H. Hobro, “Laser-based
standoff detection of explosives: A critical review,” Analyt. and Bioanalyt.
Chem. 395(2), 259–274 (2009).
10. M. Nordberg, M. Åkeson, H. Östmark, and T. E. Carlsson, “Stand-off
detection of explosive particles by imaging Raman spectroscopy,” Proc.
SPIE 8017, 80171B (2011) [doi: 10.1117/12.883179].
11. R. Furstenberg, C. A. Kendziora, J. Stepnowski, S. V. Stepnowski,
M. Rake, M. R. Papantonakis, V. Nguyen, G. K. Hubler, and
R. A. McGill, “Stand-off detection of trace explosives via resonant
infrared photothermal imaging,” Appl. Phys. Lett. 93, 224103 (2008).
12. M. B. Pushkarsky, I. G. Dunayevskiy, M. Prasanna, A. Tsekoun, R. Go,
and C. K. N Patel, “High sensitivity detection of TNT,” Proc. of the
National Academy of Sciences of the United States of America, 103(52),
19630–19634 (2006).
Broadband-Tunable External-Cavity Quantum Cascade Lasers… 667
Stefan Hugger received the Ph.D. degree in physics from the Heinrich-Heine-
University in Duesseldorf, Germany, in 2008. He joined the Fraunhofer IAF
in Freiburg, Germany in 2008, where he works on quantum cascade lasers and
their applications. Currently, he focuses on sensing of hazardous substances
using quantum-cascade-laser-based spectroscopic techniques.
26.1 Introduction
26.2 Overall Considerations
26.3 System Configuration
26.4 Demonstrations and Discussion
26.5 Summary and Future Perspectives
Acknowledgments
References
26.1 Introduction
Emission spectroscopy, including the photoluminescence (PL) spectrum, the
electroluminescence (EL) spectrum, and lasing spectra (LS), has become a
useful tool in the scientific research and industry development of semicon-
ductor optoelectronic materials and devices. Among different spectroscopy
schemes, Fourier transform infrared (FTIR) spectrometers are widely used,
especially in the mid-infrared (MIR) band. Through a simple review of FTIR
and detailed analysis of its merits and limitations in emission spectroscopy, an
improved FTIR spectroscopy system adopting a simple, convenient, and
universal emission accessory attached to a commercial FTIR spectrometer
with a diode-pumped solid state (DPSS) laser for PL is introduced in this
chapter. Based on this system, three different FTIR spectroscopy modes,
673
674 Chapter 26
rapid scan (RS), double modulation (DM), and step scan (SS), are adequately
demonstrated on various PL, EL, and LS samples.
The FTIR spectrometer, which is based on a Michelson interferometer
adopting a beam spliter with a light source, moving mirror, fixed mirror, and
photodetector on four opposite arms, is a mature and powerful tool used in
numerous fields to distinguish diversified features of materials and devices.
While dispersive spectroscopy uses a prism or grating to differentiate the
wavelengths of light in an actual spectrum, in FTIR, only an interferogram is
acquired, from which a virtual spectrum is then gained by using a Fourier
transform algorithm. Compared to a dispersive spectrometer, the high flux,
low equivalent noise, fast scan speed, and parallel beam features of FTIR
spectroscopy make it an optimal option, especially in the MIR or far-infrared
(FIR) bands, although in certain cases some artifacts in the measurement
might exist. Starting in the mid-1900s, these features have contributed to the
multiplexing (Fellgett)1 and throughput (Jacquinot)2 advantages of FTIR.
Since then, FTIR instrumentation has evolved to include computer
technology for fast Fourier transforms.3 Because of high demand,
commercial FTIR spectrometers are mainly designed for obtaining absorption
or reflection measurements using an internal polychromatic infrared radiation
source, with the specimen mounted into the sample compartment inside the
spectrometer.
Emission spectroscopy such as for PL, EL, or LS is indispensable in the
research and development of optoelectronic materials and devices. However,
most commercial FTIR spectrometers are unsuitable for direct application to
these spectra. Fortunately, for most modern FTIR instruments, an emission
port for accepting an external light signal has been incorporated, through
which the spectral features of materials and devices from external emission
sources can be characterized. Based on this arrangement, acquisition
configurations for different approaches have been developed,4 22 whereas
commercial accessories with versatility to apply to PL, EL, or LS are still rare.
In performing spectroscopy in the MIR, two difficulties are often
encountered: a weak signal from the sample and a relatively strong
background from the scene, especially uncontrollable room-temperature
(RT) blackbody radiation from surrounding objects. For example, measuring
PL in the MIR is difficult, especially at RT. PL is a strong nondestructive tool
for both scientific research and daily production examination of semiconduc-
tor materials. The combination of PL with FTIR increases the functionality
and is therefore becoming popular for use in the near-infrared (NIR)
extending to the MIR bands. Acquiring relatively strong PL signals from a
background using a conventional RS mode of FTIR is feasible.4,5 The RS
mode is also effective for device characterization, such as for the narrow-band
emission spectra of quantum well lasers in the NIR6,7 and for quantum
cascade lasers (QCLs) with sufficient lasing power in the MIR,8 regardless of
Emission Spectroscopy in the Mid-infrared using FTIR Spectrometry 675
use have been reviewed.21 Approaching the MIR, the thermal background
signals rise. For a 300-K RT background, the radiation peaks at 1035 cm 1,
with a magnitude greater than 0.1%, 1%, or 10% of the peak at 3831,
3144, and 2375 cm 1, respectively, from the short-wavelength side. The
long-wavelength tail extends to the FIR. It is not unusual for the emission
signal intensity to be low or much weaker than the thermal background,
especially at the long-wavelength side. As such, for FTIR emission
spectroscopy in the MIR (4000–400 cm 1) band, the RS mode should be
improved.
The DM mode adopting a phase-sensitive detection scheme is developed
with the addition of an RS mode using a modulated excitation laser, where the
modulation frequency addition to the Fourier frequency forms double
modulation. Through the introduction of a lock-in technique, the modulated
emission signal can be extracted, and the unmodulated background can be
effectively suppressed. In this approach, the modulation frequency for the
device or for the PL-excitation laser should be much higher than the Fourier
frequency (at least >five times higher, as a rule of thumb) in the spectrometer.
In this case, through a lock-in amplifier locked to the modulation frequency
with a suitable time constant, the amplified Fourier signal going back to the
spectrometer can be utilized.
A mechanical modulator (chopper) is conceivable for modulating the PL-
excitation laser, whose maximum frequency is limited to only a few kilohertz
(4 kHz in common). For example, at 4-kHz modulation, the Fourier fre-
quency should be below 800 Hz. In this case, for a wavenumber of 10,000 cm 1,
the moving mirror speed is expected to be below 0.04 cm/s. At this rather low
scan speed, the phase locking of the lock-in amplifier becomes very difficult, as
the spectrometer response is too slow for convenient alignment of the FTIR-
PL setup; additionally, the data acquisition time will be quite long. For
example, at a scan speed of 0.0475 cm/s, about 6 min was needed to acquire a
spectrum span of 4000–600 cm 1. At 16-cm 1 resolution with 32 scans in our
spectrometer, in a presetting mode, the data refresh period was more than 10 s.
For an electro-optical modulator, the modulation frequency is not a problem
(direct current to >1 MHz), whereas the higher driving voltage (>200 V),
small aperture, and relatively large size are incompatible with FTIR-PL
application. A similar case exists for acousto-optic modulators. Additionally,
the acousto-optic modulator is operated at a fixed frequency or around a
center frequency. Although its driving voltage could be lower than that of an
electro-optical modulator, this frequency is normally too high (tens of
megahertz) for lock-in amplifiers and photodetectors used in FTIR. Also,
the prices of the additional modulator and driver are quite high.
Based on the above considerations, the optimal modulation frequency for
the excitation laser of PL in DM mode should be tens of kilohertz, which is
sufficiently high to be far from the Fourier frequency of a few kilohertz for
Emission Spectroscopy in the Mid-infrared using FTIR Spectrometry 677
FTIR at a normal scan speed (e.g., 0.4 cm/s, at which speed the system
response time is adequate), and still low enough for common photodetectors/
lock-in amplifiers. Thus, an effective, maneuverable, and inexpensive laser
with a convenient electrical modulation scheme is expected. In this case, the
instrument settings could be similar to, and the data acquiring time as fast as,
those of the conventional RS mode.
Another approach to eliminate the background uses the SS mode. In this
mode, the excitation laser modulation is still needed, whereas, the scan of the
interferometer is not continuous, but step by step; therefore, the restrictions on
the modulation frequency are completely lifted. In an SS mode, a mechanical
chopper could be conveniently used, and the thermal background could be
effectively eliminated.18,19 The SS mode has become a standard configuration
for current FTIR spectrometers, or is one of the optional modes. The
instrument software is conveniently switchable between RS mode and SS
mode. However, the SS mode presents a notable problem, especially in
practical operation of the FTIR system. In SS mode, the total scan steps are
directly related to the spectral resolution, and the time spent on each step
cannot be very short, mainly due to the mechanical limitations of the
instruments. Therefore, at the same spectral span, resolution, and scan times,
the data acquisition time for the SS mode should be hundreds of times that of
the RS mode. In this extraordinarily long acquisition time, the stability of the
sample becomes compromised, the reliability of the system and random
interference from the surroundings become uncontrollable, and the patience
of the operator is tested. Furthermore, in SS mode, the “instant” response
feature of the FTIR is entirely lost, as the operator cannot instantly see the
effects of operation (such as optical path alignment) but needs to wait a few
minutes for the results. In practice, without the assistance of other modes to
acquire a weak traceable emission signal, the operation of the SS mode alone
becomes extremely difficult, especially for a new operator with an unknown
sample at hand. These nontheoretical but practical demerits remarkably
restrict the adoption of SS mode in FTIR-PL.
Figure 26.1 Schematic diagram of the FTIR emission spectroscopy setup for PL, EL, and
LS. (OAPM is off-axis parabolic mirror; FM, MM, BS, PD, and PC are fixed mirror, moving
mirror, beamsplitter, photodetector, and computer, respectively, for the FTIR spectrometer;
EP is its emission port, and RM and FL are the reflecting mirror and focus lens, respectively,
for the emission spectroscopy setup.) (Reprinted from Ref. 20.)
input port with a 0- to 5-V scale has become a common option, and the
modulation frequency can be from direct current to tens of kilohertz, fitting
the demands of the DM mode. DPSS lasers have many wavelength options,
with different power efficiencies, etc. For numerous wavelengths, the 808-nm
diode-laser-pumped 1064-nm DPSS laser and its frequency-doubled 532-nm
DPSS laser are preferred for their high power efficiencies and, therefore, lower
price of unit power. The pump power efficiency can exceed 50% for 1064 nm,
and attains 20–35% for 532-nm DPSS lasers. In general, the spectral feature
and beam quality of low-end DPSS laser products are already good enough
for PL excitation. The visible 532-nm DPSS laser could be a direct substitute
for the 514.5-nm argon laser frequently used in earlier PL systems. For
reaching deeper excitation depths and higher excitation power, the infrared
1064-nm DPSS laser should also be a good choice. Although the 1064-nm
DPSS laser is invisible to the human eye, it can be clearly seen by silicon CCD
or CMOS image sensors such as those in handsets, so the adjustment of the
system is still quite easy. Certainly, the diode laser itself could also be used
more directly for PL excitation, but its beam quality remains a problem, as
further shaping of the beam is troublesome.
Our system was based on a Nicolet Magna-IR System 860 FTIR
spectrometer, which is quite an old spectrometer model but has a
Emission Spectroscopy in the Mid-infrared using FTIR Spectrometry 679
Table 26.1 Details of the PL samples as well as their FTIR spectrometer settings. (MBE –
molecular beam epitaxy, LPE – liquid phase epitaxy. R – resolution.)
Figure 26.2 Measured PL signals of samples a, b, and c and background signals in both
RS (left) and DM (right) modes (adapted from Ref. 20).
All of the PL signals were still recorded at 16-cm 1 resolution with 32 scans at
a speed of 0.3165 cm/s, with an acquisition time of 25 s for each signal.
The 532-nm DPSS excitation laser was modulated at about 32 kHz, with an
average power of about 0.3 W in DM or SS modes. In this wavelength
span, the Fourier frequency was limited to about 2.5 kHz, and the time
constant of the lock-in amplifier was still set to 80 ms. In this case, the second
modulation frequency is >10 that of the Fourier frequency. The sensitivity
of the lock-in amplifier was set to a high but unsaturated value of 50 mV.
Sample d was an LPE-grown HgCdTe bulk layer in the midwave infrared
band of 3–5 mm, and sample e was an MBE-grown InSb pn structure with PL
at its bandgap wavelength of about 7.3 mm. A resolution of 16 cm 1 was just
enough for most of the PL samples without fine spectral structure, especially
at RT. For a higher resolution of 8 cm 1, the acquisition time correspondingly
increased to 38 s.
Normally, with the increase of wavelength in the long-wavelength side of
the MIR, the PL intensity dramatically decreases, and observation of RT PL
at longer wavelengths becomes more difficult. From the left side of Fig. 26.3,
it can be seen that, in RS mode with MCT-A detector in the MIR band,
the entire RT thermal background appears with a peak around about
1000 cm 1. For samples d and e at RT, no noticeable PL signal could be
found, excepting for the strong thermal background. The recorded
background and “PL” curves were identical. Zooming in or subtracting the
background from the signal still resulted in no trace of the PL signal, implying
that the PL of those samples was much weaker than the RT background.
By adopting the DM mode, the PL signals were clearly extracted, and the
684 Chapter 26
Table 26.2 Comparison of the data acquisition time for each mode under certain
conditions.
In the right side of Fig. 26.3, the PL of sample e in SS mode is also shown;
this data was also acquired at a resolution of 16 cm 1 but only in one scan.
The DPSS laser was still modulated at the same frequency. From this trace, it
can be seen that, even using only one scan, the SNR in SS mode is still better
than that in the DM mode. Adopting this approach, in the SS mode, the
excitation laser modulation frequency is more appropriate than the DM mode
for the lock-in amplifier to reach lower noise. Also, the Fourier frequency is
“zero,” so there is no possible cross interference with the excitation laser
modulation frequency; therefore, a better SNR was achieved, even at only one
scan. However, the data acquisition time for one scan in SS mode at the same
spectral span was 7.4 min, a value that is still acceptable but much longer than
those of 32 scans in RS or DM modes. For 32 scans in SS mode, data
acquisition time was increased to an unpractical value of 3.8 h. Table 26.2
compares the data acquisition time for etch modes under certain conditions on
this spectrometer.
It should be noted that the time consumption in the SS mode was an
obvious practical demerit. For FTIR PL measurements at this lower
resolution, at least several minutes will be needed in SS mode. This longer
time period is not only a challenge to the stability of the system, including
laser and spectrometer, but also a perilous factor for data acquisition failure
in catching a random strong electromagnetic interference from the power
line and so on, despite the patience of the operator. Depending on the
circumstances, the SS mode could be an effective spare mode for FTIR PL in
this approach. That is, when a weak PL signal has been caught in DM mode
but with poor SNR, switching the FTIR spectrometer directly to the SS mode
with a few scans could be attempted in order to acquire better data. In each of
the RS, DM, and SS modes, the data acquisition time is directly related to the
spectral span and resolution. For device characterization using an emission
port, such as the lasing spectrum of semiconductor laser, the highest
resolution (0.125 cm 1 in this spectrometer) with a limited spectral span is
commonly recommended for differentiating the lasing modes and line shape in
a narrower-wavelength range. At this highest resolution, the repeatability of
different scans and stability of the device under test cannot be guaranteed;
686 Chapter 26
Figure 26.5 Measured lasing spectra of a low-temperature QCL around 7.9 mm in RS (left)
and DM (right) modes. (Adapted from Ref. 8.)
of 4 cm 1, with one scan, as also shown in the right side. In this case, the SNR
was good, but the acquisition time increased dramatically to 1268 s at the same
wavelength span. Only an envelope of the spectral feature could be acquired.
Notice that the highest resolution in SS mode was limited to a lower value of
0.5 cm 1 in this instrument for the unacceptably long acquisition time (>2.5 h at
0.5 cm 1 resolution, 100-cm 1 wavelength span with one scan). The sample
chirp and random effect during this long acquiring time was uncontrollable;
therefore, the SS mode was unsuitable for acquiring such fine spectral features.
The DM mode was also recommended for LS, especially for the device
with a lasing wavelength >5 mm (<2000 cm 1) and with low output power.
Figure 26.5 shows the LS of a Fabry–Pérot QCL at a wavelength of about
7.9 mm measured in both RS and DM modes. This device was lasing at a very
low temperature of 20 K with quite weak intensity. In the measurement, the
device was driven using a square pulse of 50 ns at a repeat frequency of
100 kHz, and the resolution was still at 0.125 cm 1 with one scan. From the
left side it can be seen that in RS mode the lasing intensity is still comparable
to the thermal background, in this case, even if the signal can be observed but
is quite difficult to catch in the measurement. By switching to the DM mode,
as shown in the right side, the background was effectively eliminated; the
signal was amplified by more than three orders of magnitude. In this case,
catching the lasing signal became considerably easier.
Figure 26.6 shows the measurement results of another LS sample.17 The
sample was a DFB-QCL lasing around 7.8 mm that was driven using square
pulses at a repeat frequency of 10 kHz. All of the spectra were acquired at DM
mode, and the resolution was still at 0.125 cm 1 with one scan. The left side
shows that the lasing spectra of the device was driven at a relatively short
688 Chapter 26
Figure 26.6 Measured lasing spectra of a RT QCL around 7.8 mm in DM mode at various
temperatures (left) and pulse durations (right). (Reprinted from Ref. 17.)
pulse duration of 100 ns with different currents. In this short pulse duration,
the self-heating effect was inconspicuous, and a apparent lasing line width of
about 0.24 cm 1 could be estimated. The heat sink temperature of the laser
was changed from 0 8C to 50 8C, and the temperature coefficient of the lasing
peak is about –0.089 cm 1/K (0.54 nm/K).
