Prodromakis T 2009 PHD Thesis

Application of Maxwell-Wagner
Polarisation in Monolithic
Technologies
by
Themistoklis Pro dromakis
December 2008
A thesis submitted for

the degree of Doctor of Philosophy of the University of London
Department of Electrical and Electronic Engineering

Imperial College of Science, Technology and Medicine
University of London
Contents
1 Introduction 6
1.1 Motivation 6
1.2 Research Objectives 8
1.3 Overview 8
1.3.1 Electromagnetic Wave Propagation on MIS Structures 9
1.3.2 MIS Modeling and Devices 9
1.3.3 High-K Structures Exhibiting Maxwell-Wagner Polarisation 9
1.3.4 Engineering the Maxwell-Wagner Polarisation Effect 10
1.3.5 Miniaturising Passive Components by Utilisation of MIS Microstrip 10
1.3.6 Conclusion 10
2 Electromagnetic Wave Propagation on MISM Structures 12
2.1 Introduction 12
2.2 Space-Charge Layer Effects at Insulator-Semiconductor Interfaces 14
2.2.1 Ideal MOS Diode 15
2.2.2 The SiO2 — Si MOS Diode 19
2.3 Polarisation in Dielectrics 25
2.3.1 Maxwell-Wagner Polarisation Mechanism 26

CONTENTS ii
2.4 Propagation modes of MIS Microstrip 30
2.4.1 Slow-Wave Propagation Mode 32
2.4.2 Attenuation Due to Semiconductor and Metallic Conductor 36
2.5 Summary 39
3 MIS Modeling and Devices 44
3.1 Introduction 44
3.2 History of the MIS Lines Mathematical Analysis 45
3.2.1 Empirical Lumped Circuit Models 45
3.2.2 Parallel-Plate Wave Guide Approach 46
3.2.3 Full-wave Analysis 48
3.2.4 Quasi-Static Modeling 50
3.2.5 Single Layer Reduction Process 51
3.2.6 Device-Level Simulation 52
3.3 Devices Utilising MIS Microstrip 57
3.3.1 Voltage-Variable Capacitor 57
3.3.2 Delay Lines and Electronically Variable Phase Shifter 57
3.3.3 Voltage-tunable MIS filters 63
3.3.4 Variable coupling-coefficient directional couplers 64
3.3.5 Distributed FET transistors 64
3.4 Summary 64
4 High-K structures Exhibiting Maxwell-Wagner Polarisation 73
4.1 Introduction 73
4.1.1 Technology Modelling 74

CONTENTS iii
4.1.2 Measurement Apparatus 75
4.2 Silicon Based Structures 76
4.2.1 Characterisation Devices 76
4.2.2 Effect of Substrate Conductivity 79
4.2.3 Effect of Semiconductor to Insulator Thickness Ratio 82
4.2.4 Effect of Transmission Line Dimensions 86
4.3 GaAs Membranes 90
4.3.1 Implemented Process Flow 91
4.3.2 Results Discussion 91
4.4 Slow-wave Characteristics of Microstrip Transmission Lines on Aqueous Di-

electrics 93
4.4.1 Experimental Platform 93
4.4.2 Apparatus 95
4.4.3 Simulation Results 96
4.4.4 Experimental Results 97
4.5 Summary 103
5 Engineering the Maxwell-Wagner Polarisation Effect 106
5.1 Introduction 106
5.2 KOH on Si before oxidation 107
5.2.1 Process Description 108
5.2.2 Results 109
5.3 Unbuffered HF 5% on Si/Si02 111
5.3.2 Results 112

CONTENTS iv
5.4 Reactive Ion Etching of Si before oxidation 113
5.4.2 Results 114
5.5 Au nano-spheres embedded in the oxide 116
5.5.2 Results 118
5.6 Germanium spheres embedded in the oxide 119
5.6.2 Results 120
5.7 Na+ Ions on Si/Si02 122
5.7.2 Results 122
5.8 Ar, As and P Implantation on Si 124
5.8.2 Results 126
5.9 SrTiO3 on Si02 /Si Substrate 128
5.9.2 Film Characterisation 128
5.9.3 Results 129
5.10 Summary 131
6 Miniaturising Passive Components by Utilisation of MISM Microstrip 134
6.1 Introduction 134
6.2 Miniature Delay Lines 135

CONTENTS
6.2.1 Design and Fabrication 136
6.2.2 Experimental Results and Discussion 137
6.3 Miniature Distributed Filter Design on Silicon CMOS 139
6.3.1 CMOS Manufacturing Process 141
6.3.2 Measurements, Simulations and Discussion 141
6.4 Microstrip Stepped Impedance LowPass Filters 144
6.4.1 Stepped Impedance Filters on Conventional Substrates 145
6.4.2 Stepped Impedance Filters on Variable Dielectric Substrates 147
6.5 Miniature Microstrip Antennas 149
6.5.1 Microstrip Antennas on Laminar Substrates 150
6.6 Miniature Quadrature Hybrid Couplers 153
6.6.1 Simulations and Discussion 153
6.7 Summary 156
7 Conclusion 161
7.1 Contributions 162
7.2 Recommendations for Future Work 163
7.2.1 Maxwell-Wagner Polarisation on Controlled Damaged Specimens . 163
A Interfacial and Space-Charge Polarisation 166
B MATLAB Modeling of MIS Microstrip Using Hasegawa's Model 169
C Complementary Measurements 172
D Gnici, Measurements 174

CONTENTS vi
E Capacitance-Voltage Measurements 181
F Publications 191
List of Figures
2.1 Metal-Insulator-Semiconductor-Metal (MISM) structure. 13
2.2 (a) Perspective view of a MIS diode. (b) Cross-section of a MIS diode . . 14
2.3 Energy band diagrams and charge distributions of an ideal MIS diode in (a)
accumulation, (b) depletion, and (c) inversion cases. 16
2.4 (a) High-frequency MIS CV curve. Inset shows the equivalent circuit for the
two layer condenser. (b) Effect of measurement frequency on the CV curve
[13] 19
2.5 Charges and their location for thermally oxidized silicon [15]. 21
2.6 Capacitances of an 1\40S capacitor for the different bias conditions [12]. . 23
2.7 Capacitance stretch-out due to interface trapped charges. 24
2.8 Polarisation mechanisms and their effect on the frequency dependence of the
relative electric permittivity [19]. 26
2.9 Interfacial polarisation on a two layer condenser. 27
2.10 Two-layer condenser and its equivalent circuit. 28
2.11 Relaxation spectrum of the two-layer condenser [25] 29
2.12 Possible shapes and orientations of particles of the conducting medium [24]. 30
2.13 Resistivity-frequency domain chart showing different modes of propagation

in MIS microstrip. 31
2.14 Quasi-static e (left) and „X' (right) field distribution in a MIS microstrip. 32
vii
LIST OF FIGURES viii
2.15 Frequency response of effective index of refraction with respect to the semiconductor-
insulator thickness ratio 34
2.16 E field analysis on the interface between two dielectrics. 35
2.17 Frequency response of relative effective dielectric constant (left axis) and
attenuation per wavelength (right axis) for substrate's conductivities of a-8,=
20 S/cm. Inset: equivalent circuit of the two layer lossy condenser [25]. . . . 37
2.18 Simulated attenuation for different insulators 38
2.19 Loss tangent frequency response for several semiconductor-insulator thickness

ratios as well as different substrate conductivities. 39
3.1 MIS equivalent circuits for q-TEM, slow-wave and skin-effect mode, proposed
by Hasegawa [7] 47
3.2 Jager's slow-wave equivalent circuit [9]. 47
3.3 Equivalent lumped-element model of a two-dimensional lossless TLM network

[20] 49
3.4 SLR based circuit model of multilayer slow-wave microstrip line. 52
3.5 Gaofeng's equivalent circuit model for MIS interconnects. 54
3.6 Schematic drawing of MIS meander-line showing biasing scheme with block-
ing capacitors and RF choke. M, I, and S represent metal, insulator, and
semiconductor, respectively [53]. 59
3.7 Block diagram of a phase-shift oscillator. 61
3.8 Pictorial illustration of a STWA. 62
4.1 Calibration sequence of semi-empirical routines. 74
4.2 CV measurement set up. 75
4.3 Maxwell-Wagner polarisation mechanism characterisation mask. 77
4.4 Microstrip Transmission Lines of several widths and lengths. 78

LIST OF FIGURES ix
4.5 Coplanar waveguides of several widths, lengths and gaps. 78
4.6 Top metal electrode of MOS capacitors with the electrode area varying. . 78
4.7 Measured group delay and effective length of a 100 tan MSTL on substrates
of various conductivity. 81
4.8 Measured insertion loss of a 100 pm long MSTL on substrates of various

conductivity. 81
4.9 Measured and simulated unwrapped phase-shift of a 100 pm long MSTL with
the semiconductor to insulator thickness ratio varying 83
4.10 Measured group delay and effective length of a 100 pm long MSTL with the
semiconductor to insulator thickness ratio varying. 84
4.11 Measured group delay and effective length of a 100 pm long CPW with the
semiconductor to insulator thickness ratio varying. 84
4.12 Measured and simulated insertion and return loss of a 100 pm long MSTL
with the semiconductor to insulator thickness ratio varying 85
4.13 Optional caption for list of figures 87
4.14 Measured unwrap-phase shift of MSTLs for various widths: 5 pm, 20 pm, 40
pm and 60 pm. Inset: Zoomed-in caption of linear phase variation up to 1
GHz 87
4.15 Measured group delay and effective length of MSTLs for various widths: 5
pm, 20 pm, 40 pm and 60 pm 88
4.16 Measured insertion loss of MSTLs for various widths: 5 pm, 20 pm, 40 pm
and 60 pm. Inset: Zoomed-in caption of insertion loss up to 10 GHz 89
4.17 Measured and simulated unwrap phase-shift of MSTLs for various lengths:
100 pm, 200 pm, 400 pm 89
4.18 Schematic illustration of GaAs membrane structure. Shown are: (a) top
metal patterns, (b) cross-section of membranes and (c) SEM microphotographs
of the back-side during etching (vias are visible as dark squares). 90
LIST OF FIGURES
4.19 Measured group delay and effective length of MSTLs for various widths: 5
pm 10 jim, 40 tim and 60 ,um. 92
4.20 Measured insertion loss of MSTLs for various widths: 5 pm 10 ,um, 40 pm

and 60 pm. 92
4.21 Experimental structure with a copper plate forming the back metallization. 94
4.22 Top view of the experimental structure showing the SMA connectors as well
as the printed lines. 95
4.23 Schematic illustration of coaxial cylinder electrodes as deployed in measuring

the electrical conductivity of a liquid [7] 96
4.24 g(left) and<7.1

1' (right) field distribution of the liquid substrate with crsub =
0.09S/m. 97
4.25 S(left) and dr (right) field distribution of the liquid substrate with crsub -=
4S/m. 97
4.28 Mismatch loss due to the slow-wave mode impedance decrement. 101
5.1 Si surface after KOH processing. The increased roughness can be observed
in the zoomed microscope photograph 107
5.2 AFM photograph of Si surface after KOH processing. 108
5.3 Measured group delay and effective length of a 100tim long MSTL on Si02
based on standard Si and on KOH processed Si. 110
5.4 Measured insertion loss of a 100i.cm long MSTL on Si02 based on standard
Si and on KOH processed Si. 110
5.5 Microphotograph of oxide surface defects introduced by unbuffered HF wet

etching. 111
LIST OF FIGURES xi
5.6 Measured group delay and effective length of a 100 pm long MSTL on good
quality Si02 as well as surface defected Si02 on Si. 112
5.7 Measured insertion loss of a 100 pm long MSTL on good quality Si02 as
well as surface defected Si02 on Si. 113
5.8 Measured unwrapped phase-shift of a 100 pm long MSTL on Si02 based on

standard Si as well as on RIE defected Si 115
5.9 Measured insertion loss of a 100 tim long MSTL on Si02 based on standard
Si as well as on RIE defected Si. 115
5.10 Au nanodots on Si02 surface. Zoomed section is a SEM microphotograph

demonstrating Au spheres dissolved in the oxide layer 116
5.11 AFM photograph of Au nanodots on Si02 surface 117
5.12 Measured group delay and effective length of a 100 pm long MSTL based on
standard Si02 and on Si02 with buried Au nanodots 118
5.13 Measured insertion loss of a 100 pm long MSTL based on standard Si02 and
on Si02 with buried Au nanodots 119
5.14 AFM photograph of diffused Ge in Si02 120
5.15 Measured group delay and effective length of a 100 pm long MSTL on Si02
based on standard Si02 and on Ge rich SiO2. 121
5.16 Measured insertion loss of a 100 11,M long MSTL on Si02 based on standard
Si02 and on Ge rich Si02. 121
5.17 Measured group delay and effective length of a 200 pm long MSTL on
Si02 /Si sample before and after inducing Na+ ions 123
5.18 Measured insertion loss of a 200 pm long MSTL on Si02 /Si sample before
and after inducing Na+ ions. 123
5.19 Profile code retained dose estimates. 125
5.20 Displacements per atom 125

LIST OF FIGURES xii
5.21 Measured group delay and effective length of a 100 pm long MSTL on
Si02/Si and on Si02 /Si where the Si surface has been doped with Ar,
As and P. 127
5.22 Measured insertion loss of a 100 pm long MSTL on Si02 /Si and on Si02 /Si
where the Si surface has been doped with Ar, As and P. 127
5.23 X-ray diffraction analysis of the processed sample. 129
5.24 Measured group delay and effective length of a 100 pm long MSTL on a
Si02 /Si and a SrTi031Si021Si system. 130
5.25 Measured insertion loss of a 100 pm long MSTL on a Si02 /Si and a SrTi031Si021Si
system. 130
6.1 SEM microphotograph of the designed delay lines. 136
6.2 Measured vs Simulated (analytical & equivalent model) phase shift. 137
6.3 Measured vs Simulated (analytical & equivalent model) 821 138
6.4 Simulated and Measured insertion loss of the proposed filter (L=100 pm) for
various slowing factors. 142
6.5 Simulated and Measured insertion loss of the proposed filter (L=200 pm) for
various slowing factors. 143
6.6 Measured return loss of the proposed filters for various slowing factors. . . 143
6.7 5th order low pass microstrip stepped impedance filter on RT Duroid 5880,
with fo=1 GHz. 146
6.8 Simulated and measured return loss and insertion loss of the stepped impedance
filter shown in figure 6.7. 146
6.9 Top and side view of proposed stepped impedance filter 147
6.10 Simulated return loss and insertion loss of the proposed stepped impedance
filter. 148
6.11 Far field radiation and directivity of a 10x10 mm patch antenna on Si substrate.151
LIST OF FIGURES xiii
6.12 Far field radiation and directivity of a 1x1 mm patch antenna on proposed
substrate. 151
6.13 Insertion loss of a 10x10 mm patch antenna on Si substrate 152
6.14 Insertion loss of a lx1 mm patch antenna on proposed substrate. 152
6.15 S-parameter magnitudes versus frequency for a quadrature hybrid coupler on

typical substrate 154
6.16 S-parameter magnitudes versus frequency for a miniature quadrature hybrid

coupler on a laminar substrate 155
6.17 Unwrapped Phase-shift for the couplers shown in figures 6.15 and 6.16. . . 155
C.1 Measured and simulated unwrap phase-shift of a 100 pm long microstrip

transmission line (MSTL) with the semiconductor to insulator thickness ratio
varying. 172
C.2 Measured and simulated insertion loss of a 100 pm long microstrip transmis-
sion line (MSTL) with the semiconductor to insulator thickness ratio varying. 173
C.3 Measured group delay and effective length of a 100 pm long MSTL on Si02
based on standard Si as well as on RIE defected Si 173
D.1 Measured Gmax losses of a 100 pm long MSTL on substrates of various

conductivity. 174
D.2 Measured Gmax losses of a 100 pm long MSTL with the semiconductor to
insulator thickness ratio varying. 175
D.3 Measured Gmax losses of a 100 pm long MSTL for various widths: 5 pm, 20
pm, 40 pm and 60 pm 175
D.4 Measured Gmax losses of a 100 pm long MSTL on GaAs membranes for
various widths: 5 pm 10 pm, 40 pm and 60 pm. 176
D.5 Measured Gmax losses of a 100 pm long MSTL on Si02 based on standard
Si and on KOH processed Si. 176
LIST OF FIGURES xiv
D.6 Measured Gmax losses of a 100 pm long MSTL on good quality Si02 as well
as surface defected Si02 on Si. 177
Si as well as on RIE defected Si. 177
D.8 Measured Gmax losses of a 100 pm long MSTL based on standard Si02 and
on Si02 with buried Au nanodots 178
Si02 and on Ge rich SiO2. 178
D.10 Measured Gmax losses of a 100 pm long MSTL on Si02/Si sample before
and after inducing Na+ ions. 179
D.11 Measured Gmax losses of a 100 pm long MSTLon Si02/Si and on Si02 /Si
where the Si surface has been doped with Ar, As and P 179
D.12 Measured Gmax losses of a 100 pm long MSTL on a Si02/Si and a SrTiO3/ Si02/ Si
system. 180
E.1 CV measurement of a MOS diode at 1 kHz, with t s.b/tox =2100 181
E.2 CV measurement of a MOS diode at 1 MHz, with t sub/tox =2100. 182
E.3 CV measurement of a MOS diode at 1 kHz, with t sub/tox =1140 182
E.4 CV measurement of a MOS diode at 1 MHz, with t s ,,b/tox =1140. 182
E.9 CV measurement of a MOS diode at 1 kHz, with KOH 20% processed Si. 184
E.10 CV measurement of a MOS diode at 1 MHz, with KOH 20% processed Si. 184
E.11 CV measurement of a MOS diode at 1 kHz, with HF 5% processed SiO2. 185
E.12 CV measurement of a MOS diode at 1 MHz, with HF 5% processed Si02. 185

LIST OF FIGURES xv
E.13 CV measurement of a MOS diode at 1 kHz, with RIE processed Si. 185
E.14 CV measurement of a MOS diode at 1 MHz, with RIE processed Si. . 186
E.15 CV measurement of a MOS diode at 1 kHz, with Au nanodots in Si. 186
E.16 CV measurement of a MOS diode at 1 MHz, with Au nanodots in Si. . 186
E.17 CV measurement of a MOS diode at 1 kHz, with Ge spheres in Si02 . . 187
E.18 CV measurement of a MOS diode at 1 MHz, with Ge spheres in Si02. . 187
E.19 CV measurement of a MOS diode at 1 kHz, with induced Na+ in Si02. 187
E.20 CV measurement of a MOS diode at 1 MHz, with induced Na+ in Si02 . . 188
E.21 CV measurement of a 1\40S diode at 1 kHz, with Ar implanted Si. 188
E.22 CV measurement of a MOS diode at 1 MHz, with Ar implanted Si 188
E.23 CV measurement of a MOS diode at 1 kHz, with As implanted Si. 189
E.24 CV measurement of a MOS diode at 1 MHz, with As implanted Si 189
E.25 CV measurement of a MOS diode at 1 kHz, with P implanted Si 189
E.26 CV measurement of a MOS diode at 1 MHz, with P implanted Si. 190
E.27 CV measurement of a MOS diode on SrTiO3/Si02/Si at 1 kHz 190
E.28 CV measurement of a MOS diode on SrTiO3/Si02 /Si at 1 MHz. 190

List of Tables
2.1 Design Considerations 40
3.1 Methods of MIS lines mathematical analysis. 55
3.2 Derived Quantities from MIS models. 56
3.3 Comparative review of true analogue delay lines 60
4.1 Wet-Oxidation of Si at 1100°C 82
4.2 Semiconductor to Insulator Thickness Variation Effect 85
4.3 Characterisation Devices Dimensions (Am) 86
4.4 Calculated relaxation frequency of interfacial polarisation in MHz 98
4.5 Calculated controllable range of slow-wave factor 102
5.1 RIE processing parameters. 114
6.1 LowPass Characteristic Performance of MISM lines 144
xvi
Nomenclature
a Width of microstrip
(to Skin effect attenuation factor
a„,„ Semiconductor attenuation factor
as Attenuation factor of slow-wave mode
Phase constant
Os Phase constant of the slow-wave mode
X Semiconductor electron affinity
63 Skin depth
Dielectric constant
Er Complex effective dielectric constant in slow-mode

r ef f
fox Electric permittivity of oxide
enef fsktn Relative effective dielectric constant in skin mode
Er,ef f Effective relative electric permittivity
er,e f f Relative effective dielectric constant

Cr,eq Equivalent relative electric permittivity

Er Relative Dielectric Constant

ES Electric permittivity of semiconductor

Ei Electric permittivity of medium with index i
xvii
LIST OF TABLES xviii
Intrinsic impedance per unit area of substrate
• Real part of effective electric permittivity
• Complex electric permittivity
• Imaginary effective electric permittivity
Do Optical electric permittivity
fes Static electric permittivity
Sox Electric field in the oxide
Magnetic permeability of free space
Phase shift
O Work function difference between metal and semiconductor
Metal work function
Os Semiconductor work function
Electrostatic potential
B Potential difference between EF and Et
Vis Electrostatic potential of semiconductor surface
ps (x) Charge density per unit volume at x
a, Substrate conductivity for which f max is maximum
a Conductivity of medium with index i
o-„„„, Semiconductor's conductivity
tan oeq Equivalent loss tangent
• Relaxation time
Ti Time constant of RiCi circuit
rn Electron collision time

LIST OF TABLES xix
Tp Hole collision time
AC Alternating current
AFM Atomic force microscopy
C Total Capacitance per unit area
Co Velocity of light in free-space
CD Depletion capacitance per unit area
Cit Interface trapped charge capacitance
Ci Capacitance per unit area of layer i
Ci Capacitance per unit length of layer i
Cn Inversion layer capacitance
Cox Oxide capacitance per unit area
Cp Surface charge accumulation capacitance
CMOS Complementary Metal-Oxide-Semiconductor
CPW Coplanar waveguide
CV Capacitance-voltage
Dit Interface trapped charge density
Dni Normal electric flux density of layer i
Dn Electron diffusion coefficient
dox Oxide thickness
Dp Hole diffusion coefficient
d„,„ Semiconductor thickness
DC Direct current
DI De-ionised
LIST OF TABLES xx
dpa Displacement per atom
DUT Device under test
di Thickness of layer i
EF Fermi level
E9 Bandgap energy
Ei Intrinsic Fermi level
ei Transporting charges in layer i
Eni Normal electric field in layer i

Eti Transverse electric field in layer i
EC Equivalent Circuit
EL Electrical length
EM Electromagnetic
eV Electronvolt
f Frequency of applied signal
fa Onset frequency of the skin-effect mode
fmax Maximum attainable cut-off frequency of the slow-wave mode
fs Relaxation frequency of the interfacial polarisation
FDM Finite difference method
FDTD Finite-difference time-domain
FDTLM Frequency-domain TLM
FEM Finite element method
FET Field-effect transistor
GaAs Gallium Arsenide

LIST OF TABLES xxi
GD Group delay
HF Hydrofluoric
HFSS High-frequency structure simulator
IC Integrated Circuit
ISS Impedance standard substrate
Jr, Electron current density
JP Hole current density
Boltzmann's constant
Line length
LDB Debye length
L, Inductance per unit length of layer i
Ls Semiconductor inductance per unit length
LT Low-temperature
MIS Metal-Insulator-Semiconductor
MMIC Monolithic microwave integrated circuit
/11/1///14. Mode-matching method
MOL Method of lines
MOS Metal-Oxide-Semiconductor
MOSFET Metal-Oxide-Semiconductor Field-Effect Transistor
MSTL Microstrip transmission line
mT miliTorr
NA Acceptor impurity density
Ni Mobile charge density in layer i

LIST OF TABLES xxii
ni Intrinsic carrier concentration
nP Electron concentration
Pe Electronic polarisation
P,„, Ionic polarisation
Pi Interfacial polarisation
P0 Orientation polarisation
p p Carrier density of a p-type semiconductor
Ptot Total polarisation
PDE partial differential equation
PLD Pulsed laser deposition
PNA Personal network analyser
ppt parts per thousand
PV Cauchy principal value
Q Quality factor
q Electronic charge
q — TEM Quasi-Transverse Electromagnetic

Qb Bulk charge density

Qb Space charge density
Fixed oxide charge per unit area
Qf

QG Gate charge per unit area

Qtt Interface trapped charge per unit area
Q.
Charge per unit area in the metal
Mobile ionic charge per unit area
Qum
LIST OF TABLES xxiii
Qn Electron charge density
Q. Negative charges in the semiconductor per unit area

Qot Oxide trapped charge per unit area
Qp Hole charge density
Qse Space charge per unit area
Qs Charge per unit area in the semiconductor

Ri Resistance per unit length of layer i
Ri Resistance per unit length of layer i
RCA Radio corporation of America
RF Radio Frequency
RIE Reactive ion etching
RTA Rapid thermal annealing
Si Slowing factor
SAMP Switchable attenuating medium propagation
sccm Standard Cubic Centimeters per Minute
scr Space charge region
SCTL Schottky contact transmission line
SDA Semiconductor device analyser
SDA Spectral domain approach
SEM Scanning electron microscope
S/ Semi-insulating
Si Silicon
SLR Single layer reduction

LIST OF TABLES xxiv
SMA SubMiniature version A
SO/ Silicon on insulator
STWA Semiconductor travelling-wave amplifier
T Absolute temperature
tans Loss tangent
TCE Trichloroethylene
TLM Transmission line matrix
V Applied Voltage
VF B Flat-band voltage
Vg,slow Wave velocity in slow-wave mode

VG Gate voltage
Vy Wave velocity
Vox Potential across the oxide layer

VT Threshold voltage
VVC Voltage variable capacitor
W Watt
W Width of the depletion region
Wm Maximum depletion width
xi Width of inversion layer
Y Admittance
Z Impedance
Z* Complex characteristic impedance

To my parents and Maria!
Acknowledgements
First and foremost, I would like to express my sincere appreciation to my supervisor, Dr.
Christos Papavassiliou, for his guidance, inspiration, and support through the past years at
Imperial College. Conversations with him have always been stimulating and enlightening,
and are the sources of new ideas and creative thoughts.
I would also like to thank my colleagues and friends Drs. Kostis Michelakis, George
Konstantinidis and Munir Ahmad for their support in fabricating all specimens presented
in this thesis. They granted me access in their cleanroom facilities and introduced me into
the art of micro-processing techiques.
Special mention must also go to Drs. Timothy Constandinou and Pantelis Georgiou
(Ki omos teleiose!) for their continual support and advice during the course of this work.
Brain-storming during "stuff"-time has always been enlightening.
I would also like to thank my friends and colleagues for their outstanding love and
support: Amir Eftekhar, Eduardo-Aguilar Pelaez, Sofia Vatti, Franscesco Cannillo, Michalis
Frangos, Suket Singhal (the cleaner!), Carlos Frangiadakis, Theodoros Zoumpoulidis, Sanjiv
Sharma...and the list goes on! I would also like to express my gratitude to Mrs. Wiesia
Hsissen for looking after us during our PhD with such kindness and warmth!
Many thanks are also due to Mr. Phil Jones as well as Mr. Mike Harbour for their
valuable help and advices through the development of experimental structures. Moreover,
I would like to thank Dr. Stepan Lucyszyn, for the many useful discussions we had as well
as for allowing us to use their HFSS software license for simulations throughout the period
this research was conducted.
Special mention must also go to Professor Chris Toumazou for his generous support
during my PhD and all the opportunities he offered me ever since. I also wish to acknowledge
2
Acknowledgements 3
the support of the Leventis foundation as well as the EU AMICOM Network of Excellence.
Finally, this thesis would not have been possible without the love and support of my
family. I consider as a member of my family Mrs. Anna Gabriel who has been taking
good care of me (and many many others...) while in London. I am grateful to my sisters
Dimitra and Kalliopi for sharing my enthusiasm during my PhD. I should mention here that
without my grandfather's Themistoklis Prodromakis inspiration on Byzantine music, my
life would have been completely different. Also, special thanks go to my other grandfather
Dimitrios Tetradis for all his wonderful stories... And most of all, I thank my fiancee Maria
Stivachtopoulou and my parents Konstantinos and Mavra Prodromakis, for all their love,
patience and the sacrifices they made for me to be here.
Abstract
This work is a study on the waveguiding properties of laminar substrates consisting of

metal-insulator-semiconductor (MIS) layers. The propagation characteristics of MIS lines
are controlled by the resistivity of the substrate, the operating frequency and the ratio of
the semiconductor to insulator layer thicknesses. What is intriguing about MIS microstrip
structures is the very large slowing factor occurring at a band of intermediate frequen-
cies (-500 MHz). A strong interfacial polarisation (Maxwell-Wagner polarisation) taking
place at dielectric interfaces in multilayer structures is responsible for the slow propagation
mode. This phenomenon has previously been extensively examined, both theoretically and
experimentally in the context of data dispersion for high speed data buses.
Structures supporting the Maxwell-Wagner polarisation, exhibit abnormally small phase

velocities proportional to the square root of insulator to semiconductor thickness ratio. To
the lowest approximation, microstrip lines on laminar substrates, resemble microstrip lines
on high-K dielectrics. This concept has many ramifications and is the central theme behind
most of the ideas presented here.
Presently, an investigation has been undertaken, covering the important parameters that
establish slow-wave modes. Measured results on Si-based structures demonstrate that the
available theory may well be too pessimistic in some of its predictions. Specimens based
on GaAs membranes can show larger than expected slowing-factors with the attainable
dielectric losses being at lower levels than in Si, for the same predicted electric permittivity.
Additionally, a novel experimental platform has been developed, based on a macroscopic
model of the MIS microstrip, with the semiconductor layer emulated by a liquid "crystal"
whose conductivity can be varied. The physical dimensions of each layer can be easily
adjusted, which facilitates the investigation of slow-wave characteristics over a wide range
of parameters.
4
Abstract 5
Of course, any element or mechanism that causes a spatial variation of charge density,
contributes to the dielectric susceptibility of the medium. In fact, every polarisation mech-
anism may support a resonance. The Maxwell-Wagner polarisation is limited within the
low GHz region (3-5 GHz). Several processing techniques are considered which attempt to
extend the transition of the Maxwell-Wagner relaxation from the interfacial to the atomic
level. These can be divided into two categories, those that increase the roughness of the
semiconductor surface (i.e. affect the insulator-semiconductor interface) and those that in-
troduce defects in the insulating medium. Measured results demonstrate that substrates
engineered/fabricated with these methods, exhibit higher stray capacitances and are there-
fore an indication that Maxwell-Wagner polarisation can be indeed engineered.
Finally, laminar substrate materials are used for the construction of physically small
but electrically large passive devices. Potential applications of this method, such as delay
lines, distributed low-pass filters, patch antennas and couplers, alongside with the expected
benefits of these devices are examined.
Chapter 1
Introduction
1.1 Motivation
As technology moves forward we increasingly aim for devices that offer more functionality,
consume less power and have a smaller physical size. Portability has become a very impor-
tant task for engineers, especially nowadays that the need for integrating all components
into a single device is increasing more and more.
Portable devices require some type of radio frequency (RF) connectivity, which involves
passive reactive components, such as inductors, capacitors, delay lines and resonators. It
would appear that Moore's law [1] allows indefinite miniaturisation of all electronic devices.
This is not entirely true. Miniaturisation has led to a point that the area occupied by the
active part of a radio, including digital signal processing in the baseband, is small compared
to the area required for passive monolithic components. External discrete passives are often
used to alleviate the silicon area required for monolithic passives. The need for methods to
reduce the size and improve the performance of monolithic passive components is therefore
clear. The physical size of passive components is proportional to the effective wavelength
(A) which in turn is proportional to the speed of light (co ) and inversely proportional to
the frequency of the applied signal (1) and the dielectric constant (Er ) of the substrate on
which the passives are built.
co
A
f \AT-
In order to miniaturise distributed passives one needs to drastically reduce the on-
chip electromagnetic (EM) propagation velocity by engineering an artificial waveguiding
6
Introduction 7
medium. The development of high permittivity, low loss dielectrics; best known for their
introduction into the gate stack of modern metal-oxide-semiconductor field-effect transis-
tors (MOSFETs), has also enabled the production of miniaturised dielectric resonators and
filters. A high dielectric constant substrate or cover layer is an effective means of achieving
resonator miniaturisation, since the physical size of the circuit elements is inversely propor-
tional to the effective index of refraction. A sufficiently high dielectric constant substrate
will, in principle, allow adequate miniaturisation at any operating frequency. However, most
naturally occurring materials that are practically suitable, have relatively small dielectric
constants and are therefore not appropriate for obtaining significant reduction in size. For
example, titanium dioxide TiO2 with an Er ^ 85 is typically used for miniaturising passives.
Ferroelectric materials, which possess even higher dielectric constants, can be used, but at
the cost of increased sensitivity to temperature variation and very high dielectric losses.
Recent advances in micromachining have opened new possibilities for developing com-
posite materials with extremely high dielectric constants. A material, for example, recently
reported by Yang Rao et al. [2], was measured to have a high dielectric constant Er of
1000. This material does not exhibit the temperature sensitivity of ferroelectric materials.
Nevertheless, the use of composite materials for integrating active as well as passive com-
ponents in standard technologies is currently unpractical, as it involves a large number of
process steps. Additionally, "exotic" materials are usually incompatible with the chemical
cleanliness required of, for example, standard Complementary-Metal-Oxide-Semiconductor
(CMOS) processes.
The present work seeks an alternate solution to the high-dielectric constant substrate
question by investigating a band limited approach. This thesis focuses on slow-wave polar-
isation modes as a way to enhance effective dielectric constants. Such polarisation modes
naturally occur in layered structures made of materials with very different dielectric con-
stants and conductivities. Effective dielectric constants exceeding values of 1000 are at-
tainable with relatively low dielectric losses. The ideas presented here are not new [3].
Polarisation modes are known to increase bus mode separation and consequently to aug-
ment data dependent dispersion in high speed digital buses on silicon substrates [4]. To the
best of my knowledge, this is the first systematic attempt at engineering passive components
relying on slow-wave modes for their operation.
It is well known that when a wave propagates through a medium with a dielectric
constant Er , its velocity is reduced by the factor Er . It is therefore observed that the
Introduction 8
higher the dielectric constant of a material the lower the propagation velocity of a wave
travelling through it. Consequently, structures causing large delays to propagating waves
will efficiently behave as materials with very large dielectric constants. This approach is
indeed an utilisation of the slow-wave propagation occurring in highly anisotropic multilayer
structures for realising ultra-high dielectric constants in a given frequency range.
Unfortunately, nothing comes for free in nature. The Maxwell-Wagner polarisation

can be profitably used in the MHz and low GHz frequency bands, contributing to a big
increase of the effective electrical length of passives and hence minimising their required
physical dimensions. At the same time, the mechanics of this polarisation mode set a cut-
off point in the mid GHz band (3-5 GHz) that limits potential applications. Additionally,
the excessively large slowing-factors observed in the MHz region, come with analogously
large dielectric losses as the frequency increases. Therefore, the Maxwell-Wagner effect and
its potential applications have seen limited use to date.
1.2 Research Objectives
This research is aimed at exploring the Maxwell-Wagner polarisation effect and its applica-
tion in monolithic technologies. This is achieved by developing a thorough understanding of
the polarisation mechanism, validating previously reported assumptions, proposing meth-
ods that can be used to appropriately tailor this effect and, finally, by investigating potential
applications. Subsequently, this work attempts to provide some insight into answering the
following questions:
• Can Maxwell-Wagner polarisation be used for efficiently miniaturising passive com-

ponents?
• Are there any limitations in applying interfacial polarisations for miniaturisation pur-
poses and how can we overcome these?
• Are there any new applications that can benefit from this polarisation mechanism?
1.3 Overview
The dissertation is organised over 6 chapters as outlined below.

