Professional Documents
Culture Documents
EuCARD BOO 2010 004 PDF
EuCARD BOO 2010 004 PDF
PUBLICATION
17 August 2012
The research leading to these results has received funding from the European Commission
under the FP7 Research Infrastructures project EuCARD, grant agreement no. 227579.
This work is part of EuCARD Work Package 4: AccNet: Accelerator Science Networks.
The electronic version of this EuCARD Publication is available via the EuCARD web site
<http://cern.ch/eucard> or on the CERN Document Server at the following URL :
<http://cdsweb.cern.ch/record/1473436
EuCARD-BOO-2010-004
Pawe³ Gryboœ
Front-end Electronics
for Multichannel
Semiconductor
Detector Systems
CONTENTS
1. Introduction ................................................................................................................................1
2. Semiconductor detectors.............................................................................................................5
2.1. Materials for semiconductor detectors ............................................................................6
2.2. Reverse bias p-n junction ................................................................................................8
2.3. Charge generation in detector........................................................................................13
2.4. Charge transport ............................................................................................................15
2.5. Ramo theory and signal formation ................................................................................19
2.6. Detector geometry .........................................................................................................21
2.7. Important detector parameters.......................................................................................23
7. References ..............................................................................................................................173
Front-end electronics for multichannel semiconductor detector systems iii
Acknowledgements
This monograph is the result of countless interactions with many people who de-
voted their precious time and effort trying to teach me electronics. At various occasions
I have met them at universities, research institutes, conferences, meetings, or just on the
web. I would like to thank them all for their open mind, patience and cordial assistance.
I have also benefited from suggestions made by my reviewers: Prof. Kazimierz Korbel
and Prof. Andrzej Napieralski.
I wish to extend my appreciation to Robert Szczygieł for proof-reading and Mar-
cin Gryboś for his help with drawing figures. I am also grateful to Janusz Oleś and Boż-
ena Bryzek-Oleś for linguistic corrections. Finally, I would like to thank my wife Joanna
and family for their continuous support.
The part of the work presented in this monograph was supported by the Ministry
of Science and Higher Education, Poland, Projects no. N515 262 235 in the years
2008−2010 and Project no. N515 243 037 in the years 2009−2011.
Paweł Gryboś
iv Front-end electronics for multichannel semiconductor detector systems
LIST OF SYMBOLS
A fitting parameter
AF SPICE exponent constant for flicker noise
Aβ area proportionality constant of variation of current factor
ACox area proportionality constant of variation of gate oxide capacitance
Aγ area proportionality constant of variation of body factor
Ai real part of pole
AL proportionality constant of variation of channel length
Aμ area proportionality constant of variation of mobility
AP area proportionality constant of variation of parameter P
AW proportionality constant of variation of channel width
AVT0 area proportionality constant of variation of threshold voltage
a constant
α exponent constant of flicker noise
αN fitting parameter
b constant
β current factor in MOS transistor
βn, βp beta of transistor npn, beta of transistor pnp
C capacitance
Cb capacitance to the backplane
Cbu capacitance to the backplane per unit strip length
Cβ , CVT0 matching parameters
Cc coupling capacitance
Cdet detector capacitance
CF feedback capacitance
Cgs gate-source capacitance
Cgd gate-drain capacitance
Cin input amplifier capacitance
Cn capacitance to the neighbour strip (interstrip capacitance)
Cnu capacitance to the neighbour strip per unit strip length
Cox oxide capacitance per unit area
Cov overlap gate-diffusion capacitance per channel width
Cpar parasitic capacitance of connection detector - CSA
CT sum of capacitances Cdet, C1 and Cpar
Ctest test capacitor
C1 input transistor capacitance
D spacing distance
De,Dh diffusion coefficient for electrons, holes
DI threshold adjust implant dose
d detector thickness
E, Emax, Emin electric field, maximum electric field, minimum electric field
Eg energy bandgap
Eph photon energy
Ew weighting field
ENC equivalent noise charge
ENCf ENC contribution from flicker voltage noise
ENCi ENC contribution from current noise
ENCw ENC contribution from thermal voltage noise
εSi silicon permittivity
εox oxide permittivity
Front-end electronics for semiconductor detector systems v
F Fano factor
F(ω) amplitude characteristics
Fv,Fvf, Fi filter constants for ENCw, ENCf, ENCi
f frequency
fin average frequency of input pulses
fD duty factor of input pulses
f0 noise count rate at zero threshold level
fnoiMod switch for flicker noise model in BSIM4
φF Fermi potential
φMS gate-semiconductor work function difference
ϕT thermal voltage ϕT=kT/q
GBW gain bandwidth product
GDSNIO channel thermal noise coefficient in HSPICE
gds source-drain conductance
gm gate transconductance
gmb body transconductance
H0 constant
H(s) filter transfer function
γ body factor
γn thermal noise factor changing from weak to strong inversion
IBEAM, IBEAM0 beam intensity, primary beam intensity
ID0 process dependent parameter for subthreshold current
IDB drain-bulk current
Idet detector leakage current
IDS drain-source current
iD detector current pulse
ik current induced at k electrode
if normalized forward current
inoW ratio of if /W
di 2
power spectral density of current noise
df
j imaginary unit
J current density
JR, JRg reverse current density, reverse generation current density
k Boltzman constant
K constant
Kf flicker noise constant
Kfn , Kfp flicker noise constants of NMOS and PMOS transistors
KF SPICE flicker noise constant
Kv voltage gain
K0 DC gain
L channel length
Leff effective channel length
Lind inductance
Lmin minimum channel length
λ channel length modulation parameter
m fitting parameter
μ mobility
μatt attenuation coefficient
μe , μh mobility of electrons, holes
μeff effective mobility
vi Front-end electronics for multichannel semiconductor detector systems
1. INTRODUCTION
that each detector electrode is readout by an independent electronic channel. Such posi-
tion sensitive semiconductor systems have been used for many years in High Energy
Physics (HEP) experiments, where the number of readout electronic channels comes up
to several millions [4,5]. Nowadays, similar systems are used in different X-ray imaging
techniques in solid-state physics, material science, medicine, etc. In these applications,
the trend is to build the position sensitive detection system, which will provide also in-
formation about energy of photons.
For many years, the author has worked on different aspects related to front-end
electronics for semiconductor detector systems, namely:
− designing and testing silicon position sensitive detectors for HEP experiments and
X-ray imaging applications,
− designing and testing of multichannel readout electronics for semiconductor detectors
used in X-ray imaging applications, especially for noise minimization, fast signal proc-
essing, crosstalk reduction and good matching performance,
− optimization of semiconductor detection systems in respect to the effects of radiation
damage.
The presented monograph is the result of the author's experience in the
above-mentioned areas and it is an attempt of a comprehensive presentation of issues
related to the position sensitive detection system working in a single photon counting
mode and intended to X-ray imaging applications. The structure of this book is sche-
matically shown in Fig. 1.1.
2. SEMICONDUCTOR DETECTORS
electronics can provide not only binary information (YES/NO pulse on a given elec-
trode) but also information about energy deposited in the detector or about time event.
In particle physics experiments several layers of such detectors allow to reconstruct the
particle track. In X-ray imaging techniques these detectors can work in a single photon
counting mode providing information about the spatial photon distribution and some-
times also about photon energies. An advantage of single photon counting detectors is
essentially an infinite dynamic range contrary to the integration-type detectors (like
those used in CCD cameras), which usually have problems with a limited dynamic range
and a low contrast of an image.
This chapter presents briefly basic physic phenomena in a semiconductor detector
together with a process of pulse generation in the detector medium. Different possible
geometry (strip, pixel, pad) of these detectors is presented. The detector geometry and
the type of material used determine not only spatial resolution of the detector and its
efficiency, but also influence its electrical parameters like capacitance or leakage current
per single electrode. These parameters can be modeled in an equivalent electric scheme
of the sensor and later, easily implemented together with the front-end electronics dur-
ing numerical simulation.
A good material for a solid-state detector should possess the following features:
− large signal in response to particle/photon deposited energy, which requires
a small energy bandgap of a given material to ensure low average energy for the
hole-electron generation,
− low atomic number Z and low density in case of tracking application for particle
physic experiments (particle energy is measured in the calorimeter system),
− high atomic number Z and high density in case of X-ray and γ-ray spectroscopic
and imaging applications; high density leads to a large energy loss per traversed
length and higher probability of absorbing all photons in a beam,
− high mobility of charge carriers and no trapping effects to collect all the gener-
ated charge in a short period of time (important for an operation with high radia-
tion intensity),
− large carrier lifetime to increase so-called charge collection efficiency, defined
as a collected charge on the electrode to a total deposited charge,
− low leakage current in room temperature; too small bandgap can result in a sig-
nificant thermally generated current,
− large and high quality crystal (good homogeneity, low impurity levels, high re-
sistivity) to produce a large area detector,
Semiconductor detectors 7
For nearly three decades, the silicon has been the most popular material used for
semiconductor detectors [3, 6]. A very advanced silicon technology is driven by elec-
tronics industry. The silicon material fulfills nearly all of the above points, with the ex-
ception of high Z and radiation hardness. Because of low Z =14 and matured stable
technology, it is very attractive for a tracking detector in particle physics experiments.
These experiments are often performed on high luminosity machines like the Large
Hadron Collider (LHC), where the expected radiation doses are very high. For silicon
tracker doses up to 10 Mrad of ionizing particles and fluences of 1013−1014 neutron/cm2
are expected in over ten years of LHC operation [7]. In such conditions both the bulk
damages [8, 9] and ionization effects in silicon oxide [10, 11] are observed. The low
Z and low silicon density limit its X-ray applications mainly to low energy photons.
A standard 300 μm thick detector converts nearly all 8 keV X-ray, but only ∼26.7% of
20 keV X-ray and ∼2% of 60 keV X-ray. Therefore, many laboratories make an effort to
produce detectors more efficient for high X-ray energy, with the use of other semicon-
ductor materials [12, 13], such as high purity germanium (Ge) and compound semicon-
ductors, like gallium arsenide (GaAs), cadmium telluride (CdTe), cadmium zinc tellu-
ride (CdZnTe).
Germanium has smaller bandgap than silicon (0.67eV vs. 1.12eV for Si) and has
a very good energy resolution [14]. However, low energy bandgap greatly increases
a reverse current and a Ge detector typically is cooled to nitrogen temperature (77 K).
Germanium, because of high photoabsorption cross section σphoto ∼ Z 4-5, is more attrac-
tive for higher X-ray energies than silicon [15].
Gallium arsenide has been studied as the material for a semiconductor detector
for γ-ray since the early 1960’s [12, 13, 16]. Because of its potential radiation hardness,
8 Semiconductor detectors
it is also used in military applications and tested for possible applications in particle
physics experiments. GaAs has relatively high mobility. Because of impurities in the
order of 1015 cm-3, its carrier lifetime is only 10 ns. Due to trapping effects, it also suf-
fers from an incomplete charge collection.
CdTe and CdZnTe have high density and high Z (ZCd = 48, ZTe = 52, ZZn =30).
Because of large photon absorption cross section and possibility of operation at room
temperature, they are used for X-ray and γ-ray spectroscopic and also for medical imag-
ing applications [17-19]. CdTe and CdZnTe generally suffer from poor hole collection.
The hole mobility is very low μh ≈ 100 cm2/Vs and tends to be much smaller than for
electrons μe ≈ 1000 cm2/Vs. The detectors with CdTe Schottky contacts have lower
leakage currents than ohmic devices. However, Schottky detectors have a problem with
polarization effects [20]. The technology of CdTe and CdZnTe detectors is still limited
to small crystal and the connections with front-end electronics are more difficult than in
case of Si detectors.
Diamond, as a material with low Z, is a good candidate for tracking applications
in particle physics [21]. It has very good radiation hardness, even for radiation expected
at LHC. Diamond is classified as an insulator and in order to create hole-electron pair,
an average energy of 13.1 eV is required. Large detector samples have not been
achieved yet. Because of very long trapping times, the signal from a diamond detector
increases during radiation ("pumping" or "priming" effect [22]).
A reversed biased pn junction is a basic element for silicon detector and its pa-
rameters influence the detector characteristics. Let us consider an abrupt p+n junction
under the reversed bias as shown in Fig. 2.1. The following assumption is made: con-
stant doping densities of ND in the n-type side (i.e. ND = 1012 atoms/cm3) and NA in the
p-type side (i.e. NA = 1016 atoms/cm3). Even for zero applied bias voltage, there is a re-
gion at the junction where the mobile charges (electrons and holes) are removed, leaving
fixed acceptor and donor ions. This region is called the depletion region or space charge
region. Because of the fixed ions in this area, there is a built-in potential equal to [23,
24]:
N AND
Vbi = ϕT ln (2.1)
ni2
Vbi ≈ 1 V. If we apply reversed bias voltage VR , the total voltage across the junction
increases to VR+Vbi. The overall charge neutrality requires that
W1 N D = W2 N A (2.2)
where W1 and W2 are the width of the depletion region in p+ an n side of the junction.
Because NA >> ND the depletion region extends predominantly into the n-side region and
the width of depletion layer Wdep = W1 + W2 ≈ W1.
Fig. 2.1. The abrupt p+n junction under the reversed bias condition: a) junction, b) charge density,
c) electric field, d) potential.
10 Semiconductor detectors
d 2V qN D
=− for −W1 ≤ x < 0 (2.3)
dx 2
ε Si
d 2V qN A
= for 0 < x ≤ W2 (2.4)
dx 2 ε Si
where q is electron charge and εSi is the permittivity of silicon (1.04×10-12 F/cm). Inte-
gration of the above equations with boundary conditions E = 0 for x =W1 and x =W2
gives
qN D (x + W1 )
E ( x) = for −W1 ≤ x < 0 (2.5)
ε Si
qN A (x − W2 )
E ( x) = − for 0 < x ≤ W2 (2.6)
ε Si
qN DW1 qN AW2
E max = =
ε Si ε Si (2.7)
Integration of equations (2.5) and (2.6) gives voltage drops in the p+ side and n side of
the junction equal to
qN DW12
V1 = (2.8)
2ε Si
qN AW22
V2 = (2.9)
2ε Si
Using the eq. (2.2), (2.8) and (2.9) the total voltage drop across the junction is:
Semiconductor detectors 11
qN DW12 ⎛ N ⎞
Vbi + VR = ⎜⎜1 + D ⎟⎟ (2.10)
2ε Si ⎝ NA ⎠
Because for asymmetrical p+n junction (NA >> ND), the above equation can be rewritten
as
qN DW12
Vbi + VR = (2.11)
2ε Si
The width of a depletion layer Wdep ≈ W1 is given as:
2ε Si (Vbi + VR )
Wdep = (2.12)
qN D
There are two important conclusions from the above equation. The first conclusion is
that for the abrupt junction, the width of depletion region is proportional to square root
of the applied reversed voltage √VR. The second one is that a depletion voltage Vdep,
which guarantees the total depletion of the whole detector area (with the thickness d), is
lower for a purer detector material (lower ND or higher resistivity silicon). The full de-
pletion voltage is given as:
qN D d 2
Vdep = VR |Wdep = d = − Vbi (2.13)
2ε Si
For detector production silicon wafers with resistivity in the range from 5 to 10 kΩcm
are used. The relation between the resistivity ρn (n-type silicon) and the dopant concen-
tration ND is
1
ρn = (2.14)
qμe N D
where μe is the mobility of electrons (for low electric field μe is equal to 1350 cm2/Vs at
T = 300 K). Using formulae (2.13) and (2.14) the depletion voltage Vdep can be rewritten
12 Semiconductor detectors
d2
Vdep = − Vbi (2.15)
2ε Si ρ n μe
For example, for the given above resistivity of a detector wafer of 300 μm thick (based
on n-type material) the depletion voltage Vdep is in the range from about 31 V to 63 V.
The depleted junction volume is free from the mobile charges and forms a capaci-
tor. Since there is a voltage-dependent charge QC associated with the depletion region,
the junction capacitance can be calculated as
According to the above equation, the capacitance is inversely proportional to square root
of the applied reversed voltage √VR .
The reversed bias junction has always a dark or leakage current which can be ap-
proximated by the sum of diffusion components in the neutral region and the generation
current in the depletion region (for pno >> npo and |VR| > 3kT/q) [23]
where JR is the reversed current per unit area, ni is the intrinsic carrier concentration,
τg is the carrier generation lifetime, Dh is diffusion coefficient for holes and τh is carrier
lifetime for holes. In case of semiconductor with small values of ni (like Si), the genera-
tion current may dominate [23]. The additional components to the detector leakage cur-
rent are surface generation current and avalanche breakdown current at high voltage.
A method to reduce the detector current is to reduce the temperature. Consider the
generation current JRg where the temperature dependence is hidden in the intrinsic carrier
concentration ni and in the generation lifetime τg.The JRg is proportional to [4]
⎛ E ⎞
J Rg ∝ T 2 exp⎜⎜ − g ⎟⎟ (2.18)
⎝ 2kT ⎠
Semiconductor detectors 13
where Eg is the energy bandgap and k is the Boltzman constant. For silicon
(Eg = 1.12 eV) the temperature drop by every 8 K means the generation current reduc-
tion by factor of 2. So, in order to reduce the detector leakage current and noise associ-
ated with it, the detector is often cooled.
where IBEAM0 is the primary beam intensity and IBEAM(x) is beam intensity after crossing
the detector medium in x distance and μatt is an attenuation coefficient. The photon cross
section vs. energy in silicon is shown in Fig. 2.2.
The average number of e-h pairs produced by a single photon of energy Eph ab-
sorbed in a semiconductor detector is given as
E ph
N= (2.20)
Eeh
where Eeh is energy needed for ionization (energy required to create an e-h pair). In sili-
con Eeh = 3.6 eV and it is different from band gap energy Eg = 1.12 eV because an exci-
tation of an electron to a conduction band requires a simultaneous transfer of both en-
ergy and momentum. In silicon the minimum of the conduction band and the maximum
of the valence band have a wave vector offset and during the excitation the momentum
is transferred to lattice vibration [4]. For example, the X-ray photon of 8 keV (absorbed
in Si by photoelectric effect) produces about 2200 e-h pairs and the charge deposition is
located nearly in a single point (see Fig. 2.3(a)). The number of produced pairs is also
subject to statistical fluctuation described by Fano factor F according to [26]
ΔN eh2 = FN eh (2.21)
The Fano factor is a function of temperature and for Si the measured F value is between
0.07 and 0.16 [27−29].
(b)
(a)
Fig. 2.3. Generation of a hole-electron pair in silicon strip detector for: a) low energy X-ray - charge is
generated nearly in a single point, b) minimum ionization particle - charge is generated along particle
track.
A high energy charge particle (like in high energy physics experiments) traverses
the sensor and deposits there a part of its energy. The energy is deposited by many scat-
tering processes with electrons of sensor material and the energy deposition is described
by "long" Bethe-Bloch formula [30]. Along the particle trajectory hole-electron pairs are
Semiconductor detectors 15
produced (see Fig. 2.3(b)). A minimum ionizing particle produces about 77 e-h pairs per
μm path length in Si, which means about 23000 e-h pairs for the popular 300 μm thick
silicon detector. This is only the most probable value of generated charge because the
process of energy deposition for minimum ionizing particles is subject to statistical fluc-
tuation described usually by Landau-Vaviliov distribution (or Bichsel distribution - see
Fig. 2.4). These distributions show a long tail, so there is a significant probability of
higher energy deposition and more generated charge.
Fig. 2.4. A comparison of noiseless Bichsel and Vavilov distributions for a single 300 μm long track
segment (reprinted from [31] © 1995, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
There are two mechanisms of charge carrier transport drift and diffusion. In the
presence of an external field the charge moves parallel to the electric field and it is ac-
celerated between random collisions with the lattice. The average carrier drift velocity is
given by
r
ue = − μ e E (2.22)
r
uh = μ h E (2.23)
In silicon at low field (up to about 104 V/cm) the mobility is constant. At 300 K
the mobility is 1350 cm2/Vs for electrons and 480 cm2/Vs for holes. For high electric
field (E > 105V/cm) the carrier drift velocity saturates at the level of 107 cm/s. The holes
and electrons move in opposite directions and the time required for carriers to traverse
16 Semiconductor detectors
the detector volume is called collection time. Collection time depends on a charge gen-
eration point, detector thickness and the applied electric field.
The electric field in reversed biased p+n diode depends on the applied VR voltage.
Fig. 2.5 shows only the n-side of the diode for three different cases: partial depletion
voltage, full depletion voltage and overdepleted voltage. Compared to Fig. 2.1 the x-axis
is shifted of detector thickness to simplify the further calculations.
To shorten collection time the overdepleted case with VR > Vdep is mostly used. In
that case the field distribution can be written as
qN D
E ( x) = x + E min (2.24)
ε Si
where
VR − Vdep
Emin ≈ (2.25)
d
Let us consider a pair of carriers (electron and hole) generated at point x0 (see
Fig. 2.5(c)). The motion equations are the following [32]
dx ⎛ qN ⎞
= μ h ⎜⎜ D x + Emin ⎟⎟ for hole (2.26)
dt ⎝ ε Si ⎠
dx ⎛ qN ⎞
= − μe ⎜⎜ D x + Emin ⎟⎟ for electron (2.27)
dt ⎝ ε Si ⎠
Semiconductor detectors 17
Integrating the above equations with the initial condition x = x0 at t = 0 one obtains for
hole 0 ≤ t ≤ th
ε Si ⎛ ε ⎞ ⎛ qμ N ⎞
xh = − Emin + ⎜⎜ x0 + Si Emin ⎟⎟ exp⎜⎜ h D t ⎟⎟ (2.28)
qN D ⎝ qN D ⎠ ⎝ ε Si ⎠
ε Si ⎛ ε ⎞ ⎛ qμ N ⎞
xe = − Emin + ⎜⎜ x0 + Si Emin ⎟⎟ exp⎜⎜ − e D t ⎟⎟ (2.29)
qN D ⎝ qN D ⎠ ⎝ ε Si ⎠
⎛ qN D ⎞ ⎛ qμ N ⎞
uh = μ h ⎜⎜ Emin + x0 ⎟⎟ exp⎜⎜ h D t ⎟⎟ (2.30)
⎝ ε Si ⎠ ⎝ ε Si ⎠
⎛ qN D ⎞ ⎛ qμ N ⎞
ue = − μ e ⎜⎜ Emin + x0 ⎟⎟ exp⎜⎜ − e D t ⎟⎟ (2.31)
⎝ ε Si ⎠ ⎝ ε Si ⎠
The collection times th and te can be calculated directly from (2.28) and (2.29)
ε Si d + (ε Si / qN D )Emin
th = ln (2.32)
μ h qN D x0 + (ε Si / qN D )Emin
ε Si x0 + (ε Si / qN D )Emin
te = ln (2.32)
μe qN D (ε Si / qN D )Emin
ε Si d + (ε Si / qN D )Emin
th max = ln (2.33)
μh qN D (ε Si / qN D )Emin
18 Semiconductor detectors
ε Si d + (ε Si / qN D )Emin
te max = ln (2.34)
μe qN D (ε Si / qN D )Emin
The examples of the maximum charge collection time (assuming that charge cross the
whole detector thickness) for different bias voltages of the detector are calculated in
Table 2.2. For the calculation it is assumed that the silicon resistivity is 10 kΩcm
(Vdep = 31 V) and the detector thickness is d = 300 μm.
Table. 2.2. Examples of the maximum collection times in Si detector with Vdep = 31 V and d = 300 μm
calculated according to equations (2.33) and (2.34).
Applied voltage VR [V] temax for electrons [ns] thmax for holes [ns]
45 17.9 50.2
60 12.1 34.1
90 7.6 21.5
120 5.6 15.9
180 3.7 10.5
The long collection time can influence the energy resolution of the position sensi-
tive detector because of a diffusion process. In case of gradient concentration, the ran-
dom movement of the charge carries is more probable in the direction of lower concen-
tration. So, the cloud of charge carriers spreads out with respect to the trajectory without
diffusion. The average square deviation Δr 2 during the time interval t is [32]
Δr 2 = 2 De, ht (2.35)
⎛ kT ⎞
De , h = ⎜⎜ ⎟⎟ μe , h (2.36)
⎝ q ⎠
Semiconductor detectors 19
In silicon the diffusion coefficients are De ≈ 35 cm2/s (3.5 μm2/ns) for electrons and
Dh ≈ 12 cm2/s (1.2 μm2/ns) for hole. The spread of thermal diffusion is larger for longer
time collections, so for this reason, in position sensitive detectors the collection time
should be minimized.
The second important case where collection time plays an important role, is the
case of using very fast readout electronics. The shaping time in the readout electronics
should be longer than the collection time to integrate all the charge generated in the de-
tector. Otherwise, the pulse amplitude in the readout electronics is proportional only to
the fraction of generated charge and this phenomenon is called ballistic deficit [4].
After the charge generation, the holes and electrons start to move in the detector
medium because of the applied electric field. A movement of generated charge carriers
induces current pulse in detector electrodes (a current flow i(t) in the circuit connected
to the detector electrode). The current flow begins instantaneously and finishes when all
the charges (both electrons and holes) are collected. The polarity of induced currents by
holes and electrons on a given detector electrode are the same, because the charges of
opposite signs move in opposite directions. The current ik(t) induced at k electrode of the
detector by the infinitesimal movement of the charge q is described by the Ramo theo-
rem [35]
i k (t ) = quk Ew (2.37)
where uk is drift velocity and Ew is a weighting field. The weighting field is obtained
by applying unit potential to k electrode and zero to all other electrodes and it is differ-
ent from the electric field inside the sensor. For practical cases (such as strip or pixel
detector) the numerical calculation is necessary to obtain pulse shape at different elec-
trodes. The pulse shape strongly depends on initial distribution of electrons and holes,
detector geometry and electric field distribution [32, 36].
Consider the simple case with parallel plate geometry and linear field distribution
for VR > Vdep as shown in Fig. 2.5(c). The weighting field is in this case Ew = 1/d. Ac-
cording to the Ramo theorem the following expression can be written for current in-
duced by the motion of charge carriers (see eq. (2.30)−(2.31))
20 Semiconductor detectors
q ⎛ qN ⎞ ⎛ qμ N ⎞
ih (t ) = μ h ⎜⎜ Emin + D x0 ⎟⎟ exp⎜⎜ h D t ⎟⎟ (2.38)
d ⎝ ε Si ⎠ ⎝ ε Si ⎠
q ⎛ qN ⎞ ⎛ qμ N ⎞
ie (t ) = μe ⎜⎜ Emin + D x0 ⎟⎟ exp⎜⎜ − e D t ⎟⎟ (2.39)
d ⎝ ε Si ⎠ ⎝ ε Si ⎠
q
ih (t ) = μ hVR (2.40)
d2
q
ie (t ) = μeVR (2.41)
d2
The examples of the current pulses for the uniform field are shown in Fig. 2.6(b).
(b)
(a)
Fig. 2.6. Examples of pulse shapes for parallel plate geometry and: a) linear field distribution, b) uniform
field [32, 37].
Semiconductor detectors 21
(a) (b)
Fig. 2.7. Simplified view of strip position sensitive detector: a) single sided strip, b) double sided strip.
Fig. 2.8. Ambiguity of position reconstruction in double sided silicon strip detectors.
22 Semiconductor detectors
The double sided detector (see Fig. 2.7(b)) is an interesting option for 2-D imag-
ing. The second orthogonal set of strips is made on the backside of the detector, how-
ever this solution is limited to a low intensity radiation experiment [42]. Assume that
two photons hit the detector at the same time (see Fig. 2.8). The two photons produce
signals in two upper and bottom strips. From the reconstruction of the hits, four posi-
tions are possible: two real hits and two "ghost" hits. To minimize the number of "ghost"
hits, radiation intensity must be limited or a very fast and precise coincidence system
has to be used.
Fig. 2.10. Hybrid pixel detector - sensor and readout chip are connected together using bump-bond
technique.
A real two dimensional imaging is possible using pad or pixel detector - see
Fig. 2.9 and Fig. 2.10. A pad detector requires a trace to each electrode, because the
readout electronic channels are connected at one, two or four sides of this detector (usu-
ally by a simple wire bonging technique) [43]. This limits the maximum number of the
readout channels (to about several hundreds of channels). A pixel detector does not have
Semiconductor detectors 23
The 3-D model of a DC coupled Si strip detector is shown in Fig. 2.11 [53]. The
components shown in this scheme represent:
− Cbu - capacitance to the backplane per unit strip length,
− Cnu - capacitance to the neighbour strips per unit strip length (interstrip capacitance),
− Rsu - effective resistance of strip per unit strip length.
