47 / 2 Nov 2008

SILICON SEMICONDUCTOR DETECTORS FOR PARTICLE IMAGING

ERIK H.M. HEIJNE, CERN PH – DEPARTMENT

CH1211 Geneva 23 Switzerland
Abstract
Large-scale applications of silicon nuclear particle detectors have been developed during the recent
decade for elementary particle physics experiments and for space-based systems dedicated to studies of
cosmic radiation. These detectors typically consist of arrays of highly segmented sensors with matching
integrated circuit chips for the processing of the electric signals. Some of these systems cover an area
exceeding 100m2 with millions of sensor cells. In the past, the strong point of semiconductor detectors
was their capability to measure with high precision in a small volume the energy of ionizing particles.
Now these radiation detector systems, in addition, perform imaging functions that allow the
reconstruction in space and time of radiation emitting objects and processes or elementary particle
interactions. The imaging detectors also begin to find applications in materials analysis at synchrotron
light sources and in some types of industrial equipment. Basic principles are briefly described and a
few examples are discussed. The emphasis is on the instruments and reasons for their development are
indicated, but little is provided here about the significant results that they have enabled.

Submitted to Wiley VCH for Encyclopedia of Applied Physics, ed. R. Stock

1
Introduction

1. Basics of silicon detectors

- Device structure

- Detector segmentation and edges

- Energy deposition and signal generation in silicon detectors

- Manufacturing of semiconductor detectors

- Degradation of detectors

2. Various sensor designs and the associated readout electronics

- Custom-design of chips

- Imaging matrix with linear or 2D sensor geometry

- Factors affecting signal/noise

3. Typical silicon detector arrays in a particle physics experiment

- Tracking detectors for momentum measurement

- High precision detectors for interaction vertexing

- Semiconductor detectors for particle energy measurement

4. Other applications of semiconductor detectors

- Satellite - based experiments

- Detectors at synchrotron light sources

- Radiation imaging detectors in industrial or research equipment

Glossary

List of Works Cited / References

2
Introduction

Radiation, light quanta as well as energetic particles interact with semiconductor materials by liberating
free charge carriers, negative electrons as well as positive holes. In very pure silicon or germanium
crystals such charge carriers can move with a high mobility and a lifetime of ms. They can be
transported in an electrical field towards over distances of several mm on a ns timescale, and induce
electrical signals in the external contacts. Miniaturized electronic circuits then perform the charge signal
processing, and in this way it has become possible to study various properties of the incoming radiation.

In the early years, 1950-1970, semiconductor detectors were primarily used as a 'solid state' ionization
chamber. They offer the advantages of enhanced stopping power and compactness, compared with
magnetic spectrometers or improved precision in the energy measurement compared with other solid
state detectors such as sodium iodide scintillating crystals. This article describes developments of
semiconductor detector systems since ~1980, after a number of fundamental improvements were
introduced in detector manufacturing and in systems architecture [1]. In particular, mastering of custom
design of analog and mixed mode integrated signal processing chips in commercial CMOS
technologies proved to be an essential boundary condition for the introduction of large-scale arrays.
The properties of the semiconductor materials and the device technology are sufficiently complicated to
justify an intensive effort in R&D, with thousands of technically oriented articles already dedicated to
the study of these detectors. In exploiting these devices a multitude of scientific and practical results
have been obtained, often linked in a unique way to the innovation in these instruments.

The new applications of semiconductor detectors can be characterized as particle or photon radiation
imaging. The detectors now consist of a regular structure of segmented sensor cells, usually diodes,
each connected to a dedicated amplifier on an electronics readout chip. The amplifier circuit is matched
to the characteristics of the sensor element. The measured positions of the radiation quanta are
combined into a 2D or a 3D reconstruction of some object, process or interaction of elementary
particles. These imaging detectors use parallel processing of the signals from simultaneously incident
ionizing particles, sometimes even at rates in excess of 1010 cm-2s-1. Besides the hybrid detector
systems, in which the sensors and the readout electronics are still distinct components, one is developing
also monolithic detectors in which all functions are integrated in the same piece of silicon. In the hybrid
approach one can optimize functions in the separate parts, at the cost of complexity in interconnections,
whereas the advantages of full integration come with loss of flexibility in the electronics design and
manufacturing. However, the fully integrated devices profit from the rapid advances in silicon imagers
for consumer applications.

In outer space and in elementary particle physics experiments, several large arrays have been
constructed with a silicon sensitive area from ~m2 to over 200m2, consisting of thousands of modules
with fairly large Si sensor chips and Si CMOS electronic readout chips. Such a large system is
illustrated in Fig.1a and a close-up of a detector module in Fig 1b.

3
Fig. 1a. People are shown assembling the Atlas silicon tracking detector Fig. 1b. Part of a sector of the Atlas
SCT in a clean room at CERN. This detector system consists of a central forward silicon detector. This module
barrel with several cilindrical layers and perpendicular forward discs at both has been assembled at the NIKHEF
sides. Fully assembled, it has been placed in the center of the Atlas particle physics laboratory, in
experiment at the Large Hadron Collider LHC. The evacuated beam pipe Amsterdam, using boards tested in
passes along the axis of the cilinder. The energetic proton beams are several labs around the world, and
colliding around the mid-point of the barrel. Along this longitudinal axis equipped with Si detectors that were
the interactions take place, and particles originating from the vertex point manufactured by industrial suppliers,
fly in all directions. These will generally cross several planes of sensors readout chips custom designed in a
and allow at least 5 space points to be determined. More details are radhard, 0.25 µm CMOS technology,
discussed in section 3. (Photo CERN) and components for powering, alignment
and cooling. (Photo NIKHEF)

The turning point in elementary particle physics for abandoning photographic methods in favor of fully
electronic experiments has been around 1985. The ultimate, very large bubble chambers have been
decommissioned and replaced by instruments that feature combinations of silicon vertex tracking
detectors, gas-filled time-projection chambers and heavy calorimeters that measure total particle energy.
Large improvements in rate capability and in spatial precision have been achieved with this new
equipment. A major driving force for the development and building of the large silicon detector arrays
has been the need for improved precision in track reconstruction in physics experiments and the
identification of short-lived charm and beauty particles. Particle physics applications will be discussed
in some detail in section 3. A few other applications are mentioned in section 4, such as space-based
research and materials analysis with synchrotron light sources. Section 1 describes basic aspects of the
semiconductor devices and technology and sensor designs are enumerated in section 2.

The textbook Semiconductor Detectors by Bertolini and Coche [2] described principles and the
developments until ~1965. A general, comprehensive book on detectors is Radiation Detection and
Measurement by Knoll [3]. Recent books on silicon detectors have been written by Lutz [4] and by
Spieler [5].

1. Basics of silicon detectors

The physics of semiconductor devices is described, for example, in the famous textbooks of Grove [6]
or Sze [7]. Device properties are created by local modifications of electrical charge carrier
concentrations at micrometer and recently even at nanometer scale, in an initially homogeneous piece of
semiconductor. The extreme purity achieved in manufacturing of silicon crystals, which can be less

4
than one electrically active impurity atom on 1014, has allowed reliable and reproduceable silicon devices
to be used for countless applications.

OXIDE -
EDGE FRONT SIDE METAL CONTACT

xL p-TYPE IMPLANTATION
xL
p-n JUNCTION
xD
DEPLETION REGION

n-TYPE BULK

REAR SIDE METAL CONTACT +

Fig. 2 Schematic drawing of a silicon p-n junction diode. The rectifying p-n junction can be obtained by controlled
introduction of a p-type impurity such as ion-implantation of boron into the bulk n-type silicon. The p-n diode junction
area is defined by the opening that is etched in the surface oxide layer. The metal contacts usually are made with
aluminium. The diode is reverse biased by applying a potential between the + and - contacts. For high resistivity n-type
silicon this is usually a voltage between 20V and 500V. A depletion region extends a distance xD into the n-type bulk as
well as some lateral distance xL under the surrounding oxide edge.

- Device structure

The p-n junction diode, treated by Sze in his chapter 3 (chapter 2 in Sze's 3rd edition), is the basic
building block for a silicon detector. A schematic drawing is shown in figure 2. The rectifying
property is needed to limit the current in the sensing element which at room temperature would flow
when an electrical field is applied between ohmic contacts on Si. For many years the metal-
semiconductor Schottky diode gave the best achievable detector performance but in the recent imaging
detector systems one employs semiconductor-semiconductor p-n junctions obtained by local
implantation of doping ions with a keV beam. In both approaches a depletion layer is formed in which
practically no mobile charge carriers exist, so that no current flows unless charges are created by
incoming radiation. The concentration of free charge carriers in a semiconductor depends on the
bandgap, the temperature, doping with impurity atoms and crystal defects. A rectifying junction is not
needed for semiconductor materials with a large bandgap, such as diamond, as long as the contacts are
non-injecting.