Introduction 9
1.3.1 Electromagnetic Wave Propagation on MIS Structures
Chapter 2 introduces the basic theory for analysis of MIS interconnects. The analysis is
broken into three parts, as encountered in the literature. First, the device physics approach
is considered, that ascribes the observed slowing-effect to the formation of a space-charge
layer at the insulator-semiconductor interface. In the second part, the concept of polarisabil-
ity of a medium is introduced with focus on the Maxwell-Wagner polarisation mechanism.
Finally, different propagating modes commonly used by the RF and Microwave research
community, are considered.
1.3.2 MIS Modeling and Devices
In chapter 3, a thorough literature review of the theory and application of slow modes is at-
tempted, as several models have been proposed to study MIS lines. Formulae for analyzing
MIS interconnects are established by either solving Maxwell's equations or combining them
with the motion equations of charged carriers. Limiting approximations of both approaches
lead to equivalent circuits for different propagating modes. The first part of the chapter is
a review of the available models found in the literature. The second part surveys devices
that have been built in the past, utilising the slow-wave behaviour of MIS microstrip lines.
Delay lines, variable phase shifters, the voltage tunable phase-shift oscillator, the semicon-
ductor travelling-wave amplifier, voltage-tunable MIS filters, variable coupling-coefficient
directional couplers and distributed FET transistors are briefly reviewed.
1.3.3 High-K Structures Exhibiting Maxwell-Wagner Polarisation
We focus on the slow-wave mode because in the frequency region where this mode domi-
nates MIS structures exhibit ultra-high effective dielectric constant (Er ,eff ) values. Although
this phenomenon can be very useful in many applications, large attenuation occurring in
propagating waves necessitate an assiduous study of the sources of these losses. In the
fourth chapter of this thesis, structures that support the propagation of slow-waves and
effectively behave as high-K equivalents are presented. A comprehensive comparison is con-
veyed through a large variety of samples that accommodate major parameters that establish
Maxwell-Wagner propagation characteristics. Si substrates of moderate conductivity and
GaAs membranes are authoritatively used. Finally, an experimental platform, based on
aqueous solutions, is shown to support slow-wave propagation and can be effectively used
Introduction 10
to mimic smaller systems for a great variety of parameters.
1.3.4 Engineering the Maxwell-Wagner Polarisation Effect
In chapter 5, novel substrates that were engineered by applying various processing tech-
niques in order to enhance the slow-wave mode are presented. These substrates were
optimised to meet three, partially contradictory design targets: to minimise the guide
wavelength and the loss tangent and maximise the upper cut-off frequency of the mode.
The presented results revealed that the theory available to describe the behaviour of these
substrates, may well be too pessimistic in some of its predictions.
1.3.5 Miniaturising Passive Components by Utilisation of MIS Microstrip
In chapter 6, potential applications which exploit the Maxwell-Wagner effect are proposed.
Miniaturised passive devices, such as patch antennas, distributed filters and delay lines are
presented. These devices, when built on slow-wave substrates, behave as electrically larger
devices, due to the large slowing factor of the substrate. To support all claims, simulated
data as well as measurements of developed prototypes are presented.
1.3.6 Conclusion
Important conclusions from the work presented in this thesis are summarised in chapter
7. Contributions to current research as well as suggestions for future work are included.
Undoubtedly, these preliminary results are very promising and there is considerable margin
for improvement. Potential challenges of this project are identified and we envisage to
continue working towards their solutions. Literature references are listed at the end of each
chapter.
References
[1] G. Moore, "Progress in digital integrated electronics," in Proc. IEEE International

Electron Devices Meeting, pp. 11-13, 1975.
[2] Y. Rao and C. P. Wong, "Development of novel ultra-high dielectric constant polymer
based composite," in Proc. IEEE International Symposium on Electronic Materials and
Packaging, vol. 1, pp. 335-340, May 2002.
[3] H. Hasegawa, M. Furukawa, and H. Yanai, "Properties of microstrip line on si-sio2

system," Transactions on Microwave Theory and Techniques IEEE, vol. 19, pp. 869-
881, Nov. 1971.
[4] J. P. K. Gilb and A. Balanis, "Mis slow-wave structures over a wide range of parameters,"
Transactions on Microwave Theory and Techniques IEEE, vol. 40, pp. 2148-2154, Dec.
1992.
11
Chapter 2
Electromagnetic Wave Propagation

on MISM Structures
2.1 Introduction
Microstrip lines on semiconductor-insulator substrates were introduced in the late 60's af-
ter Hyltin [1] proposed the use of high-resistivity semiconductor substrates instead of the
conventional insulating substrates. This approach enabled the monolithic integration of
passive and active devices on a semiconductor substrate. The remarkable reduction of par-
asitic capacitances, the elimination of bond wires and the reduction of mismatch losses made
microwave monolithic circuits superior to hybrid assemblies on ceramic substrates. As a
result, a significant volume of research has been carried out since on accurately modeling
propagation on microstrip and other conductor patterns defined on semiconductor-insulator
substrates (explained in more detail in chapter 3).
In recent years, there has been an increasing interest in minimising or even controlling
delays introduced by interconnects in digital integrated circuits. Surprisingly, under certain
circumstances transmission lines built on such substrates, can cause appreciable delays to
propagating signals. Such interconnects typically consist of conductor tracks on top of a
multilayer structure that comprises one or more thin insulator layers on top of a relatively
thick semiconductor substrate (figure 2.1) of modest conductivity (typically 1< o-„,„ <3000
S/m). Signals propagating on data buses made on such structures often exhibit large
"data dependent dispersion", suggesting that some bus modes have very small propagation
velocities, significantly slower than the modes of tracks printed on a homogeneous silicon
12
Electromagnetic Wave Propagation on MISM Structures 13
substrate of comparable thickness. Bus modes were studied to some depth by Gilb and his
co-workers in [2]. This "slow" behaviour of metal-insulator-semiconductor-metal (MISM)
lines and the more general "slow" nature of MISM structures, was first studied, theoretically
and experimentally, by Guckel [3] and Hasegawa et al. [4]. The system they considered was
the microstrip line on a Si02 — Si substrate.
pipmprium
Figure 2.1: Metal-Insulator-Semiconductor-Metal (MISM) structure.
Slow-wave propagation can be intuitively attributed to the finite conductivity of the

semiconducting substrate, which together with the dielectric layer permittivity has a "fil-
tering effect" on the electromagnetic fields. When certain conditions involving, the signal
frequency and the physical characteristics of the substrate are met, the electric fields in
the substrate are screened by the interface charge, while magnetic fields penetrate the sub-
strate. In other words, reactive power is exchanged between the dipole moment density at
the semiconductor-insulator interface and the magnetic energy stored in the semiconductor
layer. The sum of these energies is transmitted along the propagation direction (z-direction
as depicted in figure 2.1). The large time constant associated with this energy exchange
mechanism is responsible for a significant slow-down in wave propagation. This mechanism
is, of course, reminiscent of a resonance; nonetheless, a frequency band exists where losses
are relatively low and dispersion is insignificant.
It is perhaps interesting that this peculiar slowing phenomenon has been identified by
three separate scientific communities and as such, it has been interpreted in three different
ways. Physicists attribute this effect to the presence of free carriers at the insulator-
semiconductor interface, the RF and Microwave community describe it in terms of a propa-
gation mode (slow-wave mode), while the materials community refers to it as the Maxwell-
Wagner' polarisation mechanism. Each interpretation is valuable for it offers insight to

different aspects of this phenomenon. This chapter presents all three theoretical interpreta-
tions in a fairly unified way. The mechanics of the large slowing factor, the large attenuation
of propagating signals and the sources of these losses are unfolded as well.
2.2 Space-Charge Layer Effects at Insulator-Semiconductor

Interfaces
The metal-insulator-semiconductor (MIS) diode2 is probably the most useful device in the
study of semiconductor surfaces. Since it is directly related to most planar devices and in-
tegrated circuits, it has been extensively studied. The MIS structure was first proposed as a
voltage-variable capacitor in 1959 by Moll [5] and by Pfann and Garett [6]. Its characteris-
tics were then analyzed by Frank' [7] and Lindner [8] followed by Terman who investigated
the MIS diode in the context of a thermally oxidized silicon surface [9]. All standard text-
books on Semiconductor Physics, such as [10], [11], [12], dedicate at least a chapter on the
metal-oxide-semiconductor (MOS) diode. The keen reader can find a comprehensive and in
depth treatment in MOS Physics and Technology by Nicollian and Brews [13].
Insulator Metal
Insulator
rMetal
Ot
Semiconductor semi
Ohmic contact
(b)
Figure 2.2: (a) Perspective view of a MIS diode. (b) Cross-section of a MIS diode
The following sections highlight the characteristics of an ideal MOS diode under various
bias conditions. Then the appropriate considerations are made to include the effect of
metal-semiconductor work-function difference, interface traps and oxide charges.
'Also defined as Maxwell-Wagner-Sillars polarisation mechanism
2 The metal-insulator-semiconductor (MIS) diode is a more general class of device. Since in most exper-
imental studies the insulator has been silicon dioxide, the term metal-oxide-semiconductor (MOS) diode is
used interchangeably with MIS diode
2.2.1 Ideal MOS Diode
A cross section of a MOS diode is shown in figure 2.2, where d„ and dsemi is the thickness
of the oxide and the semiconductor respectively and V is the applied voltage on the metal
field plane. Throughout this section we consider the voltage V to be positive when the
metal plate is positively biased with respect to the ohmic contact and V is negative when
the opposite occurs.
An ideal MOS diode is defined as follows:
1. When there is no applied voltage (V=0), the energy difference between the metal work
function q and the semiconductor work function 0, is zero (flat-band condition).

Orns = (Om — Os) = — (X + 2q ±0B) = 0 (2.1)
where Om, is the work-function difference, x the semiconductor electron affinity, Eg

the bandgap, q the magnitude of electronic charge, and '013 the potential difference
between the Fermi level EF and the intrinsic Fermi level E. The plus or minus sign
holds when the semiconductor is p-type or n-type respectively.
2. The only charges that are present under any biasing conditions are those in the semi-
conductor and those at the metal surface adjacent to the oxide that are equal but of
opposite sign.
3. Under direct current (DC)-biasing conditions, there is no carrier transport through

the insulator. Or in other words, the resistivity of the insulator is infinite, or at least,
many orders of magnitude larger than that of the semiconductor.
Depending on the biasing conditions (positive or negative voltages), there exist three
distinct cases at the semiconductor surface. Here we consider the semiconductor to be p-
type. When a negative voltage (V < 0) is applied to the metal plate, the top of the valence
band bends upward and thus gets closer to the Fermi level (figure 2.3a). Since there is no
current flowing in the structure3, the Fermi level in the semiconductor remains constant.
And as the carrier density depends exponentially on the energy difference4 (E, — EF ), this
band bending causes an accumulation of majority carriers (holes) at the semiconductor
3 This holds under the assumption that the MIS diode is ideal.
4 The carrier density of a p-type semiconductor is given by pp = ?lie( E-
, E F)/ kT
where n, is the intrinsic
carrier concentration, k is Boltzmann's constant and 7' is the absolute temperature.
surface. This is called the accumulation case. The corresponding charge distribution is
shown on the right side of figure 2.3a, with Q, being the positive charge per unit area in
the semiconductor and the corresponding equal but of opposite sign charge per unit area in
the metal Qm .
IF
Semiconductor
1111.. 1e11111r-_
Metal m1COnuuctor
p-type p-type
V> 0 if
EF
V///////d
(b)
Ec.
I
E.
' EF x
Ev qNAW
(c)
Figure 2.3: Energy band diagrams and charge distributions of an ideal MIS diode in (a)
accumulation, (b) depletion, and (c) inversion cases.
When the ideal MIS diode is biased with a small positive voltage (V > 0), the energy
bands bend downward and thus the majority carriers are depleted (figure 2.3b). This is
called the depletion case. The space charge per unit area (Q„) in the semiconductor is equal
to —qNAW, where NA = rtieq1P1 3 MT is the acceptor impurity density and W is the width
of the depleted region. Increasing the applied positive voltage, results in further downward
bending of the intrinsic Fermi level Ei at the semiconductor's surface, that eventually crosses
over the Fermi level, as shown in figure 2.3c. Since the positive biased gate induces excess
negative carriers, the number of electrons5 (np) at the insulator-semiconductor interface is
larger than that of the holes, and thus the surface is inverted. This is called the inversion
case. All the additional negative charges in the semiconductor (Qn ) form a very narrow
n-type inversion layer 0 < x < x i , where xi is the width of the inversion layer. Initially,
the surface is "weakly" inverted as the electron concentration is small. However, once
strong inversion is reached, the width of the depleted region remains nearly constant at the
maximum value.
es kTln(NAIni)
Wm = 2 (2.2)
q 2NA
where es is the dielectric permittivity of the semiconductor. This occurs because a very
small band bending corresponds into a very small increase of the depletion width but in
a large increase of Qn. Therefore, under strong inversion, the charge per unit area in the
semiconductor can be summarized as:
Qs = Qn + Q s, — qNA IV,, (2.3)
Finally, it should be noted that similar results can be obtained for n-type semiconductors
with the polarity of the voltage changed.
Ideal MOS curves
The electrostatic potential is defined as zero at the bulk of the semiconductor where at
the semiconductor's surface 1/.)= os and it is called the surface potential. Using the one-
dimensional Poisson's equation, the potential can be defined as a function of distance:
c12 11) —ps(x)

(2.4)
dx2 es
where ps(x) is the charge density per unit volume at position x. When the semiconductor
is depleted to a width of W and the charge within the semiconductor is ps = —qNA the
surface potential 0, is given by:
qNA W2
Os — 2e5 (2.5)
5 The electron concentration in the semiconductor is given by n, =

nze(EF -E,)/ kT

When a voltage V is applied, it will partially appear across the oxide and across the
semiconductor.6

V = vox + (2.6)
where Vox is the potential across the oxide and is given by:
P sidox
vox — goxdox — — IQ (2.7)
fox Cox
where gox is the electric field in the oxide, cox is the dielectric permittivity of the oxide, and
C, is the oxide capacitance per unit area.
The total capacitance C is therefore a series combination of the oxide capacitance Cox
and the depletion layer capacitance CD
ox CD
C = C (2.8)
(cox + CD)
where

CD = EsIW (2.9)
Following equations 2.5, 2.6, 2.7, and 2.14 we obtain:

C 1
= (2.10)
k-iox V 1 + q2E2
c's v
Npf,d2
When the applied voltage is negative, there is an accumulation of majority carriers at

the semiconductor surface and no depletion region is formed. Thus, the total capacitance
is solely attributed of the oxide capacitance Cox = /dox (figure 2.4a). In the other
extreme, strong inversion occurs when the electron concentration at the surface is equal to
the substrate impurity concentration, i.e.:
ns = nieq(6-0B )IkT = N A (2.11)
It follows that the onset of strong inversion is when the bands are bent downward by
= 20B. Consequently, using equations 2.6 and 2.7, and bearing in mind that the
depletion layer width is fixed at Wm (equation 2.2), the metal-plate voltage at the onset of
strong inversion, also called the threshold voltage, is given by:
qNAW, -V2EANA(20B)
VT = ± + 20B (2.12)
Cox
k-,ox
6 This is only true for the ideal MOS diode, in the absence of any work function differences.
For a metal-plate voltage V > VT , the total capacitance will remain at a minimum value,
as shown in figure 2.4a, given by equation 2.13 with CD = Es/Wm ,
Eox
Crain (2.13)
d+ (cox/es)Wm
C/C
Accumulation --lpft—Depletion-0-4— inversion
1.0
-0.8
a.
U C.
-0.6

0 -20 -10 0 10 20
V (V) V (V)
(a) (b)
Figure 2.4: (a) High-frequency MIS CV curve. Inset shows the equivalent circuit for the
two layer condenser. (b) Effect of measurement frequency on the CV curve [13].
The above analysis is valid under the assumption that when the biasing voltage changes,
all the incremental charge appears at the edge of the depletion layer. This is indeed the case
when the measurement frequency is high. However, at low frequencies, minority carriers can
follow the alternating current (AC) signal and lead to charge exchange with the inversion
layer. This has a direct affect on the total capacitance in strong inversion, which for low
frequencies is equal to the capacitance of the oxide layer alone. This is demonstrated in
figure 2.4b ([10]), where several capacitance-voltage (CV) curves are plotted for different
measurement frequencies. Finally, the above analysis is valid not only for p-type but for
n-type semiconductors as well, with the appropriate changes in signs. The CV curves will
have identical shapes but will be mirror images of the ones for p-type semiconductors, with
the threshold voltage being negative.
2.2.2 The SiO2 — Si MOS Diode
The discussion in this section is applicable to all metal-insulator-semiconductor systems.

Since the metal-Si02 — Si diode is commonly used in all standard technologies and as
such is the most extensively studied, thus all considerations are directed accordingly. The
electrical behaviour of the Si02 — Si system is very close to the previously discussed ideal
MOS diode but different in an important way. The work function difference, for commonly
used metal pads, is not zero (q0nis 0 0). Additionally, charges are induced in the oxide and
at the insulator-semiconductor interface of a practical MOS diode during processing or ever
post-processing of the device. In the following sections these charges are identified and the
accompanied effects to the ideal MOS characteristics are considered.
The Work Function Difference
The work function of a semiconductor (q0s ), which is the minimum required energy to
move an electron from the Fermi energy level into vacuum level, varies with the doping
concentration. The work function of a metal varies for different metals. For example, the
work function for n+ and p+ polysilicon are 4.05 and 5.05eV, respectively. Depending on
the electrodes used and the doping concentration the work function difference (gq5,) can
vary up to 2V [10].
In the ideal case where the metal and the semiconductor are isolated, with an insulator
sandwiched between them, all the energy bands should be flat. However, in a practical
p-type MOS diode and under zero biasing, the energy bands are bent downwards due to
the non-zero work function difference. Thus, at thermal equilibrium, the metal is positively
charged while the semiconductor surface is negatively charged. In order to obtain the ideal
flat band condition, a voltage equal to qcbm,,, but of opposite sign has to be applied. This
voltage VFB is called the flat-band voltage. The effect on the flat band voltage is even
greater when charges exist in the oxide or at the insulator-semiconductor interface. Oxide
and interface charges contribute to a significant increase of VFB, since it images all of its
charges in the semiconductor.
Oxide and Interface Trapped Charges
There are four general types of charges associated with the Si02 — Si system shown in figure
2.5. They are mobile ionic charge, oxide trapped charge, fixed oxide charge and interface
trapped charge. This nomenclature was standardized in 1978. Modern deep submicron
technologies with extremely thin oxides necessitate a thorough investigation of each of these
types of charges and the effect they have on the ideal MOS diode performance.
The interface trapped charges Qtt are charges nominally located at the Si02 — Si inter-
face. They originate in structural defects, oxidation-induced defects, metal impurities, or
other defects caused by radiation or similar bond breaking processes. Orientation depen-
dence is particularly important. In the < 111 > orientation the interface trapped charge
density Dit is about an order of magnitude larger than in < 100 > orientation [14]. These
interface charges have energy states in the silicon bandgap which allow them to exchange
charges with silicon in a short time. Interface states can be charged or discharged by
applying an appropriate surface potential. These kind of charges can be neutralized by
low-temperature annealing.
e 0 Mobile Ionic Charge (Qm) Cl 0
+ Oxide Trapped Charge (Q)) +
Fixed Oxide Charge (Qf)
El E El I=1 CI CI El El El El El
gut 11, •
Figure 2.5: Charges and their location for thermally oxidized silicon [15].
During thermal oxidation, a monolayer of SiOs is sandwiched between the stoichiometric

Si02 and the single silicon crystal, as shown in figure 2.5. This layer is incompletely oxidized
silicon in which "fixed"7 oxide charges Qf may appear. The formation of this type of charge
depends on the diffusive supply of oxygen in the oxidation process. The higher the oxidation
temperature, the lower is Q. However, if a low oxidation temperature is required, Qf can
be reduced by low-temperature annealing. In order to be detected, one has to first make
sure that the structure is free of interface trapped charges.
7 These charges are called fixed as they are immobile to any applied electric field.
Oxide trapped charges Qot can consist of both positive and negative charges that are
distributed within the oxide. Trapping may result from X-ray radiation exposure or other
mechanisms. Low temperature (< 500°C) treatment seems to eliminate this type of charge.
The fourth category of charges that might exist in an oxide are the mobile ionic charges
Qin. They are primarily caused by ionic impurities such as N a+ , Li+ , K+, H+ or even by
the interaction of the oxide with heavy metals or negative ions. Qn, charges are usually
mobile within the oxide, leading to hysteresis of the threshold voltage.
Effect of oxide and interface trapped charges on MOS diode
The electrical performance of a practical MOS diode is quite close to that presented in
section 2.2.1. However, the existence of charges in the oxide as well as at the oxide's
interface with the semiconductor, pose a significant influence on the device's characteristics.
Obviously, when such charges are present, the key parameter is the identification of the
corresponding charge densities. Once these are known, the MOS diode can be characterised
rather easily, in an intuitive way.
Let us first consider the case in which interface trapped charge (Qit ) exists, with a charge
density Dit, while the oxide is free of trapped or mobile charges. The gate capacitance of a
MOS diode is defined as:
dQG
C— (2.14)
dVG
where QG is the gate charge and VG is the gate voltage. Obviously, a charge variation will
give rise to the capacitance. From charge neutrality, it holds that the charge at the gate
QG is equal to the sum of the charge appearing in the semiconductor Qs and the interface
trapped charge Qit.
QG = — (QS Qit) (2.15)
And since the gate voltage is partially dropped across the oxide and partially across the
semiconductor, it holds that:
VG =VFB +Vox + (2.16)
allowing equation 2.14 to be rewritten as:
C = d( dQ it (2.17)
dVox
The semiconductor charge density stands for the sum of the charges that are present
in the semiconductor under any biasing conditions. Qs therefore consists of a hole charge
density Q p , a space-charge-region (scr), a bulk charge density Qb, and an electron charge
density Qn . Thus, substituting Q, = Q p Qb Qr, into equation 2.17 and by use of 2.14
it yields:
c„(C p + + Cr, Cit) (2.18)
C=
Cox + Cp + Cb + + Cit
Equation 2.18 can be represented by the equivalent circuit shown in figure 2.6a. Con-
sidering a p-type MOS diode, when a negative voltage is applied at the gate, the positive
charge (Qp) that accumulates at the semiconductor's surface will dominate and as such, Cp
will be shorted. This is depicted by the track of the heavily line in figure 2.6b. Thus, the
total capacitance of the diode is quite close to the oxide's capacitance Cox.
(a)
OX
Cit T ox
Cit
Accumulation Depletion
(b) (c)
ICox
T
IC
Cit 1Cb 1C.
ox
Cb

Inversion - Low Frequency Inversion - High Frequency
(d) (e)
Figure 2.6: Capacitances of an MOS capacitor for the different bias conditions [12].
For small positive gate voltages, the semiconductor surface is depleted and as a result
the charge within the space-charge region Qb will dominate. The interface trapped charge
contributes a capacitance equal to Cit. In this case the equivalent circuit in depletion
reduces to that shown in figure 2.6c.
In weak inversion the inversion layer charge Q„, starts to pile up at the semiconductor's
surface and as the gate voltage increases, Q„ increases and eventually strong inversion oc-
curs. Q, is the dominant charge and Cr, is very high approaching a short circuit. This
is the case when the measurement frequency is low enough for Q, to follow. Figure 2.6d
illustrates the equivalent circuit reduction to a sole capacitance C,. However, when the
inversion charge is unable to follow the ac voltage, the total capacitance is a series combi-
nation of C, and Cb, as shown in figure 2.6, with Cb = cs /Wm, where Wm the inversion
space-charge region width is as in equation 2.2.
Ciot(pF)
•••
Stretch-out
due to
interface trapped
• charges
•
High Frequency Low Frequency
1 MHz —1 kHz
0
V (V)
Figure 2.7: Capacitance stretch-out due to interface trapped charges.
Most importantly, the influence of interface traps on the biasing voltage, causes a shift
of the ideal MOS C-V curve along the voltage axis. Apart from this, in the presence of
traps, the bias range for which the semiconductor's surface is depleted increases [16]. This
is due to the fact that when interface charges are present, more charges on the metal gate
are required to create a given surface potential. Therefore, provided that the frequency is
low enough for the interface traps to follow, the trap density can be found by comparing
the stretch-out on the C-V curve from low-frequency (solid line) to high-frequency (dotted
line) which is free of any traps capacitance, as shown in figure 2.7.
Oxide charges include the oxide fixed charge Q f , the oxide trapped charge Qot , and
the mobile ionic charge Qm, as shown in figure 2.5. When fixed charges are present inside
the oxide layer, electrons are attracted towards the insulator-semiconductor interface and
contribute to the development of a thin inversion layer [16]. This low resistivity layer covers
the entire silicon surface and its thickness is an increasing function of the oxide charge
density. Oxide charges cause a voltage shift of the MOS C-V curve, which indicates that
the semiconductor's surface can be inverted at zero or negative bias. The voltage shift is
measured at sufficiently high frequency with respect to an ideal MOS diode C-V curve.
Goetzberger in [17] has published a complete set of such curves. Depending on the polarity
of the charge, the voltage shift can be negative or positive for positive or negative charge
respectively, which is valid for both n-type and p-type semiconductors [11].
2.3 Polarisation in Dielectrics
The dielectric constant € of a material, is a term that refers to the material's ability to con-
centrate an electric field. As an example, the higher the dielectric constant of an insulator,
the higher the capacitance it can provide when placed between two conducting plates, or
in other words, it can store more charge for the same thickness of insulator. The dielectric
constant is related to how much a material can be polarised. When placed in an electric
field, the atoms, molecules, or ions of a dielectric, respond to the applied electric field (6').
It is actually the charges in the dielectric's atoms, molecules, or ions that will reorient
themselves in the direction of the applied field. The more responsive these charges are, the
higher the electric permittivity € of the dielectric will be.
Since polarisation is the "alignment of permanent or induced atomic or molecular dipole

moments with an externally applied electric field" [18], several sources of polarisation exist:
interfacial, electronics, ionic, and orientation polarisation. Typically, dielectric materials
exhibit at least one of these polarisation mechanisms depending on the material and the
manner of the external field application (i.e. frequency, figure 2.8). The total polarisation
of a dielectric is consequently equal to the sum of all the polarisation mechanisms the
material exhibits, such as, interfacial (Pi ), electronic (Ps ), ionic (Pi „), and orientation (P0 )
polarisation.
Ptot = Pi + Pe + Pion + Po (2.19)
8 This polarisation mechanism is also called atomic polarisation.

Ei/E0 Vibrations
Interfacial Dipoles Atoms Valence Electrons Inner Electrons
/ 1
1 6n
0 •or"."-"••••... f
Relaxations I
I Resonances
Figure 2.8: Polarisation mechanisms and their effect on the frequency dependence of the
relative electric permittivity [19].
The last three mechanisms are amenable to basic considerations and calculations and
the keen reader can find more information in standard materials textbooks, such as [18],
[20] and [21]. However, the interfacial polarisation and the distinct case of the so-called
Maxwell-Wagner polarisation mechanism is of great significance to the development of this
thesis and as such it is unfolded in the following section.
2.3.1 Maxwell-Wagner Polarisation Mechanism
The electronic, the ionic and the dipole orientation polarisation mechanisms have in common
that they are caused by the displacement or orientation of bound charge carriers. These
carriers are under the influence of the local field which depends on the total applied field and
the polarisation of the surroundings. However, the space-charge or interfacial polarisation
mechanism, is produced by traveling charge carriers. Thus, the piling up of space-charge in
a volume, or of surface charges at an interface, incorporates large-scale field patterns and
as such it greatly affects the overall polarisability of a material/structure (figure 2.8).
The classical example of interfacial polarisation is the Maxwell-Wagner two layer con-
denser that consists of an insulator of vanishing conductivity, and a semiconducting medium
that exhibits finite conductivity and vanishing electric susceptibility. In such structures, the
charge carriers can be blocked at an inner dielectric boundary layers on a mesoscopic scale,
or at the external electrodes contacting the sample (electrode polarisation) on a macroscopic
scale, as shown in figure 2.9. In both cases this leads to a separation of charges which in
turn gives rise to an additional contribution to the polarisation. The charges blocked at
internal phase boundaries (mesoscopic scale), generate the Maxwell-Wagner polarisation.

The dielectric strength of this mechanism is strongly dependent on the effective sur-
face between the phases. Above the phase transition, the phase boundaries disappear
and therefore the charges cannot be blocked anymore. However, the charges may be sepa-
rated over a considerable distance. Therefore the contribution to the dielectric loss can be
by orders of magnitude larger than the dielectric response due to molecular fluctuations.
This mechanism typically appears in inhomogeneous materials such as blends of polymers,
semicrystalline polymers, and biological samples.
Electrode Electrode
Insulator Semiconductor
eIe e i e 4. e4__.c) e.4_____.® -

eec) e + e o e<-----o -
0 e e e + 8*--0 e*----o -
ee e + 42 0 e.4________0 _
e a® ee e + e4 'De< E e-
e e + 0 0 e.----e -
00 e e o + e--e e
e e 4----- C) -
+ at- 0 0 4------ 0-
00 000 + o 00 0-
Fixed Charge
Mobile Charge Accumulated Charge
(a) (b)
Figure 2.9: Interfacial polarisation on a two layer condenser.
The simplest model to describe an inhomogeneous structure is a double layer arrange-

ment. Maxwell [22], followed by Wagner [23] and Sillars [24] have utilised this approach in
order to model the response of an inhomogenus medium to the electric perturbation. Each
layer is characterised by its electric permittivity ej, relative conductivity a, and thickness
di (figure 2.10). Following the analysis presented in Appendix A, the two layer condenser
appears to the outside observer as a single dielectric medium with a dielectric constant:
, 2
i T1 + 72 — -r T1T2T
=
W27-2) (2.20)
Co(Ri + R2)(1
where 7-1, T2 are the individual time constants of the RC circuits shown in figure 2.10, and

T being the relaxation time,

R1R2(C1 + C2) eid2 + €12 d1
T= (2.21)
R1 + R2 a1 d2 0-2d1
0 _ 0 „ 0 .; 0 - 0 - 0
0 0.
0 _ 0 0
0
) 0 0 0 0.° o0 a 0 e0 0 ° 0 0 0
0 0 0 .o 0
0 -0 0c c; 0 -0 000:
'- 0:3 0 0 s'0 C
00 0 0
i R4 T C1
d
o. 0 0 00 0 0 0 0 0 20 00 o . 0 0 o 0
0 - 0 o o o_o _0000 ° o o o 0 o 0 o D o n o
0 0
o 0 o ° 0 n 11
12 R2
Figure 2.10: Two-layer condenser and its equivalent circuit.
The electric permittivity (equation 2.20) changes from its static value (w = 0)
+ T2 — T (di + d2)(ElOidi C20 M2) (2.22)

kt =
Co(Ri R2) Co(Crid2 0-2d1)2
to its optical value (w co)

T2 1
= (2.23)
cc) Co(R1 R2 )T
Additionally, the two-layer dielectric will exhibit a dissipation factor given by:
— 1 + w2(7-7-1 + TT2 — T1T2)
(2.24)
wCo(R1 + R2)(1 + w27-2)
And the complex permittivity may thus be written (see Appendix A),
+ — 0-
(2.25)
1 +
jw'r
By observation, it is evident that the two-layer condenser experiences a relaxation spectrum
similar to the simple orientation polarisation of the Debye theory [25]. As such, the real
and imaginary components of the effective electric permittivity, given by equations 2.20 and
2.24, would be interrelated by the Kramers-Kronig relationship [26].
(w) — 1 =7r x—(L,,) dx

PV fc")o x
(2.26)
ic"(w) = —1
7T PV — 00
1
k' (x)-1
X-0)
dx
where PV denotes the Cauchy [27] principal value. The physical meaning of the Kramers-
Kronig relations is that energy storage and dissipation are two aspects of the same phe-
nomenon; hence if one is given over the whole frequency spectrum the other can be pre-
scribed. This is rather helpful in checking the reliability of measurements, however it can
be strenuous when various polarisation mechanisms are superimposed.
Conductivity term
K'
Debye term
log OJT log GOT
Figure 2.11: Relaxation spectrum of the two-layer condenser [25].
Figure 2.11 demonstrates the relaxation spectrum of the two-layer condenser. As fre-
quency increases, the interfacial polarisation begins to lag and the dielectric constant re
decreases to the midpoint between tc's and 14, when w = 1/7-. It is evident that the static
dielectric constant n's is much larger than the optical one Icio„ cx, is determined by the real
permittivities of the two media, owing to the static contributions of the electronic, atomic
and orientation polarisations. However, res is determined by the mobile charge carriers
contained in the two media and thus to their conductivities.
al = N1e1 b1
(2.27)
0-2 = N2e2b2
where Ni are the densities of the mobile charges, ei are the transporting charges and b, the
equivalent mobilities. In general, interfacial polarisation occurs when adjoining dielectrics
differ in conductivity and therefore require different voltage gradients to transconduct a
current of constant density. As a result, charges pile up at the media's interface until
constant current is established and the static conductivity results.
d
Q= ( +
d2
(2.28)
crt cr2
Finally, it is critical to investigate the dependence of the dielectric response on the shape
and orientation of the two layers. By dividing the conducting and insulating layers into n1
and n2 equal sub-layers of thicknesses Ai and 02 (figures 2.12b and 2.12c) the dielectric
response of the multi-layer condenser will be identical to that of the two-layer condenser.
In fact, substituting
d1 = niAi
(2.29)
d2 = n 2.6.2
in all previous equations results in the same total impedance. The significance of this is that
by subdividing to molecular thicknesses a continuous transition from interfacial
to atomic polarisation is theoretically viable. However, the field distortion caused by
a polarised particle depends on its shape and orientation of the applied field. Consequently,
according to Sillars [24], dispersing one medium in the other as demonstrated in figures
2.12d, 2.12e and 2.12f, will have an impact on the effective dielectric response. It is inter-
esting though that the dielectric response of this system does not depend on the absolute
particle size, but only on the mass ratio of the two media.
Conducting Sheet
Insulating Sheet

(a) (b) (c)
O 0000 Conducting 000000 Conducting

Ellipsoids
Conducting
Cylinders
O 0000

Spheres
0000 0
1 1
(d) (e) (0
Figure 2.12: Possible shapes and orientations of particles of the conducting medium [24].
2.4 Propagation modes of MIS Microstrip
A four layer Metal-Insulator-Semiconductor-Metal microstrip transmission line, similar to

the one depicted in figure 2.1, supports 3 modes [4]. The transition from one propagation
mode to the other is determined by several parameters, such as the frequency of the applied
signal as well as the electric and magnetic characteristics of the materials evolved. Figure
2.13 demonstrates the dependance of each mode on the frequency as well as the conductivity
of the substrate. And since the dominance of each mode is strongly dictated by these two
parameters, an intuitive rule of thumb is the comparison of the substrate's thickness with
the skin depth.

1
6, = (2.30)
Tfiioa2
where, f is the frequency of the applied signal, i0 the magnetic permeability of free space
and a2 is the conductivity of the substrate.
LogflAttenuationffactor) Slow-waveffactorfika)
102 102
20
16
10'
16
N
14
c 10 1 10 12
41
0
10
11
10
1 1O1
6
1-2 4
,o' 10-3 104 $6. 10. 10, 10. 101 102 1 010-7
10'
SubstratefiesistivityfiQ-m)
'
103 104 103 102 101 io°
Substratefresistivityfig2-m)
Figure 2.13: Resistivity-frequency domain chart showing different modes of propagation in

MIS microstrip.
At lower frequencies, where the semiconductor is thinner than the skin depth, the famil-
iar "Quasi-Transverse Electromagnetic (q-TEM) Mode" is supported. In this mode energy
propagates mostly through the two substrate layers and the propagation velocity is deter-
mined by the weighted average of the permittivities of the two layers. At sufficiently high
frequencies, or equivalently for high substrate conductivities, the substrate is relatively
thick compared to its skin depth and energy is again propagated within the semiconduc-
tor. Despite that, the propagated energy is now largely confined within one skin depth of
the semi-conducting substrate. This mode is usually called the "Skin-Effect Mode" and is
usually characterised by a phase velocity lower than of the q-TEM mode.
Of particular interest is the "polarisation mode" supported at intermediate frequencies9.