The effective strip resistance is a parallel connection of a highly doped semiconductor
strip (diffusion/implantation p+ or n+ type) and a low resistivity metal strip. The resis-
tance of the metal strip (usually Al) per unit length depends on the metal thickness and
width. For example, for 1 μm thick and 20 μm wide Al strip has resistance of
15−20 Ω/cm.
(a) (b)
If the current pulse shape is not important for the performed experiment (in most
cases the current pulse is integrated in a charge sensitive amplifier), then the strip resis-
tance can be neglected and the model can be significantly simplified to the scheme
shown in Fig. 2.12(a), where Cb and Cn represent the total capacitance of the strip to the
backplane and to the neighbour strips respectively. An AC-coupled detector model
(Fig. 2.12(b)) contains additionally coupling capacitance Cc (between p+ strip and metal
strip) and the bias resistance Rbias.
The parameters of the detector are usually measured before connections of the
readout integrated circuit, however, some of them can be approximated in advance. The
backplane capacitance per strip length is
Cb ε Si p
≈ (2.42)
l d
where p is the strip pitch and d is the detector thickness. This capacitance slightly de-
pends also on the strip width. For example, for a 280 μm thick detector with 50 μm pitch
the measured Cb/l is 0.15−0.18 pF/cm (for strip width w is in the range from 10−30 μm)
[53]. The capacitance to the neighbour strip (interstrip capacitance) in a Si detector can
be approximated as [54]
Cn ⎛ w + 20μm ⎞
= ⎜⎜ 0.03 + 1.62 ⎟⎟ (2.43)
l ⎝ p ⎠
where, in the above formula, the strip width and pitch in μm and Cn/l are obtained in
pF/cm. This capacitance depends on the quality of field oxide between strips and usually
increases after strong irradiation. In most cases this capacitance is in the range of
0.8−1.5 pF/cm [52−55].
Semiconductor detectors 25
In case of an AC-coupling detector, one should also take into account also de-
coupling capacitance Cc/l = 10-30 pF/cm and bias resistance Rbias. The bias resistance
depends on the used structure to bias the detector (diffusion resistor, FOXFET structure,
etc.) and it varies in the range from 200 kΩ up to over 100 MΩ.
An additional detector parameter is detector leakage current Idet which strongly
depends on the bulk material quality (high resistivity Si is better but more expensive),
process production, temperature and radiation damage. For example, for a good new Si
detector with 50 μm pitch and standard 300 μm thickness, this leakage current should be
below 100 pA/cm (at room temperature).
A simplified model for a pixel detector is similar to the one shown in
Fig. 2.12 (a). The capacitance to the backplane for pixel area A is
ε Si A
Cb ≈ (2.44)
d
For example, for a 300 μm thick detector with the pixel area A equals to
100×100 μm2, this capacitance is about 3.5 fF. The more important component is ca-
pacitance to the neighbour pixel and it strongly depends on implant/diffusion area. One
has to expect the total capacitance for a 125×125 μm2 square pixel is of ≤ 100 fF, and of
about ≤ 200 fF for a long and narrow 400×50 μm2 pixel [5]. Because the capacitance of
a pixel detector itself is really small, other parasitic components of the closely spaced
detector − readout electronics should be taken into account (capacitance between the
detector and ground/bias plane of readout IC, capacitance of bump bonds etc.).
As for the noise performance of the detector − readout system, three parameters of
position sensitive detector are important:
− total capacitance per strip/pixel (sum of bulk capacitance, interstrip capacitance, para-
sitic capacitance of the connection with readout IC),
− shot noise of detector leakage current (per strip/pixel),
− thermal noise of the bias resistor (AC coupled detectors only).
The noise performance of readout electronics is discussed in detail in Chapter 3.
The second effect related to detector capacitance is the crosstalk between detector
electrodes. Due to interstrip/interpixel capacitance, the charge deposited on a single
electrode can induce parasitic signals on neighboring strips/pixels. Consider the simple
scheme shown in Fig. 2.12(a). In most systems, the relationship between input capaci-
tance Cin of readout electronics and detector capacitances are the following
Fig. 2.13. Simplified model of one detector electrode together with the input capacitance Cin of readout
electronics.
The simplified scheme of one electrode node can be drawn as shown in Fig. 2.13.
The charge Qtot deposited at node X leads to the voltage
Qtot
VX = (2.46)
Cin + 2Cn + Cb
Qtot ⎛ 2Cn + Cb ⎞
Qin = × Cin = Qtot ⎜⎜1 − ⎟⎟ (2.47)
Cin + 2Cn + Cb ⎝ Cin + 2Cn + Cb ⎠
The remaining charge fraction (Qtot − Qin) is located by half on left and right elec-
trodes. To minimize this crosstalk between the electrodes, the effective input capaci-
tance Cin of readout electronics must be much higher than the total strip/pixel capaci-
tance (2Cn + Cb).
Architecture of front-end electronics 27
3. ARCHITECTURE OF FRONT-END
ELECTRONICS
The current pulse provided by the sensor is amplified and shaped in the front-end
electronics. Different modes of signal acquisition are possible: current mode, voltage
mode or mode with Charge Sensitive Amplifier (CSA) at the input. An example of
front-end electronics using the mode with CSA is schematically shown in Fig. 3.1.
A current signal generated in the silicon strip/pixel detector is integrated in a charge
sensitive preamplifier. At the output of the preamplifier one obtains a voltage step with
amplitude proportional to the total charge generated in the detector. The voltage step is
fed to the main amplifier called a shaper, which provides pulse shaping according to the
timing requirements and the filtration of noise to maximize the signal to noise ratio.
Further processing of the shaped signal can be done in different ways depending
on specific applications. Possible options are schematically shown in Fig. 3.1. The first
one is based on so-called binary readout architecture. In that case the comparator detects
28 Architecture of front-end electronics
the presence of the signal of amplitude above the preset threshold and in response pro-
vides 1-bit yes/no information. The second way of processing the shaped signal employs
an analog-to-digital converter (ADC), where the amplitude of the signal corresponding
to each individual photon is measured, and then the information is stored and used for
off-line processing. The third option is used for timing measurements - determination of
the time of occurrence.
The number of front-end channels in readout systems in most cases is equal to the
number of strips/pixels in the semiconductor detector. In case of readout electronics for
strip detector, multichannel integrated circuits are mostly designed as 32, 64 or 128 –
channels integrated circuits, which are the basis for building larger modules consisting
of several hundreds up to a few thousands readout channels (see Fig. 3.2) [56].
Fig. 3.2. Fragment of multichip module: 512- strip detector connected to eight 64-channel readout ASICs.
Architecture of front-end electronics 29
The signal reception from a detector can be realized in the front-end electronics in
three main modes [32, 37]:
− current mode: to preserve pulse shape,
− voltage mode: to obtain high signal amplitude at the amplifier input,
− mode with charge sensitive amplifier to integrate a detector current pulse.
As the option with the CSA is most widely used in the front-end electronic system
it is presented in detail in the next chapter. To distinguish between the current and volt-
age modes, a simplified scheme of a detector and input amplifier is useful (see Fig. 3.3).
The capacitance CT is a sum of detector capacitance Cdet, input transistors capacitance C1
and stray capacitance Cpar of the connection between the detector and front-end electron-
ics. The resistance Rin represents the input resistance of the amplifier. For the detector
pulse, the current iD provided to the amplifier can be written as
1
i R ( s ) = iD ( s ) (3.1)
1 + sCT Rin
In the current mode the signal from the detector is amplified without change in its
shape. This requires a very low input time constant τin= RinCT and this means the re-
quirements of small input resistance Rin of the amplifier.
In the voltage mode the voltage signal produced at the input of the amplifier is
much higher thanks to high value of the input resistance Rin and it can be written as
30 Architecture of front-end electronics
1
v R ( s ) = iD ( s ) Rin (3.2)
1 + sCT Rin
However, the above increase of pulse amplitude vR results in long time constant
τin at the amplifier input and the detector pulse shape is no longer preserved but it is
determined by this long time constant. The possible practical realization of current and
voltage modes are schematically shown in Fig. 3.4 (timp is the charge collection time at
detector electrode).
(a)
(b)
Fig. 3.4. The examples of realization of an input stage of front-end electronics: a) current configurations,
b) voltage configurations.
Architecture of front-end electronics 31
vout ⎛ v ⎞
iD ( s ) = − sCT + ⎜⎜ − out − vout ⎟⎟ sC F (3.3)
K0 ⎝ K0 ⎠
where vout is the voltage at CSA output. The above equation gives the CSA transfer
function
vout ( s ) 1 K0
=− (3.4)
iD ( s ) s ( K 0 + 1)C F + CT
vout ( s) 1
≈− (3.5)
iD ( s ) sCF
Using the eq. (3.4) the signal at CSA output can be written as
timp
K0
vout (t ) = −
( K 0 + 1)C F + CT ∫i
0
D (t )dt (3.6)
where timp is the charge collection time at the detector electrode. Assuming at the input
the delta-like current pulse iD(t) = Qinδ(t) one obtains at the CSA output the voltage step
which is proportional to total charge carried by the detector pulse
K0
vout (t ) = − Qin (3.7)
( K 0 + 1)CF + CT
For the ideal integrator its input impedance Zin is capacitive and the effective high
input capacitance is guaranteed by the high gain K0 of the amplifier (due to the Miller
effect [24])
1 1
Z in ( s ) = ≈ (3.10)
( K 0 + 1) sCF K 0 sCF
The large input capacitance of CSA and the requirements (3.9) guarantee that
most of the charge produced by the detector is transferred to the amplifier (see
eq. (2.47)).
Architecture of front-end electronics 33
v out ( s ) K0
= KV = − (3.11)
vin ( s) 1 + s ω0
where K0 is the DC gain and ω0 is the dominant pole of the amplifier. The gain band-
width product of the amplifier is GBW = K0ω0. With the above aspects in mind let us
consider the CSA scheme shown in Fig. 3.6 with additional unity gain buffer to ensure
small output impedance [5]
Fig. 3.6. Charge sensitive amplifier based on a core amplifier with frequency dependent voltage gain.
⎛ 1 ⎞
iD ( s ) = vin ( s) sCT + (vin ( s ) − vout ( s ) )⎜ + sCF ⎟ (3.12)
⎜R ⎟
⎝ f ⎠
34 Architecture of front-end electronics
vout ( s ) K0
=− (3.13)
iD ( s ) 1 + K0 ⎡ 1 ⎤ 2 CT + CF
+ s ⎢CT + (1 + K 0 )CF + ⎥+s
Rf ⎢⎣ ω0 R f ⎦⎥ ω0
Because K0 >> 1 and assuming that the requirement ( K 0 + 1)C F >> C D is ful-
filled the equation (3.13) can be rewritten as
vout ( s ) Rf
≈− (3.14)
iD ( s ) ⎡ 1 ⎤ 2 (CT + CF )R f
1 + s ⎢ R f CF + +s
⎣ GBW ⎥⎦ GBW
The above transfer function has two poles ω1 and ω2 which are usually real and
widely separated (ω1 << ω2). The denominator of this function can be written as
⎛ s ⎞⎛ s ⎞ ⎛1 1 ⎞ s2 s s2
⎜ ⎟⎜ ⎟ ⎜
D ( s ) = ⎜1 + ⎟⎜1 + ⎟ = 1 + s⎜ + ⎟ + ⎟ ≈ 1+ + (3.15)
⎝ ω1 ⎠⎝ ω2 ⎠ ⎝ ω1 ω2 ⎠ ω1ω2 ω1 ω1ω2
1
ω1 = (3.16)
1
R f CF +
GBW
Because the feedback time constant R f CF is much higher than 1/GBW the dominant
pole can be rewritten as
1
ω1 ≈ (3.17)
R f CF
Architecture of front-end electronics 35
CF
ω2 ≈ GBW (3.18)
CT + CF
There are two time constants connected with the above poles feedback time con-
stant τf = 1/ω1 and τ2 = 1/ω2 and the transfer function of the CSA can be rewritten as
vout ( s ) Rf 1 τ f ⎛⎜ τ f τ 2 ⎞⎟
≈− =− −
iD ( s ) (1 + sτ f )(1 + sτ 2 ) CF τ f − τ 2 ⎜⎝ 1 + sτ f 1 + sτ 2 ⎟⎠ (3.19)
In the time domain in response to the input current pulse iD(t) = Qinδ(t) one ob-
tains at the CSA output
Qin τ f
vout (t ) = −
CF τ f − τ 2
[ ]
exp(− t τ f ) − exp(− t τ 2 ) (3.20)
The exemplary time responses are shown in Fig. 3.7. The time constant τ f is re-
sponsible for slow signal decay and τ 2 determines the rise time at CSA output.
For very high feedback resistance Rf → ∞ (according to (3.17) the time constant
τf → ∞) one obtains
Qin
vout (t ) ≈ − [1 − exp(− t τ 2 )] (3.21)
CF
For a very fast core amplifier GBW → ∞ (according to (3.18) the time constant
τ2 → 0) and finite τf the equation (3.20) simplifies to
exp(− t τ f )
Qin
vout (t ) = − (3.22)
CF
36 Architecture of front-end electronics
(a) (b)
Fig. 3.7. Time response of the CSA for different time constants (horizontal scales in both figures are
different): a) τf is responsible for slow signal decay, (b) τ2 determines the rise time at CSA output.
The realistic CSA amplifier has also different input impedance than specified by
eq. (3.10). For low frequency ω << ω0 the input impedance is given as
1 ⎛⎜ Rf ⎞
⎟
Z in ( s ) ≈ (3.23)
K 0 ⎜⎝ 1 + sCF R f ⎟
⎠
For large Rf the above equation gives the same result as eq. (3.10) and the input
impedance appears capacitive with the effective input capacitance Cin ≈ K0CF.
For high frequency ω >> ω0 the gain drops linearly with frequency and it can be
expressed as
K 0ω0 GBW
KV ≈ − =− (3.24)
s s
Assuming large Rf and for high frequency ω >> ω0 the input impedance of CSA is
s 1 1
Z in ( s ) ≈ = (3.25)
GBW sC F C F GBW
Architecture of front-end electronics 37
This impedance appears as a resistance Rin = (CF GBW)-1. The situation at the in-
put of the amplifier becomes similar to situation shown in Fig. 3.3. The time constant at
the CSA input τin= RinCT equals
CT
τ in = (3.26)
C F GBW
In order to transfer quickly the charge generated by the detector to the charge sensitive
amplifier, the GBW of the core amplifier must be sufficiently large.
Cascode amplifier architecture is one of most commonly used solutions for CSA
core [57−69] because of its simplicity, high output resistance, reduction of unwanted
Miller multiplication of gate-drain capacitance Cgd of the input transistor and possible
operation at high frequency. A differential amplifier configuration is also used espe-
cially in case of required good power supply rejection ratio [44, 68−71], however it
gives worse noise performance than single ended stages. There are also some successful
designs using a CMOS inverter for CSA core (NMOS and PMOS transistors with both
gates connected at the input and their drains connected as an output). This solution pro-
vides low noise performance, however with limited open-loop gain [72,73].
(a) (b)
Fig. 3.8. Examples of core amplifiers: a) cascode stage, b) folded cascode stage.
38 Architecture of front-end electronics
The simple cascode is shown in Fig. 3.8(a). The main transistor M1 sometimes
operates at lower supply voltage to save power consumption. Its dimensions are sized
according to input detector capacitance to minimize the noise (see Chapter 3.3.2). The
primary function of M2 is to keep small signal resistance at the drain of M1 low. There-
fore, the cascode transistor M2 has sufficiently high transconductance and collects all
signal current from M1. In the simple cascode the same current flows through M3 and
M1 transistors, which makes it difficult to obtain simultaneously high transconductance
gm1 of input transistor M1 and high output resistance rds3 of transistor M3 at the same
time. An additional current source connected like transistor M4, can help in this case
[61, 74]. The M4 which sinks the significant part of current from transistor M1 and al-
lows obtaining high gm1/gds3 ratio and high gain in low frequency region. The output
source follower M5−M6 works as an output buffer.
For a connection of DC feedback loop and for an operation of the input transistor
with lower supply voltage, the folded cascode configuration (shown in Fig. 3.8(b)) is
more convenient. To obtain higher output resistance of the folded cascode stage a cur-
rent cascode source is often used, instead of a simple current mirror M4.
The input transistor of the cascode stage can be NMOS or PMOS and the choice
of the transistor type depends on:
− noise performance of both types of transistor in the selected technology,
− sensitivity of a given transistor type to a substrate noise,
− other aspects, for example, quality of NMOS or PMOS current sources in case of us-
ing enclosed gate transistor layout, etc.
Taking into account the folded cascode only i.e. the transistors M1−M4 (without
output source follower M5-M6 and neglecting the bulk effect of M2 gmb2 = 0) the low
frequency voltage gain can be written as
vX g m1 ( g m 2 + g ds 2 )
=− (3.27)
vin g ds 2 ( g ds1 + g ds 4 ) + g ds 3 ( g m 2 + g d 1 + g ds 2 + g ds 4 )
where gm1− gm4 are transconductances of transistors M1−M4 and gds1− gds4 are their out-
put conductances. Assuming that gm2 >> gds1, gds2, gds4 and gds3 ≈ gds2 , the above equa-
tion can be rewritten as
vout g
≈ − m1 (3.28)
vin g ds 3
The dominant pole of the folded cascode can be expressed as
1
ω0 = (3.29)
C X RX
Architecture of front-end electronics 39
where the total capacitance seen from the node X equals to CX. RX is given as
The simplified noise scheme of the folded cascode is shown in Fig. 3.9. The noise
calculated to the cascode input is given as
where gds1 =1/rds1 and gds4=1/rds4. In a good design the dominant part of the noise comes
from the main transistor M1, because gm1 is high. However, special attention should be
paid to current source M4 and its reference (not shown in Fig. 3.8(b)). The problem is
that drain currents of transistors M1 and M4 (for cascode design used in CSA applica-
tion) are nearly equal and to reduce the last term in eq. (3.31), a designer should keep
the transconducance gm4 of current source M4 as small as possible (i.e. using transistor
M4 with relatively long channel L4).
After integration of a current pulse in the CSA, the feedback capacitor CF should
be discharged by the reset block (Fig. 3.5) during a short period of time to prevent satu-
ration of the amplifier [75]. There are two basic techniques implemented for discharging
the feedback capacitance: switch reset and continuous discharge.
(a)
(b) (c)
(e)
(d) (f)
Fig. 3.10. Most frequently used reset systems in CSA feedback - see description in text below.
a saturation limit of the CSA, however such a solution results in additional deadtime of
the whole system. The disadvantages of this solution are sampled noise and a possible
problem with charge injection from switch control voltage.
Continuous discharging can be completed either by a resistor parallel to the ca-
pacitor or by a controlled current source (see Fig. 3.10(b)−(f)). In either case, the dis-
charging component contributes to the parallel noise at the CSA input. In order to limit
this noise source, one should use a large value resistor or a low discharging current.
Using a physical resistor (Fig. 3.10(b)) seems to be a simple solution, however the resis-
tance should be in the range from hundreds of kΩ to a few GΩ (resistor value depends
on noise requirement of an experiment and a peaking time of the shaper). It is difficult
to obtain large value resistance in a monolithic process with a low parasitic capacitance.
Instead of a simple resistor, many designers use a feedback MOS transistor working in
triode or saturation region [57, 78, 79]. This is a compact solution with the possibility to
control feedback resistance, however nonlinear effects must be taken into account.
More complex feedback solutions are shown in Fig. 3.10(d)−(f):
− the configuration shown in Fig. 3.10(d) uses a current conveyor feedback [80, 81]. In
response to a signal at CSA output a reference current is produced in a low value resistor
R. This reference current is significantly reduced in the network based on current mirror
and discharges the feedback capacitor,
− the technique shown in Fig. 3.10(e) is also based on a current mirror and uses a cur-
rent source [82, 83]. With no activity, the feedback transistor M2 stays in the linear re-
gion. When the signal appears, the M2 enters the saturation region and the copy of bias
current Ibias discharges the capacitor,
− discharge system shown in Fig. 3.10(f) uses a differential stage. The baseline recov-
ery after signal integration is achieved by the low frequency feedback loop that sets the
output voltage of the CSA to the reference voltage VREF [84]. Proper circuit compensa-
tion is an important issue in this case. The effective feedback resistance is equal to
Rf = 2/gm1. Other solutions using a differential stage are also possible [58, 85], but with-
out the mentioned above low frequency feedback.
There are also two aspects which should be taken into account while choosing one
of the above options as reset system in given applications, namely:
− for a DC coupled detector leakage current should be automatically accommodated by
the CSA - the good candidates are e.g. the solutions in Fig. 3.10(c) (especially with the
feedback transistor working in saturation region) and in Fig. 3.10(f),
− the long decay time constant of the preamplifier output signal, which produces the
limitations of the pulse rate due to the pulse pile-ups. The implementation of the Pole-
Zero Cancellation (PZC) circuit can significantly help in this case (see Chapter 3.4.2)
and adding of a PZC circuit is relatively easy in the cases shown in Fig. 3.10(b) and
3.10(c).
42 Architecture of front-end electronics
Fig. 3.11. Small capacitor Ctest at CSA input can work as charge injector.
Ctest
Qtest = ΔVtest ≈ ΔVtest Ctest (3.32)
Ctest
1+
Cdet + (KV + 1)CF
Since in practice Ctest << Cdet << (KV+1)CF the injected charge Qtest is totally fed
into the CSA. The voltage step ΔVtest can be applied externally or internally. The possi-
ble options of internal implementation of the chopper circuits are shown in
Fig. 3.12(a)−(b).
In the first option a step signal is produced on resistor R in response to STROBE
signal. The step value is controlled by changing the value of a current source in the
differential amplifier. In the second option the STROBE_P and STROBE_N signals
switch the test capacitor Ctest between two lines CAL_P and CAL_N with different DC
potentials. In both options it is easy to generate square wave signals, which means that
for the rising edge of the test pulse the charge is injected to the CSA and for the falling
edge of the test pulse it is extracted. In most cases it is enough to test a basic
functionality of the front-end electronics. However, in other cases it is insufficient, for
instance, for checking high rate performance of the front-end electronics. The
Architecture of front-end electronics 43
application of voltage step function seems more appropriate in this case. Generating
a voltage step function inside the integrated circuit is difficult, but it can be applied
using an external arbitrary waveform generator. Another option is to feed a known small
current Iinj to the preamplifier input during a known time interval Tinj. The injection
charge is equal to Qtest = IinjTinj.
(a)
(b)
Fig. 3.12. Possible realization of the test injection circuits: a) generation of voltage steps on resistor R [86],
b) switching between two lines with different DC potentials [87].
3.3. SHAPER
The shaper stage after the CSA is added to perform the following tasks:
− to filter the CSA output signal in order to improve the signal to noise ratio in the sys-
tem,
− to add more gain in the signal processing chain,
− to shorten the pulse duration and to reduce the possibility of pile-up pulses.
The choice of the filter type, order and its time constants strongly depends on
specified energy resolution of the readout system and its high rate operation require-
ments [2, 32, 88−92]. There is a wide range of shapers built in hybrid technologies and
which use components of the shelf. However, in case of a multichannel ICs there are
additional very strong requirements on a low power budget and a small area occupied by
the single channel, and a practical realization of all filter types in the IC technology is
limited. For the purpose of this book three types of filters often used in the multichannel
ICs are described:
44 Architecture of front-end electronics
(a) (b)
Fig. 3.13. Timing definition in pulse shaping: a) unipolar pulse, b) bipolar pulse [94, 95].
For the CR-(RC)n filter with the same integrator and differentiator time constants
τi = τd = τ the transfer function is given by [92, 96]
n
⎛ sτ ⎞⎛ 1 ⎞
H ( s) = ⎜ ⎟⎜ ⎟ (3.33)
⎝ 1 + sτ ⎠⎝ 1 + sτ ⎠
Assuming an ideal unity voltage step at the shaper input and taking the transfer
function of the filter given by (3.33), one obtains the shaper output signal in the time
domain as
n
1⎛t⎞ ⎛ t⎞
vout (t ) = ⎜ ⎟ exp⎜ − ⎟ (3.34)
n! ⎝ τ ⎠ ⎝ τ⎠
nn
vmax = (3.35)
n!e n
with the peaking time tp = nτ . The family of shaper output pulses for a given time con-
stant τ is shown in Fig. 3.15(a). Increasing the filter order results in decreasing the sig-
nal amplitude, but makes the pulse more symmetrical. Higher filter order is more suit-
able for high rate application, however to obtain the same peaking time for the higher
shaper order, one should shorten the time constant of the filters. The family of the pulses
with the normalized amplitude and the time scale, but different orders, is shown in
Fig. 3.15(b). For higher order filters the shaper output pulse returns to the baseline
faster, and this directly influences the high rate operation and reduces the probability of
pile-up pulses.
46 Architecture of front-end electronics
(a)
(b)
Fig. 3.15. Examples of shaper output for different CR-(RC) filter orders with: a) the same time constant τ ,
n
b) normalized output .
The pulse width t0.01 (calculated according the criteria 1% to 1% of the maximum
pulse amplitude) for different filter order as the ratio to peaking time is summarized in
Table 3.1.
Table 3.1. Pulse width for different filter CR-(RC)n.
Filter order n 1 2 3 5 7
t0.01/tp 7.66 5.04 4.17 3.46 3.14
Higher filter order requires more power and occupies more silicon area. For that
reason many multichannel readout ICs use simple CR-(RC)2 shaper, while the readout
electronics for pixel detector has a very simple shaper stage or has nearly no separate
shaper stage at all.
Bipolar semi-Gaussian pulse shaper. The semi-Gaussian pulse shaper of type
(CR) -(RC)n, consists of two CR differentiators and n integrators (see Fig. 3.16). For the
2
same integrator and differentiator time constants τi = τd = τ the transfer function is given
by [5, 92]
2 n
⎛ sτ ⎞ ⎛ 1 ⎞
H ( s) = ⎜ ⎟ ⎜ ⎟ (3.36)
⎝ 1 + sτ ⎠ ⎝ 1 + s τ ⎠
Assuming an ideal unity voltage step at the shaper input and taking into account
the above transfer function of the filter, one obtains the shaper output signal in the time
domain as
n
(n + 1)τ − t ⎛ t ⎞ ⎛ t⎞
vout (t ) = ⎜ ⎟ exp⎜ − ⎟ (3.37)
(n + 1)!τ ⎝ τ ⎠ ⎝ τ⎠
This pulse has a negative undershoot and crosses zero at tox = (n+1)τ, and the maximum
and minimum pulse values are obtained at time coordinates
(
tmax = n + 1 − n + 1 τ ) (3.38)
tmin = (n + 1 + n + 1) τ (3.39)
The family of bipolar pulses at shaper output for a given time constant τ is shown
in Fig. 3.17(a). The family of the pulses with the normalized amplitude and the time
scale, but different filter orders, is shown in Fig. 3.17(b).
(a) (b)
Fig. 3.17. Examples of shaper output for different (CR)2-(RC)n filter orders with: a) the same time constant
τ , b) normalized output.
The pulse width t0.01 for different filter order as the ratio to tp1 is summarized in
Table 3.2. The ratio of amplitude overshoot to pulse amplitude is also given in the table
below.
Table 3.2. Pulse width for different (CR)2-(RC)n filter.
Filter order n 1 2 3 5 7
t0.01/tp1 16.6 9.85 7.67 5.96 5.22
|vmin/vmax| 0.344 0.436 0.495 0.569 0.616
48 Architecture of front-end electronics
The bipolar shaping results in a longer pulse width than unipolar shaping. As
shown later, the bipolar shaper offers also a worse signal to noise ratio. However, the
bipolar shaping has two advantages:
− zero crossing time txo does not depend on pulse amplitude and this feature of bipolar
shaper is used in timing measurements,
− it eliminates the problem of a baseline shift (for high rate of input pulses) in the sys-
tem with AC coupling [97].
Nearly true Gaussian filter obtained using Ohkawa synthesis method [93]. In
order to minimize the effects of pile-up pulses, the minimum pulse width t0.01 is prefer-
able, as for example in the case of Gaussian shaped pulses. Using CR-(RC)n configura-
tion to obtain the true Gaussian pulse shape, a CR differentiator and the infinitely large
number of RC integrators are necessary. Some systematic studies can be found in litera-
ture (in time or frequency domain) to obtain the optimal pulse shaping network, e.g.