The right side of Fig. 26.6 shows that the laser spectra of the device were
driven at different pulse durations, from 50 to 600 ns, while the driving current
and heat sink temperature were fixed. It can be clearly seen that, with the
increase in pulse duration, the apparent lasing spectra broadened dramatically
toward the red side, revealing the self-heating and wavelength scan effects of
the driving pulse. An average wavelength scan rate of about 0.012 nm/ns
(–0.002 cm 1/ns) could be estimated from those measurements. All of the fine
features were valuable for adapting the device to tunable-diode-laser
absorption spectroscopy applications.
It should be noticed that, if the output power of the laser is high enough,
then the RS mode is sufficient for LS measurement, so the DM mode becomes
unnecessary. In certain cases, if the output power of the laser is too high, it
should be attenuated to prevent saturation of the photodetector in the FTIR
spectrometer. Nevertheless, the SS mode is almost futile in LS measurements
for its limited resolusion and extremely long acquision time.
approach developed for emission spectroscopy of PL, EL, and LS from the
NIR to MIR has been introduced, in which diode-pumped solid state
excitation lasers were adopted for PL excitation. In this approach, three
different FTIR modes of rapid scan, double modulation, and step scan were
software switchable without changing of the hardware or connections. The
advantages and limitations of each mode were analyzed in detail. Using this
approach, a group of III-V and II-VI PL samples from the NIR extending to
the MIR with PL intensities in a wider range were characterized at RT to
demonstrate the validity and overall performances of the system. The EL and
LS of different quantum cascade samples in the MIR band were also surveyed
at different modes and resolutions. Results show that for samples with
relatively strong PL, EL, or LS signals outside the background signals, the RS
mode is the most preferable. For weaker PL, EL, or LS signals overlapped
with the background signals, double modulation is the most effective mode.
To obtain a better SNR when weaker PL or EL signals have been observed in
the DM mode, switching to the SS mode should be an advisable option,
despite the long data acquisition time and limited resolution.
Through this work, a simple, convenient, and universal emission accessory
attached to a commercial FTIR spectrometer has been developed; its
feasibility and performance have been adequately validated, although it has
not progressed past the laboratory. Accordingly, installing this accessory into
a hermetic box connected to the emission port of the FTIR spectrometer in an
air-tight manner to prevent the influence of water and CO2 absorption
features through nitrogen purge will be necessary. Furthermore, a commercial
integrated FTIR system including PL, EL, LS and conventional absorption/
reflection functions for semiconductor material and device specifically are
expected imminently, although this may still depend on volume demands.
Acknowledgments
I am very grateful to my colleagues and students for their cooperation and
encouragement. This work is supported in part by the National Basic
Research Program of China under grant No. 2012CB619200 and the National
Natural Science Foundation of China under grant No. 61275113.
References
1. P. Fellgett, “The Multiplex Advantage,” Ph.D. thesis, Cambridge
University, Cambridge, UK (1951).
2. P. Jacquinot, “New developments in interference spectroscopy,” Rep.
Prog. Phys. 23, 267–312 (1960).
690 Chapter 26
Xiaojuan Cui
Laboratory of Physical Chemistry of the Atmosphere, University of the Littoral Opal
Coast, Dunkerque, France and Anhui Institute of Optics & Fine Mechanics,
Chinese Academy of Sciences, Hefei, China
Tao Wu
Laboratory of Physical Chemistry of the Atmosphere, University of the Littoral Opal
Coast, Dunkerque, France and Key Laboratory of Nondestructive Testing,
Nanchang Hangkong University, Nanchang, China
Eric Fertein
Laboratory of Physical Chemistry of the Atmosphere, University of the Littoral Opal
Coast, Dunkerque, France
693
694 Chapter 27
Yingjian Wang, Weijun Zhang, Xiaoming Gao, Wenqing Liu, and Fengzhong Dong
Anhui Institute of Optics & Fine Mechanics, Chinese Academy of Sciences,
Hefei, China
27.1 Introduction
27.2 State of the Art Instruments for Measurement of Atmospheric HONO
27.2.1 Wet chemical analytical methods
27.2.2 Gas phase spectroscopic analytical methods
27.2.2.1 Long path length absorption based direct spectroscopic detection (DOAS,
TLAS, IBBCEAS, CRDS, FTIR)
27.2.2.2 Indirect spectroscopic detection (PF LIF, CIMS, TDC)
27.3 HONO Sample Production and Quantification
27.3.1 Production of HONO samples in the laboratory
27.3.2 Quantification of HONO concentration
27.3.3 HONO losses on absorption cell wall
27.4 Photonic Monitoring using Infrared Laser
27.4.1 Environmental HONO monitoring by multipass cell based long path absorption
spectroscopy using an 8 mm QCL
27.4.2 HONO monitoring near 2.8 mm
27.5 Photonic Monitoring using LED based IBBCCEAS
27.5.1 Concentration retrieval of multiple absorbers from a structured broadband
absorption spectrum
27.5.2 Determination of cavity mirror reflectivity
27.5.3 Allan variance
27.5.4 Instrumental development and application
27.5.4.1 Open cavity configuration
27.5.4.2 Closed cavity configuration
27.6 Summary and Outlook
Acknowledgments
References
27.1 Introduction
Most pollutants emitted into the atmosphere by natural sources or human
activities are removed by atmospheric oxidants (OH and NO3 radicals, O3, Cl
atom, etc.) via chemical oxidation processes. For instance, the hydroxyl free
radical (OH), as a primary “cleansing agent,” removes 85% CO, 90% CH4,
30% SO2, 50% NO2, and most volatile organic compounds (VOCs) from the
atmosphere.1 It is well known that, besides ozone (O3) and formaldehyde
(CH2O), nitrous acid (HONO in gas phase and HNO2 in liquid phase) acts as
an important source of OH radical.2 4 HONO is usually considered to be an
important OH source by its photolysis in the early morning.5,6 Recent
measurements revealed, however, a strongly enhanced formation of HONO
during daytime via unknown mechanisms, the photolysis of which accounts
for up to 60% of the integrated hydroxyl radical source strengths,4,7 10
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 695
made using an ultraviolet (UV) detector at 290 nm.60 The detection limit is
less than 5 pptv for a sample collected in 5 min.
The wet denuder48,61 63 is an effective and relatively inexpensive wet-
chemical sampling approach that allows separating particles from the trace
gases to be analyzed. The detection limit can be as low as a few pptv with a
time resolution ranging from a few minutes to several hours. A variety of
denuders have been developed and coupled to different chemical analytical
techniques for HONO quantification, such as:
• Air-dragged aqua-membrane denuder (ADAMD) combined with FL
detection,64
• Wet effluent diffusion denuder (WEDD) coupled to flow-injection CL
detection65 or coupled to IC.47,62,66 68, and
• Rotated wet annular denuder (RWAD) coupled with IC.47,63
The chemical instruments are typically inexpensive, easy to use, and very
sensitive, with detection limits as low as a few pptv with a time resolution in
the range of minutes to hours. However, since HONO is sampled on humid/
aqueous surfaces and can be likely formed heterogeneously on the surfaces,
these wet-chemical-based instruments often suffer from unquantified
chemical interferences and sampling artifacts.48,69 Significant differences
were observed in previous intercomparisons involving different measure-
ment techniques.68,70,71 Validation and calibration of chemical instruments
against self-calibrated spectroscopic instruments free of sampling artifacts
are of paramount importance.
kilometers) is the key issue to reach the required sensitivity for successful
tropospheric sensing.
One well-established technique is based on differential optical absorption
spectroscopy (DOAS), involving long-path (some tens of meters to kilometers)
light absorption in the visible and near-UV wavelength range. Absorption by
molecules along the open path through the atmosphere is recorded with a
grating spectrometer and multivariate-fitted to reference spectra of the
corresponding absorbers to identify the molecular species and retrieve their
path-integrated concentrations.6,70,72 74 By use of a relatively wide spectral
source, DOAS allows simultaneous quantification of multiple absorbing species
(typically 2–10) having structured broadband absorption features. The typical
detection limit for HONO is from 30 to 200 pptv, depending on the optical
length of DOAS, with an average time resolution of about 2 min. So far, it is an
optical instrument well adapted for gradient measurement of HONO
concentration.38,42,43,75 Though the DOAS technique is selective, free of
sampling induced artifacts and chemical interferences,72 with a good detection
limit,73,76 the small-scale spatial distribution of HONO in the atmosphere
cannot be well resolved with such a long optical absorption configuration.
For high-spatial-resolution in situ measurements, long-path-length absorp-
tion can be achieved by means of multiple reflections of the probing light in a
multipass optical cell (Ref. 77 and references therein), providing effective optical
path length from 18 to 200 m, or in a high-finesse optical cavity formed with 2 or
3 high-reflectivity dielectric mirrors, on which the light may bounce back and
forth up to one thousand times, leading to 1–10-km-long effective absorption
path lengths. The spectroscopic techniques used for in situ concentration
measurements are discussed next.
Tunable laser absorption spectroscopy (TLAS) in the infrared was first
applied to the measurement of ambient HONO in 2001 using the absorption of
an R-branch line of the n3 band near 1280.4 cm 1 in a 126-m cell. A detection
limit of 300 pptv [1 s, for signal-to-noise ratio (SNR) ¼ 1] was obtained using
30-min-time-averaged data.78 Though the temporal resolution was low, this
technique demonstrated higher sensitivity than FTIR (Fourier-transform
infrared) spectroscopy and greater spatial resolution than DOAS. Li et al.79
reported an improved 1 s detection limit of <200 pptv for a 1-s integration time
by using a lead-salt tunable diode laser operating around 1713.5 cm 1
combined with a 153-m-long absorption cell. Recently, Lee et al.80 developed a
two-channel quantum cascade laser (QCL)-based spectrometer for simulta-
neous measurement of HONO (1660 cm 1) and NO2 (1604 cm 1) absorptions
in a 210-m-long multipass cell at a reduced pressure. The resulting 1-s 1 s
detection limits were 100 and 10 ppt for HONO and NO2, respectively.
Incoherent broadband cavity-enhanced absorption spectroscopy (IBB-
CEAS), introduced by Ruth and coworkers in 2003,81 is an approach
employing a cavity-enhanced absorption scheme associated with a broadband
700 Chapter 27
Table 27.1 Comparative overview of the main techniques available for routine atmospheric
measurements of HONO (with typical concentration ranging from 30 pptv to 15 ppbv).
Techniques’ acronyms are (re)defined below the table.
(continued)
702 Chapter 27
(continued)
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 703
Figure 27.1 (a) Apparatus for HONO production via reaction Eq. (27.1) by slowly dripping
dilute NaNO2 solution into diluted H2SO4 solution. (b) Denuder in combination with a NOx
analyzer used in the present work for absolute HONO concentration determination.
(c) HONO concentration deduced from the difference in the CL signals measured by the NOx
analyzer without and with the carbonate denuder in the sampling lines.
second neck to flush the gaseous products (containing gaseous HONO) from
the solution surface to a dark bag through the third neck of the flask at a rate
of 0.16 l/min. The HONO synthesis was carried out in the dark to avoid
photolysis of HONO. Although this synthesis produced not only HONO
[Eq. (27.1)], but also NO, NO2, and H2O by self-reactions [Eq. (27.2)], these
gaseous byproducts do not affect the HONO measurements by spectroscopy
at 8 mm since there is no spectral interference of such components with
HONO in the relevant laser wavelength range. By controlling the solution
concentration of NaNO2 and H2SO4, some tens of ppmv (parts per million by
volume) of HONO can be produced.
Figure 27.2 (a) HONO absorption spectra at different concentrations (100, 70, 40 ppmv),
sampled after the alkalic denuder, showing that HONO samples with concentration less than
40 ppmv might be totally trapped by the denuder. (b) Plot of HONO absorption intensity
versus time that allows deduction of a lifetime of about 318 s in a closed absorption cell
made of borosilicate glass heated at about 303 K. [(b) reprinted from Ref. 127.]
708 Chapter 27
Figure 27.3 (a) QCL-based laser instrument setup. (b) Top panel: direct absorption
spectrum of the n3 band trans-HONO in the range of 1254.8–1255.1 cm1 (at a scan rate of
2.5 kHz recorded at 18.68 mbar in a 125-m-long absorption cell; spectrum is an average of
210 spectra). Bottom panel: fit residual. [(a) reprinted from Ref. 45, and (b) reprinted from
Ref. 127.]
Figure 27.4 Monitoring of HONO diurnal variation (blue lines) associated with the
corresponding solar radiation curve (black lines). (a) A decrease in HONO concentration due
to photolysis under solar radiation on a no-snow day. (b) Photochemically enhanced HONO
production by solar radiation on a snow day. (See color plate section.)
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 711
Figure 27.5 (a) A typical periodically poled LiNbO3 (PPLN) DFG architecture100 operating
in the mid-IR of 2.5–5 mm (Ge – germanium filter, PC – polarization controller, WDM –
wavelength-division multiplexer, AFS – achromatic focusing system.105 (b) A portion of the
DFG spectrum of 625-ppmv HONO at 2.78 mm, recorded in a 106-m absorption cell at 10
mbar by means of a PPLN-based DFG laser spectrometer.104 [(a) reprinted from Ref. 100,
and (b) reprinted from Ref. 128.] (See color plate section.)
712 Chapter 27
Figure 27.6 (a) Strong and structured broadband molecular absorptions in the UV and
visible spectral regions, arising from the fundamental electronic transition, allow for high-
sensitivity detection of key atmospheric species at the ppbv–pptv levels. (b) Characterization
of some LEDs used in our IBBCEAS experiments.89,107,108 (Reprinted from Ref. 87.)
(See color plate section.)
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 713
ð27:10Þ
The cross section of the oxygen collisional pair reported by Greenblatt
et al.,116 the cross section of NO2 from Ref. 117, and the Rayleigh-
scattering cross section of N2 and O2 from Ref. 118 were used for the mirror
reflectivity determination. The effective path length can be expressed as
follows:
d
Leff ðlÞ ¼ P : ð27:11Þ
1 RðlÞ þ aRay ðlÞd þ aMie ðlÞd þ si Ni d
27.5.3 Allan variance
The measurement accuracy of an instrument system usually depends on
calibration, while the measurement precision can be improved by spectral
averaging. By averaging k spectra, the white noise may be reduced by a factor
of k1/2. However, the maximum averaging number is limited by the stability of
both the LED emission and the charge-coupled device (CCD) spectrometer
system. The optimal averaging time, corresponding to the maximum
averaging number, can be determined by an Allan variance analysis.107,113
For simultaneous detection of HONO and NO2, the minimum absorption
coefficient can be expressed by the following equation:
1 RðlÞ DI ðlÞ
amin ðlÞ ¼ NHONOmin sHONO ðlÞ þ NNO2min sNO2 ðlÞ ¼ ,
d I0 ðlÞ min
ð27:12Þ
where [DI(l)/I0(l)]min is the smallest detectable change DI(l) in the light
intensity I0(l) when an absorber is present. NHONOmin and NNO2min are the
minimum detectable number densities (in molecules/cm3) of HONO and NO2.
According to Eq. (27.12), besides enhancement of the cavity length d and the
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 715
M 2
1 X
s2ANO2 ðtÞ ¼ XNO2jþ1 ðtÞ XNO2j ðtÞ : ð27:14Þ
2M j 1
Plots of these Allan variances as a function of the averaging time allow for
determination of the system stability and hence the optimum averaging time.
Figure 27.7 (a) Schematic diagram of the developed open-path IBBCEAS setup.
(b) Photograph of the infrastructure deployed for open-path environmental testing in ambient
air measurements. An LED emission image is shown in the inset. (Reprinted from Ref. 89.)
provided power for both the TEC and UV LED. The optical cavity was
formed with two high-reflectivity mirrors (Layertec GmbH) separated by a
distance of 1.85 m. The mirrors had a diameter of 25 mm, 2-m radius of
curvature, and a reflectivity of 99.97% between 340 and 370 nm (specified
by the manufacturer). The light emitting from the LED was directly focused
into the cavity with a UV lens ( f ¼ 75 mm). In order to avoid CCD saturation
at the edges of the high-reflectivity range of the cavity mirrors, a UV bandpass
filter (Semrock) was placed between the lens and the cavity to block the light
at undesirable wavelengths. The 8-mm diameter of the light beam injected
into the cavity was controlled with an iris. Light transmitted through the
cavity was collected, by means of a second UV lens ( f ¼ 75 mm), into a
multimode optical fiber (1000 mm in diameter with a numerical aperture of
0.22), and then transferred to a RT CCD spectrometer (Avantes, AvaSpec-
2048-USB2) for spectral dispersion and absorption signal detection. The
spectrograph was equipped with a 25-mm entrance slit, a 600-groove/mm
grating, and a 2048-pixel linear CCD array that allowed covering the entire
220- to 740-nm spectral range with a spectral resolution of 0.77 nm around
360 nm. This spectral resolution, experimentally checked in the present work
using known atomic transition lines, is sufficient for selective recognition of
the structured broadband absorption of HONO and NO2. The whole system,
including LED source, optical cavity, focusing lens, and fiber coupler, was
mounted on a 2.6-m-long optical rail [Fig. 27.7(b)]. A He-Ne laser and a CCD
camera were used to help initial optical cavity alignment.