This mode, when compared with the q-TEM and skin effect modes, has a very low group
velocity which is approximately constant with frequency. This behaviour is attributed to
the dispersion of energy among the insulating and semiconducting layers and as such this
mode is often designated as the "Slow-Wave Mode".
9 Typically 100 MHz-3 GHz for the layer thicknesses and substrate conductivities commonly found in
conventional Silicon integrated circuits.
2.4.1 Slow-Wave Propagation Mode
At intermediate frequencies and intermediate values of substrate conductivity, the skin

depth is comparable to the thickness of the substrate and thus propagation is due to the
slow-wave mode. This mode owns its existence to a thin space-charge layer that is formed at
the insulator-semiconductor interface. These charges act as a lossy ground plane, allowing
magnetic fields to penetrate it. However, the intrinsic impedance of the substrate is given
by
7=
7 (2.31)
a2 +jw€2
and at intermediate frequencies and for relatively conductive substrates it is rather low.
Thus, the electric field lines are greatly confined within the insulating layer, as demonstrated
in figure 2.14.
• • N oovo.—
Figure 2.14: Quasi-static 6 (left) and A' (right) field distribution in a MIS microstrip.
The confinement of the electric field within the top metal layer and the insulator-
semiconductor interface, introduces a large distributed capacitance in accordance with the
small distance di separating these two boundaries.
Cs cocoa (2.32)
di
where W is the width of the microstrip. Additionally, the full penetration of magnetic
fields, implies an inductance in accordance with the large thickness of the substrate d2.
dl + d 2
L s w (2.33)
In a perfect homogeneous dielectric, where the energy is guided through the entire substrate,
the electric and magnetic fields penetrate to the same depth. That being so, for a lossless
case, the wave velocity is equal to:
1 1 w
Vg = = T3 (2.34)
Elio .N/LC
where L and C are the equivalent inductance and capacitance per unit length and 13 is the
phase constant [28]. Similarly, by utilisation of equations 2.32 and 2.33, the propagation
velocity of a wave through a microstrip line operating in the slow-wave mode region is:
1 1 1 w
Vg,slow (2.35)
Ls Cs (di+d2) VEr,ef f PO Os
P d i €0Eox O
where Os is now the phase constant of the slow-wave mode.
Clearly, the wave velocity is a function of the substrate. And as such, the combination
of the associated large capacitance and large inductance of the slow-wave mode, manifest an
abnormally low phase velocity. This phase velocity is much slower than the permittivity of
either the insulator or the substrate alone would indicate. This peculiar slowing mechanism
is consistent with the extremely large static permittivity (equation 2.20), dictated by the
Maxwell-Wagner polarisation relaxation. Since the semiconductor to insulator thickness
ratio d2 /di can be very large, the effective relative permittivity of the substrate can be
increased by several orders of magnitude, leading to slowing factors in excess of 100.
The effective dielectric constant of a microstrip line on a homogeneous dielectric sub-

strate is always less than the relative permittivity of the substrate and it increases with
frequency. A number of analytic approximations to the dispersion of the microstrip effec-
tive dielectric constant are available, for example the one given in equation 2.36 after Pozar
[29]. However, such a microstrip line only supports the dielectric mode.
6, + 1 Er — 1 (2.36)
Er'eff 2 + 2-V1 + 12d/W
Where d is the substrate's thickness, Er is the relative dielectric constant of the substrate
and W is the conductor width.
In contrast, a microstrip line on either a MIS or a Schottky structure supports all

three modes introduced earlier. At intermediate frequencies, where the slow-wave mode
dominates, the transmission line behaves as if it was built on a homogeneous substrate of
very high permittivity. And in the case where the two layers are Si and SiO2 of thicknesses
cl„ and dox , respectively, the effective dielectric constant can be approximated by:
d3i dox
fr,ox (2.37)
Er'eff dox
Based on equations 2.20 and 2.37 the effective dielectric constant of typical silicon-
silicon dioxide substrates is simulated in figure 2.15. One can observe that the greater the
anisotropy between the layers forming this structure, the larger the slowing effect will be.
Effective index of refraction frequency response with respect to aspect ratio

200
tsemi
it =10
ex
....tsemi/to.=100
,,„,,,tsemi/tox=1000
c 150
0 tsemitto..10000
"5 100
•
C
11111,,[[1,1i
t 50
0
I I [

10' 108 109 1010
Frequency (Hz)
Figure 2.15: Frequency response of effective index of refraction with respect to the
semiconductor-insulator thickness ratio.
Assuming that this is indeed the case, when applying the boundary conditions on the
dielectric's interface separating the metal strip from this synthetic high-dielectric substrate,
it follows that Dril = Dn2 (requirement of constant flux density) and Eti = Et2 (current
continuity), which can be equally expressed as:
Enl
= fr,2 (2.38)
En2 Er,1
From previous assumption it holds that Erg >> eni , which yields En1 >> Ert2 (figure 2.16);
concluding that electric field penetration in the substrate is minimal. This is in accordance
with figure 2.14 where the electric and magnetic field distributions of a MIS structure,
operating in the slow-wave mode, are illustrated.
Following [3] and [4] it is shown that the slow-wave mode is present at frequencies where
the substrate is nearly the same thickness as the skin depth and the maximum frequency
CO/ [411
I.
n2
E
t2 r,2' 112
L L
Figure 2.16: E field analysis on the interface between two dielectrics.
at which the slow-wave mode occurs is given by:

1 , 30-2
f max (2.39)
27 r 3€0Eox Pou3d2d1
An intriguing characteristic of this polarisation mode is that it exhibits a low pass

response. The upper cut-off frequency f max is constrained by the relaxation frequency of
the interfacial polarisation f, which signifies the onset of the quasi-TEM mode, and the
onset of the skin effect mode h.
0- 2 d1
= (2.40)
27r E0Eox d2
1 2
fb = (2.41)
itoo-2 4
The slow-wave mode cut-off frequency is maximised at a substrate conductivity cr and

layer thicknesses that set the two characteristic frequencies such as they satisfy:
fs = 3/2 f a (2.42)
/ 3E0fox
ac (2.43)
V itod2di
Furthermore, the low-pass response of the Maxwell-Wagner polarisation mechanism is

supplemented by the fact that the slow-wave mode exhibits significant losses at higher
frequencies. In equation 2.35 we came across the phase constant /3, of this mode. However,
this mode is not lossless and as such an attenuation factor a, will delineate this mode
together with /3g . And as demonstrated in figure 2.13, losses are particularly high in the
transition regions between modes. Therefore, the sources of these losses alongside the
fundamental trade-off between the maximum frequency of operation and the attainable
effective dielectric constant are discussed in the following section.
2.4.2 Attenuation Due to Semiconductor and Metallic Conductor
An optimist might say that the high effective dielectric constant a semiconductor-insulator
substrate may exhibit, while in slow-wave region, is the ideal scenario for miniaturising
passive components at no cost. Unfortunately, nothing comes for free in nature. The
phenomenon we have been discussing is a resonance, and we fundamentally expect the mul-
tilayer structures to exhibit high losses, especially outside the slow mode band, as demon-
strated in figures 2.13 and 2.17.
Additionally, according to equation 2.37, one might be inclined to use an insulator

with a high electric permittivity in order to maximize the overall effective dielectric con-
stant of the laminar structure. Even though this is true, a high permittivity insulator
will also increase losses. The way losses are increased for several practical insulators, such
as Si02, Si3N2, A1203, is summarized in figure 2.18. It should be noted that Si02 has
the smallest dielectric constant; therefore a microstrip line lying on Si02 would effectively
exhibit the lowest loss.
The first measurements of losses in microstrip structures employing semiconducting

substrates were carried out by G. W. Hughes in [30] followed by Hasegawa et al. in [4].
The latter work assumed that losses in MIS microstrips were entirely due to current flow in
the semiconductor layer, where the electric field is normal to the interface, often related to
as eddy currents. Eddy currents are caused when a time-varying magnetic field intersects
a conductor, or vice-versa. This motion causes a circulating flow of electrons, or current,
within the conductor. These circulating eddies of current create in turn electromagnets
with magnetic fields that oppose the effect of the applied magnetic field. Nevertheless,
longitudinal as well as transverse components of the conduction currents are not the only
source of attenuation. Conductor losses at the top metal as well as the ground metallization
plane need to be considered as well.
When losses are small, it can be assumed that both semiconductor and conductor at-
tenuation components are additive. In this case attenuation due to the skin effect can be
Substrate's Conductivity asi=20 S/cm

4500 103
4000
3500 102
3000
-o
: 2500
101-9.
as
a2, 2000
a)
1500
10°
1000
2
tsi/tox=1000
500 - - - tsi/tox=500
10-1
108 109 101 °
Frequency (Hz)
Figure 2.17: Frequency response of relative effective dielectric constant (left axis) and at-
tenuation per wavelength (right axis) for substrate's conductivities of usi = 20 S/cm. Inset:
equivalent circuit of the two layer lossy condenser [25].
calculated using the method proposed by Wheeler [31]. Otherwise, analytical models, such
as those summarized in Table 3.2, have to be used.
It is worth noting that both semiconductor (a„,„), and the skin effect losses (as), vary
with frequency, and are proportional to f 2 and f 1/2, respectively.
as = aserni(f/fs)2 ao(f/h)2 (2.44)
where,
+ d2)
aserni = 7rfs lioeoEox (2.45)
2 (di + d2)
ab = -7rf(5 POE0cox (2.46)
3 d1
Substituting 2.45 and 2.46 in 2.44, it follows:
(di + d2) 1 2
as = 7r.f2 itocgox + —) (2.47)
d1 3fd
GSi=100 S/m
40
••
35
l•
41"
"T-
30 •
•
E
E 25
4t
••
.32' 20 .4
•
•
0 ••
ffi 15
•
••
a) 1 0 • ® SO2
•
•
5 S i 3Ni 2
A 203

0 , A -
-5 6 7 109
111
5 10 108
10 10
Frequency (Hz)
Figure 2.18: Simulated attenuation for different insulators.
which justifies the relation given in 2.42 for maximizing the slow-wave bandwidth while
achieving minimal losses. At low frequencies, the metal strip losses dominate, while at high
frequencies the semiconductor losses dominate.
Finally, an auxiliary though simplistic treatment of the losses is to consider that the only
losses present in this system are due to collisions of the material's atoms during polarisation.
Dielectric dissipation is described by the imaginary part of the complex effective permittivity
of the medium (equation 2.24) and a direct measure of these losses is via the loss tangent
tar- S of this artificial medium. Figure 2.19 demonstrates the frequency response of tans,
for several semiconductor to insulator thicknesses as well as for different values of substrate
conductivity. Contrary to intuition, this figure concludes that losses are relatively low when
the substrate's conductivity is high as well as when the semiconductor to insulator thickness
ratio is low (for example t32/t„,=1000 with us,=2781 S/m).
./tox =1000 with a =1284 S/m t ./t =1000 with a =100 S/m
t si si si ox Si
• t It =500 with CY =1284 S/m t ./tox =1000 with a =1000 S/m
si ox si Si
............. ., t ./t =100 with a .=1284 S/m t ./t =1000 with a =2781 S/m
si ox si si ox si
1
0.9
0.8
0.7
rc, 0.6
C
ct5 0.5
0.4
0.3
0.2
0.1
0 1 010
108 109
Frequency (Hz)
Figure 2.19: Loss tangent frequency response for several semiconductor-insulator thickness
ratios as well as different substrate conductivities.
2.5 Summary
A microstrip line in either a MIS or a Schottky structure supports three different propaga-
tion modes. At intermediate frequencies, where the slow-wave mode dominates, the phase
velocity is much lower than the velocity of a microstip on a homogeneous substrate. This
is due to the Maxwell-Wagner interfacial polarisation mechanism, caused by the piling up
of interface charge at the semiconductor-dielectric interface. This polarisation mechanism
typically occurs in heterogeneous structures and its dielectric contribution depends on sev-
eral parameters such as the frequency of the applied signal, the semiconductor to insulator
thickness ratio, the conductivity of the substrate as well as the electric permittivity of the
insulator. Table 2.1 summarizes the effect of each parameter on the attainable slowing
factor, the dielectric dissipation factor as well as the bandwidth of the slow-wave mode.
Microstrip lines or, more generally, passive devices on laminar substrates can be engi-
Table 2.1: Design Considerations

Slowing Factor Dielectric Loss Bandwidth
d2/di I I 1
(72 I I r
6, I T I
neered to be electrically large. Given that oxides are typically several orders of magnitude
thinner than the semiconductor substrate, devices utilising this approach, can be electri-
cally 2-3 orders of magnitude larger than their physical dimensions. However, dispersion
associated with the slow-wave effect suggest significant phase distortion in the received wave-
form. In summary, this approach is a band-limited solution to the large effective electric
permittivity problem.
References
[1] T. M. Hyltin, "Microstrip transmission on semiconductor dielectrics," Transactions on

Microwave Theory and Techniques IEEE, vol. 13, pp. 777-781, Nov. 1965.
[2] J. P. K. Gilb and A. Balanis, "Mis slow-wave structures over a wide range of parame-
ters," Transactions on Microwave Theory and Techniques IEEE, vol. 40, pp. 2148-2154,
Dec. 1992.
[3] H. Guckel, P. A. Brennan, and I. Palocz, "A parallel-plate waveguide approach to

microminiaturized, planar transmission lines for integrated circuits," Transactions on
Microwave Theory and Techniques IEEE, vol. 15, pp. 468-476, Aug. 1967.

881, Nov. 1971.
[5] J. L. Moll, "Variable capacitance with large capacity change," in Proc. IEEE Interna-
tional Symposium on Electronic Materials and Packaging, vol. 3, p. 32, 1959.
[6] W. G. Pfann and C. G. B. Garrett, "Semiconductor varactor using space-charge layers,"

in Proc. IRE, vol. 47, pp. 2011-2012, 1959.
[7] D. R. Frankl, "Some effects of material parameters on the design of surface space-charge
varactors," Solid State Electronics, vol. 2, p. 71, 1961.
[8] R. Lindner, "Semiconductor surface varactor," The Bell System Technical Journal,
vol. 41, pp. 803-831, 1962.
[9] L. M. Terman, "An investigation of surface states at a silicon/silicon dioxide interface

employing metal-oxide-silicon diodes," Solid State Electronics, vol. 5, no. 285, 1962.
[10] S. M. Sze, Semiconductor Devices Physics and Technology. John Wiley and Sons, 2001.
41
REFERENCES 42
[11] S. M. Sze, Physics of Semiconductor Devices. John Wiley and Sons, 1981.
[12] D. K. Schroder, Semiconductor Material and Device Characterization. John Wiley and
Sons, 2006.
[13] E. H. Nicollian and J. R. Brews, MOS(Metal Oxide Semiconductor) Physics and Tech-
nology. John Wiley and Sons, 2003.
[14] B. E. Deal, M. Sklar, A. S. Grove, and E. H. Snow, "Characteristics of the surface-state

charge (qss ) of thermally oxidized silicon," Journal of Electrochemical Society, vol. 114,
p. 266, 1967.
[15] B. E. Deal, "Standardized terminology for oxide charge associated with thermally ox-
idized silicon," Transactions on Electron Devices IEEE, vol. ED-27, p. 606, 1980.
[16] D. Lederer and J. P. Raskin, "Substrate loss mechanisms for microstrip and cpw trans-
mission lines on lossy silicon wafers," Solid State Electronics, vol. 47, pp. 1927-1936,
June 2003.
[17] A. Goetzberger, "Ideal mos curves for silicon," The Bell System Technical Journal,
vol. 45, pp. 1097-1122, 1966.
[18] W. D. Callister, Materials Science and Enineering An Introduction. John Wiley and
Sons, 2003.
[19] G. J. Papaioannou, "Dielectric charging." 2nd AMICOM Summer School, Romania,

Sept. 2005.
[20] D. R. Askeland and P. P. Phule, The Science and Engineering of Materials. Thomson,
2006.
[21] H. Foll, "Electronic materials." http I Iwunv .t f .uni

kiel.delmatwis I amat I elmate n I index.html.
[22] J. C. Maxwell, A Treatise on Electricity and Magnetism, vol. 1. Clarendon Press, 1891.
[23] K. W. Wagner, "Erklarung der dielektrischen nachwirkungsvorgange auf grund

maxwellscher vorstellungen," Arc. Elektrotech., vol. 2, no. 9, pp. 371-387, 1914.
[24] R. W. Sillars, "The properties of a dielectric containing semi-conducting particles of

various shapes," Journal IEE, vol. 80, p. 378, 1937.
REFERENCES 43
[25] A. R. V. Hippel, Dielectrics and Waves. The MIT Press, 1954.
[26] H. A. Kramers, "La diffusion de la lumiere par les atomes," Transactions of Volta
Centenary Congress, Atti Cong. Intern. Fisica, vol. 2, pp. 545-557, 1927.
[27] L. D. Landau and E. M. Lifschitz, Electrodynamics of Continuous Media,. Pergamon

Press, 1984.
[28] J. D. Kraus and D. A. Fleisch, Electromagnetics with Applications. McGraw-Hill, 1999.
[29] D. Pozar, Microwave Engineering. John Wiley and Sons, 2005.
[30] G. W. Hughes and R. M. White, "Microwave properties of nonlinear mis and schottky-
barrier microstrip," Transactions on Electron Devices IEEE, vol. 22, pp. 945-956, Oct.
1975.
[31] H. A. Wheeler, "Formulas for the skin effect," Proc. IRE IEEE, vol. 30, pp. 412-424,
Sept. 1942.
Chapter 3
MIS Modeling and Devices
3.1 Introduction
MIS structures were introduced in the late 60's and soon became a mature technology
since they were heavily used in Monolithic Microwave Integrated Circuit's (MMICs) ever
since. Reasonably enough, diverse approaches for analyzing MIS structures abound in
the literature. The slow-wave mode, supported in MIS structures, was regarded in most
of the earlier studies (until the mid 80's) as a limitation which needed to be avoided by
application of appropriate design considerations after meticulous modeling. Eventually it
was understood that, what used to be a limiting factor to overall system performance,
appeared as the holy grail for miniaturisation purposes. The Maxwell-Wagner polarisation
mechanism and its slowing effect in propagation velocity was exploited (chapter 6), to realise
abnormally large delay elements as well as compact passive components.
This chapter is a survey on MIS modeling and devices utilising the excessive slow prop-
agation of such structures. A synopsis of different modeling approaches is given which
enables an assiduous appreciation of Maxwell-Wagner characteristics. In the second half
of this chapter, numerous devices that turn into advantage this slowing mechanism are re-
ported; such as the voltage variable capacitor, delay lines, the electronically variable phase
shifter, microwave filters, variable-coupling coefficient directional couplers and distributed
field-effect transistors (FETs).
44
MIS Modeling and Devices 45
3.2 History of the MIS Lines Mathematical Analysis
Slow-wave propagation in MIS and Schottky-contact interconnects has been studied in

depth using several analytical models [1]—[41]. Despite the fact that these structures can
be very useful in many applications, the energy dissipation in both semiconductor layers
and conductor lines can have a significant impact on their performance. Therefore, in order
to understand the mechanisms behind various effects such as the slow-wave effect, loss and
dispersion, it is necessary to accurately simulate the MIS interconnect structures.
There is a plethora of papers presenting different approaches for analysing MIS struc-
tures, either with greater accuracy in calculations or less computation time required. It is
my intention to briefly review some of the major approaches in order to build up the reader's
comprehension to this subject. Finally, it should be pointed that although all approaches
use different assumptions for simplifying the analysis of the problem, they all make use of
the Maxwellian equations for solving the electromagnetic fields in the media, which in a
homogeneous medium take the following form:
il aH(r,t)
V x E(r,t)= (3.1)
at
aE(r
V x H(r,t) = , t) J(r,t) (3.2)
6 at
c
V • E(r, t) = p(r,t) (3.3)
V • H(r,t) 0 (3.4)
3.2.1 Empirical Lumped Circuit Models
Analytical and empirical lumped circuit models are based on the assumption that all the
conditions for slow-wave propagation are satisfied. These models are typically not applicable
to either the skin-effect nor the q-TEM mode and are restricted to rather simple applications.
However, such models, provide fast calculation and are adequate for comparing potential
characteristics of MIS lines. Nevertheless, such models can be also used to initialise more
powerful numerical studies.
A typical example of an empirical lumped circuit model, describing solely the slow-wave
mode, is given in [1]. Tedjini et al. in [2] used polynomial functions to describe the effective
electric permittivity, the attenuation factor as well as the characteristic impedance of a MIS
microstrip line. These empirical formulae are valid in both the slow-wave mode and the
skin-effect mode. However, as frequency increases, the accuracy falls off when compared
with full wave analyses and correction factors are required. In [3], an equivalent circuit (EC)
model has been developed based on the power and magnetic energy conservation laws. The
substrate effect is accounted within the elements of the EC by considering mutual inductance
with other lines. Finally, in [4], the authors claim that they have developed a transmission
line circuit model which is valid over all three regions. This model has its origin on the
parallel plate waveguide structure analysis, under the assumption that the microstrip is
quite wide.
3.2.2 Parallel-Plate Wave Guide Approach
Guckel et al. [5] and later Ho and Mullick [6], first investigated the MIS transmission line
based on a one-dimensional analysis. This approach does not account for the finite width of
the top conductor; instead it assumes all layers of the MIS structure having equal widths, i.e.
forming a parallel-plate waveguide. Under this approximation, any fringing fields occurring
at the edges of the microstrip are neglected, which in turn results in a simplified problem
to solve. Additionally, the driving impedance port is defined as the input impedance of an
infinitely long structure. Clearly, this method can be used as a general rule-of-thumb but
significant divergence from the actual solution must be expected due to the assumptions
that were made.
In 1971, Hasegawa in [7] and [8] extended Guckel's work by extrapolating equivalent
circuits for each propagating mode (figure 3.1). Clearly, the low frequency equivalent circuit
is very similar to the two-layer condenser equivalent circuit proposed by Von Hipel in [42].
However, these circuits are valid under the assumption of very wide microstrips without
accounting any conductor loss. It should be noted here that Hasegawa's predecessors, Ho
and Mullick in [6], had qualitatively described the contribution of the conductor's width on
the conductor losses as well as in the overall delay the line exhibits.
In 1976, D. Jager [9] focused on the slow-wave mode region and demonstrated that
the attenuation due to series losses in the semiconductor is proportional to the square of
frequency. Additionally, he defined the frequency limits in which the slow-wave mode is
dominant, i.e. upper frequency limit due to the interfacial polarisation and lower frequency
limit due to the finite conductivity of the metallic conductors. Jager deviated significantly
from his predecessors by proposing an equivalent-circuit model (figure 3.2) which helped
1:1=2motto2ds,2usi(d..+dsi)f/3W
L=no(d +dsi)/W RS
wottu/2W
L=tto(dox+d,i)/W L=tio(dox+dsi)/W
Cl-WEr,oxIdox CI=WErox /dox
C1=WErmidox
CS WErs1
/d si cr51.
=dSI./W Rs=d5i/Wcysi
• • •
• •
Ax Ax
Dielectricfi)uasi-TEMfrnode Slow-Wavefmode Skin-effectfmode
Figure 3.1: MIS equivalent circuits for q-TEM, slow-wave and skin-effect mode, proposed
by Hasegawa [7].
him to demonstrate the significance imperfect conductors have on slow-wave mode domi-
nation. However, the effect of fringing fields is still omitted, as his analysis is based on the
assumption of very wide microstrips.
Zmetal=2ffmetalcothynietaldmetai/W
L=Rodsi/W . 1
• %At •
Fts=3/0y(ds,-ddep,e.„„,) C=WEr,oiddepletion
ddepletion/Wosi
14 Ax
Figure 3.2: Jager's slow-wave equivalent circuit [9].
More recently, a unified equivalent circuit was presented in [10] describing MIS lines over
their entire range of operating conditions. The series impedance and admittance of this
one-dimensional model are determined via the surface impedance and by utilisation of the
parallel plate capacitance model respectively. William's model is based on the assumption of
a single mode propagation (TM° mode), and the multimodal character of MIS transmission
lines is attributed in its high-loss region of operation.
3.2.3 Full-wave Analysis
In the early 90's, as the microprocessor clock speeds were increasing, the limitation of low
data transfer speeds due to interconnects on data buses had to be addressed. The need for
more accurate models for computing delays and losses necessitated the improvement of the
previous analysis methods. In 1981, Kennis et al. [11] conducted the most rigorous to that
day analysis, by aid of the Spectral Domain Approach (SDA), explained in [43]. Since SDA
is mainly used in calculating the propagation characteristics of lossless structures, Kennis
dealt only with the bulk losses and ignored losses associated with the conductors.
On the same track, Fukuoka in [12] achieved more accurate results, that were compared
with measurements, by assuming twice the depth of the doped region, since the semiconduc-
tor is not uniformly doped. However, the metallic strips were assumed to be infinitesimally
thin perfect conductors which implies that the metallic losses were not accounted for. On
the other hand, in the same communication, this problem was solved by application of the
Mode-Matching Method (MMM). This method utilises the eigenfunction expansion of the
field. The structure is subdivided into six regions and each region is characterised by a
complex dielectric constant [13]-[14].
A notable improvement in MIS modeling came with the employment of Green's function
for solving inhomogeneous differential equations subject to boundary conditions. Livernois
and Katehi in [15] and [16] proposed a method for deriving Green's function in the space
domain, for an arbitrarily oriented current in a rectangular waveguide which consists of a
number of lossy, isotropic dielectric layers in an inhomogeneous manner. In [17], Liou et
al. derived a correction term for the impedance Green's function that represents conductor
losses. This term is then incorporated in the SDA calculation and more accurate results,
when compared with [11], can be obtained. Similarly, in 1992, Gilb et al. [18] proposed
a new model for calculating the MIS slow-wave characteristics valid over a wide range of
parameters. His full wave analysis was again based on SDA, incorporating Green's function
for structures with multiple substrates. His model successfully confirmed the expectations
that "even" slow-wave modes will result in a higher ET,ef f than "odd" modes, due to the
higher field concentration in the lower substrate. Measurements supporting their results
were made on Si-SiO2 as well as on GaAs configurations.
A remarkable full-wave analysis was presented by Wu in [19], in which non-uniform

doping profiles were considered. Additionally, the conductor is treated as a lossy dielectric
layer of high conductivity and hence is not based on any approximation. Simulation re-
sults obtained with the aid of the Method of Lines (MOL) indicate attractive propagation
characteristics, even at high frequencies.

L M/2 3
L AL/2 L AL/2
2 4
L At/2
1 = 2 CM
y
• •
►z
Figure 3.3: Equivalent lumped-element model of a two-dimensional lossless TLM network
[20].
A different flavour of analysis method, namely the Transmission-Line Matrix (TLM)

method, was proposed in 1985 by Hoefer [20] which is applicable to the MIS propagation
problem. TLM method represents wave propagation in time domain by simulating two- as
well as three-dimensional propagation of delta function impulses on Cartesian meshes of
two- or three-wire transmission lines respectively (see figure 3.3). Electric and magnetic
boundaries are represented by short and open circuits where lossy boundaries can be ac-
counted by setting the appropriate reflection coefficients. Moreover, a more advanced model
has been presented in [21], based on the frequency-domain TLM (FDTLM) method, where
inhomogeneous doping profiles have been implemented. The discrete nature of the TLM
method can cause irregularities which can however be prevailed by adaptive meshing. In
general, TLM method illustrates slightly better accuracy than conventional finite element
and finite difference methods, nonetheless its main drawback is the increased computational
expenditure due to its complexity.
A significant breakthrough in the history of analytical methods is with no doubt the

Finite Element Method (FEM). FEM is used for finding approximate solutions of partial
differential equations (PDE) as well as of integral equations. The solution approach is
based on approximating continuous quantities as a set of quantities at discrete points,
often regularly spaced into a so-called grid or mesh [22]. Aubourg [23] benefited from this
numerical technique and utilised it in modeling MIS microstrip lines. The advantage of
this approach lies with the ability to subdivide the cross section of the waveguide into
triangular regions. Then by use of polynomials the solution is approximated. Also, this
method permits the thickness of the strip to be taken into account. FEM might work well
for layered media but in practice, devices such as Schottky contact microstrip or coplanar
waveguides (CPW) can not be modeled as laminar structures. Tzuang and Itoh [24] appear
to overcome this limitation of FEM by application of "localized models". These models take
into account any localized depletion regions in the semiconductor and in conjunction with
higher order elements result into greater accuracy. Additionally, Kucera and Gutmann
in [25] have extended previous FEM based methods by including inhomogeneous doping
profiles as well as conductors with finite thicknesses and conductivity.
An alternative analysis approach to the FEM is the Finite Difference Method (FDM).
While FEM is an approximation to the differential equation solution, FDM is an approxima-
tion to the differential equation. The main drawback of FDM is its restriction in handling
rectangular shapes and simple alterations thereof. However, FDM is very attractive for its
ease of implementation [44]. In [26], Shibata et al. utilised FDM for exploring the propaga-
tion characteristics of MIS CPW. They considered both the semiconducting effect as well
as the conductor losses and paid particular attention to the transition regions between the
distinct propagation mode. Their analysis was summarized in the form of a quasi-static
equivalent circuit model.
3.2.4 Quasi-Static Modeling
An attempt has been made to decrease computation time required for modeling MIS struc-
tures by using quasi-static analysis instead of the non trivial full wave approach. In quasi-
static analysis the original vector field problem is reduced to a scalar potential problem.
This approach is preferred for analyzing low loss MMIC structures and there is generally
good agreement with results obtained by full-wave models. However, since the propagation
modes within an inhomogeneous structure are hybrid, quasi-static approaches are generally
used in limited cases. Hasegawa and Seki [27] have shown that common lumped-capacitance
models for interconnections are not adequate when switching speeds are less than 100 ps.
Moreover, it is reported in [28] that the validity of the quasi-TEM assumption holds for
switching speeds below 7ps.
In general, under the assumption that the transverse dimensions are much smaller than
a wavelength, several noteworthy analyses have been reported. In [29], Kwon et al. inves-
tigated the effect of conductor losses as well as longitudinal currents due to the interaction
between the magnetic field and the doped semiconductor by means of an equivalent cir-
cuit. Theoretical predictions are correlated with measurements from 1 GHz to 12.4 GHz.
MIS propagation characteristics of CPWs have been similarly explored in [30] and the re-
sults are compared with those obtained by application of full-wave approaches. Mesa and
his coworkers conclude that as the substrate's conductivity rises the obtained results can
deviate significantly from those obtained utilising a full-wave approach.
The solution to this was partially given in Tuncer's quasi-static model [31]. In order
to evaluate the impact of a semiconductor layer of conductivity a on the transmission
line, changes in both the electric and magnetic fields should be determined. Depending
on the operating frequency, the total impedance per unit length for the microstrip can be
calculated. In contrast, Mesa did not fully consider the impact of the semiconductor on Z.
Tuncer in [31] and more analytically in [32] assured that the quasi-statics account for changes
in both Y and Z. And as such, the results obtained by estimating the semiconductor's
conductivity impact on the transmission line are in the vicinity of the results obtained by
the tedious full-wave approach.
3.2.5 Single Layer Reduction Process
The single Layer Reduction (SLR) process [33], as its name proclaims, converts a multi-
layer microstrip structure into a single layer one, where the equivalent single-layer substrate
has a relative permittivity (€,,,q ) and a loss tangent (tan Jeq) . The SLR model is able to
handle the multilayer substrates, and the effect of the top conductor on the propagation
characteristics. This is accomplished in two steps: first by converting the inhomogeneous
multilayer substrate into a homogeneous one utilising Wheeler's transformation and then
by converting the homogeneous medium microstrip into a standard microstrip in the inho-
mogeneous medium. Then the obtained single layer Er ,eq and tan Seq will account for the
resistivity of the substrate as well as the thickness of each layer in the original multilayer
structure. However, this approach is only valid in the quasi-TEM and slow-wave modes.
An improved SLR method which is valid for all three modes of propagation is pre-
sented in [33]. According to the substrate's conductivity, three regions are defined on which
different assumptions are made. These estimations result in more realistic results when
compared with the standard SLR method. Furthermore, an updated version of Jager's
slow-wave equivalent circuit is presented which also accounts for the finite width of the
microstrip (figure 3.4). The resistance R5 accounts for any loss in the semiconductor due
to a longitudinal current. The R, L, C and G equivalent circuit components are defined
through the SLR formulations. All the required expressions for computing the primary line
constants values can be found in [33]. Finally, the improved SLR approach might be valid
for all three modes of propagation but unfortunately, it does not explain the dispersion
behaviour of a slow-wave microstrip at lower frequencies.
Fts=1 /o2d2W
02 :conductivity of substrate
d2 :substrate thickness
W :width of the conductor
Z :characteristic impedance of the multilayer
slow-wave microstrip line
Vereff :s I owing-factor
C :free-space velocity
o, :conductivity of the strip conductor
t :conductor thickness
Ax a :conductor loss
Figure 3.4: SLR based circuit model of multilayer slow-wave microstrip line.
3.2.6 Device-Level Simulation
All mathematical models presented so far, have something in common. They all make use of
Maxwell's equations (3.1-3.4) for describing the fields' distributions and the actual solution
is given after some approximations are made. However, MIS propagation characteristics
are substantially affected by the screening effect of the carriers in the semiconductor that
prohibit the electric field from penetrating deep into the semiconductor. For that reason,
the coupled electromagnetic and device simulation models need to include the interaction
mechanism between the electromagnetic field and the carriers in the semiconductor.
Several transport-based analyses [34] exist that tackle layered problems. The first device-
level analysis with direct application to MIS transmission line was presented in 1989 by
Krowne and Tait [35]. Despite the fact that this kind of analysis can be rather rigorous,
it proves extremely powerful. Since the electron mobility varies with the applied electric
field, negative differential conductance effects can be detected. Thus, it is possible for the
attenuation constant of the substrate to be negative and thereby denoting distributed gain
for the structure. This is a unique feature that conventional full-wave analyses do not
possess and that it is not clear from measurements, since ohmic loss usually prevails over
any gain contribution.
In [36], a biased parallel-plate MIS waveguide was investigated using a transport-based

analysis. A formulation incorporating Maxwell's equations and the equations of motion of
the carriers was solved using the 1-D finite-difference scheme. A 3-D combined electromag-
netic and device simulation model was presented in [37] and [38] to study the performance
of microwave devices. Maxwell's equations in conjunction with a 3-D hydrodynamic model
are solved using the Finite-Difference Time-Domain (FDTD) method. This approach in-
cludes the interactions of the conducting carriers with the electromagnetic wave. However,
this approach has been found to be computationally intensive.
For a device-level simulation model to be accurate enough, it should provide knowledge

on the following:
• field-carrier interactions
• screening effect of charged carriers
• slow-wave effect
• semiconductor substrate loss
• nonlinearity
• external bias effect
In [39, 40], the motion equations of charged carriers (equations 3.5-3.6) alongside with
Maxwell's equations are solved simultaneously.
8J„(r,t)
Tn Jn(r, t) = qpn(r,t)E(r,t) + qD,Vn(r,t) (3.5)
at
aJ p( ,t)
Jp(T,t) = qp,p(r,t)E(r,t) + qDpVp(r,t) (3.6)
P at
where q is the elementary charge, J„ and J p are the electron and hole current densities, ittri
and pp are the effective carrier mobilities of electrons and holes, Dn and D p are diffusion
coefficients for electrons and holes, and rn and Tp are the average collision times of electrons
and holes respectively.
By virtue of the finite element scheme and multi-dimensional Newton's method, the wave
propagation characteristics, electromagnetic fields, and carrier concentrations can be solved
using the frequency domain formulae under appropriately specified boundary conditions.
It should be noted that this method preserves the nonlinearity in the motion equation of
charged carriers and therefore, it is suitable for small as well as large signal analysis, which
used to be a drawback of all previously presented models. Additionally, an equivalent circuit
for MIS interconnects has been presented in [39, 41] based on this approach (figure 3.5).
Rox
+Rs Lox+LS
C G
T
S S
• Ax
Figure 3.5: Gaofeng's equivalent circuit model for MIS interconnects.
In figure 3.5, circuit elements Ro, Lo, and Co are, respectively, the resistance, inductance
and capacitance per unit length due to electromagnetic field in the insulator layer, whereas
circuit elements R3 , L3 , C8 , and C3 denote, respectively, the resistance, inductance, capaci-
tance, and conductance per unit length due to the electromagnetic field in the semiconductor
substrate. Furthermore, circuit element R, denotes the resistance per unit length over the
signal line.
saa!Aaa puu 2uHapow SIJAJ
Method Description References

Empirical lumped circuit models Usually applicable only within the slow-wave mode region. [1]—[4]
The analytical lumped circuit models provide fast calcula-
tion and first-hand insight to the performance of MIS inter-
connects.
Parallel-plate waveguide approach One-dimensional analysis. Does not account for the finite [5]—[10]
width of the conductor. Assumes an infinitely long struc-
ture.
Full-wave analysis 2-D Analysis. Spectral Domain Approach (SDA). Green's [11]—[26].
functions. Method of lines. Transmission Line Matrix
(TLM) Method. Finite-difference time-domain (FDTD)
method. Finite Element method.
Quasi-static modelling Preferred for low-loss structures. [27]—[32]
Single layer reduction (SLR) process Converts a multilayer structure into a single layer one with [33]
Er,eq and tan (5eq that account for the substrates resistivity
as well as the thickness of each layer. Valid only in the

dielectric and slow-wave modes.
Device-level simulation The wave propagation characteristics, the EM fields and the [34]—[41]
carrier concentrations are solved under appropriate bound-
ary conditions.
Table 3.1: Methods of MIS lines mathematical analysis.

saoinaa puv 2unapomi sin'
Method Primary Information Derived Quantities

Parallel-plate waveguide models E and H field. Line voltage and line current.
Propagation constant, attenuation factor and slow-wave fac-
tor.
Characteristic impedance.
Current distribution in semiconductor layer.
Equivalent circuit.
Planar multi-layered multi conduc- E and H field. Line voltages and line currents.
tor transmission line models Scattering parameters.
Characteristic impedance matrix.
Impedance matrix or admittance matrix.
Equivalent multi-port circuit network.
Combined electromagnetic and de- E and H field. Line voltages and line currents.
vice simulation models Electron and hole Scattering parameters of multi-port circuit network.
concentrations. Characteristic impedance matrix.
Impedance matrix or admittance matrix.

Equivalent multi-port circuit network.
Table 3.2: Derived Quantities from MIS models.