Some unusual aspects of the diode structure for semiconductor nuclear particle detectors have to be
briefly mentioned here. First of all, whereas most other devices use a surface layer of a few µm at most,
a thick sensitive volume is needed for many of the detector applications and ideally the full wafer
thickness of several hundred µm is exploited. Reasons are that one may wish to slow down and stop
the incoming ionizing particle so that its total energy can be determined; one may need a high efficiency
for conversion of X-ray or g-ray photons or for traversing, swift particles one needs a sufficiently large
signal charge in comparison with noise fluctuations. For traversing particles, the signal is proportional
to the energy deposition along the pathlength in the sensitive thickness of the device. Characteristics for
energy loss are given later in this section. A second, related aspect is, that the rear side contact of the

5
thick device wafer needs special attention, contrary to the usual situation for electronics chips, where the
wafer acts simply as a mechanical support and grounded shield. For optimal functioning, only a small
reverse current in the reverse-biased diode can be tolerated, in order to detect the randomly arriving,
non-repetitive signal charge above the noise background. The undesirable reverse diode current is
increased by crystal or surface impurities which act as generation-recombination centers for free
carriers, by injection from contacts and by ambient factors such as temperature or light. Under optimal
conditions the generation current IG is the main component, proportional to the sensitive volume:
thickness w times area A, and is determined by the intrinsic equilibrium carrier concentration ni and
inversely proportional to the minority carrier lifetime t

IG = ( ni / 2t ) q wA

The electron unit charge is q (q=1.602 10-19C). In pure Si ni = 1010cm-3 and with the years, the
measured values tend to become a little lower as better silicon purity is achieved. By introducing n-type
or p-type doping impurity atoms one can increase the majority carrier concentration n or p over a large
range, even close to 1020cm-3 . The product of n and p remains a constant, equal to the value of ni at the
given temperature. The value of the intrinsic carrier concentration is a function of the semiconductor
bandgap Eg, the absolute temperature T and the effective densities of state in the valence band NV and
the conduction band NC (see Table1)

ni2 = n p = NC NV exp ( - Eg / k T )

The bandgap Eg in silicon is 1.12eV at room temperature. When the temperature is increased the
concentrations of electrons n and of holes p increase both and eventually become equal, at which
temperature the material is again intrinsic and highly conductive. As the total number of charge carriers
in a piece of semiconductor is n+p, the minimum number of carriers is obtained when n and p are
nearly equal, resulting in a low conductivity s or reciprocally a high resistivity r. The resistivity (unit
Ωcm) is also a function of the respective carrier mobilities me and mh (unit cm2V-1s-1)
1 1
r= =
s q(m e n + mh p)

and for detectors often values are used in excess of 1 kΩcm, corresponding to 1011 cm-3 electrons in n-
type Si. With such a low carrier concentration one can achieve low reverse current, and also obtain a
thick depletion region xD as shown in figure 2. When the diode is heavily doped at the rectifying
contact, the extension of the depletion layer is towards the lowly doped side of the diode, because of the
equilibrium between the amounts of charge at either side of the p-n junction, as described by the
Poisson equation (see Sze). For a p+ n junction* one obtains

2 e
xD = (V + VB ) = 2eme r (V0 + VB )
q n 0

* adding + to the carrier concentration indicates which is the more highly doped side, it does not indicate the sign of the
carriers.
6
where e is the permittivity (for Si 1.054 10-12Fcm-1), VB is the externally applied diode reverse bias
voltage and V0 is the built-in voltage of the unbiased junction. A typically applied voltage is in the range
20V to 500V, which can achieve several 100µm of depletion layer thickness in kΩcm Si. An essential
feature is the lateral extension xL of the depletion region sideways under the surface oxide. The
distance xL might be equal to xD but can also be different, depending on fixed charge that is present at
the interface or in the oxide itself. As will be discussed later, the lateral extension allows the complete
volume to be depleted in detectors with segmented diode contacts.

Such a diode detector presents, proportional to the surface area A of the device, a lower electrical
capacitance CD, the thicker the depletion region xD.

e A e
CD = = A
xD 2m er (V0 + VB )

The capacitance of a detector is an important factor in the noise performance for small signals, which
will be discussed in section 2.

A different situation arises in transistors in deep sub-micron CMOS technology, where the
miniaturization implies very shallow junctions. These require a concentration approaching 1020cm-3
which brings a large capacitance per unit of area. Yet, because of the nanovolume of such devices, the
actual values in a device remain small and only ~1000 unit charges q may determine the function of a
transistor. Fluctuations on such small numbers cause a significant spread in characteristics from one
device to another. In heavily doped Si the conductivity is not far from that caused by the ~1023 free
electrons cm-3 in metallic materials. The minority carrier concentration then becomes lower than 10 or
100cm-3 because under steady state conditions the product n.p remains constant as mentioned above.
In fact, this can be used to prevent minority carrier injection from an ohmic metal contact by creating
such a heavily doped semiconductor layer between the metal and the lowly doped bulk part of the
device.

Table 1 provides values for a number of properties of pure crystalline silicon. Some of these have a
determining impact on detector characteristics, as mentioned in the text.

7
Table 1 Properties of silicon

Atomic number 14
Atomic weight 28.09
Stable isotopes, natural abundance 28-29-30 (92.21% - 4.70% - 3.09%)
Density (300K) g cm-2 2.328

Atoms cm-3 5.0 x 1022

Crystal structure Diamond type, space group Fd 3m
Lattice constant (300K) cm 5.43 x 10-8
Melting point ˚C 1410
Linear thermal expansion coefficient ˚C-1 2.6 x 10-6
Specific heat (25˚C) J kg-1˚C-1 760

Thermal conductivity (25˚C) J m-1˚C-1 148

(77K) 2000
Dielectric constant 11.9
Permittivity Fcm-1 1.054 x 10-12
Bandgap (300K) eV 1.12
(0K) eV 1.16
Intrinsic carrier density (300K) cm-3 1 x 1010
Intrinsic resistivity (300K) Ωcm 3.2 x 105
Electron diffusion coefficient (300K) cm2s-1 35

Hole diffusion coefficient (300K) cm2s-1 12

Electron mobility (300K) cm2V-1s-1 1500

Electron mobility (77K) cm2V-1s-1 30 000 (approximate)

Hole mobility (300K) cm2V-1s-1 600

Hole mobility (77K) cm2V-1s-1 20 000 (approximate)

Carrier saturation velocity (300K) cm s-1 1 x 107

Breakdown field V cm-1 ~5 x 105
Electron affinity V 4.05
Work function V 4.8
Effective density of states NC in 3.2 x 1019
conduction band (300K) cm-3
idem NV in valence band (300K) cm-3 1.8 x 1019
* *
Effective mass m*/m0 ml = 0.98 m t = 0.19
Electrons * *

Holes
mlh = 0.16 m h h = 0.49
Raman phonon energy eV 0.063
Vapour pressure (930˚C) Torr 10-8
Energy required per electron - hole pair
(300K) eV 3.62
(77K) eV 3.76
dE/dx m.i.p. MeV g-1cm2 1.66

Nuclear collision length gcm-2 69

Si cm 29.64

Radiation length gcm -2 21.8

Si cm 9.36

8
- Detector segmentation and edges

An array of adjacent diodes, each with a separate contact, on a single piece of high-resistivity silicon
presents several interesting features for use in physics experiments or radiation imaging applications.
The positions of several simultaneously incident particles can be determined simply with reference to
the positions of the responding cells in the array. The segmentation, moreover, leads to miniaturization
of the elements, and this reduces their capacitance and noise, and also their leakage currents, even if the
current per cm2 is high. With the use of high resistivity and a reverse bias voltage that achieves full
depletion of the wafer thickness, one can avoid insensitive regions in the silicon volume, between the
segments, except around the outer edges of the sensor chip. Currently, special techniques are being
developed to increase the sensitive volume even quite close to the edges. This is important for
elimination of dead areas in applications with multi-sensor arrays and when detection is needed of
interaction products in proximity to a high intensity beam, which itself should not pass through the
detector.

Already around 1960 the first segmented surface barrier silicon detectors have been made. However,
the much more recent development of custom integrated circuits for signal processing and readout was
needed to allow a large scale use of such devices. It should be obvious that the characteristics and
possibilities of the signal processing chips basically determine the possibilities for different geometries
and properties of the sensor matrix. The segments have to be connected to the signal amplifiers and
further information processing. Most often the rectifying side of the sensor is segmented, because there
the electrical field is strongest, but it is also possible to segment the side with the ohmic contact, when
some precautions are taken. Finally, 'double-sided' detectors can be made where both sides have been
segmented.

The sensor manufacturing process imposes boundary conditions, such as needs for edge protection,
maximum wafer size and minimum achievable dark current. Also the irradiation of the sensors imposes
requirements for the sensor architecture. Various designs of segmented sensors and the associated
readout circuits are described and illustrated in sections 2 and 3.

9
 MOS CAPACITOR
ACCUMULATION
C OXIDE CAP V+ GATE
ox
excess e - OXIDE

Si n-type FIXED CHARGE

neutral
FLAT BAND Charge neutrality
DEPLETION
-
e carriers repelled V GATE
Cmin OXIDE

+ space charge
Si n-type FIXED CHARGE
neutral

INVERSION
V GATE
C
ox + OXIDE

excess e
FIXED CHARGE
neutral

Fig. 3 Schematic drawing of the MOS capacitor structure on n-type Si with capacitance C ox. in four different biasing
conditions From top to bottom are sketched the charge and carrier distributions at the oxide-silicon interface layer, for
accumulation, (flat band, no sketch), depletion and inversion. Fixed unit charges are encircled, the mobile charges are not.
The capacitor plate 'gate' on top of the oxide can be a doped, conducting polysilicon or an aluminium metallic contact.
The bulk volume, far from the interface, is electrically neutral n-type Si, with mobile carriers (electrons) compensating
the positive fixed charges.