[98, 99]. However, some results of these studies are difficult to implement in integrated
circuits (i.e. filters contain inductors).
An interesting approach for the Gaussian filter which can be applicable in the IC
can be found in [93]. A nearly true Gaussian filter with a limited number of stages can
be realized using active filters with resistors and capacitors. The filters have complex
poles which are obtained on the basis of the frequency domain analysis. Let us analyze
shortly the idea of this kind of filter according to [93].
The pulse after charge sensitive amplifier is fed to the shaper which is assumed to
be composed of an ideal differentiator, followed by the network whose pulse response is
the Gaussian waveform
⎛ t2 ⎞
vout (t ) = a0 exp⎜⎜ − ⎟
2 ⎟
(3.40)
⎝ 2σ ⎠
where a0 is a constant and σ is the rms deviation of the normal distribution. The above
Gaussian waveform has a peak a0 at t = 0. In the frequency domain the above equation
can be written as
⎛ σ 2ω 2 ⎞
F (ω ) = a0 2π σ exp⎜⎜ − ⎟ (3.41)
⎝ 2 ⎟⎠
Let us assume the following transfer function of the Gaussian filter
H0
H (s) = (3.42)
Q( s)
of the complex plane). As it can be easily shown [93] there is the following relation
between the transfer function and amplitude characteristics
H ( jω ) H (− jω ) = [F (ω )]
2
(3.43)
Taking into account eq. (3.41) and (3.42), the above relation can be rewritten as
2
⎛ H0 ⎞
⎟⎟ exp(− σ 2 s 2 )
1
Q( s )Q(− s ) = ⎜⎜ (3.44)
2π ⎝ a0σ ⎠
By proper normalization the above equation can be transformed into
2
Q( p )Q(− p ) = e − p (3.45)
and p = σs. The above exponential function can be approximated by the finite terms of
Taylor series as
p4 p6 p 2 nG
Q( p)Q(− p ) = 1 − p 2 + − + ... + (−1) nG (3.46)
2! 3! nG !
For nG = 1 the above equation is
Q( p )Q(− p) = 1 − p 2 (3.47)
Q( p) = 1 + p (3.48)
and the network function has the real pole at -1 on the p plane.
For nG = 2 the equation (3.46) gives
p4
Q( p )Q(− p) = 1 − p + 2
(3.49)
2!
1 ⎛
Q( p) = ⎜ 2 + (2 + 2 2 ) p ⎞⎟ + p 2 (3.50)
2⎝ ⎠
50 Architecture of front-end electronics
p1 =
(2 + 2 2 )
2
−1±[ ( 2 −1)] (3.51)
For higher order nG = 3−7 the pole constellation can be obtained by numerical
analysis. The pole constellation plotted on the complex plane p is shown in Fig. 3.18.
Poles are also listed in Table 3.3, where the real pole is p0 =− A0 , Ai and the complex
pole pairs are pi =− Ai ±Wi .
Table 3.3. Pole locations of the Gaussian filters (reprinted from [93] © 1997, with permission from
Elsevier, http://www.sciencedirect.com/science/journal/ ).
nG = 3 nG = 4 nG = 5 nG = 6 nG = 7
A0 1.2633573 1.4766878 1.6610245
A1 1.1490948 1.3553576 1.4166647 1.5601279 1.6229725
W1 0.7864188 0.3277948 0.5978596 0.2686793 0.5007975
A2 1.1810803 1.2036832 1.4613750 1.4949993
W2 1.0603749 1.2994843 0.8329565 1.0454546
A3 1.2207388 1.2344141
W3 1.5145343 1.7113028
With the pole locations, the transfer function of the H(s) filter for the odd number
nG of poles can be expressed as
Architecture of front-end electronics 51
A0 ∏ (Ai2 + Wi 2 )
k
H (s) = i =1
(3.52)
(σs + A0 )∏ [(σs + Ai ) ]
k
2
+ Wi 2
i =1
∏ (A + Wi 2 )
k
2
i
H (s) = i =1
(3.53)
∏ [(σs + A ) ]
k
2
i + Wi 2
i =1
The constants in the numerators of equations (3.52) and (3.53) are added to normalize
the H(s) value to 1 when s = 0.
Ohkawa et al. [93] use of an equal-area time constant to CR-RC filter in order to
compare different waveforms. The area of the pulse after CR-RC filter with peak height
1 is given by
∞
⎛ t ⎞ ⎛ t ⎞
SCR = e ∫ ⎜⎜ ⎟⎟ exp⎜⎜ − ⎟⎟dt = eC x Rx (3.54)
CR
0⎝ x x ⎠ ⎝ C x Rx ⎠
∞
⎛ t2 ⎞
SG = ∫ ⎜⎜⎝ − 2σ 2 ⎟⎟⎠dt = σ 2π
−∞
exp (3.55)
e
σ= C x Rx = σ 0τ x (3.56)
2π
where τx = CxRx is time constant of the CR-RC filter and σ0 = e/√(2π) = 1.0844 is so
called an equal-area time constant.
52 Architecture of front-end electronics
The pulse response approaches the Gaussian waveform when the filter order in-
creases (see Fig. 3.19), so for the pulse shaping the preferable filter order is nG ≥ 5.
Fig. 3.19. Pulse responses for a different filter order as a function of time normalized with τ0 (reprinted
from [93] © 1997, with permission from Elsevier, http://www.sciencedirect.com/science/journal/ ).
Let us follow now the procedure of Gaussian active filter design. The common
practice in nuclear amplifier to obtain a differentiator is to use a simple CR circuit with
zero at origin and a real pole which determines the time constant. For this type of
a shaper, the pole constellations with a real pole such as nG = 5 or nG = 7 are very con-
venient (see Table 3.3). The relations between the time constant C0R0 of the differenti-
ator and the real pole from Table 3.3 are as follows
C0 R0 = σ 0τ x A0 (3.57)
The next step is to build stages with complex poles. There are different types of
filters with complex pole pairs. Some examples together with the schemes (see
Fig. 3.20, Fig. 3.21 and Fig. 3.22.) and equations for transfer functions and poles are
presented below.
Architecture of front-end electronics 53
The transfer function and pole positions for an active filter with multiple negative
feedback are
R1 1
H (s) = − (3.58)
R3 1 + sC1R1 + s 2C1R1C2 R2
⎛ 1 ⎞⎡ ⎛ 4C R ⎞⎤
s1 = ⎜⎜ ⎟⎟ ⎢− 1 ± j ⎜⎜ 2 2 − 1⎟⎟ ⎥ (3.59)
⎝ 2C2 R2 ⎠ ⎢⎣ ⎝ C1R1 ⎠ ⎥⎦
The transfer function and pole positions for an active filter with bridged-T feed-
back are
54 Architecture of front-end electronics
R1 R2
sC2 +1
R1 + R2 R1 + R2
H (s) = − (3.60)
R3 1 + sC1 (R1 + R2 ) + s 2C1 R1C2 R2
⎛ R + R2 ⎞ ⎡ ⎛ 4C2 R1 R2 ⎞⎤
s1 = ⎜⎜ 1 ⎟⎟ ⎢− 1 ± j ⎜⎜ − 1 ⎟⎥
⎟⎥ (3.61)
⎝ C1 (R1 + R2 )
2
⎝ 2C2 R1 R2 ⎠ ⎢⎣ ⎠⎦
The transfer function and pole positions for an active filter with positive feedback
are
1
H (s) = − (3.62)
1 + sC2 (R1 + R2 ) + s 2C1 R1C2 R2
⎛ R + R2 ⎞ ⎡ ⎛ 4C1R1R2 ⎞⎤
s1 = ⎜⎜ 1 ⎟⎟ ⎢− 1 ± j ⎜⎜ − 1⎟⎥
⎟⎥ (3.63)
⎝ C2 (R1 + R2 )
2
⎝ 2C1R1R2 ⎠ ⎢⎣ ⎠⎦
Architecture of front-end electronics 55
Let us consider an active filter with multiple negative feedbacks (Fig. 3.20) to
build the Gaussian filter. The real and imaginary parts of the complex poles are in the
following relations to the coefficient in Table 3.3
σ 0τ x
Ai = (3.64)
2C2 R2
⎛ στ ⎞ ⎛ 4C2 R2 ⎞
Wi = ⎜⎜ 0 x ⎟⎟ ⎜⎜ − 1⎟⎟ (3.65)
⎝ 2C2 R2 ⎠ ⎝ C1 R1 ⎠
The example of the Gaussian filter for nG = 7 realized according to the procedure
described is shown in Fig. 3.23 and the exemplary values of resistors and capacitors are
shown in Table 3.4. For τx = 100 ns the peaking time of this Gaussian filter is about
tp = 330 ns.
Table 3.5. Exemplary values of calculated resistors and capacitors of the Gaussian filter for nG = 7.
Pole n=7 C0R0 C2R2 C1R1 Examples of calculated Ci and
from eq. from eq. from eq. Ri
(3.57) (3.66) (3.67) for τx = 100 ns
A0 1.6610245 0.652850τx − − C0 = 13.057 pF , R0 = 5 kΩ
A1 1.6229725 − 0.334078τx − C2C = 8.148 pF , R2C = 4.1 kΩ
W1 0.5007975 − − 1.220138τx C1C = 8.134 pF , R1C = 15 kΩ
A2 1.4949993 − 0.362676τx − C2B = 6.476 pF , R2B = 5.6 kΩ
W2 1.0454546 − − 0.974266τx C1B = 6.495 pF , R1B = 15 kΩ
A3 1.2344141 − 0.439237τx − C2A = 3.993 pF , R2A = 11 kΩ
W3 1.7113028 − − 0.601302τx C1A = 4.009 pF , R1A = 15 kΩ
R1C/R3C, R1B/R3B, R1A/R3A set the filter gain, for example, these resistors can be R3C = R3B = R3A = 4.7 kΩ.
The pulse width t0.01 for the Gaussian filter is significantly smaller than in case of
CR-(RC)n filter, so they are better for applications working with high rate of input pulses
(see Table 3.6).
Table 3.6. Pulse width for the Gaussian filter.
Filter order nG 5 7
t0.01/tp 2.84 2.55
The main noise contribution to the detector readout system comes from the detec-
tor itself and from components of the charge sensitive amplifier. The simplified noise
model of a detector and the CSA is shown in Fig. 3.24. The noise performance of this
system can be analyzed using the equivalent input current and voltage noise sources.
A current noise is associated with the detector leakage current Idet, the detector
bias resistance Rbias (in case of an AC coupling detector only) and the effective feedback
resistance Rf in the CSA, so the power spectral density is given by
din2
=a (3.69)
df
In the well-designed CSA the voltage noise is dominated by an input transistor.
For CMOS technology, the voltage noise of the input MOS transistor can be expressed
as [96]
dvn2 γn Kf 1 1
= 4kT + (3.70)
df g m1 Cox2 WL f
dvn2 AF
=b+ (3.71)
df f
where constants b and AF are responsible for the thermal and flicker noise respectively.
If the current and voltage noise sources are uncorrelated, they can be calculated to
the CSA output independently. The parallel current noise is multiplied by a square of
module of CSA transfer function (see eq. (3.5)) and this gives
58 Architecture of front-end electronics
2
dvCSAout _ parrallel a a
= = (3.72)
df ω CF (2πfCF )2
2 2
To transfer the amplifier voltage noise to the CSA output let us look at first at the
relationship between CSA input voltage and CSA output voltage (for K0 >> 1), which
are given by simple impedance divider
vin 1 sCT CF
= = (3.73)
vout 1 sCT + 1 sCF CT + CF
The gate itself of input transistor works as virtual ground and therefore the series
voltage noise gives at the CSA output
2 2
dvCSAout _ series ⎛ A ⎞⎛ C + CF ⎞
= ⎜⎜ b + F ⎟⎟⎜⎜ T ⎟⎟ (3.74)
df ⎝ f ⎠⎝ CF ⎠
The total noise at the CSA output is the sum of eq. (3.72) and (3.73)
2 2
dvCSAout a ⎛ A ⎞⎛ C + CF ⎞
= + ⎜⎜ b + F ⎟⎟⎜⎜ T ⎟⎟ (3.75)
df (2πfCF ) ⎝ f ⎠⎝ CF
2
⎠
The noise at CSA output is filtered by the following shaper stage. For the gain of
the CSA which is high enough and careful design of the shaper stage, the noise of the
shaper stage can be negligible.
Let us consider first a simple RC-CR filter (with the identical time constants
τi = τd = τ) where the peaking time is equal to filter time constant tp = τ. The module of
its transfer function is given by
2πft p
H (2πf ) = (3.76)
1 + (2πft p )
2
Architecture of front-end electronics 59
The rms value of noise at the shaper output with the above assumption can be cal-
culated as
2
∞⎡ a ⎛ A ⎞⎛ C + CF ⎞
2
⎤ ⎛ 2πft p ⎞
2
dvSHout = ∫⎢ + ⎜⎜ b + F ⎟⎟⎜⎜ T ⎟⎟ ⎥×⎜ ⎟ df (3.77)
⎣ (2πfCF ) ⎝
0⎢
2
f ⎠⎝ CF ⎠ ⎜ ( )2 ⎟
⎥⎦ ⎝ 1 + 2πft p ⎠
The solution of the above integral is given by the following equations [37]
1 ⎡ at p 2⎛ b AF ⎞⎟⎤
2
dvSHout = ⎢ + (C + C ) ⎜
⎜ 8t
+ ⎥ (3.78)
2 ⎟⎠⎦⎥
F T
CF2 ⎣⎢ 8 ⎝ p
In the front-end electronics the noise from the shaper output is recalculated to the
CSA input. Because the input signal is a charge pulse, the noise calculated to the CSA
input is expressed as Equivalent Noise Charge (ENC). The ENC is equal to the detector
signal that yields the signal-to-noise ratio of 1. The ENC is expressed in electrons or in
Coulombs, or sometimes in equivalent energy (eV) deposited in the detector. For this
calculation the gain of the chain CSA−shaper is necessary. At the output of the CSA the
signal step is equal to Qin/CF and at the shaper output the peak amplitude is proportional
to the charge Qin generated in the detector. In practice, the ENC of the detector system is
calculated as the ratio of the total integrated rms noise at the shaper output to the peak
amplitude (at the shaper output) for an input charge Qin of 1 electron (1.602×10-19 C)
2
dvSHout
ENC =
vSH max (Qin = 1 electron)
(3.79)
In the case of simple RC-CR filter analyzed above, the output signal in the time
domain (according the eq. (3.34) − (3.35)) is given as
t ⎛ t ⎞
vout = vSH max × exp⎜ − ⎟ (3.80)
tp ⎜ t ⎟
⎝ p ⎠
with peak amplitude
Using the equations (3.78), (3.79) and (3.81) results, the ENC is given as
⎡ at 2⎛ b A ⎞⎤
ENC 2 = e 2 ⎢ p + (CF + CT ) ⎜ + F ⎟⎥ (3.82)
⎜ 8t 2 ⎟⎠⎥⎦
⎢⎣ 8 ⎝ p
To analyze the obtained results let us split the above equations to three compo-
nents: ENCi due to current parallel noise component (eq. (3.68) and (3.69)), ENCw due
to voltage thermal noise component (the first term in eq. (3.70) and (3.71)) and ENCf
due to voltage flicker noise component (the second term in eq. (3.70) and (3.71))
ENC 2
= 0.924
(CF + CT )
2
×b (3.85)
w
tp
ENC 2f = 3.695(CF + CT ) × AF
2
(3.86)
Fig. 3.25. Contributions of different noise components to the total ENC vs. peaking time.
ENCw2 = Fv
(CF + CT )2 × b (3.88)
tp
where constants Fi , Fv , Fvf for semi-Gaussian pulse shapers analyzed earlier are sum-
marized in Table 3.7.
Table. 3.7. Noise coefficients for a different filter type.
Filter type Fi Fv Fvf
CR-RC 0.92 0.92 3.70
CR-(RC)2 0.64 0.85 3.41
CR-(RC)3 0.52 0.93 3.32
CR-(RC)5 0.40 1.11 3.25
CR-(RC)7 0.34 1.27 3.22
(CR)2-RC 1.00 1.03 4.70
(CR)2-(RC)2 0.72 1.16 4.89
(CR)2-(RC)3 0.60 1.44 5.12
(CR)2-(RC)5 0.48 2.00 5.49
(CR)2-(RC)7 0.41 2.52 5.75
of noise at CSA output (and also ENC at CSA input). This means that there is the opti-
mal transistor dimensions for which the PSD of the noise at CSA output and the ENC is
minimal. Let us analyze this problem for the cases when the input transistor operates in
strong, moderate or weak inversion regions.
Table 3.8. Summary of voltage noise performance of system CSA-shaper
Noise Thermal noise 1/f noise
PSD of MOS transistor γn AF K f 1 1
b = 4kT = 2
dvn2 AF g m1 f Cox WL f
=b+
df f
PSD at CSA output 2 2
⎛ C + C1 + CF ⎞ AF ⎛ Cdet + C1 + CF ⎞
b⎜⎜ det ⎟⎟ ⎜⎜ ⎟⎟
⎝ CF ⎠ f ⎝ CF ⎠
ENC2 for RC-CR filter
0.924
(Cdet + C1 + CF )2 × b 3.695(Cdet + C1 + C F ) × AF
2
tp
Term to be optimized
for transistor dimensions (Cdet + C1 + CF )2 (Cdet + C1 + CF )2
g m1 W1L1
W1
g m1 = 2 μ Cox I DS 1 (3.90)
L1
− the minimum transistor gate length L1 = Lmin is desired. However, for some technolo-
gies and bias conditions, the choice of minimum transistor length results in noise in-
crease due to short channel effect (see discussion in the Chapter 4.1.3). In such cases the
noise measurements of the MOS transistors in a given technology (with the dimensions
and bias conditions similar to the application in the CSA) are necessary [104−106] and
the L1 has to be chosen as small as possible to make the excessive noise connected with
the short channel effects negligible.
Calculating from equation (3.91) the transistor width is
C1
W1 = (3.92)
3 C ox L1 + 2C ov
2
I DS 1 C1
g m1 = 2 μ Cox ×2 (3.93)
Lmin 3 Cox Lmin + 2Cov
ENC 2
~
(Cdet + C1 + CF )
2
(3.94)
w
C1
The minimum value of the ENCw is calculated by solving the equation of the derivative
of (3.94) with respect to C1. It results in
Cdet + C F
C1 = (3.95)
3
The optimal WSIopt in strong inversion from (3.92) and (3.94) is given as
Cdet + CF
WSIopt = (3.96)
2Cox Lmin + 6Cov
Architecture of front-end electronics 65
Flicker noise. Calculating the flicker noise factor AF as a function of detector ca-
pacitance and assuming L1 = Lmin , one obtains
Kf 2
Cox L1 + 2Cov
AF = 2
× 3
(3.97)
C Lmin
ox C1
ENC 2f ~
(Cdet + C1 + CF )2 (3.98)
C1
The minimum value of the ENCf is calculated by solving the equation of the de-
rivative of (3.98) with respect to C1. It results in
C1 = Cdet + CF (3.99)
The equation for the optimum input transistor capacitance is different for thermal
and flicker noise. For the CSA analyzed above, with the input MOS transistor operating
in strong inversion, the optimum value of input transistor capacitance is in the range of
(Cdet+CF)/3 to (Cdet+CF) - see Fig. 3.26. Because for short peaking time the voltage
thermal noise dominates over the flicker noise, the optimum transistor capacitance is
closer to eq. (3.95), while for the longer peaking time the formula (3.99) is more appro-
priate [97].
Fig. 3.26. ENC as a function of C1/(Cdet+CF) ratio. Simulation performed for NMOS transistor with
Lmin = 0.35 μm and IDS1= 1 mA. Other parameters: Cdet = 2 pF, CF = 0.1 pF, CR-(RC)2 shaper with
tp = 300 ns.
Another aspect is the choice between NMOS or PMOS transistor as an input de-
vice. Usually, a PMOS transistor with smaller Kf decreases flicker noise, however its
transconductance in strong inversion region is smaller than an NMOS transistor (with
66 Architecture of front-end electronics
the same W/L ) and an NMOS transistor will have smaller thermal noise. So, the choice
between input transistor types is not so obvious, and depends on: technology, peaking
time, current density in the input transistor and methodology of protection of an input
device from substrate noise (see Chapter 4.2) [104−109].
Fv γ K
ENC ≈ Cin × 4kT n + Fvf × 2 f (3.100)
tp gm CoxWL
Fv γ C 2 WL
tp < × 4kT n × ox (3.101)
Fvf gm Kf
the contribution of 1/f noise is always much smaller than the channel thermal noise
which dominates [109]. The formula (3.96) was derived with the assumption that the
input transistor works in strong inversion and then the appropriate equations for gm1 and
C1 were used (eq. (3.90) and (3.91)). However, in a modern multichannel CSA it is fre-
quent that the input transistor operates in moderate or weak inversion regions, where the
above mentioned formulae are not valid.
An adequate variable used to determine the actual inversion level of a MOS tran-
sistor operating in saturation is its normalized forward current if [110]
I DS
if = (3.102)
2ns μ Cox (W / L)ϕT2
where IDS is the drain current, ns is the subthreshold slope factor (usually in the range
1.1 to 1.3), μ is the carrier mobility, ϕT is the thermal voltage.
Architecture of front-end electronics 67
Fig. 3.27. Limits of a moderate inversion region calculated from eq. (3.102) at if = 0.1 and at if =10. The
calculations were performed for three CMOS technology generations 0.13 μm, 0.35 μm and 0.8 μm,
assuming in each case the minimum transistor length available [109] © 2007, IEEE.
It is commonly assumed that weak inversion takes place for if < 0.1, while strong
inversion region is for if > 10 [111]. Therefore, an MOS transistor has a two-decade
current transition region called moderate inversion region bounded by if = 0.1 and
if = 10. In a modern multichannel CSA the input transistor often operates in this region
(see Fig. 3.27). The if current is relatively low due to the following factors:
− high density front-end systems are designed for a low-power operation, so the IDS
current is strongly limited,
− thermal noise optimization of the CSA input transistor often leads to a high value of
gm and large W/L ratio,
− in modern submicron technologies, smaller channel length L and higher capacitance
Cox are available.
Calculation of optimum transistor width using an EKV model. In order to ac-
count for the changes of gm, Cg and γn when departing from a strong inversion region
the simplified analytical EKV model is used [109, 110, 113]. The EKV model with its
relatively simple formulae is found to be quite successful in a moderate inversion region
[111]. In its simplified version the basic parameters gm, Cgs, Cgb, γn of the MOS transis-
tor operating in saturation, may be calculated as shown below. Transconductance is
given by
I DS
gm = f (i f ) (3.103)
nsϕT
where
68 Architecture of front-end electronics
1
f (i f ) = (3.104)
i f + 0 .5 i f + 1
Total gate capacitance is the sum of gate-source capacitance Cgs, gate-bulk ca-
pacitance Cgb and overlap capacitance
C g = C gs + C gb + 2CovW (3.105)
where
−1
⎛3 1 ⎞⎟
C gs = CoxW L ⎜ + (3.106)
⎜ 2 i f (i ) ⎟
⎝ f f ⎠
ns − 1 ⎛⎜ i f f (i f ) ⎞
⎟
C gb = CoxW L 1− (3.107)
n s ⎜⎝ 1 + 1.5i f f (i f ) ⎟⎠
1 1 if
γn = + (3.108)
2 6 if +1
In order to calculate these MOS parameters one needs to have two additional
technological parameters Cov and subthreshold slope factor ns. These parameters can be
extracted from numerical simulations.
Having the analytical expression for gm, C1 and γn , the function of the ENC as the
transistor width can be easily plotted as shown in Fig. 3.28. These plots can also be eas-
ily obtained with numerical simulations (with the use of the HSPICE or SPECTRE).
From the results in Fig. 3.28 the optimum transistor Wopt width can found.
Architecture of front-end electronics 69
Fig. 3.28. ENC vs. PMOS transistor width for detector capacitance Cdet =2 pF, feedback capacitance CF
= 100 fF and drain current IDS = 500 μA. The points denote HSPICE simulations and the lines simplified
EKV model calculations [109] © 2007, IEEE.
Fig. 3.29. Optimum PMOS transistor width for 0.35 μm process vs. drain current for Cdet = 2 pF and
CF = 100 fF [109] © 2007, IEEE.
From the simulations for a given detector capacitance and different drain currents
(see Fig. 3.29) it can be noticed that:
− optimum transistor width depends on the drain current. It is not predicted by the
strong inversion formula (3.96),
− wide flat ENC range around Wopt exists (marked as error bars - changing the channel
width inside these limits results in the noise increase below 3%). This means that there
is a large margin for the choice of the input transistor width around Wopt, in which the
ENC is practically not affected. It should be pointed out that a similar, very broad mini-
mum for a product called "timing noise factor" (input referred, noise voltage spectral
density multiplied by the total input capacitance) is also reported in [112].
Universal formula for Wopt. As shown in [109], it possible to find a new analyti-
cal formula for Wopt, which with the results of eq. (3.94) allows to calculate the optimal
transistor width also for moderate and weak inversion operation of an MOS transistor.
This formula is as follows
70 Architecture of front-end electronics
1
Wopt = WSIopt m
(3.109)
⎛W ⎞
1 + A × ⎜⎜ SIopt ⎟⎟
⎝ inoW ⎠
where
I DS
inoW = (3.110)
2nμ Cox (1 / L)ϕT2
and m = 0.61 and A = 0.25 are the fitting parameters (obtained by checking different
technologies). This formula is valid for input transistor working in weak, moderate and
strong inversion. The comparisons of the results obtained from above formula with nu-
merical simulations (HSPICE - transistor model BSIM3v3) and analytical calculation
using an EKV model are shown in Fig. 3.30.
(a) (b)
Fig. 3.30. Optimum width of PMOS transistors for two CMOS technologies (minimum transistor length is
used): a) 0.35 μm, b) 0.13 μm (dotted lines show the limits of moderate inversion region) [109] © 2007,
IEEE.
During the designing stage of low noise readout electronics for high-rate opera-
tion, negative effects of pulse pile-up in the signal processing chain should be taken into
account. There are two types of pile-ups
Architecture of front-end electronics 71
The pile-up on pulse tail produces the "positive" base line shift (with the same
sign as a pulse amplitude). This type of pulse pile-ups can take place both at the CSA
and shaper outputs. In case of shaper output the pulse pile-up is reduced by using the
filter of a proper type and order to obtain short peaking time tp and short pulse
width t0.01.
The pile-up on pulse undershoots is mainly observed in AC coupled amplifiers.
The remedy for this effect is to use pole-zero cancellation circuit (see Chapter 3.5.2) and
base line restores (see Chapter 3.5.3).
(a) (b)
Fig. 3.31. Two types of pulse pile-up: a) on pulse tail at CSA output, b) on pulse undershoot at shaper
output.
Qin
VoutCSA (t) ≈ exp( −t / τ f ) (3.111)
CF
where τf = CF Rf and the negative charge at the input is assumed to simplify further
equations and plots. Noise optimization of the CSA stage together with the feedback
circuit often gives the feedback resistance in hundreds of MΩ range and then the feed-
72 Architecture of front-end electronics
back time constant τf is consequently long. The train of input pulses of high average
frequency fin generates a "positive" DC voltage shift VshiftCSA at the CSA output (pile-up
on pulse tail), which can be calculated as [114]
∞
VshiftCSA ≈ f in ∫ VoutCSA (t)dt (3.112)
0
Using the formula (3.111) and assuming that all the input pulses carry the same
charge −Qin , the equation (3.112) can be rewritten as
From the above equation it can be seen that the DC voltage shift VshiftCSA is propor-
tional to the charge Qin of the input pulse, the average frequency fin and to the feedback
resistance Rf. A compromise should be found for the Rf value in order to obtain an ac-
ceptable VshiftCSA without significant deterioration of circuit noise.
Let us estimate the influence of feedback resistance on the voltage shift at the
CSA output and the noise performance, considering the following example: charge pul-
ses of −Qin = 1 fC arrive at the CSA input with the average frequency fin = 2 MHz, while
the CSA has an ideal feedback resistance of Rf = 200 MΩ and a RC-(CR)2 shaper has
a peaking time tp = 100 ns. For these conditions, according to formula (3.113), the DC
voltage shift is VshiftCSA = 400 mV, while the contribution feedback resistance Rf is
ENCi ≈ 15 e− rms [79].