3.3 Devices Utilising MIS Microstrip
There are numerous planar microwave devices that can benefit of the variable phase velocity
and the slow-wave properties of MIS microstrip. In the following sections the usefulness
of MIS structures in microwave engineering is illustrated by reviewing devices found in the
literature that utilise the Maxwell-Wagner polarisation mechanism.
3.3.1 Voltage-Variable Capacitor
Reasonably enough, the first device that was exploited, utilising MIS structures, was the
Voltage-Variable Capacitor (VVC). A VVC, also known as a varactor, variable capacitance
diode or varacap, is a semiconductor device characterised by a voltage sensitive capacitance
which resides in the space charge region at the surface of a semiconductor bounded by an
insulating layer. And as a matter of fact, all other applications that follow, rely on this
mechanism. VVCs are extremely popular devices and as such there is a plethora of papers
describing such devices. An excellent report on VVCs can be found in [45] by Mortenson.
However the main contributors have to be briefly outlined in the following paragraph.
Pfann et al. [46] first described the use of a MIS structure as a VVC. Lindner [47]
followed by suggesting its use as a tuning element. Lindner's approach, however, was in
a sense limited, as the achievable capacitance range was restricted due to the inversion of
the structure, even for small biasing voltages. This was resolved by Moll in [48], where the
Si02 insulator was stressed beyond breakdown, allowing sufficient current to flow so that
nonequilibrium conditions are obtained and the device is operated in the deep depletion
mode instead of the inversion mode. Maclver in [49] moved forward by combining p-n
junction and MIS techniques to obtain operation in the deep depletion mode. This resulted
into an extended capacitance range (15:1) in 14 V.
3.3.2 Delay Lines and Electronically Variable Phase Shifter
A phase shifter is a control circuit that can be found in many microwave communica-
tion, radar and measurement systems. Especially nowadays, smart antennas and generally
electronic beam steering techniques, utilise phase shifters for delaying the signals before
combining. Up to now, several methods for phase shifting have been proposed [50]. Phase
shifters are usually divided into analogue and digital depending on how the phase shift is
obtained. In an analogue phase shifter a set of voltage settings is used to obtain the required
phase shift. Specific phase shifts (0°, 22.5°, 45°, etc.), though, are obtained when a digital
phase shifter is used. Both types of phase shifters can be monolithic. However, digital
designs tend to occupy larger chip areas, especially if a large number of bits is required to
achieved the desired resolution. Also it should be noted that an ideal phase shifter should
be dispersion-less so that the phase shift is a linear function of frequency.
So far, the most popular application of the Maxwell-Wagner polarisation in MMIC's has
been the true analogue delay line. If the substrate's conductivity is in the right range and
the insulator thickness is thin enough, a space-charge layer exists at the interface which can
be modulated by either DC or AC voltages along the line. Since this layer's conductivity
will never be high enough to prevent magnetic fields from penetrating it, the inductance
per unit length of the line will be independent of bias conditions. On the other hand, the
capacitance per unit length of the line is voltage dependent, which implies that the phase
shift of a line element can be varied.
Hasegawa, Furukawa and Yanai in [7, 8], were the first on reporting on enhanced phase
shifts of microstrip transmission lines due to slow-wave propagation. Depending on the line
dimensions, they illustrated a phase variation of c/10 to c/30 (3:1). A year later, Jaffe
[51] demonstrated a phase variation of 8:1 with attenuation varying from 0.75 to 1 dB/cm.
This delay line had a non-dispersive character from 70 MHz to 130 MHz. In [52], Jager
presented a schottky microstrip line, which at 100 MHz exhibited a phase velocity up to 80
times slower than speed of light.
The first electronically variable phase shifter, utilising the dispersion characteristics of
MIS microstrip lines, was reported by Hughes in [53, 54] and more analytically in [55] .
Figure 3.6 presents schematically this device. Assuming no line-to-line coupling in this
meander-line structure, the phase shift through the device is given by:
27rf
= 1SF (3.7)
vco
Where f is the frequency of operation, co is the velocity of light in free space, 1 is the
unfolded line length, and SF is the structure's slowing factor (./;).
Even though this device was built back in 1975, its compactness makes it very attractive
for applications such as phased-array radar where small size might be important. This device
also has the advantage that the phase can be changed very rapidly. However, since the phase
variation depends on the applied bias voltage, an accurate as well as large phase variation
Figure 3.6: Schematic drawing of MIS meander-line showing biasing scheme with blocking
capacitors and RF choke. M, I, and S represent metal, insulator, and semiconductor,
respectively [53].
is extremely hard to achieve, particularly over a broad bandwidth. Additionally, another

drawback of this device is the coupling between the adjacent segments of the meander line.
The larger the slowing factor, the larger the electrical length of each line unit will be and
in consequence the larger the coupling between them.
In 1977, Hasegawa, who without doubt is the most significant pioneer of slow-wave
propagation and associated devices, presented a delay line in [56] based on either a Schottky
contact or an epitaxial layer on a GaAs substrate. By doing so he succeeded in reducing the
losses by almost an order of magnitude. Similarly, Neidert and Krowne [57, 58] reported on
a voltage variable microwave phase shifter built on a GaAs substrate as well. Even though
they claim that their device is operable from DC to 150 GHz, data are presented only for
the 2-18 GHz frequency band. Hietala et al. [59] presented a CPW which in the 1GHz
to 12.4 GHz frequency band exhibits a slowing factor of 6. This relatively low slowing
factor allowed them to report an increased value of the quality factor (Q1) (around 4.8 at
12.4 GHz) when compared with the results published previously by Hasegawa (Q=1.6 at
1.6GHz and Q=0.9 at 2GHz) in [56]. A year later, the same authors presented a similar
device, though based on an epitaxial layer, that demonstrated an enhanced slowing factor
in a broader bandwidth (1-20 GHz) but at the cost of increased losses [60].
1Where Q is defined as Q = j3 /2a. With a being the attenuation constant.

Year Ref. Sub. Insulator Line )o/A,1 a2

Psub
(Ohm • cm) Type (GHz) (dB I AO
1996 [61] GaAs 1.5 SCTL3 CPW4 8-24 10 57.5-72

1990 [62] GaAs 1.5 SCTL CPW 2.7 40 2344
1987 [60] Si 5 x 10-3 SCTL CPW 11.7-17 1-20 9.8-70.4
1986 [58] GaAs - SCTL MS5 6 2-18 57.5-240
1986 [59] Si 0.02 Si02 CPW 6 1-12.4 9.6-23
1985 [57, 58] GaAs - SCTL CPW 16.5 2-18 20.9-87.2
1977 [56] GaAs - SCTL CPW 30 1 6.7
1975 [53]—[55] Si 10-3-0.8 Si02& SCTL MS 20 0.5 86.3
1973 [52] Si 0.1-1 SCTL MS 80 0.03-0.1 -
1972 [51] Si 1 SCTL MS 50 0.07-0.12 17.3-23
1971 [7, 8] Si 10-3-85 SiO 2 MS 30 0.03-4 0.076-38.4
1 Maximum normalized slowing factor

2 Reported attenuation range
3 Schottky Contact Transmission Line (SCTL)
4 Coplanar Waveguide (CPW)
5 Microstip Transmission Line (MSTL)
Table 3.3: Comparative review of true analogue delay lines
A rather interesting approach for varying the electrical length of a delay line was intro-
duced by Cheung, Neikirk and Itoh in [62]. A coplanar waveguide (CPW) was built on a
Cr-doped GaAs substrate and its propagation constant was altered by varying the intensity
of the optical illumination applied on top of the conductor. This can be achieved since the
conductivity of the semiconducting layer depends on the density of the electron-hole pairs
generated, which in turn is controlled by the intensity of the optical illumination. How-
ever, this approach is not of any practical use since it experiences a very low slowing factor
accompanied by an extremely high attenuation factor.
The slow character of microstrip transmission lines built on a doped GaAs substrate
was also presented by Ingram and Clifton in [61]. They indicate that their transmission
lines possess a slow-wave factor of 8 to 24 at 10 GHz with a loss per slow-wave factor
of 0.7dB/mm. Additionally, the significant low impedance and associated mismatches are
addressed. Table 3.3 synoptically presents a comparison of all the pre-discussed up to date
available delay lines.
Devices that partially utilise the Maxwell-Wagner polarisation mechanism have system-
atically been tagged as slow-wave devices. In the following sections, we briefly review two
devices for historical reasons; the Voltage Tunable Phase-Shift Oscillator and the Semicon-
ductor Travelling-Wave Amplifier. Both of them utilise a slow-wave element similar to the

ones previously discussed.
Voltage Tunable Phase-Shift Oscillator
A useful application of the variable phase velocity that MIS structures exhibit, is the devel-
opment of a voltage tunable phase-shift oscillator. The block diagram, shown in figure 3.7,
briefly explains the principle of operation of such an oscillator, where the triangular block
represents an amplifier with voltage gain of:
co _ A(e)e _30 A
(3.8)
et
and the square block represents a phase shifter. Therefore, instead of using a conventional
phase shifter, one of the delay elements presented in section 3.2.1 can be used, as demon-
strated in [55]. Clearly the gain of the this circuit is:
co A(e)e— ni'A
(3.9)
ea 1— aTA(e)e-3(0+0A)
Therefore, self-sustained oscillation of this circuit is possible only when the denominator
of equation 3.9 vanishes. It follows that the frequency and amplitude of oscillation must
satisfy:
1 — aT A(e)e-3( °+°A) = 0 (3.10)
Figure 3.7: Block diagram of a phase-shift oscillator.
Furthermore, as the phase shift of the variable phase shifter is given in equation 3.7, we
can calculate the oscillation frequency of this circuit by the following formula:
27r — coo
fasc = (3.11)
l Si — KA
where 00 and KA are constants that can be determined from measurement. Finally, it has
been demonstrated in [55] that when a wideband amplifier is used, sinusoidal oscillations
can be observed at frequencies in the range of 10.65-12.80 MHz.
Semiconductor Travelling-Wave Amplifier
Wideband amplification of microwave signals is quite a difficult task when semiconductors

are used. This is due to the fact that most practical semiconductor devices are either two
or three terminal transit-time limited devices with only a localized interaction taking place.
This results in poor high-frequency operation and frequency dependent input and output
impedances which limits the amplifier bandwidth.
The solution came in 1966 when L. Solymar et al. [63] introduced the solid-state
travelling-wave amplifier (STWA). The amplification is based on the interaction of drifting
carries in semiconductors with travelling waves in an external slow-wave circuit. This in-
teraction leads to the formation of electron bunches, which in turn enhance the fields. This
device (shown in figure 3.8) was subsequently analysed by others including Furukawa [64]
and Sumi [65].
Metal finger
El Base plate
Semiconductor
(GaAs)
Figure 3.8: Pictorial illustration of a STWA.
Although Solymar proposed that a meander-line can be used as the slow-electromagnetic

circuit, it is virtually impossible to entirely rely upon its geometrical slowing factor. Instead,
a meander-line circuit which is fabricated directly on top of an insulator-semiconductor
structure, as the one shown in figure 3.6, might be used. This was proposed by G. W.
Hughes in 1975 [53].
3.3.3 Voltage-tunable MIS filters
The concept of voltage-variable capacitance has also been implemented for the development
of voltage-tunable microwave filters. To the best of my knowledge this concept has been
introduced by Swart et al. in [66] in the flavor of a tunable notch attenuator. In the low-
MHz region, their filter demonstrated a peak attenuation of >50 dB which was maintained
over a 12 % frequency shift effected by a bias voltage change of 15 V. On the same track,
Fleming et al. [67] reported on an attenuator which they like to refer to as a switchable
attenuating medium propagation (SAMP) device. Their device is nothing more than a CPW
on a GaAs substrate in which the off-state loss is determined by the carrier concentration in
the space-charge layer. In the 13 to 18 GHz range, the insertion loss of the on-state ranges
from 1.5 to 3 dB where the off-state losses were measured to be 17 to 20 dB.
In 1975, Hughes [53] reviewed some possible applications of MIS slow-wave propagation
and claimed that the large slowing factors observed in such structures can be effectively
used for constructing miniature microstrip filters. Microstrip filters are typically composed
of open-circuit half-wavelength stubs, fabricated on insulating substrates. However, the en-
hanced dielectric behaviour of MIS substrates results in a compressed effective wavelength
and consequently the stubs of a distributed filter can exhibit a much larger electrical length
than their actual dimensions indicate. Although the effect of the Maxwell-Wagner mech-
anism and the accompanied benefits are known for more than three decades, a miniature
distributed filter based on MIS slow-wave propagation has not been reported so far.
Instead, some attempts have been reported in which the slow-wave phenomena are
created by the spatial separation of energy storage by means of a periodic structure. Seki
and Hasegawa [68] utilised a cross-tie periodic structure to electrically enhance a CPW.
They achieved a slow-wave factor of 12.4 while the attenuation constant of their design
is more than one order of magnitude smaller than conventional MIS CPWs. Based on
this technique, Wang and Itoh [69] presented a miniature band-reject filter with a centre
frequency of 26 GHz. In similar fashion, Yang et al. [70] reported on a low-loss slow-wave
microstrip structure where the ground plane is comprised by a periodic structure which
results in a 1.2-2.4 miniaturization factor. On the same track, Wu [19] presented a band-
pass filter where the physical length is 26% less than that of a conventional microstrip
band-pass filter on the same substrate. Furthermore, this paper assesses the performance
of many previously reported designs based on slow-wave lines.
3.3.4 Variable coupling-coefficient directional couplers
The directional coupler is an important device in MMICs as a number of microwave circuits

such as power splitters/combiners filters and phase shifters can be built out of it. Since
1969, when Julius Lange introduced his very famous quadrature coupler [71], there has been
a great interest in introducing techniques for amplifying the coupling-coefficient between a
coupler's arms. Clearly, the Maxwell-Wagner mechanism and its associated large slowing-
effect can be advantageously used to enhance coupling-coefficients of a coupler or in other
words offer miniature designs.
To date, there is little research focusing on the application of slow-wave structures in

directional coupler designs. The first record of such a device is after Ogawa et a/. [72] back
in 1986. The idea of slow-wave directional couplers is introduced and the associated benefits
are interpreted. However, they do not present any measured data or other indication that
this device has been fabricated. More recently, the technique of periodic loading has been
utilised by Fujii et al. [73] for developing miniature 3-dB branch-line couplers. They report
on several patterns that result in size-reductions up to 25%.
3.3.5 Distributed FET transistors
Another useful device which might take advantage of the variable phase velocity concept is
the distributed field-effect transistor. The advantage of this device was originally proposed
by McIver in [74]. He proposed to use strip-type transmission lines of voltage-controlled
capacitance as the input and output terminals of the FET. By doing so, the drain, source
and substrate associated parasitic capacitances are eliminated qualifying this device for
wide-band operation.
3.4 Summary
During the past years, in an attempt to examine MIS lines performance, several models have
been proposed. Formulae for analyzing MIS interconnects are established by either solving
Maxwells equations or combining them with the motion equations of charged carriers as well
as by extrapolating parameters that constitute equivalent circuits for different propagating
modes. The first half of this chapter is a review of the available models found in the
literature.
While the trend of analyzing the Maxwell-Wagner mechanism and its propagation char-
acteristics was focused on minimising the propagation velocity, other researchers optimisti-
cally utilised this mechanism for enhancing the electrical performance of microwave devices.
It has also been shown that interfacial polarisations can be applied for miniaturisation pur-
poses. In the second half of this chapter, such devices were briefly reviewed.
References
[1] C. Seguinot, P. Kennis, P. Pribetich, and J. F. Legier, "Analytical model of the schot-
tky contact coplanar line," in Proc. IEEE European Microwave Conference, vol. 14,
pp. 160-465, Sept. 1984.
[2] S. Tedjini and M. V. Veen, "Empirical formulas for mis microstrip line parameters,"
Electronic Letters IET, vol. 25, pp. 739-741, May 1989.
[3] J. K. Wee, Y. J. Park, H. S. Min, D. H. Cho, M. H. Seung, and H. S. Park, "Modeling

the substrate effect in interconnect line characteristics of high-speed vlsi circuits,"
Transactions on Microwave Theory and Techniques IEEE, vol. 46, pp. 1436-1443, Oct.
1998.
[4] E. K. Sharma, A. Bhadauria, and A. K. Verma, "Optically controlled microstrip slow

wave structures: A circuit model," in Proc. IEEE Microwave Conference, Asia-Pacific,
pp. 557-560, June 2000.
[5] H. Guckel, P. A. Brennan, and I. Palocz, "A parallel-plate waveguide approach to

microminiaturized, planar transmission lines for integrated circuits," Transactions on
Microwave Theory and Techniques IEEE, vol. 15, pp. 468-476, Aug. 1967.
[6] I. T. Ho and S. K. Mullick, "Analysis of transmission lines on integrated-circuit chips,"

Transactions on Solid-State Circuits IEEE, vol. 2, pp. 201-207. Dec. 1967.

881, Nov. 1971.
[8] H. Hasegawa, M. Furukawa, and H. Yanai, "Slow wave propagation along a microstrip
line on si-sio2 systems," Proc. IEEE, vol. 59, pp. 297-299, Feb. 1971.
66
REFERENCES 67
[9] D. Jager, "Slow-wave propagation along variable schottky-contact microstrip line,"

Transactions on Microwave Theory and Techniques IEEE, vol. 24, pp. 566-573, Sept.
1976.
[10] D. F. Williams, "Metal-insulator-semiconductor transmission lines," Transactions on

Microwave Theory and Techniques IEEE, vol. 47, pp. 176-181, Feb. 1999.
[11] P. Kennis and L. Faucon, "Rigorous analysis of planar m.i.s. transmission lines," Elec-
tronic Letters IET, vol. 17, pp. 454-456, June 1981.
[12] Y. Fukuoka, Y. Shih, and T. Itoh, "Analysis of slow-wave coplanar waveguide for
monolithic integrated circuits," Transactions on Microwave Theory and Techniques
IEEE, vol. 31, pp. 567-573, July 1983.
[13] Y. Fukuoka and T. Itoh, "Analysis of slow-wave phenomena in coplanar waveguide on

a semiconductor substrate," Electronic Letters IET, vol. 18, pp. 589-590, July 1982.
[14] R. Sorrentino, G. Leuzzi, and A. Silbermann, "Characteristics of metal-insulator-

semiconductor coplanar waveguides for monolithic microwave circuits," Transactions
on Microwave Theory and Techniques IEEE, vol. 32, pp. 410-416, Apr. 1984.
[15] T. G. Livernois and P. B. Katehi, "A generalized method for deriving the space-domain
green's function in a shielded, multilayer substrate structure with applications to mis
transmission lines," Transactions on Microwave Theory and Techniques IEEE, vol. 37,
pp. 1761-1767, Nov. 1989.
[16] T. G. Livernois and P. B. Katehi, "Characteristic impedance and electromagnetic field

distribution in metal-insulator-semiconductor microstrip," Transactions on Microwave
Theory and Techniques IEEE, vol. 38, pp. 1740-1743, Nov. 1990.
[17] J. C. Liou and K. M. Lau, "Analysis of slow-wave transmission lines on multi-layered

semiconductor structures including conductor loss," Transactions on Microwave The-
ory and Techniques IEEE, vol. 41, pp. 814-829, May 1993.
[18] J. P. K. Gilb and A. Balanis, "Mis slow-wave structures over a wide range of parame-
ters," Transactions on Microwave Theory and Techniques IEEE, vol. 40, pp. 2148-2154,
Dec. 1992.
[19] K. Wu and R. Vahldieck, "Hybrid-mode analysis of homogeneously and inhomoge-

neously doped low-loss slow-wave coplanar transmission lines," Transactions on Mi-
crowave Theory and Techniques IEEE, vol. 39, pp. 1348-1360, Aug. 1991.
REFERENCES 68
[20] W. J. R. Hoefer, "The transmission-line matrix method—theory and applications,"

Transactions on Microwave Theory and Techniques IEEE, vol. 33, pp. 882-893, Oct.
1985.
[21] S. Chen, R. Vahldieck, and J. Huang, "Rigorous analysis of mode propagation and field
scattering in silicon-based coplanar mis slow wave structures with abrupt transitions
to transmission lines on normal substrate," Transactions on Microwave Theory and
Techniques IEEE, vol. 44, pp. 2487-2494, May 1996.
[22] J. E. Akin, Finite Elements for Analysis and Design. Academic Press, 1994.
[23] M. Aubourg, J. Villotte, F. Godon, and Y. Garault, "Finite element analysis of lossy
waveguide applications to microstrip lines on semiconductor substrate," Transactions
on Microwave Theory and Techniques IEEE, vol. 31, pp. 326-330, Apr. 1983.
[24] C. Tzuang and T. Itoh, "Finite element analysis of slow-wave schootky contact printed
lines," Transactions on Microwave Theory and Techniques IEEE, vol. 34, pp. 1483-
1489, Dec. 1986.
[25] J. J. Kucera and R. J. Gutmann, "Effect of finite metallization and inhomogeneous dop-
ings on slow-wave-mode propagation," Transactions on Microwave Theory and Tech-
niques IEEE, vol. 45, pp. 1807-1810, Oct. 1997.
[26] T. Shibata and E. Sano, "Characterization of mis structure coplanar transmission

lines for investigation of signal propagation in integrated circuits," Transactions on
Microwave Theory and Techniques IEEE, vol. 38, pp. 881-890, July 1990.
[27] H. Hasegawa and S. Seki, "Analysis of interconnection delay on very high-speed lsi/vlsi
chips using an mis microstrip line model," Transactions on Microwave Theory and
Techniques IEEE, vol. 32, pp. 1721-1727, Dec. 1984.
[28] K. W. Goossen and R. B. Hammond, "Modeling of picosecond pulse propagation in

microstrip interconnections on integrated circuits," Transactions on Microwave Theory
and Techniques IEEE, vol. 37, pp. 469-478, Mar. 1989.
[29] Y. R. Kwon, V. M. Hietala, and K. S. Champlin, "Quasi-tem analysis of "slow-wave"

mode propagation on coplanar microstructure mis transmission lines," Transactions
on Microwave Theory and Techniques IEEE, vol. 35, pp. 545-551, June 1987.
REFERENCES 69
[30] F. L. Mesa, G. Cano, F. Medina, R. Marques, and M. Horno, "On the quasi-tem and
full-wave approaches applied to coplanar multistrip on lossy dielectric layered media,"
Transactions on Microwave Theory and Techniques IEEE, vol. 40, pp. 524-531, Mar.
1992.
[31] E. Tuncer and D. P. Neikirk, "Highly accurate quasi-static modeling of microstrip lines
over lossy substrates," Microwave and Guided Wave Letters IEEE, vol. 2, pp. 409-411,
Oct. 1992.
[32] E. Tuncer, Extraction of Parameters for High Speed Digital Interconnects. PhD thesis,
The University of Texas at Austin, Austin, Dec. 1995.
[33] A. K. Verma, Nasimuddin, and E. K. Sharma, "Analysis and circuit model of a mul-
tilayer semiconductor slow-wave microstrip line," Proc. Microwaves, Antennas and
Propagation IEE, vol. 151, pp. 441-449, Oct. 2004.
[34] Y. Fukuoka and T. Itoh, "Field analysis of a millimeter-wave gaas double-drift impatt
diode in the traveling-wave mode," Transactions on Microwave Theory and Techniques
IEEE, vol. 33, pp. 216-222, Mar. 1985.
[35] C. M. Krowne and G. B. Tait, "Propagation in layered biased semiconductor struc-

tures based on transport analysis," Transactions on Microwave Theory and Techniques
IEEE, vol. 37, pp. 711-722, Apr. 1989.
[36] K. Han and T. T. Y. Wong, "Space-charge wave considerations in mis waveguide analy-
sis," Transactions on Microwave Theory and Techniques IEEE, vol. 39, pp. 1126-1132,
July 1991.
[37] M. A. Alsunaidi, S. M. S. Imtiaz, and S. M. El-Ghazaly, "Electromagnetic wave ef-

fects on microwave transistors using a full-wave time domain model," Transactions on
Microwave Theory and Techniques IEEE, vol. 44, pp. 799-808, June 1996.
[38] S. M. S. Imtiaz and S. M. El-Ghazaly, "Performance of modfet and mesfet: a compara-

tive study including equivalent circuits using combined electromagnetic and solid-state
simulator," Transactions on Microwave Theory and Techniques IEEE, vol. 46, pp. 923-
931, July 1998.
[39] G. Wang, Coupled Electromagnetic and Device Level Investigations of Metal-Insulator-

Semiconductor Interconnects. PhD thesis, Stanford University, Feb. 2001.
REFERENCES 70
[40] G. Wang, R. W. Dutton, and C. S. Rafferty, "Device-level simulation of wave propaga-

tion along metal-insulator-semiconductor interconnects," Transactions on Microwave
Theory and Techniques IEEE, vol. 50, pp. 1127-1136, Apr. 2002.
[41] G. Wang, X. Qi, Z. Yu, and R. W. Dutton, "Device level modeling of metal-insulator-
semiconductor interconnects," Transactions on Electron Devices IEEE, vol. 48,
pp. 1672-1682, Aug. 2001.
[42] A. R. V. Hippel, Dielectrics and Waves. The MIT Press, 1954.
[43] T. Uwano and T. Itoh, Numerical Techniques for Microwave and Millimeter-Wave
Passive Structures. John Wiley and Sons, 1989.
[44] 0. Rbenknig, "The finite difference method (fdm) - an introduction." Albert Ludwigs
University of Preiburg, 2006.
[45] K. E. Mortenson, Variable capacitance diodes: the operation and characterization of

varactor, charge storage and PIN diodes for RF and microwave applications. Artech
House, 1974.
[46] W. G. Pfann and C. G. B. Garrett, "Semiconductor varactor using space-charge layers,"

in Proc. IRE, vol. 47, pp. 2011-2012, 1959.
[47] R. Lindner, "Semiconductor surface varactor," The Bell System Technical Journal,
vol. 41, pp. 803-831, 1962.
[48] J. L. Moll, "Variable capacitance with large capacity change," in Proc. IEEE Interna-
tional Symposium on Electronic Materials and Packaging, vol. 3, p. 32, 1959.
[49] B. A. Maclver, "A hybrid voltage-variable capacitor," Transactions on Electron Devices

IEEE, vol. 18, pp. 401-408, July 1971.
[50] I. Robertson and S. Lucyszyn, RFIC and MMIC Design and Technology. Institution
of Engineering and Technology, 2001.
[51] J. M. Jaffe, "A high-frequency variable delay line," Transactions on Electron Devices
IEEE, vol. 19, pp. 1292-1294, Dec. 1972.
[52] D. Jager and W. Rabus, "Bias-dependent phase delay of schottky contact microstrip
line," Electronic Letters IET, vol. 9, pp. 201-203, May 1973.
REFERENCES 71
[53] G. W. Hughes and R. M. White, "Microwave properties of nonlinear mis and schottky-
barrier microstrip," Transactions on Electron Devices IEEE, vol. 22, pp. 945-956, Oct.
1975.
[54] G. W. Hughes and R. M. White, "Mis and schottky barrier microstrip devices," Proc.
IEEE, vol. 60, pp. 1460-1461, Nov. 1972.
[55] G. W. Hughes, Electromagnetic Slow-Wave Devices Utilizing Metal-Insulator-

Semiconductor Microstrip. PhD thesis, University of California, Berkeley, June 1973.
[56] H. Hasegawa and H. Okizaki, "M.i.s. and schottky slow-wave coplanar striplines on
gaas substrates," Electronic Letters IET, vol. 13, pp. 663-664, Oct. 1977.
[57] R. E. Neidert and C. M. Krowne, "Voltage variable microwave phase shifter," Electronic
Letters IET, vol. 21, pp. 636-638, July 1985.
[58] C. 1\4. Krowne and R. E. Neidert, "Solid state monolithic variable phase shifter with
operation into the millimeter wave wavelength regime," Int. Journal of Infrared and
Millimeter Waves, vol. 7, pp. 715-728, May 1986.
[59] V. M. Hietala, Y. R. Kwon, and K. S. Champlin, "Low-loss slow-wave propagation

along a microstructure transmission line on a silicon surface," Electronic Letters IET,
vol. 22, pp. 755-756, July 1986.
[60] V. M. Hietala, Y. R. Kwon, and K. S. Champlin, "Broadband continuously variable

microwave phase shifter employing a distributed schottky contact on silicon," Electronic
Letters IET, vol. 23, pp. 675-677, June 1987.
[61] S. G. Ingram and J. C. Clifton, "The use of active travelling-wave structures in gaas
mmic's," Transactions on Microwave Theory and Techniques IEEE, vol. 44, pp. 956-
960, June 1996.
[62] P. Cheung, D. P. Neikirk, and T. Itoh, "Optically controlled coplanar waveguide phase
shifters," Transactions on Microwave Theory and Techniques IEEE, vol. 38, pp. 586-
595, May 1990.
[63] L. Solymar and E. A. Ash, "Some travelling-wave interactions in semiconductors-theory

and design considerations," Int. Journal in Electronics, vol. 20, no. 2, pp. 127-148,
1966.
REFERENCES 72
[64] S. Furukawa, T. Hata, and S. Horiguchi, "The loss characteristics of a semiconductor

travelling wave amplifier," Jap. Journal in Applied Physics, vol. 8, pp. 739-748, June
1969.
[65] M. Sumi, "Travelling-wave amplification by drifting carriers in semiconductors," Jap.

Journal in Applied Physics, vol. 6, pp. 688-698, June 1967.
[66] P. L. Swart and C. K. Cambell, "A voltage-controlled tunable distributed rc filter,"

Transactions on Solid-State Circuits IEEE, vol. 7, pp. 306-308, Aug. 1972.
[67] P. L. Fleming, T. Smith, H. E. Carlson, and W. A. Cox, "Gaas samp device for ku-
band switching," Transactions on Microwave Theory and Techniques IEEE, vol. 27,
pp. 1032-1035, Dec. 1979.
[68] S. Seki and H. Hasegawa, "Cross-tie slow-wave coplanar waveguide on semi-insulating

gaas substrates," Electronic Letters IET, vol. 17, pp. 940-941, Dec. 1981.
[69] T. H. Wang and T. Itoh, "Compact grating structure for application to filters and
resonators in monolithic microwave integrated circuits," Transactions on Microwave
Theory and Techniques IEEE, vol. 35, pp. 1176-1182, Dec. 1987.
[70] F. R. Yang, Y. Qian, R. Coccioli, and T. Itoh, "A novel low-loss slow-wave microstrip
structure," Microwave and Guided Wave Letters IEEE, vol. 8, pp. 372-374, Nov. 1998.
[71] J. Lange, "Interdigitated stripline quadrature hybrid," Transactions on Microwave

[72] H. Ogawa and T. Itoh, "Slow-wave directional coupler and its applications," int. Jour-
nal of Infrared and Millimeter Waves, vol. 7, pp. 1417-1427, Oct. 1986.
[73] T. Fujii, I. Ohta, T. Kawai, and Y. Kokubo, "Miniature broad-band cpw 3-db branch-
line couplers in slow-wave structure," Journal IEICE Trans. Electron., vol. E90-C,
pp. 2245-2253, Dec. 2007.
[74] G. W. McIver, "A travelling-wave transistor," Proc. IEEE, pp. 1747-1748, Nov. 1965.
Chapter 4
High-K structures Exhibiting

Maxwell-Wagner Polarisation
4.1 Introduction
There is continuous interest in slow-wave propagation arising due to the Maxwell-Wagner

polarisation mechanism in layered substrates. As demonstrated in previous chapters, re-
searchers have extensively investigated this phenomenon, both for preventing slow-propagation
in digital integrated circuits (ICs) and the associated data dependent dispersion as well as
for utilising such propagation in microwave-analogue ICs for passive components. This the-
sis is an investigation of the latter approach and more particularly the exploitation of MISM
structures in emulating high-K substrates.
This chapter presents high-K structures by utilisation of the Maxwell-Wagner polar-

isation mechanism on different substrate materials. Simple devices, such as transmission
lines and diodes, have been designed, examined and characterised both on silicon substrates
as well as on GaAs membranes. Apparently, the presented results demonstrate that these
structures can indeed serve as high-K equivalents. Finally, a novel experimental platform
which supports slow-wave propagation is proposed. This platform allows easy adjustment
of the parameters affecting the interface polarisation mechanism, and consequently the slow
mode characteristics.
73
High-K structures Exhibiting Maxwell-Wagner Polarisation 74
4.1.1 Technology Modelling
During the course of this work, a 3-D electromagnetic simulator, Ansoft's High Frequency
Structure Simulator [1] (HFSSTM ) has been used for simulation tasks in the first instance.
A general purpose electromagnetic simulator, however, requires excessive time (order of
days) to accurately model a structure exhibiting a large dimensional contrast. Furthermore,
full field simulation accuracy is very sensitive to the exact values of layer thicknesses and
material properties. Also, a simplified full-wave field solver, without the more detailed
physical transport processes, usually overestimates the wave slowing. For these reasons,
quasi-static approaches based on physical considerations that take account of the Maxwell-
Wagner odd mechanics seem to be much more practical.
Correction
Coefficients
Design Parameters: Hasegawa's Artificial Dielectric AWR Calibrated

Two-layer with Equivalent Microwav Output
Electrical and Magnetic condenser eaf) and tanojf Office
Properties model
of Materials
Ansoft
Physical Dimensions HFSS
Figure 4.1: Calibration sequence of semi-empirical routines.
Throughout this thesis, most simulations were carried out on a 2.5-D simulation plat-
form, Applied Wave Research Microwave Office [2]. This package is normally used for
microwave circuit simulation and measurement analysis. However, its strength resides with
its ability to integrate externally developed models. Semi-empirical simulation routines,
based on Hasegawa's model [3], were developed in MATLAB [4] (see Appendix B). This
model incorporates the effect of Maxwell-Wagner polarisation mechanism in the simulation
by means of an equivalent artificial dielectric that exhibits a frequency dependent cneff (f)
and tan6(f). The 3-D simulator is still necessary to calibrate the faster routines, and also
model launcher structures and when necessary, correction coefficients are applied (see figure
4.1).
4.1.2 Measurement Apparatus
Scattering Parameter Measurement Apparatus
The S-parameter measurements presented in sections 4.2 and 4.3, were performed on a Cas-
cade Microtech probe station using Agilent's E8361A computer-assisted Personal Network
Analyser (PNA), which can be used between 10MHz and 67GHz. Contacts to the lines
were made with a pair of Cascade Microtech microwave probes, of 200,am pitch Ground-
Signal-Ground (GSG), which provided rapid, reproducible measurements free of packaging
considerations. The system was calibrated on wafer with a Cascade Microtech Impedance
Standard Substrate (ISS) calibration kit. Measurements were taken over the range from
1MHz to 30 GHz.
All measurements presented in section 4.4 were carried out using an HP8753D 6 GHz
Network Analyzer calibrated with 3.5 mm coaxial calibration standards. The volumetric
conductivity of the solutions were also measured as a function of frequency with the aid of
Agilent's 85070E Dielectric Probe Kit.
Capacitance-Voltage Measurement Apparatus
Capacitance-voltage (CV) measurements is a widely used method to gain information on the

electrical properties of insulating materials. These measurements allow the determination
of insulator charges like interface traps, volume traps or mobile ionic charges, which are
responsible of the Maxwell-Wagner polarisation extent. Such measurements were conducted
on a semi-automatic probe station using Agilent's B1500A Semiconductor Device Analyser
(SDA). The MOS capacitors were contacted with a pair of RF probes (figure 4.2) and the
system was calibrated at 1 KHz and 1 MHz, where measurements were taken.
Metal
Insulator
LC-Meter
Semiconductor
Ohmic contact
Figure 4.2: CV measurement set up.