Segmentation of silicon detectors relies to a large extent on the unique properties of the silicon dioxide
surface layer, which is created during very clean oxidation of a silicon wafer. Understanding the
physics of the Metal-Oxide-Semiconductor (MOS) system is useful for the design of segmented
devices. The interface between the oxide and the silicon acts as one plate of the capacitor structure, and
a metal or polysilicon plate on top of the oxide layer is the other side. The classical book MOS Physics
and Technology by Nicollian and Brews [8] treats this in detail. The schematic, simplified drawing of
figure 3 for the case of n-type Si shows the accumulation, flat-band, depletion and inversion states,
depending on the external voltage applied to the top capacitor plate. When the swing of the external
voltage is applied in a pulsed mode, the behaviour depends on the pulse frequency, and at >MHz the
buildup of the inversion state can not be achieved if the cycle time is shorter than the thermal generation
time constant for the minority carriers that have to populate this inversion layer. Then the depletion state
with a low capacitance value is maintained throughout the cycle.

The usual acceptor-type surface defects at the interface between the oxide and the silicon attract negative
charges, which always provoke a thin depletion layer in the n-type Si. Therefore, an oxide on n-type Si
10
naturally leads to a surface region with high resistance, while on p-type Si it provokes an accumulation
layer that can conduct current between segments. To avoid such short-circuiting channels between
diodes in p-type Si , special treatments or layouts must be introduced.

In the surface barrier technology the separation between evaporated metallic gold or aluminium contacts
depends on the quality of the native oxide layer of only some 3-4nm, which is formed as soon as a Si
surface is exposed to air. It was found that better stability can be obtained by oxidation in boiling water.
The resistance of such a native oxide can be ~ MΩ for 0.1mm. Much better controlled and higher
resistance, GΩ instead of MΩ, is achieved with modern oxidation technology.

In most segmentation schemes the electrodes on one plane are all connected to the same potential,
provided via the input contacts at the readout chip, so that there is no lateral electrical field between the
contacts in the matrix. All field lines then end in a regularly distributed way on the segmented diode
contacts.

- Energy deposition and signal generation in silicon detectors

Swift ionizing particles lose energy while traversing material. Various mechanisms are at work and an
introductory description can be found for example in 'Passage of particles through matter' by Bichsel,
Groom and Klein [9]. At low momentum the energy loss in Si is some keV per µm, and when the
momentum of the particle is much higher, ~0.5GeV/c, the mean specific energy loss goes to a minimum
of 0.38keV/µm. From that momentum range onward one speaks of 'minimum ionizing particles' m.i.p.
or mip's. One has to distinguish the energy loss of the particle itself and the energy that is deposited in
the absorbing medium. Some of the lost energy escapes from the medium in the form of energetic
secondary radiation. Also, the actual energy deposit in a thin silicon layer varies for each particle and
statisticallly it follows the so-called Landau energy distribution, shown in figure 4. In practice one must
consider the 'most probable energy deposition' for a particle which corresponds to the value at the peak
of the Landau energy loss distribution. Occasionally, large fluctuations may occur, which are caused by
close encounters with atomic electrons. Such an electron may receive a large energy transfer, and
escapes from the sensor volume as a 'delta-ray' electron. The specific energy deposited in a thin silicon
layer decreases with the layer thickness as shown in figure 5. While for a thick silicon layer of several
mm the peak of the distribution amounts to 0.36keV/µm, the value becomes as low as 0.23keV/µm for a
layer of ~50µm thick. The actual value of the energy deposition also depends on the immediate
surrounding of the sensitive sensor material.

In a semiconductor on average about one third of the deposited energy is spent on excitation of
electrons out from the valence band into the conduction band, where carriers can move through the
crystal. The remaining positive hole also becomes a mobile charge carrier. The average energy loss per
generated e-h pair is 3.62 eV. Of this, only 1.12 eV corresponds to the band gap energy in Si,

11
Fig.4 Measured and calculated Landau energy loss Fig.5 Most probable energy deposition in silicon detectors
distribution for 280 GeV muons traversing a 300µm of different thickness from 0.1 to 3.5 mm, that is the
thick Si detector in air, from [10]. The most probable values in eV/µm for the peak of the Landau distributions.
energy deposit is at the peak in channel 290, to which The curve is calculated for 200GeV muons. The + are
the calculation is normalized. One observes some measured muon data, determined as shown in Fig. 4, the x
accompanying secondary particles that give rise to an are electron data and the 0 are earlier calculations by Paul.
increased energy deposit probabilty at exactly twice the Detailed explanations can be found in [9] and [10].
Landau peak.

and the energy difference is dissipated as phonons and eventually as heat. In a detector of 300µm
thickness a minimum ionizing particle that traverses perpendicularly, loses ~80keV and generates
~22000 electron-hole pairs. These free electrons (as well as the holes) move under the influence of an
electrical field E, with a typical drift speed v, and a mobility µe depending on the temperature T

v = µe(T) E

The holes that are left in the valence band are mobile as well with mobility µh(T). Values for the
mobilities in high purity Si have been measured by Jacoboni et al. [11] and are 1500 Vcm-2s-1 for e-
and 600 for h+ . at room temperature. There is a strong temperature dependence of the carrier mobilities
in Si so that the transit time around room temperature varies ~1% for a 0.1 ˚C temperature variation.
The mobilities in Si increase with decreasing temperature and reach a quite high value at liquid
N2.temperature ~ 77K. Also the thermal conductivity is an inverse function of temperature, and at 77K
Si conducts actually better than Cu, although it can not compete in thermal condctivity with diamond or
modern diamond-like carbon materials.

The moving charge carriers represent a signal current in the diode and image charges are induced on the
diode contacts during the movement as described by Ramo's theorem. The situation for a multi-
electrode geometry with arbitrary potentials was described by Gatti et al. [12]. The actual measured
signal may be reduced by 'ballistic deficit' if the integration time of the amplifier is shorter than the
duration of the signal current, given by the charge transit time in the detector. In practice, for the usual
Si detector thickness of 300µm the hole transit time is <20ns for typical bias voltages, and full charge
integration can be achieved. In this situation the integral on nearby electrodes is zero, because the
initially induced charge is compensated when the moving carriers reach the collecting electrode(s).

12
- Manufacturing of semiconductor detectors

Most of the large radiation imaging detectors currently in use have been made with relatively cheap,
commercially manufactured, standard polished thin silicon semiconductor wafers, sliced from industrial
single-crystal ingots. Due to an immense effort in research and development of crystal manufacturing,
the silicon industry has achieved extreme electrical and chemical purity and crystal homogeneity, even
for a diameter of 300mm. In some specific applications, different semiconductor materials may present
advantages. However, with the exception of hyperpure germanium crystals, all other semiconductor
materials can not yet be obtained with similar purity. Also, compound semiconductors such as CdTe
often are not monocrystalline, but present a mosaic of sub-crystals with different orientations.

The Si crystals are grown in ever increasing dimensions, over 100cm long and up to Ø 45cm, most
often along <100> direction, although high-resistivity, >1kΩcm, is traditionally grown along <111>.
High resistivity n-type or p-type wafers are float-zone purified and they are available in diameters 2" or
3" and 100mm, 125, 150mm. The uniformity of resistivity is better in neutron-transmutation-
doped(NTD), n-type silicon than in float-zone n-type silicon wafers, where circular striations of ~mm
periodicity are observed with tens of % of resistivity variation. For recent, deep submicron CMOS
processing with 90, 65 or 45 nm feature size one uses now mostly 300mm diameter wafers, with
resistivity ~Ωcm. The progress in silicon material is described in the series of 4-yearly conference
proceedings 'Semiconductor Silicon 19xx', ed. R.Huff et al. [13].

Standardization in materials and in processing equipment for the CMOS industry has led to much
improved device quality and it has lowered production cost. Recent device manufacturing technologies
and parameters are treated in Silicon VLSI Technology by Plummer, Deal and Griffin [14]. This
industrial development also has facilitated the manufacturing of silicon particle detectors, which started
in 1980 with the introduction of 'planar' processing by Kemmer [15] and Burger [1].

In comparison with advanced CMOS technology, the processing of silicon particle detectors is
relatively straightforward. However, special precautions have to be taken because of the unusual
materials aspects such as the high resistivity and the need for a long minority carrier lifetime in the bulk
silicon. Also the applied reverse bias to the diodes, of several hundred volts, impose a special approach,
but which can be similar to that for power devices, as described in Semiconductor Power Devices by
Ghandhi [16] and Baliga [17]. The main breakthrough by Kemmer was the introduction of improved
cleanliness during the first passivating oxidation step at ~1200˚C. This became possible by advances in
MOS technology, but which had not yet been transferred to the detector manufacturing environment.
However, additional long exposures to high temperature, e.g. the usual 1100˚C anneal step, still have to
be avoided as these generally lead to the introduction and diffusion into the bulk volume of metallic
impurity defects which shorten the minority carrier lifetime and lead to an increase of dark reverse bias
current.

Kemmer described schematically the basic manufacturing steps for silicon detectors, as shown in
figure 6. Some critical details have not been included, such as the profiling of the diode edges and the

13
parameters of the cleaning steps. Commercial manufacturers understandably have been reluctant to
publish such manufacturing details.

Fig.6 Schematic process flow for 1cm2 diodes following the description by Josef Kemmer in US Patent 4,442,592[18]

After thorough cleaning of the silicon wafers, the first processing step is the growth of a passivating
oxide layer that completely covers the surface. Oxidation takes place in a quartz tube in a high-
temperature furnace. The Si wafers are exposed to a homogeneous flow of oxygen or water vapour. At
e.g. 1200˚C a typical SiO2 thickness of 25nm is grown in 10min. Often one adds HCl or, preferably, a
Cl containing compound in the gas flow, such as trichloroethylene, in order to prevent diffusion of metal
atoms from the outside heating coils through the hot quartz tube into the Si wafers.