The baseline shift at CSA output is non-negligible in modern CMOS technologies
which are usually characterized by a low value of power supply voltage. It must also be
taken into account in case of active elements in CSA feedback (see Fig. 3.10), because
this DC voltage shift VshiftCSA can modify the effective feedback resistance.
2
Qin 1 s ⎛ 1 ⎞
voutSH (s) = × ×⎜ ⎟ (3.114)
Cf s + 1 1 ⎝ sCR + 1 ⎠
s+
CF R f CR
In the time domain the above pulse has a long "negative" undershoot, whose am-
plitude and width depend on the time constant τf and its relation to filter time constant
τ - see Fig. 3.32.
(a) (b)
Fig. 3.32. Pulses for different CSA feedback time constant: a) at the CSA output, b) at the shaper output -
the second order shaper with tp ≈ 180 ns.
At low rates of input pulses, this pulse overshoot is usually tolerable, however, in
high rate experiments the consecutive input pulses produce a negative baseline shift at
the shaper output and loss of the amplitude resolution of the system (see Fig. 3.33(a)).
Additionally, if the readout system operates at high fluctuating random rates, a signifi-
cant base line variation is produced. The undershoots can be eliminated (see
Fig. 3.33(b)) by applying the pole-zero cancellation circuit [115,116].
(a) (b)
Fig. 3.33. Examples of waveforms at shaper output for high rate of input pulses:
a) without PZC circuit, b) with PZC circuit.
74 Architecture of front-end electronics
The idea of a PZC circuit is shown in Fig. 3.34. By adding an extra resistor Rpz
(parallel to capacitor C) the long time constant τf at CSA output can be cancelled by the
subsequent stage if the following condition is fulfilled
C F R f = CR pz (3.115)
2
Q 1 ⎛ 1 ⎞
voutSH ( s ) = in ×⎜ ⎟ (3.117)
CF s + 1 ⎝ sCR + 1 ⎠
C (R pz || R )
The "pole" of the CSA is cancelled by the "zero" of the PZC circuit. The new time con-
stant after the PZC is equal to C(Rpz||R) and it is smaller than τf.
Different practical implementation of the PZC circuit for an integrated circuit can
be found in [62, 78, 79, 81, 115−119]. One of the most popular solutions in an inte-
grated circuit is to use MOS transistors in CSA feedback and a PZC circuit working in
linear or saturation region - see Fig. 3.35. The response at the shaper output is
Rf sC R pz + 1
voutSH ( s ) = −Qin × × Z1 (3.118)
sCF R f + 1 R pz
Architecture of front-end electronics 75
C
voutSH ( s) = −Qin × Z1 (3.119)
CF
where Rf and Rpz represent equivalent small signal channel resistance of the MF and
MPZ transistors respectively.
The baseline problem is less severe when bipolar shaping is employed. However,
the unipolar pulse shaping is more advantageous than bipolar shaping when considering
noise performance and pulse width. Adding a PZC block (like one in Fig. 3.34 or
Fig. 3.35) means that the CSA−shaper block becomes DC coupled (without Rpz the
CSA-shaper block was AC coupled). The AC coupling between the stages in the front-
end electronics is often used in multichannel systems to avoid offset propagation
through the entire signal path and to minimize the problem of the leakage current in case
of a DC coupled detector. A sequence of unipolar pulses passing through an AC coupled
stages leads to the base line shift. Similar effects occur in the systems with a continuous
DC path from input to output which modify the frequency response of the front-end
electronics through feedback network (usually to reduce the gain in low frequency
range). In these cases the restoration of the base line is necessary and Base Line Restor-
ers (BLR) or a Base Line Holder (BLH) are used for this purpose. The idea of the BLR
is shown in Fig. 3.36. The basic components of the BLR are memory C capacitor and
a switch of small resistance R. Between the pulses, the switch is closed along the base-
line. When the signal arrives the switch opens and the output signal is given by the input
signal minus its base line [120,121].
76 Architecture of front-end electronics
Four basic configurations of the BLR are shown in Fig. 3.37. The BLR shown in
Fig. 3.37(a) works with a signal of positive polarity and clipping diode shorts to ground
the negative undershoots. These circuits work effectively for input pulse of amplitude
considerably higher than
kT
vmax >> (3.120)
q
where kT/q = 26 mV in room temperature. Applying here bipolar pulses at the input can
lead to double pulses at the output [37].
The circuit shown in Fig 3.37(b) was suggested by Robinson, et al. [122] and it
works for input pulses of both polarities. In the base configuration the current I2 = ID is
a half of the current I1 = 2ID. Without input signals, both diodes are forward bias, each
carrying current ID and the node X is shorted to ground via small rD resistance of bias
forward diodes. The time constant of the input circuit is τ ≈ 2rDC. Input signals (greater
than a few hundreds mV) of either polarity cut off one of the diodes and change the time
constant τ. Positive input signals cut off the diode D1 and the current I2 = ID flows into
the coupling capacitor C. Negative input signals cut off the diode D2 and the current
ID = I1−I2 charges coupling capacitor C. The output pulse is given as
1
vout (t ) = vin (t ) − I Dt (3.121)
C
Between pulses the charge leaks off through small resistance rD of two series of
connected diodes and a small residual baseline shift is observed even for the quite high
rate of input pulses [123]
2rD I D f D 2 Kf D
ΔVshift = = (3.122)
1− fD 1− fD
where fD is the duty factor of input pulses and diode resistance rD = K/ID (for semicon-
ductor diodes the K value is between 25 mV and 50 mV).
Architecture of front-end electronics 77
(a) (b)
(c) (d)
Fig. 3.37. Examples of different configurations of the BLR: a) with a clipping diode, b) Robinson
configuration with a two diodes, c) Chase-Poulo restorer, d) Gere-Miller restorer.
Together with the increase of the duty factor fD > 10% the performance of the
Robinson restorer becomes significantly degraded because of the finite diode incre-
mental resistance. A modification to decrease the effective diode resistance was pro-
posed by Chase and Poulo [123] - see Fig. 3.37(c). By adding an amplifier with diodes
in the feedback loop their effective dynamic resistance is decreased by the factor equal
to the amplifier gain, without increasing the diode current.
An interesting concept of the BLR was proposed by Gere and Miller [124] - see
Fig. 3.37(d). Between the signals current Ig is pulled from the amplifier input. The cur-
rent value is chosen so that Ig > C(dVin/dt)max and the diode remains closed and the node
X is held at ground potential. Sometime before the peak of the input pulse the gate cur-
rent is switched off. For the negative-going input pulse the point X remains at ground.
However, as soon as the input signals starts to go positive, the diode opens. As a result,
the full peak-to-peak amplitude is obtained at the output.
78 Architecture of front-end electronics
Fig. 3.38. BLR with switch and transconductor structure [121] © 2005, IEEE.
Development of high rate spectrometry has caused that there are some new reali-
zations of the circuits which are BLR or perform similar function [121, 125−132].
Let us analyze two examples of these circuits which were realized in CMOS tech-
nology. The first circuit (0.35 μm CMOS technology) was suggested by Pulia, et al.
[121] and is presented in Fig. 3.38. The BLR consists of memory capacitor C, switch S
and an active circuit (amplifier A1, transistors M1−M6) which works as unity gain
buffer or large value resistor. A two-position switch S is driven by the arrival of the
input signal. When the switch is closed to position a the transconductor (A1-M1-R)
together with two current mirrors (M2−M3 with ratio K1 = 25 and M4−M5 with the
ratio K2 = 20) is equivalent to large dynamic resistance Req connected to node a given as
Req = R × K1 × K 2 (3.123)
For the integrated resistor R = 60 kΩ the above equivalent resistance Req equals
30 MΩ. When the switch is connected to node c, the voltage drop across the capacitor C
is frozen, and the structure A1-M1-R works like a unity gain buffer. The switch S is
driven by discriminator A2 which compares the input current with a preset threshold
current Ith.
The BLRs have usually large bandwidth in order to provide a quick return to
baseline between the pulses. A different concept is used in the base line holder (BLH) as
proposed by Geronimo, et al. [132, 133]. The BLH concept is based on the feedback
loop closed on shaper amplifier and the output stage - see Fig. 3.39. The BLH loop con-
sists of a differential amplifier, non-linear buffer and low-pass filter. The output voltage
Architecture of front-end electronics 79
Vout is compared to the reference voltage. The following non-linear buffer reduces the
gain of the feedback loop only in the presence of a large and fast signal. The loop pass
filter reduces the loop bandwidth and ensures the stability of the loop. The current IF
(produced by a low pass filter) is subtracted from the main signal current IIN at the
shaper input (input of the shaper is assumed to be virtual ground).
Fig. 3.39. Simplified scheme of part of front-end channel containing BLH [132] © 2000, IEEE.
A detailed scheme of the BLH is shown in Fig. 3.40. The non-linear buffer is
based on transistor M1 with controlled bias current (M2 controlled by Vg1) and C1B ca-
pacitor. Low bias current of source follower M1-M2 and relatively large C1B cause that
the response of the buffer to large and fast pulses is one directional slew-rate limited to
IdM1/C1B (IdM1 is a bias current of M1 transistor). The next source follower M3 (bias from
current source M4) with the capacitance C2B forms a low pass filter. Low bias current of
this stage and properly set C2B provide the dominant pole of the loop at very low fre-
quency. The transistors M5 are used for voltage to current conversion. For small and
slow signals the high loop gain keeps Vout = VBL (no slew rate limit occurs). Large and
fast signals are unaffected by the BLH. The main drawback of this solution is residual
fluctuation of the baseline discussed in detail in [132].
Further signal processing at the shaper outputs in the multichannel chip depends
strongly on specific requirements of the foreseen applications. The ideal system should
ensure position, energy and timing measurements. And that means that each channel
should contain fast high resolution analog-to-digital converter (ADC), time-to-digital
converter (TDC) and a big memory buffer. Nowadays, the constraints associated with
the area and power consumption of multichannel IC make this approach difficult to real-
ize in practice. Instead, the existing multichannel ASICs for fast digital imaging use an
intermediate solutions, which reduce the energy (or timing) resolution or single channel
maximum throughput.
Let us concentrate on applications for fast X-ray digital imaging. Solutions most
frequently used include:
− one or a few discriminators per single channel [44, 46, 134, 135] - see Fig. 3.41,
− a simple low resolution ADC per single channel [39−41] - see Fig. 3.42,
− multiplexing technique with high resolution ADC [42,43] - see Fig. 3.43.
The architecture with one or a few amplitude discriminators per single channel is
used in various imaging techniques, where it is sufficient to measure spatial distributions
of the X-rays of energies above a given threshold or within a given energy window. The
existing solutions have up to 9 discriminators per single channel [135]. The counters
after the discriminators store the information about the number of signals, which have
been higher than the given threshold level. As each channel works independently, this
Architecture of front-end electronics 81
solution is suitable for very high intensity of X-rays (even more than 1 Mcps per single
channel). This is a significant advantage in systems comprising hundreds or thousands
of channels. Additionally, using this simple approach, by scanning the discriminator
threshold level and measuring the integral distribution of pulse amplitudes, one could
extract spectroscopic information of X-ray radiation. Examples of discriminator archi-
tectures are presented in Chapter 3.6.1.
In the time-over-threshold method [144, 145], the analog signal from the front-
end circuit is applied to a simple threshold discriminator. Signals above the threshold
generate a logic pulse at the discriminator output. The duration time of this pulse is
measured in a simple way by counting pulses from the clock generator over the period
equal to the duration of the discriminator response. The width of the discriminator pulse
is a non-linear function of pulse amplitude and knowing this function for a given pulse
shape one could determine the pulse amplitude at discriminator input. The basic limita-
tions of ToT processor are: the measurement range limited by the discrimination level,
82 Architecture of front-end electronics
low accuracy for small signals just above the threshold and sensitivity to the time jitter
of the discriminator, especially for low amplitudes. The system with a ToT processor is
relatively simple and because of low power consumption it can be implemented in every
channel. This solution has been implemented in several ASICs for readout of strip/pixel
detector in the particle physics experiments [144−146].
A fully analog readout scheme, employing a standard ADC, is shown schemati-
cally in Fig. 3.43. Each channel is equipped with a peak detector and a Sample and Hold
(S&H) circuit. Because in X-ray applications the input pulses are statistically distributed
in time, one needs to implement a threshold discriminator in each channel to generate
a trigger signal for the S&H circuit. Analog signals from a certain number of channels
are then multiplexed into one ADC, which, in most cases, is an external device, although
one can consider integrating it in the front-end ASIC. An obvious limitation of the
maximum channel throughput is the multiplexing rate, which is less critical when we
reduce the number of channels per ADC and increase the number of ADCs in the sys-
tem. Another aspect which is specific for such architecture concerns the control of the
multiplexer and the ADC operation. One can either run the multiplexer and the ADC
continuously [149] allowing for some probability of the pile-ups in the S&H circuits, or
trigger the multiplexer and the ADC upon a signal occurring in the detector [42]. The
most commonly used scheme is based on OR gate taking inputs from all the channels.
Then each signal occurring in any of the channels triggers the readout sequence. An-
other trend of the development is to integrate in the front-end ASICs analog memory
buffers which can serve as derandomizers and in that way increase the channel through-
put (see Chapter 3.6.2).
Fig. 3.43. Analog architecture employing multiplexing of analog signals and fast ADC.
Architecture of front-end electronics 83
3.6.1. Discriminators
The discriminators used in multichannel binary and counting systems are rela-
tively simple, however often contain blocks for offset correction. The threshold adjust-
ment can be done using:
− trimming DACs (3-bit up to 8-bit) [44, 150],
− analog approach to store correction voltage on the capacitor [151].
(a)
(b)
(c) (d)
Fig. 3.44. Possible solutions of applying threshold voltage: a) appling controlled current to resistive
structure, b) applying two differential voltages, c) changing the switching point of differential amplifier,
d) with feedback diode on simple inverter.
Threshold voltage or trim voltage can be applied in different ways (see Fig 3.44).
These are possible solutions:
− controlled current is fed into the resistor (which can be based on the MOS structure) -
Fig. 3.44(a) [151],
− a controlled current source connected to one branch of differential pair shifts its
switching point - Fig. 3.44(b) [44, 70],
84 Architecture of front-end electronics
(a)
(b)
Fig. 3.45. Two differential discriminators used in a single photon counting system: a) schematic, b) pulse
waveform at input, after setting the threshold and at the output [152] © 2007, IEEE.
Architecture of front-end electronics 85
The correction circuit consists of transistors M20 and M21 as well as a tunable
current source ICOR (set by the internal trim DAC). The correction circuit is common for
both discriminators because the whole front-end electronics channel is DC coupled and
dominant contribution of offset spread comes from the earlier CSA and shaper stages.
The correction circuit trims a channel-to-channel DC level spread at the discriminator
inputs (VINREF voltage is set in each channel independently). The input differential stages
M22−M26 and M36−M40 have gain close to 1 and are used to provide differential
threshold voltages (common value for all 64 channels) for the first and the second dis-
criminator respectively
Fig. 3.46. Differential discriminators used in a single photon counting system PX90 [87], © 2010, IEEE.
86 Architecture of front-end electronics
This discriminator has a differential input and consists of three main parts:
− two input source followers (transistors M21 and M22) with controlled currents
sources I1 and I2 to trim a channel-to-channel DC level spread,
− differential stage (transistors M23−M29) which add some gain and set the thresh-
old level,
− core of a discriminator (transistors M34−M44).
There is an OTA at the core discriminator input (transistors M34−M38) to convert
voltage signal to current signal. The discrimination is then performed by subsequent
current comparator (transistors M39-M42) [70, 154]. The last output amplifier (transis-
tors M43 and M44) is added to drive the logic gate properly. The maximum current in
the last stage is limited to 0.7 μA to reduce the current spikes (on power supply lines)
generated during comparator switching and the current consumption.
(a) (b)
Fig. 3.47. Simplified scheme of a two-phase peak detector: a) WRITE phase, b) READ phase (reprinted
from [155] © 2002, with permission from Elsevier, http://www.sciencedirect.com/science/journal/ ).
Architecture of front-end electronics 87
The basis of the chip is a two-phase peak detector (PD) with high absolute accu-
racy (0.2%) and linearity (0.05%) which is shown schematically in Fig. 3.47. In the
WRITE phase PMOS transistors M1 load the hold capacitor CH up to the peak value of
the input pulse. In the READ phase it works as unity gain buffer. This PD scheme is an
offset-free configuration, because in the WRITE-READ cycle the error of amplifier
offset is cancelled.
The PDD ASIC contains 32 input channels and performs simultaneous amplitude
and time measurements, together with delivery of the channel address information. The
chip has been fabricated in 0.35 μm CMOS technology. This fully self-triggered ASIC
multiplexes 32 shaped input signals into an array of eight peak detectors (which act as
a derandomizing analog memory) with associated time-to-amplitude converters (TAC) -
see Fig. 3.48. 32 comparators monitor the inputs for any activity. When the input pulse
arrives, the arbitration logic routes the pulse to the next available PD/TAC. The connec-
tion is maintained until the peak amplitude is found. This value is stored on the PD hold
capacitor until the external ADC is ready to convert this sample.
Fig. 3.48. Block diagram of amplitude and time measurement ASIC [157] © 2003, IEEE.
Integrated circuits for strip/pixel detector consist of many channels. For strip de-
tector, the chips often contain 32, 64 or 128 readout channels in a single die (see Chap-
ter 6), while for the pixel detector, it can reach several thousands readout channels work-
ing parallelly and independently, e.g. MEDIPIX2 IC has 65536 identical front-end
channels [44]. The number of readout channels in a single integrated circuit puts con-
straints on the maximum available area and power dissipation per single channel. Be-
sides, to make the system effective, the spread of analog parameters from channel to
channel must be minimized; therefore, the good matching performance as the additional
requirement should be taken into account starting from the early design stage.
Another aspect is a mixed-mode character of readout IC. In order to improve the
functionality and reliability of multichannel systems, the single chip should be equipped
with digital blocks for control and testing the chip itself and some digital data processing
90 Important aspects of multichannel IC
directly on the chip. This results in extra crosstalk between sensitive analog circuits and
fast digital blocks, usually known as switching noise.
In most cases the main requirements for multichannel mixed-mode readout elec-
tronics for semiconductor detectors are:
− low level of noise and high speed pulse processing,
− strict limits on maximum power dissipation and area of single readout channel,
− minimization of crosstalk effects in the mixed-mode integrated circuits,
− good matching performance of analog parameters (gain, offsets, noise) from channel-
to-channel.
To the community of electronic VLSI designers, each of the above requirements
is very often a challenging task. There are number of interesting and valuable papers,
which describe each of the above aspects, but separately. In case of multichannel mixed-
mode integrated circuits, all of these aspects are important and they often necessitate
conflicting requirements [158].
Fig. 4.1. Relations among different aspects of multichannel mixed-mode ICs design.
The relations among different design optimization aspects are shown schemati-
cally in Fig. 4.1. For example, biasing an input stage of front-end electronics with high
current is advantageous for the noise optimization, but in multichannel system it is lim-
ited by available power dissipation in a single channel. Some elements of circuit archi-
tecture which are preferred from the point of view of crosstalk minimization (e.g. differ-
ential structure in analog blocks) increase the noise. The limitation of a single channel
area is in opposition to the low level of 1/f noise and to the reduction of mismatch ef-
fects.
Taking into account the above-mentioned aspects, we would like to concentrate
here on three problems important for such mixed-mode multichannel ASIC designs,
viz.[159]:
− SPICE modeling of noise in MOS transistors and short channel effects,
− crosstalk phenomena in mixed-mode integrated circuits,
− mismatch modeling in the MOS transistor and circuits.
Important aspects of multichannel IC 91
Noise sets a lower limit to the accuracy of any measurements and to the amplitude
of signal that can be processed electronically. By understanding the noise origin and
models used by the circuit simulators, one can significantly improve noise parameters of
the circuit. As we analyze the CMOS circuits, this chapter concentrates on a brief review
of noise models for the MOS transistor, including simple models for first-hand calcula-
tion, the example of numerical model (used by HSPICE) and their limitation due to
short channel effects.
There are several noise sources in MOS transistors:
− thermal noise in the drain-source channel,
− flicker noise in the drain-source channel,
− induced gate noise at high frequencies,
− shot noise due to leakage current through the SiO2 gate,
− shot noise associated with leakage current of the drain (source) reverse biased diodes,
− thermal noise due to the source, drain and gate contact resistance.
In most cases only the first two items are important, so they are discussed in detail
in the next subsections. Induced gate noise becomes non-negligible at high frequencies
when the MOS transistor must be considered as an RC distributed network. The capaci-
tive coupling to the gate is modeled as a distributed capacitance and the channel itself is
represented as a distributed resistance [160, 161]. The gate admittance has capacitive
and conductive components. The conductance has noise associated with it. To calculate
the induced noise at higher frequency one should determine the degree of correlation
between the gate and drain thermal induced noise current, since both noise currents have
the same physical origin. The leakage current through the SiO2 gate and drain (source)
reverse biased diodes are in most cases negligible, and the shot noise associated with
them can be also neglected. The source, drain and gate contact resistances contribute to
the thermal noise and their power spectral densities are provided by well-known Nyquist
equation.
dvth2
= 4kTR (4.1)
df
where k is Boltzman constant and T is an absolute temperature.
92 Important aspects of multichannel IC
According to the Norton’s theorem, the voltage noise source, as above, can be
represented as a parallel current source with the power spectral density equal to
dith2 4kT
= (4.2)
df R
The MOS transistor has a resistive channel between the drain and the source, and
the random thermal motion of carriers in the channel results in thermal noise. This noise
depends on bias conditions, transistor dimensions and the properties of a given technol-
ogy. For the first-hand noise calculation the thermal channel noise in strong inversion in
the saturation region can be expressed as
dith2 8
= kTg m (4.3)
df 3
where gm is gate transconductance of the MOS transistor. In the linear region, when
drain-source voltage VDS is close to zero, the power spectral density is given as
dith2
= 4kTg ds (4.4)
df
where gds is source-drain small signal conductance of the MOS transistor.
In the weak inversion region for VDS > 5kT/q, the corresponding expression for the
thermal noise is
dith2 n
≈ 2qI DS ≈ 4kT s g m (4.5)
df 2
dith2 8
= kTg m (4.6)
df 3
Important aspects of multichannel IC 93
The above equation is used both in the saturation and the linear regions, but in the
latter case it can lead to wrong results. The model incorrectly predicts that the thermal
noise of the MOS transistor falls to zero, when VDS = 0 (because then gm equals 0). If the
NLEV model parameter is set to 3, HSPICE uses a different equation which is valid in
both the linear and saturation regions
dith2 8kT W 1 + a + a2
= μ Cox (VGS − VT ) GDSNIO (4.7)
df 3 L 1+ a
where μ is mobility, Cox is oxide capacitance per unit area, W, L are width and length of
MOS transistor, VGS is gate-source voltage, VT is threshold voltage, GDSNIO is HSPICE
channel thermal noise coefficient (default equals 1), a is parameter defined as
a = 1 − VDS / VDSsat , VDS is drain-source voltage, VDSsat is saturation drain-source voltage.
In the linear region with VDS = 0, we have a = 1 that gives [163]
dith2
= 4kTg ds (4.8)
df
For the HSPICE NLEV equaling 3, the noise model works reasonably well for
long channel devices, but is not adequate for short channel ones.
The other equations are used by the BSIM4 model from UC Berkeley [164].
There are also two channel thermal noise models, which can be selected by the model
flag tnoiMod. For the default tnoiMod = 0, the charge base model is used (similar to that
used in BSIM3v3.2). Power spectral density of thermal noise given by the formula
dith2 4kT
= NTNOI (4.9)
df L2eff
Rds (V ) +
μeff Qinv
where Rds(V) is the bias-dependent LDD (light doped drain) source/drain resistance, μeff
is effective mobility, Leff is effective channel length, Qinv is total inversion channel
charge and the NTNOI parameter is introduced for more accurate fitting of short channel
devices. The inversion channel charge is computed from the charge-based capacitance
model equations [163], taking into account the effective channel width and length, the
operation point and the bulk charge effect.
For tnoiMod = 1, the holistic model is used which in the scheme gives noise volt-
age source partitioned to the source side and noise current source which is put in the
channel region. This model takes into account:
− short-channel effects and velocity saturation,
94 Important aspects of multichannel IC
− the amplification of the channel thermal noise through gm and gmb, as well as the in-
duced gate noise with partial correlation to the channel thermal noise.
dv12/ f K f 1 1
= 2 (4.10)
df Cox WL f α
dv12/f K f (VGS ) 1 1
= (4.11)
df Cox WL f α
where Kf (VGS) is a bias dependent factor and α ≈ 1. The inverse proportionality with Cox
is not universally accepted [163].
The above models are generally accepted for all regions of channel inversion. In
most cases it is noticed that the PMOS transistors have significantly less noise than the
NMOS ones (by one order of magnitude or more). That is attributed to the buried chan-
nel character of the PMOS transistor, because the channel is farther from the Si-SiO2
interface and the charge carriers in the transistor channel are less affected by the inter-
face traps. In modern technologies where the buried channel effect is no longer present,
the flicker noise for the PMOS and NMOS transistors can be similar [168].
The HSPICE MOS transistor model for the flicker noise simulation uses a current
noise source placed between the drain and the source. That enables calculation of the
Important aspects of multichannel IC 95
power spectral density of the total noise as the sum of white and flicker noise compo-
nents given by
To obtain the representation of the equivalent input noise voltage source in series
with the transistor gate, the following relation should be used
dv n2 1 di 2
= 2 total (4.13)
df g m df
For the flicker noise, two parameters are used by the HSPICE models: flicker
noise coefficient KF and flicker noise exponent AF (default equals 1). The model equa-
tions are selected by the NLEV parameter. For the NLEV = 0, the power spectral density
of the flicker current noise is equal to
AF
di12/ f KF I DS 1
= 2
(4.14)
df Cox Leff f
For the NLEV = 1 L2eff in the above equation is replaced by Weff Leff and then
di12/f KF I DS AF
1
= (4.15)
df Cox Weff Leff f
where Weff is the effective channel width. For the NLEV equaling 2 or 3, the equation is
as follows
di12/ f KF g m2 1
= (4.16)
df Cox Weff Leff f AF
The BSIM4 gives two flicker noise models depending on the fnoiMod selector.
For fnoiMod = 0 the simple model similar to the equation (4.14) is used (with an addi-
tional frequency exponent close to 1). For fnoiMod = 1 a unified model is used with
a fairly complicated formula for the flicker noise calculation with three parameters:
NOIA, NOIB, NOIC. The unified model takes into account the effective channel width
96 Important aspects of multichannel IC
and length as well as the effective mobility at given bias conditions and the charge bulk
effect [164]. The BSIM4 flicker noise model is smooth over all bias regions.
ceeds several millions) and many papers focus their experimental noise study on transis-
tors with geometry and biasing suitable for the CSA input device.
Table 4.1. Transistor dimensions used during the noise measurement tests [106].
Technology node Transistor length [μm] Transistor width [μm]
CMOS 130 nm 0.13 / 0.2 / 0.35 / 0.5 / 0.7 200 / 600 / 1000
CMOS 90 nm 0.1 / 0.13 / 0.2 / 0.35 / 0.5 / 0.7 200 / 600
(a) (b)
Fig. 4.2. Measurement results of 1/f coefficient Kf at VDS = 0.6 V as a function of: (a) gate length for NMOS
transistors, (b) gate overdrive voltage for PMOS transistors [106] © 2007, IEEE.
In case of thermal noise there were no short channel effects in the measured oper-
ating region except for NMOS transistors with the minimum feature size in 130 nm
process. For 90 nm process the thermal noise in NMOS devices with L ≤ 130 nm were
not presented because the noise corner frequency (boundary between 1/f and white
98 Important aspects of multichannel IC
noise) was located at several tens of MHz (close to the higher frequency limit of the
measured system).
The measurements of noise of technologies from the same nodes but delivered by
different vendors (different technology steps) can provide different results concerning
the short channel effects. Some designers claim [175] that using minimum transistor
length with a small current density IDS/W and a low VDS voltage (just above VDSsat) will
cause only a modest increase in noise due to short channel effects. Without dedicated
noise measurements, some designers prefer to use an input transistor with the length of
30% − 100% higher than Lmin.
capacitance to overload and the larger voltage swing, the higher switching noise. The
noise can be transferred to the analog blocks in two major ways: through the common
power supply lines or through the common substrate shared by the analog and the digital
circuits (see Fig. 4.3).