The cavity mirrors were housed in custom-made mounts that were purged
with pure nitrogen to prevent contamination of the optical surfaces by aerosol
particles. During system calibration, a 185-cm-long polytetrafluoroethylene
(PTFE) mobile tube with an inner diameter of 25 mm was inserted
between the cavity mirror mounts. The tube was removed from the cavity
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 717
Figure 27.8 (a) Top panel: reference absorption cross sections of NO2 (black curve) and
O2-O2 (blue line) used for determination of the cavity mirror reflectivity. Middle panel:
measured (black) and fitted (red) absorption of O2-O2 in 1-atm pure oxygen, and the
absolute mirror reflectivity (blue). Bottom panel: fit residual (calculated as the difference
between the measured and fitted spectra). (b) Simultaneous measurements of HONO and
NO2 concentrations in the environment. Measured (gray) and fit (red) absorption spectra of a
mixture of 3.1 0.3-ppbv HONO and 22.2 0.5-ppbv NO2 in laboratory ambient air. The
absorption coefficients of NO2 (blue)117 and HONO (purple)75 as well as the background
(green) from the fit are included for reference. Lower panel: residual spectrum with a
standard deviation of 5.7 109 cm1. (Reprinted from Ref. 89.) (See color plate section.)
for open-path measurement of HONO and NO2 in ambient air. In the present
work, the mirror reflectivity was determined according to Eqs. (27.9) and
(27.10). Figure 27.8(a) shows the measured and fitted absorption of O2-O2 at
1-atm (atmosphere) pure oxygen. The fit result (middle panel) provided the
absolute mirror reflectivity scale (blue curve). The maximum mirror
reflectivity was found to be 0.99945 around 360 nm. Considering the
uncertainties in the cross section of the oxygen collisional pair (2%), in the O2
pressure within the cavity (3%), in the cavity length (1%), in DI/I (0.5%), and
in the statistical uncertainty from the fit (4%), the total uncertainty in the
factor of (1 R) was approximately estimated to be 6.2%. Rayleigh extinction
amounts of pure N2 or air (aRay 7 10 7 cm 1 at 20 8C and 1 atm) can be
up to 23% of the cavity mirror losses ([1 – R(l)]/d), and its contributions are
certainly non-negligible. By taking into account Rayleigh extinction in
aerosol-free samples and in an urban environment at modest aerosol levels
(i.e., aMie < 5% of the cavity mirror losses), the effective optical absorption
length was estimated to be 2.8 km.
Figure 27.8(b) shows a simultaneous measurement result of HONO
and NO2 concentrations. The atmospheric HONO and NO2 concentra-
tions are retrieved via a multivariate fit to the experimental spectrum a(l):
aðlÞ ¼ NHONO sHONO ðlÞ þ NNO2 sNO2 ðlÞ þ BðlÞ, ð27:15Þ
718 Chapter 27
where NHONO and NNO2 are the number densities (in moleculecm 3) of
HONO and NO2, respectively. sHONO and sNO2 are the cross sections
obtained by convolution of high-resolution cross-section spectra of HONO75
and NO2,117 respectively, with an instrument function of 0.77 nm full-width at
half-maximum (FWHM). B(l) is a polynomial function used to account for
smooth variation in spectral background. The 1s MDC of 430 pptv for
HONO and 1 ppbv for NO2 were experimentally determined for a 90-s
optimum acquisition time (determined by an Allan variance analysis). The
total measurement uncertainty, including statistical uncertainty from the fit
and the systematic uncertainty, was approximately estimated to be 10%. The
HONO detectivity could be improved to several tens of pptv by using higher
reflectivity mirrors (99.99%) and a TEC CCD detector.
Figure 27.9 The LED-based IBBCEAS instrument deployed for a field intercomparison
campaign in Hong Kong (reprinted from Ref. 87).
Photonic Sensing of Environmental Gaseous Nitrous Acid (HONO) 719
Figure 27.10 (a) Plot of the Allan variance s2AHONO for HONO as a function of the signal
integration time, associated with a t–1/2 curve (red curve). The optimum averaging time was found
to be 120 s. (b) Time series measurements of environmental HONO and NO2 concentrations from
14:20 on 5 May 2012 to 14:20 on 6 May. [(b) reprinted from Ref. 87.] (See color plate section.)
Acknowledgments
This work was mainly supported by the IRENI program of the Région Nord-
Pas de Calais and the ANR NexCILAS project. The support of the
Groupement de Recherche International SAMIA between CNRS (France),
RFBR (Russia), and CAS (China) is acknowledged.
References
1. R. G. Prinn, “The cleansing capacity of the atmosphere,” Annu. Rev.
Environ. Resour. 28, 29–57 (2003).
2. B. J. Finlayson-Pitts, and J. N. Pitts, Jr., Chemistry of the Upper and
Lower Atmosphere, Academic Press, New York, 273–276 (2000).
3. J. Stutz, B. Alicke, and A. Neftel, “Nitrous acid formation in the urban
atmosphere: Gradient measurements of NO2 and HONO over grass in
Milan, Italy,” Geophys. Res. 107, 8192 (2002).
4. J. Kleffmann, “Daytime sources of nitrous acid (HONO) in the
atmospheric boundary layer,” Chem. Phys. Chem. 8, 1137–1144 (2007).
5. U. Platt, B. Alicke, R. Dubois, A. Geyer, A. Hofzumahaus, F. Holland,
M. Martinez, D. Mihelcic, T. Klüpfel, B. Lohrmann, W. Pätz, D. Pernez,
722 Chapter 27
28.1 Introduction
28.2 Near Field Scanning Optical Microscopy (NSOM)
28.3 Optical Force
28.4 Deep Subdiffraction Mechanical Frequency and Amplitude Modulation
28.5 Optical Switching via Near Field Interaction
28.6 Conclusions
References
28.1 Introduction
A laser can be thought of as an optical cavity with two mirrors within which
light bounces back and forth many times. Because the optical mode is strongly
coupled to the cavity, a slight change to either mirror can cause a change to
the cavity and to the operation of the laser. Increasing the interaction strength
between changes to the mirror and the cavity can lead to interesting
operational methods for the laser. Recently developed fabrication techniques
have made it possible to integrate metallic nanostructures with lasers, creating
a hybrid device with applications in the medical, communication, commercial,
and defense fields. Using an optical antenna creates an area of very high
electric-field concentration; by placing one of these antennas in close
proximity to the laser cavity, the antenna can perturb this area, influencing
the whole cavity mode. We will show that by modulating the laser mode by
interacting it with the near field of the antenna, one can perform the
following: determine the near-field and optical force profiles of different
739
740 Chapter 28
Figure 28.1 Various examples of optical antennas integrated on the facet of a laser device.
(a) Quantum cascade laser (QCL) showing the scale of the device. (b) Bare facet of the
laser. (c) and (d) Single nanorod antenna. (e) Coupled nanorod antenna (the scale of
the yellow bar is 50 nm). (f) Photonic crystal antenna. (g) Bulls-eye antenna. (h) Bowtie
antenna. (Reprinted from Ref. 52.) (See color plate section.)
Integrated Plasmonic Antennas with Active Optical Devices 741
volume (Vmode). Q factor is a measure of the total energy stored relative to the
amount of energy lost per each cycle of the oscillator. In high-Q-factor
cavities, the optical mode is underdamped and can survive for a long time,
thus increasing the chance of interaction with an electronic resonator placed
within said cavity. This resonator does not necessarily need to be an atomic
one but could be a molecular one, whose vibrational states resonate at the
optical frequency of the cavity. In this case, a high Purcell constant would not
denote an increase in the spontaneous emission rate, but in the absorption rate
of the molecular resonator. Another way of thinking about an increased
Purcell constant is that the time the photon spends interacting with the
molecule within the cavity is increased; therefore, the probability of
interaction by absorption is also increased. This absorption is usually
measured by optical spectroscopy and, by collecting over a broad spectral
range, can be used to distinguish molecules from one another:
3
3 Q l
FP ¼ 2 : ð28:3Þ
4p Vmode n
However, in order to collect a broad spectrum, the cavity must have a
broad spectral resonance, and the Q factor is inversely proportional to the
linewidth. Therefore, one cannot rely on a high Q factor for a high Fp but
rather must find some way of reducing Vmode. The mode volume is usually
diffraction limited and cannot be pushed much lower than one-half its
wavelength in the material within which the mode is propagating.
To achieve subdiffraction-mode volumes, we used plasmonic antennas.
Some structures can be so effective that they can achieve Vmode on the order of
single molecules and, because plasmonic cavities have Q factors somewhere
on the order of 10 to 100, they have high Fp and broadband sensing
capabilities. The volume of the squeezed mode for the coupled nanoparticles is
approximately (l/30)3, which is 6000 times the incident mode volume from a
free-space wavelength for a plasmonic system resonating in the mid-infrared
region. A plasmonic photonic crystal (PPC) with a cavity defect is a
metal–dielectric–metal (MDM) sandwich with a very thin dielectric region
(10–40 nm) and a photonic crystal drilled through the three layers. The central
hole is made very small and acts as a cavity defect, wherein the light can be
squeezed more than 300,000 times the volume of the free-space wave. The
excited gap mode resonates through the middle dielectric layer where, because
of the extremely high refractive index, it propagates very slowly, giving itself
every opportunity to interact with a molecule. Furthermore, broadband cavities
are resistant to manufacturing variances and environmental changes such as
temperature or humidity. More importantly, this broadband capability allows
us to sweep over a range of wavelengths to excite the molecular resonances
around that resonance; and, because not all molecules will have identical
spectral responses, this sweeping will dramatically improve the specificity.
Integrated Plasmonic Antennas with Active Optical Devices 743
It has already been said that a-NSOM often uses an AFM in noncontact
mode to scan the surface of a device. Typically, the way this works is that there is
a small tip at the end of a long cantilever that is vibrating slightly above the
resonant frequency of the cantilever while scanning the surface.23 A laser beam is
reflected off of the back of the cantilever onto a quad detector, and the amplitude
and phase of vibration of the beam, and therefore the cantilever, are measured.
Surface interaction forces such as the electrostatic force and the van der Waals
force make the tip have some interaction with the surface, even though it is not in
contact with the surface.24 Because the tip is vibrating slightly above resonant
frequency, the fundamental forces acting on the tip are out of phase with each
other, lowering the amplitude and keeping the tip in noncontact mode.25 The
amplitude of vibration is fed into a feedback loop, and a piezo and servo keep the
tip out of contact with the surface of the sample as it scans.
The final piece of an a-NSOM setup, and indeed many experimental
setups, is a lock-in amplifier. A lock-in amplifier basically measures the
amplitude of a sinusoidal voltage. However, what makes using a lock-in
amplifier different from using the “measure amplitude” function of an
oscilloscope is that the lock-in amplifier can measure the amplitude of a
sinusoid that is only one small component of a signal that contains a much
broader frequency spectrum. A lock-in amplifier works on the principle of
phase-sensitive detection (PSD), which is basically a multiplier that multiplies
the input signal by a reference signal. At the output of the PSD is a DC
component giving the amplitude of the input signal and higher harmonics.
The output of the PSD is then low-pass filtered to obtain the final output.26
Examples of completed near-field maps taken concurrently with AFM
topography scans can be seen in Fig 28.2.
Figure 28.2 Examples of three antennas fabricated onto a QCL facet with their
corresponding AFM topography scans and near-field maps. All three antennas resonate
at the working wavelength of the QCL, with the bowtie antenna exhibiting the highest
confinement. (Reprinted from Ref. 52.) (See color plate section.)
Integrated Plasmonic Antennas with Active Optical Devices 745
It is important to note that there are other types of NSOM. First, there is
aperture-probe NSOM, in which a tip containing a small aperture is scanned
across the sample. Light is pushed through the aperture and collected with
special lenses or mirrors after interacting with the sample.27
Secondly, there is scattering-NSOM or s-NSOM, in which the scattered
signal is measured as opposed to the reflected signal. Thus, the sample can be
illuminated from the top with an external laser, and the scattered signal from
the AFM tip is measured, again, using mirrors or lenses to collect the
scattered light.28
A special method for measuring antenna-integrated devices through
a-NSOM is a backscatter-type a-NSOM method. In this method the light is
scattered back through the device and passes through a beamsplitter, where
the light is collected into a detector to be measured. Figure 28.3 shows the
setup of this method in the case where the source is a midwave infrared
(MWIR) QCL. Another method of backscattered NSOM uses a nonactive
element (waveguide) such as an optical fiber (see Fig. 28.4). This method
has been used in the shortwave infrared (SWIR) near the telecommunica-
tions band.
Aside from measuring the near-field radiation patterns of such devices, it
is also possible to modify and measure the amplitude and frequency of
antenna-integrated laser devices through special methods. Yet another
possibility that will be discussed is measuring the optical force created by
the strong gradient of the electric field created through near-field plasmonics.
Figure 28.3 Backscatter-type a-NSOM setup developed for imaging SPs in the MWIR
created by optical-antenna-integrated QCLs. (Reprinted from Ref. 16.)
746 Chapter 28
Figure 28.4 Modified backscatter-type a-NSOM setup for detection in the SWIR and for
use with optical fibers (reprinted from Ref. 30).
Lastly, the goal of molecular detection through amplitude and frequency shift
of these devices will be discussed, as many molecules have vibrational
resonance in the MWIR.14,16,29
different molecules; once they are in the trap, the gradient force is strong
enough to keep the molecule there, even if it is off resonance. The same system
used for filtering the molecules can also be used to retrieve the spectroscopic
information of that molecule. By using the optical force created by the
gradient of the electrical field created by such an antenna, the antenna can
pull in the correct molecule to be sensed. Here we introduce a method for
mapping the optical force of a plasmonic nanoantenna.30
Photons have a momentum (h/l), and transfer of this momentum to other
objects due to elastic scattering is the physical origin of radiation pressure and
optical force. These forces have been known to exist since their deduction by
Maxwell in 1871 and have been explored in many applications,31 such as solar
sails for space propulsion32 and, more recently, radiation cooling.33,34
However, the force caused by unfocused light is very small, such that for an
application such as a solar sail a huge area is needed.32 Previous work has
shown that if the light is focused by traditional lens optics, a force large
enough to measure with a probe particle is possible and is called optical
trapping.35 This phenomenon has been well developed by the far-field optics
community.36 38 Developmental work was done with a probe particle held by
an optical trap in 2006 to map the force induced by SP radiation.39 Recently,
there has been a lot of interest in exploiting the near-field region to generate
an optical force to similarly trap particles.40 42 Near-field trapping was
demonstrated by Righini et. al.43 in 2007 and has more recently been used to
trap even smaller particles.44 Here, one uses a nanoantenna to focus the light,
instead of a lens, and the light can be focused past the diffraction limit using
an integrated source. Once the light is focused to this point, a local force in the
piconewton range may be expected, as shown by simulation, so far.40 42,45
Furthermore, one can then use the local trapping forces toward building a
biosensor utilizing SPR.46,47 Similar structures have been used for sensing.48
In fact, in a microfluidic setup, the optical force could be used to enhance the
attraction of small particles, and therefore molecules, to the hotspot beyond
the diffusion rate.43,44,46,47 Also, unlike the usual trapping methods,
plasmonic nanostructures are ideal for “chip-scale” optical trapping. Most
importantly, this allows trapping in volumes that are many orders of
magnitude smaller than what could be achieved with far-field optical
trapping.14,16,48 So far, experimental work done to map optical forces in the
near and far fields has used large probe particles at the expense of high spatial
resolution.35,39 Little experimental work has been done in measuring the
optical force in the near field with a high spatial resolution. Because the
optical force originates from the divergence of the electromagnetic energy
density,49 the fact that plasmonics can be used to concentrate the electric field
to a very small spot can lead to a very large force density.
Based on our simulation and experimental analysis,14,16 we have chosen
an MDM bowtie antenna design for this optical force experiment, due to its
748 Chapter 28
Figure 28.5 (a) SEM image showing a top-down view of the fiber before FIB milling. The
schematic circle shows the core of the fiber. (b) Top-down view of the bowtie antenna
fabricated on the core of the fiber using FIB. The darker area is the milled region where the
laser core material (SiO2) has been exposed. (c) Sideways view of the bowtie antenna
showing the gap between two the arms at 50 nm. (Reprinted from Ref. 30.)
Integrated Plasmonic Antennas with Active Optical Devices 749
antenna axis. We placed six plane monitors around the sphere to obtain the
electric and magnetic field components at the different points of that monitor.
The field components are related to the Maxwell stress tensor42 by
1
Tij ¼ «0 «m Ei Ej þ m0 mm Hi Hj dij ð«0 «m jEj2 þ m0 mm jHj2 Þ, ð28:4Þ
2
where Ei and Hi correspond to the electric and magnetic field, «m represents
the electric permittivity, and mm represents the magnetic permeability for the
object on which optical force is applied (AFM tip in our case). The
electromagnetic force acting on the AFM tip was calculated by
Z
1
hFi i ¼ Re Tij nj dS , ð28:5Þ
2 S
where nj is the outward normal to the surface of the six planes mentioned
before.
Simulations show that the z component of the calculated optical force is
10 times larger than the in-plane (x and y) components, and that this
component would also have the largest effect on the amplitude of the AFM
tip. Thus, for simulated force mapping (Fig. 28.6), we only considered the
vertical component (z) of the optical force exerted on the AFM tip. While
keeping the vertical distance constant at 50 nm (near the average position of
the AFM tip from the antenna top surface25), we simulated the z component
of the optical force at different lateral positions (x and y). It is important to
note that the actual force is attractive, creating a trap that can be used for
biosensing, but we have shown the amplitude of the force. The final simulated
optical force mapping result is shown in Fig. 28.6.
Figure 28.6 Simulated optical force intensity map. The green dashed lines demarcate the
antenna structure, while the white dashed lines demarcate the box carved out by the FIB
milling. (Reprinted from Ref. 30.) (See color plate section.)