The definition of the diode pattern on the wafer is obtained by illuminating a lightsensitive photoresist,
which after development presents a barrier layer of polymer material on the non-illuminated places and
open windows where the light came through the mask (positive resist). The creation of diodes needs the
implantation of doping atoms with an ion-implantation machine, which is in fact an ion accelerator with
a high current beam that is swept over the wafer area. By chosing a low ion energy, e.g. 15keV, the
penetration of the ions is only some tens of nm and one obtains a thin implanted thickness with high
concentration of doping atoms. In order to be electrically active, the implanted ions have to be located

14
on lattice positions and for this re-arrangement one needs an activation anneal at elevated temperature,
usually above 1000˚C. Care has to be taken that during this step also the radiation damage from the
implantation is repaired, but that no diffusion of undesirable atoms such as Na or Fe takes place.
Sometimes one provides special 'gettering' structures between the devices or on the back side of the
wafer, which can catch such undesirable atoms.

After processing, the individual sensors on the wafer have to be separated. Traditionally this is done
with a wafer-saw that cuts a ~60µm kerf in the 'dicing street' between the chips. Recently, singulation is
also done with laser cutting or with a plasma ion-etching. These methods promise to bring less crystal
damage along the edge of the cut , so that the insensitive edge width can be reduced.

The precision assembly of the sensors and the electrical, mechanical and thermal contacts require
careful design and construction. The position-sensitive devices aim at measurements with 2-10µm
accuracy. Manipulation and assembly are executed under dust-free conditions and at controlled
temperature, in medium-quality cleanrooms, as was already illustrated in figure 1.

- Degradation of detectors

During operation detectors may degrade for various reasons, such as a prolongated heat load that
causes metal diffusion into the pure Si material, or humidity that short-circuits isolation layers in critical
places of the device. Besides such accidental degradation, the changes induced by particles incident on
the sensor material are a fundamental and unavoidable degradation mechanism. Radiation quanta
generate free electrons by ionization of the semiconductor material, but at the same time a fraction of the
energy deposition leads to dislocations of atoms from their original crystal lattice positions. The ejected
atom is mobile as an interstitial, and also the created lattice vacancy can move around for a while, until it
combines into a stable defect. The movement of atoms, vacancies and complex defects is strongly
dependent on the temperature and the concentrations of impurity atoms. While in hyperpure Si the
concentration of electrically active impurities initially can be as low as 1012cm-3 there are usually
chemical impurities such as C or O at a concentration of order 1015cm-3 and these play an important
role in the stabilization of the radiation-induced defects. Many studies have been devoted to radiation
effects in semiconductors, and Si in particular (see Series of Conferences on Defects in
Semiconductors[19] ). The effects of radiation damage in high resistivity Si did not receive a lot of
attention until the recent decades. Some defects that at first are stable at a certain temperature, may
gradually convert into a different type of defect complex, for example by combining with a diffused
oxygen atom. Such a process, called annealing, is usually thermally activated, and often leads to
reduction of the reverse bias current. When, on the contrary, the new defects lead to a continuing
increase of dark current one speaks of 'reverse annealing'.

Unfortunately, the defect complexes in Si that result from the irradiations, such as vacancy-vacancy,
oxygen-vacancy, phosporus-vacancy, etc. present a variety of electrically active energy levels in the
semiconductor bandgap. Because the functioning of semiconductor detectors depends on a small
concentration of electrical charges, typically <1013cm-3, these devices are quite sensitive to changes in
surface charge and to radiation-induced defects. It is found that active defects are introduced at a rate of

15
~0.1 cm-3, per incident particle cm-2. Therefore, it is not surprising that a fluence (integrated flux) of
~1013cm2 particles leads to significant changes in the operation characteristics of a Si detector. It is
found that many radiation-induced defects exhibit acceptor-type behavior, and compensate the donor
characteristics of the traditional, n-type starting material for Si detectors. It was also found that an
operating temperature well below -10˚C can prevent formation of the most agressive type of defect,
which is believed to be a carbon-related complex. An excess of O in the starting Si seems also to lower
the introduction rate of this agressive defect. Watts [20] gave an overview of work on defects in the
particle physics community.

Recently, studies have started on the use of silicon crystals that have been grown in a different way,
such as epitaxially grown layers, or p-type instead of n-type material. First results show that different
types of stable defects are being obtained, due to different background conditions, and that in some
cases operation at +15˚C would be preferable above cold operation, due to beneficial annealing.

2. Various sensor designs and the associated readout electronics

The architecture of the readout electronics is really determining for the different sensor designs.
Initially, a single amplifier with discrete components was used for each detecting unit. For position-
sensitive devices it was possible to limit the number of readout channels to two, or maximum four, by
incorporating sophisticated, resistive charge division in the sensor construction, as reviewed by
Laegsgaard[21]. In many nuclear physics measurements such detectors with charge division are
adequate, but they can not handle a situation in which several particles are incident at the same moment,
at different positions. Yet, already around 1970 it was possible to exploit detectors with tens to
hundreds of segments using voluminous electronics, as in the BOL project in Amsterdam [22].

Table 2 Segmented silicon detectors - Naming convention based on cell dimensions

Detector Type Coordinate Long Edge Short Edge Associated Chip

Pad Detector 2D ≥ 1mm ≥ 1mm 16 to 64 channels

Microstrip Detector 1D ≥ 1mm < 1mm 128 channels

Hybrid Pixel Detector 2D < 1mm < 1mm 2048 to 65 536 ch

Monolithic Pixel Chip 2D < 30µm < 30µm >250 000 cells

Using custom-designed integrated circuits, with 64, 128, 256 or even tens of thousands of amplifiers
per chip, it has become much easier to apply parallel signal processing for a multitude of microscopic
sensor elements. This alternative approach opened up applications in multi-particle experiments in
particle physics, light detection and space, with thousands of cells per sensor. Although there are no

16
formal standards, one may distinguish sensor layouts following a classification of the cell dimensions,
as shown in Table 2, with 1 mm as the distinguishing criterion for the naming convention.

A basic distinction exists between a hybrid detector system and a monolithic detector system+ . The first
three rows in Table 2 refer to hybrid detector assemblies, in which the sensor cells, manufactured on a
separate silicon chip, are each separately connected to the first stages of the amplifier chains, located on
another chip. A typical assembly is illustrated in figure 7, which represents a so-called ladder, a module
for a detector array in the form of a barrel. Pioneering work on module assembly and readout chip
development was undertaken in a collaboration at SLAC, during the mid-eighties[23].
Fig. 7 Sketch of one of the first
ladder assembly designs for a silicon
detector module. This was made
~1987 for the silicon tracker in the
MARKII experiment at SLAC. The
module contains a single silicon
microstrip detector with 512 strips.
At both sides of the sensor alternate
strips are connected by wirebonding
to the inputs on the two 128 channel
Microplex readout chips, which are
glued on a ceramic support that also
carries the supply and filtering
components, and the connector for
the cable to the off-detector
electronics. Reproduced from
LaBarga et al. [23]

- Custom-design of chips

Segmented silicon sensors such as microstrip detectors or hybrid pixel detectors need a large number
of parallel amplifiers, and parallel-to-serial conversion of the signals in the chip circuit, so as to allow
more economical sequential readout by the data acquisition system. Hyams, Parker and Walker
initiated at Stanford the design of an NMOS chip, called Microplex [24], which was with 128 amplifiers
the first integrated design for silicon microstrip detector readout. Their design used a switched
capacitor feedback amplifier and this evolved later, thanks to Kleinfelder [25], Spieler [26] and
coworkers at Lawrence Berkeley Lab into a family of CMOS chips. These are used still now at
Fermilab in the CDF and D0 Tevatron collider experiments. In Germany, a team at the Werner
Heisenberg Laboratory, Max Planck Gesellschaft, Munich and the Microelectronics Institute in
Duisburg, with Lutz and Buttler, together with Manfredi et al. at the University of Pavia, developed a
128 channel CMOS circuit with JFET frontend transistors [27]. Later this chip was called CAMEX
and was used in the electron collider experiment Aleph at CERN, from 1989 to 2001.

An additional feature necessary with high-rate interactions in particle physics experiments is an on-chip
temporary data-storage memory, from which selected data that belong to a precise timeframe with a
positive trigger can be retrieved later. For the Large Hadron Collider experiments these memories

+ Monolithic designates the situation that all elements are built on a single 'stone' of silicon. Also a segmented sensor is
monolithic, but here we mean that sensor as well as amplifiers and other readout electronics is built in one chip.
17
usually operate at 40MHz frequency and provide storage for the signal charge quantity that is measured
every 25ns in each of the sensor elements. The memory has a ~180 cell depth that allows all data to be
kept until ~4µs later a first level trigger has determined which data set may be of interest.