The crosstalk through the common supply lines is often called a supply bounce
[178]. When many gates change states, a large cumulative current spike flows through
parasitic inductance and resistance of bias lines creating power supply voltage spikes.
The main source of the series inductance in the integrated circuits is the parasitic induc-
tance of the bond wires and package leads. The current spikes drawn from the supply
lines of inductance Lind generate the voltage drop V = Lind dI/dt, and in consequence,
these supply lines can be very noisy. Because the analog blocks always have a limited
Power Supply Rejection Ratio (PSRR), the disturbances on the supply lines can corrupt
their achievable accuracy. The main method to minimize the supply bounce is to reduce
the value of the power supply connection inductance.
Even if power supply noise is significantly reduced, there is another method of
the crosstalk due to the non-ideal isolation provided by the common substrate. Every
switching activity of the digital blocks injects current into the substrate and causes the
fluctuation of the substrate voltage. This is known as a substrate noise [179] despite the
fact that it is not a real noise. Because of the non-zero dielectric constant and the con-
ductivity of the substrate material, the parasitic currents can reach different parts of the
chip.
There are three main mechanisms for injecting substrate noise. The first one is the
capacitive coupling from switching nodes of active and passive devices. For example, in
case of the NMOS transistor made in p-type substrate, noise is coupled to the substrate
via source/drain-to-substrate capacitance, while the vertical NPN transistor interacts
with the substrate through the collector-to-bulk junction capacitance [180].
100 Important aspects of multichannel IC
The second source of substrate noise is the coupling from the clock lines and
power supply lines through the line-to-substrate capacitance. Moreover, in most cases
the digital ground is connected to the substrate in every standard digital cell, and then
the supply bounce on digital ground is directly coupled to the substrate [181].
The third mechanism responsible for the noise injection is the impact ionization in
MOS transistors. For the short channel devices working in saturation, the electrical field
strength in the drain-end of the channel can be high enough to cause impact ionization
and to generate electron-hole pairs. The holes form a drain-to-substrate current which is
always positive for both 1−0 and 0−1 transitions in the drain node. The effect of the
impact ionization is often taken into account by the SPICE transistors models provided
by the foundry. Whether or not the impact ionization is an important source of the sub-
strate noise depends on the technology, especially on the combination of the supply
voltage and the channel length. For example, for the epi-type 0.5 μm 3.3 V technology
presented in [181], the substrate current caused by impact ionization is negligible.
Transfer of the switching noise through the substrate depends strongly on its elec-
trical properties. In most cases bulk CMOS technologies use one of two kinds of sub-
strate: a low resistivity substrate or a high resistivity substrate shown schematically in
Fig. 4.4. In a high resistivity substrate, the bulk region of 200−400 μm thick is lightly
doped silicon of resistivity about ρ = 20 Ωcm with a thin (1 μm) implant (or epi) layer
on top. In a low resistivity substrate the bulk is heavily doped and has very low resis-
tivity of the order of 1 mΩcm. Above it there is an epi-type layer of the thickness
Tepi = 5−10 μm with the resistivity in the range ρ =10−15 Ωcm. For the p-type bulk there
is also a thin surface implant on the top (known as p-tub or channel stop) to avoid para-
sitic inversion of the silicon by potential of the lowest metal layers. Technology based
on the low resistivity substrate is the one mostly used for the digital CMOS design due
to its immunity to latch-up.
a) b)
Fig. 4.4. Substrate cross-section: a) high resistivity substrate; b) low resistivity substrate.
Important aspects of multichannel IC 101
Current density J flowing through the substrate can be expressed by the formula
J = (σ + jωε Si ) E (4.17)
The switching noise coupling degrades the performance of the analog blocks. If
the switching noise is of the order of magnitude or higher than thermal, shot or flicker
noise of electronic devices, the analog circuit can loss accuracy, dynamic range, gain
and bandwidth [180]. In addition to the above effects, one should take into account the
possibility of circuit oscillation. The total on-chip power-to-ground capacitance and the
inductance of the bond wires form a resonant LC circuit, which can oscillate at higher
frequency (for example, the total capacitance of 10 pF and the bond wires inductance of
3 nH give the resonant frequency of about 900 MHz). A remedy for the problem is to
reduce the Q-factor of such a circuit by using series resistance or the RLC parallel
branch, as it has been suggested in [184]. There is also another source of possible oscil-
lation when the high gain amplifiers are implemented. The substrate or the power supply
lines on the chip can create a positive feedback path for the signal that leads to the loss
of circuit stability. To cut the positive feedback path, proper isolation techniques dis-
cussed in detail in section 4.2.4 should be applied.
consumption (except for [187]), they are not useful for large digital integrated circuits
and their application is limited to small blocks.
Special attention has to be paid to the nodes with large parasitic capacitance to the
substrate (large fanout) like buses, I/O drivers and clock distribution networks. The rise
and fall times for drivers of these nodes should be as large as the design constrains allow
[178]. The voltage swing on these nodes should be minimized, while the distribution of
the digital signals and clocks in complementary form reduces the net amount of cou-
pling noise.
external world [181]. These parasitics depend on the type of package, type of bonding
and the number of pads used for the power supply lines. For the bonding wire connec-
tions, the typical inductance is 2.6 nH per single wire (45 μm diameter and 2.5 mm
long), while the flip-chip connection has the inductance below 0.2 nH [189]. To de-
crease the inductance of the bonding wire connection, multiple bonding pads and the
bonding wires for each power supply bus can be used.
Still, the application of the above steps may not reduce the noise of the power
supply lines on the chip. If this is the case, it is reasonable to separate power supply
buses, pads and bonding wires for noisy digital and sensitive analog blocks provided
there is enough space on the chip. Some designers may argue that sets of power and
ground connections should be multiplied in order to further divide circuit blocks, for
example fast output drivers [178].
The distribution of power supply on the chip is closely related to careful floor
planning (see Fig. 4.5). The analog blocks need to be categorized by the sensitivity to
the switching noise, while the digital blocks need to be classified by the amount of gen-
erated noise. The most sensitive analog circuit (i.e. high gain preamplifier) should be as
far from the noisy digital circuit (i.e. output buffers) as possible, and the least sensitive
analog blocks should be placed next to the least offensive digital circuit [190]. The
power supply lines must be sufficiently wide, especially in large chips, to avoid un-
wanted voltage drops across the chip, while the current loop area should be kept small.
The floor planning should enable proper signal routing and pads assignment. Digital
signals should not be routed over the analog portion of the chip or close to sensitive
lines. Sensitive and noisy pads must be kept as far away from one another as possible
due to mutual inductance of the bonding wires or package pins.
Fig. 4.5 Example of layout floorplan of 64-channel mixed-mode ASIC RX64 [159].
Important aspects of multichannel IC 105
Placement and connecting substrate contacts are critical for proper isolation (see
Fig. 4.6). Here the techniques applied are different for high and low resistivity sub-
strates, but in both cases every effort should be made to connect the substrate near the
sensitive analog region to a quiet supply [178]. The high resistivity substrate requires for
the latch-up protection that the multiply substrate contacts be tied to the ground periodi-
cally. The substrate needs to be split into quiet and noisy regions. Analog and digital
regions should have separate substrate contacts with separate bonding pads for the
ground. The backside contact to the low impedance ground is recommended, but for the
sake of effectiveness, it usually requires the wafer to be thinned [186].
Fig. 4.6. General rules for separate analog and digital power supply pads and guardrings [184] © 1997,
IEEE.
For the low resistivity substrate, the best solution is the placement of substrate
contacts in the analog part of the chip with a connection to a separate substrate pad, or if
it is not possible, to the analog supply [178]. In that case, a low impedance connection
for the substrate is of great importance, because in this way the isolation can be im-
proved further on by thinning the wafer and by adding the backside low impedance con-
tact (with metallization) tied to the quiet ground by conductive epoxy. Most of standard
cell libraries automatically connect digital ground to the substrate, which is not recom-
mended however because the low impedance path is formed between the quiet analog
and the noisy digital ground buses [186]. The recommended solution is to use a dedi-
cated cell library with separate connections for bulk and source of NMOS transistors (in
the case of p-type common substrate).
Using guard rings can reduce the transfer of switching noise. There are two kinds
of these. Let us consider a p-type substrate. A guard ring may be a simply continuous
ring of substrate contacts (p+ diffusion) that surrounds the circuit providing a low im-
pedance path to the ground for the charge carriers produced in the substrate. The guard
ring can also be formed as n-well ring and as a result the noise currents flowing near the
surface will stop. Methods of using both of them depend strongly on the kind of sub-
strate. Because the substrate current flow is lateral and concentrated near the surface, in
high resistivity substrate, the guard rings are very effective, especially if biased via sepa-
rate bonding pad. In case of p-type substrate, the p+ ring diffusion can reduce the cou-
pling of the switching noise by almost an order of magnitude [179]. The guard rings act
as an efficient current sink, and it is better to put them close to the protected objects. The
106 Important aspects of multichannel IC
best solution is to put two rings around analog and digital blocks with separate package
pins. The floating guard rings are less effective and when tied to the power supply buses
they can even have a detrimental effect. The n-well rings are also recommended, be-
cause they break the low resistivity surface implant layer and force the substrate current
to flow in the substrate underneath the well where the resistivity is significantly higher
[182].
In the low resistivity substrate, most of the current flows vertically to the low re-
sistivity bulk and then through the heavily doped bulk, over the entire chip. The
p+ guard rings reduce the substrate crosstalk only by about 20% when placed close to
the analog circuits and biased via the dedicated pins. The p+ guard rings can be used
separately for analog and digital blocks, but then the separate bonding pads for each of
them are necessary. Connecting the p+ guard ring to the digital ground or large substrate
contact results in the increase of the observed noise [184]. The n-well guard ring in the
low resistivity bulk has nearly no effect [179].
Guard rings are not the only way of shielding. Diffusion, implant, polysilicon or
metal layers tied by the low impedance connection to the power supply non-
contaminated with noise can form vertical Faraday shielding. The n-wells can protect
the device inside very effectively and for that reason PMOS transistors are recom-
mended for signal handling. Large area sensitive devices (including input pads) and
noisy routing channels for clocks should be vertically or horizontally shielded
[191, 192].
The use of different power supply filters is also important. In most cases, it is
done by adding a large decoupling capacitance between the power supply and ground on
the chip and off the chip on the board. The capacitance on the chip can be obtained by
stacking supply rails, using the MOS or polysilicon capacitors. Simple capacitance can
be replaced by more efficient RLC filters [184]. There are other sophisticated techniques
as “active supply bypass” [193] or “active guard ring” [194] which can also be taken
into consideration.
Fig. 4.7. Small area transistor in submicron technology (W = L = 0.25μm) - Poisson statistics gives the
relative fluctuation of active dopant about 3% [205].
AP2
σ (ΔP ) =
2
+ S P2 D 2 (4.18)
WL
current factor β. The threshold voltage of the NMOS transistor (with long and uni-
formly doping channel) built on p-type substrate can be expressed as [147]
VT = VT 0 + γ ( 2φ F + VSB − 2 φ F ) (4.19)
Qox 4qε Si φF N a
VT 0 = φMS + 2 φF − + (4.20)
Cox Cox
2qε Si N a
γ= (4.21)
Cox
where VT0 is threshold voltage for VSB = 0, VSB is source-bulk voltage, γ is the body factor
of MOS transistor, φF is Fermi potential in the bulk, φMS is gate-semiconductor work
function difference, Na is channel doping density, Qox is fixed oxide charge density, εSi
silicon permittivity, Cox is gate oxide capacitance per unit area equaling εox/tox , εox is
silicon dioxide permittivity and tox is oxide thickness.
Let us analyze the contributions of possible errors to the threshold voltage mis-
match:
− because φMS and φF have logarithmic dependence on doping in the substrate [23], they
can be considered as constants with no contribution to any mismatch,
− in the well controlled process the oxide thickness tox is reproducible and its contribu-
tion to the variation of VT can be ignored,
− the density of fixed oxide charges Qox in the modern VLSI MOS processing is much
smaller than the depletion charge density equal to 4qε SiφF N a (see the third and
fourth term in equation (4.20)), and it can be said that the variation of channel doping
level Na is a dominant source of mismatch,
− the variation of body factor γ (caused mainly by the variation of the doping level Na
and oxide thickness tox) is important only for the transistors bias with non-zero VSB volt-
age.
In case of the PMOS transistors, additional term equal to qDI /Cox should be added
in equation (4.20), where DI is the threshold adjust implant dose [207]. The main factors
causing mismatch of the threshold voltage and the bulk factor satisfy the assumptions
(in the first order approximation) of the above model of random matching. According to
equation (4.18) the variance of VT0 and γ can be written as
Important aspects of multichannel IC 111
2
AVT
σ 2 (VT 0 ) = 0
+ SVT
2
0D
2
(4.22)
WL
Aγ2
σ (γ ) =
2
+ Sγ2 D 2 (4.23)
WL
where W and L are width and length of MOS transistor, AVT0 is area proportionality con-
stant of variation of threshold voltage VT0 , SVT0 is constant of variation of threshold volt-
age VT0 with distance, Aγ is area proportionality constant of variation of body factor γ
and Sγ is constant of variation of body factor γ with distance.
The current factor β is given by
W
β = μ Cox (4.24)
L
The mobility μ, oxide capacitance per unit area Cox and the dimensions W and L of the
MOS transistor are all independent. For example, the definition of width W and length L
is determined by different steps in different conditions during IC processing, so they
may be treated independently. With that assumption in mind the variance of current
factor β can be expressed as
σ 2 (β ) σ 2 (μ ) σ 2 (Cox ) σ 2 (W ) σ 2 (L )
= + + + (4.25)
β2 μ2 Cox2 W2 L2
Physical effects responsible for the variation in the mobility and oxide capaci-
tance can be treated according to equation (4.18), while the variation in transistor di-
mensions requires some additional comments. Variations of the width and length origi-
nate from the variations in photolithographic process. From one dimensional analysis of
random error due to the edge roughness it can be inferred that σ2(L)∼1/W and
σ2(W)∼1/L. The equation (4.25) can be rewritten as
σ 2 (β ) Aμ ACox
2 2
AW2 AL2
= + + + + S β2 D 2 (4.26)
β 2 2
WL WL W L WL 2
where Aμ, ACox, AW, AL are technology dependent constants and Sβ is constant of variation
of current factor β with distance.
The paper [199] suggests that for W and L large enough, the above equation can
be approximated as
112 Important aspects of multichannel IC
σ 2 (β ) Aβ
2
≈ + S β2 D 2 (4.27)
β 2
WL
where Aβ is area proportionality constant of variation of current factor β.
Some other authors [207, 210] come to the conclusion that the relative variation
of current factor should be scaled with 1/(W2+L2) rather than with 1/(WL). Moreover, the
experimental results [211] show that even for equal area transistors, shorter channel
lengths and wider channel widths result in poorer matching than longer channel lengths
and narrower channel widths.
The variation of gate oxide capacitance is a common factor in VT and β, so one
could expect the correlation in the mismatches in VT and β. However, both the theoreti-
cal and experimental values of the correlation coefficient are close to zero [207], so for
the first approximation it can be assumed that the variations of VT and β are almost inde-
pendent.
tox 4 N a
σ (VT ) ≈ const (4.28)
WL
Important aspects of multichannel IC 113
Improvement of matching due to the thinner oxide layer is slowed down by the increase
of the doping level under the oxide. The more detailed analysis requires a more accurate
model of statistical dopant fluctuation of MOS transistors given by Stolk, et al. [213].
Fig. 4.8. Threshold voltage matching for NMOS transistors in 180 nm CMOS process. The W/L [μm/μm]
ratio of the devices has been included. Errors bars represent 99% confidence interval [204] © 2002, IEEE.
For the PMOS transistors the situation in VT matching is more complicated and
strongly depends on the technology used. Table 4.3 presents some data for the different
CMOS technologies.
Table 4.3 Matching of VT for NMOS and PMOS for different technologies.
Technology / AVT for NMOS AVT for PMOS Reference
oxide thickness [mV μm] [mV μm]
0.8 μm/15 nm 10.7 18 [168]
0.6 μm/12 nm 11.0 8.5 [168]
0.25 μm/4.4 nm 3.6 − [200]
0.18 μm/3.2 nm 3.0 − [201]
0.15 μm/3.3 nm 3.6 − [201]
0.12 μm/− − 3 [205, 203]
0.1 μm/− 2.5 − [201]
90 nm/− ~3.8 − [202]
65 nm/− 3.3 − [214, 215]
− pockets or halo implants cause high impurity concentration in the space charge region
and strongly influence the matching in short channel devices, etc.
It should be stressed that even if technologies belong to the same technology node, some
of them provide acceptable matching, while others are much worse. The matching de-
pends on process steps and, as it is shown in [201, 216] by careful process optimization,
it can be significantly improved (e.g. for 180 nm CMOS process the threshold voltage
coefficient was reduced from around 8 mVμm to 3 mVμm [201]).
By scaling the technology down, the improvements in β variation are very slow,
i.e. for NMOS transistors scaled down to the oxide thickness from 50 nm to 12 nm the
Aβ parameter is still around 2 %μm [203]. For submicron technologies for a well engi-
neered and controlled process, the Aβ seems to hover around 1 to 2 %μm [205].
β
I DS = (VGS − VT )2 (4.29)
2
As the correlation coefficient between mismatch in VT and β almost equals zero, the
relative source-drain current IDS mismatch may be expressed as
σ 2 ( I DS ) σ 2 (VT ) σ 2 (β )
=4 + (4.30)
2
I DS (VGS − VT )2 β2
At a low value of (VGS−VT ), the dominant factor in the mismatch comes from the vari-
ance of VT, while for the higher (VGS−VT ), the variance of current factor β dominates.
Better matching is obtained at the higher difference (VGS−VT), which means the higher
current. At the low value of VGS , the mismatch does not go to infinity [203,217], be-
cause transistors enter the weak inversion region where square law model (see equation
(4.29)) is no longer valid and ID depends exponentially on VGS
W ⎡ q(V − VT )⎤
I DS = I D 0 exp ⎢ GS ⎥ (4.31)
L ⎣ ns kT ⎦
where ID0 is a process dependent parameter. Calculating the relative source-drain current
IDS mismatch in weak inversion one obtains
Important aspects of multichannel IC 115
σ ( I DS ) σ (VT )
= (4.32)
I DS ns kT q
The subthreshold slope factor ns is usually about 1.1 to 1.3. From above formula for
example for σ(VT) ≈ 5 mV, T = 300 K and ns ≈ 1.2 one obtains σ(IDS)/IDS = 16 %.
(b)
(a)
Fig. 4.9. Mismatch in circuits: a) differential pair, b) current sources.
VGS − VT ⎛ ΔR Δβ ⎞
Vosr = ± ΔVT + ⎜± ± ⎟ (4.33)
2 ⎜ R β ⎟
⎝ ⎠
The equation (4.33) shows that random offset depends both on device mismatches
and bias conditions. The contributions of mismatches ΔR and Δβ increase with the tran-
sistor overdrive (VGS−VT ), while the threshold voltage mismatch ΔVT is directly referred
to the input. To minimize the offset, it is desirable to use low values of (VGS−VT ) by
lowering the bias current or increasing the W/L ratio. A random mismatch results not
only in a random offset but it also reduces the common-mode rejection ratio of the dif-
ferential amplifier [218].
Let us analyze the mismatch of the nominally identical current sources M1 and
M2 shown in the Fig. 4.9(b). Let us assume that both transistors work in strong inver-
sion with the same drain source voltage VDS1 = VDS2. The device mismatch is incorpo-
rated again as VT1 = VT, VT2 = VT±ΔVT, β1 = β, β2 = β±Δβ, that results in different currents
IDS1 = IDS and IDS2 = IDS ± ΔIDS. The relative variation of the output currents can be written
as [210]
ΔI DS 2 Δβ
=± ΔVT ± (4.34)
I DS VGS − VT β
To minimize the random current variation the overdrive voltage (VGS −VT ) should
be maximized. And that is in the opposition to the rule of minimizing the random offset
in the differential amplifier (see equation (4.33)). In practice there is also a systematic
current variation for the case of VDS1 ≠ VDS2, coming from the finite output resistance of
the current source.
skew due to parasitic resistance and capacitance [219]. Sometimes manual correction
must be made in electrical scheme for the proper evaluation of these effects.
Identical temperature. The current of the MOS transistor or the resistor value
depends on temperature. For the MOS transistor, the main temperature dependent pa-
rameters are the mobility μ and the threshold voltage VT. Near room temperature, the
temperature coefficient for the threshold voltage is in the range of 0.5−3 mV/°C [163],
while the mobility depends on the temperature as μ ∼ T −1.5 [23]. So, the matched de-
vices must operate at the same temperature, which does not pose a problem if the power
dissipated by each block on the chip is low. Otherwise, one should identify the large
power sinks on the chip and the distribution of isotherms on the chip. If possible, the
sensitive circuits should be placed at the largest possible distance from power blocks
and realigned on the same isotherms.
The same orientation. IC processes exhibit anisotropy which results from certain
processing steps or lattice orientation and has great influence on matching. For that rea-
son, it is important how the devices are oriented on the layout relative to one another
and what the relative directions of the current flow are. Comparative measurements of
the standard deviation of current factor β matching for transistor placement rotated by
90 degrees show that the matching is several times worse than in case of parallel transis-
tor placement [199]. Because the threshold and substrate factor mismatch is identical for
the rotated and parallel placements, the local mobility variations can be a possible ex-
planation for the rotation dependent mismatch. In case of parallel placement, the better
matching is for the transistors with the same direction of current flow. This is due to
the subtle effect called “gate shadowing” [220]. To avoid the channeling effect during
the drain/source ion implantation, the implant beam (or the wafer) is tilted by 7−9°. For
that reason one obtains small asymmetry between the source and drain diffusion result-
ing in different capacitive coupling as well as different transconductance between the
transistors with opposite current directions.
Common centroid layout. Because of the gradients in temperature, oxide thick-
ness and other process variations, it is difficult to ensure symmetry for large transistors.
To reduce these global errors, the common centroid layout is the most effective solution,
however, routing of interconnections is considered to be a complex process. Also, it is
important to keep the symmetry in the routing of interconnections on the layout and to
avoid differences in the parasitic resistance or capacitance [190, 221].
Unit cells. All devices to be matched should have the same area to the perimeter
ratio. In case of two devices of different dimensions, it is a good practice to base the
layout of both of them on the same unit cell, which can be a simple capacitor, resistor,
transistor or a more complicated structure such as a current cell [222]. Let us consider
the DAC using the binary weighted current source architecture. After setting the mini-
mum dimension of the unit current source according to the required accuracy and
matching parameters of the technology [223], each binary weighted current source is
constructed by putting the appropriate amount of N unit current sources in parallel (for
example a source of weight equaling 16, is implemented as 16 unit cells together). To
minimize the effect of linear gradient, the cells which switch simultaneously in a binary
118 Important aspects of multichannel IC
array are arranged in common centroid geometry [224]. Apparently, the use of only
translated copies is allowed with no rotation or mirroring. Excellent examples of the unit
cells use can be found in [196, 225].
The same surrounding. Adjacent structures have a systematic influence on the
matched devices, so symmetry must be applied not only to the devices under considera-
tion but also to their surrounding environment. Each line of metal, diffusion, polysilicon
etc., has width variation at the IC processing. There is also a line width variation de-
pendent on the location of adjacent structure. This is the proximity effect due to the
variable light interference in exposure and to the variations in chemical flow for photo-
resist, developers and etchants [226]. So, the environment should be identical outside
of matched to the distance of at least 30−50 μm and the matched devices should be
“infinitely far away” from the crystal edge. The remedy for that problem is to surround
the matched devices with dummy structures. The dummy device should be of equal
nature as the matched device, for example for matched transistors (cells) the dummy
structures are also transistors (cells). How many rings of dummy cells are required de-
pends on the requested accuracy and technology [225]. Obviously, the course of this
solution costs an extra area of silicon. However, there is often not enough space for
dummy structures. Then a designer should consider what kind of proximity effect is
important in their design and follow the rules specified by a technology vendor. For
example, taking into account the well proximity effect [227] the recommendation are the
following:
− keeping the same transistor orientation to the well edge (preferable direction is
source/drain perpendicular to the well edge),
− placing matched components at the equal distance (> 3 μm) from the well edge.
Metal coverage. Symmetry is also important in metal coverage. The asymmetri-
cal metal lines over identical MOS transistors which cover one of the transistors (par-
tially or completely) can be disastrous for matching performance. The papers [228, 229]
show that the shape and the degree of metal 1 overlap influence the relative dramatically
mismatch current. The rule is fairly straightforward: it is ideal to avoid metal lines over
matched components. Sometimes it is impossible, for example for high accuracy DAC
[225], and then it is better if the matched cells are covered by higher metal layers than
metal 1, and to obtain the best symmetry all of them should be covered with the same
metal layer.
The mismatch among identical components from different dies, wafers or batches
is always bigger than among components placed on the same chip. As a result, the
spread of channel parameters inside the chip is small, while the differences from chip to
chip on the module are too large to be accepted. The solution is to design on each chip
the correction DACs for tuning the gain, passband, discriminator level, etc., and imple-
ment the local address set during bonding a die on a module. Then, after the calibration
measurements, the correction DACs are loaded one by one. That protocol guarantees the
cancellation of unwanted offsets from chip to chip, and ensures proper work of the
whole module.
The effects of radiation damage were observed in 1962 for the first time during
the mission of Telstar 1 telecommunication satellite. Because of the earlier series of
nuclear weapon explosions performed by both the USA and the Soviet Union, radioac-
tivity of the van Allen Belts surrounding the Earth increased significantly.
Telstar 1 crossed these belts several times which caused significant radiation damages in
satellite electronics. Since then the era of intensive research on the radiation damage
started. The radiation damages are important not only for space systems, but also for
nuclear plants, military equipments, high energy particle experiments and other radiation
detection systems in medicine, material science, etc. Radiation damages are usually
divided into two classes. The first one covers the total dose effects, both displacement
damages and ionization damages, while the second class is connected with Single Event
Effects (SEE), caused by high energy particles. Both types of radiation damage effects
122 Radiation damage
are shortly described below taking into account the physic phenomena in silicon and
MOS structure and degradation of parameters of detector and readout electronic compo-
nents.
A high energy particle can cause displacement of crystal atoms from their lattice
positions. The minimum recoil energy required to remove a silicon atom from its lattice
position is about 25 eV [5]. The displacement damages depend on a particle type, its
mass and energy. All summed energies deposited in a crystal, which has not been used
for a fully reversible process of ionization, are called NonIonizing Energy Loss (NIEL).
The displacement damages are scaled with NIEL. Often, as a reference for a displace-
ment damage 1 MeV neutrons are used. Fig. 5.1 presents relative displacement damages
(to 1 MeV neutrons) vs. energy for neutrons, protons, pions and electrons.
Fig. 5.1. Relative displacement damage for different particles vs. their energies [233].
The displacement damages result in additional energy levels in a silicon band gap.
Some of the displacement damages are not stable and move throughout the crystal. This
movement is strongly temperature and bias depended leading to a complex annealing
behavior.
The main effects of bulk damages in silicon detectors are:
Radiation damage 123
− increase of a leakage current because the additional energy levels in a band-gap act as
generation- recombination centers,
− change of the net effective impurity concentration (see Fig. 5.2) and change of a detec-
tor full depletion voltage (see eq. (2.13)),
− increase of charge trapping centers which hold a part of a signal charge for a time
longer than charge collection time which reduces the signal amplitude.
Fig. 5.2. Absolute value of the effective impurity concentration Neff versus fluence Φ up to 1015 n/cm2 [8]
© 1992, with permission from Elsevier, http://www.sciencedirect.com/science/journal/ ).
In case of integrated circuits the displacement damages are mainly visible in bipo-
lar technologies, because of the lifetime reduction of minority carriers. This lifetime
reduction results in βn,p decrease of a bipolar transistor. This effect depends both on total
particle fluence and on technology of the IC - see Table 5.1.
Table 5.1. Changing in DC current gain of bipolar transistors for different technologies* [234].