750 Chapter 28
Our primary goal is to observe the effect of the optical force on the
amplitude modulation of the AFM tip when it is brought in this region of
optical confinement. We used a simple modification of the a-NSOM setup to
measure the optical force generated by the near field of the bowtie
nanoantenna. Previously, we showed that amplitude and phase require some
time (1 ms) to respond to an external force.25 Thus, we modulated the laser
in 50% duty cycle at low frequency (1 kHz). The operating frequency of the
laser was then used as the reference signal for the lock-in (see Fig. 28.7). We
then fed the output of the lock-in back to the computer to map to the current
position of the AFM, giving the optical force intensity.
The minimum force sensitivity of our AFM system, limited by thermal
fluctuations,50,51 can be calculated from
s
4kB TkB
Fmin ¼ , ð28:6Þ
v0 Q
where kB is the Boltzmann constant, T is room temperature, k is the spring
constant of the cantilever (3 N/m), B is the bandwidth of measurement (set by
lock-in amplifier) used in the force measurement (7.8 Hz), v0 is the resonant
Figure 28.7 Experimental setup for measuring optical force. The antenna is fabricated on
the cleaved tip of an optical fiber that the AFM scans in noncontact mode. The signal from
the quad detector is fed to the signal input port of the lock-in, and the trigger out of the laser
is fed to the reference input port of the lock-in. The laser is operated at 1 kHz at 50% duty
cycle. (Reprinted from Ref. 30.)
Integrated Plasmonic Antennas with Active Optical Devices 751
frequency of the AFM tip (101 kHz), and Q is the quality factor of the
cantilever (160). This leads to a force sensitivity on the order of 62 fN in
our setup. Thus, measuring optical force on the order of a fraction of a
piconewton is not limited by the thermal sensitivity of our system.
Figure 28.8 shows the experimental optical force map. Comparing to the
simulation (Fig. 28.6), both show a central spot on the order of 0.5 pN and
lobes at the ends of the bowtie arms where the force is diminished compared
to the central spot, showing good agreement. The central spot of the
experimental image is likely larger compared to the simulation due to the fact
that in the simulation we considered the AFM tip as a sphere instead of using
its actual pyramidal structure.
To calculate the force from the lock-in signal, we used the following
equation:
cantilever amplitude ðmÞ lock-in signal ðVÞ spring constant ðN=mÞ
F¼ :
max peak-to-peak voltage from quad detector ðVÞ
ð28:7Þ
Note that this conversion is unlike a-NSOM, where it is impossible to convert
the lock-in signal into units of electric field intensity (V/m).
The optical force was only observed in the bowtie whose long axis was
aligned with the incident electric field [Fig. 28.8 (a) top]. In the lower bowtie,
where the polarization was perpendicular to the long axis, no optical force was
observed. The spatial resolution of this technique is limited by two factors:
(1) the radius of curvature of the AFM tip used for the scan and (2) the
bandwidth of the lock-in amplifier used in the experimental setup. For the
radius of curvature of the AFM tip, there is a tradeoff: the smaller the radius,
the greater the resolution but also the smaller the interaction with the optical
force generated by the nanoantenna, explaining why we prefer a tip with a
radius of about 100 nm. There is also a tradeoff with bandwidth of the lock-in:
the greater the bandwidth of the lock-in, the smaller the spot size of the scan,
Figure 28.8 (a) Topography showing two perpendicularly aligned bowties and (b) experimental
optical force map. (Reprinted from Ref. 30.)
752 Chapter 28
but the noise will be higher. Note that the long-term mechanical stability of
the sample determines a lower bound on the scan speed, which determines the
minimum bandwidth that can be chosen.
With dual lock-in amplifiers, this method could simultaneously map the
topography, near-field, and optical force. We believe that this work has many
applications in areas ranging from optical trapping in biosensing46,47,52 to
optical switching in telecommunication.53
In the following sections we discuss special methods for modifying the
amplitude and frequency of an antenna-integrated device through external
means. Recently, there has been increasing activity using SP-based devices for
optical modulation and switching.54 56 The modulation of integrated optical
signals (i.e., diode lasers) is generally achieved by directly modulating the
current that drives the light source. All optical modulation methods have
become more attractive for applications in compact optical circuits and
integrated communication devices. A silicon-based MDM channel with
subwavelength slits that act as an optical source and drain to create an SP-
based optical modulator has been created.54 Other researchers have coupled
SPs into and out of a grating and along their path and have modulated the
signal using a polymer loaded with photochromic molecules.55 Theoretical
work shows that ultrafast and compact all-optical switches can be realized by
means of a semiconductor gap-loaded nanoantenna.56 In the following
section, we show how a laser is significantly modulated by a volume that is
eight orders of magnitude smaller than the laser itself.
The AFM tip is placed over the hotspot of the bowtie nanoantenna. The
measured near field is shown in Fig. 28.9(b). The experimental setup is shown
in Fig. 28.10. A lock-in amplifier allows us to synchronize two frequencies: the
driving frequency of the laser and the frequency of the AFM tip. This
capability allows us to vary the trigger signal delay of the QCL driving pulse
to measure the spectrum at each vertical position of the AFM tip over the
hotspot. The QCL is operated just above the threshold current. The spectrum
was recorded at six equidistant delay points between 0 and 2p with respect to
Figure 28.9 (a) 3D simulation showing the intensity enhancement above the antenna
at resonance condition. (b) 3D NSOM image showing the squeezing optical mode of
the device. Full-width at half-maximum (FWHM) was found to be 100 nm. (Reprinted from
Ref. 80.) (See color plate section.)
Figure 28.10 Setup used to simultaneously measure NSOM, topography, and time-
resolved spectrum of the device. (Reprinted from Ref. 80.)
754 Chapter 28
Figure 28.11 (a) Schematic diagram of the setup used to measure the spectrum of the laser
at different positions of the AFM tip. At 2p/5 phase, mode hopping causes the fit line to split the
two modes in the figure. The white dotted line shows the calculated theoretical model for the
shift in frequency. (b) A shift in laser frequency and amplitude was observed due to feedback
from the apex of the vibrating AFM tip. (Reprinted from Ref. 80.) (See color plate section.)
the AFM tip cycle. The time-resolved step scan for all measurements is shown
in Fig. 28.11. As the delay was varied, the laser’s central wavelength of
operation shifted. In effect, this has created mechanical frequency modulation
in the laser device.
For our previous work, we used a backscattered a-NSOM to study the
near field of the device.22,58 However, for this work, we used a modified a-
NSOM that can also simultaneously measure the spectrum of the device with
respect to time and AFM tip position. During the measurement, the laser was
operated in pulsed mode and biased using a voltage pulse generator and
enclosed in a temperature- and airflow-stabilized environment. To observe the
lasing spectrum at a specific position of the AFM tip, both the voltage pulse
generator and the Fourier transform infrared (FTIR) spectrometer were
synchronized with the AFM tip’s vibration frequency. The setup has been
elaborated in Fig. 28.10. The apex of the vibrating AFM tip backscatters the
near-field light intensity back through the laser, and the backscattered light
intensity is collected using a liquid-nitrogen-cooled mercury–cadmium–
telluride (MCT) detector. The background noise removal and near-field
Integrated Plasmonic Antennas with Active Optical Devices 755
We believe that the reason for the higher g in our device versus the previous
design is the high mode confinement via the plasmonic nanoantenna and the
strong coupling of that mode to the cavity mode of the QCL. When the
plasmonic mode is modified by the AFM tip, it strongly changes the cavity
mode of the laser through this strong coupling. Both devices differ from
previous designs that use the optical gradient force to modulate the
frequency.71 In those cases, the devices can actually be mechanically self-
actuated by the optical gradient force, while the devices discussed here are
externally actuated.
In conclusion, the laser cavity mode has shown an extremely high
sensitivity with respect to the position of the probing AFM tip over the optical
antenna hotspot. The laser power is reduced by 74% and shifts by nearly 30
GHz with a change in the AFM tip position near the hotspot, creating
mechanical frequency and amplitude modulation. We have developed a
theoretical model based on the rate equations for a QCL and dynamic mirror
reflectivity that confirm our experiments. This strong influence relies on a
MDM plasmonic bowtie antenna integrated onto the facet of a QCL that can
squeeze the optical mode to within a spot size of 100 nm, which is 60 times
smaller than the operating wavelength. We used FDTD software to optimize
the design of such a device and measured it using a modified a-NSOM that
can also measure the amplitude and spectrum of the device as a function of
time and AFM tip position. We compared our device to a previous frequency-
modulated QCL57 and calculated a figure of merit that is an order of
magnitude higher, while our design uses a volumetric change per l3 that is five
orders of magnitude smaller. Our device differs from optical-gradient-force-
actuated devices in that our device is externally mechanically actuated, while
those devices are self-actuated through the optical force.71 This sensitivity of
the laser cavity mode to the fine position of a nanometer-scale metallic
absorber opens up the opportunity for modulating large amounts of optical
power by changing the optical properties of a miniscule volume in an
integrated, chip-scale device. Our experiment demonstrates the strong
influence of the hotspot on the laser cavity modes, despite the fact that the
former is many orders of magnitude smaller than the latter. Although this
motion is more than two orders of magnitude smaller than the wavelength of
the laser, the strong coupling changes the laser intensity by more than a factor
of three. Without the antenna, this interaction would have produced a nearly
four-orders-of-magnitude smaller amplitude change according to our simula-
tions, a change of only 0.007%.
This measurement demonstrates the sensitivity of such an antenna-
integrated device to be used for molecular sensing. In particular, if the laser is
operated very near to threshold, the tip has the ability to modulate by this
strong coupling between the two. A molecule that is sitting at or much nearer
758 Chapter 28
to the cavity will act as a strong absorber and vibrate at the driven frequency.
This will modulate the laser the in same manner as the AFM tip.
Figure 28.12 Device schematic (not to scale). (Reprinted from Ref. 72.)
Integrated Plasmonic Antennas with Active Optical Devices 759
Figure 28.13 Simulated backscattered light intensity signal near the antenna when (a) the
1550-nm probe signal is turned off and (b) when the 1550-nm probe signal is turned on.
(Reprinted from Ref. 72.) (See color plate section.)
760 Chapter 28
(80 nm, as in the simulation). After retesting the laser, the final antenna
structure was milled out of the aluminum using a two-step process in FIB:
first, the smaller antenna aligned to the germanium rectangle was milled
using a low current (9.7 pA), then the large antenna was aligned to the
smaller antenna and milled at a high current (93 pA). The final antenna
design is shown in Fig. 28.14.
After fabrication, our sample was experimentally tested in a two-way
microscope setup, schematically pictured in Fig. 28.15. From the top of the
setup, an NIR camera allows one to view the laser facet as well as
simultaneously focus the modulation (1550-nm wavelength) laser light onto
the laser facet. From the bottom of the setup, a broadband inverted
microscope coupling the light from the laser facet to a visible camera and
MCT detector allows one to simultaneously view the bottom facet of the
laser and focus the beam of the QCL onto the MCT detector. We first
observed the signal from the MCT detector on the oscilloscope with the
1550-nm source off and recorded its signal. Subsequently, we turned
the 1550-nm source on and recorded the output of the MCT detector on the
oscilloscope. When the 1550-nm source is on, there is a nearly 15% decrease
in the peak signal of the QCL output [Fig. 28.17(b)]. We have further
characterized the device through a sideband measurement method by
modulating the 1550-nm laser at 200 Hz and recording the output of the
MCT detector in an RF spectrum analyzer.
As the QCL is operating in pulsed mode at 0.5% duty cycle (103 kHz,
50-ns pulse width), we observe a main peak at 103 kHz, and when the 1550
laser is on, two sidebands at 200 Hz with respect to the main peak.
Figure 28.14 SEM image showing the completed device. (Reprinted from Ref. 72.)
Integrated Plasmonic Antennas with Active Optical Devices 761
Figure 28.15 Schematic diagram of the experimental setup. An NIR camera allows one to
view the top QCL facet as well as simultaneously focus the modulation (1550-nm
wavelength) laser light onto the laser facet. A broadband inverted microscope coupling
the light from the bottom QCL laser facet to a visible camera and MCT detector allows one to
simultaneously view the bottom facet of the laser and focus the beam of the QCL onto
the MCT detector. The signal from the MCT detector goes through a bias-tee and low-noise
amplifier (LNA) before being recorded on an RF spectrum analyzer. The experiment
allows for polarization control, wavelength tunability, and power measurement. (Reprinted
from Ref. 72.)
Figure 28.16 (a) Sideband power versus power of the 1550-nm modulation signal. Left
axis shows experimental signal power, and right axis shows simulation data. Experimental
signal power is limited by duty cycle and total power, and is reduced by the responsivity of
the MCT detector. (b) Sideband power versus polarization of 1550-nm modulation signal.
Left axis shows experimental signal power, and right axis shows simulation data. (Reprinted
from Ref. 72.)
Figure 28.17 (a) Sideband power versus wavelength of modulation signal. Data is
centered and averaged over several measurements. Left axis shows experimental signal
power, and right axis shows simulation data. (b) Time response of modulated signal,
showing difference in [Off] minus [On] states. [Off] nominal peak is near 8 mV, so depth of
modulation nears 15%. (Reprinted from Ref. 72.)
Integrated Plasmonic Antennas with Active Optical Devices 763
of the smaller bowtie is the main factor in the frequency response of the
device. The simulation and experimental data agree well, both peaking near
1580 nm.
Figure 28.17(b) shows the difference signal on the oscilloscope when the
switch beam is switched on. A difference in magnitude of 1.25 mV is obtained,
in reference to a “before” signal of 8 mV. This accounts for a depth of
modulation of 15%, with an input switch power of 70 mW/mm2. The length
of the pulse is 50 ns, which is limited by heating in the laser core that is
operated at 103 kHz for a 0.5% duty cycle. In Figs. 28.16(a) and (b), the
power of modulation (left axis) is partly limited by the duty cycle and total
power output of the QCL; it is also limited by the responsivity of the MCT
detector used in the setup.
We considered different recombination mechanisms that are proportional
to first, second, and third powers of free carrier density75 (e.g., Shockley–
Read–Hall, radiative, and Auger recombination). In bulk single-crystal
germanium, the linear term is dominant up to carrier densities of N 0.5
1019 cm 3.76 Therefore, we first assumed that the linear term was dominant in
our experiment and calculated the density that fits our experimental results.
The resulting value is about 1 1019 cm 3 at a carrier lifetime of about 15 ns.
Considering the large surface effect in our submicron germanium piece, and
the fact that it is amorphous, we think that the linear term is certainly
dominant. Using this linear term for the carrier lifetime also produces a very
good fit to our experimental data.
We observed a maximum modulation depth of 15% at the maximum
power of 70 mW/mm2. This results in a switching energy of 8.4 pJ, in line with
other all-optical switches.55,56 One way to improve the modulation depth
would be to use a pulsed laser, which would allow for a higher carrier density.
Lastly, by reducing the volume of the germanium to cover only the volume
beneath the smaller bowtie hotspot instead of the entire area below the smaller
bowtie, we could increase the depth of modulation. Doing this would create a
higher carrier density because the carriers would have less volume to diffuse
over and the loss would increase, thus increasing the change in the optical
properties of the larger antenna when switched.
28.6 Conclusions
We have designed, simulated, fabricated, and experimentally character-
ized a QCL with an integrated nonlinear optical antenna that is switchable
via a telecommunications-wavelength laser. This device shows a viable
optical method to modulate the far field of a laser through a near-field
interaction. The effect is not due to absorption in the quantum wells.
Because of the wide-ranging usefulness for terahertz and shortwave
infrared laser modulators, we believe that such a device could be very
764 Chapter 28
References
1. R. D. Grober, R. J. Schoelkopf, and D. E. Prober, “Optical antenna:
Towards a unity efficiency near-field optical probe,” Appl. Phys. Lett. 70,
1354–1356 (1997).
2. J. Kohoutek, D. Dey, R. Gelfand, A. Bonakdar, and H. Mohseni, “An
apertureless near-field scanning optical microscope for imaging surface
plasmons in the mid-wave infrared,” Proc. SPIE 7787, 77870R (2010)
[doi: 10.1117/12.861582].
3. H. Raether, Surface Plasmons on Smooth and Rough Surfaces and on
Gratings, Springer, New York, pp. 118–122 (1988).
4. T. W. Ebbesen, H. J. Lezec, H. F. Ghaemi, T. Thio, and P. A. Wolff,
“Extraordinary optical transmission through sub-wavelength hole
arrays,” Nature 391, 667–669 (1998).
5. A. Lesuffleur, H. Im, N. C. Lindquist, and S. H. Oh, “Periodic nanohole
arrays with shape-enhanced plasmon resonance as real-time biosensors,”
Appl. Phys. Lett. 90(24), 243110 (2007).
6. Y. Liu, J. Bishop, L. Williams, S. Blair, and J. Herron, “Biosensing based
upon molecular confinement in metallic nanocavity arrays,” Nanotechnol.
15, 1368–1374 (2004).
7. C. T. Campbell and G. Kim, “SPR microscopy and its applications to
high-throughput analyses of biomolecular binding events and their
kinetics,” Biomater. 28, 2380–2392 (2007).
Integrated Plasmonic Antennas with Active Optical Devices 765
2006. Mohseni has served on the Advisory Board and as Program Chair and
Co-chair in several major conferences including IEEE Photonics, SPIE Optics
and Photonics, and SPIE Security and Defense. Dr. Mohseni has published
more than 110 peer-reviewed articles in major scientific journals including
Nature, Nano Letters, Small, and ACS Nano. He holds 13 issued US and
International patents on novel optoelectronic devices and nanoprocessing. He
has presented more than 51 invited and keynote talks at different commercial,
government, and educational institutes. He is a Fellow of SPIE, and of the
Optical Society of America, and a Senior Member of IEEE.