Many of the front-end circuits use an amplifier with switching feedback and double correlated
sampling. A different line of devices, such as the 16-channel AMPLEX chip was initiated by
Jarron[28]. This family uses the more classical continuous feedback, which can be implemented in
different ways. For example, in the first AMPLEX a long transistor is used as the feedback element, as
shown in figure 8 [28]. This approach allows direct connection between sensor and amplifier input,
with compensation of the effects of the sensor leakage current. A current up to 450 nA per segment can
thus be compensated. The sensor dark current at the input modulates the effective resistance of the
feedback, so that the operational level at the output can be maintained, at the cost of a slight reduction of
amplification. The chip was produced in 1987 in the 3um CMOS technology of MIETEC in Belgium.
Fig. 8 Schematic diagram of the connection
of segmented diode elements of a silicon
microstrip detector to an Operational
Transconductance Amplifier (OTA) with a
feedback element presenting resistance Rf and
capacitance Cf . This type of feedback restores
the operating point also if there is a large
diode leakage current Idark in the sensor
element at the input of the amplifier [28].

A variety of readout chips with a linear array of amplifiers has been designed since ~1990 for use with
silicon microstrip detectors. The pitch of the diode elements on the sensor is not always identical to the
pitch of the amplifier input pads on the readout chip. For reliable wire-bonding connections, it would be
best to make pitch adaptation on the border of the sensor, using metal pads on a thick oxide in
combination with a separate depletion region in order to make the border independent of the sensitive
cells. In some systems one has introduced pitch adapters on a separate substrate such as quartz, but
then one has to double the number of connections.

- Imaging matrix with linear or 2D sensor geometry

Microstrip detectors provide a 1D projection of the particle positions. For 2D position measurement, a
second detector is needed, with the strips orthogonal, or under a different angle. When several particles
are simultaneously incident on such a sensor sandwich, ambiguities arise and a third sensor plane with
another strip orientation is needed to resolve these. The use of a double-sided microstrip detector
reduces the number of planes, but does not resolve ambiguities.

A true 2D device with a matrix of sensitive cells makes it possible to assign unambiguous positions to
individual particles, even if there are many co-incident on the same device. The first experiment that
practically adopted this approach at full scale was NA11. Damerell, Watts and colleagues [30] adapted
standard Charge Coupled Devices CCD for use with mips. They showed that with true 2D detectors
one can obtain a clean pattern recognition. The fully serial readout of the CCD makes it a quite slow
18
detector, and increased parallelism is required for high-rate experiments. The integrated nature and the
small pixel capacitance of the CCD result in low electronic noise, and a thin sensitive layer of ~20µm of
silicon delivers a sufficiently large signal.

While in a CCD the signal charge is transported from pixel to pixel, in a clocked parallel-serial
operation, all over the matrix towards a single output node, a different charge transport mechanism can
be implemented by the use of potential gradients in a homogeneous silicon volume. In the early 70's
this approach had already been adopted for light-sensitive devices. It has been introduced for position-
sensitive particle detection by Gatti and Rehak [31]. A lateral drift potential is superimposed on the
voltage that depletes the silicon bulk volume. Signal charges are drifted by the potential gradient as
well-defined packets towards the readout nodes, which form a linear matrix along one or two borders of
the silicon chip, as in a microstrip detector. A drawing of a practical drift detector for the Alice
experiment is shown in figure 9, reproduced from a description by Nouais et al.[32].
Fig. 9 Schematic drawing of a silicon
drift detector for the Alice experiment
[32], with 256 anode readout elements
along the top and the bottom. The
drift potential is vertical from the
center, both to the top and to the
bottom anodes. The drift potential is
distributed along the sides, and a
typical value is 2000V over the 3.5cm
drift region. Several rows of MOS
injection capacitors are indicated.
These can be used to monitor the
conditions for the drift velocity, which
is strongly dependent on the silicon
doping, radiation defects and the
ambient temperature.

While the drift detector geometry has a linear, one-dimensional matrix of readout pads similar to that of
a microstrip detector, it presents true 2D information on particle impact positions. The second
coordinate can be derived from the drift time, if the timing of the impact is measured independently, for
example from a common pulse pickup at the detector backside, or from the known bunch crossing time
in a collider experiment. An important advantage of the drift detector is the very low capacitance of the
readout node, in comparison with a microstrip detector of similar dimensions. However, the sensitivity
of the drift velocity on the uniformity of the devices and on the ambient temperature present practical
limitations which require continuous calibration procedures. Calibration is performed by charge
injection via built-in capacitors.

19
For high-rate experiments fully parallel hybrid pixel detectors have been developed, which consist of a
silicon sensor chip and a matching chip with readout circuits, as illustrated in figure 10. Each sensor
Fig. 10 Schematic drawing of a 'flip-
chip' hybrid silicon pixel detector
showing a partially transparent sensor
matrix chip and a matching CMOS
electronics readout chip underneath
Solder bumps are indicated, which
connect the individual sensor cells to
the corresponding readout elements in
a fully parallel fashion. The sensor
is supposed to be a matrix of p-n
junctions with a common ohmic rear
side contact covered with an
aluminium metallization.

cell is individually connected to a readout cell, using a microscopic solder connection, called bump-
bond. The first such hybrid pixel detectors had ~1000 pixels of dimensions 75µmx500µm, but the
most recent one has a matrix of 256x256 pixels of 55µmx55µm. Besides the need to register the
positions of incident particles one now aims also to incorporate a variety of additional functions in the
pixel circuitry, such as local encoding of the timing and the signal amplitude. For example, a 100 MHz
clock distribution has been implemented in the Timepix assembly [33]. The schematic circuit diagram
of the pixel electronics in this Timepix chip is reproduced in figure 11. This drawing shows the
complexity that can be achieved on a small, 55µmx55µm area with recent submicron CMOS
technologies.

Fig.11 Schematic diagram of the pixel electronics in the Timepix chip [33], which is situated in a 0.003mm2 area, and
reproduced 256x256 times on this 0.25µm technology CMOS chip.

20
- Factors affecting signal/noise

The signal amplitude in nuclear physics applications is mostly determined by the energy of the
incoming particle, often several MeV, provided that the particle comes to a complete stop inside the
detector. In particle physics experiments, on the contrary, the ionizing particles generally traverse the
silicon detector while undergoing only a fractional, small energy loss, typically of order 30keV per
100µm of Si thickness. The signal charge is practically independent of the particle energy, and only
determined by the pathlength in the detector, and the thickness of the sensor chip. In order for such a
mip particle to be detected with 100% efficiency, the noise of the signal processing chain has to be
sufficiently low. For readout electronics of particle detectors, the noise distribution is usually
characterized by the sigma of the variance, expressed as the number of equivalent noise electron charges
(ENC) at the input of the chain. For reliable discrimination of signals above a threshold from the noise
distribution it is desirable that the signal exceeds the noise baseline value by at least six sigma. For
reliable detection of a swift particle in a segmented detector, one also has to take into account the signal
charge division, when this is shared between adjacent segments.

The noise in the signal processing chain has various sources, and optimization can be achieved in the
sensor as well as in the front-end amplifier. The sensor capacitance, the signal shaping time, the power
dissipation and the dark current are prime objectives in the noise reduction. In the pixel detector one
can reduce both capacitance and dark current for a single pixel, while the signal amplitude remains
practically identical. Therefore, pixel detectors offer a very robust approach to detection with practically
100% efficiency for ionizing particles.

3. Typical silicon detector arrays in a particle physics experiment

During the recent decade the silicon detector arrays have become a standard feature of particle physics
experiments, around fixed targets as well as in the collider experiments. Immediately after the first
publications in 1980 of test results, the first fixed target experiments to embrace this new technology
were WA82 (in the Omega spectrometer), WA75 and NA11 at CERN, and the photoproduction
experiment E691 at Fermilab. About 1982 the CERN proton-antiproton collider experiments did not
yet use these devices, but in the final stage, 1986-1988, UA2 installed two layers of Si pad detectors in
order to recognize leptons and to reject signals from photon conversions in the inner volume. These
arrays had ~1.1m2 area with 6000 sensing elements. The MARKII experiment at SLAC and the
Fermilab collider experiments CDF and D0 then adopted from the beginning silicon microstrip
detectors for their inner trackers, all with tens of thousands of amplifier channels. By the end of the
80's most LEP experiments were preparing inner silicon detector barrel arrays, equipped with CMOS
readout chips, as mentioned in the previous section. Also the BaBar experiment at SLAC, and
experiments at several other, smaller accelerators incorporated silicon detector arrays. Now with the
new generation of experiments at the Large Hadron Collider LHC at CERN the collaborations have
installed silicon tracker arrays, unprecedented in size and complexity. A few examples will be

21
mentioned, but already hundreds of pages have been published to provide details of the design,
construction, system monitoring and expected performance for all these silicon detector systems.

The design of the silicon tracker arrays is usually based on the physics goals for particle tracking near
the interaction vertex. The precise measurement of the coordinates of the primary interaction point in
particle physics experiments and those of possible secondary vertices became more important with the
discovery of short-lived particles with a decay length of a few mm. A secondary vertex indicates the
presence of a charm, beauty or tau particle among the interaction products. Such a particle is a
messenger for a potentially interesting, special reaction. A second goal of the silicon imaging detectors
is the improvement on the momentum measurement of the outgoing particles. The precision needed to
achieve a given momentum resolution is a function of the radius at which the detector plane is installed,
and somewhat relaxed precision can be tolerated towards larger radii. This has led in several
experiments to a multitude of different systems, that are enclosing one another in the way of onion-
shells. Practical experience will show if this is the optimal approach. Maintenance, replacement and
extension operations and cost may eventually lead to modified designs. It may be mentioned that the
use of silicon tracking devices has led to order-of magnitude improved precision compared to earlier
employed gaseous detectors or scintillators, and this has allowed the overall volume of an experiment at
a given energy to shrink significantly. The higher cost of the silicon-based systems in the center can be
offset by the smaller volume of the magnet and other parts of the detector, including the outer muon
chambers.