4.2 Silicon Based Structures
Over the past 50 years, semiconducting materials have played a significant role in microelec-
tronics and ICs. Silicon has been heavily used ever since, as it maintains its semiconductor
properties at high temperatures, native oxides can be easily grown and forms a better semi-
conductor/dielectric interface than any other material. Likewise, since Si is the second most
abundant element in the earth's crust, it is also inexpensive. Rationally, the first MMICs
were implemented on Si and thus it has been heavily used in MIS structures.
In this section, MISM morphologies where the semiconductor layer is formed by doped Si
have been investigated, designed and characterised. Such structures abound in the literature
and as such there is no desire for "reinventing the wheel". It is however my intention to
thoroughly investigate the Maxwell-Wagner mechanism in prospective media and therefore,
silicon based structures provide useful patterns for comparison. Furthermore, later on this
thesis, novel techniques for improving the slow-wave propagation characteristics in Si-based
structures are proposed.
4.2.1 Characterisation Devices
A multilayered structure that consists of metal, insulator and semiconductor layers, under
certain conditions, is shown to support the Maxwell-Wagner polarisation mechanism. The
direct effect of this mechanism is the significant slow-down of the propagation velocity,
which can be interpreted by the effect a single high-K dielectric has. In other words, passive
devices built on such substrates tend to behave as if they were electrically larger than when
developed on standard substrates. Since the scope of this thesis is the investigation of
Maxwell-Wagner polarisation and its application to MMICs, the investigation is extended
on potential layered morphologies, by means of some representative passive devices, through
which information about the actual polarisation mechanism can be extrapolated.
Simple passive devices, that are based on MOS layered structures, such as microstrip
lines, diodes, interdigitated capacitors, coplanar waveguides, patch antennas and directional
couplers have been designed and modeled under various conditions. The top metal layout
of these devices was transfered on a 5x5 inch chrome on glass mask, shown in figure 4.3 with
the individual devices highlighted. This mask was then utilised to transfer the conductive
patterns on different substrate morphologies.
Devices such as microstrip transmission lines and coplanar-waveguides are useful for
characterisation purposes as their measured effective length and their expected value can be
easily compared. The geometry of the experimental microstructure transmission lines is also
shown in figure 4.4, while the coplanar equivalents are presented in figure 4.5. Additionally,
biasing effects are investigated through MOS capacitors (see figure 4.6). This way all
possible design aspects and their direct effect on interfacial polarisation can be investigated.
▪ Microstrip Lines 3 Interdigitated Capacitors • Patch Antennas

• CV dots

4 Coplanar Waveguides ri
Directional Couplers II Calibration Impedances
•. •
MarelOginwr Merl:Mon Meehanlie
Chinceidoim Mak
Therrisiddleholunilik
Impel" Cabo London
•
••
Figure 4.3: Maxwell-Wagner polarisation mechanism characterisation mask.
Figure 4.4: Microstrip Transmission Lines of several widths and lengths.
•
Figure 4.5: Coplanar waveguides of several widths, lengths and gaps.
•
•
Figure 4.6: Top metal electrode of MOS capacitors with the electrode area varying.
4.2.2 Effect of Substrate Conductivity
The main function of substrate conductivity is to establish the frequency band over which
the slow mode is observed. As discussed previously, increasing the substrate conductivity of
a metal-insulator-semiconductor system results in an increased frequency band in which the
slow-wave mode dominates (equation 2.39). Therefore, low resistivity substrates offer large
slowing factors over broader bandwidths. Additionally, the associated high losses, which
grow rapidly outside the polarisation mode, are shifted towards higher frequencies.
Implemented Process Flow
The geometries of the experimental microstructure transmission lines, shown in figures 4.4
and 4.5, were separated from a silicon substrate by an insulating SiO2 layer. A 700nm thick
oxide was grown by wet oxidation at 1100°C, with the substrate conductivity ranging from
0.002 to 200 S/cm. These values were also confirmed with a 4-probe surface conductivity
measuring system. The centre conductor and ground strips were fabricated by evaporating
Ag or Au onto the SiO2 layer and the conducting structures were defined via standard
photolithographic techniques. The ohmic ground plane was formed on the unpolished back
side of the wafer by evaporating a 300 nm thick layer of Al after treating the surface with a
buffered Hydrofluoric (HF) solution to remove the unwanted grown oxide. For all specimens
described in this section, the wafer thickness was 525 ,um and the top metal thickness was
300 nm. Finally, the large ohmic contact frame was connected to ground over the full area
using conductive epoxy around the outside edge of the chip along both sides.
Results and Discussion
The resulting slow-wave mode occurs when electromagnetic fields partially penetrate the
substrate. This is valid under two cases:
• When the intrinsic impedance (equation 2.31) is rather low, preventing electric fields
penetrating into the substrate.
• When the skin depth (equation 2.30) is comparable to the substrate's thickness, i.e.
magnetic fields penetrate the entire substrate.
Both of the above criteria are dependent upon the operating frequency of the signal as well
as the substrate's conductivity. And when both parameters are fulfilled, the wave velocity
becomes a function of the substrate and not just the good dielectric insulation between the
trace and the top layer of the substrate.
The effective effective length of a 100 pm long MSTL, as well as the group delay of
a propagating signal through this device, were measured with the substrate conductivity
varying. Figure 4.7 illustrates the remarkable dependance of the propagation characteristics
of the MSTL on the substrate's conductivity and hence on the Maxwell-Wagner polarisation
mechanism. The sample with the lowest substrate conductivity (0.002 S/cm), exhibits a
comparable to the substrate thickness skin-depth at low frequencies and reasonably enough,
the slow-wave mode is partially supported in the low MHz region. The moderate conduc-
tivity (20 S/cm) specimen, fulfills the intrinsic impedance as well as the skin effect criteria,
leading to an evidently large energy dispersion from 100 MHz and up to a few GHz. It is
significant to point that this particular conductivity value, is quite close to the maximum
conductivity limit (given by equation 2.43) which maximizes the upper cut-off frequency
limit of this polarisation mode. Therefore, increasing the substrate conductivity (200 S/cm),
one would expect the slow-mode bandwidth to become narrower. Additionally, this con-
ductivity increment, reduces the skin depth and thus the substrate's volume within which
nearly all the energy propagates. In other words, this can be very roughly approximated by
a Metal-Insulator-Metal structure instead of the conventional MOS-metal pile. Therefore
the smaller effective slowing-factor of the substrate as well as the narrower bandwidth is
due to the slow-wave to skin effect mode transition.
The substrate conductivity effect can also be delineated via figure 4.8, where the mea-
sured insertion loss of the previously discussed devices is depicted. It can be observed that
the microstrip line on the high conductivity substrate experiences a slightly larger dissipa-
tion factor than the moderate conductivity sample, in the MHz region. However, from 2
GHz to 10 GHz the MSTL on the 20 S/cm substrate demonstrates a lower forward trans-
mission coefficient. This is well expected as it is attributed to the higher dielectric losses on
the mode transition region. Additionally, since the slow-wave mode is partially supported
in the high resistivity MSTL, the associated equivalent tans of the structure is not as large
as in the other two cases.
60
Group Delay (ps) as 20 S/cm
—c— c5.=200 S/cm
40 — - as,=0.002 S/cm
20 —
0
15
Effective Lengt h (mm)
—0-- crs,-20 S/cm

- crsi-200 S/cm
10 as,=0.002 S/cm
0
01 10
Frequency (GHz)
Figure 4.7: Measured group delay and effective length of a 100 pm MSTL on substrates of
various conductivity.
0
Insertion Loss (dB)
-6
—e— 0-S.=20 S/cm
- 65i=200 S/cm
-8 =0.002 S/cm
si
-10
0.1 1 10
Frequency (GHz)
Figure 4.8: Measured insertion loss of a 100 itin long MSTL on substrates of various con-
ductivity.
4.2.3 Effect of Semiconductor to Insulator Thickness Ratio
Under appropriate conditions the Maxwell-Wagner polarisation mechanism is supported in

layered media. However, in structures where the semiconductor to insulator thickness ratio
is relatively large, the corresponding detected slowing effect is substantially large. This is in
accordance with equation 2.37, where the dependance of the relative effective permittivity
on the individual layer thicknesses is depicted.
In similar fashion with the process flow presented in section 4.2.2, specimens that encom-
passed oxides of several thicknesses were developed. These structures employed microstrip
as well as coplanar Ti/Au strips separated from a p-type silicon substrate by a thin layer of
Si02 . The Si02 was grown on Boron-doped, p-type silicon (0.05-1.0 Slcm) by wet oxidation
at 1100°C in a carbolyte furnace. The oxidation time was varied from 10 minutes to 3.5
hours, which resulted in 250 nm to 1.05 pm thick oxides (table 4.1). The top metal layer
was fabricated by evaporating a 25 nm thick adhesion layer of Ti followed by a 100 nm
Au layer and all conducting patterns were defined through standard photolithographic and
etching techniques. The back side ohmic contact was formed by a 3000A thick Al layer.
Table 4.1: Wet-Oxidation of Si at 1100°C

Oxidation Time Oxide Thickness
10 min 250 nm
30 min 460 nm
1 h and 45 min 850 nm
3 h and 30 min 1.05 um
A 100 pm long MSTL on homogeneous substrates, typically exhibits a phase shift of 0.5°
at 2 GHz1. However, in practice, semi-conductive substrates lead to polarisation effects
that significantly delay propagating waves and thus transmission lines on such substrates
demonstrate considerably larger phase shifts. Figure 4.9 exhibits the unwrapped phase-shift
'This is valid for Si or Si/Si02 substrates and under the assumption that any polarisation effects are
minimal.
variation of a 100 pm long MSTL for various semiconductor to insulator thickness ratios.
The Si substrate is rather conductive (o = 2000S/m) and thus the slow-wave mode is well
supported within the measured frequency spectrum. The manifested phase-shifts are up
to two orders of magnitude larger when compared with standard devices. The appealing
attribute of figure 4.9 is the obvious relation of the substrate's geometrical ratio with the
effective length of the device and illatively the Maxwell-Wagner polarisation magnitude.
25
Phase Shift (Deg)
50
- Measured tsub/tox=2100
— Simulated tsubltox=2100
Measured tsub/tox=1140
-75 -- —s— Simulated tsubltox=1140
- Measured tsub/tox=620
--0— Simulated tsub/tox=620
Measured tsub/tox=500
--,-- Simulated tsub/tox=500
-100
01 10
Frequency (GHz)
Figure 4.9: Measured and simulated unwrapped phase-shift of a 100 pm long MSTL with
the semiconductor to insulator thickness ratio varying.
Table 4.2 summarises the investigated semiconductor to insulator thickness ratios, along-
side with the expected slowing factors of each structure. Apparently, the more anisotropic
the substrate is, the higher the effective electric permittivity of the structure will be (see
equation 2.37). These claims are also verified via figure 4.10 in which the correspond-
ing measured effective lengths (EL) as well as group delays (GD) of the devices under test
(DUT) are displayed. As a comparison figure, the corresponding electrical length and group
delay of each specimen at 100 MHz2 are given in Table 4.2. Similarly, figure 4.11 depicts
the measured group delay and effective length of a CPW (L=100 tim, G=20 pm and W=60
pm). Evidently, Maxwell-Wagner polarisation boosts the effective length of the CPW in a
similar fashion with the MSTL case presented earlier.
2 At 100 MHz the dominating slow mode is nearly dispersionless, i.e. at 100 MHz a fair comparison of
the slowing factors can be done.
120
Group De lay (p s) - tsubitox=21 00

tsub/tox=1140
80 tsub/tox=620
tsub/tox=500
40
Effective Length (mm)
—a— tsub/tox=2100
- tsub/tox=1140
- tsubitox=620
- tsubltox=500
0
01 10
Frequency (GHz)
Figure 4.10: Measured group delay and effective length of a 100 pm long MSTL with the
semiconductor to insulator thickness ratio varying.
120
Group Delay (ps )
- tsub/tox=2100
—0— tsub/tox=1140
80 — — tsub/tox=620
- tsub/tox=500
40
0
30 —0— tsub/tox=2100
- tsub/tox=1140
- tsubitox=620
20 --v— tsub/tox=500
10
0

01 10
Frequency (GHz)
Figure 4.11: Measured group delay and effective length of a 100 pm long CPW with the
semiconductor to insulator thickness ratio varying.
As anticipated in chapter 2, the perceived large delays are interlinked with the energy
storage mechanism of Maxwell-Wagner polarisation. However, this energy exchange also
Table 4.2: Semiconductor to Insulator Thickness Variation Effect

tsub/tox er,eff Effective Length (mm) Group Delay (ps) f max (MHz)
2100 8194 33.3 111 596
1140 4455 20.7 69 635
620 2413 14 39.7 660
500 1954 9.7 32.2 664
translates into increased dielectric loss. Therefore, a structure with a large semiconductor
to insulator thickness ratio may resemble a high-k dielectric material, but also demonstrates
a large dissipation factor (Table 2.1). Indeed, this is the case and it is verified through
insertion loss measurements, presented in figure 4.12. Consequently, the corresponding
bandwidth over which the Maxwell-Wagner polarisation can be efficiently utilised3, becomes
narrower if a large slowing factor needs to be achieved (see Table 4.2). All measured data
are in excellent agreement with simulations.
0 0
-10
U)
-20
C
Measured MSTL with tsubitox=2100 cll

— - Simulated MSTL with taut:Aux...2100
Measured MSTL with tsubitox=1140 -30
- Simulated MSTL with tsubltax=1140
- Measured MSTL with taubttox=620
Simulated MSTL with tsubltax=620
Measured MSTL with tsubitoxxSOD
- Simulated MSTL with tsubitox=500
-15 40

01 10
Frequency (GHz)
Figure 4.12: Measured and simulated insertion and return loss of a 100 pm long MSTL
with the semiconductor to insulator thickness ratio varying.
3 Here it is defined as "efficient bandwidth" the 3 dB cut-off point in which the relaxation of the Maxwell-
Wagner polarisation exhibits a nearly constant effective permittivity as well as dissipation factor (i.e. is
nearly dispersionless).
4.2.4 Effect of Transmission Line Dimensions
It is well known in MMIC design that a wider linewidth tends to reduce the resistance R of
the line, which can be desirable. However the capacitance per unit length of the line also
increases. It is therefore meaningful to investigate the influence of a device's dimensions on
the Maxwell-Wagner polarisation mechanism.
The top metal patterns of the utilised characterisation devices (4.4 and 4.5) were defined
through standard photolithographic techniques on a 250 nm thick SiO2, grown on a Si
substrate of a = 20S/cm. The resulting line dimensions are summarized in Table 4.3. For
all devices presented in this section, the wafer thickness (dsi ) is 525 pm, the oxide thickness
is t0x=0.25 pm and the top metal thickness is 100 nm.
Table 4.3: Characterisation Devices Dimensions (pm)

Devices Length Width Gap Diameter
MSTL 100, 200, 400 1, 2, 5, 10, 20, 40, 60
CPW 100, 200, 400 2, 10, 20, 60 20, 100, 250, 500
CV-dots 100, 300, 500, 1000
A high-k dielectric, when sandwiched between two conducting plates, can store more charge
and consequently can provide a larger capacitance when compared with a low-k dielectric
of the same thickness. This is illustrated in figures 4.13(a) and 4.13(b), where the electric
field distribution of a MSTL built on a low-k as well as a high-k dielectric are depicted.
Clearly, in the high-k case a stronger E field can be supported, which means that the field
lines are more confined within the insulating media. Undoubtedly, the fringing field lines
will get more sparse when the dielectric constant of the insulator increases or when the top
metal conductor becomes wider, i.e. enclosing more dynamic field lines (figure 4.13(c)). It
can be therefore anticipated that the E-mode slow-wave propagation is more dominant on
wider lines than on narrower lines.
(a)

fig (b)

(c)
Figure 4.13: Field line distribution of a MSTL on (a) low-k dielectric (b) high-k dielectric
and (c) high-k dielectric with wider top conductor.
- c- MSTL with W=5µrn

- MSTL with W=20).un
MSTL with W=40µm
- - MSTL with W=60p.m
-50 -
cc)
a)
0 - - MSTL wit W- pm
- • - MSTL Milk W=200rn
-100 MSTL v.ith W=40,rn
_c ----MSTL v.ith W=60,m
U)
a) 0
in
co
0- 2 -20-
-150
oo
-200
01 10
Frequency (GHz)
Figure 4.14: Measured unwrap-phase shift of MSTLs for various widths: 5 itm, 20 pm, 40
pm and 60 pm. Inset: Zoomed-in caption of linear phase variation up to 1 GHz.
The unwraped phase-shift of a 100 t.tm long MSTL has been measured, with the line's
width varying from 5 pm to 60 pm and the results are depicted in figure 4.14. Since all
the devices have been developed on the same structure, the measured phase-shift for all the
lines unfolds in a rather similar way. In the low-GHz region, where the slow-wave mode is
dispersion-less, the phase-shift of the lines varies up to 5° at 1 GHz, with the wider MSTL
manifesting the largest delay (inset of figure 4.14). However, above a few GHz, dispersion
occurs, affecting the propagation characteristics of the lines. And since the Maxwell-Wagner
polarisation mechanism is more presiding on the wider lines, the deterioration of wider
line's electrical length is observed to occur at lower frequencies than the narrower ones as
depicted on figure 4.15. Clearly, this has a direct effect on the experienced group delay of
propagating signals, as shown on the same figure as well as through the phase-shift of all
MSTLs at relatively high frequencies. It is quite interesting that above 5 GHz the narrow
lines are still under the influence of the polarisation mode and as such they look electrically
larger than wider MSTLs which their slow- wave propagation characteristics have started
to deteriorate.
150
-a-W=60µm
- -W=40 pm
U) -0-W=20pm
9; 100 W=5pm
>,
(7)
0
0_ 50
0
(9
Effe ctive Length (mm)
30
20
10
0
01 10
Frequency (GHz)
Figure 4.15: Measured group delay and effective length of MSTLs for various widths: 5 pm,
20 pm, 40 pm and 60 pm.
On the other hand, as frequency increases, the effective dielectric of the substrate drops
while the overall effective loss tangent increases. Therefore, comparing a wide and a narrow
MSTL of the same length, the wide one is expected to exhibit lower insertion loss at low
frequencies, due to the larger effective dielectric constant. This is actually demonstrated
in the inset of figure 4.16. Then, as frequency increases, the comparable dielectric losses of
the wide MSTL become larger and as expected, the line becomes more lossy. This is indeed
the case and it is fully demonstrated on figure 4.16.
0
MSTL with W=5µm
MSTL with W=20µm
MSTL with W=40µm
MSTL with W=60µm
• • MS11,4411W 45,4n
uan with W4201.
--- W44ow
rasa .4111W40Own
sertion Loss (dB)
C
O
t
u)
(1) -20
C
2 4 6 40 11
Frequency (GHz) Ir
30 I I 1 S
01 1 10
Frequency (GHz)
Figure 4.16: Measured insertion loss of MSTLs for various widths: 5 pm, 20 pm, 40 pm
and 60 pm. Inset: Zoomed-in caption of insertion loss up to 10 GHz.
-50
0)
a)
0
to
a)
uJ
_c -100
(13
0_ —a—Measured L= 1 00 µm
— Simulated L=1 00µm
— Measured L=200um
Simulated L=200um
Measured L=400um
Simulated L=400µm
-150

01 10
Frequency (GHz)
Figure 4.17: Measured and simulated unwrap phase-shift of MSTLs for various lengths: 100
pm, 200 pm, 400 pm.
4.3 GaAs Membranes
Gallium arsenide (GaAs) is an important semiconductor that has been extensively used in
MMICs [5]. When compared with Si, GaAs demonstrates superior electronic properties.
It has a higher saturated electron velocity and higher electron mobility. Therefore, GaAs
devices generate less noise than silicon devices when operated at high frequencies. It also
has a higher breakdown voltage than Si, allowing higher power levels.
(a)
2pm n' GaAs (1x10" cm ')
(b) 2ttm LT GaAs
(c)
Figure 4.18: Schematic illustration of GaAs membrane structure. Shown are: (a) top metal
patterns, (b) cross-section of membranes and (c) SEM microphotographs of the back-side
during etching (vias are visible as dark squares).
In this section, the propagation characteristics of similar to the previous section devices
(see figure 4.3) that were developed on 4 ,um GaAs membranes are presented. The devices
were designed at Imperial College and were fabricated by our overseas collaborators at
IESL-FORTH4, within the EU AMICOM Network of Excellence.
4lnstitute of Electronic Structure and Laser, Foundation for Research and Technology-Hellas, Greece
4.3.1 Implemented Process Flow
Three layers of A/0.55Ga0.45 As, low-temperature (LT) GaAs and n+ GaAs of 200nm, 2pm
and 2,um respectively, were grown on a semi-insulating (SI) GaAs substrate. A SiO2 layer
of 200nm thickness was deposited everywhere. Vias patterns were implemented by pho-
tolithography, wet etching SiO2 in a buffered HF solution, followed by a H2SO4/1/2 02/1/2
for etching the two GaAs based layers, about 4 pm deep and below the A1GaAs etch stop
layer. The top conductive patterns were defined through photolithography and a sandwich
of Ti/Au (50A /1000A ) was evaporated. The excess Au and Ti were back-etched using an
iodine based solution and a 1% HF solution respectively. The SI GaAs substrate was then,
lapped down to about 200 pm and the membrane was defined by a back-side lithography
followed by a fast wet etching and a membrane selective reactive-ion-etching (RIE). Finally,
a blanket layer of Au was deposited on the back-side of the sample.
4.3.2 Results Discussion
The fabricated membrane structures, shown on figure 4.18, have a relatively small thickness
contrast between the oxide and the semiconductor. As a result, one would expect that
the effective slowing-factor of the structure will be moderate. It is however known that
GaAs/oxide interfaces have a rather high trap density, which can have a significant effect
on the overall effective dielectric strength of the membranes.
In figure 4.19, the group delay as well as the measured effective length of MSTLs with
standard L=100 pm are depicted. Clearly, the presented measurements indicate slowing
factors that profoundly exceed anything reported so far (see figures 4.7, 4.10, and 4.15 as
well as Table 4.2). These results are in direct contradiction with the geometrical factors of
the laminate materials evolved. However they are attributed to the large trap density of the
interface between the semiconductor and the oxide. Additionally, in this case, where the
semiconductor is formed by doped GaAs and the insulating media is formed by a deposited
oxide, there is a great discontinuity between the phase-transition of the materials [6].
The frequency of operation of the slow mode can in theory be maximised while the losses
can be simultaneously reduced by using a thin, high conductivity substrate. In figure 4.20,
similarly with figures 4.12 and 4.16, distinct lossy-bands can be identified. In the low-MHz
region, the insertion loss of the lines roughly increases by 5 dB/dec and over the 3 GHz to
30 GHz bandwidth, the insertion loss augments by almost 30 dB/dec.
400
0
0. 300
>
200
2 100
SL
0
-^ 120 • • •
—*—W=60prn
E —e—W=40pm
_c —6--W=10pm
45) 80 —r—W-5prn
a)
a)
> 40
a)
0

01 1 10
Frequency (GHz)
Figure 4.19: Measured group delay and effective length of MSTLs for various widths: 5 pm
10 pm, 40 pm and 60 pm.
10
Insertion Loss ( dB)
20
30
—s— W=60µrn
-40 - W=40µrn
W=101im
— 4P— W=51101
50

01 1 10
Frequency (GHz)
Figure 4.20: Measured insertion loss of MSTLs for various widths: 5 pm 10 pm, 40 pm and
60 pm.
The measured losses of the GaAs membrane-based devices are more or less double the
losses that identical devices, based on Si-substrates, exhibit (see figure 4.16). However, the
corresponding measured slowing-factors of the GaAs membranes are tripled when compared
with those reported in figure 4.15. Moreover, the third lossy region appearing in figure 4.16
is not depicted in figure 4.20 and it is believed that is retracted at even higher frequencies.
This fact supports all expectations that the cut-off frequency of the slow-wave mode can be
extended by utilisation of this type of substrates.
4.4 Slow-wave Characteristics of Microstrip Transmission Lines

on Aqueous Dielectrics
Observing the slow-wave propagation as well as measuring the attenuation of MIS microstrip
lines is quite a difficult task. In order to correlate the theory with the actual behaviour
of these lines, experiments must be carried out for different width strips over a wide range
of frequencies. On the other hand, the strong frequency dependence of the semiconductor
attenuation dictates that short lengths of line must be used if a wide range of frequencies
are to be measured. This in turn will create measurement problems, since radiation loss
due to discontinuities becomes significant. Furthermore, the slow-factor dependence upon
the physical thicknesses of the insulator-semiconductor layers as well as the substrate's
conductivity requires the development of a large number of specimens.
In an attempt to save resources on developing slow-wave specimens, an experiment has

been set up that mimics the slow performance of these devices without great cost. This
platform proves quite useful for investigating the MIS microstrip characteristics for a large
variety of parameters, such as substrate conductivity, semiconductor to insulator thickness
ratio as well as line thickness.
4.4.1 Experimental Platform
The experimental structure, shown in figures 4.21 and 4.22, replicates the insulator - semi-
conductor system which is well-known for supporting the propagation of slow-waves. In
this model the semiconductor layer is simulated by an ionic solution while the insulator
and metal layers are formed on FR4 substrates instead of typical insulators such as SiO2.
The microstrip lines are excited through SMA launchers fitted on the top of the platform.
Furthermore, for ensuring a uniform ground return path as well as preventing leakages to
the top of the platform, the copper ribbons shown in figures 4.21 and 4.22 were utilised.
The following design considerations of the experimental platform, offer significant ad-
vantages when compared with the standard MIS system:
• The salinity of the solution can be varied by adding an electrolyte, such as NaCI or an
acid, like HC1. This allows controlled variation of the aqueous dielectric conductivity.
• Different materials can be employed to construct the top dielectric layer; laminates of
several thicknesses and permittivity can be easily fitted on the experimental platform
which results in greater flexibility.
• The ionic solution's thickness can be adjusted manually by using supporting posts of
different height, or even by utilising an adjustable height support mechanism.
• Different conductor patterns can easily be introduced, as they only require printing
the necessary pattern on a laminate.
FR4 laminate
Supply tank
Figure 4.21: Experimental structure with a copper plate forming the back metallization.
4 GND metalization
Figure 4.22: Top view of the experimental structure showing the SMA connectors as well
as the printed lines.
4.4.2 Apparatus
In order to measure the conductivity of the ionic solution and verify the predicted values
given by the salinity of the aqueous substrate, a standard technique which has been de-
scribed, in [7] has been utilised. This volumetric conductivity meter consists of two coaxial,
cylindrical electrodes, fixed with dielectric separators that do not contact the liquid (figure
4.23). The electrodes are immersed in the liquid to an arbitrary initial depth, and AC
impedance is measured over a wide frequency range. This process is repeated over many
immersion locations, and the electrical conductivity is calculated by linear regression on
the measured variation of conductance with immersion depth. Nonetheless, the volumetric
conductivity of the solutions was also measured as a function of frequency with the aid of
Agilent's 85070E Dielectric Probe Kit.
electrode
leads
dielectric
separator
outer
electrode
inner
electrode
Figure 4.23: Schematic illustration of coaxial cylinder electrodes as deployed in measuring

the electrical conductivity of a liquid [7].
4.4.3 Simulation Results
The physical principle of slow-wave propagation is based on the semiconducting nature of

the thin epitaxial layer, which performs a kind of "filtering" against the field lines. Under
certain conditions the epitaxial layer does not allow the electric-field lines to pass through
it, but it does allow the magnetic field lines to pass. This behaviour was verified using the
HFSS finite element solver [1].
The experimental structure shown in figures 4.22, 4.21 was simulated for a copper line of
thickness w=2 mm printed on a FR4 laminate with thickness t FR4=0.5 mm, containing pure
distilled water with a conductivity of o-H20=0.09 S/m, covering a depth of 4/20 =20 mm.
The electric and magnetic field distributions for this configuration are shown on figure 4.24.
Increasing the water's conductivity from 0.09 S/m to 4 S/m, while maintaining the physical
dimension parameters unaltered, results into less electric field penetrating the substrate
(figure 4.25).
Figure 4.24: S(left) and ,91' (right) field distribution of the liquid substrate with asub =
0.09S/m.
Figure 4.25: 63(left) and dr (right) field distribution of the liquid substrate with asub =
4S/m.
4.4.4 Experimental Results
The results presented in the following sections demonstrate the ability of the experimen-
tal platform to support slow-wave propagation, and serve as an adequate platform for a
detailed investigation of the slow-wave behaviour in metal-insulator-semiconductor printed
waveguides. In the discussion it is demonstrated that all obtained data are in excellent
agreement with the theory. All measurements were carried out using an HP8753D 6 GHz
Network Analyzer calibrated with 3.5 mm coaxial calibration standards.
Effect of Substrate Conductivity
The main function of substrate conductivity is to establish the frequency band over which
the slow mode is observed. As discussed previously, increasing the substrate conductivity of
a MIS system results to an increased frequency band in which the slow-wave mode dominates
(equation 2.39). Therefore, as demonstrated in table 4.4, low resistivity substrates offer large
slowing factors over broader bandwidths. The associated high losses, which grow rapidly
outside the polarisation mode, are shifted towards higher frequencies.
Table 4.4: Calculated relaxation frequency of interfacial polarisation in MHz

Conductivity in S/m
tH2 O/tFR4
0.09 5.3 10.6 21.6 81
157.48 2.4 126 252 514 1930
78.74 4.8 252 504 1030 3850
40 9.4 500 992 2020 7600
20 19 992 1980 4040 15000
12.9 29 1540 3080 6270 23500
6.45 58 3080 6150 12500 47000
In figures 4.26(a) and 4.26(b) measurements of propagation of a 96 mm line on the

prescribed system are plotted. In this measurement a 20 mm thick layer of water plays
the role of the semiconductor and standard FR4 laminate of thickness -I FR, -=0.5 mm was
used as the insulator. The distilled water conductivity was measured to be approximately
01120=0.09 S/m. The slow-wave mode is partially supported in the low MHz region, namely
between 250-350 MHz. A direct comparison with simulation of an idealized quasi-TEM line
of the same dimensions on a homogeneous water substrate, plotted on the same graphs,
indicates that the slow mode is approximately 2 times slower than the quasi-TEM mode.
The high frequency limit of the slow-wave mode lies at about 350 MHz. At that point the
effective dielectric constant of the line starts decreasing and eventually settles at the much
lower value given by Wheeler's approximation (equation 2.36). Very large losses at the high
end of the band are evident in figure 4.26(b) where the forward transmission coefficient of
the line is plotted.
The addition of 70 g of NaC1 yielded a saline solution of 35 parts per thousand (ppt),
which resulted to a measured conductivity of asub 5 S/m. As expected, increasing the
conductivity of the ionic solution broadens the frequency range in which the Maxwell-
Wagner polarisation occurs. A 50 MHz shift of the relaxation of the structure is clearly
0-
....................... • a.
.................... A
-100 -
.............. • 1••
• ....
-10 -
• • •
131
- • tr. ;
-••••
re -200 - ".0 • /
• / • •
•/ ••
cn
cu -300 •
Jr.
•
0
-Simulated-Dieted Water .•" 30
-400 - • Measured-Distilled Waist — Simulated-Distilled Water ■
- - SImulated-Sallnity• 35 pet • Measured-Distilled Water
• Measured-Salinity= 35ppt - - Simulated-Salinity= 35ppt
Ideet simulated Behaviour • Measured-Salinity= 35ppt
- *- Idea: simulated Behaviour on Er,eff=190 • Ideal simulated Behaviour
-500 -40
00 0.2 0.4 0.6 02 0.4 0!6
Frequency (GHz) Frequency (GHz)
(a) (b)
Figure 4.26: The experimental platform has been set with tFR4 =-0.5 mm and tH20=20 mm
for a 96 mm long microstrip line. Measurements were taken with the salinity of the ionic
solution varying up to 140 ppt. Shown are: (a) measured effective lengths of the microstrip
line in degrees and (b) forward transmission coefficient in dBs.
visible in figure 4.26(a), with the effective length of the line tracking that of the line based
on the less saline solution, with the exemption of the extended band. Additionally, in
agreement with theory, the high loss transition region, from slow-wave to skin-effect mode,
has shifted upwards in frequency by approximately 50 MHz.
Contrary to intuition, a further increase of the solution's salinity did not result into an
extended frequency band in which the dominant mode is the polarisation mode. This is due
to the limited solubility of NaC1 in water. It was confirmed that the additional NaC1 did
not proportionally increase the solution conductivity, as the solubility limit of the solution
was reached.
Effect of transmission line dimensions
To further illustrate the usefulness of this experimental platform, the layer thicknesses (tH20
and tF R4) were kept as before and the same measurements were performed with the length
and the width of the line as variables. Figure 4.27(a) presents the insertion phase angle
of a microstrip line similar to the previous one but with its length halved. As expected,
the measured phase shift for both the distilled water as well as the 35 ppt saline solution
is almost halved when compared with that presented in figure 4.26(a). On the other hand,
increasing the width of the line directly affects the slow-mode behaviour and consequently
the phase shift through the microstrip line. This fact is well illustrated in figures 4.27(a),
4.27(b) and 4.27(c).
-50
■
......, -100 •;•••• -100 -
ne ••••••• a)
■
0
O 1vic
• . 0
• e
-150 n -150 -
_c
(0 UT.. .. • • V)
a)
riviya: .•
sie <I,
2 -200 • g,,,mm
1 -200 -
:
2_
0_
v,
-250 • Me/noted-Distilled Water sr.T -250 - • Meesured-Distilled Water
• Measonad-Sallnity• 3sppt •,_ • Measured-Salinity 35 pet
Simulated Salinity. 35ppt
-
••,,,, —v— Shinuleted-Salinity• 35ppt
-A-- Ideal simulated
Behaviour on Er,eff=190 - • kiwi simulated Behaviour on ErAff=190
-300 -300
0 0 0.2 0.4 0.6 0 0 0.2 0.4 0.6


(a) (b)
0
-50-
•
-100-
'En so
1454i .1
. • • .18•0
s•
V -150 -
•
2 -200
ct_
-250 - • Meesuntel-Distillei Water

• Bleasured,Sallnity• 35 pet
Sinneetad-Selinity= 35ppt
- • -- Ideal einulatai Behaviour on Ereittete0
300
0 0 0.2 OA 0.6
Frequency (GHz)
(C)
Figure 4.27: The experimental platform has been set with tFH4=0.5 mm and tH20=20
mm for a 46 mm long microstrip lines. In these figures the measured effective lengths (in
degrees) of the microstrip lines are shown for various line widths: (a) w=0.5 mm, (b) w=1
mm and (c) w=2 mm.
Before devices or circuits exploiting the Maxwell-Wagner polarisation for increased de-
lays can be designed, the characteristic impedance of the slow mode must be examined. It
is surprising, perhaps, that the usual approximation for the characteristic impedance of the
quasi-TEM mode applies:

Z* Pz-% Z°
(4.1)
Er,eff
Where, Z, is the characteristic impedance of the microstrip line on an air substrate and
€r,ef f is the complex effective dielectric constant of the substrate in the slow-wave mode.
It follows that when the effective permittivity is high, the line's impedance drops sig-
nificantly, making impedance matching difficult to implement. In figure 4.28 the measured
mismatch loss for the 96 mm long microstrip line discussed earlier is presented. By obser-
vation, the mismatch loss increases with frequency and obtains a maximum at 250 MHz,
namely at the onset of the slow-wave mode. What is interesting is that all saline solutions
acquire the same maximum. This is consistent with the anticipation that all structures
examined exhibit similar maximum slowing-factor, thus the impedance of the lines will be
rather similar. Finally, the distilled water configuration achieves, as anticipated, a better
loss factor above 250 MHz.
Mismatch Loss
Distilled H2O
0.9 —e—Salinity of Hp=35 ppt
0.8 — B—Salinity of Hp=70 ppt

—9ArSalinity of Hp=140 ppt
0.7
Loss Factor
0.6
0.5
0.4
0.3
0.2
0.1
0
100 150 200 250 300 350 400
Frequency (MHz)
Figure 4.28: Mismatch loss due to the slow-wave mode impedance decrement.
Effect of semiconductor to insulator thickness ratio
Wave propagation on MIS lines is mainly controlled by the substrate's resistivity, the operat-
ing frequency and the semiconductor to insulator thickness ratio. Although, the existence
of the slow-wave mode is mainly dependent upon the first two parameters, it is the lat-
ter that is a direct measure of the speed of wave propagation (equation 2.37). Table 4.5
demonstrates the achievable thickness ratios of the presented experimental platform and the
yielded slow-wave factors. Undoubtedly, at the limit where the thickness of the insulator is

comparable to that of the semiconductor, the slow-wave mode's impact is minimum.
Table 4.5: Calculated controllable range of slow-wave factor

tH2 0/tFR4 Er,eff
157.48 760
78.74 383
40 197
20 101
12.9 67
6.45 36
Measurements presented in figures 4.29(a) and 4.29(b) depict the effect of varying the
thickness of the insulating and semiconducting layers on the relative effective electric per-
mittivity of the line. The measurement is repeated at a number of substrate salinities. For
a semiconductor to insulator thickness ratio of 40, the effective length response of a 96 mm
long transmission line will be as shown in figure 4.26(a). Reducing this ratio by a factor of
two also reduces the er,eff of this structure by a factor of two. The effective length of this
line is indeed observed to be half its reference value (figure 4.29(a)).
-50
•••.,
A...,
......, -100 •••A,,,. ,.., -100
CO • .. AAA., AAA CO
01. • 0
0 IS..
N ,,, 0
• ,..;••• v
g -150 • .:•'• • dlk
1/4 2 . .• .
••,,. =
CI) c0
a)
, .. : . . . 1 :„ .141
•69 cn
i -200 2 -200
0- v,v,,,,
0_
-250 • Measured-DatBed Water • Measured-Disalled Water

• Measured-Saiirdly• 35 pat • Mattered-Salinity= 35ppt
SImulated-Sallnity• 35ppt Ivy, —•— Simutated-Salinity. 35 pot
- A - Ideal simulated Behaviour on Er,eff=100 A - Ideal simulated Behaviour on Er,eIF135
-300 300
0 0 0.2 0.4 0.6 0 0 0.2 0.4 0.6

(a) (b)
Figure 4.29: Unwrapped phase of a microstrip line with L=96 mm and W=0.5 mm. During
measurements, the substrate was formed by distilled H2O as well as a 140 ppt saline solution,
with the thickness of both layers varying. Shown are two configurations: (a) 4/20=10 mm
with tFR4=0.5 mm and (b) .4-120=20 mm with tFR4=1.55 mm.
In the case where the H2O — FR4 ratio is set by utilisation of a 1.55 mm thick FR4
laminate on top of 20 mm of water, the slowing factor will be significantly smaller when
compared with the case where the thinner (i F R4-= 0.5 mm) insulator is utilised. Therefore,
even though these lines are of the same dimensions, their effective lengths differ substantially
(figures 4.29(a) and 4.29(b)). However, as in all cases presented up to now, the same
succession of events is repeated in this case as well, i.e. the effective length of the line is
enhanced at low frequencies, deteriorates as the attenuation maximizes and then increases
linearly according to the skin effect dielectric constant Er,ef f 3kzn of the substrate for which
fr,efht. >> € r,ei fskzn •
In conclusion, in this section the flexibility of this experimental platform on investigating
the propagation characteristics of microstrip transmission lines, has been demonstrated
successfully. Measured results for a wide range of parameters are exhibited which are in
excellent agreement with the theory.
4.5 Summary
The dielectric propagation properties of transmission lines on laminar substrates have been
known to microwave engineers for several decades. In particular, the slow-wave charac-
teristics of such lines have attracted much attention due to the large delays caused by
interconnects on ICs, as well as the unique opportunities for developing novel devices based
on this effects. The slow-wave mode of a metal-insulator-semiconductor transmission line
is dependent on various parameters, including: i) substrate resistivity, ii) the ratio of semi-
conductor to insulator thickness, and iii) the actual dimensions of the device.
In this chapter, the effect of Maxwell-Wagner polarisation mechanism under various

conditions has been thoroughly investigated. Standard devices have been designed and
developed for characterisation purposes on different substrates. All theoretical predictions
that slow-wave propagation is supported in laminar structures were confirmed and the
slowing-mechanism on silicon-based substrates was evaluated.
In the second section of this chapter, the superior electrical performance of GaAs has
been exploited for reducing the high dielectric losses observed in Si-based multi-layer sub-
strates. The irregular interface between the oxide and the GaAs layer is shown to enhance
the effective dielectric response of the structure, while the utilisation of a membrane-based
design demonstrates higher cut-off frequency for the slow-mode.
5 Maxwell-Wagner polarisation mechanism.
Finally, a macroscopic model structure for MIS transmission lines, supporting slow-wave
propagation, has been designed and characterised. This structure uses an aqueous dielectric,
whose conductivity can easily be altered and offers flexibility of replacing the insulating layer
by any laminate material. These design considerations make the experimental structure
ideal for exploiting slow-waves on laminar substrates over a wide range of parameters. The
presented results demonstrate the ability of this platform to support slow-wave propagation,
following all theoretical predictions, and yield the effect of the Maxwell-Wagner polarisation
mechanism under various conditions.
References
[1] "Ansoft corporation, high-frequency structure simulator (HFSS)." Version 9.0.
[2] "Applied wave research, microwave office." Version 7.05.