Technology DMILL AT&T TEKTRONIX
Dose 1.6 ×1014 neutrons 1.6 ×1014 protons 1.1 ×1014 protons
NPN βn(pre)/ βn(post) 150/30 95/55 95/40
PNP βp(pre)/ βp (post) 40/25 45/23 −
*
Data was normalized for the same collector current density 2.5 μA/μm2
Ionization radiation liberates charge carriers in SiO2 which finally results in posi-
tive charge trapped in oxide and an increase of interface traps at Si-SiO2 border. Let us
consider an MOS structure with p-type bulk (see Fig. 5.3). Average energy needed to
generate an e-h pair in silicon dioxide is about 18 eV. After the charge generation some
of the electron-hole pairs recombine. The mobility of holes and electrons in silicon diox-
124 Radiation damage
ide differs considerably: the mobility of holes ranges from 10-4 to 10-11 cm2/Vs and the
mobility of an electron is about 20 cm2/Vs. Most of the electrons move rapidly towards
the positive bias gate (within a few picoseconds). The remaining holes move slowly
towards Si-SiO2 interface by hopping through localized states in oxide. A fraction of the
holes which reach the interface are trapped there, forming a positive oxide trap charge.
It seems that hydrogen atoms (protons) are likely to be released as "hop" holes through
the oxide or they are trapped near Si-SiO2 interface. The hydrogen ions can also drift to
Si-SiO2 interface where they may react to form interface traps [11]. The charge of inter-
face traps is changed by applying external bias and the interface traps can be negative,
neutral or positive. For PMOS transistors at the threshold, the interface traps are pre-
dominantly positively charged, while for NMOS transistors at the threshold the interface
traps are predominantly negatively charged. The number of positively generated charge
oxide traps and interface traps depends not only on total ionization dose (given usually
in rad(Si) or rad(SiO2)), but it also depends on bias conditions, temperature, technology
parameters (especially oxide processing) and dose rate.
Fig. 5.3. Band diagram of MOS capacitor with a positive gate bias. The main processes for radiation-
induced charge generation are shown [11] © 2008, IEEE.
In case of silicon detectors the ionisation damages cause the increase of detector
leakage current (surface component mainly) and of the interstrip/interpixel capacitance
and they can cause the problem with the proper interstrip isolation in case of a double-
sided detector [7, 53, 55].
For the CMOS technology the ionization damages manifest in [7, 10, 170,
235−237]:
− threshold voltage shift,
− transconductance reduction,
Radiation damage 125
(a)
(b)
Fig. 5.4. Changes in transfer characteristic of MOS transistors - threshold voltage shift and
transconductance reduction are shown: a) PMOS transistor, b) NMOS transistor.
q q
ΔVT = − ΔN ot ± ΔN it (5.1)
Cox Cox
where Cox is gate oxide capacitance per oxide area, ΔNot and ΔNit are numbers of oxide
and interface traps per gate area. In the above equation for the NMOS transistor, the sign
before the term with interface traps is positive, while for the PMOS transistors, this sign
is negative. For the PMOS transistor, the threshold voltage shift is always negative,
while for the NMOS one, the situation is more complex (see Fig. 5.4). In the beginning,
when the positive charged oxide traps dominate, the ΔVT is negative, while later, when
the number of negative charged interface traps increases, the ΔVT becomes more posi-
tive. As it is shown in Fig. 5.4, the increase of the interface traps also reduces the tran-
sistor transconductance, because of mobility degradation according to the formula [238]
μ
μ after = (5.2)
1 + α N ΔN it
where αafter is the mobility after irradiation, αN is a semi-empiric fitting parameter in the
range from 0.35×10-12 [239] to 5.26×10-11 cm2 [240].
In submicron technologies with the oxide thickness below 12 nm the threshold
voltage shift is dramatically reduced (see Fig. 5.5). This is because of electrons which
by a tunneling effect neutralize the trapped holes by recombination.
Fig. 5.5. Threshold voltage shift after irradiation for technologies with different oxide thickness [241-243].
Legend shows the minimum gate length in μm in given technology.
The negligible threshold voltage shift (thank to the thin gate oxide) is a big advan-
tage of deep submicron technologies. However, CMOS technologies use also a thick
field oxide. The positive induced charge in this field oxide creates parasitic NMOS tran-
sistors. These parasitic NMOS transistors are the sources of additional leakage currents,
both in standard NMOS transistors and between two NMOS transistors with different
Radiation damage 127
biased sources/drains (see Fig. 5.6). To eliminate these parasitic and leaky NMOS tran-
sistors a special treatment at the layout drawing is necessary (enclose layout transistors,
additional guarding, etc. - these mitigation techniques are discussed in Chapter 5.3).
(a)
(b)
Fig. 5.6. Radiation damages create leaky parasitic NMOS transistors: a) on the edges of standard NMOS
transistor, b) between NMOS transistors.
Single event effects are caused by a high energy ionizing particle, which crossing
the integrated circuit, creates a lot of e-h pairs. The charge generated by such a particle
can results in:
− soft errors like Single Event Upsets (SEU), Multiple Bit Upsets (MBU), Single Event
Transient (SET) and Single Event Function Interrupts (SEFI) – all these errors are re-
versible and non-destructive,
128 Radiation damage
− hard errors like Single Event Latchups (SEL), Single Event Snapbacks (SES), Single
Hard Errors (SHE), Single Event Gate Ruptures (SEGR) and Single Event Burt-Out
(SEBO) - these errors are non-reversible and can be destructive.
A comprehensive overview of the above-mentioned SEE can be found in [7, 70,
244]. Let us consider the SEU as an example of the most popular SEE. A particle cross-
ing, for example SRAM memory (shown in Fig. 5.7), creates charge along its track.
Linear Energy Transfer (LET) is a measure of the energy transferred to material as an
ionizing particle travels through it. If the linear energy transfer is higher than the critical
LET, the node is vulnerable to the SEU and the memory state can be changed from 0 to
1. The critical LET is lower for modern technologies, where the power supply voltage
and capacitance of the storage node are much lower than in older technologies.
The dedicated rad-hard technologies are relatively expensive and not so easily ac-
cessible as a commercial process. From this point of view, the natural choice is to use
deep submicron technologies where the threshold voltage shift (due to ionizing dam-
ages) is negligible. However, there are two effects which must still be considered to
make the design radiation tolerant. The first one is the accumulation of positive charge
in field oxide which creates the parasitic leaky NMOS transistors (see Fig. 5.6). The
second effect is the problem with increased sensitivity of modern technologies to single
event effects, because of small capacitances in storage nodes and small power supply
voltages (see Fig. 5.7).
One of possible solutions for the leaky parasitic NMOS transistors with thick field
oxide is drawing the layout of NMOS transistors in so-called enclosed gate geometry
and putting them inside p+ guard rings. No special treatment is required in case of
PMOS transistor. A simple layout of an inverter using enclosed layout transistor (ELT)
of NMOS type and standard layout of PMOS transistor is shown in Fig. 5.8.
Fig. 5.8. Simplified layout of inverter: NMOS enclosed gate transistor in a p+ guard ring and standard
PMOS transistor.
However, the enclosed layout NMOS transistors have some limitations, namely
[5, 237, 245]
− the device should be properly modeled,
− small ratio W/L below ~2 is not possible and this for example limits the possibility of
building good current sources based on NMOS transistors,
− capacitances on drain and source nodes are different and this must be taken into ac-
count at the designing stage,
− larger area of an ELT transistor - because of the transistor layout and necessary
p+ guard rings (for example, two series connection of NMOS transistors requires addi-
tional p+ guard rings),
130 Radiation damage
− in CMOS gates - because NMOS ELTs are larger (compared to standard transistors),
and larger PMOS transistors are required for good balance - which results in larger area
and increases dynamic power consumption.
The following steps are recommended with view of the SEE [5, 237, 246−248]:
− eliminate completely the "forbidden" state in a state machine to omit a permanent
lockup,
− increase the critical LET for the SEU by enlarging the width of transistors driving
a sensitive node, increasing capacitive load at this node, or by adding feedback resis-
tance between cross-coupled inverters,
− using correction logic (e.g. Hamming codes) or self-correcting triple redundancy cells
(see example in Fig. 5.9); however, using real redundancy is a very costly solution due
to the silicon area, and it should be used only in the most important blocks of the IC.
Fig. 5.9. D flip-flop with majority voting, self-recover system and information bit of SEU event [248] ©
2002, IEEE.
Examples of multichannel counting IC 131
The most advanced solutions for the multichannel readout architecture of silicon
strip or pixel detectors are known from the instrumentation for particle physics experi-
ments [24, 48, 49, 70, 74, 83, 249−253]. The intensive research of groups working on
such detectors and readout electronics were the inspiration for the application of similar
solutions in X-ray imaging techniques used in solid state physics, material science,
medicine, etc. Especially attractive for these applications are multichannel systems with
a detector of high granularity and a readout electronics working in a single photon
counting mode with a large dynamic range as they are able to cope with high intensity of
X-ray radiation. The examples of this type of mulichannel electronics are presented in
this chapter.
132 Examples of multichannel counting IC
manding. Additionally, HEP experiments are synchronous thus trigger in such system is
available. In X-ray imaging, the input pulses have the stochastic character with the Pois-
son distribution in time.
Single photon counting systems often use a binary readout scheme in each read-
out channel. From the point of view of the ASIC design, the binary readout architecture
means that a multichannel integrated circuit should be able to amplify and filter small
signals from the sensor, perform amplitude discrimination and store the data in binary
form on the integrated circuit in each channel independently at the same time. Such ar-
chitecture provides fully parallel signal processing including data storage from all sensor
elements. And this is suitable for the high count rate applications, since the amount of
data to be handled is already minimized in the front-end ASIC.
Although in the binary readout architecture the front-end channel is ended with
a discriminator, the signal-to-noise ratio remains one of the most critical problems to be
solved. A typical plot of a number of counts at the discriminator output as a function of
threshold voltage level is shown in Fig. 6.1. For low threshold values (in Fig. 6.1 below
30 mV), a rapid increase of counts is observed. This is due to the noise which for low
threshold level switches the comparator on and off and contributes to the total number of
counts.
Fig. 6.1. Total number of counts at the discriminator output vs. threshold level for a given acquisition time.
By differentiating the total number of counts vs. the threshold level, one obtains
what is called pulse height spectrum shown in Fig. 6.2(a). The noise and the signal
smeared by the noise are clearly visible in Fig. 6.2(a). Sufficient separation between the
noise level and the signal level is required in order to provide high detection efficiency
and limit noise count rate. For a system with white input noise and a first order band-
pass filter, the noise count rate at the comparator output is given by the Rice formula
[255]
⎛ V 2⎞
f n = f 0 exp⎜⎜ − TH 2 ⎟⎟
⎝ 2σ n ⎠ (6.1)
where f0 is noise count rate at zero threshold level, VTH is comparator threshold, σn is
voltage noise rms at the comparator input.
134 Examples of multichannel counting IC
(a)
(b)
(c)
(d)
Fig. 6.2. Pulse height spectrum for binary readout architecture for: a) reference channel, b) channels with
different gains, c) channels with different offsets, d) channel with a crosstalk problem. In these cases setting
of the common threshold voltage is impossible.
According to the Rice formula (6.1), in the counting system, we should be able to
set a threshold voltage in the discriminator at (3-5)×σn, where σn is a voltage noise rms
at comparator input. Then, the registration of low level signals with minimizing rate of
false noise hits is possible. However, setting such a low value of the threshold for all
channels (in a multichannel ASIC) may be impossible, if we have problems with chan-
Examples of multichannel counting IC 135
nel-to-channel matching of analog parameters (gain, offset), or with the switching noise.
Various possible amplitude distributions of noise and signals in the channel affected by
mismatch and crosstalk are illustrated in Fig. 6.2 (b)−(d).
Another parameter crucial for various practical applications is a possible opera-
tion of the detector with high intensity of X-ray radiation. Taking into account a fully
parallel architecture of the readout system (including counters which serve as memory
buffers), the counting rate considered for the total area of a detector depends on the fol-
lowing:
− detector segmentation,
− maximum count rate of a single channel.
Detector segmentation is limited mainly by a charge sharing effect [33]. Regarding the
counting rate limit for a single readout channel, a peaking time and a pulse width must
be shortened in a shaper stage and possible effects of pulse pile-up should be eliminated
(by using pole-zero cancellation techniques, avoiding AC coupling circuits between the
stages or using base line restorers/holders). However, a very short peaking time often
means higher noise, while with the absence of AC coupling between stages
(CSA−shaper−discriminator), DC offsets propagate.
CASTOR IC
The CASTOR chip (Counting and Amplifying SysTem fOr Radiation detection)
was one of the first multichannel mixed-mode ASIC working in a single photon count-
ing mode for X-ray application and optimized for silicon strip detector [256, 257].
CASTOR 1.0 version contained 32 channels and was fabricated in 1.2 μm CMOS tech-
nology. The single channel consisted of a charge sensitive amplifier, an RC-CR shaper,
a discriminator and a 16-bit ripple counter (see Fig. 6.3). It used a "classical" architec-
ture with AC couplings between the CSA-shaper and shaper-discriminator stages to
avoid offsets propagation. Resistors in the CSA, shaper feedbacks and in the AC cou-
pling at the discriminator input are realized as simple MOS transistors with controlled
gate potentials. The CSA and shaper cores are based on folded cascode topology. The
CSA input transistor is optimized for input capacitance of 5-10 pF and has dimensions
of W/L = 1000μm/1.5μm with a transconductance of 5.2 mS (for CSA bias current of
1 mA). The CSA feedback capacitance is 200 fF, while the effective feedback resistance
exceeds 100 MΩ. For the peaking time of 850 ns, the noise is ENC = 60+17×Cd e− rms
with the maximum counting rate up to 100-200 kcps/channel. The discriminators have
a common threshold for all channels. To minimize the discriminator offsets to a few mV
input, transistors (in discriminator differential pair) have the size of
W/L = 720μm/1.5μm. The channel gain is 180 mV/fC with the power consumption of
3.8 mW/channel.
Fig. 6.3. Block diagram of CASTOR channel (reprinted from [257] © 1997, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
Examples of multichannel counting IC 137
DEDIX IC
DEDIX IC (Dual Energy Digital Imaging of X-ray) has been developed at the
AGH UST Cracow, Poland as a fast readout ASIC for silicon strip detector [152]. The
chip comprises six basic blocks (see Fig. 6.4(a)): 64 analog front-end channels,
2×64 counters with RAM, an input-output block, a control command decoder, control
DACs and a calibration circuit. The ASIC is designed in 0.35 μm austriamicrosystems
CMOS process and the total layout area is 3900 μm × 5000 μm. The block diagram for
a single channel is shown in Fig. 6.4(b). Each channel consists of a charge sensitive
amplifier with a pole-zero cancellation circuit, a shaper CR-(RC)2 with a peaking time
of 160 ns, two discriminators and two independent 20-bit counters. All stages are DC
coupled.
(a)
(b)
Fig. 6.4. Block diagram of: a) DEDIX chip, b) single channel [152] © 2007, IEEE.
138 Examples of multichannel counting IC
The CSA is based on the folded cascode configuration (see Fig. 6.5) with
a PMOS input transistor M1 with W1/L1 = 600μm/0.45μm. The transconductance of M1
is gm1 = 9.45 mS for the current IDS1 = 0.9 mA and gm1 = 12.17 mS for IDS1 = 1.47 mA.
The feedback loop of the CSA is formed by the capacitor Cf of 100 fF and the channel
conductance of the array of ten PMOS transistors Mf0−Mf9 with Wf/Lf = 0.4μm/10μm
connected in series. The pole-zero cancellation circuit consists of a capacitor Cd = 24 pF
and twenty PMOS transistors Mpz0−Mpz19 of Wpz/Lpz = 0.4μm/10μm connected in
parallel. For negligible detector leakage current, the transistors in CSA feedback and in
PZC circuit are working in the triode region. The gates and sources of the transistors
Mf9 and Mpz0−Mpz19 are connected together. This enables setting the effective CSA
feedback resistance RMf in a wide range (from tens of MΩ to tens of GΩ) ensuring
proper functioning of the PZC circuit even for high rates of input pulses which generate
a DC voltage shift at CSA output.
Fig. 6.5. Scheme of charge sensitive amplifier with PZC circuit [152] © 2007, IEEE.
By the use of the PZC circuit, undershoots can be eliminated after the differentiat-
ing filter stage, which are the results of the long time decay of pulses at the CSA output.
However, this results in DC coupling between CSA input and discriminator inputs. The
offsets propagate from CSA input to discriminator inputs and result in a spread of the
effective discriminator threshold in an ASIC multichannel.
The PZC structure with the tracking system is followed by the second-order filter
(shaper) with a peaking time of 160 ns. The simplified scheme of the filter is shown in
Fig. 6.6. The capacitors Csh1 = 500 fF and Csh2 = 2 pF are based on poly1-poly2 structure,
while for the resistors Rsh1 = 200 kΩ, Rsh2 = 30 kΩ and Rsh3 = 12 kΩ are implemented
using high resistivity poly layer. The OTA stages are based on a folded cascode ampli-
fier with the input transistors of Wsh1/Lsh1 = 88µm/0.45µm and the bias current about
6.5 times smaller than in the CSA stage. The OTA gain is about 1700 V/V. The imple-
mented buffers separate the feedback of OTAs from the next stages.
Examples of multichannel counting IC 139
The shaper signal is fed to two discriminators (described earlier in chapter 3.6.1).
The threshold voltages for two discriminators in a single channel are set independently,
but are common for all 64 channels of the ASIC. As the signal processing chain is
DC-coupled from the CSA input up to the discriminator inputs, a DAC correction
(working in each channel independently) is necessary to minimize the effects of the DC
level spread at the discriminator inputs.
The examples of the X-ray spectra measured with a 64-channel DEDIX chip with
the correction circuit ON and OFF are shown in Fig. 6.7.
(a) (b)
Fig. 6.7. Spectra of Pu-238 radioactive source and Cu Kα line measured with silicon strip detector and
DEDIX IC: a) correction OFF, b) correction ON [152] © 2007, IEEE.
140 Examples of multichannel counting IC
The spectra have been obtained by scanning the discriminator threshold and
measuring the integral distribution of pulse amplitudes. The differences of intensity in
different channels are due to small dimensions of X-ray source (different X-ray intensity
across the strips of the detector). The effective threshold voltage spread calculated to
CSA input is below 7 e− rms.
The DEDIX chip is able to work properly up to the rate of 1 Mcps per channel for
statistically distributed photons from the X-ray tube without gain, offsets and noise deg-
radation. The exemplary results obtained using 8 keV photons from X-ray tube are
shown in Fig. 6.8. The plots show the DEDIX high count rate performance for different
settings of CSA feedback resistance RMf.
(a) (b)
(d)
(c)
Fig. 6.8. High count rate characteristic of DEDIX chip: a) registered counts vs. input counts for different
threshold settings, b) gain, c) DC level shift, d) noise vs. average rate of input pulses [152] © 2007, IEEE.
The chip uses two power supply voltages: 2.2 V (analog parts) and 3 V (discrimi-
nators and digital blocks), and consumes about 5 mW per channel. This IC is mainly
used in diffractometry measurements. The main chip parameters are summarized in
Table 6.3.
Examples of multichannel counting IC 141
Parallelly to multichannel ICs for SSDs used in X-ray imaging applications, sev-
eral interesting solutions of readout electronics for pad and small array of pixel detector
have been developed [67, 133, 135, 265-271]. The power consumption per single chan-
nel in these cases is much higher than for readout electronics of pixel architecture devel-
oped for large array of pixel detector. Exemplary designs are presented below.
The Instrumentation Division from BNL, Upton, USA designed 64-channel ASIC
for high rate photon counting with multiple energy discriminators for the use with pixe-
lated CdZnTe sensor [135]. The IC with the area of 6.65 mm × 6.65 mm was fabricated
in commercial CMOS 0.25 μm 1P5M technology. The main components of single chan-
nels are (see Fig. 6.9): low-noise two stage charge amplifiers, high order shaper (9th)
with baseline stabilizer, five window-discriminators with the same number of 16-bit
counters and 16-bit memory blocks.
The CSA has PMOS input transistors with W/L = 192μm/0.24μm optimized for
2 pF input capacitance. The input transistor with drain current of 200 μA has a trans-
conductance of gm ≈ 3 mS and a gate capacitance of Cg ≈ 400 fF. The CSA feedback is
based on a compensated continuous reset circuit [78] with adjustable gain. The shaper
142 Examples of multichannel counting IC
has one real pole and eight complex conjugate poles with four different settings for
peaking times of 40, 80, 160 and 320 ns. The overall channel gain is adjusted at 250,
500, 750 and 1000 mV/fC. A band gap referenced baseline holder stabilizes the shaper
output for process variation, offsets, temperature and high rate operation [132]. The five
window discriminators have adjustable threshold using five on-chip 10-bit DACs com-
mon to all channels. Additionally, there are five 3-bit trim DACs (with 10 mV step)
working in each channel independently.
The analog output of each channel can be monitored through an on-chip multi-
plexer (common buffer for all channels is used). The data is read out by 16-bit output
bus with a clock up to 60 MHz resulting in the total readout time below 20 μs. The
memory blocks in each channel allow simultaneous measurement as well as readout;
and in this mode dead time is reduced to 20 ns. For floating input at 40 ns peaking time
and gain 1 V/fC an ENC ≈ 140 e− rms is measured. The ENC is also measured for ASIC
connected to CdZnTe arrays of 16 elements (with the size of 2.2×0.7×3 mm3) and using
uncollimated 241Am source. For peaking time of 320 ns, the measured noise is
200 e− rms (with about 70 e− rms contributed from ~1 nA detector leakage current) and
495 e− rms effective (with ballistic deficit) for 40 ns peaking time (without ballistic defi-
cit ENC ≈ 390 e− rms). The average power consumption per channel is about 4.9 mW
(1.1 mW for CSA, 2.8 mW for shaper, 0.45 mW for comparators, 0.36 mW for trim
DACs and the rest for common chip blocks).
Examples of multichannel counting IC 143
CERN_DxCTA IC
Fig. 6.11. Scheme of the CSA and shaper (reprinted from [269] © 2003, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
The single discriminator scheme is shown in Fig. 6.12. The first differential block
converts a single shaper output to a differential signal. A VT2-VT1 threshold applied
differentially produces a DC shift at the output of the differential pair. This DC shift can
be modified in each channel independently by 5-bit trim DAC. The buffered differential
signal is applied to a two-stage comparator with a high DC gain (72dB). The total ASIC
area is 5.0 mm × 8.5 mm.
Fig. 6.12. Discriminator scheme (reprinted from [269] © 2003, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
Examples of multichannel counting IC 145
The measurements with a calibration pulse shows the noise of equivalent noise
charge ENC = 430+56×Cd e− rms and an average gain of 143.5 mV/fC, with channel-to-
channel gain variation of 1.5 %. The comparator offset is 5.2 mV rms and the power
consumption is 2.1 mW/channel. The counting rate measurement has been performed
with CdZnTe detector biased up to 1200 V. The count rate of 5 Mcps has been achieved
for maximum X-ray energy of 140 keV.
CIX 0.2 IC
A new approach for a high rate X-ray imaging system has been proposed in the
CIX (Counting and Integrating X-ray) integrated circuit, which combines the advantages
of both counting and integrating in each pixel [270, 271]. Counting systems can provide
spectral information about the photon flux. For these systems the lowest measurable flux
is a single photon in a measurement interval, while the maximum flux per pixel is lim-
ited to a few Mcps (in the best systems) as a result of pile-up effects. The integral sys-
tems provide information about the total deposited energy. However, they are well
suited for large rates and signal currents. Measurement of small signals is difficult be-
cause of electronic noise. By simultaneous counting and integrating in every pixel, the
dynamic range can be extended. Additionally, average energy of the deposited photon
can be calculated in this flux range where photon counting and integrating overlap (with
increasing pile-up, the correlation between counting and integrating signals get weaker).
Fig. 6.13. Block diagram of a CIX pixel cell [271] © 2007, IEEE.
146 Examples of multichannel counting IC
The CIX 0.2 version IC has been implemented in AMS 0.35 μm technology with
the matrix of 8 × 8 pixels. The pixel dimensions are 250 μm × 500 μm and the block
diagram of the CIX pixel cell is shown in Fig. 6.13. The pixel contains both counting
and integrating channels. The key element of this solution is a configurable feedback
circuit of the CSA which provides continuous reset, leakage current compensation and
produces the input signal replica for the integrator. To reduce the problem of the sub-
strate noise generated by the digital block, the following have been applied [187]: a low
swing differential logic for the design of the counters and a digital circuit.
The simplified scheme of the photon counting block is shown in Fig. 6.14. The
counting channel contains a charge sensitive amplifier with a feedback capacitor of
10 fF, two stage comparator with differential output and 16-bit ripple counter. A con-
tinuous feedback of the CSA is based on a differential pair working as a voltage current
source. Negative input pulse of charge causes the voltage drop on feedback capacitor
CFb. This voltage drop activates the feedback current source. The current source delivers
the IFb current to the input node and discharge the CFb capacitor. In a simple model, the
trb time necessary to return to the baseline depends on the pulse size and it is given as
Qin
t rb = (6.2)
I fb
If the CSA output exceeds the threshold voltage VCountTh, the pulse is registered in the
counter.
The integrator circuit is shown in Fig. 6.15. The feedback circuit is a charge pump
operating in a clock-synchronized mode. The charge pump removes the Qpkt charge
packet of the defined size from the Cint capacitor each time a predefined threshold VIntTh
is passed. Two counters record the number of charge pumps events Npkt and the time Δt
during which the pump events has occurred. The measured ISignal current is given as
Examples of multichannel counting IC 147
N pkt Q pkt
I Signal = (6.3)
Δt
The charge pump clock frequency and the size of charge packets can be tuned.
This allows setting the minimum and the maximum integrator currents.
The core of the feedback circuit consists of two differential pairs (see Fig. 6.16).
The first one provides the shaping of the signal in the counter channel and replicates the
input signal for the integrator. The second one is responsible for detector leakage current
compensation. Both differential pairs work in a similar way: two currents at the bottom
of each branch drain precisely half of the current source above. Any difference in gate
148 Examples of multichannel counting IC
voltages of two transistors in differential pair shifts the current from one branch into the
other. The additional/missing current has to leave/enter the branch through the node
above the respective current drain.
With bump bonded CdZnTe sensor (3 mm thick) the obtained energy resolution
was 0.3 keV. The maximum count rate of 3.3 Mcps (at 10 keV threshold) has been
measured. However, the combined dynamic range of the counter and integrator cover
5 orders of magnitude up to approximately 22.5 Mcps [272]. The double data readout
buffer enables dead time free data acquisition with the frame rates up to 20 kHz. The
power consumption in the CIX 0.2 is about 3.2 mW per pixel and almost all of the
power (98%) is consumed by the digital circuitry.
Probably one of the most demanding solutions of the readout electronics is the one
for large area pixel detectors, because of limited area and the power per single channel,
together with the requirements for good pixel-to-pixel matching and good final yield.
The era of pixel readout electronics started in the nineties of 20th century and the last
technology developments (deep submicron and 3-D technologies) seem to be the most
appropriate for this kind of readout electronics. However, it should be mentioned that
only a few designs published in literature have succeeded in building large area detec-
tors. Some exemplary designs have been presented below.
(b)
(a)
Fig. 6.17. The CSA with active feedback which automatically compensates for detector leakage current a)
simplified circuit (reprinted from [84] © 1991, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ) - the currents in the brackets were
added for small signal analysis, b) equivalent circuit diagram [273] © 2001, IEEE.
Let us perform a small signal analysis of active feedback [273] and assume for the
AC- analysis an input signal iin, input voltage vin and vout = gmvinZ0 at CSA output. Small
signal currents in each branch (as indicated in brackets in Fig. 6.17(a)) can be written as
1
i1a = g m1vout (6.4)
2
1
i1b = − g m1vout (6.5)
2
i f = (vout − vin )sC f ≈ vout sC f (6.6)
i1b
i2 = g m 2 (6.7)
sC
Summing the current at CSA input gives
1 1 g m1 g m 2vout
iin + g m1vout + vout sC f + =0 (6.9)
2 2 sC
150 Examples of multichannel counting IC
vout 2 sC
=− 2 (6.10)
iin 2 s C f C + sCg m1 + g m1 g m 2
The stability criterion requires adequate separation of the poles frequency. As-
suming that the poles are real and widely separated, one obtains
g m2
p1 ≈ − (6.11)
C
g
p2 ≈ − m1 (6.12)
2C f
The stability criterion
C 2C f
>> (6.13)
gm2 g m1
must be satisfied for the maximum value of expected detector leakage current. The
equivalent scheme of the active feedback contains both the Rf ≈ 2/gm1 equivalent feed-
back resistance and the Lf inductance of the value (see Fig. 6.17(b)).