In recent studies about requirements for future electron collider experiments, the need for improved
precision in the particle total energy measurement has become an issue. Some designs are being
evaluated of a 'silicon-based' calorimeter with tracking capability. Also the identification of original
leptons from the primary or secondary interaction vertex may be improved by the use of radiation
imaging arrays.

- Tracking detectors for momentum measurement

Silicon tracker detectors consist of several layers of silicon position sensitive devices with associated
readout chips. The layout for a typical example is shown in figure 12, after the description of the device
in the D0 experiment at Fermilab by Hanagaki et al. [34]. The outer layers in this tracker barrel were
supplemented at a later stage with layers 0A-B and 1A-B in the inner volume, very close to the
beampipe. Now being longer and carrying more sensors and readout chips, the modules in such a
barrel geometry still continue to follow the original design that was developed for MARKII, as shown in
figure 7.

22
Fig. 12 Schematic drawing of the 6 silicon microstrip detector layers in the Fermilab D0 experiment. The tracker is
divided in two groups, with the L0-1 layers in the centre and the earlier installed layers 2-5 at radii between 53mm and
164mm. From Hanagaki [34]

A photograph of part of the real silicon tracker for the CERN CMS collider experiment is reproduced
in figure 13 before complete cabling and encapsulation. Here one still can see the silicon sensors and
readout chips, very much arranged in a similar way to that of the previous figure 12. Once all
connections for power, signals and cooling have been installed, and thermal insulation is in place, not
much of the devices remains visible. The CMS silicon tracker has been described in numerous

Fig. 13 Photograph of the inside, the Tracker-Inner-Barrel, of the CMS silicon microstrip detector at CERN, during
construction in a cleanroom. Photo CERN, also published by Pooth [35].
23
publications, e.g. by Pooth [35[, and is the largest silicon array to date. The CMS tracker array has 10
cylindrical, overlapping layers in the barrel section between radius 20cm and 120cm, and 3 inner disks
plus 9 end-cap disks on either side of the barrel. The overall length is 5.6m. It covers an active sensor
area of 200m2, divided over more than 15 000 microstrip sensor units, with diode strip pitch between
100µm and 200µm, depending on the distance of the sensor to the interaction point. All sensors are
read out using the 128-channel APV chip, of which ~80 000 have been installed next to the sensors. A
critical aspect is the cooling of the instrument, which dissipates several tens of kW, and which has to be
maintained below -10˚C in order to preserve the quality of the high-resistivity silicon under irradiation.

The design of the CMS microstrip tracker allows between 8 and 14 measurement points on each
outgoing track, with position precision of the order 10-20µm. This is complemented by ~3 points
closer to the beampipe, measured in the pixel detector, which is situated within the tracker barrel.

- High precision detectors for interaction vertexing

For high precision vertex measurements one now often uses pixel detector planes, which offer direct 2D
space points with precision better than 10µm. The first use of a pixelized matrix detector was the CCD
in the NA11 experiment at CERN, already mentioned [30]. Due to the shifting of the charge signals
and serial readout via one or a few output nodes, the application of CCD is limited to some specific
types of experiments. Some can tolerate low rates, or, as in NA11, the beam can be deflected after a
positively triggered interaction so that the detectors are not flooded with additional, confusing signals.
Fig. 14 Perspective view of the
reconstruction of 153 tracks in a partial
angular cone of a lead-lead interaction in
the fixed target experiment WA97 [36],
using 7 planes of silicon hybrid pixel
detectors. The registered hits in the
planes are indicated by small squares,
with smaller squares for the planes closer
to the interaction vertex. Most tracks
originate at this vertex, and are shown
fully drawn. A few tracks cross only a
subset of the planes and do not pass
through this origin. These tracks are
shown with a hatched line. In the
analysis it was found that all hits
correspond to a physical particle track,
and no uncorrelated noise hits have been
observed. This is expected due to the
large noise margin in such a pixel
detector.

The magnetic field of the Omega

spectrometer had been switched off when
this interaction took place.

In high rate experiments one can use the hybrid pixel detectors described above in section 2. The first
application was in the high intensity environment of the CERN WA97 lead-ion beam, impinging on a
lead target [36]. A reconstruction of an event is shown in figure 14. The resolving power compared to
projective microstrip detectors as well as the low noise are clearly illustrated.
24
The pixel detector technology is a sophisticated integration of the sensor and the electronics readout
function, and it is attractive to think of a full monolithic device, wherein all functions are implemented
together. CMOS sensors for visible light achieve now this integration, and have become competitive
with CCD in many applications. However, in nuclear and particle physics experiments the devices have
to fulfill the need for single quantum processing, where each particle is recorded with its time of
incidence and other characteristics. Continued research is ongoing, in order to come to such monolithic
detectors for particle physics.

In the LHC collider experiments Alice, Atlas and CMS at CERN, one has constructed large arrays of
hybrid pixel detectors, with sensitive areas of respectively 0.24, 1.7 and 1.2m2. A part of the Atlas pixel
detector is shown during assembly in the cleanroom, in figure 15.

Fig. 15 The Atlas silicon pixel

detector during assembly in the
cleanroom. One sees a half-shell of
the barrel part. Ladders are supported
on a lightweight but rigid carbon-
fiber structure. On each ladder one
can distinguish a number of through-
connected modules. The white tubes
are the connections for the cooling
fluid.

(Photo CERN)

A comprehensive description of work on pixel detectors has been published as Pixel Detectors, From
fundamentals to applications by Rossi et al [37].

- Semiconductor detectors for particle energy measurement

The energy loss and the conversion into free electric charges in silicon is very reproducible and well-
understood. If an energetic particle can be made to stop in the detector volume, a constant fraction of its
energy is converted into free electron-hole pairs, on average 3.62eV per pair. For particles of much
higher energy one has built so-called calorimeters for the measurement of their total energy, and these
may be more than a meter thick to stop the electromagnetic or the hadronic shower that develops when
the particle comes to rest. A tedious calibration process is needed for the current calorimeter devices
that employ alternating sheets of passive absorber (Fe, W) and active detector material such as light-
emitting scintillator crystals or liquid argon, in which the ionizing secondaries generate mobile free
electrons and slowly moving ions.

Simulations show that improved performance can be obtained if the sensor would be made out of
silicon semiconductor detectors. Already ~1985 some prototypes had been tested but the technology
was practically abandoned in view of the high cost of silicon detectors. Now since 2006 the
development has been taken up again, and in the framework of the EUDET detector R&D for a future

25
international linear collider (ILC) new prototypes are being evaluated [38]. Besides the improved
energy measurement, also the increased calorimeter segmentation is an attractive feature, and in addition
such a highly segmented device acts as a continuation of the inner tracker.

4. Other applications of semiconductor detectors

The extensive efforts on silicon detector development in particle physics have paved the way for
applications in other fields. A few space-borne projects will be mentioned here, which sometimes have
been initiated by particle physicists. In other applications, for example protein diffraction with
synchrotron light, or materials analysis, scientists from particle physics have been involved as well.

- Satellite - based experiments

Since a long time silicon detectors have been used on-board satellites for intensity and energy
measurement and identification of cosmic rays. One is increasingly aware of the possibilities for
physics exploration that are offered by the natural accelerators that exist far away in the universe.
These generate particles with much higher energies than possible on earth, although the fluxes are
very small when reaching the solar system. The particle imaging capabilities of multi-layer silicon
tracker arrays allow a much improved directional detection of these incoming particles, so that one
may correlate them with a cosmic source. With this low flux, the sensitivity has to be increased by
enlarging the sensitive area. In recent years, balloon experiments such as PAMELA, described by
Straulino et al.[39] have used silicon detector arrays for the measurement and identification of
cosmic particles. One of the PAMELA detector modules is shown in the photograph in figure 16.
Fig. 16 Photograph of a silicon
detector plane for the PAMELA
experiment. Three ladders, each with
two double-sided detectors, are
mounted together on an aluminium
frame. The diode strips are wire-
bonded from sensor to sensor at the
left and then bonded to the readout
chips at the right. Flexible foils carry
the amplified signals further [39].

The basic ladder structure is maintained also here, as it was adopted originally for the MARKII particle
physics experiment at SLAC.

Between 1995 and 2008 two much larger experiments have been built for use in space, in which silicon
detector arrays are the main component of the instrumentation. The Fermi Gamma Ray Space
Telescope [www-glast.stanford.edu/], formerly called GLAST, has a Large Area Tracker LAT that
consists of 87m2 of Si micostrip detectors, 20 000 sensors and their associated readout chips. The

26
instrument has been described by Sadrozinski [40], and more recently by Hiroyasu Tajima[41]. Total
number of sensor readout channels is ~ 10 million, but the overall power budget is only ~250W, which
is provided by the solar panels. The GLAST/Fermi telescope has been succesfully launched on 13 June
2008.

The Alpha Magnetic Spectrometer project is a two-step experiment, described e.g. by Viertel and
Capel[42], with AMS01 (3200kg) and AMS02 (6000kg). AMS02 is to be launched by the NASA
shuttle and then to be operated at the International Space Station. The website [http://ams-
02project.jsc.nasa.gov/] gives numerous details. The experiment searches for new physics phenomena
in cosmic radiation. Inside a 0.8T superconducting magnet is an 8-layer ionizing particle tracker built
from double-sided Si microstrip detectors. Some practical aspects of the Si tracker and the construction
have been described in publications by D.Haas [43] and by M. Ionica [44]. In this instrument the total
silicon sensor area is 7.2m2 and it contains 2286 Si detectors, placed on 192 ladders with a total of
196608 sensing channels. For the readout one uses the VA-hdr chip [http://www.ideas.no/] with 32
channels per chip.