Chapter 29
Quantum-Dot Biosensors using
Fluorescence Resonance
Energy Transfer (FRET)
James W. Garland and Dinakar Ramadurai
Episensors, Inc., Bolingbrook, Illinois, USA and Sivananthan Laboratories, Inc.,
Bolingbrook, Illinois, USA
Siva Sivananthan
Sivananthan Laboratories, Inc., Bolingbrook, Illinois, USA and Physics
Department, University of Illinois at Chicago, Chicago, Illinois, USA
29.1 Introduction
29.2 Conjugated QDs
29.3 Fluorescence Resonance Energy Transfer (FRET)
29.4 Biosensor using FRET and Antibody Conjugated QDs: Concept and Bench top Results
Acknowledgments
29.5 EpiSENSE Prototype Biosensor for Rapid Detection of Airborne Biological Pathogens
29.5.1 Sensor design
29.5.2 Testing of the EpiSENSE biosensor
Acknowledgments
29.6 Summary
References
29.1 Introduction1
This chapter primarily describes a biosensor developed by EPIR Technolo-
gies, Inc. (EPIR) that allows multichannel real-time detection of biological
agents (antigens) with exceptional specificity and sensitivity. However, this
biosensor should be considered only as a prototype proof of concept and as an
example of the power of nanobiotechnology. Nanobiotechnology and
bionanotechnology are very new fields (the term nanobiotechnology was first
773
774 Chapter 29
used in the year 2000, and the term bionanotechnology was not introduced
until 2004). As many readers of this book may have had no previous
introduction to the field, this introduction is meant primarily to briefly
introduce the reader to the field, hopefully stirring enough interest to motivate
the reader to investigate it further. Reference 1, from which much of the
material in this introduction was culled, offers an excellent, accessible basic
introduction to the field.
Both nanobiotechnology and bionanotechnology are recent marriages of
biotechnology and the traditional, but also young, field of nanotechnology.
However, they differ substantially in both their approach and their goals.
Nanobiotechnology describes the application of nanotechnological materials
and techniques for the development and improvement of biotechnological
processes and products. On the medical front, it involves the fabrication and
manipulation of substances and devices on a nanoscale for such purposes as
the development of more sensitive and accurate diagnostic tools and methods
and more targeted and effective treatments for diseases, e.g., the use of
nanoscale matrices for the controlled release of drugs and enzymes, for tissue
engineering, and for tissue regeneration. In this chapter we present a
nonmedical application: the use of nanoparticles and biological entities to
develop real-time biosensors with very high specificity and sensitivity for
environmental monitoring. Bionanotechnology in a way is the converse of
nanobiotechnology; it is the use of naturally occurring biological building
blocks such as DNA, RNA, peptides, and biological nanotubes having high
specificity for nanotechnological applications, including electronic and
optoelectronic applications. Bionanotechnology takes advantage of the
remarkable specificity and diversity of biological organisms and structures
on a nanoscopic level and of the well-organized self-assembly of biological
supramolecules.
The diagnostic aspect of biotechnology primarily consists of the detection
and quantification of biological substances using biochemical techniques such
as immunological recognition assays, enzymatic reactions, and DNA- or
RNA-based technology. The use of biological nanoscale assemblies,
nanotechnological structures such as quantum dots (QDs), and nanoscale
fabrication techniques is increasingly improving both the sensitivity and the
specificity of medical diagnosis and biological detection and is rapidly
increasing the specificity and effectiveness of disease treatments. Familiar
examples of diagnostic immunoassays include home pregnancy test kits,
diagnostic kits for the determination of HIV and hepatitis, and kits for the
determination of blood glucose levels. Nanotechnological lab-on-a-chip
devices to be worn on the skin or implanted in acute type-I and chronic
type-II diabetes patients to continuously monitor glucose levels already exist.
These devices could be used to control an insulin delivery system to be worn
by the patient, assuring a correct insulin level at all times. An example of a
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 775
are the use of peptide nanotubes rather than carbon nanotubes to enhance the
sensitivity and specificity in nanobiological medical diagnoses, and the use of
biologically decorated silicone nanowires for the detection of various diseases
and biological and chemical agents. When decorated with high-affinity
antibodies, these nanowires are so sensitive to the binding of a single
biological molecule that they can detect a single virus or other biomolecule.
Figure 29.1 shows some generic tools used in nanobiotechnology and
bionanotechnology.
quantitative reagents. Thus, the QDs used in the biosensor described later in
this chapter do not need to be frequently replaced.
The interaction of QDs with other structures can be controlled by
applying capping materials such as polymers, surfactants, etc.3 5 These
capping materials not only affect the properties of the colloids, but also make
them soluble in various solutions such as water, ethanol, toluene, etc. Wu
et al.6 have demonstrated the advantages of using semiconductor QDs as
fluorescent biotags. Jaiswal et al.7 have recently demonstrated that ZnS-
coated CdSe QDs may be used in long-term multicolor imaging of structures
in living cells. Dahan et al.8 subsequently used colloidal nanocrystals to image
glycine-receptor diffusion in neural processes.
Standard fluorescence microscopes can be easily used in research for the
detection of QD emission. Lamp- or LED-based excitation can be applied
through an excitation filter to excite the QDs without interfering with the
observation of their emission. For single-color applications, a narrow
emission filter can be used to maximize the signal-to-background ratio,
while, for multiple-color applications, a broader emission filter can be used.
For long-term observation and/or data recording, a narrow emission filter and
visible-light detector that generates a current proportional to the input photon
flux can be used to convert the QD emission into an electrical signal that can
be measured, recorded, and transmitted. For multicolor applications, a set of
detectors can be used, each with an emission filter allowing a different narrow
band of light to be detected.
The biosensor described in this chapter uses antibody-conjugated QDs.
The conjugation of QDs with antibodies was pioneered by Goldman et al.,9,10
who developed the conjugation of QDs with antibodies for the detection of
proteins or small molecules in fluoroimmunoassays in 2002 and later extended
it to the detection of explosives.10 The conjugation strategy used by Goldman
et al. was based on electrostatic self-assembly between negatively charged
dihydrolipoic acid (DHLA)-capped CdSe-ZnS core-shell QDs and positively
charged proteins that served to bridge the QD and antibody. However, the
conjugation strategy used for the development and fabrication of the
biosensor described here was substantially different.11 That strategy is
outlined in Section 29.4.
been studied by Bruno et al.15 using organic dyes and quencher molecules.
FRET is a nonradiative transfer of energy that occurs even between
fluorescent chromophores. FRET and radiative energy transfer are the short-
and long-range asymptotes of a single unified mechanism. As the interaction
between chromophores is a dipole–dipole interaction, the FRET quantum
yield E is inversely proportional to the sixth power of the distance r between
them:
1
E¼ , ð29:1Þ
1 þ ðr=R0 Þ6
where R0 is defined as the Förster radius. Because of the strong dependence
of E on r, the most common use of FRET is to measure distances in
biological systems by measuring the dependence of E on distance. For
example, FRET can be used to measure distances between domains in a
single protein and therefore to provide information about protein
conformation, and also can be used to detect interactions between proteins.
Applied in vivo in living cells, it has been used to detect the location of and
the interactions between genes and cellular structures including integrins and
membrane proteins.
Figure 29.2 Schematic of the FRET-based detection system for detection of target
antigens in solution. A signal is detected from QD conjugates when quencher-labeled
biological agents are replaced by free unlabeled agents.
because other components in a given solution are not able to displace the
inactive pathogen. This technology results in reduced background noise in
comparison to other FRET techniques and improves the SNR of the device.
This allows the system to detect smaller quantities of a pathogen in a given
volume of fluid. All other FRET-based detection devices begin in an “on”
state and are turned off by a binding event of interest, resulting in a higher rate
of false positives and a lower signal to noise ratio. As a result, we obtained
detection limits as low as 102–103 colony-forming units (CFUs) per ml and the
ability to measure the number of CFUs/ml up to 105 times the minimum
detectable number. Such a sensor also displays an almost immediate response.
The initial investigations described here employed FRET-based interac-
tions between fluorescent QDs and organic quencher molecules to detect
antigens such as E. coli 0571H7 (1–3 mm in diameter), Bacilluscereus
(B. cereus) (1 mm in diameter), and the MS2 bacteriophage (30 nm in
diameter), which were used as model biological contaminants for large,
midsize, and small antigens, respectively. FRET-based detection can be used
for the simultaneous detection of different biological contaminants such as
bacteria and viruses using QDs with different characteristic emission
wavelengths. By conjugating different antibody fragments specific to each
type of target antigen, a highly sensitive technique can be developed for the
FRET-based detection of multiple antigens. Using QDs conjugated to
antigen-specific antibody fragments, the detection of all three types of
antigens (E. coli 0571H7, B. cereus, and MS2 virus) were demonstrated both
individually and simultaneously, both in aqueous solution and immobilized
(trapped) on the surface of a porous filter.
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 781
The FRET reaction between the QDs and the conjugated quenchers
inhibits the fluorescence of the QDs. The reagent solution contains inactivated
target antigens, represented in Fig. 29.2 as biological-agent (BA) analogs
bound to quencher molecules, and QDs bound to antibodies specific to those
BAs (or BA analogs). Initially, the QD–antibody complexes were conjugated
with the BA-analog–quencher complexes. The proximity of the quenchers to
the QDs then allowed FRET to inhibit their fluorescence. When a test sample
containing unlabeled BAs (without bound quenchers) is added to the assay
solution, equilibrium reactions cause them to displace a fraction of the
quencher-labeled BAs from the QD-labeled antibodies. The QDs no longer
adjacent to a quencher then fluoresce upon excitation by a suitable signal, and
the presence of a target BA can be detected either through PL measurements
in aqueous solution or through fluorescence imaging of conjugates trapped on
the surface of a porous filter. PL is a simple technique for the detection of free
target antigens (BAs), but its usefulness is hampered by a large background
signal in the samples, as initially prepared. However, as was discovered
through subsequent PL and fluorescence measurements, that PL background
arose from QDs that had remained unconjugated. Even the largest
unconjugated QDs were much smaller than any of the conjugated QDs and
therefore could be filtered out. Fluorescence imaging measurements on a
porous filter through which the unconjugated QDs passed freely and on which
only the conjugated QDs were captured gave an almost-zero background
signal. Quantitative results with almost no background signal could be
obtained by passing the original assay solution through a filter to capture only
the conjugated QDs and then either (1) backwashing the filter to create a new
assay solution for antigen detection using PL, or (2) using software to
integrate the signal from the filter used for fluorescence imaging. Elimination
of the background allows orders-of-magnitude improvement in the detector
sensitivity and a more accurate measurement of antigen concentrations with
no danger of false positives.
E. coli 0157:H7 monoclonal antibodies were conjugated to 605-nm QDs,
B. cereus antibodies were conjugated to 565-nm QDs, and MS2-virus
antibodies were conjugated to 525-nm QDs. The method used for the binding
of the antibodies to QDs is schematically illustrated in Fig. 29.3. A detailed
description of the QDs, cross-linking reagents, antibodies, deactivated
antigens, and quenchers used in the assay solution and of the procedures
used for the conjugation of the QDs and antibodies, the attachment of the
quenchers, and the final preparation of the assay solution is given in Ref. 11.
The solution containing these antibody-conjugated QDs served as the reagent
for detecting the antigens corresponding to the antibodies. Initially, the
detection of a concentration of 105 CFU/mL of single types of antigens was
performed through both fluorescence imaging and PL measurements;
subsequently, the simultaneous detection of the same concentration of three
782 Chapter 29
Figure 29.3 Illustration of the binding reaction between QDs and antibodies. Carboxyl-
coated QDs bind to the primary amine of the antibodies to form amide derivatives in the
presence of EDC and sulfo-NHS. (EDC is 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide
hydrochloride, and sulfo-NHS is N-hydroxysuccinimide.)
different types of antigens was demonstrated. Figures 29.4 and 29.5 show the
results from the fluorescence imaging and PL measurement of 105 CFU/mL of
B. cereus; the results for the detection of E. coli 0157:H7 and the MS2 virus
were similar. The signal present before the addition of the B. cereus came from
the unconjugated QDs present in the original unbackwashed reagent. The PL
signal from the simultaneous detection of E. coli 0157:H7, B. cereus and MS2
virus is shown in Fig. 29.6.
For the detection of lower concentrations of antigens and to demonstrate
the determination of the concentration of antigens in a given sample, we
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 783
Figure 29.5 PL from quenched and unquenched B. cereus 565-nm QDs labeled with
BHQ-2 quenchers (reprinted from Ref. 11).
Figure 29.6 PL from quenched and unquenched E. coli 605-nm QDs labeled with BHQ-2
quenchers þ B. cereus 565-nm QDs labeled with BHQ-2 quenchers þ MS2 525-nm QDs
labeled with BHQ-2 quenchers. (Reprinted from Ref. 11.)
performed a simple scaling study using serial dilutions of free antigens added
to the solution containing the complexes of QDs and quencher-labeled
antigens. This was performed at the U.S. Army Corps of Engineers, Engineer
Research and Development Center (ERDC) facility in Champaign, IL. In
order to prove that the conjugation chemistry does not affect the sensitivity of
detection and is dependent only on the characteristic emission of the QDs and
the concentrations used for detection, a set of slightly different QDs from a
different manufacturer was used, and the crosslinkers used for conjugation
784 Chapter 29
Figure 29.7 Fluorescence intensity versus antigen concentration for serial dilutions of
108–102. (Reprinted from Ref. 11.)
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 785
Acknowledgments
The authors gratefully acknowledge the contributions of Prof. Mitra Dutta
and Mr. Jianyong Yang from the University of Illinois at Chicago and
E. Norton, J. Hale, L. D. Stephenson and A. Kumar of the U.S. Army Corp
of Engineers Engineer Research and Development Center’s Construction
Engineering Research Laboratory (CERL) to the work reported in this
section, and gratefully acknowledge the support of the Department of Defense
under STTR contract no. W9132T-06-C-0032.
antigens and its successful testing. A prototype biosensor for the rapid detection
of aerosolized biological pathogens has been developed based on the FRET
immunoassay technology, using QDs for the detection of biological antigens in
fluids outlined in Section 29.4. That sensor has demonstrated great potential for
the detection of bacteria and bacteriophages in an aerosolized cloud under
ambient field test conditions. The system was developed as a biological
pathogen collector and identifier with lightweight capabilities for field
operation. The ability of the system to trap airborne bacteria in a cloud and
identify them with accuracy in a rapid manner has been investigated, and its
detection capacity and speed have been measured by comparison with a particle
counter system. The aerosolized bacteria to be detected are trapped in a fluid
and detected on disposable filter surfaces as in the benchtop studies reported in
Section 29.4. The fluorescent intensity of the trapped QDs is integrated across
the filter and detected by silicon photodetectors. The reported results were
obtained in field tests in the Joint Ambient Breeze Tunnel (JABT) at the U.S.
Army’s Dugway Proving Grounds (DPG), which had the capability of
performing timed releases of bacterial clouds.
Figure 29.8 Schematic illustration of the basic EpiSENSE system design. The system is
based on a three-stage operation: collection, filtering, and detection.
tested for the presence of antigens, as in Section 29.4. As soon as that testing is
completed, the collection cycle is restarted.
When the air-collection pump is turned off, the liquid is transferred
through a filter (1-in diameter, 0.4-mm polycarbonate filter from Millipore
Corp.) to the filtering and backwash stage, and a few drops (1 ml) of
quencher-labeled QDs are injected into the liquid. At this point, a fraction of
the quencher-conjugated inactivated antigens that labeled the QDs are
replaced by a fraction of any free antigens that were trapped in the liquid from
the air. A third Sensidyne miniature pump (860–2600 cc/min) was used to
filter the reagent (to draw any unconjugated QDs through the filter while
trapping the conjugated QDs on the front surface of the filter) and to
backwash the conjugated QDs (by reversing the flow of the pump and forcing
2 ml of fresh buffer from the bottom). This process of filtering and
backwashing is repeated five or six times to completely remove any free QDs
from the solution. The filtrate after each cycle is emptied into a waste
reservoir. At this point in the process, the top section of the filter holder holds
the quenched and unquenched conjugated QDs in a 2-ml buffer solution.
The final stage of the sensor is the detection stage. The solution from the
top of the filter holder is forced through a rectangular narrow aperture flow
through a quartz cuvette (Starna Cells, Inc.) using a model Pk solenoid pinch
valve (Clark Solutions) that is normally closed but can be opened by an
788 Chapter 29
external 12-V DC trigger. The valve is triggered repeatedly, thus pinching and
relaxing a 1.6-mm silicone tubing. This results in a peristaltic action that
drives the QD solution into the cuvette, where the QDs can be imaged. The
QDs are excited using a 1-W 365-nm UV LED from Nichia Corp. and
detected using a PCDMini photon counting detector (SensL Technologies).
The detector converts the incident photon energy into a digital voltage signal
that is measured by a Keithley 2100 digital multimeter. The detector and LED
are placed at right angles to each other, with the detector at the bottom of the
cuvette looking up at the QD solution and the LED placed along the
rectangular side of the cuvette.
Figure 29.9 Schematic of the test setup at the JABT. (ECBC is U.S. Army Edgewood
Chemical Biological Center; ELPI is electrical low-pressure impactor; BTK is Bacillus
thuringiensis kurstaki.)
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 789
This data was then retrieved from the laptop and plotted using standard
Microsoft Excel software. The plots were then compared with the plots
obtained from the ELPI. Figures 29.10 and 29.11 compare test plots from the
EpiSENSE and ELPI systems.
The experiments at DPG showed that the EpiSENSE biosensor is capable
of detecting airborne antigens within a time of less than 5 min; as measured by
the ELPI, the delays in detection after the first exposure to the antigens were
1 min and 57 s for a 1-min exposure to E. herbicola, 2 min and 30 s for a 10-s
Figure 29.10 Comparative results from the (a) ELPI and (b) EpiSENSE systems during the
detection of E. herbicola. The data corresponds to a 1-min release of E. herbicola on
September 23, 2010, starting at 23:28 h. The EpiSENSE system detected the pathogens
1 min and 50 s after the ELPI did.