2C
Lf = (6.14)
g m1 g m 2
The detail analysis of the Krummenacher active feedback can be found in [274].
The shaper stages in single ended and differential configurations proposed by
Krummenacher are shown in Fig. 6.18. The transfer function of a single-ended stage is
given
g mi Z D
Kv ≈ − (6.15)
1 + g mi Z S
where
1
ZD = (6.16)
g mo + sC LP
Examples of multichannel counting IC 151
1
ZS = (6.17)
sCHP
After substitution of ZD and ZS to eq. (6.15) the transfer function of the shaper is
g mi
s
C LP
Kv ≈ − (6.18)
⎛ g ⎞⎛ g ⎞
⎜⎜ s + mi ⎟⎟⎜⎜ s + mo ⎟⎟
⎝ C HP ⎠⎝ C LP ⎠
The high pass characteristics is determined by time constant τHP = CHP/gmi and low pass
characteristics by time constant τLP = CLP/gmo.
(a) (b)
Fig. 6.18. Shaper amplifier: (a) single-ended and (b) differential circuit (reprinted from [84] © 1991, with
permission from Elsevier, http://www.sciencedirect.com/science/journal/ ).
Digital Pixel Array Detector (DPAD) has been designed for static and time re-
solve protein crystallography [61, 275, 276]. The detector has a pixel pitch of 150 μm.
Two prototype readout ASICs with pixel array of 8×8 and 16×16 have been designed in
HP 0.5 μm 3-metal layer process. The simplified scheme of a single readout pixel is
shown in Fig. 6.19. It consists of a preamplifier with a DC adjustable reset, shaper with
a peaking time controlled in the range of 50 ns − 150 ns, a differential comparator and
a 3-bit prescaler. The prescaler divides the event rate by 8 and reduces bandwidth re-
152 Examples of multichannel counting IC
quirements.When a pixel has accumulated 8 counts then an additional overflow bit starts
the readout logic sequence to generate the pixel address in less than 80 ns.
(a) (b)
Fig. 6.20. Simplified scheme of: (a) integrator, (b) shaper [61] © 1996, IEEE.
The circuit gain is 690 mV/fC and the measured noise is 60 e− rms (@ 5.9 keV)
when a Si pixel detector (Cdet = 0.3 pF) is used. The single channel can accommodate up
to 1 Mcps, while the power consumption in analog part is 50 μW/channel (VSS = − 3 V).
Examples of multichannel counting IC 153
The measurements of 8×8 chips show the shaper's output voltage spread of about
σ = 7 mV (from pixel to pixel).
MPEC 2.3 IC
MPEC (Multi Picture Element Counters) ICs family started with the first IC [277]
by adopting BIER&PASTIS chip architecture developed for the ATLAS pixel detector
at the LHC at CERN [146]. The latest version of the MPEC 2.3 contains an array of
32 × 32 pixels with a pitch of 200 μm × 200 μm [278, 279]. Additional blocks are input
and output buffers and six 8-bit DACs to generate internal bias currents. The MPEC 2.3
IC works in a single photon counting mode with energy windowing and operates up to
1 MHz rate of input pulses per single channel. The chip was designed in AMS 0.8 μm
CMOS process with only two metal layers.
Fig. 6.21. Schematic of one pixel of MPEC 2.1/2.3 chip [279] © 2004, IEEE.
The main parts of a single pixel are: the CSA with a current feedback, two inde-
pendent discriminators, two 18-bit counters and a dedicated window logic which allows
the counters to record only photons within a set energy region (see Fig. 6.21). The front-
end and the feedback circuit are designed for positive input charges. The discriminators
are AC-coupled to the CSA to eliminate DC voltage offsets propagation. The discrimi-
nator thresholds are set globally. Additionally, a correction voltage is stored on a capaci-
tor independently for each discriminator (see Fig. 6.22(a)). The correction voltage stored
dynamically on a capacitor must be periodically refreshed due to leakage current
(mainly into the bulk of a switch transistor). The simple circuit with additional buffer
(between store node and bulk of switch transistor) reduces the leakage current and
minimizes the necessary refresh rate (see Fig. 6.22(b)). In order to save the silicon area
the counters are realized as pseudorandom counters (properly connected feedback in the
shift register - Fig. 6.23(a)) and a single phase flip-flop contains only two inverters and
two transistors (see Fig. 6.23(b)) [280].
154 Examples of multichannel counting IC
(b)
(a)
Fig. 6.22. Correction of discriminator offset: a) discriminator with threshold adjustment, b) threshold drift
compensation circuit (reprinted from [280] © 2001, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
(b)
(a)
Fig. 6.23.Simplified scheme of counter architecture: a) 18-bit linear feedback shift register counter,
b) single phase flip-flop cell of the shift counter [280].
The test of MPEC 2.3 has been performed both with Si and CdTe sensors, how-
ever in the second case, the hole trapping and low hole mobility reduce the imaging
performance. The measured ENC without sensor is 60 e− rms and about 110 e− rms
when connected to the sensor. The minimum threshold can be set only to 1300 e−, be-
cause of digital switching noise in the chip (the design uses the process with only two
metal layers and an effective shielding poses difficulty in this case). The threshold dis-
persion before the correction reaches 180 e− rms and after adjustment it is reduced to
10 e− rms. Multichip modules with 4 single photon counting MPEC 2.3 chips bump
bonded to Si and CdTe detectors have been successfully built and operated.
MEDIPIX2 IC
The MEDIPIX family has been developed at CERN [281, 282] and the first
MEDIPIX1 chip contains the matrix of 64 × 64 pixels and 170 μm × 170 μm in size.
It has been designed in the 1 μm SACMOS process. The next generation of MEDIPIX2
[44] IC has been designed in the 6-metal CMOS 0.25 μm process. Its readout electronics
Examples of multichannel counting IC 155
consists of an array of 256 × 256 readout channels which gives the matrix of 65536
identical elements in total, with the same geometry as the pixel detector. Each readout
pixel with 500 transistors occupies the area of 55 μm × 55 μm and has a static power
consumption about of 8 μW. A bump bonding technique is used to connect the detector
to the 20 μm wide octagonal input pads of the readout electronics. The architecture of
the single cell is shown in Fig. 6.24.
The floor plan of MEDIPIX2 chip is shown in Fig. 6.25. Its total area is
14.1 mm x 16.1 mm, and the sensitive matrix area of pixels covers 87% of the entire
chip (i.e. 1.98 cm2). The periphery area, which contains fast shift registers, DACs, I/O
control logic and LVDS drivers and receivers, is located at the bottom part of the chip.
Such a floor plan enables minimization of the dead area in multichip module.
Fig. 6.25. Schematic floor plan of MEDIPIX2 chip [44] © 2007, IEEE.
When pixels are read out or their registers are loaded, the whole pixel matrix is
organized in 256 columns. The readout of the chip can be performed in a serial mode
using fast LVDS logic or in parallel via 32-bit CMOS bus. When the 100 MHz clock is
used, the first mode of the readout takes less than 9 ms, while in the second mode it is
done in 266 μs. The whole chip contains about 33 million transistors.
The MEDIPIX2 is very attractive for different applications and it is used in many
scientific experiments [282]. The RELAXD project (high REsolution Large Area X-ray
Detectors) aims to develop a four-side tillable photon-counting module with minimum
dead spaces [283]. These modules will be used to build an arbitrary large-area detector.
Other projects based on MEDIPIX2 IC, which aim to add new functionality or increase
chip performance, have been developed. These include: TIMEPIX chip [284] and com-
pletely new MEDIPIX3 chip [285].
Examples of multichannel counting IC 157
TIMEPIX IC
The TIMEPIX chip [284] contains an array of 256 × 256 pixels with the size of
55 μm × 55 μm. Each pixel can be used independently for arrival time, as well as for
energy and/or photon counting measurements. The chip has been designed in a commer-
cial 6 metal CMOS 0.25 μm technology. The floorplan and readout architecture are the
same as the MEDIPIX2.
The scheme of a single cell is shown in Fig. 6.26. Comparing with the MEDIPIX2
cell, there is only a single threshold with 4-bits threshold adjustment; each pixel can be
configured in three different operation modes and the counting clock is synchronized
with the external Ref_Clk clock reference.
Fig. 6.26. TIMEPIX pixel cell schematic (reprinted from [284] © 2007, with permission from Elsevier,
http://www.sciencedirect.com/science/journal/ ).
The analog part of the pixel contains a preamplifier based on the cascode differen-
tial amplifier and Krummenacher feedback [84]. The preamplifier gain for nominal bias
condition is ~16.5 mV/ke− with a linear dynamic voltage range up to ~50 ke−. The peak-
ing time can be set from 90 ns to 180 ns by the Preamp DAC, while the return to zero
baseline of ~10 ke− input charge can be adjusted from 500 ns to 2500 ns depending on
Ikrum DAC setting. The preamplifier is DC coupled to the discriminator. The discrimi-
nator has a differential OTA at the input, 4-bits trim DAC and its core is based on cur-
rent discriminator with hysteresis.
The digital part of the pixel contains TIMEPIX Synchronization Logic (TSL), the
14-bit shift register, the overflow control logic, the Ref_Clk pixel buffer and 8-bit Pixel
Configuration Register (PCR). The TIMEPIX uses an external clock with the frequency
of up to 100 MHz as a time reference. Depending on the Shutter signal, the shift register
is used to shift the data from pixel to pixel (Shutter = 1) or works as a counter (with
a XOR tap) with a dynamic range of 11810 counts (Shutter = 0). During the data acqui-
158 Examples of multichannel counting IC
sition the pixel counter is incremented by the Ref_Clk depending of the operation mode
bits (P0 and P1):
− event counting mode (P0 = 0, P1 = 1) - each hit above the threshold increments the
counter by 1,
− time-over-threshold (ToT) mode (P1 = 1, P1 = 0) - the counter is incremented con-
tinuously while the preamplifier output signal is over the threshold,
− arrival time mode (P1 = 1, P1 = 1) - the counter is incremented from the first time the
discriminator goes high to the closing of the Shutter.
The measured pixel noise is ~100 e− rms and the threshold spread after trimming
is ~35 e− rms. The chip operates for both input charge polarities with the minimum de-
tectable charge of ~650 e−. In the ToT mode, the energy resolution (ΔToT/ToT) is better
than 5% if the input charge is ≥ 1k e− above the threshold. The measured time walk is
≤ 50 ns. Using a 100 MHz clock, the chip can be read out serially (via on-chip LVDS
drivers) in less than 10 ms, or in parallel (32-bit CMOS port) in less than 300 μs. The
TIMEPIX chip operates with 2.2 V power supply. Power consumption for the analog
part is ~440 mW and ~450 mW for the digital part.
MEDIPIX3 IC
The MEDIPIX3 chip [285, 286] has been designed to eliminate the spectral dis-
tortion produced by the charge diffusion in highly segmented pixel detector (see
Fig. 6.27). The effects strongly influence the measured energy spectrum as the pixel
pitch is decreased with respects to the thickness of the detector material.
(a) (b)
Fig. 6.27. Charge diffusion effect in highly segmented detector - charge generated at the pixel border is
shared between pixels [286].
The chip has been developed in 8-metal CMOS 0.13 μm process and its core con-
tains a matrix of 256 × 256 pixels, which is 55 μm × 55 μm in size. In a new architec-
ture proposed in MEDIPIX3, the pixels communicate with one another. At the corner of
each pixel, summing circuits add the total charge deposited in each sub-group of 4 pix-
Examples of multichannel counting IC 159
els. An arbiter circuit assigns the hit to the summing circuit with the highest charge. The
chip is highly configurable and can operate with a single pixel mode or charge summing
mode with the effective pixel size of 55 μm × 55 μm or 110 μm × 110 μm.
Any single pixel (see Fig. 6.28) contains a charge sensitive preamplifier, shaper,
two discriminators with 5-bit threshold adjustment, pixel memory (13-bits), arbitration
logic for charge allocation, control logic and configurable counter (about 1600 transis-
tors per pixel). At the bottom of the pixel matrix there are peripheral circuits: LVDS
drivers and receivers, bandgap reference, 25 DACs (10 9-bit and 15 8-bit), 32 e-fuse
bits, EoC, 2 test pulse generators per pixel column, temperature sensor, full IO logic and
command decoder.
PILATUS IC
The fast readout electronics for pixel detector has been developed for the X-ray
measurements at the beamlime of the Swiss Light Source (SLS), as part of the PILA-
TUS project (PIxeL ApparaTUs for the SLS) [46, 153, 153, 287, 288]. The goal of this
project was to build a hybrid pixel system covering approximately the area of
40 × 40 cm2 with 2000 × 2000 pixels [153]. The first generation PILATUS I chip was
designed in 2000 at the Paul Scherrer Institute (PSI), Villigen, Switzerland, using
DMILL radiation tolerant CMOS process (Atmel Temic SA, Nantes, France).
PILATUS I contains an array of 44 × 78 = 3432 pixels with a pixel size of
217 μm × 217 μm. The active area spans over 10 × 17 mm2. Each pixel contains a low
noise CSA, a single-level comparator with a 4-bit individual threshold adjustment and
15-bit shift register counter (see Fig. 6.29). The noise of bump-bonded chip is
75 e− rms. The total power consumption is 100 μW per pixel. The lowest achievable
threshold of the chip is about 3 keV, which allows to measure 22Ti Kα radiation
(4.5 keV).
Fig. 6.29. Schematic view of the pixel unit cell [46] © 2006, IUCr Journals, http://journals.iucr.org/ .
To improve the yield of good pixels the chip has been redesigned. The second
version of the PILATUS II chip was designed in 2004 at the PSI, using the UMC
0.25 μm technology in which radiation tolerance was achieved by design [288]. The
active chip area is 10 × 17 mm2 and it consists of an array of 60 × 97 pixels. Each pixel
has the size of 172 μm × 172 μm. A single pixel contains similar blocks as the first ver-
sion, but a comparator has a 6-bit individual threshold adjustment and the counter capac-
ity is 20 bit. The authors claim the count rate up to ∼1.5 MHz/pixel/s. The chips
mounted in the PILATUS module (an array of 8 × 2 chips) are read out parallelly within
the readout time of about 2 ms. Using PILATUS modules, PILATUS 6M detector sys-
tem has been built. This system is composed of 5 × 12 modules with 2463 × 2527 pixels
and has a total active area of 424 × 435 mm2.
XPAD3 IC
XPAD IC family has been developed at CPPM in France [45, 289]. The latest
XPAD3 version has been realized in 0.25 μm IBM technology with the pixel matrix of
80 × 120 elements of single pixel size of 130 μm × 130 μm. The chip is in two versions:
− XPAD3-S (S - as in Si) version works with positive input charge generated by photons
in the energy range from 4 keV to 40 keV and has a single threshold,
− XPAD3-C (C - as in CdTe) version works with negative input charge generated by
photons in the energy range from 6 keV to 60 keV and has two independent thresholds.
The architecture of XPAD-C is shown in Fig. 6.30. The pixel contains a CSA, an
operational transconductance amplifier and two independent working current low and
high discriminators with 5-bit and 2-bit DACs trim respectively. The XPAD3-C has an
Examples of multichannel counting IC 163
active OTA feedback while the XPAD-S uses in CSA feedback a single MOS device
operated in triode region. The active OTA feedback allows proper XPAD3-S operation
with negative input charge, however, it results in an increase of noise and output offset.
The summary of measured XPAD3 parameters is presented in Table 6.6. The sin-
gle pixel dissipates 40 μW at 2 V power supply. The ENC is 127 e− rms for XPAD3-S
and after the adjustment (6-bit trim DAC) the effective threshold spread is 57 e− rms.
For XPAD3-C the noise is higher (ENC =185 e− rms), because of the OTA feedback
being used in the CSA stage. The XPAD3 architecture allows for the in-flight data read-
out every 2 ms/frame (9600 pixels with 12-bit register each) which results in up to
500 frames per second.
Table 6.6. Characterization results of XPAD3 circuits [45] © 2007, IEEE.
Version XPAD3-S XPAD3_C
Number of pixels 80 × 120 = 9600
Pixel size 130 μm × 130 μm
Expected readout time 2 ms/frame
Counting rate > 106 photons/pixel/s
On the fly readout Yes
Power consumption 40 μW/pixel (2V)
Input polarity Holes collections Electrons collections
Gain 89 nA/keV to 35 keV 46 nA/keV to 60 keV
Selectivity mode Single threshold Double threshold
Non linearity < 4% over 35 keV < 2.5% over 60 keV
Minimum threshold < 4 keV < 10 keV
Global electronic noise 127 e− rms 185 e− rms
Threshold adjustment dispersion 57 e− -
Threshold adjustment resolution 0.2 keV rms -
The XPAD3 chips operate with Si, CdTe and GaAs detectors. A multichip mod-
ule (up to eight XPAD3) has been constructed. A PIXSCAN project aimed to build
a small animal computed tomography scanner demonstrator based on the XPAD3 chip
with an active area of 7.5 cm × 12 cm [290].
The PSI-SLS detector group has also developed a new integrated circuit for high
frame rate X-ray applications called EIGER [291]. The main features of the readout chip
are presented in Table 6.7. The EIGER chip is a completely new design when compared
to the PILATUS II: the pixel size is only 75 μm × 75 μm, the pixel matrix is 256 × 256,
the chip has double buffering and very high readout speed.
The pixel architecture is presented in Fig. 6.31. The low noise preamplifier is fol-
lowed by shaper with tunable shaping time. The discriminator has 6-bit DAC trimming.
A maximum count rate per pixel is more than 1 MHz. During the exposure time the
comparator pulse increments the digital counter by 1. At the end of the exposure time,
164 Examples of multichannel counting IC
the counter content is loaded to the temporary pixel buffer and the counter is reset. Buff-
ering and counter reset last about 1 μs after which new exposure is possible.
Table 6.7. Main features of the new chip and of its pixel (reprinted from [291] © 2010, with permission
from Elsevier, http://www.sciencedirect.com/science/journal/ ).
Readout chip features
Technology UMC 0.25 μm
Power supplies 1.1V (analog), 2V (digital), 1.8V(I/O)
Radtion tolerance Rad-tolerant design (> 44 Mrad)
Pixel array 256 × 256 = 65 536
Chip size 19.3 mm × 20 mm
Readout speed up to 24 k frames/s
Pixel features
Pixel size 75 μm × 75 μm
Gain 44.6 μV/ e−
Peaking time 31 ns
Return to 0 at 1% 151 ns
Noise (simulation) 135 e− rms
Static power 8.8 μW/pixel
Transistor count 430/pixel
Pixel counter Configurable (4, 8, 12-bit mode), binary,
double buffered for continuous readout
Threshold adjust 6-bit DAC/pixel
Other features Overflow control, single pixel addressing
and analog out for testing
Simulations are done with "standard" settings. "Low noise" or "high speed" settings can improve perform-
ance for applications with specific needs.
Fig. 6.31. Pixel architecture of EIGER IC for high frame rate X-ray applications (reprinted from [291]
© 2010, with permission from Elsevier, http://www.sciencedirect.com/science/journal/ ).
The chip can be read out via 32-line parallel bus line with 100 MHz Double Data
Rate (DDR). The double data storage allows minimizing the dead time. The maximum
frame rate can be adjusted using a selectable length of the pixel counter from 4 bits to
Examples of multichannel counting IC 165
12 bits. The maximum frame rates for 4, 8 and 12 bit modes are 24, 12 and
8 k frames/sec respectively. The authors are planning to build 8-chips module bump-
bonded to a silicon sensor of 78 mm × 39 mm with about 0.5 Mpixel in total. Several
modules (up to 18 modules) can be tiled to build a large area detector (9 Mpixel,
~550 cm2).
PX90 IC
The PX90 IC is the first implementation of the readout electronics for hybrid
pixel detector in the CMOS 90 nm technology node [87, 292]. The prototype has been
built to investigate the possibilities of new deep submicron technologies and also to
check such critical parameters for front-end readout electronics like noise, matching and
crosstalk. The aim of the project was to build a readout electronics with a pixel of rela-
tively small area, with high count rate and a readout architecture which allowed a dead
time free operation.
Fig. 6.32. Simplified scheme of pixel cell in PX90 IC [87] © 2010, IEEE.
(a)
(b)
Fig. 6.33. Simplified scheme of PX90 pixel blocks: (a) CSA, (b) AMPII [87] © 2010, IEEE.
The core of the charge sensitive amplifier is based on the folded cascode configu-
ration (M1−M4 transistors - see Fig. 6.33(a)). The CSA M1 input transistor is PMOS
one of W/L = 20μm/0.2μm and operates with the nominal drain current of 5.2 μA. The
designers do not use at the CSA input MOS transistors with the minimum channel
length to omit the possible noise increase connected with the short channel effect and
because of the low output drain-source resistance of the transistor with the minimum
gate length. The feedback loop of the CSA consists of a capacitor Cf = 5 fF and the
Krummenacher feedback circuit [10]. For IKrum = 21.5 nA the simulated peaking time is
tp = 27 ns with pulse width of t0.01 = 365 ns, while for IKrum = 36.5 nA these parameters
are tp = 26 ns with pulse width of t0.01 = 216 ns. To reduce power consumption, the left
branch of the folded cascode operates with power supply voltage of Vddm = 0.8 V,
while the right one with Vdda = 1.2 V.
The second AMPII stage is a fully differential stage (see Fig. 6.33(b) and has
three main functions:
− add some voltage gain,
− reduce (by digital trimming) offset spread in the front-end electronics,
− set the threshold level for the discriminator stage.
The AMPII consists of two source followers (with control bias current) at the in-
put, a simple differential amplifier and two source followers at the output. The fully
differential architecture of this stage allows the operation both, with positive and nega-
tive pulses. The offset spread in the signal processing chain is compensated by the input
source followers of the AMPII (M21−M22 transistors), because the I1 and I2 currents
can differ according to the setting of the DAC trim. In this design currents are I1 = Ibase
and I2 = Ibase+Itrim, where Ibase = 0.5 μA and Itrim is controlled by the DAC trim. By apply-
ing two different voltages Vt and Vtr (see Fig. 6.33(b)), the effective threshold level,
which equals VTH =Vt−Vtr , is set for the next discriminator stage. The differential output
Examples of multichannel counting IC 167
signals from the AMPII are fed into the current discriminator (discussed in chapter
3.6.1.).
Discriminator pulses are counted by two 16-bit counters. A single counter unit is
presented in Fig. 6.34. Each unit contains a 16-bit binary counter, which can be con-
verted into a shift register. In the counter mode, the unit input is fed with discriminator
pulses. In the shift mode, the unit reads the data from the output of the previous unit and
provides the data for the next unit. The units are grouped according to the thresholds,
e.g. the units counting low threshold discriminator pulses are connected and so are the
units counting high threshold discriminator pulses. In this way, the data for both thresh-
olds can be read out independently. Therefore, for the same setting of the thresholds, the
integrated circuit can operate in the continuous readout mode. Additionally, there is an
8-bit latch in each unit which provides the data for the trimming DACs. The latch is
parallelly loaded from the register.
The ASIC is designed in the 90 nm TSMC CMOS process and its total area is
4 mm × 4 mm. Each pixel contains about 1800 transistors and measures 100 μm by
100 μm. 2/3 of the pixel area is occupied by analog blocks and 1/3 by the digital blocks
(see Fig. 6.35(a)). The cross-section of the pixel cell showing the distribution of all
metal layers is shown in Fig. 6.35(b). The metal layers M1 to M3 are used for routing
inside the pixel and distribution of control signals. The metal layers M4 to M9 are used
for the distribution of power supplies and shielding of the IC. The chip has been manu-
factured with the use of the Multi-Project-Wafer (MPW) run. In order to enable tests
with pixel detectors, each readout pixel is provided with a large pad of 60 μm × 60 μm
for stud bump bonding. The large area of input pads allows the effective stud bump
bonding (and tests of a small prototype with a detector), however this results in rela-
tively high parasitic capacitance at the input of the CSA - in this case, 230 fF.
To reduce the effects of the injecting substrate noise the following steps have
been applied:
− all NMOS transistors in the analog block are shielded with a Deep N-Well (DNW)
layer and the analog ground has a separate wire-bonding pad from digital ground,
168 Examples of multichannel counting IC
− PMOS transistors in analog and digital blocks have separate N-Well contacts con-
nected to appropriate positive supply lines (except for some floating N-Wells in analog
blocks),
− guard rings and decoupling capacitors are implemented according the rules suggested
in [184].
(a) (b)
Fig. 6.35. Layout of a single pixel: (a) 1 - CSA & CSAREF, 2 - second stage amplifiers, 3 - discriminators,
4 - trim DACs, 5 - reference blocks, 6 - counters (MET4-MET9 are removed for the better visibility),
(b) cross-section with metal M1-M9 layers and NW and DNW layers [87] © 2010, IEEE.
For nominal bias conditions, the power consumption is 47 μW per pixel. The
digital blocks, i.e. the shift register tests and read/write counters tests work without er-
rors up to 200 MHz clock (limitation in the test set-up). The mean gain measured on
different modules is 28 μV/e−. The offset spread for the threshold of the IC before cor-
rection is about 35 mV (on one sigma level) and after correction it is reduced to 1.8 mV
(calculated to the input is equal to 64 e− rms). For the pixel which is not connected to
the detector, the noise is ENC = 204 e− rms. For the PX90 IC with a detector, the noise
is ENC = 240 e− rms. According to the simulation the total detector capacitance is about
50 fF, while the parasitic capacitance of an input pad is approx. 230 fF.
Using the 200 Mbps during the data transfer via single LVDS data output results
in about 218 μs readout time. A dead time free readout is possible with the PX90 IC
using the continuous readout mode, however the observed effective noise of the system
increases by about 15 %. In order to characterize the count rate performance of the
PX90 IC, certain tests were carried out with signals arriving randomly using photons
from rotating anode high power X-ray generator. The count rate of 2 Mcps/pixel was
obtained which resulted in 200 Mcps/mm2. The PX90 IC performance is summarized in
Table 6.8.
Examples of multichannel counting IC 169
VIPIC IC
For a few years the Fermi National Accelerator Laboratory (FNAL) has success-
fully tested the possibility of using 3D-IC technology for pixel readout electronics both
for high energy physics and related fields [293]. In 2009 the Fermilab organized a Mu-
lit-Project-Wafer using 3D-IC technology provided by Tezzaron Semiconductor. The
wafers were fabricated in a commercial 0.13 μm bulk CMOS provided by Chartered
Semiconductor. This CMOS process uses 6 Cu metal layers per wafer. The 3D integra-
tion is done by Tezzaron Semiconductor using wafer with TSVs (Through Silicon Vias)
added after completion of the Front-End-Of Line (FEOL) part of the process. TSVs are
1.3 μm in diameter, 6 μm deep and 3.8 μm minimum spacing is required. The Tezzaron
demonstrated the possibility of stacking up to five layers, however during this MPW run
only two wafers were stacking face to face.
An example of a readout chip is a VIPIC IC (Vertically Integrated Pixel Imaging
Chip) designed for X-ray Photon Correlation Spectroscopy (XPCS) experiments by
FNAL in collaboration with AGH UST [294]. A VIPIC chip is a prototype matrix with
64 × 64 pixels with 80 μm × 80 μm pixel size and consists of two layers: analog and
digital. The simplified scheme of analog pixel cell is shown in Fig. 6.36.
170 Examples of multichannel counting IC
Fig. 6.36. Simplified scheme of an analog pixel cell of the VIPIC chip [294].
The single pixel cell consists of two (CSA and CSA_REF) charge sensitive am-
plifiers, two stages of AMP I and AMPII amplifiers (they are AC coupled to cancel the
offset propagation and shaping a signal) and single DISCR current discriminator with
differential threshold setting. The CSA feedback contains an 8 fF capacitor
(MET4−MET5 structure) and a simple MOS transistor working in the linear region. The
differential or single-ended operation of front-end is possible to test digital crosstalk in
the chip. There are two trim DACs: 7-bit DAC for threshold correction and 3-bit DAC
for trimming CSA feedback time constant individually in each pixel. The simulated gain
is 52 μV/e− and noise ENC < 150 e− rms (with Cdet = 100 fF) and peaking time
tp < 250 ns. The power consumption in the analog part is 25 μW/pixel.