881, Nov. 1971.
[4] "Mathworks, matlab." Version 7.0.
[5] P. Ladbrooke, MMIC Design: GaAs FETS and HEMTs. Artech House Publishers, 1989.
[6] R. Ding and H. Wang, "Microstructure and optical properties of gaas/sio2 nanogranular
films prepared by magnetron co-sputtering," Materials Chemistry and Physics, vol. 77,
pp. 841-845, Jan. 2003.
[7] S. L. Schiefelbein, N. A. Fried, K. G. Rhoads, and D. R. Sadoway, high-accuracy,

calibration-free technique for measuring the electrical conductivity of liquids," RSI,
vol. 69, pp. 3308-3313, Sept. 1998.
105
Chapter 5
Engineering the Maxwell-Wagner

Polarisation Effect
5.1 Introduction
Every polarisation mechanism supports a resonance. As such, polarisations due to inter-

faces, dipoles, atoms or ions, contribute on the effective electric permittivity of a structure
in a frequency band set by the dimensions evolved. Since the interfacial dimensions are rel-
atively large, Maxwell-Wagner polarisation and thus the associated large dielectric response
is limited to relatively low frequencies.
This chapter is focused on increasing the high frequency cut-off limit of the Maxwell-
Wagner polarisation, i.e. extending the transition of the Maxwell-Wagner mechanism, from
interfacial to atomic polarisation. This is achieved by subdividing the large interfaces
into molecular thicknesses. Several micro-fabrication techniques that are competent in
developing small interfaces are presented and simple devices are utilised for characterisation
purposes.
Finally, since the magnitude of all polarisation modes relies on the physical properties of
the medium, such as the relative electric permittivity Er , the conductivity as and the Debye
lengths LDB, several methods are presented for enhancing the Maxwell-Wagner effect.
1 Debye Length LDB is the scale over which mobile charge carriers screen out electric fields in plasmas and
other conductors. In other words, the Debye length is the distance over which significant charge separation
can occur. Often it is useful to consider the shielding of the fields as a dielectric phenomenon, i.e. it is
the polarisation of the plasma medium, and the associated redistribution of space charge, which prevents
penetration by an external electric field.
106
Engineering the Maxwell-Wagner Polarisation Effect 107
5.2 KOH on Si before oxidation
The semiconductor's surface quality plays an extremely significant role in the extent of
the Maxwell-Wagner polarisation mechanism both in magnitude and frequency. Typically,
good-quality semiconductor-insulator interfaces are smooth and continuous over the areas
where an oxide is grown. However, disturbing the uniformity of the interface proves benefi-
cial in expanding the Maxwell-Wagner mechanism and the associated slowing phenomenon.
This is realized by increasing the roughness of the semiconductor surface by exposing

it in an anisotropic etchant such as potassium hydroxide (KOH) which is widely used for
bulk etching of silicon. Figure 5.1 depicts the Si surface after receiving a KOH treatment
while figure 5.2 demonstrates the roughness of the surface through Atomic-Force-Microscopy
(AFM). Growing an oxide on this defected surface results in an inhomogeneous interface be-
tween the semiconductor and the oxide, or equivalently into a multitude of small interfaces.
Figure 5.1: Si surface after KOH processing. The increased roughness can be observed in
the zoomed microscope photograph.
5.2.1 Process Description
A 4-inch n+ silicon wafer was thoroughly cleaned using the RCA cleaning process. The
sample was then immersed in a 20% KOH solution (1:5 KOH : H2O) for 30 minutes at 60°C.
After rinsing it with fresh DI water the sample was dipped in a 1:1:5 H202 : HC1 : H2O at
70°C to remove KOH contaminants (potassium-ions) and was again rinsed thoroughly with
de-ionised (DI) water. The roughness of silicon's surface was measured to be RA = ±113A.
An oxide was grown by dry oxidation in a carbolyte furnace at 1100°C. The oxide thickness
was characterised with an elipsometer and was measured to be 250 nm thick. The unwanted
deposited oxide on the unpolished surface of Si, was removed by exposing the sample in
a buffered HF solution (14:100 HF : NH3F) while protecting the front-side grown oxide by
spin coating a thick photoresist (AZ-4562) and hard baking it at 120°C for 1 minute. The
etching shielding photoresist was removed by degreasing the sample in Acetone, Isopropanol
and DI water which also refined the sample. An ohmic contact was formed on the back-side
of the sample by coating a 150 nm thick layer of Aluminum using a BOC-500 Physical
vapor deposition system. The characterisation devices' patterns were defined on a thin
photoresist (AZ-5214) layer using standard photolithographic rules followed by evaporating
a 25nm adhesion Ti layer and a 100 nm thick layer of Au. Finally, the excess deposited
metal was lifted-off by dissolving the photoresist in acetone followed by isopropanol and
thorough DI water rinsing.
Figure 5.2: AFM photograph of Si surface after KOH processing.

5.2.2 Results
In a laminar structure, the surface roughness of all the media evolved, can play a signifi-
cant role in the overall performance of a device. Especially nowadays, with the strive to
scale down standard CMOS technologies, surface roughness scattering becomes the limiting
scattering mechanism for the channel mobility. Lately, these issues have been addressed by
processing techniques that reduce silicon's surface roughness down to RA < ±1A [1].
On the other hand, a curt surface is in favor with interface states and consequently a
high trapped charge density (Qit ) should exist. Therefore, the deliberate perturbation of
the substrate's uniformity results in a significant increase of the distributed capacitance
of the line which apparently increases the effective effective length as well as the observed
group delay of the line.
As shown in figure 5.3, a 100 om long MSTL based on a Si02 /Si substrate demon-
strates a 20% larger effective length and group delay, when the substrate's surface has been
processed with KOH2. At low frequencies where there is a single mode propagation, the
propagation characteristics of the "processed" microstrip line are comparable to those that
a microstrip line, with twice the length, exhibits on a Si02/Si system with a rather smooth
interface. At relatively high frequencies, where the slow-wave mode is not supported, the
excess interface charge becomes insignificant and thus the measured results of the line look
like that of a 100 ttm long device on the polished Si.
As expected, in the low GHz region, the measured insertion loss of the KOH processed
device is worse than the propagation loss of the standard device due to the increased effective
slowing factor of the line (figure 5.4), which is also valid within the 15-30 GHz frequency
band. However, within these two bands, the insertion loss for all specimens alters in the
same manner. Surprisingly, within the 3 to 15 GHz band, the KOH processed sample
exhibits almost 3 dB/dec less attenuation when compared with the standard sample. It is
speculated that this is an indication that the upper frequency limit of the slow-mode has
been extended into a higher value due to the smaller physical dimensions evolved in the
polarisation mechanism. Therefore, the forward transmission coefficient (S21 ) of a 100 itm
long MSTL, based on a KOH processed geometry that experiences the same slowing effect
as in the standard, should be as projected in figure 5.4.
2 Note that the substrate's conductivity as well as the oxide thickness are the same for both specimens.
150
—0-- Standard sample L=100µm
-o—KOH processed Si L=100µrn
a
9-- loo
co
To
O
50
0
0
0
40
E
E —0—Standard sample L=100µrn
■ —0— KOH processed Si L=100µrn
—
_c 30
0)
° 20
a)
t 10
Lu
0

01 10
Frequency (GHz)
Figure 5.3: Measured group delay and effective length of a 100pm long MSTL on Si02
based on standard Si and on KOH processed Si.
0
U)
0 15
t
a) -20 --m— Standard sample L=100Rm
—0— KOH processed Si L=100pm
Standard sample L=200Am
-25 —. Projection
30
01 10
Frequency (GHz)
Figure 5.4: Measured insertion loss of a 100pm long MSTL on Si02 based on standard Si
and on KOH processed Si.
5.3 Unbuffered HF 5% on Si/Si02
The Maxwell-Wagner polarisation effect can be expanded by introducing irregularities in

the Si — Si02 interface that act as local small interfaces. In this section the effect that
similar defects have when they are located on the oxide's surface is investigated. The defects
are created by attacking the oxide's surface with an unbuffered 5% HF solution that tends
to be rather abrasive as shown in figure 5.5.
Figure 5.5: Microphotograph of oxide surface defects introduced by unbuffered HF wet

etching.
A 4-inch n-type silicon wafer was thoroughly cleaned using the RCA cleaning process. A
250 nm thick oxide was grown on top by wet oxidation at 1100°C. The sample was then
immersed in a 5% HF solution (5:100 ml HF : H2O) for 5 minutes at room temperature.
The roughness of the oxide surface was measured to be RA = +76221. An ohmic contact
was formed on the back-side of the sample by evaporating a 150 nm thick layer of Al. The

characterisation devices' patterns were defined using standard photolithographic rules and
a layer of Ti/Au (25nm/100nm) was utilised as the conducting media.
5.3.2 Results
Solutions based on hydrofluoric acid are commonly used as SiO2 etchants, therefore treating
Si/Si02 specimens with such solutions alters the semiconductor to insulator thickness ratio.
Eventually, this leads into a higher slowing factor of the structure. However, during this
investigation an unbuffered HF solution which is shown to have a low oxide etching rate, yet
quite abrupt [2], was utilised. Following the previously described process, it was confirmed
that the oxide thickness was reduced from 250 nm to 200 nm which is translated into a
doubling of the observed slowing factor of the utilised structure. This is well illustrated in
figure 5.6 where the effective length as well as the group delay is plotted for two microstrip
lines that have the same dimensions but are based on different substrates.
250
—0— Standard Sample
13. 200 —0— 5% HF processed Si
>. 150
1)
100 .°
o.
o 50
0
60
—0—Standard Sample
--0— 5% HF processed Si
40
20
0
0 1 1 10
Frequency (GHz)
Figure 5.6: Measured group delay and effective length of a 100 pm long MSTL on good
quality SiO2 as well as surface defected SiO2 on Si.
Apparently, the HF processed device exhibits twice the effective length when compared
with the model device and accordingly it possesses almost double group delay at low fre-
quencies. Also the effective bandwidth, in which the line is dispersionless, is halved. This
is due to the thickness variation of the evolved oxides and clearly it has nothing to do
with the defected oxide surface. Figure 5.7 depicts the measured insertion loss of the same
devices. The device on the HF processed structure exhibits a larger insertion loss than the
standard device. This deviation fully complies with the observed large slowing factor of
the HF processed specimen. And since the propagation characteristics of both specimens
demonstrate an identical variation it can be safely concluded that the planted defects on the
oxide's surface do not affect the Maxwell-Wagner polarisation effect neither in magnitude
nor in bandwidth.
10
U)
U)
0
C
0
t
-20 --s— Standard Sample
—0— 5% HF processed Si
30

01 1 10
Frequency (GHz)
Figure 5.7: Measured insertion loss of a 100 pm long MSTL on good quality SiO2 as well
as surface defected SiO2 on Si.
5.4 Reactive Ion Etching of Si before oxidation
Reactive Ion Etching (RIE) is typically used to remove deposited material on wafers. De-
pending on the gases evolved in the process, a variety of materials can be etched. Carbon
Hydro-Trifluoride (CHF3 ) is commonly used for removing oxide layers, however as shown in
[3], very often the underlying Si surface can be affected. In this section the potential of engi-
neering the Maxwell-Wagner polarisation by developing specimens in which the substrate's
surface has been damaged in a controlled way is exploited.
A 4-inch n-type Si wafer was diced into four pieces and the samples were degreased using
Trichloroethylene (TCE), Acetone and Methanol and rinsed in DI water. The samples were
placed in a reactive ion etching chamber and were processed as shown on table 5.1. Then a
250 nm thick oxide was grown on top of all the specimens by wet oxidation at 1100°C. An
ohmic contact was formed on the back side by evaporating 1500A of Aluminum. Finally,
the top metal patterns were transfered via photolithography on a MA6/BA6 Suss Microtech
mask-aligner and the top metal layer was formed by 25 nm of Ti and 100 nm of Au.
Table 5.1: RIE processing parameters.

# Gases Flow(sccm) Power(W) Pressure(mT) Time(min)
1 CHF3/Ar 25/25 200 80 0.6
2 CHF3/Ar 25/25 370 80 1
3 CHF3 30 200 40 2.5
4 CHF3/02 25/25 100 100 10
5.4.2 Results
Through this investigation, four specimens were developed in accordance with the different
RIE processes presented in table 5.1. All the samples were examined under an optical
microscope, using the Nomarski3 mode, and surface defects were only identified on the
fourth specimen. This is attributed to the long exposure time as well as the use of 02.
The MSTL on the defected substrate exhibits a higher effective slowing factor from the
one on the "clean" substrate. This behaviour is observed at relatively low frequencies as
depicted on figure 5.8. Once more, this effect is attributed to the defected surface that
introduces an accessory capacitance, causing an extra 10° phase-shift at 1 GHz. On the
other hand, this excess capacitive loading has an accompanying unfavorable impact on the
overall performance of the line. As demonstrated on figure 5.9. the insertion loss of the
defected MSTL deteriorates from 1 to 3 dBs within the frequency band where the extra
capacitance is observed.
3 Differential Interference Contrast mode. This mode is ideal for identifying defected surfaces.
0
Unwrap Phase Shift (Deg)
-40
-80
—a—Standard Sample
—0— CHF3 & 02 processed Si
-120 —
-160
01 i 10
Frequency (GHz)
Figure 5.8: Measured unwrapped phase-shift of a 100 itm long MSTL on SiO2 based on
standard Si as well as on RIE defected Si.
0
0
_i
C
—a—Standard Sample
c —0— CHF3 & 02 processed Si
30

01 10
Frequency (GHz)
Figure 5.9: Measured insertion loss of a 100 pm long MSTL on SiO2 based on standard Si
as well as on RIE defected Si.
5.5 Au nano-spheres embedded in the oxide
So far, the implantation of defects that serve as miniature interfaces, which under certain
parameters can extend/expand the slowing effect of the exhibited polarisation mechanism,
has been investigated. Some very promising results have been presented demonstrating that
defects in the semiconductor-insulator interface affect the interfacial polarisation mechanics.
Being intrigued by these experiments, more specimens were developed on highly controlled
defected substrates.
This section was inspired by the metamaterials community, embracing a three dimen-
sional approach for defects, instead of a defected monolayer. Metal "nano-spheres" are
distributed in the insulating layer, emulating a crystal lattice of an artificial material. Fig-
ure 5.10 illustrates the top of the oxide layer as well as the distribution of the tiny spheres
in the oxide layer. Additionally, the surface roughness is acknowledged in figure 5.11 by aid
the of Atomic Force Microscopy (AFM).
Figure 5.10: Au nanodots on SiO2 surface. Zoomed section is a SEM microphotograph

demonstrating Au spheres dissolved in the oxide layer.
A 4-inch n-type Si wafer was degreased using TCE, Acetone and Methanol and rinsed in
DI water. A Si02 layer was evaporated on top, of total thickness 2500A in 3 stages (1000
A, 1000 A and 500 A) with a pause of 15 minutes between them for cooling. A 50A layer of
Au was evaporated on top without breaking the vacuum. A rapid thermal annealing (RTA)
process was utilised for embedding the Au layer in the oxide. The specimen remained at
500°C for 30 seconds and then the system was ramped up to 900°C in 1 minute and held
there for 5 minutes. During the process nitrogen flowed in the chamber (500 sccm). An
ohmic contact was formed on the back side by evaporating 1500A of Aluminum. Finally,
the top metal patterns were transfered via photolithography on a MA6/BA6 Suss Microtech
mask-aligner and the top metal layer was formed by 25 nm of Ti and 100 nm of Au.
s NanoScoc-,
Scan size 500.0 nm
Scan rate 0.2501 Hz
Number of samples 512
Image Data Height
Data scale 150.0 nm
Engage X P05 -19783.4 um
Engage Y Pos -42151.3 um
LI view angle
-:,t57 light angle
411141ppri”1 - nm
400
100 0 °
X: 100.000 nm/div
150.000 nm/div
,u nanodots 'n ,g1.0 -

;1539.001
Figure 5.11: AFM photograph of Au nanodots on Si02 surface.

5.5.2 Results
Since the diameter of the spheres are very small, when compared with the evolved dimen-
sions of the whole system, it is expected that there will be minimal contribution on the
interfacial polarisation mechanism. By observation (figure 5.12), this is indeed the case. A
100pm long MSTL on a Si02/Si substrate, both on a standard oxide and a nano-sphere
diffused oxide, demonstrates an almost identical effective length, as demonstrated in figure
5.12.
150
Standard Sample
Nanodots
Group Delay (ps)
100
50
0
40
—0—Au Nanodots
30
20
10
0
01 10
Frequency (GHz)
Figure 5.12: Measured group delay and effective length of a 100 tim long MSTL based on
standard SiO2 and on SiO2 with buried Au nanodots.
At relatively low frequencies, the reported insertion loss of the aforementioned devices
are identical. Surprisingly enough, in the GHz region, there exists a small variation in the
measured losses. The device that is based on the Au diffused oxide exhibits on average 1 dB
less losses over the whole region that this variation is observed (see figure 5.13). It should
be noted, however, that the insertion loss of both devices varies with a similar pattern,
denoting a small broadening of the device's bandwidth.
10
0)
_J
t
.
a.) -20 —♦— Standard Sample
Au Nanodots
30
01 10
Frequency (GHz)
Figure 5.13: Measured insertion loss of a 100 pm long MSTL based on standard Si02 and
on Si02 with buried Au nanodots.
5.6 Germanium spheres embedded in the oxide
In this section the diffusion of Ge in the oxide layer was attempted. This results in a rather
similar structure to the one presented in section 5.5. Accordingly, the compatibleness of
this structure with the propagation of the slow-wave mode and all associated effects is
investigated.
A thin layer (50 nm) of strained Silicon-Germanium (SiGe), with 22% of Ge, was grown on
a 6-inch N+ silicon wafer. A 46 nm thick Si02 layer was grown on top by dry oxidation
at 1100° C. The high temperatures of the oxidation process forced Ge to diffuse into the
oxide. An Aluminum (150 nm thick) ohmic contact was evaporated on the back side of
the specimen, acting as the ground plane. The top metal consists of a Ti and a Au layer
with thicknesses 25 nm and 100 nm respectively. The devices were formed by standard
photolithography rules.
5.6.2 Results
The diffusion of Ge in the oxide layer proved to be not as reliable as the diffusion of Au
spheres. It was discovered that the SiGe layer was not fully diffused in the oxide. Also,
there were areas that the diffusion rate varied significantly causing irregularities on the
oxide surface. Figure 5.14 demonstrates two distinct areas of the specimen's surface. The
area, shown on the left side, exhibits an abnormally rough surface while the one on the
right side depicts a rather smooth surface.
Figure 5.14: AFM photograph of diffused Ge in SiO2 .
Figure 5.15 demonstrates the measured group delay and the effective effective length of
a device based on the model structure as well as on the Ge/SiO2 /Si laminate. A small
slowing-factor was retraced in the low MHz band but it does not comply with the expected
large effective electric permittivity that the dimensional contrast of the structure dictates.
Similarly, the measured loss, depicted in figure 5.16, does not have a similar frequency
response with the model specimen, implying a mixed mode propagation. This is also af-
firmed by the fact that the low-frequency losses of the DUT appear to be larger than the
model device, which is in direct contradiction with the relatively small effective effective
length of the line.
180
—0— SiGe sample
120
co
60
o-
60
—0— SiGe sample
40
20
0

01 10
1
Frequency (GHz)
Figure 5.15: Measured group delay and effective length of a 100 um long MSTL on Si02
based on standard Si02 and on Ge rich SiO2.
— 10
O
C
O
a) -20 --•— Standard Sample
C —.— SiGe sample
30
0 1 1 10
Frequency (GHz)
Figure 5.16: Measured insertion loss of a 100 pm long MSTL on Si02 based on standard
Si02 and on Ge rich SiO2.
5.7 Na+ Ions on Si/Si02
Intentionally planted defects could be used for exploiting the Maxwell-Wagner polarisation
mechanism. It should be noted though, that one of the adjoining effects of imperfections
is the formation of interface trapped charges that reflect as an increase of the per unit
length capacitance of the DUT. Analogously, a microstrip line can be capacitively loaded
by simply inducing mobile charges, such as Nat, in the oxide layer. This is the approach
that is delineated in this section.
The MISM structure was formed by a 250 nm thick oxide on an n-type, 4-inch Si wafer with
525 pm thickness. The SiO2 layer was grown in a carbolyte furnace at 1100°C. 20 grams
of NaCl were dissolved into 100 ml of DI water forming a 20% saline solution. The solution
was warmed up to 110°C. The sample was then immersed in the saline solution and was
kept there for 30 minutes at the same temperature. After thoroughly cleaning the sample,
the ground blanket metallization was formed by evaporating an Aluminum layer of 150 nm
thickness, that acts as an ohmic contact. The top metal of the characterisation devices was
transferred on the oxide layer via standard photolithographic rules. An adhesion layer of
Ti followed by a 100 nm thick Au layer was coated on top and the devices formed by lifting
off the excess deposited metal.
5.7.2 Results
The measured results presented in figures 5.17 and 5.18 confirm all expectations that oxide
mobile charges have a direct affect on the Maxwell-Wagner polarisation relaxation. It can
be observed that the measured propagation characteristics of a 200 in long MSTL fluctuate
with a reciprocal fashion for both cases. As expected, the transmission line on the Si02 /Si
substrate, where Na+ ions have been induced, experiences a 15% dimensional improvement
at the cost of 1 dB higher losses. This translates into a 200 MHz narrowing of the effective
bandwidth.
150
, —0— Standard Sample
—c— Sample with Na'
TD
in 60
a.
30
0
50
E
E
--- 40
.c
co
c 3
a)
-.1
O 20
>
a 10
w
0
01 1 10
Frequency (GHz)
Figure 5.17: Measured group delay and effective length of a 200 iim long MSTL on Si02 /Si
sample before and after inducing Na+ ions.
0
_1
c
O
a)
cn
-20 —s— Standard Sample
c
--•— Sample with Na+
30

01 10
Frequency (GHz)
Figure 5.18: Measured insertion loss of a 200 pm long MSTL on Si02 /Si sample before
and after inducing Na+ ions.
5.8 Ar, As and P Implantation on Si
Inhomogeneous doping has been shown in [4] to minimize the overall dielectric loss. On the
other hand, the induced charge causes an accumulation of space charge which in turn intro-
duces an added layered capacitance in the MISM system. However, this layered capacitance
can be further enhanced, and hence the slowing factor of the structure can be increased by
introducing some defects alongside with the implantation process. This can be achieved by
implanting large enough atoms4 that can cause dislocations in the Si crystal.
Six 4-inch Si wafers were thoroughly cleaned in HNO3 99% for 10 minutes at ambient
temperature, DI water and 10 more minutes in HNO3 99% at 110°C. The native oxide was
removed by dipping the samples in a 0.55% HF solution for 4 minutes at ambient temper-
ature. P, As and Ar were selected as the implants and the implantation was performed on
2 wafers for each atom. The utilised energy and dose were the same for all implants, E=10
KeV and D=1015 ions I cm' . According to these parameters, the predicted range of ions in
Si is as illustrated in figure 5.19. The SiO2 layer was formed by wet thermal oxidation at
1100°C. The oxide thicknesses were characterised with an elipsometer and were measured to
be 310.35 nm, 313.1 nm and 285.21 nm on the P, As and Ar implanted wafers respectively.
The unwanted deposited oxide on the unpolished surface of Si, was removed by exposing
the sample in a buffered HF solution (14:100 HF : NH3F) while protecting the front-side
grown oxide by spin coating a thick photoresist (AZ-4562) and hard baking it at 120°C,
for 1 minute. The etching shielding photoresist was removed by degreasing the sample in
Acetone Isopropanol and DI water which also refined the sample. An ohmic contact was
formed on the back-side of the sample by coating a 300 nm thick layer of Aluminum using
a BOC-500 Physical vapor deposition system. The characterisation devices' patterns were
defined on a thin photoresist (AZ-5214) layer using standard photolithographic rules fol-
lowed by evaporating a 25 nm adhesion Ti layer followed by a 100 nm thick layer of Au.
Finally, the excess deposited metal was lifted-off by dissolving the photoresist in acetone
followed by isopropanol and thorough DI water rinsing.
4 This is relative to the Silicon atom.
9.00E+020
8.00E+020
7.00E+020
6.00E+020
o 5.00E+020
A
4.00E+020
o
3.00E+020
- --As
P
2.00E+020
b
1.00E+020 •
•
,.0
0.00E+000

0 100 200 300 400
Depth (Angstroms)
Figure 5.19: Profile code retained dose estimates.
co
ca.
D
--•—Ar
1 - o-As
0
0 50 100 150 200
Depth (Angstroms)
Figure 5.20: Displacements per atom.

5.8.2 Results
The three selected implants P, Ar and As, are considered to be "heavy" atoms, when com-
pared with Si, and they all have different atomic weights. As such, the expected dislocations
in the silicon's lattice should vary accordingly in each case. As a quantitative figure, the
effect of the formed defects can be evaluated by the displacements per atom (dpa) in the Si
crystal lattice. The predicted dpa for each case is illustrated in figure 5.20.
The model structure is based on a 250 nm thick oxide on Si. However, the grown oxides
on the 3 disparate doped Si substrates are approximately 25% thicker for P and As and 14%
thicker in the Ar case. In consequence, the predicted slowing factor in the P, As and Ar
case should accordingly be 11.3%, 11.9% and 6.7% less. Nonetheless, as depicted in figure
5.21, this is not the case. The measured effective length of a 100pm long MSTL is slightly
less than that of the model structure while in the Ar implanted case it is significantly larger.
Qualitatively, these results are associated with the increased layered capacitance due to the
increase of the accumulated space charge.
Some very interesting results are presented in figure 5.22. In the P and As doped Si
cases, the effective dielectric loss is expected to be slightly less than in the model case
which is in perfect accordance with the low-GHz region results, presented in figure 5.22.
Furthermore, since the Ar doped Si structure experiences a larger slowing-factor than the
model structure, one would expect that the dielectric losses would be analogously larger.
However, the measured insertion loss at relatively low frequencies, presented in figure 5.22,
signifies an almost identical response with the model MISM structure. Yet the compelling
findings occur in the middle GHz region where the measured insertion loss of the doped
structures varies in a different manner than the standard sample, with an insertion loss
improvement of better than 10 dB.
It is strongly believed that the presented results are quite encouraging, indicating that
the Maxwell-Wagner polarisation mechanism can be efficiently tailored to suit all desires
for large "miniaturization factors" while the dielectric losses remain in satisfactory levels.
150
—ID-- Standard Sample
— -- As
Group Delay (ps)
Ar
100
—P
50
0
40
--m— Standard Sample

—s— As
30
- Ar
P
20
10
0

01 10
Frequency (GHz)
Figure 5.21: Measured group delay and effective length of a 100 /.cm long MSTL on Si02/Si
and on Si02/Si where the Si surface has been doped with Ar, As and P.
0
10
CO
U)
0
—J
0
t
-20 —0—Standard Sample
- As
Ar
- P
30

01 10
Frequency (GHz)
Figure 5.22: Measured insertion loss of a 100 tm long MSTL on Si02/Si and on Si02/Si
where the Si surface has been doped with Ar, As and P.
5.9 SrTiO3 on Si02/Si Substrate
The large effective electric permittivity of a laminate structure, under the influence of large
interfacial polarisations, strongly depends on the characteristics of the individual layers.
Clearly, the electric permittivity of the individual layers is one of the most significant
parameters. On that basis, the dielectric enhancement of MISM structures was investigated
by introducing a ferroelectric layer.
A target for the Pulsed Laser Deposition (PLD) system was made from ceramic powder
prepared using the mixed oxide route. Thin films were grown by laser ablation (Neocera
PLD system with a Lambda Physik KrF laser, A=248 nm) on Si02/Si (250 nm/525 pm)
substrate. The substrates were secured by silver paste on to the stainless-steel resistive
heater with a target to substrate distance of 70 mm. Thin films were deposited from 20
mm diameter stoichiometric STO target in an oxygen pressure of 300 mTorr. The substrate
temperature (T8 =740 K) during deposition was controlled using a thermocouple embedded
in the heater. The energy density of the laser spot (2x10 mm2 ) was 2.5 J/cm2. From the
sample thickness measured using a Dektak 11A the film growth rate was estimated to be
approximately 0.03 nm/pulse. The total number of pulses was 4000 with a repetition rate of
8 Hz. Once the ablation was over, the samples were then cooled down at a rate of 10°C/min
in an oxygen rich environment (760 Torr).
5.9.2 Film Characterisation
The sample was examined by conventional X-ray diffraction, performed on a PAnalytical

X'Pert MRD system. It was revealed that the samples are single crystal SrTiO3 films with
(001) and (110) oriented grains. A typical XRD pattern of SrTiO3 thin films is presented
in figure 5.23. It shows set of (h00) and (hk0) peaks. The precise analysis of their positions
confirmed that both set of peaks are belonging to the same SrTiO3 crystal cell with lattice
parameter 3.904 A which indicates the high crystal quality of the deposited films.
Counts 557445 (20-75 deg 70s pe 0.01 step)
100
50
111111111111111111111 1111111
30 40 50 60 70
Position r2Theta]
Figure 5.23: X-ray diffraction analysis of the processed sample.
5.9.3 Results
At room temperature, strontium titanate5 has a very large dielectric constant E r

However, the crystal lattice of deposited thin films tends to be strained, thus exhibiting
lower Er values. In this case, it was calculated that the SrTiO3 layer demonstrates an Er
200. Since this is a multilayer system, the effective electric permittivity of the insulating
layers of the structure will be approximating the mean value of the independent dielectric
constants evolved. It is therefore, expected that the overall Maxwell-Wagner effective per-
mittivity of the structure will be significantly larger when compared with a Si02/Si system
of comparable thicknesses (i.e. isrTio, tox = tSi02,equivalent)•
The group delay and effective length of a 100 pm long microstrip line on the fabricated
SrTiO3/Si02 /Si substrate are measured and are summarised in figure 5.24. It is apparent
that this particular structure supports the Maxwell-Wagner polarisation, since the large
measured GD and EL indicate a larger attainable slowing factor than indicated by the
ferroelectric.
° The SrTiO3 is considered to be bulk material.
150
Standard Sample with cilt.=2100
—0— SrTiOiSi02 Sample
100 with tsmolts,olts,=120nm/250nm/525pm
--- Standard Sample with t../t=1140
a)
50
7
0
0
40
E —e— Standard Sample with t/t=2100
—e—SrTiO/S102 Sample
cr) with tw,03/ts,02/ts,=120nm/250nm/5251.tm
- Standard Sample with cit. j-1140
20
a)
>
0
a)
LL
01 1 10
Frequency (GHz)
Figure 5.24: Measured group delay and effective length of a 100 pm long MSTL on a
Si02/Si and a SrTiO3/Si02 /Si system.
0
10
Cl)
0
-1
0
w
o 20
—e— Standard Sample with tsubItox=2100

—0— SrTiO3/S102 Sample
with twiO3/ts0Jts,=120nm/250nm/525µm
—e— Standard Sample with tsoltox=1140
30 I I g

01 1 10
Frequency (GHz)
Figure 5.25: Measured insertion loss of a 100 pm long MSTL on a Si02/Si and a
SrTi031Si021Si system.
For comparison, the measured GD and EL of a similar microstrip line based on standard
Si02/Si substrates are depicted as well in figure 5.24. These specimens have a semiconduc-
tor to insulator thickness ratio (t sub/tox ) of 2100 and 1140 respectively. In the SrTiO3 case,
the corresponding thickness ratio is t 1420. -ox = However, the presented results demon-
strate significantly larger er,ef f than the specimen of comparable thickness ratio (t sub/t ox=
1140). As expected, the bandwidth in which the exhibited large cneff is rather constant,
is narrower for the ferroelectric specimen. Evidently, in the SrTiO3 specimen, dispersion
seems to be rather crucial.
Surprisingly, figure 5.25 illustrates that the measured insertion loss of the MSTL on the
proposed structure is quite similar to the Si02/Si sample of t- sub, /t0x=
- 1140, which is in
direct disagreement with the low er ,ef f this specimen exhibits. It can be therefore concluded
that although ferroelectrics are not compatible with standard CMOS technologies they can
be advantageously used for achieving even higher slowing factors at the cost of less dielectric
losses.
5.10 Summary
The propagation characteristics of layered morphologies are dictated by the electric and
magnetic characteristics of the materials involved, as well as the physical dimensions of
the individual layers. More precisely, it is the mass ratio of the media involved that set
the effective dielectric response of the structure. Since, the interfacial dimensions are quite
large, the Maxwell-Wagner relaxation and thus the large dielectric response are limited to
relatively low frequencies.
In this chapter novel substrates are presented that are optimized to enhance the slow-
mode to meet the three, partially contradictory design targets: minimize guide wavelength,
minimize loss tangent and maximize the upper mode cut-off frequency. This is achieved
by applying various processing techniques for substrate engineering. This work's aim is to
subdivide the large interface areas into smaller interfaces or induce excess mobile charges
that contribute to the overall effective dielectric response of the medium at higher frequency
bands than the Maxwell-Wagner polarisation.
This chapter clearly demonstrated that suitable polarisation states can be induced to
a structure, by utilisation of substrate processing techniques, and thus extend the effect
of the Maxwell-Wagner polarisation to higher frequencies. This can also be paraphrased
into large attainable effective dielectric constants at the cost of relatively low corresponding
dielectric losses.
References
[1] D.-Y. Lee, C. C. Chen, C. H. Huang, P. S. Lim, M. H. Chan, M. S. Yeh, C. S. Huang,