Figure 29.11 Comparative results from the (a) ELPI and (b) EpiSENSE systems during the
detection of MS2 phage. The data corresponds to a 3-min release of MS2 phage on
September 29, 2010, starting at 21:50 h. The EpiSENSE system detected the pathogens
3 min and 25 s after the ELPI did.
Quantum-Dot Biosensors using Fluorescence Resonance Energy Transfer (FRET) 791
exposure to E. herbicola, and 3 min and 25 s for a 3-min exposure to the MS2
phage. The somewhat longer time for detection of the MS2 phage was
because, as a virus, it has a much smaller size. Other tests have revealed that
the EpiSENSE is capable of detecting airborne bacterial concentrations below
103 organisms per liter. Thus, the EpiSENSE satisfies all five requirements for
a successful biosensor: high sensitivity, high specificity, rapid detection, a low
false-positive rate, and multiplexing ability. The essential components for
meeting all of these requirements are (1) a technology in which the normal
state is an “off” state, (2) the use of FRET, and (3) the use of QDs to allow
multiplexing.
Acknowledgments
The authors gratefully acknowledge invaluable assistance from many others
both in designing and constructing the final EpiSENSE biosensor and in
conducting the tests at the Dugway Proving Grounds. We gratefully
acknowledge the assistance of William Orosz in designing the EpiSENSE
biosensor and Robert Crosby, who played a major role in the design and
construction of the EpiSENSE biosensor. We also gratefully acknowledge the
invaluable assistance of Michael Williamson and Paul De Luca of CERL and
of the entire JABT team in setting up the testing at DPG. We also wish to
thank Robert Doherty of Science Applications International Corp. (SAIC) for
supporting ECBC with effective dialogue and valuable comments and
suggestions regarding aerosol physics.
29.6 Summary
Both nanobiotechnology and bionanotechnology are very rapidly growing
fields of research, and both have already generated technological advances of
great practical importance in medicine, biotechnology, and nanotechnology,
in general. The use of QDs and other nanoparticles is central to many
emerging technological advances in medicine, biological research, electronics
and optoelectronics, as well as many other fields. The use of FRET is
relatively new, even in these new fields, and has not yet found many
applications but, as illustrated by the biosensor applications discussed in this
chapter, has great promise.
References
1. Much of the material in the introduction is taken from E. Gazit, Plenty of
Room at the Bottom: An Introduction to Biotechnology, Imperial College
Press, London (2007).
2. N. C. Seeman, “De novo design of sequences for nucleic acid structural
engineering,” J. Biomol. Struct. Dyn. 8, 573–581 (1990); N. C. Seeman,
792 Chapter 29
30.1 Introduction
30.2 Monodisperse Carbon Nanomaterials
30.3 Assembly Strategies
30.4 Electronics of Semiconducting SWCNT Films
30.5 Optoelectronics with Semiconducting SWCNT Films
30.6 Applications of Metallic SWCNTs
30.7 Applications of Solution Processed Graphene
30.8 Summary and Future Outlook
References
30.1 Introduction
For the past 40 years, silicon-based electronics has followed Moore’s law,
which predicts doubling of transistor density every two years.1 These
sustained advances have provided ever increasing processing capabilities for
both information and consumer electronics. However, as device features
approach the 10-nm-length scale, conventional silicon electronics will face
both technological and fundamental challenges, such as short-channel effects,
parasitic resistance and capacitance, and power management issues.2 5
Similarly, optoelectronic applications have been dominated by III-V
semiconductors for the past several decades, but the complex growth and
fabrication techniques for these materials are not well suited for the rapidly
795
796 Chapter 30
Figure 30.1 Monodisperse carbon nanomaterials are isolated using DGU. (a) DGU sorting
single-walled carbon nanotubes by electronic type (reprinted from Ref. 39). (b) DGU sorting
single-walled and double-walled carbon nanotubes by number of walls and outer-shell
electronic type (reprinted from Ref. 45). (c) DGU sorting graphene by thickness (reprinted
from Ref. 50). (See color plate section.)
resilient mechanical properties, all of which make them attractive for use in
flexible and transparent electronics. Of particular interest are DGU-sorted
SWCNTs since their electronic homogeneity implies improved performance in
thin film electronic applications.
Indeed, DGU-sorted SWCNT-TFTs have shown concurrently high
mobilities, current densities, and on/off ratios. For example, the first reports
of random-network DGU-sorted SWCNT-TFTs39 possessed on/off ratios of
2 104, and improvements in sorting and device fabrication42,53 have since
enhanced this ratio to 105. Recently, 99% semiconducting, single-chirality
(6,5) SWCNT-TFTs were shown to retain this high on/off ratio while also
producing high drive currents and field-effect mobilities (up to 37 cm2/V/s).42
These random networks of sorted SWCNTs have also been combined with
ultrathin, low-leakage, and high-k self-assembled nanodielectrics into TFTs
that show significant improvements in all performance metrics such as a
subthreshold swing of 150 mV per decade and an intrinsic field-effect mobility
of 150 cm2/V/s.53 Beyond random-network TFTs, DGU-sorted SWCNT-
TFTs have also been fabricated on aligned films produced via EDSA.59,60 In
these top-gated devices, which are shown in Fig. 30.3(a), a high on/off ratio
(104) as well as a large on-state conductance (25 mS) was measured, as shown
in Fig. 30.3(b). Finally, 98% semiconducting SWCNT inks were recently
incorporated into printed TFTs that operate at sub-3-V gate voltages.20 These
TFTs were fabricated on both SiO2 and polyimide substrates, allowing for the
production of digital circuits and five-stage ring oscillators.
In addition to transistors, thin films of semiconducting SWCNTs are
attractive for radio frequency (RF) electronic applications since individual
SWCNT devices suffer from high input impedance and low drive current.49
Recently, high-performance, random-network TFTs were fabricated using
99% semiconducting SWCNTs. The surfactant encapsulated nanotubes were
deposited onto the gate area of a prepatterned substrate using a combination
of surface chemistry and dielectrophoresis.24 Following fabrication, the direct-
current (DC) and alternating-current (AC) characteristics were measured.
This characterization showed that these devices possess high on-currents and
transconductances, and extrinsic and intrinsic gain cutoff frequencies of 15
and 80 GHz, respectively. While this study showed that random networks of
SWCNTs could enable GHz frequency operation, it was expected that aligned
films would lead to further performance increases. Consequently, TFTs using
aligned arrays of semiconducting SWCTs were recently produced via low-
frequency dielectrophoresis in the dual-device channel geometry shown in
Fig. 30.3(c).58 These aligned devices not only show high current densities and
transconductance values, but also current saturation (in contrast to graphene
RF devices). Additionally, the extrinsic and intrinsic current gains [shown in
Fig. 30.3(d).] are 7 and 153 GHz, respectively, which indicates that the
alignment of SWCNTs does improve intrinsic RF device performance.
Optoelectronic Applications of Monodisperse Carbon Nanomaterials 803
Figure 30.3 Electronics based on semiconducting SWCNT thin films. (a) Aligned SWCNT-
FET with a top gate (reprinted from Ref. 60). (b) Transfer curve of the aligned SWCNT-FET
(reprinted from Ref. 60). (c) Dual-channel SWCNT transistor for radio-frequency (RF)
applications (reprinted from Ref. 58). (d) Intrinsic short-circuit power gain and the maximum
available power gain in the dual-channel transistor (reprinted from Ref. 58). (e) SWCNT-FET
for hydrogen-sensing applications (adapted from Ref. 18). (f) Sensitivity of the SWCNT-FET
to hydrogen (adapted from Ref. 18). (See color plate section.)
Figure 30.4 EL from SWCNT thin films. (a) PL and EL spectra from a top-gate ambipolar
SWCNT-TFT (reprinted from Ref. 60). (b) Thin film semiconducting SWCNT p–i–n diode.
The length of the scale bar is 2 mm (reprinted from Ref. 74). (See color plate section.)
SWCNT thin films since a p–n junction can be produced via electrostatic
doping. This junction forms when one top-gate is biased with a negative
voltage, inducing holes into the channel, while the other top-gate is biased
with a positive voltage, inducing electrons into the channel.75 Upon the
application of a bias, electrons and holes are driven to radiatively recombine
at the p–n junction, thereby emitting light. Indeed, EL measurements from
these devices show a thresholdless light-emission band in the near-infrared
spectrum. Additionally, the light generation efficiency of these thin film,
split-gate devices74 improves on single-nanotube devices16 by a factor of 15,
thus illustrating the potential of semiconducting SWCNTs for infrared
optoelectronics.
The resulting conductive films are translucent with a visible color since DGU
affords angstrom-level control over the SWCNT diameter, as shown in
Fig. 30.5(a). Additionally, these predominantly metallic films enhance the
conductivity of the transparent conductor in comparison to unsorted samples.
As shown in Fig. 30.5(b), the sheet resistance of a metallic SWCNT film is
much lower than that of an unsorted film, leading to a conductivity
enhancement factor of 5.6 in the visible region.80
With impressive performance as transparent conductors, SWCNT thin
films have recently been incorporated into the transparent anode of organic
photovoltaics (OPVs).61,81 Recent studies have replaced an ITO anode with
an unsorted SWCNT film, but the performance of these OPVs was limited by
the high sheet resistance of the SWCNT film.82 84 Since the contact resistances
between SWCNTs of different electronic types is 100 times higher than those
of the same type, it is expected that electronically monodisperse films will have
a lower sheet resistance and lead to better-performing OPVs. Towards that
end, DGU-sorted SWCNTs have recently been employed to study the effect
of electronic type on OPV anodes. The DGU-sorted SWCNTs were
Figure 30.5 Applications of metallic SWCNT thin films. (a) Translucent, visibly colored,
conductive films from diameter-sorted metallic SWCNTs (reprinted from Ref. 80). (b) The
sheet resistance for a given optical transparency is reduced for metallic SWCNTs compared
to unsorted SWCNTs (adapted from Ref. 80). (c) The power conversion efficiency of organic
photovoltaics increases with increasing metallic content of the SWCNT thin film anode
(adapted from Ref. 54). (d) Phonon-mediated EL spectra from metallic SWCNTs (reprinted
from Ref. 87). (See color plate section.)
Optoelectronic Applications of Monodisperse Carbon Nanomaterials 807
assembled into thin films via a dip-coating technique, which helps produce
smooth films.54 Additionally, the SWCNT films were treated with a nitric acid
solution that densifies the film and further reduces surface roughness. After
fabrication, the performance of the SWCNT films with increasing concentra-
tions of metallic nanotubes (0.1% metallic to 99.9% metallic) was compared,
and it was determined that the 99.9% metallic films outperformed the 99.9%
semiconducting films by 50-fold, as shown in Fig. 30.5(c). This performance
difference was attributed to an increased sheet resistance following dedoping
of the semiconducting films by the hole transport layer in the OPV. Finally,
it was shown that the performance of the metallic SWCNT anodes is
comparable to conventional ITO anodes, thereby validating the potential of
metallic SWCNT films in OPVs.54
As indicated above, much of the optoelectronic work on sorted metallic
films has focused on transparent conductors because light emission is not
expected from metallic SWCNTs due to nonradiative relaxation pathways.85
However, weak EL from biased metallic SWCNT-TFTs was recently
observed.85,86 A proposed mechanism for this luminescence was thermal
population and depopulation of states at van Hove singularities close to the
K point, from which radiative recombination can occur via an interband
transition. For such a model to be valid, the emitted photon energy must scale
inversely with nanotube diameter. Therefore, a study on the light emission
properties of diameter-sorted SWCNTs was conducted.
First, the DGU-sorted metallic nanotubes, which had diameters ranging
from 0.7 to 1.5 nm, were assembled into thin film TFTs using dielectrophor-
esis.87 Following fabrication, the light emission properties of these devices
were studied. As shown in Fig. 30.5(d), two EL peaks appear in the 1.3- to 2.5-
eV range for a medium-diameter metallic SWCNT-TFT, and these two peaks
appear in the EL spectrum regardless of the diameter of the nanotube. Since
these peak positions do not scale with nanotube diameter, K-point interband
transitions cannot be the origin of the light emission. Additionally, the
substrate and electrode materials are not direct or indirect sources of light, as
different material choices did not affect the location of the emission peaks.
Thus, light emission must originate from the nanotubes, and it is likely that
this emission is due to phonon-assisted radiative decay from p* band states at
the M point to the Fermi level at the K point.87 Finally, this observation was
complemented by measurements on multiwalled-nanotube and few-layer
graphene devices, which also show two emission peaks in this range.
References
1. G. E. Moore, “Cramming more components onto integrated circuits,”
Proc. IEEE 86(1), 82–85 (1998).
2. K. K. Likharev, “CMOL: Second life for silicon?,” Microelectron. J.
39(2), 177–183 (2008).
3. K. J. Kuhn, “Considerations for ultimate CMOS scaling,” IEEE Trans.
Electron Devices 59(7), 1813–1828 (2012).
4. A. W. Topol, D. C. La Tulipe, L. Shi, D. J. Frank, K. Bernstein,
S. E. Steen, A. Kumar, G. U. Singco, A. M. Young, K. W. Guarini, and
M. Ieong, “Three-dimensional integrated circuits,” IBM J. Res. Dev. 50,
491–506 (2006).
5. J. J. Welser, S. Kosonocky, T.-J. K. Liu, T. Sakurai, R. Thewes, and
B. Zhao, “Special issue on device technologies and circuit techniques for
power management,” IEEE Trans. Electron Devices 55(1), 4–7 (2008).
6. P. R. Wallace, “The band theory of graphite,” Phys. Rev. 71, 622–634
(1947).
7. K. S. Novoselov, “Electric field effect in atomically thin carbon films,”
Science 306(5696), 666–669 (2004).
8. J. Hass, W. A. de Heer, and E. H. Conrad, “The growth and
morphology of epitaxial multilayer graphene,” J. Phys.: Cond. Matter
20(32), 323202 (2008).
9. X. Li, W. Cai, J. An, S. Kim, J. Nah, D. Yang, R. Piner,
A. Velamakanni, I. Jung, E. Tutuc, S. K. Banerjee, L. Colombo, and
R. S. Ruoff, “Large-area synthesis of high-quality and uniform graphene
films on copper foils,” Science 324(5932), 1312–1314 (2009).
10. K. I. Bolotin, K. J. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone,
P. Kim, and H. L. Stormer, “Ultrahigh electron mobility in suspended
graphene,” Solid State Commun. 146(9–10), 351–355 (2008).
11. S. Morozov, K. Novoselov, M. Katsnelson, F. Schedin, D. Elias,
J. Jaszczak, and A. Geim, “Giant intrinsic carrier mobilities in graphene
and its bilayer,” Phys. Rev. Lett. 100(1), 016602 (2008).
812 Chapter 30
Shadi A. Dayeh
Center for Integrated Nanotechnologies, Los Alamos National Laboratory,
Los Alamos, New Mexico, USA and Department of Electrical and Computer
Engineering, University of California, San Diego, La Jolla, California, USA
S. Tom Picraux
Center for Integrated Nanotechnologies, Los Alamos National Laboratory,
Los Alamos, New Mexico, USA
31.1 Introduction
31.2 Device Fabrication
31.3 Estimation of Depletion Region
31.4 Optical Absorption Simulation
31.4.1 Effect of nanowire length
31.4.2 Effect of pitch size
31.5 Conclusion and Outlook
Acknowledgments
References
823
824 Chapter 31
31.1 Introduction
The quantum efficiency of solar cells, like that of any photon detector, is
dictated by the ability to absorb photons to create conducting carriers, and the
efficiency to drive such carriers to electrodes for collection. Having a medium
that enables full photon absorption in a short length, together with a long
carrier lifetime that allows photogenerated carriers to reach electrodes before
recombining is ideal but is not always realistic. For example, silicon
photovoltaics, despite being a major player in the solar cell market, suffer
from their low absorption coefficient, thus requiring a thick absorbing layer
that impairs the efficiency with which photogenerated carriers are collected.
Radial p–i–n silicon nanowires (NWs) have been proposed as a candidate for
reducing the optical absorption length and required processing purity in
silicon-based solar cells without compromising their quantum efficiency and
yet reducing the overall cell cost.1 In this scheme incident light propagates
along the axial dimension of the NWs and thus has a greater chance of being
absorbed when the NW length extends beyond 10 mm due to interarray light
scattering effects [Fig. 31.1(a)]. At the same time, the core–shell radial p-i-n
structure leads electrical current flow along submicron radii, enabling rapid
collection of most photogenerated carriers, as the transport length is typically
less than the diffusion lengths of minority carriers [Fig. 31.1(b)].
Since the first discussion on the device operation of a radial NW geometry
for photovoltaic cell,1 much work has been done to experimentally realize the
advantages of this NW array system.2 14 A detailed survey on current
activities and progress on NW array solar cells can be found in Ref 15. In the
current work discussed in the chapter we perform finite-difference time-
domain (FDTD) simulations to investigate the absorption process in arrayed
Figure 31.1 Schematic illustration showing the advantages of a core–shell NW solar cell.
(a) Array arrangement enhances light scattering and photon absorption. (b) Short collection
length of charge carriers along radial direction is decoupled from the long absorption length
along the axis.
Design of Radial p–i–n Silicon Nanowires for High-Performance Solar Cells 825
radial NWs. The goal of this work is to gain insight on absorption processes in
NW arrays and to develop strategies for enhancing absorption efficiency. The
effects of light scattering and the material filling ratio (ratio of the cross-
sectional area that is occupied by the nanowires to the total area of the array)
at different NW spacings will be discussed. Evolution of absorption with NW
length, particularly in the long-wavelength range (700–1100 nm) will be
shown to illustrate the advantages of NWs as opposed to conventional planar
structures. In addition, actual NW geometries after shell overgrowth using
chemical vapor deposition (CVD) for different NW lengths and spacings have
been studied.