Fig. 17 Schematic drawing of the AMS02 detector with a cut-away that allows a view on the inner parts. The tracker is
indicated in the centre with a few sheets, surrounded by a Veto Counter and the magnet in its liquid He Vessel. Additional
particle identification instruments are a Time of Flight ToF and a Transition Radiation Detector TRD at the top (entrance
side) and a Ring Imaging Cerenkov Hodoscope RICH, Electromagnetic Calorimeter Ecal and another ToF at the bottom
(exit side). Reproduced from Haas [43].

To give an impression of such a space-based instrument, the schematic drawing of AMS02 is shown in
figure 17. The distinctive feature is the superconducting, liquid He cooled magnet, that should allow
charge determination and precise momentum analysis of the cosmic particles as they can be detectted
still outside the earth atmosphere. The instrument should be transported to the International Space
Station by one of the Space Shuttle Flights.

27
- Detectors at synchrotron light sources

Imaging of synchrotron light and X-ray diffraction operate often with photon signals that are similar to
that of minimum ionizing particles in silicon, in the range 5-100keV. For the lower end of this range
the absorption of Si for X-rays is quite satisfactory, and the silicon devices that have been developed for
particle physics experiments can serve well for different imaging applications at synchrotrons.

Medipix single photon processing detectors have been tested for low intensity imaging and prove very
efficient even for single quanta, as described by Tlustos[45]. A second imaging detector for high
photon intensity, with storage capability for up to 8 successive timeslots has been developed by Gruner
and his team[46].

'Pilatus' is the first large area silicon array detector for protein crystallography with direct photon
conversion, designed at the Swiss Light Source in the Paul Scherrer Institute in Villigen and now
produced by Dectris AG [http://www.dectris.com/]. The design is modular and an array can be
composed of a certain number of basic pixel detector modules that each carry 16 readout chips on a Si
detector with a matrix of pixels. The cell dimension in the most recent version is 172µmx172µm for a
60x97 matrix on a readout chip of appr. 10x17mm2. The modular structure is illustrated in figure 16.
Fig. 18 Assembly of a number of
'Pilatus' basic modules, which shows the
structure that allows a large array to be
constructed. A basic module carries a
single sensor chip with 2x8 readout chips
attached to it. The sensor faces the beam
with the homogeneous rear side, while
the segmented side is towards the readout
chips. The dimensions of the sensor on
top of the module on the rightmost rail
are 84.5mm x33.5mm.

from: http://www.dectris.com/

- Radiation imaging detectors in industrial or research equipment

CCD have been used since quite some time in research equipment, and in medical and industrial X-ray
imagers. Often the sensor is a scintillator with high-Z atomic number, in which the incident X-rays
generate scintillation light. This light is transported and focused onto and detected by the CCD. It
would be advantageous to replace such an indirect imaging by a sensor system that transforms the
radiation directly into electrical signals. Silicon can be used for X-rays in the range 4-20keV, and these
quanta produce signals that are quite similar to those from minimum ionizing particles. Based on the
experience and advances in particle physics instrumentation, it is then even possible to implement X-ray
imaging systems with single quantum processing. Imaging with individual treatment of each photon
still has to be studied in more detail to appreciate all advantages. An early, striking example is the
'Medipix' range of detectors, which is a spin-off from the Si pixel detector development for the Alice
28
experiment at the LHC. This hybrid pixel assembly has a monolithic silicon sensor matrix of 256x256
cells of 55µmx55µm, which is mated by solder bump bonding with a separate readout matrix in
0.25µm CMOS technology. A sophisticated signal processing for individual, high rate signals is
incorporated in each pixel, such as was illustrated in figure 11. In the most recent version Medipix3,
now using 0.13µm CMOS, one aims to acquire spectroscopic energy information on the individual
photons while building up the X-ray image, as described by Ballabriga et al. [47]. Also the previously
mentioned Pilatus imager system is based on this approach. For efficient absorption of energetic X-ray
photons in the range 30-150keV, or even higher, it is needed to use different semiconductor materials
such as CdTe.

A variety of applications can be thought of, including the use for visible light imaging with single
quantum photon detection. In this case one uses a photocathode that produces a primary electron,
which then is amplified in a gas-filled structure or in a vacuum tube with a microchannel plate or a kV
accelerating field.

29
Glossary
Bias voltage The usual silicon particle sensor has the structure of a rectifying diode and needs a
reverse bias voltage, that depletes the sensor volume of free charge carriers.

CCD Charge Coupled Device Matrix of imaging cells consisting of capacitors with transparent
polysilicon contact plates, so that light can enter in the silicon and generate free charge carriers,
proportional to the intensity.

Double-sided microstrip detector Semiconductor detector with a microstrip diode matrix on the
rectifying front side, as well as on the ohmic rear side. Often the microstrip patterns make an angle of
90˚. It may be useful to have a much smaller angle, such as 7˚, which reduces the problem of
ambiguities if several particles cross the detector matrix at the same time.

Drift detector Position sensitive detector in which small readout electrodes are placed at the end of one
or more carrier drift paths. These drift paths are defined by an electrode structure on the sensor
surfaces, and on which a potential gradient is applied, typically ~1V per µm.

Efficiency Detection efficiency is the percentage of particles or photon quanta that is detected in the
instrument, as a fraction of the number of quanta that are incident on the sensitive detector area. This is
equivalent to the probability that the signal charge generated by a particle in the detector volume can be
distinguished above the noise charge.

Ion-implanted detector Rectifying diode obtained by creating a p-n junction on a high resistivity
silicon slice, using an ion implantation accelerator. Boron is often the species that is used to create the
p-regions, and phosporous or arsenic is used to create the n+ ohmic rear contact of the detector diode.

Ladder Linear assembly of two or more sensors on some substrate, often also including the signal
readout circuits.

Leakage current Current that flows in a reverse biased diode when no incident radiation is present.

Microstrip detector One-dimensional matrix of contiguous sensitive cells (strips) with long dimension
above 1000µm and short dimension (pitch) below 1000µm.

Minimum Ionizing Particle m.i.p. or mip Swift particle with momentum at or above the value for
which the specific energy loss reaches a minimum. The material surroundings of the detector modulate
the actual value of the energy that is deposited in the sensitive volume. In a 'thin' (<20µm) Si layer in
vacuum the average specific energy deposit is ~230eV per µm, corresponding to an average generation
of 63 electron-hole pairs. In an 'embedded, thick' layer the values can be 360eV and 99 e-h pairs on
average.

Neutron Transmutation Doped NTD silicon Under neutron irradiation, usually close to a nuclear
reactor, silicon atoms can be converted into phosphorous atoms after neutron capture. This results in a
very homogeneous n-type doping profile if the crystal is rotated in the neutron flux. The radiation

30
defects have to be annealed and the doping atoms have to be electrically activated by a high temperature
treatment, usually above 1000 ˚C.

Noise charge The noise in the detector system is expressed as the Equivalent Noise Charge ENC at the
input of the amplifier in unit electron charges. The noise signal distribution is Gaussian and the width
can be expressed by the r.m.s. sigma value or by the Full Width at Half Maximum FWHM. The
relation between these is FWHM = 2.35 x sigma.

Pad detector One-dimensional or two-dimensional matrix of contiguous sensitive cells with all lateral
dimensions above 1000µm.

Pixel detector Two-dimensional matrix of contiguous sensitive cells with dimensions below 1000µm.

Surface barrier detector Rectifying diode obtained by metal deposition on a high resistivity silicon
slice, which is used as particle detector.

Type inversion During prolonged irradiation of n-type silicon the progressive introduction of
acceptor-type radiation defects leads to effective compensation of the original type and the material
exhibits properties of p-type silicon.