H. J. Tao, and Y. J. Mii, "Novel silicon surface pre-treatment (sspt) technique for
cmos device performance boosting," in International Symposium on VLSI Technology,
Systems and Applications, pp. 42-43, Apr. 2008.
[2] K. R. Williams, K. Gupta, and M. Wasilik, "Etch rates for micromachining processing-
part ii," Journal of Microelectromechanical Systems, vol. 12, pp. 761-788, Dec. 2003.
[3] H.-H. Park, K.-H. Kwon, S.-H. Lee, B.-H. Koak, S. Nahm, H.-T. Lee, K.-I. Cho, O.-J.
Kwon, and Y.-I. Kang, "A study on modified silicon surface after chf3/c2f6 reactive ion
etching," ETRI Journal, vol. 16, pp. 45-57, Apr. 1994.
[4] J. J. Kucera and R. J. Gutmann, "Effect of finite metallization and inhomogeneous

dopings on slow-wave-mode propagation," Transactions on Microwave Theory and Tech-
niques IEEE, vol. 45, pp. 1807-1810, Oct. 1997.
133
Chapter 6
Miniaturising Passive Components

by Utilisation of MISM Microstrip
6.1 Introduction
The trend for miniaturisation necessitates the establishment of a technique for developing
passive devices that have small physical dimensions but behave as electrically larger devices.
In this chapter,the exploitation of large interfacial polarisation occurring in anisotropic
structures for enhancing the effective electric permittivity of the substrate is proposed.
These layered substrates are constituted by a succession of conductive and insulating lay-
ers, which is a typical characteristic of CMOS integrated circuits and thus proves to be
compatible with monolithic implementations. This technique is unique as it offers the ad-
vantage of tuning the dielectric behaviour of the substrate by appropriate design.
Maximising the effective electric permittivity of a structure, also maximises the imagi-
nary part of the dielectric constant. Thus, as the imaginary part of er,ef f is associated with
the loss tangent (tangy) of the material, the associated dielectric loss gets larger, because
the real and imaginary parts of Er are related by the Kramers-Kronig relation 2.26. The
increased losses will slightly affect our design as the losses reach a maximum outside the
frequency band of interest. This offers the flexibility of designing for a maximum scaling
factor or for a minimum loss, when sensitivity is an issue. On the other hand, what used
to be a limiting factor to overall system performance can be advantageously used in several
devices where attenuation is desired.
134
Miniaturising Passive Components by Utilisation of MISM Microstrip 135
6.2 Miniature Delay Lines
Delay lines offering a broad range of phase shifts are desirable for emerging applications such
as portable phased array antennas. Additionally, they are often employed for minimising
the clock skew in order to avert false switching of digital logic circuits. To be attractive,
a delay line should provide large predictable phase shifts, exhibit low loss, have a regular
shape and be compact in size. The latter one has focalised research into compact designs
such as the meander or serpentine line [1, 2] as well as the spiral lines [3, 4]. Obviously,
the phase shifts provided by these "true" delay lines are proportional to their dimensions
which is a limiting factor when large phase shifts are required.
Attempts to squeeze the dimensions of unit elements of a meander line may result in
a size effective device but also enhances the coupling between adjacent elements [5], [6].
Eventually, if the coupling coefficient is large enough, less delay than expected is observed
which can cause false switching of active devices or, on a phased array system, radiation in
an unwanted direction.
In this work, the enhancement of the electrical performance of conventional microstrip

delay lines is investigated by exploiting high dielectric substrates. It is well known that
the effective dimensions of passive components are inversely proportional to the square root
of the effective dielectric constant of the substrate, on which these components are built.
Therefore, a transmission line built on a substrate of high electric permittivity will be corn-
parable with a larger line built on a low dielectric medium. Even though this approach is
utilisation of this method, exotic high-K dielectrics, such as ferroelectrics, can be avoided.
Instead, a band limited solution to the problem of realising a high dielectric constant sub-
strate is recommended by application of the Maxwell-Wagner polarisation mechanism.
Microstrip lines on multilayered substrates are known to support a number of prop-

agation modes. At least one of these modes exhibits low propagation losses and a very
low phase velocity. In the following section, miniaturised delay lines are presented which
demonstrate the potential of the proposed method. Several specimens were developed so
that the dimensional dependence of the slowing factor can be observed. All theoretical
predictions are verified through a comparison between simulated data, based on numerical
models, and measured results.
6.2.1 Design and Fabrication
The dominance of the slow-wave polarisation mode depends upon a wide range of pa-
rameters, such as substrate resistivity, semiconductor to insulator thickness ratio and the
operating frequency. For this reason several specimens, invoking as many possible design
variables, were fabricated and characterised.
3526 2810 X100 l B0iJm 14048

Figure 6.1: SEM microphotograph of the designed delay lines.
Miniaturised delay lines were fabricated on highly conductive silicon substrates with
a substrate conductivity o-sub=20 S/cm. SiO2 was thermally grown and etched down to
various thicknesses (0.7-1.4 gm). Then 100 pm long Ag microstrip lines were deposited on
top of the oxide (figure 6.1), with their widths varying from 6 ,um up to 120 ,um. Finally,
the back metallization was formed by evaporating a Cr adhesion layer followed by Au. All
these characteristics allowed us to extend the upper limit of the slow-wave mode up to 3
GHz and observe a rather constant cneff up to at least 2 GHz.
6.2.2 Experimental Results and Discussion
Simulations were carried out using a 2.5D solver included in AWR's Microwave Office [7] and
a 3D FEM solver (Ansoft's HFSST M [8]). In order to verify all expectations two different
models were set up. The first one was based on an artificial dielectric with frequency
dependent er ,eff and tanS. These set of values were given through an approximate model
for the slow-wave behaviour of the MISM structure and the actual line's dimensions (100x60
pm) were used. The second model was based on the predicted effective dimensions of
the delay line and a standard Si substrate with a constant Er and tans was used. All
measurements were carried out using the following apparatus: HP8753D Network Analyzer,
Cascade Microtech probe station, 200/cm pitch coplanar RF probes and calibrated on wafer
with a Cascade Microtech ISS.
Phase Shift
-F tox=a7um Measured
4-tox=0.7um Simulated
(analytical model)
-I- Simulated
Eq. Dimensions
-v• tox=1.4um Measured
++44-
▪ tox=1.4um Simulated
-20 (analytical model)
-0 Simulated
Eq. Dimensions
-25
0 1 2
Frequency (GHz)
Figure 6.2: Measured vs Simulated (analytical & equivalent model) phase shift.
Measured results, as well as simulated data, make it evident that the 100 pm long
transmission line on a 0.7 pm thick insulator can provide a phase shift as large as 16.5°
at 2 GHz (figure 6.2). This is significantly larger than the phase shift the same line would
provide if it was built on a standard substrate (0 0.6°). Clearly this is due to the ultra-
high E r ,e f f this substrate morphology exhibits. On the other hand, if the semiconductor to
insulator thickness ratio is reduced by thickening the oxide layer (tox ) from 0.7 Am to 1.4
pm, it would be expected to get less phase difference than the previously presented delay
line, as this line will be electrically shorter. Indeed, the same delay line built on the layered
substrate of tox=1.4pm and tsz=525t,tm shows a phase shift of less than 3° at the same
frequency.
Up to 2 GHz, the phase shift increases linearly (within ±1°) for the line with the large
slowing factor whilst for the line with the lower slowing factor the phase shift increases
almost linearly (±0.2°) but for a broader frequency range.
Forward Transmission Coefficient
-0.2
-I- tox=1.4um Measured

-8-tox=1.4um Simulated
(analytical model)
Simulated
Eq. Dimensions
-0.6
-0.5
-1
tox=0.7um Measured
tox=0.7um Simulated
(analytical model)
+ Simulated
Eq. Dimensions
-2.5
0 1 2 3
Frequency (GHz)
Figure 6.3: Measured vs Simulated (analytical & equivalent model) 821 .

A measure of how lossy these delay lines are is the forward transmission coefficient (S21)
which is plotted on figure 6.3 for both lines. Measured and simulated data differ slightly at
the lower frequency range whereas for high frequencies, i.e. close to the slow-wave to skin
effect transition region, 821 has a 0.2 dB variation for the small slowing factor line and a
1 dB variation for the electrically larger line. This behaviour is very much expected as the
larger the Er.,ef f , and consequently the larger the scaling factor, the higher the dielectric
losses become (figure 2.15).
It is evident that the measured data are in very good agreement with the simulation
results which were based on the analytical slow-wave model. Additionally, the equivalent
dimensions model, is in good agreement, especially for the phase response (figure 6.2), which
indicates that the designed delay line behaves as a much larger one.
In conclusion, the fabricated miniature delay line is based on a conventional microstrip

line on a multilayered substrate that can be easily fabricated in standard CMOS tech-
nologies. The large interfacial polarisation, occurring at the layers interface, dramatically
enhances the effective electric permittivity of the substrate up to f r,ef f 2900 and thus
between 10 MHz and 2.5 GHz a scaling factor of about 54 with a 1.2% variation has been
achieved. Therefore, this line can provide a phase delay of 16.5° at 2 GHz with a +1°
variation. In conclusion, a compact phase shifter that will provide 45°, 90° or even 180°
can be fabricated by incorporating multiple lines of several lengths.
6.3 Miniature Distributed Filter Design on Silicon CMOS
There has been extensive interest on monolithic lumped element LC filters using standard
CMOS technologies. The lumped element approach is reputed to offer a number of ad-
vantages compared to distributed filters, including compactness, design simplicity and ease
of manufacture. Recent reports of lumped filters for microwave frequencies up to 18 GHz
promise that low frequency circuit design techniques, coupled with standard CMOS manu-
facturing, are adequate for all the requirements of foreseeable high frequency applications.
It is not surprising that the filter design method of choice for microwave filters, namely
distributed circuits, has generally been avoided in CMOS.
Distributed filters are indeed preferred when low insertion loss, high operation frequency
and good linear power handling are the dominant consideration [9]. For this reason they
are ubiquitous in III-V MMIC's applications. However, conductor losses, prohibitive sub-
strate absorption and cross-coupling in conjunction with the rather large physical size of
distributed microstrip or coplanar waveguide circuit elements on a conductive silicon sub-
strate make distributed filters infeasible in CMOS technologies.
The substrate loss and cross-coupling limitation of CMOS is adequately being addressed
by the introduction of Silicon-on-insulator (SOI) CMOS technologies in RF applications.
This approach still does not address the fact that distributed circuits are prohibitively large
at frequencies of technological interest, currently up to 20 GHz. This is indeed due to the
relatively low value of the substrate dielectric constant, motivating in part the development
of high permittivity and low loss dielectrics. This approach has enabled the production of
miniaturised dielectric resonators and filters.
Several techniques have been reported in the literature, recommending possible solutions
to a filter's compactness. A good review can be found in [10] and more recent designs are
given in [11]-[15]. These designs are based on either new topologies or on "advanced"
materials exhibiting ultra-high dielectric constants; with the latter category being more
effective in attaining larger scaling factors.
An alternative solution to the problem of realising a high dielectric constant substrate

was proposed in [16] similar to the one proposed in previous sections. Instead of looking for
a material which exhibits a high dielectric constant at all frequencies, the vertical structure
of a CMOS circuit is exploited to obtain a high effective dielectric constant over a band of
frequencies. Synthesised high-dielectric constant substrates can be implemented by exploit-
ing interfacial polarisation wave modes naturally occurring in a layered substrate consisting
of a sequence of conductive and insulating layers. CMOS integrated circuits are indeed in
this category.
Distributed filters can be considered as a two-port network used to provide transmission

at frequencies within the passband of the filter and attenuation in the stopband of the filter.
They can be classified as combline, interdigital, parallel-coupled line bandpass and bandstop,
ring and patch filters, and stepped-impedance filters and several can be implemented using
waveguides, dielectric resonators, coaxial lines, evanescent-mode filters, and various printed
circuit filters in microstrip, stripline and suspended substrates [9]. There exist several
techniques for developing "pseudo" coaxial lines in CMOS; however it is my intention to
provide a fundamental understanding of the concept. For this reason a distributed filter
was designed using printed microstrip lines instead of coaxial lines.
6.3.1 CMOS Manufacturing Process
There are two possible ways for developing slow-wave structures in CMOS. The main dif-
ference depends upon the oxide layer which is used. One way is by depositing polysilicon
on top of the transistor area, as is the case for MOSFETs gates, which are deposited on top
of the thin oxide layer. An advantage of this technique is the very large semiconductor to
insulator thicknesses ratios that can be achieved as thin oxides have typical thicknesses of
a few nanometers, which will eventually result in a better control of the design.
Alternatively, polysilicon can be deposited off the active transistor area, i.e. utilising
the field oxide area used for insulation between different devices. In this technique, the
semiconductor to insulator thickness ratio control is limited when compared with the for-
mer technique, due to the thicker Si02 (typical field oxide thicknesses are in the range of
hundreds of nanometers).
In both processes, polysilicon can be patterned using standard CMOS fabrication pro-
cedures. It seems that the first approach will be better due to the higher dielectric control.
Alternatively, the thicker field oxide could also be used, giving a good semiconductor to
insulator thickness ratio, which means less compatibility issues in the standard CMOS pro-
cess (avoidance of using active areas). What is important for minimising fringing fields is to
make sure that the insulator layer covers a larger area than the area which will be occupied
by the filter.
6.3.2 Measurements, Simulations and Discussion
Provided that the substrate is sufficiently conductive (eqation 2.43), a large semiconductor
to insulator thickness ratio will eventually result into a large interfacial polarisation, which
contributes a significant augmentation of the effective electric permittivity of the structure
up to a frequency limit, given by equation 2.39. However, the larger the effect of this po-
larisation mechanism is, the narrower the bandwidth in which this phenomenon is observed
becomes. Moreover, a large Er ,e ff dictates that the dielectric losses of this structure will
be significantly increased in the frequency band where the transition from slow-wave to
skin-effect mode occurs (figure 2.15).
Figure 6.4 illustrates the forward transmission coefficient for two 100 pm long transmis-
sion lines, built on substrates with different slowing factors. For tox=0.7 pm, the semicon-
ductor to insulator thickness ratio is fixed to 750 while for a 1.4 pm thick oxide this ratio is
halved. As a result, at low frequencies, the supporting structure behaves as if it was a single
material, which for the first case exhibits an effective dielectric constant of Er ,eff ti 2900
where for the latter case Er,ef I drops to approximately 1460. Indeed, by observing figure
6.4, one can realize that both the measurements as well as the simulated data are in good
agreement with these values, as the cut-off frequencies of these lines are scaled down in
analogy to these dimensional ratios. Furthermore, as stated earlier, the dielectric losses
of the "slower" substrate will be larger than in the "fast" substrate. By observation, one
can realize that the gradient of the attenuation of the first line is indeed greater than the
"faster" line.
In order to validate this allegation, the same measurements were repeated for longer
lines (L=200 µm) based on the same substrates previously used. The discussion in the
previously presented data also holds for the data presented in figure 6.5. However, due to
the doubling of the length, the cut-off frequency of these lines occurs at lower frequencies
than previously.
0.0
-2.5
-7.5 - tox=1.4µm (Measured)

—e— tox=0.7pm (Measured)
—A— tox=0.7µm (Simulated)
—*— tox=1.4µm (Simulated)
-10 0
1 10
Frequency (GHz)
Figure 6.4: Simulated and Measured insertion loss of the proposed filter (L=100 um) for
various slowing factors.
0.0
-2.5
72 -5.0
-7.5 - tox=1.4µm (Measured)

- tox=0.71.im (Measured)
- tox=0.7µ,m (Simulated)
—*— tox=1.4µm (Simulated)
-10 0

1 10
Frequency (GHz)
Figure 6.5: Simulated and Measured insertion loss of the proposed filter (L=-200 itm) for
various slowing factors.
—■— L=1 00µm & tox=0.71.im

—•— L=1 00µm & tox=1 .411m
- A- L=20Own & tox=0.7pm
—*— L=200pm & tox=1 .4pm

1 10
Frequency (GHz)
Figure 6.6: Measured return loss of the proposed filters for various slowing factors.
Design Parameters fc (GHz) fe (GHz) RL (dB) IL (dB)

of Specimen (Measured) (Simulated) (0.1-2GHz) (0.1-2GHz)
L=100pm, tsio,/tox = 750 3.34 3.12 >7 <1.4
L=100,am, tsio,/tox = 375 6.65 5.31 >12.48 <0.48
L=200,um, tsio,/tox = 750 3 2.55 >7.2 <1.73
L=200pm, ts,02 /t0x = 375 5.85 5.4 >12.48 <0.63
Table 6.1: LowPass Characteristic Performance of MISM lines
It is observed that both the 100 pm as well as the 200 pm long transmission lines behave
as second order lowpass filters and their performance characteristics are mainly subject to
the extent of the Maxwell-Wagner effect (figures 6.4, 6.5 and 6.6). Table 6.1 summarizes
the performance of all the measured devices. Comparing, the 100 pm and the 200 pm line
based on the tszo2/t0x = 750 thickness ratio substrate, it is found that fe varies from 3.34
to 3 GHz, with better than 7 dB of return loss and an insertion loss of less than 1.73 dB.
On the other hand, if the thickness ratio (ts,02 /t0x ) is halved, a cut-off frequency varying
from 6.65 to 5.85 GHz, with better than 12.5 dB of return loss and less than 0.63 dB of
insertion loss is achieved.
6.4 Microstrip Stepped Impedance LowPass Filters
Stepped impedance filters are implemented through a succession of high and low charac-
teristic impedance microstrip line segments. This is why this type of filters are often called
hi-Z, low-Z filters. Strictly speaking, this is nothing more than a transmission-line represen-
tation for inductive and capacitive filter elements. Therefore, practical design rules impose
the application of a large high-to-low impedance ratio. Clearly, the subordinateness of these
filters is mainly imputed to impedance range limitations which result in shallow roll-off and
poor attenuation in the stop band. Moreover, the low-Z sections are usually formed by
extremely wide transmission lines which oppose the overall compactness of the filter. On
the other hand, a large high-Z section, i.e. a very small feature size might be unfeasible,
depending on conventional process limitations.
A number of techniques to tackle these issues have been proposed. The employment of
a substrate, having a plurality of regions with each having respective substrate properties,
namely the relative dielectric constant, has been identified as the most promising suggestion.
This can be accomplished in several ways. Drayton et al. [18] first introduced the idea of
varying the effective permittivity of the substrate by appropriate micromachining which
results in different substrate heights. A similar approach can be found in [19], in which
now the ground metalization has been etched at some areas. However, both techniques are
limited by the fact that a small impedance ratio can be achieved, Zmax /Zmin ^ 5.
A rather clever approach, due to Johansson et al. [20], introduces the deposition of
islands of a material with a high dielectric constant on a relatively low dielectric substrate.
The "islands" were employed under the low-Z sections of the filter, with the high-Z sections
being developed directly on the substrate. Utilisation of feroelectrics or any kind of high-K
composite materials, eventually results in increasing the attainable high-to-low impedance
ratio. However, this approach is not advisable when CMOS compatibility is an issue or
when temperature variations, which are known to have a direct effect on the electrical
performance of ferroelectric materials, are expected.
The idea presented in [20] can be further exploited by employment of the Maxwell-
Wagner polarisation mechanism. Instead of utilising exotic materials, it is proposed to
locally exploit the large interfacial polarisation mechanism that occurs at layered sub-
strates. This approach utilises thin layers of Si02 on top of a conductive Si substrate
which makes it fully compatible with standard CMOS technologies. Furthermore, the ex-
tremely large effective permittivity that can be achieved proves beneficial, regarding the
high-to-low impedance ratio. Aditionally, the high dielectric losses occuring outside the fre-
quency band in which this "slowing" phenomenon occurs, are used constructively in refining
the performance of the filter in the stop band [21].
6.4.1 Stepped Impedance Filters on Conventional Substrates
Stepped impedance filters are implemented through a succession of very high and very low
characteristic impedance microstrip lines. When a uniform substrate is used, the high-Z
and low-Z sections of such a filter are implemented through very narrow and very wide
microstrip lines respectively. Therefore, this design method apart from being a spacious
one, is moreover limited to relatively low high-to-low impedance ratios.
In order to do a fair comparison between this approach and standard hi-Z, low-Z filters
on typical substrates, a reference filter is presented in this section. This filter, shown in
figure 6.7, was designed using Ansoft's Designer [22], with ZHI = 120Q, ZLOTY = 205l
Figure 6.7: 5th order low pass microstrip stepped impedance filter on RT Duroid 5880, with
fo=1 GHz.
and for a cutoff frequency of f=1 GHz. It was processed on a RT Duroid 5880 substrate,
with a thickness of t=1.57 mm, er=2.2 and a loss tangent of tand(b)=0.0009. The overall
dimensions of this filter are 91x17 mm.
The filter's low pass behaviour was simulated in AWR's Microwave Office [7]. The device
was measured with Agilents PNA E8361A Network Analyzer. Measured and simulated
results are in very good agreement and are presented in figure 6.8. From 1 MHz to 700
MHz the filter demonstrates a return loss of better than 30 dB. Additionally, at the same
bandwidth, the filter's insertion loss was found to be less than 0.1 dB.
RL and IL Chebysev 5th order

0
-10
Insertion Loss (dB)
Return Loss (dB)
-20
-12
-30
-IFOR(15(2,1)i) (R) -18
' Chebysev 5th order
-40
-e-DB(IS(1,1)) (L)
Chebysev 5th order
--DB(IS(1,1)I) (L) -24
-50 Chebysev 5th order Measured
49-oBas(2:00 (R.)
Chebysev 5th order Measured
-60 -30
0 0.5 1 1.5 2
Frequency (GHz)
Figure 6.8: Simulated and measured return loss and insertion loss of the stepped impedance
filter shown in figure 6.7.
6.4.2 Stepped Impedance Filters on Variable Dielectric Substrates
In this section a novel approach for designing stepped impedance filters is presented. In
this approach, the effective substrate permittivity can be modulated and this way achieve
a larger ZHI /ZLO ratio.
The reference filter presented in the previous section, was adapted with the use of AWR's
TxLine in order to be built on a silicon substrate. Then the low-Z sections of the filter were
resized to achieve a microstrip line of uniform width. In order to ensure the same electrical
performance of the filter, these sections have to be electrically enhanced. The required
relative effective electric permittivity was calculated to be Er,eff 4000. According to
equation 2.37, and for a Si wafer of 525 itm thickness, with a substrate conductivity of
crsub=2000 S/m, a 0.5 ium thick layer of SiO2 is postulated (figure 6.9).
The proposed topology was simulated in AWR's Microwave Office, using different sub-
strate models for the hi-Z and low-Z sections. The Si wafer is well described in the existing
libraries, however, the Maxwell-Wagner effect on the effective electric permittivity of the
substrate, was simulated in MATLAB [23] and then all the parameters were imported in to
Microwave Office.
cu
cu
Figure 6.9: Top and side view of proposed stepped impedance filter.
Figure 6.10 demonstrates the simulated insertion as well as return loss of this stepped
impedance filter. The cutoff frequency of this filter occurs at fo=1 GHz. From 1 MHz to 700
MHz the return loss is deteriorated by 10 dB from the measured return loss of the filter on
RT Duroid. Likewise, the insertion loss has increased from 0.1 dB to 1.8 dB. Nevertheless,
the high attenuation factor occurring outside the frequency band in which this polarisation
mechanism occurs, results in a steeper roll-off than previously observed.
RL and IL Chebysev 5th order Maxwell-Wagner
-10
Return Loss (dB)
-15
-20
—DB(IS(1,1)I) (L)
-25 Chebysev 5th order (Maxwell-Wagner Equivalent)
—►—DB(IS(2,1)I) (R)
Chebysev 5th order (Maxwell-Wagner Equivalent)
-30
0 0.5 1 1.5 2
Frequency (GHz)
Figure 6.10: Simulated return loss and insertion loss of the proposed stepped impedance
filter.
Currently, the proposed filter is under post-processing and its dimensions are 21x0.37
mm, which translates into a significantly smaller area than a typical implementation would
require. Also, with this technique, a large high-to-low impedance ratio has been achieved,
providing a high rejection level. Moreover, the proposed method is fully compatible with
standard CMOS technologies as it does not recourse into exotic materials for attaining
relatively high dielectric constants.
The approach of using a variable dielectric substrate with utilisation of the Maxwell-
Wagner polarisation mechanism only at the low-Z sections was presented. However, it
should be noted here that interfacial polarisation could have been used in varying the
effective electric permittivity of both the low-Z as well as high-Z sections, resulting into
greater flexibility. Finally, this approach can be of practical use in applications where
microstrip impedance transformers are required.
6.5 Miniature Microstrip Antennas
Nowadays, the trend for miniaturisation drives the physical antenna's size down while de-
mands for reduced power dissipation necessitate an increase of efficiency, i.e. electrical size.
Microstrip patch antennas have been extensively used at frequencies of interest in mobile
communications. They became extremely popular because they are easily manufactured by
the same low-cost mass-production techniques that are used to make printed circuit boards.
Despite a number of important limitations (excessive Q or, equivalently, excessively narrow
bandwidth (equation 6.5)) their convenience and compactness more than compensate in
many applications. However, their dimensions are set by the operating frequency, which for
low frequencies could be extremely large. On the other hand, reducing their dimensions will
eventually result in a significant reduction in the antenna's efficiency. R. F. Harrington [24]
and J. L. Volakis et al. [25] have explored antenna miniaturisation issues in more detail.
As an example a monopole antenna of length h, with a diameter of the antenna rod

d = 2r which operates at a frequency f (wavelength ).) has radiation efficiency given by:
Rrad
(6.1)
Rrad Rloss
Where, the radiation impedance is:
Rrad = 4072( )2 (6.2)
and the loss impedance is:

h R,
Moss = (6.3)
r
R, being the surface resistance of the antenna:
R, = (6.4)
where it and a are, as usual, the magnetic permeability and the conductivity of the material
respectively.
The trend of building antennas on high dielectric constant substrates is not a new
approach and has in fact been used since 1983 by S. A. Long et al. [26]. Recently it has
also been applied in particular for microstrip antennas by T. Lo et al. [27]. Many authors
have followed this original example but have all used a single material as the high dielectric
substrate [28]. Perhaps the highest dielectric constant material that can be found in nature
is water (cr. = 81). A salt-water coated antenna was presented by E. Paraschakis [29]. This
technique offers a size reduction factor of about 9, however many losses are introduced due
to water's low conductivity.
Instead, the Maxwell-Wagner polarisation occurring in layered structures could be used

to realise a material with a much higher value of dielectric constant that will further reduce
the effective wavelength. Clearly, this method is advantageous to the single high-dielectric
substrate because naturally occurring materials have small dielectric constants and are
not appropriate for significant size reduction. This approach is equivalent to having a
passive resonant reflector, however, there are some difficulties arising from the non-isotropic
propagation.
6.5.1 Microstrip Antennas on Laminar Substrates
In order to compare the limitations in the effort to scale down typical patch antennas, sim-
ulated results are presented demonstrating the far field radiations as well as the directivity
of a 10x10 mm patch antenna (figure 6.11) build on a typical substrate, such as silicon, and
a lx1 mm patch build on a laminar substrate that supports Maxwell-Wagner polarisation.
Surprisingly, the lx1 mm patch antenna exhibits a 4.9 dBi increase in directivity but
with an extra 0.45 dB loss (figure 6.12), due to excitation of surface waves, when compared
with the 10x10 mm patch built on a silicon substrate. Furthermore, one might expect that
reducing the dimensions of the patch antenna will result in an electrically smaller antenna.
In contradiction, the lx1 mm patch has a larger number of minor lobes when compared
with the larger patch demonstrating the opposite [30].
Although developing antennas on high dielectric constant substrates allows smaller an-
tennas to be used for lower frequency applications, several drawbacks occur that have to
be examined. One is the increase of losses due to the excitation of surface waves. Surface
wave power generated by a single element increases with substrate thickness and dielectric
constant as shown in [31, 32]. This loss will therefore be significant in this case due to the
ultra-high dielectric constant the proposed synthetic substrate exhibits. Nonetheless, this
loss could be minimised by application of a reflecting scheme. A possible solution to this
could be the use of a Left-Handed Material (LHM) [33] as a reflector.
Secondly, as stated in the introduction of this section, patch antennas typically exhibit
narrow bandwidths, in the range of 10-15 %. A large amount of work has been done in
developing techniques for improving the bandwidth of the microstrip antenna. Bandwidth
Radiation Paltena
Polar Grid Dir Theta.
0.01 9.0
Figure 6.11: Far field radiation and directivity of a 10x10 mm patch antenna on Si substrate.
Figure 6.12: Far field radiation and directivity of a lx1 mm patch antenna on proposed
substrate.
as referred to microstrip antennas may take one of several meanings. There is usually an
impedance matching network between the antenna element and its input port which must
be considered. A meaningful measure of bandwidth is that band of frequencies where the
input Voltage Standing Wave Ratio (VSWR) is less than a specified value, typically 2:1.
VSWR —1
BW = (6.5)
Q\/VSWR
Where, Q is the quality factor of the patch antenna.
Several bandwidth enhancement techniques exist such as impedance matching, use of

multiple resonance, use of a thick substrate and slot loaded patch. A good review of many
of these methods is given in [34, 35].
One might expect that utilisation of the proposed technique will further reduce the
bandwidth as the element size decreases with an increase in the dielectric constant, which
raises the Q of the resonator (equation 6.5). In contradiction, the results shown in figures
6.13 and 6.14 demonstrate the opposite. It is believed that the increased bandwidth is due
to the large losses caused by surface wave excitation. What is important though, is the
fact that the smaller radiating element resonates at lower frequencies than the large one.
Furthermore, the shift in resonant frequency is calculated to be proportional to vcr,eff of
the proposed slow wave substrate.
dB4S1WavePottl,WavePortl iI [db]
Frog !GIN
Figure 6.13: Insertion loss of a 10x10 mm patch antenna on Si substrate.
20.00
dEgiSIWavePortl.WavePortli) [d b]
0.00-
-20.00-
40.00-
4000
0.0) 2. 4.0b 6.010 o.ob 10.60
Fteq IGHz]
Figure 6.14: Insertion loss of a lx1 mm patch antenna on proposed substrate.

Finally, it should be noted that during the course of this work, several patch antennas
have been designed and fabricated on substrates that support slow-waves. These devices
were included in the process of the characterisation specimens (figure 4.3) and as such they
have not been measured yet. It is my intention to measure these devices and compare the
data with the theoretical expectations.
6.6 Miniature Quadrature Hybrid Couplers
Directional couplers are fundamentally 4-port devices used for power division. Power di-
viders are often of the equal-division (3dB) type and are usually referred to as Lange couplers
after J. Lange who introduced them [36]. These devices are very popular in modulators
[37] as well as in phase-shifters [38]. It should be noted that the coupled path is A/4 long,
therefore there is a 90° phase shift between the through and the coupled arms.
The coupling coefficients of such devices are dictated by the physical properties of the
strip lines that form the coupler (ie. the line dimensions and the substrate's physical
and electrical properties). Therefore, the Maxwell-Wagner polarisation can be applied for
enhancing the coupling of a miniature device and consequently behave as an electrically
larger device.
6.6.1 Simulations and Discussion
In order to illustrate this application of the Maxwell-Wagner polarisation effect, a 50 52

quadrature-hybrid coupler has been designed with a centre frequency fc=1GHz. The coupler
has been fabricated on a 1.55 mm thick RT/duroid 5880 substrate with ET=2.2. This coupler
requires an area of about 26 cm2. Figure 6.15 depicts the S-parameter magnitudes of this
coupler. It should be noted that a perfect 3 dB power division in ports 3 and 4 was obtained,
while isolation and return loss of the coupler reaches 40 dB at the center frequency.
Similarly, a branch-line coupler was designed at fc=1 GHz by application of the high-K
substrate concept. Since, it is generally desired to have as little dispersion as possible at
the centre frequency, a rather small semiconductor to thickness ratio has been applied that
exhibits an Er,ef f =900 . As a result the slow-wave mode is well supported in the frequency
band of interest and Maxwell-Wagner's relaxation starts to roll-off above 10 GHz. The
proposed coupler exhibits an excellent return loss of more than 40 dB and around 12 dB of
isolation at the centre frequency (figure 6.16).
-10
-a -20
DB(IS(1,1)I)
rn Quadrature Hybrid
-EI- DB(IS(1,2)I)
Quadrature Hybrid
-30
DB(IS(1,3)I)
Quadrature Hybrid
-E- DB(IS(1 ,4)I)
Quadrature Hybrid
-40
0 0.5 1 1.5 2
Frequency (GHz)
Figure 6.15: S-parameter magnitudes versus frequency for a quadrature hybrid coupler on
typical substrate.
The effective loss tangent of the structure is rather small when compared with the high-
frequency dielectric losses, within the frequency band of interest for this device. Despite
the fact that the effective loss tangent tanSM w of the structure, due to Maxwell-Wagner
polarisation, is two orders of magnitude larger than the kind of RT/duroid. This increase
is quite significant and results into the reduction of the power division in ports 3 and 4.
This topology significantly reduces the area required for a functional, if imperfect cou-
pler. The proposed coupler occupies an area of almost 9 mm2. Figure 6.17 demonstrates
the unwrapped phase-shift of both couplers, from 800 to 1200 MHz. Within this frequency
band, a 90° phase shift is observed between the through and the coupled arms in both de-
vices. This proves that the coupled arms on the high-K equivalent structure are A/4 long,
although they have been miniaturised by a factor of 17.
S-parameters magnitude
e ...
-10 —s -43 —
— S — s-
-20
DB(IS(1,1)I)
Quadrature Hybrid High Er
1 -8 DB(IS(1,2)I)
-30 -0 DB(IS(1,3)I)
DB(IS(1,4)I)
-40
0 0.5 1 1.5 2
Frequency (GHz)
Figure 6.16: S-parameter magnitudes versus frequency for a miniature quadrature hybrid
coupler on a laminar substrate.
Phase Shift
0
-50
.............................
.......................
.................................
(§) _100 ...................... ............
co
a_ -150
e AngU(S(3,1)) (Deg)
Quadrature Hybrid
co -200
AngU(S(4,1)) (Deg)
Quadrature Hybrid
*0 AngU(S(3,1)) (Deg)
-250 Quadrature Hybrid High E
-E• AngU(S(4,1)) (Deg)
Quadrature rid Hi h E
-300
0.8 1 1.2
Frequency (GHz)
Figure 6.17: Unwrapped Phase-shift for the couplers shown in figures 6.15 and 6.16.
6.7 Summary
In this chapter a technique for developing miniature delay lines, lowpass filters, patch an-
tennas as well as directional-couplers was demonstrated. This approach is based on the
Maxwell-Wagner polarisation mechanism that is supported in laminar substrates.
All designs are based on standard MISM transmission lines by utilisation of the slow-
wave propagation characteristics of the substrate. Very compact designs have been demon-
strated. Their performance was similar or even identical with their larger equivalents.
Moreover, it is anticipated that the technique demonstrated can be used in high-order fil-
ters. Finally, it must be emphasized that this approach relies on standard materials and
processing techniques that makes it fully compatible with standard CMOS technologies.
References
[1] B. J. Rubin and B. Singh, "Study of meander line delay in circuit boards," Transactions
on Microwave Theory and Techniques IEEE, vol. 48, pp. 1452-1460, Sept. 2000.
[2] R. B. Wu and F. L. Chao, "Laddering wave in serpentine delay line," Transactions on

Components, Packaging, and Manufacturing Technology IEEE, vol. 18, pp. 644-650,
Nov. 1995.
[3] W. D. Guo, G. H. Shiue, C. M. Lin, and R. B. Wu, "Comparisons between serpentine

and flat spiral delay lines on transient reflection/transmission waveforms and eye dia-
grams," Transactions on Microwave Theory and Techniques IEEE, vol. 54, pp. 1379-
1387, Apr. 2006.
[4] R. B. Wu and F. L. Chao, "Flat spiral delay line design with minimum crosstalk
penalty," Transactions on Components, Packaging, and Manufacturing Technology
IEEE, vol. 19, pp. 397-402, May 1996.
[5] W. P. Harokopus and L. P. B. Katehi, "Electromagnetic coupling and radiation loss

considerations in microstrip (m)mic design," Transactions on Microwave Theory and
Techniques IEEE, vol. 39, pp. 413-421, Mar. 1991.
[6] G. E. Ponchak, E. Dalton, E. M. Tentzeris, and J. Papapolymerou, "Coupling be-

tween microstrip lines with finite width ground plane embedded in polyimide layers
for 3d-mmics on si," Proc. IEEE International Microwave, Theory and Techniques
Symposium, vol. 3, pp. 2221-2224, 2002.
[7] "Applied wave research, microwave office." Version 7.05.
[8] "Ansoft corporation, high-frequency structure simulator (HFSS)." Version 9.0.
[9] R. Levy, R. V. Snyder, and G. Matthaei, "Design of microwave filters," Transactions

on Microwave Theory and Techniques IEEE, vol. 50, pp. 783-793, Mar. 2002.
157
REFERENCES 158
[10] J. S. Hong and M. J. Lancaster, Microstrip Filters for RF/Microwave Applications.