Figure 31.2 Illustration of NW solar cell fabrication process: (a) core formation, (b) surface
damage removal, and (c) junction formation by vapor–solid shell overgrowth.
also enable control over NW location and spacing by using patterned metal
catalysts but experiences difficulty in tuning growth conditions for a wide
range of NW diameters to achieve high yield in vertically aligned NW arrays.
The challenge for a top-down process is sidewall damage and residual
contamination during plasma etching, which hamper the nucleation of the
high-quality single-crystalline NW shell. In order to eliminate such detrimen-
tal defects, after the etch mask removal, the NW array is subjected to several
cycles of a thermal oxidation step to create a thin oxide layer followed by a
wet oxide etch step. The process also helps in removing sidewall undulation
(scallops) due to the cyclic Bosch etch process [Fig. 31.2(b)]. From the smooth
silicon surface on NW sidewalls, a single- crystalline radial shell can be grown
by SiH4-based CVD.16
The combination of top-down core etching and CVD shell growth enables
both p–n and n–p core–shell polarity with controlled doping concentration
and junction abruptness [Fig. 31.2(c)]. It also allows i regions of arbitrary
thicknesses to be inserted between the p and n regions, if desired. After shell
growth, one metal contact is deposited at the backside of the substrate, while
the other is placed on the top grown layer, outside of the array area. Ideally, a
transparent contact that fully covers the array is desired for perfect electrical
access to each individual NW, but for process simplicity, a side contact pad is
sufficient, as long as the shell doping is high enough for good electrical
conduction and the core and the shell are not too thin to be fully depleted. The
next section will investigate the relationship between doping concentration,
core–shell thicknesses, and applied voltage to satisfy the above conditions.
After processing, the NW array is characterized with standard techni-
ques16 to extract quantum efficiency and device efficiency. An example of the
external quantum efficiency of a silicon core–shell NW array cell is shown in
Design of Radial p–i–n Silicon Nanowires for High-Performance Solar Cells 827
Figure 31.3 Measured external quantum efficiency of a solar cell based on a silicon NW
array with 10-mm length, 3-mm wire-to-wire spacing, and 1.6-mm NW diameter.
Fig. 31.3. While the structure is far from optimal, a peak external quantum
efficiency of 70% is already achieved. The goal of this chapter is to elucidate
absorption processes in a NW array, and suggest an optimization scheme for
even higher quantum efficiency.
Figure 31.4 (a) Schematic cross section of the core–shell NW structure. (b) Graphical
solution to Eqs. (31.3) and (31.6) for a silicon NW with a ¼ 100 nm, Vbi ¼ 1 V, V ¼ 0 V, NA ¼
ND ¼ 1018 cm3.
solving the Poisson equation, similar to the case of 1D linear p-n junction,
except in the NW case, the Poisson equation is 1D in cylindrical coordination.
The charge density r is given by
qND for rn r a
r ¼ erE ¼ , ð31:1Þ
qNA for a r rp
where e is the dielectric constant of the material, and q is the electron charge.
The first equation that links rn and rp is dictated by the charge neutrality
condition:
p a2 r2n qND ¼ p r2p a2 qNA , ð31:2Þ
or
ND rn 2 NA rp 2
þ ¼ 1: ð31:3Þ
NA þ N D a NA þ ND a
By taking the integral for r, with boundary conditions E(rn) ¼ E(rp) ¼ 0, we
obtain
8
>
> 1 qND r2 r2n
< for rn r a,
2 e r
E¼ ð31:4Þ
>
> r2 r2
: 1 qNA p for a r rp :
2 e r
Design of Radial p–i–n Silicon Nanowires for High-Performance Solar Cells 829
Figure 31.5 Correlation between core–shell thicknesses and junction depletion. Zone 1:
When V < V1, neither the core nor the shell are fully depleted. When V ¼ V1, the shell is fully
depleted. Zone 2: When V < V2, neither the core nor the shell are fully depleted. When V ¼
V2, the core is fully depleted.
shell is fully depleted. From Eq. (31.9), the minimum core radius can be
derived as
v
u
u 4eVbi
,
amin ¼ u ð31:10Þ
t NA þ ND
qðNA þ ND Þ ln
ND
and the minimum shell radius must be
r v
NA þ N D u
u 4eV
bi :
bmin ¼ amin ¼ u ð31:11Þ
NA t NA þ ND
qNA ln
ND
Numerically, if we assume that the core and shell of silicon NWs are
doped at NA ¼ ND ¼ 1018 cm 3, we have Vbi ¼ 1 V, and the critical core–
shell radii are 44 nm and 62 nm, respectively. In order to achieve an effective
conduction channel to the electrodes, the actual radii of NWs should be larger
than the critical values by tens of nanometers.
In the case of a p–i–n radial junction, if the i region is thick enough, all of
the built-in potential and applied bias will drop at the I region, with a very thin
depletion into the p and n regions. For example, an i region of 60-nm in length
is enough to bear a built-in voltage of 1 V with only 20-nm depletion into the n
region and 10-nm depletion into the p region, assuming that NA ¼ ND ¼ 1018
cm 3 and a ¼ 100 nm. In such cases, contact depletion is not a concern.
Design of Radial p–i–n Silicon Nanowires for High-Performance Solar Cells 831
Figure 31.6 (a) Schematic layout of the NW array. (b) Simulation volume containing one
NW; R, T1, and T2 are monitor planes for reflection and transmission, respectively.
832 Chapter 31
Figure 31.7 a) Snapshots taken at t ¼ 0, t ¼ t1, and t ¼ t2 > t1 illustrating temporal field
propagation. (b) Wavelength-dependent spectra of field intensity. (c) Spatial absorption
along a slice through the center of a NW. The scale bar indicates the absorption per unit
volume in m3. (See color plate section.)
Figure 31.8 (a) Total absorption (NW þ substrate) of simulated NW arrays compared to
total absorption of an infinitely deep planar structure. (b) Absorption of NW arrays compared
only to absorption of planar bulks with the same thickness. (See color plate section.)
the small number of thermally generated carriers, which, in turn reduces its
component of dark current. For the same photocurrent generated from the
same amount of light absorption, the decrease of dark current leads to an
increase of open-circuit voltage in solar cells, as given by
nkT Iph
Voc ¼ ln þ1 , ð31:12Þ
q I
where
qV
I ¼ I0 exp 1
nkB T
is the bias-dependent dark current of a diffusion or generation–recombina-
tion-dominated photovoltaic device, n is the ideality factor, kB is the
Boltzmann constant, T is temperature, and I0 is related to the number of
minority carriers inside the devices, and this number is proportional to the
material volume where the carriers are generated.
Figure 31.9 (a) With an absorption profile at l ¼ 400 nm, the penetration depth is so
small that the amount of absorption depends on the air–silicon surface area rather than on
the volume of silicon. (b) Absorption spectra of silicon NW arrays with different pitches.
(See color plate section.)
in a smaller pitch than in a larger pitch. Another important remark drawn from
Fig. 31.10 is that in all three cases the absorption occurs more strongly near the
center of the NWs. This is as is expected because each individual NW acts as a
waveguide for light propagation, making light intensity more concentrated in
the center and, hence, resulting in a stronger absorption at the core.
836 Chapter 31
Figure 31.11 Absorption spectrum of NW arrays with different pitches and diameters. (See
color plate section.)
Observations from Fig. 31.10 raise a question about the degree to which
the outer part of the NW contributes to the total absorption. Figure 31.11
shows the absorption spectra of NW arrays with different diameter-to-pitch
(d/p) ratios for the three pitches p ¼ 1 mm, p ¼ 3 mm, and p ¼ 5 mm. For p ¼
5 mm, the pitch is too large for the scattering effect to become dominant, the
absorption is governed by the absorbing material volume; thus, the larger
diameter has a higher absorption. With p ¼ 3 mm, the effect of the material
filling ratio is still dominant, resulting in the same trends as in p ¼ 5 mm;
however, the absorption is enhanced due to the presence of light scattering.
For p ¼ 1 mm, the scattering becomes so strong that the proportionality
between absorption amplitude and material volume no longer holds. The
scattered light deviates from normal incidence, bouncing back and forth
between NWs and eventually becoming absorbed in the NWs rather than
penetrating the substrate. Also, regardless of the NW diameter, the absorption
is nearly the same.
The contribution of light scattering effects should be carefully considered
while optimizing device performance. Ideally, a smaller pitch is desired to
maximize enhanced absorption due to scattering. And at such a pitch, a small-
diameter NW can still achieve high absorption while exhibiting low dark
current due to smaller material volume. However, if technological limitation
does not allow for small pitch and small diameter, and especially if the
material is not of sufficient quality for the carrier diffusion lengths to be larger
than the NW diameter, the design for p–n NWs will need to be carefully
considered. For example, the core diameter should be small so that the
junction is closer to the core where more carriers are generated from the
absorption process. If permitted, the core material should be of p type, where
the diffusion length of minority electrons is longer than that of holes in the
Design of Radial p–i–n Silicon Nanowires for High-Performance Solar Cells 837
n-type region; thus, generated electrons near the core have a higher chance of
diffusing to the junction. In the absence of light scattering at a large pitch,
larger NW diameters will result in stronger absorption at the expense of larger
dark current. Compromises in terms of diameter and pitch will need to be
taken into account for experimental optimizations.
In this work, arrays of cylindrical NWs are utilized for the simplicity and
generality of the simulation; however, trends and interpretations can still be
applied to other experimental configurations that slightly deviate from
cylindrical form. In fact, advances in fabrication techniques can enable NW
geometries that are more beneficial for the absorption process. Conical NWs3
or dual-diameter structures4 have been shown to enhance the absorbance to
above 90%. The key in these works is the reduction of reflectance by
minimizing the top flat areas that most strongly reflect the light. The growth
conducted in this work has resulted in a different NW geometry that can lead
to similar effects. During CVD shell deposition, crystalline spherical bulbs (so-
called match heads) are built up at the top of the NWs due to faceted growth
with different growth rates along different directions [Fig. 31.2(c)]. Not only is
the reflection decreased due to literally no horizontal surface area, but the
match heads also act as microlenses that converge incident light into the NW
structure underneath, thus minimizing unabsorbed light penetrating into the
substrate. Simulation shows that NW arrays with match heads have much
higher absorption than similar arrays without match heads. Further details on
Figure 31.12 Optical photograph of a 6-in wafer with NW arrays fabricated using stepper
photolithography and dry etching. (L ¼ 10 mm, d ¼ 500 nm, p ¼ 1.5 mm.)
838 Chapter 31
match head formation and their effects on optical absorption will be reported
in a future publication.16
Acknowledgments
This research was funded in part by the Laboratory Directed Research and
Development program at Los Alamos National Laboratory and by the
Department of Energy EERE (EB2102010). Work was performed in part at
the Center for Integrated Nanotechnologies, a U.S. Department of Energy,
Office of Basic Energy Sciences, user facility at Los Alamos National
Laboratory (Contract DE-AC52-06NA25396).
References
1. B. M. Kayes, H. A. Atwater, and N. S. Lewis, “Comparison of the device
physics principles of planar and radial p–n junction nanorod solar cells,”
J. Appl. Phys. 97(11), 114302–11 (2005).
2. L. Hu and G. Chen, “Analysis of optical absorption in silicon nanowire
arrays for photovoltaic applications,” Nano Lett. 7(11), 3249–3252 (2007).
3. J. Zhu, Z. Yu, G. F. Burkhard, C.-M. Hsu, S. T. Connor, Y. Xu, Q.
Wang, M. McGehee, S. Fan, and Y. Cui, “Optical absorption
enhancement in amorphous silicon nanowire and nanocone arrays,”
Nano Lett. 9(1), 279–282 (2008).
4. Z. Fan, R. Kapadia, P. W. Leu, X. Zhang, Y.-L. Chueh, K. Takei, K. Yu,
A. Jamshidi, A. A. Rathore, D. J. Ruebusch, M. Wu, and A. Javey,
“Ordered arrays of dual-diameter nanopillars for maximized optical
absorption,” Nano Lett. 10(10), 3823–3827 (2010).
5. E. Garnett and P. Yang, “Light trapping in silicon nanowire solar cells,”
Nano Lett. 10(3), 1082–1087 (2010).
840 Chapter 31
Biographies for Jinkyoung Yoo, Paul Schuele, and David Evans are not
available.
Chapter 32
Nanostructured Electrode
Interfaces for Energy
Applications
Palash Gangopadhyay, Kaushik Balakrishnan, and Nasser Peyghambarian
College of Optical Sciences, The University of Arizona, Tucson, Arizona, USA
32.1 Introduction
32.2 0D Nanostructured Electrodes
32.3 1D Nanostructured Electrodes
32.4 2D Nanostructures and Nanostructured Electrodes
32.4.1 2D nanomaterials in energy storage
32.4.2 Supercapacitors
32.4.3 Batteries
32.4.4 2D nanostructures for flow based energy harvesting
32.4.5 Fabrication of nanostructured electrodes via nanoimprinting
32.5 3D Nanostructures
32.5.1 3D nanoarchitectures for energy harvesting
32.5.2 Sustainable integrated 3D powering solutions
32.6 Concluding Remarks
Acknowledgment
References
32.1 Introduction
One of the key challenges facing todays’ energy market is to provide highly
efficient, low-cost, and environmentally benign alternative-energy devices in
the near future, including solar1 10 and wind power,11 16 geothermal and
hydroelectric power,17,18 and batteries and supercapcitors.19 23 The impend-
ing exhaustion and the necessity to lower our dependence on fossil fuel, the
desire to develop a more sustainable transportation system, and the demand
843
844 Chapter 32
for a cleaner and secure energy future are all pushing the unprecedented
research effort and massive technological innovations that have been
experienced in the last few years.24 Concurrently, global investment initiatives
in renewable energy have seen a rapid boost in recent years, driven by
concerns about climate change, the forecast of an increasing cost of fossil
fuels, and national economic policies to create jobs. Looking forward, global
investment in renewable-energy projects alone will rise from $195 billion in
2010 to $395 billion in 2020 and to $460 billion by 2030. Over the next
20 years this growth will require nearly $7 trillion of new capital.25,26
Reflecting the rising production and investment levels, the installed capacity
of renewable power sources has been projected to climb, reaching 2.5 TW by
2030, a growth of over 800%. Although most of this market is currently
occupied by solar and onshore wind-power units, lithium (Li)-ion batteries
(LIBs), supercapacitors, and their hybrid devices have seen an upsurge in
market share in the recent years, mostly due to the resurgence of electric
vehicles and a renewed push for reducing airborne pollution from vehicles.27
In addition to electric vehicles, LIB and supercapacitors play important roles
in our daily lives by powering numerous portable consumer electronics
including laptops, personal digital assistants, and cell phones.
Given the vast expanse of the recent literature in this field and the
difficulty of addressing them all, this chapter is confined to the scope of our
group’s work on nanostructured electrode interfaces of various dimensions in
alternative-energy generation and storage. Within the last few years, our focus
has been on development of nanostructured electrodes and active materials in
organic bulk heterojunction (BHJ) solar cells, LIBs and ZnO-based energy-
harvesting devices. Nanostructured electrode interfaces, in particular, offer
unique combinations of properties in addition to the possible tunability of
their optical and electrical properties in a range of devices. However, the
design elements and fabrication processes that need to be addressed in each of
these devices are very different. In organic BHJ solar cells, the most common
donor–acceptor systems used are regioregular poly(3-hexyl)thiophene as
active donor and C60 or any of its derivatives as acceptor in a randomly mixed
nanophase. The photogenerated bound exciton is split into holes and electrons
at the donor–acceptor interface; the majority of holes transport through the
donor phase to an indium-tin-oxide electrode, whereas the electrons hop
through the acceptor phase to an aluminum electrode. In this case,
nanostructured electrodes are built to enhance absorption over a broad
spectrum of light as well as to be more efficient in charge collection from the
low-mobility polymeric semiconducting materials. A random walk model
suggests that it is essential to use a periodic and ordered electrode to improve
charge collection efficiency,28 whereas an aperiodic active material is favored
for most efficient absorption and superior antireflection properties.29,30 Such a
seemingly impossible combination, even if fabricated, runs the risk of
Nanostructured Electrode Interfaces for Energy Applications 845
imbalanced carrier transport where the effective hole mobility is far lower
than that of the electrons.31 Optimization of the nanostructured electrodes for
organic photovoltaic applications thereby must address the optical fields and
resulting photogeneration within the active material, the most efficient
pathway for maximum charge collection, and a possible hierarchical periodic–
aperiodic structure. Figure 32.1(a) presents a schematic of the working
principles of an electrode in an organic BHJ solar cell.
The design elements for the electrode of a purely electrical or
electrochemical device32 are very different from those of a solar cell.
A typical LIB is made up of a carbon–graphite electrode as an anode, a
nonaqueous liquid electrolyte, and a positive layered LiCoO2 electrode as
cathode. On charging, lithium ions are extracted from the layered LiCoO2
intercalation host (commonly carbon), and are transported through the
electrolyte and intercalate between the graphite layers in the anode. Opposite
electrochemical reactions occur during the discharging process, and the
electrons pass through the external circuit. Supercapacitors are capacitors
Figure 32.1 (a) Structure and working mechanism (top) and energy band diagram (bottom)
in a BHJ cell, where Jex is exciton diffusion flux, Jrec is recombination current, Jdark is dark
current, and Jph is the photocurrent. (b) Schematics of the electric double-layer structure
showing the arrangement of solvated anions and cations near the electrode–electrolyte
interface in the Gouy–Chapman–Stern model (top). The graph (bottom) represents the
potential across the electrical double layer. (See color plate section.)