31
List of Works Cited / References
[1] Erik H.M. Heijne, 1980, a revolution in silicon detectors, from energy spectrometer to radiation imager:
Some technical and historical details, Nucl. Instr. Meth. A591 (2008) 6
[2] G. Bertolini and A. Coche, Semiconductor detectors, North-Holland Publ. Cy, Amsterdam,1968
[3] G.F. Knoll, Radiation Detction and Measurement, 3rd ed., J. Wiley New York, 2000 ISBN 0-471-07338-5
[4] G. Lutz, Semiconductor Radiation Detectors, Device Physics, Springer 1999 ISBN 3-540-64859-3
[5] Helmuth Spieler, Semiconductor detector systems, Oxford Science Publications,
Oxford, 2005 ISBN-13: 978-90-19-852784-8
[6] A.S. Grove, Physics and technology of semiconductor devices, J. Wiley and Sons New York, 1967
[7] S.M. Sze, Physics of Semiconductor Devices, 2nd edition, Wiley-Interscience,
New York 1981 ISBN 0-471-09837-X
S.M. Sze and Kwok K. Ng, Physics of Semiconductor Devices, 3rd edition, Wiley-Interscience,
Hoboken NJ 2007 ISBN 978- 0-471-14323-9
[8] E.H. Nicollian and J.R. Brews, MOS physics and technology, New York, 1982 Wiley
[9] H. Bichsel, D.E. Groom and S.R. Klein, Passage of particles through matter, section 27, Particle Physics in J.
Phys. G 33, 1 (2006) and http://pdg.lbl.gov
[10] Erik H.M. Heijne, Muon flux measurement with silicon detectors in the CERN neutrino beams,
Yellow Report CERN 83-06 21 July 1983
[11] C. Jacoboni, C. Canali, G. Ottaviani and A. Alberigi Quaranta, A review of some charge transport properties of
silicon, Solid-St. Electron. 20 (1977) 77
[12] E. Gatti, G. Padovini and V. Radeka, Signal evaluation in multielectrode radiation detectors by means of a time
dependent weighting vector, Nucl. Instr. Meth. 193 (1982) 651
[13] H.R. Huff, R.J. Kriegler and Y. Takeishi, editors, Semiconductor Silicon 1981, The Electrochemical Soc. Proc.
Vol. 81-5
[14] J.D. Plummer, M.D. Deal and P.B. Griffin, Silicon VLSI Technology : Fundamentals, Practice and
Modeling, Prentice Hall ISBN 0-13-085037-3
[15] J. Kemmer, Fabrication of low noise silicon radiation detectors by the planar process, Nucl. Instr. Meth. 169
(1980) 499
[16] Sorab K. Ghandhi, Semiconductor power devices, J. Wiley New York, 1977 ISBN 0-471-02999-8
[17] B.J. Baliga, Power semiconductor devices, PWS publ. cy. Boston ISBN 0-534-94098-6
[18] Josef Kemmer, Passivated semiconductor PN junction of high electric strength and process for the production
thereof, US Patent 4,442,592 filed 14 Jan 1981 (DE 31 jan 1980)
[19] A. Oshiyama, K. Maeda, K.M. Itoh and H. Katayama-Yoshida, editors, Proc. 23rd Int. Conf. "Defects in
Semiconductors", Awaji Island 2005, Physica B 376-377 Elsevier, 2006
[20] S.J. Watts, Overview of radiation damage in silicon detectors - Models and defect engineering,
Nucl. Instr. Meth. A386 (1997) 149 Workshop Vertex '96
[21] E. Laegsgaard, Position-sensitive semiconductor detectors, Nucl. Instr. Meth. 162 (1979) 93
[22] L.A.Ch. Koerts, K. Mulder, J.E.J. Oberski and R. van Dantzig, The 'BOL' nuclear research project,
Nucl. Instr. Meth. 92 (1971) 157 and following articles this issue
[23] Luis LaBarga, Chris Adolphsen, Giogio Gratta, Alan Litke, Michal Turala, Carla Zaccardelli, Alan Breakstone,
Sherwood Parker, Bruce Barnett, Paul Dauncey, David Drewer, John Matthews, Robert Jacobsen and Vera Lüth,
The Mark II silicon strip vertex detector and performance of a silicon detector telescope in the Mark II detector at
the SLC, IEEE Trans. Nucl. Sci. NS-38 (1991) 25
[24] J.T. Walker, S. Parker, B.D. Hyams and S.L. Shapiro, Development of high density readout for silicon strip
detectors, Nucl. Instr. Meth. 226 (1984) 200 (3rd Munich Symposium1983)
[25] Stuart A. Kleinfelder, William C. Carithers Jr., Robert P. Ely Jr., Carl Haber, Frederick Kirsten and
Helmuth G. Spieler, A flexible 128 channel silicon strip detector instrumentation integrated circuit with
sparse data readout, IEEE Trans. Nucl. Sci. NS-35 (1988) 171
[26] H. Spieler, Analog front-end electronics for the SDC silicon tracker, Nucl. Instr. Meth. A342 (1994) 205
Hiroshima STD Symposium
[27] G. Lutz, F. Maloberti, P.F. Manfredi, V. Re, V. Speziali, W. Buttler and H. Vogt, On the design of a
JFET-CMOS front-end for low noise data acquisition from microstrip detectors, Nucl. Instr. Meth. A264
(1988) 391
[28] E. Beuville, K. Borer, E. Chesi, E.H.M. Heijne, P. Jarron, B. Lisowski and S. Singh, AMPLEX, a low-
noise, low-power analog CMOS signal processor for multi-element silicon particle detectors, Nucl. Instr.
Meth. A288 (1990) 157 (5th Munich Symposium 1989)
[29] Erik H.M. Heijne and Pierre Jarron, A low noise CMOS integrated signal processor for multi-element
particle detectors, Proc. ESSCIRC'88 Conf., p.66 UMIST, Manchester, UK
[30] C.J.S. Damerell, R.L. English, A.R. Gillman, A.L. Lintern, F.J. Wickens and S.J. Watts, Use of charge-coupled
devices as high precision detectors, IEEE Trans. Nucl. Sci. NS-33 (1986) 51 (1985 Nuclear Science Symp.,
San Francisco)
[31] E. Gatti and P. Rehak, Semiconductor drift chamber - an application of a novel charge transport scheme, Nucl.
Instr. Meth. 225 (1984) 608
[32] D. Nouais, S. Beolè, M. Bondila, V. Bonvicini, P. Cerello, E. Crescio, P. Giubellino, M. Idzik, A. Kolozhvari,
S. Kouchpil, E. Lopez Torres, M.I. Martinez, G. Mazza, S. Piano, C. Piemonte, A. Rashevsky, L. Riccati, A.

32
 Rivetti, F. Tosello, W.H. Trzaska, A. Vacchi and R. Wheadon, The ALICE Silicon Drift Detector system, Nucl.
Instr. Meth. A501 (2003) 119 Vertex 2001, Brunnen
[33] X. Llopart, R. Ballabriga, M. Campbell, L. Tlustos and W. Wong, Timepix, a 65k programmable pixel readout
chip for arrival time, energy and/or photon counting measurements, Nucl.Instr. Meth. A581 (2007) 485
[34] K. Hanagaki, Design of an upgraded DØ silicon microstrip tracker for Run IIb at the tevatron, Nucl. Instr. Meth.
A530 (2004) 12 Firenze 6th 2003
[35] Oliver Pooth, The CMS silicon tracker, Nucl. Instr. Meth. A569 (2006) 21 Vertex2005
[36] F. Antinori, D. Barberis, H. Beker, W. Beusch, M. Campbell, E. Cantatore, M.G. Catanesi, E. Chesi, G.
Darbo, D. Di Bari, S. Di Liberto, D. Elia, E.H.M. Heijne, H. Helstrup, A. Jacholkowski, W. Klempt, K.
Knudson, I. Kralik, J.C. Lassalle, V. Lenti, L. Lopez, M. Luptak, P. Martinengo, F. Meddi, P. Middelkamp, M.
Morando, F. Pellegrini, E. Quercigh, L. Rossi, K. Safarik, G. Segato and S. Simone, Development of silicon
pixels for strangeness detection in nucleus-nucleus collisions, AIP Conf. Proc. 340, ed. J. Rafelski Conf.
Strangeness in Matter, Jan 1995, Tucson
[37] Leonardo Rossi, Peter Fischer, Tilman Rohe and Norbert Wermes, Pixel Detectors, From fundamentals to
applications, Springer Verlag ISBN-13 978-3-540-28332-4
[38] Report on EUDET-Calice Collaboration arxiv.org/pdf/0707.1245
[39] S. Straulino, O. Adriani, L. Bonechi, M. Bongi, G. Castellini, R. D'Alessandro, A. Gabbanini, M. Grandi, P.
Papini, S.B. Ricciarini, P. Spillantini, F. Taccetti, M. Tesi and E. Vannuccini, The PAMELA silicon tracker,
Nucl. Instr. Meth. A530 (2004) 168 Firenze 6th 2003
[40] Hartmut F.-W. Sadrozinski, The GLAST instrument, a gamma-ray large area space telescope, Nucl. Instr. Meth.
A419 (1998) 300 Wire Chamber 98 Vienna
[41] Hiroyasu Tajima, for the GLAST tracker team, GLAST Tracker, Nucl. Instr. Meth. A569 (2006) 140 Vertex
2005
[42] G.M. Viertel and M. Capell, The ALPHA Magnetic Spectrometer, Nucl. Instr. Meth. A419 (1998) 295 Wire
Chamber 98 Vienna
[43] Daniel Haas, The Silicon Tracker of AMS02, Nucl. Instr. Meth. A530 (2004) 173 Firenze 6th 2003
[44] Maria Ionica, Experience in the assembly of the large area silicon tracker for the AMS experiment, Nucl. Instr.
Meth. A513 (2003) 222 PSD-6 Leicester
[45] L. Tlustos, Performance and limitations of high granularity single photon processing X-ray imaging detectors,
PhD Thesis, dir H. Aiginger Wien TU, 2005 CERN 2005-003 [http://cdsweb.cern.ch/record/846447/files/thesis-
2005-032.pdf]
[46] G. Rossi, M. Renzi, E.F. Eikenberry, M.W. Tate, D. Bilderback, E. Fontes, R. Wixred, S. Barna and S.M.
Gruner, Test of a prototype pixel array detector for microsecond time-resolved X-ray diffraction, J. Synchrotron
Rad 6 (1999) 1096
[47] R. Ballabriga, M. Campbell, E. H. M. Heijne, X. Llopart and L. Tlustos, The Medipix3 Prototype, a Pixel
Readout Chip Working in Single Photon Counting Mode with Improved Spectrometric Performance, IEEE
Trans. Nucl. Sci. NS-54 (Oct 2007) 1824

33