John Wiley and Sons, 2001.
[11] J. R. MacLeod, T. Nesimoglu, M. A. Beach, and P. A. Warr, "Miniature distributed

filters for software re-configurable radio applications," pp. 159-163, June 2002.
[12] Z. Ma, K. Nomiyama, and Y. Kobayashi, "Microstrip lowpass filters with reduced size
and improved stopband characteristics," Journal IEICE Trans. Electron., vol. E88-C,
pp. 62-67, Jan. 2005.
[13] W.-H. Tu and K. Chang, "Compact microstrip low-pass filter with sharp rejection,"
vol. 15, pp. 404-406, June 2005.
[14] K. Ma, J.-G. Ma, K. S. Yeo, and M. A. Do, "A compact size coupling controllable
filter with separate electric and magnetic coupling paths," Transactions on Microwave
Theory and Techniques IEEE, vol. 54, pp. 1113-1119, Mar. 2006.
[15] J. Papapolymerou, C. Lugo, Z. Zhao, X. Wang, and A. Hunt, "A miniature low-loss
slow-wave tunable ferroelectric bandpass filter from 11-14 ghz," Proc. IEEE Interna-
tional Microwave, Theory and Techniques Symposium, pp. 556-559, June 2006.
[16] T. Prodromakis and C. Papavassiliou, "Distributed filter design on silicon cmos," Proc.
IEEE International Symposium on Circuits and Systems, pp. 3309-3312, May 2006.
[17] D. Pozar, Microwave Engineering. John Wiley and Sons, 2005.
[18] R. F. Drayton, S. P. Pacheco, J. Wang, J. G. Yook, and L. P. B. Katehi, "Micro-

machined filters on synthesized substrates," Transactions on Microwave Theory and
Techniques IEEE, vol. 49, pp. 308-314, Feb. 2001.
[19] T. Akalin, M. A. G. Laso, E. Delos, T. Lopetegi, 0. Vanbesien, M. Sorolla, and

D. Lippens, "High performance double-sided microstrip pbg filter," Transactions on
Microwave and Optical Technology Letters IEEE, vol. 35, pp. 90-93, Nov. 2002.
[20] C. Johansson and M. Robertsson, "Microstrip stepped impedance filters with variable
dielectric or variable width," Electronic Letters IET, vol. 41, pp. 745-746, June 2005.
[21] T. Prodromakis, C. Papavassiliou, and G. Konstantinidis, "Micro-scale lowpass fil-

ters based on the maxwell-wagner phenomenon," in Proc. IEEE European Microwave
Conference, vol. 1, pp. 335-340, Oct. 2007.
REFERENCES 159
[22] "Ansoft corporation, designer." Version SV2.
[23] "Mathworks, matlab." Version 7.0.
[24] R. F. Harrington, "Effect of antenna size on gain, bandwidth, and efficiency," Journal
of Research of the National Bureau of Standards-D. Radio Propagation, vol. 64D, pp. 1-
12, Jan. 1960.
[25] J. L. Volakis, C.-C.-Chen, M. Lee, B. Kramer, and D. Psychoudakis, "Miniaturiza-

tion methods for narrowband and ultrawideband antennas," Proc. IEEE International
Workshop on Antenna Technology: Small Antennas and Novel Metamaterials, pp. 119-
121, Mar. 2005.
[26] S. A. Long, M. W. McAllister, and L. C. Shen, "The resonant cylindrical dielectric

cavity antenna," Transactions on Antennas and Propagation IEEE, vol. 31, pp. 406-
412, May 1983.
[27] T. K. Lo, C. 0. Ho, Y. Hwany, K. K. W. Lam, and B. Lee, "Miniature aperture-coupled

microstrip antenna of very high permittivity," Proc. Electronics Letters IEE, vol. 33,
pp. 9-10, Jan. 1997.
[28] K. R. Carver and J. W. Mink, "Microstrip antenna technology," Transactions on An-

tennas and Propagation IEEE, vol. 29, pp. 2-24, Jan. 1981.
[29] E. Paraschakis, H. Fayad, and P. Record, "Ionic liquid antenna," Proc. IEEE Interna-
tional Workshop on Antenna Technology: Small Antennas and Novel Metamaterials,
pp. 552-554, Mar. 2005.
[30] T. Prodromakis and C. Papavassiliou, "Towards a microstrip antenna on synthetic

high-dielectric constant substrates," Proc. IEEE High Frequency Postgraduate Student
Colloquium, vol. 10, pp. 57-61, Sept. 2005.
[31] D. 1\4. Pozar, "Microstrip antennas," Proc. IEEE, vol. 80, pp. 79-91, Jan. 1992.
[32] D. H. Schaubert, D. M. Pozar, and A. Adrian, "Effect of microstrip antenna substrate

thickness and permittivity: Comparison of theories with experiment," Transactions on
Antennas and Propagation IEEE, vol. 37, pp. 677-682, June 1989.
[33] J. Pendry, A. Holden, D. Robbins, and W. Stewart, "Magnetism from conductors and
enhanced nonlinear phenomena," Transactions on Microwave Theory and Techniques
IEEE, vol. 47, pp. 2075-2084, Nov. 1999.
REFERENCES 160
[34] A. Henderson, J. R. James, and C. M. Hall, "Bandwidth extension techniques in printed

conformal antennas," Military Microwaves, vol. 86, pp. 329-334, 1986.
[35] D. M. Pozar, "A review of bandwidth enhancement techniques for microstrip anten-
nas," Microstrip Antennas, IEEE Press, pp. 157-166, 1995.
[36] J. Lange, "Interdigitated stripline quadrature hybrid," Transactions on Microwave

[37] A. E. Ashtiani, S. Nam, A. d'Espona, S. Lucyszyn, and I. D. Robertson, "Direct multi-

level carrier modulation using millimeter-wave balanced vector modulators," Transac-
tions on Microwave Theory and Techniques IEEE, vol. 46, pp. 2611-2619, Dec. 1998.
[38] S. Lucyszyn and I. D. Robertson, "Two-octave bandwidth monolithic analog phase

shifter," Microwave and Guided Wave Letters IEEE, vol. 2, pp. 343-345, Aug. 1992.
Chapter 7
Conclusion
This dissertation investigated the Maxwell-Wagner polarisation effect in layered substrates

and its applications in monolithic technologies. The slow-wave propagation in laminar sub-
strates was verified and extensively explored. Methods for engineering the mechanics of
Maxwell-Wagner polarisation were examined and numerous miniature devices were devel-
oped by application of the Maxwell-Wagner polarisation effect.
The theory behind the Maxwell-Wagner polarisation mechanism has been interpreted
in several manners: the device physics, the material's science and the electromagnetic wave
propagation approach. Each research society ascribed the large slowing effect, exhibited
in laminar substrates, to the interfacial space-charge accumulation, the polarisation effects
within the media and the different propagating modes, respectively. Although all approaches
describe the same effect using different words, each approach provides unique insight to the
Maxwell-Wagner mechanics. In this thesis, all three methods have been utilised and the
Maxwell-Wagner polarisation effect has been presented in a structured and unified way,
intensifying the comprehension of the mechanics of slow-wave propagation.
The importance of accurately calculating the MISM slow-wave characteristics has been
highlighted by the need to model interconnects on digital integrated circuits. Studies pre-
senting different approaches for analyzing MISM structures abound in the literature. During
the course of this work, all analytical models ascribing the Maxwell-Wagner relaxation in
the context of empirical lumped circuit models, the parallel-plate wave approach, full-wave
analysis, quasi-static modelling and device level simulations have been reviewed. This leads
to a direct comparison with all the pros and cons of each model signified.
161
Conclusion 162
Until recently, the slow wave mode was regarded as a limitation which needed to be
avoided by meticulous modelling. However, what used to be a limiting factor to overall
system performance has been exploited, to some extent, for practical engineering. Ap-
plications that benefit from the excessive large slowing-factor are briefly discussed and a
comprehensive review of research and development in that arena is given.
7.1 Contributions
In this work, layered structures that support the Maxwell-Wagner polarisation have been
investigated. Such structures exhibit propagation characteristics similar to high-K sub-
strates. A large variety of specimens has been developed that cover significant physical
parameters that affect the Maxwell-Wagner's effect. The slow-wave mode of a metal-
insulator-semiconductor transmission line is dependent on various parameters, including:
the substrate resistivity, the ratio of semiconductor to insulator thickness, and the trans-
mission line's dimensions. During this work, these dependencies were signified for conductive
Si substrates as well as for GaAs membranes. The functionality of the Maxwell-Wagner ef-
fect has been described both qualitatively and analytically and provides both experimental
verification and an intuitive reasoning behind the excessively large slowing effect.
It is apparent that a practical investigation of the slow-wave mode requires the use of
large numbers of test specimens. To overcome this, an experimental platform has been
developed that is capable of exploiting slow-waves on laminar substrates over a wide range
of parameters. This structure serves as a macroscopic model of MISM substrates. It uses
an aqueous dielectric, whose conductivity can easily be altered and offers the flexibility of
replacing the insulating layer by any laminate material. The presented results demonstrate
the ability of this platform to support slow-wave propagation, following all theoretical pre-
dictions, and yield the effect of the Maxwell-Wagner polarization mechanism under various
conditions.
In this thesis, the large slowing-effect of Maxwell-Wagner polarisation is attributed to

the rigidity of the interfacial space-charge to be polarised when a field is applied. The
larger the extent of the interface, the more charge can be accumulated and thus a large
slowing factor is observed. However, the large dimensions involved, limit this effect within
low frequency bands. To the best of my knowledge, the method of applying substrate
processing techniques for extending the Maxwell-Wagner polarisation effect is the first to
Conclusion 163
be reported. In this report, several approaches have been presented and investigated that
fall into two main categories: the formation of excess mobile charge in the insulating medium
and the damaging of the interfaces involved. Discussed implementation issues and design
techniques are consolidated with measured data from fabricated devices.
Finally, the Maxwell-Wagner polarisation was applied for introducing miniature passive
devices. The high resonant dielectric constant of layered structures is used to obtain even
smaller effective dimensions of passive components such as delay lines, distributed low-
pass filters, patch antennas and couplers. In the case of stepped-impedance filters, the
Maxwell-Wagner polarisation mechanism has been applied for tuning the effective electric
permittivity of the substrate locally. The large attainable electric permittivity allows a
large impedance ratio and a much more compact design. The high dielectric losses are
constructively used in the stop band. The structures and the circuit topologies developed
are described in detail including simulated results. To validate the theory, measurements
will be provided in future publications.
7.2 Recommendations for Future Work
Future developments based on material described in this thesis are proposed in the following
areas:
7.2.1 Maxwell-Wagner Polarisation on Controlled Damaged Specimens
Maxwell-Wagner polarisation can be applied in numerous applications where passive com-

ponents are involved. However, the main issue is to achieve significant control on the
parameters that affect the interfacial polarisation mechanism. There is a great scope to
continue work on controlled damaged substrates.
Implantation of Very Heavy atoms in Si
In section 5.8 the effect of Si doping by heavy atoms was investigated. The lattice defects
created on the Si surface are proportional to the effective mass of the implanted atom.
During the course of this work, there was a limited access to an implanter and thus only the
available materials were utilised. However, significantly heavier atoms than Si will cause
severe damage in the Si lattice that will eventually act as tiny interfaces. Thus the effect
Conclusion 164
this has on the overall contribution to the structure's effective electric permittivity should
be examined.
Si Implantation in Si02/Si without Annealing
Si implantation in Si02 /Si, followed by an annealing process, has been thoroughly used
in SOI technologies. However, if the annealing process is omitted, defects are induced
in the Si/Si02 layer. This could be a useful technique for Maxwell-Wagner mechanism
engineering.
Low Energy Nanodots
Small islands of a material induced in another medium, create tiny interfaces which have a
relaxation at higher frequencies. Also, there is an extra layered capacitance associated with
the introduction of these small interfaces. Nassiopoulou in [1] presented a rather interesting
processing technique for developing such interfaces. Nanostructuring, in general, includ-
ing surface microtexturing could be promising in engineering the slow-wave propagation
characteristics in MISM structures.
X-ray radiation
Mobile charge within the oxide layer was found to have a direct effect on Maxwell-Wagner
polarisation (5.7). The results suggest the possibility that trapped oxide charge, induced
by X-ray radiation can support the Maxwell-Wagner mechanism.
References
[1] A. G. Nassiopoulou, "Silicon nanocrystal nonvolatile memory devices." SINANO

WORKSHOP, Montreux, Sept. 2006.
165
Appendix A
Interfacial and Space-Charge

Polarisation
The classical example of interfacial polarisation is the Maxwell-Wagner two layer condenser.
The structure consists of two layers of materials (1) and (2), characterized by their dielectric
constant, conductivity, and thickness (El, al , di ) and (E'2 , (72, d2 ) respectively. When a DC
field is applied, the initial field distribution corresponds to the electrostatic requirement of
constant flux density.
Di = D2

or (A.1)
E1/ E2 = E2/El
where the final distribution follows from the condition of current continuity.
Jl = Jz
Or (A.2)
E1/E2 = a2/ai
The transient which links the initial and final state can be derived from the equivalent
circuit of figure 2.10.
V = Vi + V2 (A.3)
dV1 V1 ,dV2 V2
(A.4)
Ri = (-12 dt 4- R2
I = Ci dt + —
A i A i _da_
where C1 = d— flci and A is the surface area of the
i El C2 = d—2 621 R1 = Aui , R2 = .n0.2
capacitor's plate. It follows,
V1 — 17- 71 11 — (1 R2C2 e -t/T)}

Ri+R2
(A.5)
CI e -t/T)}
V2 = V R 1%2 {1 (1 R1
166

Interfacial and Space-Charge Polarisation 167
where the relaxation time is:
R1R2(Ci el d2 + e2d1
T +C2) (A.6)
Ri + R2 Crld2 + cr2di
When an AC voltage V = Voeiwt is applied, the steady-state solution can be derived via the
circuit's admittance.
= Y1 Y2
(A.7)
Y V Y1 + Y2
where
1 1±juri
-
Y1 =R l = jWC1 = Ri
(A.8)
Y2 = ,÷2 = iwc2 =
Equation A.6 can be expressed in terms of Ti and T2 of the individual RC circuits.
Rir2 + R2T1
T= (A.9)
R1+ R2
and hence the admittance can be rewritten as:
1 (1 + jwri)(1 jw7-2)
Y = (A.10)
Ri ± R2 l+ jWT
However, the admittance also determines the complex permittivity of the capacitor as:
Y = jwic*Co (A.11)
where Co = '4 and d = dl + d2 Hence, the two layer condenser, appears as a single
dielectric with a relaxation described by:
= Tl + T2 — T W 2 TT1T2
(A.12)
Co(R1 + R2 )(1 + w27-2)
The two-layer condenser experiences a relaxation spectrum similar to the simple orientation
polarisation of the Debye theory'. Thus, the electric permittivity (equation A.12) changes
from its static value (w = 0)
Ti
+T —T 2
=
Co(R1 + R2)
to its optical value (w oo)
7-yr2
koo =
Co(R1 + R2)T
Additionally, the two-layer dielectric will exhibit a dissipation factor given by:
2
t/ 71+ 72 — 7 + W TrT2
=
Co(R1 + R2)
1 "Dielectrics and Waves", A. R. Von Hippel, The MIT Press, 1954
Interfacial and Space-Charge Polarisation 168
And the complex permittivity may thus be written,
=— 40 s cc) i Cr (A.16)
1 + WT w
Appendix B
MATLAB Modeling of MIS

Microstrip Using Hasegawa's
Model
This program calculates all properties of the slow-wave mode as well as all various char-
acteristic frequencies and plots the frequency response of the real as well as the imaginary
Er ,, f f of the MIS line.
169
MATLAB Modeling of MIS Microstrip Using Hasegawa's Model 170
%This program calculates all properties of the slow-wave mode

%as well as all various characteristic frequencies and plots the real
%as well as the imaginary Er,eff of the MIS line.
%(slow(s1,s2,d1,d2,erl,er2)
function [w,H]=slow(sl,s2,d1,d2,er1,er2);
e0=8.85418782*10--12; m0=4*pi*10--7; c0=1/(sqrt(e0*m0)); d=d1+d2;

Width=90*10--6; e1=er1*e0; e2=er2*e0; f(1)=1; dw=10000000; A=1;
R1=d1/s1; R2=d2/s2; C1=e1/d1; C2=e2/d2; d=d1+d2; Co=e0/d; ti=e1/s1;
t2=e2/s2; t=((e1*d2)+(e2*d1))/((s1*d2)+(s2*d1));
fs=s2*d1/(2*pi*e0*eri*d2) fe=s2/(2*pi*e0*er2) fd=1/(pi*m0*s2*(d2)-2)
fo=1/((1/fs)+(2/(3*fd)))
%Maximum effective dielectric constant

Eeff_max=(d*er1)/(d1)
for i=1:10000;
f(i+1)=f(i)+dw;
w(i+1)=2*pi*f(i+1);
%Real part of Er,effective
H(i+1)=(t1+t2-t+t1*t2*t*(w(i+1)-2))/(Co*(R1+R2)*
(1+(w(i+1)^2)*(t-2)));
%Imaginary part of Er,effective
G(i+1)=((1/w(i+1))+w(i+1)*((t*t2)+(t*t1)-(t1*t2)))/
((Co*(R1+R2))*(1+(w(i+1)-2)*(t-2)));
%Loss tangent
L(i+1)=(G(i+1))/(H(i+1));
Zos(i+1)=120*pi*d/(sqrt(H(i+1))*Width);
%Real impedance
Zreal(i+1)=Zos(i+1);
%Imaginary impedance
Zim(i+1)=Zos(i+1)/2*(f(i+1)/fs-2/3*f(i+1)/fd);
%Difference between Real and Imaginary Impedance
Zdif(i+1)=(Zreal(i+1))-(Zim(i+1));
end
Estatic=(t1+t2-t)/(Co*(R1+R2)); %calculates E static (w=0)

Eoptical=(tl*t2)/(Co*(R1+R2)); %calculates E optical (w=oo)
subplot(2,2,1); plot(f,H); grid; subplot(2,2,2); plot(f,G); grid;
f=1/t;}
This program plots the real dielectric response of a two layer structure having different
dielectrics with respect to the substrate's conductivity.
MATLAB Modeling of MIS Microstrip Using Hasegawa's Model 171
% This program plots the real dielectric response of a two layer structure
% having different dielectrics with respect to the substrate's
% conductivity. cond(dl,d2,e1,e2);
function [a,H]=cond(d1,d2,er1,er2); s1=10--13; eo=8.85418782e-12;

e1=erl*eo; e2=er2*eo; s2(1)=0; f=1*10-9; w=2*pi*f; dw=1; A=1;
C1=e1/d1; C2=e2/d2; d=d1+d2; Co=eo/d; t1=e1/s1; R1=d1/s1; for
i=1:10000
s2(i+1)=s2(i)+dw;
r(i+1)=1/s2(i+1);
R2(i+1)=d2/s2(i+1);
t2(i+1)=e2/s2(i+1);
t(i+1)=((el*d2)+(e2*d1))/((sl*d2)+(s2(i+1)*d1));
%Real part of Er,effective
H(i+1)=(t1+t2(i+1)-t(i+1)+tl*t2(i+1)*t(i+1)*(w-2))/(Co*(R1+R2(i+1))*
(1+(w- 2)*(t(i+1)-2)));
%Imaginary part of Er,effective
G(i+1)=-((1/14)-w*((t(i+1)*t2(i+1))+(t(i+1)*t1)-(t1*t2(i+1))))/
((Co*(R1+R2(i+1)))*(1+(w-2)*(t(i+1)-2)));
Z(i+1)=H(i+1)+G(i+1); %Complex Er,effective
L(i+1)=(G(i+1))/(H(i+1)); %Loss Tangent
end
subplot(2,2,1);plot(s2,H); grid; subplot(2,2,2);plot(s2,G) grid;
subplot(2,2,3);plot(s2,L); grid;
end
Appendix C
Complementary Measurements
Phase Shift (Deg)
50
—'— Measured CPW with tsub/tox=2100

--m— Simulated CPW with tsub/tox=2100
Measured CPW with tsub/tox=1140
Simulated CPW with tsub/tox=1140
--- Measured CPW with tsub/tox=620
— Simulated CPW with tsub/tox=500
-100

01 10
Frequency (GHz)
Figure C.1: Measured and simulated unwrap phase-shift of a 100 pm long microstrip trans-
mission line (MSTL) with the semiconductor to insulator thickness ratio varying.
172
Complementary Measurements 173
..---. 5
0:1
1:3
rn
0
C
O
— Measured CPW with tsub/tox=2100
a) -10 —15— Simulated CPW with tsub/tox=2100
C Measured CPW with tsub/tox=1140
—1—Simulated CPW with tsub/tox=1140
15 11111

01 1 10
Frequency (GHz)
Figure C.2: Measured and simulated insertion loss of a 100 pm long microstrip transmission
line (MSTL) with the semiconductor to insulator thickness ratio varying.
180
— —CHF3 & 02 processed Si
120
n
co
To
0
60
0
0
0
60
—a— Standard Sample

- CHF3 & 02 processed Si
40
20
0
01 10
Frequency (GHz)
Figure C.3: Measured group delay and effective length of a 100 pm long MSTL on SiO2
based on standard Si as well as on RIE defected Si.
Appendix D
G rna , Measurements
RF and microwave designers tend to use the maximum available gain (Gmax) when de-
scribing/examining passive networks and in particular amplifiers. For completeness of this
thesis, the measured Gmax of all devices exploited in this thesis is given here.
-200
-400
Frequency (Hz)
-800
as=20S/cm
asi=200S/cm
-1000 as=0.002S/cm
-1200
108 109
Frequency (Hz)
Figure D.1: Measured Gmax losses of a 100 sum long MSTL on substrates of various con-
ductivity.
174

Gmax Measurements 175
—1000- 0
-20
-40
—2000 -60
-BO
E-100
cr-1 —3000 -120
(5 -140
-160
-180
—4000
-200
10' Frequency (Hz) 10'
—5000
--1subliox=2100
--1sub/10x=1140
- tsubl1ox=620
— - tsub/tox=500
6000
—700010e
109 101 °
Frequency (Hz)
Figure D.2: Measured Gmax losses of a 100 pm long MSTL with the semiconductor to
insulator thickness ratio varying.
—1000
—2000
—3000
c"- —4000
-o
x —5000
ct
E
CD
—6000
—7000
W=60 [1M
—8000 W=40 lArri
— W=20
—9000 — W=5 ixm
—10000
10 109 101°
Frequency (Hz)
Figure D.3: Measured Gmax losses of a 100 pm long MSTL for various widths: 5 pm, 20
pm, 40 pm and 60 pm.

- 1000 -20
-40
-2000
CO-60
E80
-10
E -4000
0 120
10. Frequency (Hz) 10.
-5000
-6000 .me W=5 ti,M

W=1 0 p.m
W=40 lin
W=60 µm
-7000

108 109 1010
Frequency (Hz)
Figure D.4: Measured Gmax losses of a 100 pm long MSTL on GaAs membranes for various
widths: 5 pm 10 pm, 40 pm and 60 pm.
0 t5 /tox=2100
KOH sample
-1000-
-50
-2000-
-100
2-3000-
00 2 -150
-o E
E
0 -200
%1
< -4000-
E
-250
-5000-
-300
10. Frequency (Hz) 10
-6000-
-7000-
-8000
108 109 10'9
Frequency (Hz)
Figure D.5: Measured Gmax losses of a 100 pm long MSTL on Si02 based on standard Si
and on KOH processed Si.
tsi/c=2100
HF sample
—1000
—2000 -50
—3000 — —8-100
m
-150
—4000 E
co E
0 -200
—5000
CO -250
E
—6000
300
10 Frequency (Hz) l Os
—7000
—8000
—9000
—10000
108 109 101°
Frequency (Hz)
Figure D.6: Measured Gmax losses of a 100 pm long MSTL on good quality SiO2 as well
as surface defected SiO 2 on Si.
—1000
—2000
—3000
—4000
—5000
CO
—6000
—7000
t n =2100
—8000 si ox
CHF and 02
3
—9000

109 101 °
Frequency (Hz)
Figure D.7: Measured Gmax losses of a 100 ,um long MSTL on SiO2 based on standard Si
as well as on RIE defected Si.

_isi/tox=2100
—Au nanodots sample
-1000
-2000
7
-e-3000 -
co
-400a
E
-5000
-6000
-7000
-8000 _[ I
108 109 1070

Frequency (Hz)
Figure D.8: Measured Gmax losses of a 100 pm long MSTL based on standard SiO2 and
on SiO2 with buried Au nanodots.
-1000 -
-2000 -
ai -3000 -
-0
RY
E -4000 -
-5000 -
-6000 t It =2100
Si ox
SiGe sample
-7000
100 109 1010
Frequency (Hz)
Figure D.9: Measured Gmax losses of a 100 ,u,m long MSTL on SiO2 based on standard
SiO2 and on Ge rich SiO2.
Gina, Measurements 179
-50
-3000 - aioo
-4000 -
co
-o (5 0 -20
x -5000 -
rci
E -250
O —6000 —
-300
10. 10.
Frequency (Hz)
-7000 -
-8000 -
-9000 -
-10000
108 109 1015
Frequency (Hz)
Figure D.10: Measured Gmax losses of a 100 pin long MSTL on Si02/Si sample before
and after inducing Na+ ions.
sitox=2100
Ar sample
As sample
-1000- P sample
-20
-40 -
-2000-
6, -80
-3000- g -100 -
m 0E -120 -
-140 _
g -4000- -160-
E
-180 -
-5000- 200
10. Frequency (Hz) 1(Y
-6000-
-7000-
-8000
10' 109 10'°
Frequency (Hz)
Figure D.11: Measured Gmax losses of a 100 pm long MSTLon Si02/Si and on Si02/Si
where the Si surface has been doped with Ar, As and P.

Gffic, Measurements 180
-1000 0
20
—40
-2000 —60
2" -Bo
^.) -100
in -3000 —120
-o (.1 —140
CO —160
—180
-4000
200
09 109
Frequency (Hz)
-5000
subito.=2100
/T =120nm/250nm/525 µm
-6000 tSrT103 /tox Si
tsubnox=1140
-7000

108 109 1010
Frequency (Hz)
Figure D.12: Measured Gmax losses of a 100 pm long MSTL on a Si02/Si and a
SrTiO3/Si02/Si system.
Appendix E
Capacitance-Voltage
Measurements
In this section, CV measurements of all fabricated specimens are presented. Measurements

were taken on a semi-automatic probe station using Agilent's B1500A Semiconductor De-
vice Analyzer at 1 kHz and 1 MHz for a biasing range of ± 25 V.
Figure E.1: CV measurement of a MOS diode at 1 kHz, with tsub/tox
181
Capacitance-Voltage Measurements 182
1.52E-11
1.51E-11
Capacitance (F)
1.50E-11
C
1.49E-11
1.48E-11
1.47E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.2: CV measurement of a MOS diode at 1 MHz, with t sub/t ox =2100.
Figure E.3: CV measurement of a MOS diode at 1 kHz, with t sub/tox =1140.
1.23E-11
1.21E-11
Capacitance ( F)
1.19E-11
C
1.17E-11
1.15E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.4: CV measurement of a MOS diode at 1 MHz, with t sub/tox =1140.

7.30E-12
7.25E-12
Capacitance (F)
7.20E-12
7.15E-12
C
7.10E-12
7.05E-12
7.00E-12

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.5: CV measurement of a MOS diode at 1 kHz, with tsub /t„ =620.
7.25E-12
7.20E-12
Capacitance (F)
7.15E-12
C
7.10E-12
7.05E-12
7.00E-12

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.6: CV measurement of a MOS diode at 1 MHz, with tsub/tox =620.
Figure E.7: CV measurement of a MOS diode at 1 kHz, with tsub/tox =500.

Figure E.8: CV measurement of a MOS diode at 1 MHz, with t sub/tox =500.

2.40E-11
2.35E-11
LL 2.30E-11
cu
2.25E-11
a' 2.20E-11
m
L.)
2.15E-11
2.10E-11 •

-30 -20 -10 10 20 30
Voltage (V)
Figure E.9: CV measurement of a MOS diode at 1 kHz, with KOH 20% processed Si.
Figure E.10: CV measurement of a MOS diode at 1 MHz, with KOH 20% processed Si.
4.00E-11
3.80E-11
Capacitance (F)
3.60E-11
`C
3.40E-11
3.20E-11
100E-11

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.11: CV measurement of a MOS diode at 1 kHz, with HF 5% processed SiO2.
1.55E-11
1.52E-11
Capacitance (F)
1.49E-11
‘;" C
1.46E-11
1.43E-11
1.40E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.12: CV measurement of a MOS diode at 1 MHz, with HF 5% processed SiO2.
3.10E-11
3.00E-11
LL
2.90E-11
A
2.80E-11
2.70E-11 --
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.13: CV measurement of a MOS diode at 1 kHz, with RIE processed Si.
1.62E-11
1.59E-11
Capacitance (F)
1.56E-11
(' C
1.53E-11
1.50E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.14: CV measurement of a MOS diode at 1 MHz, with RIE processed Si.
Figure E.15: CV measurement of a MOS diode at 1 kHz, with Au nanodots in Si.
Figure E.16: CV measurement of a MOS diode at 1 MHz, with Au nanodots in Si.

2.40E-11
2.20E-11
2.00E-11
Capacitance (F)
1.80E-11
GC
1.60E-11
1.40E-11
1.20E-11
1.00E-11 - - -- --

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.17: CV measurement of a MOS diode at 1 kHz, with Ge spheres in SiO2-
2.40E-11
2.20E-11
2.00E-11
Capacitance (F)
1.80E-11
C
1.60E-11
1.40E-11
1.20E-11
1.00E-11

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.18: CV measurement of a MOS diode at 1 MHz, with Ge spheres in SiO2 .
2.30E-11
2.25E-11
2.20E-11
Capacitance (F)
2.15E-11
tc, C
2.10E-11
2.05E-11
2.00E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.19: CV measurement of a MOS diode at 1 kHz, with induced Na+ in Si02 •
1.55E-11
*
1.54E-11
a) 1.53E-11
C
1.52E-11
a.
ea
1.51E-11
1.50E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.20: CV measurement of a MOS diode at 1 MHz, with induced Na+ in Si02.
-10 0 10
Voltage (V)
Figure E.21: CV measurement of a MOS diode at 1 kHz, with Ar implanted Si.
Figure E.22: CV measurement of a 1\40S diode at 1 MHz, with Ar implanted Si.

1.86E-11
1.86E-11
1.85E-11
Capacitance (F)
1.85E-11
C
1.84E-11
1.84E-11
1.83 E-11

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.23: CV measurement of a MOS diode at 1 kHz, with As implanted Si.
Figure E.24: CV measurement of a MOS diode at 1 MHz, with As implanted Si.
1.93E-11
1.92E-11
Capacitance (F)
1.91E-11
C
1.90E-11
1.89 E-11
1.88E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.25: CV measurement of a MOS diode at 1 kHz, with P implanted Si.

1.49 E-11
1.49E-11
1.48E-11
Capacitance (F)
1.48E-11
C
1.47 E-11
1.47 E-11
1.46E-11
-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.26: CV measurement of a MOS diode at 1 MHz, with P implanted Si.
1.44E-11
1.42E-11
1.40E-11
Capacitance (F)
1.38E-11
C
1.36E-11
1.34E-11
1.32 E-11
1.30E-11 --

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.27: CV measurement of a MOS diode on SrTiO3 /Si02 /Si at 1 kHz.
1.27E-11
1.26E-11
1.25E-11
Capacitance (F)
1.24E-11
1.23 E-11 C
1.22E-11
1.21E-11

1.20E-11 ---1 •O•

-30 -20 -10 0 10 20 30
Voltage (V)
Figure E.28: CV measurement of a MOS diode on SrTiO3 /Si02 /Si at 1 MHz.

Appendix F
Publications
• T. Prodromakis and C. Papavassiliou: "Towards a Microstrip Antenna on Synthetic

High-Dielectric Constant Substrates," Proceedings of the IEEE High Frequency Post-
graduate Student Colloquium, vol. 10, pp. 57-61, September 2005.
• T. Prodromakis and C. Papavassiliou: "Distributed Filter Design on Silicon CMOS,"

Proceedings of the IEEE International Symposium on Circuits and Systems, pp. 3309-
3312, May 2006.
• T. Prodromakis, C. Papavassiliou and G. Konstantinidis: "A Miniaturized Delay Line

based on Slow-Wave Substrates," Proceedings of the IEEE International Symposium
on Circuits and Systems, pp. 369-372, May 2007.
• T. Prodromakis, C. Papavassiliou and G. Konstantinidis: "Micro-scale LowPass Fil-

ters based on the Maxwell-Wagner Phenomenon," Proceedings of the IEEE European
Microwave Conference, vol. 1, p. 335-340, Oct 2007.
• T. Prodromakis, C. Papavassiliou and K. Michelakis: "Microstrip Stepped Impedance

LowPass Filters based on the Maxwell-Wagner Polarization Mechanism," Proceedings
of the IEEE International Symposium on Circuits and Systems, pp. 616-619, May
2008.
• T. Prodromakis and C. Papavassiliou: "Slow-wave Characteristics of Microstrip Trans-

mission Lines on Aqueous Dielectrics," Transactions of the IEEE Microwave Theory
and Techniques, Awaiting Decision.
• T. Prodromakis, C. Papavassiliou and K. Michelakis: "A Ka-band true-delay-line

phase shifter using capacitive MEMS switches", MEMSWAVE, July 2008.
191
Publications 192
• T. Prodromakis, P. Georgiou, T. G. Constandinou, K. Michelakis and C. Toumazou:

"Batch Encapsulation Technique for CMOS based Chemical Sensors", Proceedings of
the IEEE Biomedical Circuits and Systems (BioCAS) Conference, November 2008.
• T. Prodromakis, C. Papavassiliou and G. Konstantinidis: "Application of Gold Nan-

odots for Maxwell-Wagner Loss Reduction", IEEE Transactions on Nanotechnology,
Awaiting Decision.
• T. Prodromakis and C. Papavassiliou: "Surface Texturing for Maxwell-Wagner Po-

larization Engineering", Micro t4 Nano Letters, Accepted for publication.
• T. Prodromakis and C. Papavassiliou: "Engineering the Maxwell-Wagner polarization

effect", Applied Surface Science, Accepted for publication.
• T. Prodromakis, P. Georgiou, K. Michelakis and C. Toumazou: "Effect of Mobile

Ionic-Charge on CMOS based Ion-Sensitive Field-Effect Transistors (ISFETs)," Pro-
ceedings of the IEEE International Symposium on Circuits and Systems, Accepted for
publication.

Prodromakis T 2009 PHD Thesis

Uploaded by

Copyright:

Available Formats

Prodromakis T 2009 PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Prodromakis T 2009 PHD Thesis

Uploaded by

Copyright:

Available Formats

Application of Maxwell-Wagner

Themistoklis Pro dromakis

A thesis submitted for

Department of Electrical and Electronic Engineering

1.2 Research Objectives 8

1.3.1 Electromagnetic Wave Propagation on MIS Structures 9

1.3.2 MIS Modeling and Devices 9

1.3.3 High-K Structures Exhibiting Maxwell-Wagner Polarisation 9

1.3.4 Engineering the Maxwell-Wagner Polarisation Effect 10

1.3.5 Miniaturising Passive Components by Utilisation of MIS Microstrip 10

2 Electromagnetic Wave Propagation on MISM Structures 12

2.2 Space-Charge Layer Effects at Insulator-Semiconductor Interfaces 14

2.2.1 Ideal MOS Diode 15

2.2.2 The SiO2 — Si MOS Diode 19

2.3 Polarisation in Dielectrics 25

2.3.1 Maxwell-Wagner Polarisation Mechanism 26

2.4 Propagation modes of MIS Microstrip 30

2.4.1 Slow-Wave Propagation Mode 32

2.4.2 Attenuation Due to Semiconductor and Metallic Conductor 36

3 MIS Modeling and Devices 44

3.2 History of the MIS Lines Mathematical Analysis 45

3.2.1 Empirical Lumped Circuit Models 45

3.2.2 Parallel-Plate Wave Guide Approach 46

3.2.3 Full-wave Analysis 48

3.2.4 Quasi-Static Modeling 50

3.2.5 Single Layer Reduction Process 51

3.2.6 Device-Level Simulation 52

3.3 Devices Utilising MIS Microstrip 57

3.3.1 Voltage-Variable Capacitor 57

3.3.2 Delay Lines and Electronically Variable Phase Shifter 57

3.3.3 Voltage-tunable MIS filters 63

3.3.4 Variable coupling-coefficient directional couplers 64

3.3.5 Distributed FET transistors 64

4 High-K structures Exhibiting Maxwell-Wagner Polarisation 73

4.1.1 Technology Modelling 74

4.1.2 Measurement Apparatus 75

4.2 Silicon Based Structures 76

4.2.1 Characterisation Devices 76

4.2.2 Effect of Substrate Conductivity 79

4.2.3 Effect of Semiconductor to Insulator Thickness Ratio 82

4.2.4 Effect of Transmission Line Dimensions 86

4.3 GaAs Membranes 90

4.3.1 Implemented Process Flow 91

4.3.2 Results Discussion 91

4.4 Slow-wave Characteristics of Microstrip Transmission Lines on Aqueous Di-

4.4.1 Experimental Platform 93

4.4.3 Simulation Results 96

4.4.4 Experimental Results 97

4.5 Summary 103

5 Engineering the Maxwell-Wagner Polarisation Effect 106

5.1 Introduction 106

5.2 KOH on Si before oxidation 107

5.2.1 Process Description 108

5.2.2 Results 109

5.3 Unbuffered HF 5% on Si/Si02 111

5.3.1 Process Description 111

5.3.2 Results 112

5.4 Reactive Ion Etching of Si before oxidation 113

5.4.1 Process Description 114