846 Chapter 32
number of dimensions (see Fig. 32.2). Being electrodes, these structures can
also be classified based on the number of preferred directions for electron
conduction/transport. The structures listed in Fig. 32.2 fall under the same
dimensional classification by either; however, there may be some exceptions.
For example, an optical hex grating with nanoholes or pillars is geometrically
1D but is 2D considering the directional diffraction properties.
transport, and thickness.48 One of the primary issues being tested to enhance
OPV performance is to enhance absorption of photons in relatively thinner
OPV active material by efficient photon harvesting. One possible solution for
capturing the light more efficiently within an ultrathin active region is the
application of plasmonic nanoparticles (PNPs) and exploitation of localized
surface plasmon resonance (SPR).49 PNPs enhance the absorption of solar
radiation by near-field enhancement as well as by increasing the forward-
scattering cross section. Moreover, since the optical absorption spectra of
PNPs depend on their size, by tuning the size of nanoparticles (NPs), large
absorption can be achieved in different parts of the solar spectrum.50
We have devised a solar cell structure that takes advantage of e-field
enhancement from 50-nm diameter gold nanoparticles (AuNPs).51 54 The
conversion efficiency of OPVs can be broken into three factors: photon
absorption efficiency (PAE), exciton dissociation efficiency (EDE), and
charge collection efficiency (CCE). By incorporating the AuNPs, both PAE
and EDE improve, while CCE is improved by using a thinner active layer.
A 65% increase in total absorption was achieved using AuNPs. Finite-difference
time-domain (FDTD) simulations show an enhancement factor of 120 in the
e-field intensity in immediate proximity to the AuNPs at their resonant
frequency, as shown in Fig. 32.3(b). The large e-field around the particles
Figure 32.4 (a) SEM micrograph of a silver NSE with 47-nm silver layer coated on top of
polyacrylonitrile (PAN) nanostructures. The Ag nanopillars have a diameter of 152 6.2 nm,
height of 360 10 nm, and a center-to-center pitch of 200 7 nm. (b) Comparison of absorption
spectra between nanostructured and planar regions in an Ag electrode; the inset shows a
photograph of the sample (reprinted from Ref. 74 with permission). (See color plate section.)
Nanostructured Electrode Interfaces for Energy Applications 853
attractive in the application of new materials for the electronics industry. The
semi-metal properties of graphene have been somewhat of a challenge in
controlling the FET operation (namely, speed of switching, i.e., on/off ratio)
despite impressive charge transport characteristics, and this challenge is
currently being addressed. Similarly, MoS2 2D nanosheets have been explored
for use in FETs; already, impressive performance metrics with high switching
speeds and charge carrier mobilities >200 cm2V 1s 1 have been achieved
and, with new advances, will immensely expand the scope of applications in
the electronics industry.86,87 The planar integration efforts of the 2D
nanomaterials are expected to favor ease of integration in comparison to
the 1D nanostructures such as carbon nanotubes, Si nanowires, etc., while
h-BN and other ceramics and chalcogenide-based 2D nanomaterials are being
explored for use as dielectric layers for electronics and energy applications.80
Perhaps the likely widespread applications of the 2D nanomaterials, particularly
graphene, are anticipated to be in nanocomposite fillers for improving the
mechanical properties. Successful demonstrations from exploratory efforts in
research labs have also paved the way for rapid commercialization of 2D
nanomaterials. Recent market estimates indicate that graphene-related products
will be worth $100 million in the next five years, with the largest impact from
electrode materials in energy-storage applications.88 This impact is largely due to
the combination of favorable properties of graphene and related materials. The
key characteristics for the use of graphene as active electrodes in energy-storage
applications are discussed below, with regards to the use in supercapacitors, LIBs,
and solar cells.
32.4.2 Supercapacitors
Activated carbon and carbon nanostructures such as CNTs are well known for
their use in electrical double-layer capacitors (EDLCs), owing to their unique
combination of pore sizes, electrical conductivities, and compatibility with a
range of electrolytes.98 The wide range of pore size distribution in carbon
nanostructures allows for maximizing the percolation of electrolyte ions within
the electrodes, leading to efficient formation of electrical double layers.
Graphene, in particular, is deemed critical for the efficient use of the available
surface area because of the ability to use both the basal and edge planes for
856 Chapter 32
the formation of the electrical double layer over the large surface area
(2600 m2 g 1).99 To this end, graphene obtained from the various methods
described above has been directly employed for the fabrication of super-
capacitors.99 101 Also, a wide range of electrolytes [aqueous, organic, room-
temperature ionic liquid (RTIL), polymer gel] have been demonstrated to be
suitable for fabricating double-layer capacitors.99,101,102 Additionally, combi-
nations of graphene with other nanomaterials (nanocomposites) have been
studied for the fabrication of redox supercapacitors (pseudo-capacitors).103
Given this vast range of exploration, we summarize some of the key metrics
achieved with graphene-based supercapacitors. The performance from
graphene devices has set record performances in terms of specific capacitance
(100–250 Fg 1),100,101 energy density (85–136 Whkg 1),104 and power
density (10 kWkg 1).99
While a majority of the supercapacitor devices have been carried out in a
stacked geometry, Yoo et al. demonstrated a planar approach that maximizes
the utilization of the available electrochemical surface area.98 The major
advantage of the in-plane geometry is the favorable movement of the ions
across the electrolyte because the applied field is in the same direction as the
high conductivity of the graphene. In this approach, both pristine graphene
and rGO films are tested and show higher capacities (three times) than the
stacked design. The concept of in-plane supercapacitors and relevant details
are illustrated in Fig. 32.5. CVD-grown pristine graphene shows areal specific
Figure 32.5 The concept and comparison of (a) stacked and (b) in-plane graphene-based
supercapacitors. The in-plane geometry is more favorable for movement of ions (upon
application of potential) because of the high in-plane conductivities and the direction of the
electric field. (c) A solid state in-plane graphene supercapacitor with a polymer-gel electrolyte.
(d) These planar graphene supercapacitors can also be fabricated in flexible geometries. Both
(e) pristine graphene grown by CVD and (f) and (g) rGO films of 10-nm thickness were
tested for their characteristic performance using in-plane and stacked geometry (reprinted
from Ref. 98 with permission; © 2011, American Chemical Society). (See color plate section.)
Nanostructured Electrode Interfaces for Energy Applications 857
32.4.3 Batteries
Graphite has been a popular choice for anodes in LIBs because of its high
electrical conductivities and its ability to intercalate and release lithium ions
(upon charging and discharging).107 Carbon nanostructures are therefore
immensely useful in battery applications, particularly as anodes.108,109 To this
end, graphene and doped-graphene structures have been widely employed as
anodes in LIB devices and show high performance.90,108,110 Graphene-based
batteries show high performance because of the dramatically increased surface
area, which is a crucial factor in LIBs. Additionally, graphene and other
carbon nanostructures have been routinely employed as conductive fillers for
cathode materials.111 Traditionally used cathodes in LIBs are often oxides
that suffer from poor electrical conductivities. Thus, graphene, given its 2D
sheet-like nature allows for improvement in conductivity by acting as a filler.
Recently, porous 3D architectures of 2D graphene sheets used as anodes in
LIBs have also shown high-performance metrics.109 Graphene is likely poised
to make a significant impact in LIB applications, given the numerous
combinations in which it can be employed (e.g., as anodes and as conductive
fillers in cathodes) and currently remains an active field of research. The
advancement of new strategies and architectures being developed for
graphene-based electrodes will be critical for next-generation LIBs.
lower operating and start-up materials cost, and the potential for high-
throughput manufacturing including possible roll-to-roll production. The
dimensions and distribution of the patterns in the NSEs are easily tuned by
simply varying the deposited metal or metal-oxide layer thickness, which in
turn allows a greater control over the optical and electrical properties of the
NSEs. An important milestone in this area of research is demonstration of a
suitable approach that allows production of high-aspect-ratio NSEs with
tunable optical and electrical properties with features below 100 nm and
covering device-relevant areas of a few square centimeters.
Interest in patterning polymer-based nanodevices and creating sub-100-
nm metal and TCO-based NSEs has led us to modify traditional imprint-
lithography techniques to enable the creation of an array of sub-30-nm-
diameter polymer nanostructures.152 We have created a thermally stable,
optically transparent, large-area, nanostructured scaffold that can be used for
fabrication of various metal and metal-oxide electrodes of interest via
traditional thermal, sputtering, e-beam, and CVD routes. In this approach, a
hard e-beam-lithographed Si or SiC master is used to directly imprint a desired
large-area nanopattern onto PAN film. The composite film is then cured at
150 8C to complete the production of nanostructured PAN films. Silver and
TCOs (such as indium tin oxide and indium zinc oxide) are then coated with
different thicknesses onto the PAN structure. A conformal coating of Ag and
TCOs is then obtained to create the representative NSEs reported here. The
mold used and the printed PAN structures are shown in Fig. 32.6. The molding
Figure 32.7 (a) Schematic of a high-aspect-ratio TCO nanostructure with a PAN core,
(b) optical bandgaps of ITO and IZO 2D electrodes, and (c) SEM images of an ITO electrode
(reprinted from Ref. 74 with permission). (See color plate section.)
32.5 3D Nanostructures
3D nanostructures offer robust mediums for a range of energy-harvesting
applications. Particularly, these structures offer favorable attributes such as
high surface area and increased compatibility with other nanostructures that
can be selectively or specifically integrated with desired structural character-
istics to enable further improvement and on current performance metrics. The
new nanoimprinting technique developed by the authors and reported in
Section 32.5.5 is advantageous because PAN, being a thermoset polymer and
otherwise unprintable using traditional imprinting techniques, can be printed
into a large range of dimensions. The PAN structure can further be
crosslinked at 150 8C, followed by stabilization at 250 8C. These 2D stabilized
nanostructures can then be carbonized at high temperature (1200 8C) to form
graphitic carbon with a hierarchical porous structure: a 3D high-surface-area
NSE. The authors are currently developing LIBs and supercapcitors using
these 3D porous graphitic carbon structures. The flexible process of the
Nanostructured Electrode Interfaces for Energy Applications 861
Figure 32.8 (a) Illustration of the 3D ZnO–cellulose-fiber nanocomposite and (b)–(d) the
corresponding morphological SEM images after the growth of ZnO nanostructures on
cellulosic fibers. (e) and (f) A flexible multisource energy-harvesting device based on 3D
ZnO–cellulose nanocomposite. [(a), (e), and (f) reprinted from Ref. 157 with permission;
© 2011 Wiley-VCH Verlag GmbH & Co.].
862 Chapter 32
Figure 32.9 (a) Schematic illustration of the integrated 3D powering solutions based on the
strategic combination of a piezoelectric polymer and an LIB. This integrated platform
benefits from the adaptation of piezoelectric polymer PVDF instead of the standard
electrolyte separators commonly employed in LIBs. (b) These 3D integrated self-powering
stacks can be attached to the sole of a shoe for battery charging during routine walking and
running. (c) Morphological characterization revealing the individual components of the 3D
self-powered stack. (d) SEM image of the TiO2 nanotubes that act as the anode in the LIB
part of the 3D self-powering stack (reprinted from Ref. 69 with permission; © 2012, American
Chemical Society). (See color plate section.)
Acknowledgment
This work was supported in part by the Center for Integrated Access
Networks / Engineering Research Center (CIAN ERC) (Grant No. EEC-
0812072) and the U.S. Air Force Office of Scientific Research / Center of
Nanostructured Electrode Interfaces for Energy Applications 865
References
1. T. Soga, Nanostructured Materials for Solar Energy Conversion, First
edition, Elsevier, Amsterdam, pp. 45–80 (2006).
2. N. Asim, K. Sopian, S. Ahmadi, K. Saeedfar, M. A. Alghoul,
O. Saadatian, and S. H. Zaidi, “A review on the role of materials science
in solar cells,” Renew. Sustain. Energy Rev. 16(8), 5834–5847 (2012).
3. J. H. Cai, H. Chen, and L. Y. Han, “Models of electron injection,
diffusion and recombination in dye sensitized solar cells,” Int. J. Mod.
Phys. B 26(17), 1230009 (2012).
4. W. Chen, M. P. Nikiforov, and S. B. Darling, “Morphology characteri-
zation in organic and hybrid solar cells,” Energy Environm. Sci. 5(8),
8045–8074 (2012).
5. X. Huang, S. Han, W. Huang, and X. Liu, “Enhancing solar cell
efficiency: The search for luminescent materials as spectral converters,”
Chem. Soc. Rev. 42(1), 173–201 (2013).
6. P. Kumar and S. Chand, “Recent progress and future aspects of organic
solar cells,” Prog. Photovolt. 20(4), 377–415 (2012).
7. S. Mokkapati and K. R. Catchpole, “Nanophotonic light trapping in
solar cells,” J. Appl. Phys. 112(10), 101101 (2012).
8. R. Po, C. Carbonera, A. Bernardi, F. Tinti, and N. Camaioni, “Polymer
and carbon-based electrodes for polymer solar cells: Toward low-cost,
continuous fabrication over large area,” Solar Energy Mater. Solar Cells
100, 97–114 (2012).
9. M. Wright and A. Uddin, “Organic-inorganic hybrid solar cells:
A comparative review,” Solar Energy Mater. Solar Cells 107, 87–111 (2012).
10. Y. Yang, K. Mielczarek, M. Aryal, A. Zakhidov, and W. Hu,
“Nanoimprinted polymer solar cell,” ACS Nano 6(4), 2877–2892 (2012).
11. M. A. Abdullah, A. H. M. Yatim, C. W. A. Tan, and R. Saidur,
“A review of maximum power point tracking algorithms for wind energy
systems,” Renew. Sustain. Energy Rev. 16(5), 3220–3227 (2012).
12. A. Arvesen and E. G. Hertwich, “Assessing the life cycle environmental
impacts of wind power: A review of present knowledge and research
needs,” Renew. Sustain. Energy Rev. 16(8), 5994–6006 (2012).
866 Chapter 32
155. K.-I. Park, S. Xu, Y. Liu, G.-T. Hwang, S.-J. L. Kang, Z. L. Wang, and
K. J. Lee, “Piezoelectric BaTiO3 thin film nanogenerator on plastic
substrates,” Nano Lett. 10(12), 4939–4943 (2010).
156. X. Wang, K. Kim, Y. Wang, M. Stadermann, A. Noy, A. V. Hamza,
J. Yang, and D. J. Sirbuly, “Matrix-assisted energy conversion in
nanostructured piezoelectric arrays,” Nano Lett. 10(12), 4901–4907
(2010).
157. A. Kumar, H. Gullapalli, K. Balakrishnan, A. Botello-Mendez,
R. Vajtai, M. Terrones, and P. M. Ajayan, “Flexible ZnO–cellulose
nanocomposite for multisource energy conversion,” Small 7(15), 2173–
2178 (2011).
158. D. Liang, H. Yang, S. W. Finefrock, and Y. Wu, “Flexible nanocrystal-
coated glass fibers for high-performance thermoelectric energy harvest-
ing,” Nano Lett. 12(4), 2140–2145 (2012).
879
880 Index
H high-resolution transmission
Hall transport, 48 electron microscopy (HRTEM),
Hamilton relation, 563 43, 138
Hamiltonian, 295 high-resolution x-ray diffraction
hard solders, 213 (HRXRD), 69, 140
Harrison, 554, 571 highly absorbing water films, 648
heat capacitance, 537, 539 HITRAN simulation, 616
heat sinking, 236 hole blocking, 353
heat waste, 215 hybrid configurations, 329
Helmholtz layer, 846 hybrid photodetectors (HPDs),
hemispherical radiation intensity, 494
543 hydrogen sensors, 803
hemoglobin, 777 hydrogenic model, 160
Hermitian symmetrized velocity hydrophobic materials, 777
operator, 562 hyperspectral imaging, 656
heterocascading, 650 hyperspectral target detection
heterojunction offset, 157 algorithms, 657
heterostructure engineering, 349 hysteresis, 31
heterostructure field-effect
transistors (HFETs), 290 I
heterostructure thermopiles image analysis, 656
(H-PILEs), 534 imaging standoff detection, 653
heterostructures, 106 impact ionization, 110–111, 116
Hg vacancy doping, 329 impact ionization threshold
HgCdTe, 324, 370, 372, 377, 442, energies, 117
445–446, 683–684 in situ tracing, 633
HgCdTe arrays, U.S. common- in vivo imaging, 777
module, 327 InAs/GaSb type-II broken-gap
HgMnTe, 340 superlattices (T2SLs), 86, 106,
HgZnTe, 340 337
hierarchical periodic-aperiodic incident radiated power, 542
structure, 845 incoherent broadband cavity-
high electron mobility, 290 enhanced absorption spectroscopy
high heat load, 232 (IBBCEAS), 699
high-heat-load (HHL) package, indirect bandgap, 492
609 indirect X and L valleys, 210
high-index plane, 358 indium tin oxide, 844, 850
high-operating-temperature (HOT), indium zinc oxide, 859
352 industrial process control, 598
high-operation-temperature (HOT) infrared (IR), 95, 99, 347
detectors, 339 infrared detectors, 41
high-performance liquid infrared focal plane array, 247
chromatography (HPLC), 697 infrared FPA image sensor, 546
Index 885
Electrons, photons, and even thermal properties can all be engineered at the
nanolevel. Possibly the simplest aspect of nanotechnology, the 2D quantum well has
dramatically enhanced the efficiency and versatility of electronic and optoelectronic
devices. Nanotechnology has now progressed to 1D (quantum wire) and 0D
(quantum dot) systems that exhibit remarkable and sometimes unexpected
behaviors. With these components serving as the modern engineer's building
blocks, it is remarkable to consider the endless possibilities that nanotechnology
holds in store.
P.O. Box 10
Bellingham, WA 98227-0010
ISBN: 9780819495969
SPIE Vol. No.: PM238