The chip is designed to yield 10 μs frame readout time at the mean occupancy of
3.8×108 photons/cm2/s. The digital layer of the VIPIC IC is divided into 16 readout
groups of pixels readout in parallel via separate serial ports with nominal frequency of
100 MHz clock using the LVDS standard (see Fig. 6.37).
Examples of multichannel counting IC 171
Fig. 6.37 Block diagram of a digital tier of the VIPIC chip [293] © 2009, IEEE.
The readout within each group is zero-suppressed. The sparsification scheme (ad-
dresses of hit pixels only) allows a dead-time free readout. The sparsification circuitry is
similar to the idea proposed in the MEPHISTO chip [295]. Each pixel in a digital tier
additionally contains a 5-bit counter. The readout is binary and each hit is represented
by a 16-bit long word (3 bits - start signs, 5 bits - content of in pixel counter, 8 bits -
pixel addresses in the sparsification mode). In imaging mode the last 8-bit of pixel ad-
dress is not required. The main features of the VIPIC IC are summarized in Table 6.9.
Table 6.9. Main features of the VIPIC IC [293],© 2009, IEEE.
Feature Comments
X-ray detection (8 keV) with Si pixel detector XPCS application
64×64 pixels, pixel area: 80 μm × 80 μm active area: 5120 μm × 5120 μm
Separate analog and digital tiers
Transistor number/pixel analog = 280 tran., digital=1400 tran.
Chip area 6.3 mm ×5.5 mm (6.3 mm ×5.5 mm) larger dicing for DBI
Single threshold for discriminator
Two trim DACs/pixel 3-bit CSA feedback and 7-bit for offset correc-
tions
Test charge injection circuitry Cinj = 1.7 fF
Front-end channel architecture single ended or differential configuration
Power consumption 25 μW/analog pixel
172 Examples of multichannel counting IC
The project aims to build a 4-side buttable pixel detector tile (see Fig. 6.38). The
detector will be fabricated using the whole available area of a wafer and the chips will
be tailed on the detector wafer by fusion bonding.
Fig. 6.38. Cross-section of a 4-side buttable pixel detector tile [293] © 2009, IEEE.
References 173
REFERENCES
[22] A. Fidanizo, G. Stimato, L. Azario, A. Piermattei: " Investigation of natural diamond detec-
tor priming effect during electron beam irradiation." Nucl. Instr. Meth. A vol. 245, 2006,
p. 421−426.
[23] S. M. Sze: "Semiconductor devices - physics and technology." 2nd edn, Wiley, New York,
2000.
[24] P. R. Gray, P. J. Hurst, S.H. Lewis, R.G. Meyer: "Analysis and design of analog integrated
circuits." 4th edn, Wiley, New York, 2001.
[25] G. F. Knoll: "Radiation detection and measurements", (3rd edn), Wiley, New York, 2007.
[26] U. Fano: "Ionization yield of radiations. II. The fluctuations of the number of ions." Phys.
Rev., vol. 72, 1947, p. 26−29.
[27] M. Krumrey, E. Tegeler, R., G. Ulm: "Complete characterization of a Si(Li) detector in the
photon energy range 0.9 - 5 keV." Rev. Sci. Instrum., vol. 60, 1989, p. 2287−2290.
[28] P. Lechner, R. Hartmann, H. Soltau, L. Struder: "Pair creation energy and Fano factor of si-
licon in the energy range of soft X-rays." Nucl. Instr. Meth. A vol. 377, 1996, p. 206−208.
[29] A. Owens, G.W. Fraser, K.J. McCarthy: "On the experimental determination of the Fano
factor in Si at soft X-ray wavelengths." Nucl. Instr. Meth. A vol. 491, 2002, p. 437−443.
[30] The Particle Data Group:" The Review of Particle Physics." Eur. Phys. J. C, vol. 15, 2000,
p. 1−878.
[31] M. L. Wood, et al.: "Charge correlation measurements of double-sided direct-coupled sili-
con microstrip detectors." Proc. 3rd Annual International Industrial Symposium on the Su-
per Colider, Atlanta, Georgia, March, 1991, p. 903−926.
[32] E. Gatti, P.F. Manfredi: "Processing the signals from solid-state detectors in elementary-
particle physics." Rivista del Nuovo Cimento, vol. 9, no. 1, 1981, p. 1−146.
[33] L. Tlustos, M. Campbell, E. Heijne, X. Llopart: "Signal variation in high granularity Si
pixel detectors." IEEE Trans. Electron Dev., vol. 50, no. 1, 2003, p. 26−31.
[34] W. Dąbrowski, P. Gryboś: "Position sensitive semiconductor strip detectors." in X-ray
Spectrometry: Recent Technological Advanced. edited by K. Tsuji and J. Injuk and R. Van
Grieken, 2004, John Wiley & Sons, Ltd, p. 247−275.
[35] S. Ramo: "Currents induced by electron motion." Proc. IRE, vol. 27, 1939, p. 584−585.
[36] W. Dąbrowski, P. Gryboś, M. Idzik: "Study of spatial resolution and efficiency of silicon
strip detectors with different readout schemes." Nucl. Instr. Meth. A, vol. 356, 1995,
p. 241−254.
[37] K. Korbel: "Układy elektroniki front-end." Uczelniane Wydawnictwa Naukowo-
Dydaktyczne AGH, Kraków, Poland 2000.
[38] J. Marczewski, et al.: "Charge sharing modeling in pixel detector with capacitive charge
division." IEEE, Trans. Nucl. Sci., vol. 51, no. 6, 2004, p. 3006−3012.
[39] W. Dąbrowski, P. Gryboś, M. Idzik: "Study of spatial resolution and efficiency of silicon
strip detectors with charge division readout." Nucl. Instr. Meth. A, vol. 383, 1996,
p. 137−143.
[40] Hilt, B., P. Fessler, and G Prévot : "The quantum X-ray radiology apparatus." Nucl. Instr.
Meth. A vol. 442, 2000, p. 355−359.
[41] C. Avila, (P. Gryboś), et al.: "Contrast cancellation technique applied to digital X-ray imag-
ing using silicon strip detector." Med. Phys., vol. 32 (12), December 2005, p. 3755−3766.
[42] Overdick, M., et al.: "A Bioscope system using double-sided silicon strip detectors and
self-triggering readout chips." Nucl. Instr. and Meth. A vol. 392, 1997, p. 173−177.
[43] P. Weilhammer, et. al.: "Si pad detectors." Nucl. Instr. and Meth. A vol. 383, 1996.
p. 89−97.
References 175
[82] L. Blanquart, A. Mekkaoui, V. Bonzom, P. Delpierre: "Pixel analog cells prototypes for
ATLAS in DMILL technology." Nucl. Instr. and Meth. A, vol. 395, 1997, p. 313−317.
[83] D. C. Christian, et al.: "Development of a pixel readout chip for BTeV." Nucl. Instr. and
Meth. A, vol. 435, 1999, p. 144−152.
[84] F. Krummenacher: "Pixel detectors with local intelligence: an IC designer point of view."
Nucl. Instr. and Meth. A, vol. 305, 1991, p. 527−532.
[85] J. Vandenbussche, L. Leyn, G. Van der Plas, G. Gielen, W. Sansen: "A fully integrated
low-power MOS particle detector front-end for space applications." IEEE Trans. Nucl. Sci.,
vol. 45, no. 4, 1998, p. 2272−2278.
[86] P. Gryboś, et. al.: "RX64DTH - a fully integrated 64-channel ASIC for a digital X-ray im-
aging system with energy window selection." IEEE Trans. Nucl. Sci., vol. 52, no. 4, 2005,
p. 839−46.
[87] R. Szczygieł, P. Gryboś, P. Maj: "A prototype pixel readout IC for high count rate X-ray
imaging systems in 90 nm CMOS technology." IEEE Trans. Nucl. Sci., vol. 57, no. 3,
2010, p. 1164−1174.
[88] E. Kowalski: "Nuclear electronics." Springer, Verlag, New York, 1970.
[89] V. Radeka: "Optimum signal-processing for pulse-amplitude spectrometry in the presence
of high-rate effects and noise." IEEE Trans. Nucl. Sci., vol. 15, no. 3, 1968, p. 455−470.
[90] F. S. Goulding: "Signal-processing for semiconductor detector." IEEE Trans. Nucl. Sci.,
vol. 29, no. 3, 1982, p. 1125−1141.
[91] Z. Y. Chang, W. Sansen: "Low-noise, wide-band amplifiers in bipolar and CMOS tech-
nologies." Kluwer Academic, Boston, 1990.
[92] K. Korbel, W. Dąbrowski: "Filtracja sygnału w spektrometrycznym torze pomiarowym.
Filtry analogowe." Skrypty Uczelniane Nr. 1318, AGH, Kraków, Poland, 1992.
[93] S. Ohkawa, M. Yoshizawa, K. Husimi: "Direct synthesis of the Gaussian filter for nuclear
pulse amplifiers." Nucl. Instr. and Meth., vol. 138, 1976, p. 85−92.
[94] International Standard. "Nuclear Instrumentation – Amplifiers and preamplifiers used with
detectors of ionizing radiation – Test Procedures." Document CEI IEC 61151, IEC 1992.
[95] IEEE Standard. "Test Procedures for Amplifiers and Preamplifiers used with Detectors of
Ionizing Radiation." IEEE Std 301−1988.
[96] W. Sansen, Z.Y. Chang: "Limits on low noise performance of detector readout front ends
in CMOS technology." IEEE Trans. Circuits and Systems, vol. 37, no. 11, 1990,
p. 1375−1382.
[97] P. O'Connor, V. Radeka: "Integrated circuit front ends for nuclear processing." Short Cour-
se. IEEE NSS and MIC, Orlando, 2009.
[98] J.L. Blankenship, C. H. Nowlin: "New concepts in nuclear pulse amplifier design." IEEE
Trans. Nucl. Sci., vol. 13, no. 3, 1966, p. 495−507.
[99] C. H. Nowlin W.: "Pulse shaping for nuclear pulse amplifier." IEEE Trans. Nucl. Sci.,
vol. 17, no. 1, 1970, p. 226−241.
[100] Z.Y. Chang, W. Sansen: "Effect of 1/f noise on resolution of CMOS analog readout sys-
tems for microstrip and pixel detectors." Nucl. Instr. and Meth. A, vol. 305, 1991,
p. 553−560.
[101] G. De Geronimo, P. O'Connor: "MOSFET optimization in deep submicron technology for
charge amplifiers." IEEE Trans. Nucl. Sci., vol. 52, no. 6, 2005, p. 3223−3232.
[102] E. Gatti, M. Sampietro, P. F. Manfredi: "Optimum filters for detector charge measurements
in presence of 1/f noise." Nucl. Instr. and Meth. A, vol. 287, 1990, p. 513−520.
[103] D. Middleton: "An introduction to statistical communication theory." MCGraw-Hill, New
York, 1960.
178 References
[104] V, Re, et. al.: "Experimental study and modeling of white noise sources in submicron
P- and N-MOSFETs." IEEE Trans. Nucl. Sci., vol. 48, no. 4, 2001, p. 1577−1586.
[105] M. Manghisoni, L. Ratti, V. Re, V. Speziali: "Submicron CMOS technologies for low-
noise analog front-end circuits." IEEE Trans. Nucl. Sci., vol. 49, no. 4, 2002,
p. 1783−1790.
[106] M. Manghisoni, et al.: "Resolution limits in 130 nm and 90 nm CMOS technologies for
analog front-end applications." IEEE Trans. Nucl. Sci., vol. 54, no. 3, 2007, p. 531−537.
[107] P.F. Manfredi, M. Manghisoni, L. Ratti, V. Re, V. Speziali: "Resolution limits achievable
with CMOS front-end in X- and γ-ray analysis with semiconductor detectors." Nucl. Instr.
and Meth. A, vol. 512, 2003, p. 167−178.
[108] P. Gryboś, M. Idzik, A. Skoczeń: "Design of low noise charge amplifier in sub-micron
technology for fast shaping time." Analog Integrated Circuits and Signal Processing, vol.
49, 2006, p. 107−114.
[109] P. Gryboś, M. Idzik, P. Maj: "Noise optimization of charge amplifiers with MOS input
transistors operating in moderate inversion region for short peaking times." IEEE Trans.
Nucl. Sci., vol. 54, no. 3, 2007, p. 555−560.
[110] C. Enz, F. Krummenacher, E. Vitoz: "An analytical MOS transistor model valid in all re-
gions of operation and dedicated to low-voltage and low-current applications." Analog In-
tegrated Circuits and Signal Processing, vol. 8, 1995, p. 83−114.
[111] S.C. Terry, et al.: "Comparison of BSIM3v3 and EKV MOSFET model for a 0.5 μm
CMOS process and implications for analog circuit design," IEEE Trans. Nucl. Sci., vol. 50,
no. 4, 2003, p. 915−920.
[112] D. M. Binkley, B.S. Puckett, M.E. Casey, R. Lecomte, A. Saoudi: "A power-efficient, low-
noise, wideband, integrated CMOS preamplifier for LSO/APD PET systems", IEEE Trans.
Nucl. Sci., vol. 47, no. 3, 2000, p. 810-817.
[113] P. O'Connor, J. -F. Pratte, G. De Geronimo: "Low noise charge amplifier in submicron
CMOS." Vth International Workshop on Front End Electronics (FEE 2003) Snowmass,
USA July 2, 2003.
[114] N. R. Campbell, V. J. Francis, "A theory of value and circuit noise." Journ. of the Institu-
tion of Electrical Engineers (IEE), vol. XCII, Part III, 45, 1946.
[115] C. H. Nowlin, J.L. Blankenship: "Elimination of undesirable undershoot in the operation
and testing of nuclear pulse amplifier." Review of Scientific Instruments, vol. 36, 1965,
p. 1830−1839.
[116] P. Gryboś, P. Maj, R. Szczygieł: "Comparison of two pole-zero cancellation circuits for
fast charge sensitive amplifier in CMOS technology", Proceedings of the 14th International
Conference Mixed Design of Integrated Circuits and Systems. MIXDES 2007, Ciecho-
cinek, Poland, June 2007, p. 243−246, IEEE Catalog Number 07EX1721.
[117] P. O'Connor, G. Gramegna , P. Rehak, F. Corsi, C. Marzocca: "Ultra low noise CMOS pre-
amplifier-shaper for X-ray spectroscopy." Nucl. Instr. and Meth. A, vol. 409, 1998,
p. 315−321.
[118] R. L. Chase, A. Hrisoho, J. P. Richter: "A CMOS detector leakage current self-adaptable
continuous reset system: Theoretical analysis." Nucl. Instr. and Meth. A, vol. 409, 1998,
p. 328−331.
[119] S. Riboldi, A. Geraci, E. Gatti, G. Ripamonti: "A new digital auto-tracking pole-zero com-
pensation technique for high resolution spectroscopy." Nucl. Instr. and Meth. A vol. 482,
2002, p. 475−490.
[120] V. Radeka: "The effects of baseline restoration on signal-to-noise ratio in pulse amplitude
measurements." Rev. Sci. Instr. vol. 38, 1967, p. 1397−1403.
References 179
[121] A. Pullia, D. Maiocchi, G. Bertuccio, S. Caccia: "A compact VLSI dc restorer for mul-
tichannel X-γ ray detectors." IEEE Trans. Nucl. Sci., vol. 52, no. 5, 2005, p. 1643−1646.
[122] L.B. Robinson: "Reduction of baseline shift in pulse amplitude measurement." Rev. Sci.
Instr. vol. 32, 1961, p. 1057.
[123] R.L. Chase, L.R. Poulo: "A high precision DC restorer." IEEE Trans. Nucl. Sci., vol. 14,
no. 6, 1967, p. 83−88.
[124] E.A. Gere, G.L. Miller: "Active D-C restoration in nuclear pulse spectrometry." IEEE
Trans. Nucl. Sci., vol. 14, no. 1, 1967, p. 89−95.
[125] E. Fairstein: "Gated baseline restorer with adjustable asymmetry." IEEE Trans. Nucl. Sci.,
vol. 22, no. 1, 1975, p. 463-466.
[126] M. Kuwata, H. Maeda, K. Husimi: "New base line restorer based on feedforward differen-
tial compensation." IEEE Trans. Nucl. Sci., vol.41, no. 4, 1994 p. 1236−1239.
[127] B. Bevensee, F.M. Newcommer, R. Van Berg, H.H. Williams: "An amplifier-shaper-
discriminator with baseline restoration for ATLAS transition radiation tracker." IEEE
Trans. Nucl. Sci., vol.43, no. 3, 1996, p. 1725−1731.
[128] J.M. Rochelle, D.M. Binkley, M.J. Paulus: "Fully integrated current-mode CMOS baseline
restorer circuits." IEEE Trans. Nucl. Sci., vol.42, no. 4, 1995, p. 729−735.
[129] A. Pullia: "The switch-off baseline restorer for 120 channel silicon detector system for
EXAFS at NSLS." Nucl. Instr. and Meth. A, vol. 370, 1996, p. 490−498.
[130] B. Yu, Z. Zojceski, J.A.Harder, A.Hrisoho, V. Radeka, G.C. Smith: "Front-end electronics
for high rate, position sensitive neutron detectors." Nucl. Instr. and Meth. A. vol. 485,
2002, p. 645−652.
[131] C. Arnaboldi, G. Pessina: "A very simple baseline restorer for nuclear applications." Nucl.
Instr. and Meth. A. vol. 512, 2003, p. 129−135.
[132] G. De Geronimo, P. O'Connor, J. Grosholz: "A CMOS baseline holder (BLH) for readout
ASICs." IEEE Trans. Nucl. Sci., vol.47, no. 3, 2000, p. 818−822.
[133] G. De Geronimo, P. O'Connor, J. Grosholz "A generation of CMOS readout ASICs for
CZT detectors." IEEE Trans. Nucl. Sci., vol.47, no. 6, 2000, p. 1857−1867.
[134] P. Gryboś, et al.: "Multichannel mixed-mode IC for digital readout of silicon strip detec-
tors." Microelectronics Reliability, vol. 42, 2002, p. 427−436.
[135] G. De Geronimo, A. Dragone, J. Grosholz, P. O'Connor, E. Vernon: "ASIC with multiple
energy discriminator for high rate photon counting applications." IEEE Trans. Nucl. Sci.,
vol. 54, no. 2, 2007, p. 303−312.
[136] A. Rivietti, G. Anelli, F. Anghinolfi, G. Mazza, F. Rotondo: "A low-power 10-bit ADC in
0.25-μm CMOS: design consideration and test results." IEEE Trans. Nucl. Sci., vol. 48, no.
4, 2001, p. 1225−1228.
[137] V. Valencic, et al.: "A low-power piecewise linear analog to digital converter for use in
particle tracking." IEEE Trans. Nucl. Sci., vol. 42, no. 4, 1995, p. 772−775.
[138] O. B. Milgrome, S.A. Kleinfelder, M.E. Levi: "A 12 bit analog to digital converter for
VLSI applications in nuclear science." IEEE Trans. Nucl. Sci., vol. 39, no. 4, 1992,
p. 771−775.
[139] A. Rossini, et al.: "A complete readout channel with embedded Wilkinson A/D converter
for X-ray spectroscopy." IEEE Trans. Nucl. Sci., vol. 54, no. 4, 2007, p. 1216−1221.
[140] M.S. Emery, et al.: "A multichannel ADC for use in PHENIX detector." IEEE Trans. Nucl.
Sci., vol. 44, no. 3, 1997, p. 374−378.
[141] J. Bouver, et al.: "A low power and low signal 5-bit 25 MS/s pipelined ADC for monolithic
active pixel sensors." IEEE Trans. Nucl. Sci., vol. 54, no. 4, 2007, p. 1195−1220.
180 References
[162] "HSPICE Users’s Manual: Elements and Device Models Volume II." Meta-Software 1996.
[163] Y. Tsividis: "Operation and Modeling of the MOS Transistor." New York, WCB/McGraw-
Hill 1999.
[164] "BSIM4.0.0 manual." Department of Electrical Engineering and Computer Sciences, Uni-
versity of California Berkeley, CA 94720, 2000.
[165] A.L. McWhorter: "1/f noise and germanium surface properties." in: "Semiconductor Sur-
face Physics." Philadelphia, University of Pennsylvania Press 1957.
[166] S. Christensson, I. Lundstrom, C. Svensson: "Low frequency noise in MOS transistors – I
theory." Solid State Electronics, vol. 11, 1968, p. 796−812.
[167] F.N. Hooge, T.G. Kleinpenning, L.K. Vandamme: "Experimental verification studies on
1/f noise." Reports on Progress in Physics, vol. 44, 1981, p. 479−532.
[168] M. Pelgrom, M. Vertregt: "CMOS technology for mixed signal ICs." Solid-State Electron-
ics, vol. 41, no. 7, 1997, p. 967−974.
[169] D. P. Triantis, A.N. Birbas, D. Kondis: "Thermal noise modeling for short channel MOS-
FET’s." IEEE Transactions on Electron Devices." vol. 43, no. 11, 1996, p. 1950−1954.
[170] G. Anelli, F. Faccio, S. Florian, P. Jarron: "Noise characterization of 0.25μm CMOS tech-
nology for LHC experiments." Nucl. Instr. and Meth., vol. A457, 2001, p. 361−368.
[171] M. Stegherr: "Flicker noise in hot degraded short channel MOSFETs." Solid State Elec-
tronics, vol. 27, 1984, p. 1055−1056.
[172] C.H. Cheng, C. Surya: "The effect of hot-electron injection on the properties of the flicker
noise in the N-channel MOSFETs." Solid State Electronics, vol. 36, 1993, p. 4750−479.
[173] Kirton M.J., Uren M.J.: "Noise in solid-state microstructures: a new perspective on indi-
vidual defects, interface states and low frequency (1/f) noise." Advances in Physics, vol.
38, no. 4, 1989, p. 367−468.
[174] C. Jakobson, I. Bloom, Y. Nemirovsky: "1/f noise in CMOS transistors for analog applica-
tions from subthreshold to saturation." Solid State Electronics, vol. 42, no. 10, 1998,
p. 1807−1817.
[175] P. O'Connor, G. De Geronimo: "Prospects for charge sensitive amplifiers in scaled
CMOS." Nucl. Instr. and Meth., vol. A480, 2002, p. 713−725.
[176] K. Verghese, D. Allstot: "Computer-aided design considerations for mixed-signal coupling
in RF integrated circuits." IEEE J. Solid-State Circuits, vol. 33, no. 3, 1998, p. 314−323.
[177] A. Afzali-Kusha, M. Nagatan, N.K. Verghese, D.J. Allstot: "Substrate noise coupling in
SoC design: modeling, avoidance, and validation." Proceedings of the IEEE, vol. 94, no.
12, 2006, p. 2109−2138.
[178] T. Blalack: "Design techniques to reduce substrate noise." in: J. Huijsing, R. van de Plass-
che, W. Sansen, eds: "Analog circuit design. Volt electronics; mixed-mode systems; low-
noise and RF power amplifiers for telecommunication." Boston, Kluwer Academic Pub-
lishers 1999, p. 193−217.
[179] K. Su, M.J. Loinaz, S. Masui, B.A. Wooley: "Experimental results and modeling tech-
niques for substrate noise in mixed-signal integrated circuits." IEEE J. Solid-State Circuits,
vol. 28, no. 4, 1993, p. 420−430.
[180] R. Gharpurey: "Modeling and analysis of substrate coupling in integrated circuits." PhD
Thesis, Electronics Research Laboratory, College of Engineering, University of California,
Berkeley 1995.
[181] Heijningen M., Compiet J., Wambacq P., Donnay S., Engels M.G.E., Bolsens I.: "Analysis
and experimental verification of digital substrate noise generation for epi-type substrates."
IEEE J. Solid-State Circuits, vol. 35, no. 7, 2000, p. 1002−1008.
182 References
[182] Clement F.: "Technology impacts on substrate noise." in: J. Huijsing, R. Van de Plassche
W. Sansen eds: "Analog circuit design. Volt electronics; mixed-mode systems; low-noise
and RF power amplifiers for telecommunication." Boston, Kluwer Academic Publishers
1999, p. 173−192
[183] F. Clement: "Substrate noise coupling analysis in mixed-signal ICs." Presentation from the
Workshop on Substrate-Noise Coupling in Mixed-Signal ICs, IMEC, Leuven, Belgium,
September 5−6, 2001.
[184] M. Ingels, M. Steyaert: "Design strategies and decoupling techniques for reducing the ef-
fects of electrical interference in mixed-mode IC’s." IEEE J. Solid-State Circuits, vol. 32,
no. 7, 1997, p. 1136−1141.
[185] K. Makie-Fukuda, T. Kikuchi, T. Matsuura, M. Hotta: "Measurement of digital noise in
mixed-signal integrated circuits." IEEE J. Solid-State Circuits, vol. 30, no. 2, 1995,
p. 87−92.
[186] T. Schmerbeck: "Practical aspects in analog & mixed-mode IC design." Laussanne, Swit-
zerland, EPFL Electronics Laboratories Advanced Engineering Course CMOS & BiCMOS
IC Design ’99: Practical Aspects in Analog and Mix-Mode ICs 1999.
[187] P. Fischer, E. Kraft: "Low swing differential logic for mixed signal applications." Nucl. In-
str. Meth. A, vol. 518, 2004, p. 511−514.
[188] B. Nauta, G. Hoogzaad: "Substrate bounce in mixed-mode CMOS." in: J. Huijsing, R. van
de Plassche, W. Sansen, eds: "Analog circuit design. Volt electronics; mixed-mode sys-
tems; low-noise and RF power amplifiers for telecommunication." Boston, Kluwer Aca-
demic Publishers 1999, p. 157−171.
[189] D. Pedder: "Interconnection and packaging of solid-state circuits." IEEE J. Solid-State Cir-
cuits, vol. 24, no. 3, 1989, p. 698−703.
[190] J. Baker, H. Li, D. Boyce: "CMOS circuit design, layout, and simulation." Piscataway, NJ,
USA, IEEE Press 1998.
[191] R. Meyer, W. Mack: "A 1-GHz BiCMOS RF front-end IC." IEEE J. Solid-State Circuits,
vol. 29, no. 3, 1994, p. 350−355.
[192] M. Pelgrom, A.C. Jeanet Rens, M. Vertregt, M.B. Dijkstra: "A 25-Ms/s 8-bit CMOS A/D
converter for embedded application." IEEE J. Solid-State Circuits, vol. 29, no. 8, 1994,
p. 879−886.
[193] R. Meyer, W. Mack: "A wideband low-noise variable-gain BiCMOS transimpedance am-
plifier." IEEE J. Solid-State Circuits, vol. 29, no. 6, 1994, p. 701−706.
[194] K. Makie-Fukuda, S. Maeda, T. Tsukada, T. Matsuura: "Substrate noise reduction using ac-
tive guard band filters in mixed-signal integrated circuits." IEICE Tran. Fundamentals, vol.
E80 A, no. 2, 1997, p. 313−320.
[195] B.P. Wong, A. Mittal, Y. Cao, G. Starr: "Nano-CMOS Circuit and Physical Design." John
Wiley and Sons, Inc., Hoboken, New Jersey, 2006.
[196] H. Schouwenaars, W. Groeneveld, H. Termeer: "A low power stereo 16-bit CMOS D/A
converter for digital audio." IEEE J. Solid-State Circuits, vol. 23, no. 6, 1988,
p. 1290−1297.
[197] G. Plas, J. Vandenbussche, A. Van den Bosch, M. Steyaert, W. Sansen, G. Gielen: "MOS
transistor mismatch for high accuracy applications." in ProRISC: IEEE Benelux Workshop
on Circuits, Systems and Signal Processing, p. 529−534, Mierlo, The Netherlands, Novem-
ber 1999.
[198] P.C.S. Scholtens, D. Smola, M. Vertregt: "Systematic power reduction and performance
analysis of mismatch limited ADC designs." ISLPED '05. Proceedings of the 2005 Interna-
tional Symposium on Low Power Electronics and Design, San Diego, 2005, p. 78−83.
References 183
[218] W. Sansen: "Analog Design Essentials.", Springer, Dordrecht, The Netherlands, 2006.
[219] J. Rabaey, A. Chandrakasan, B. Nikolic: "Digital Integrated Circuits: A Design Perspec-
tive." Upper Saddle River, NJ, USA, Prentice Hall 2003.
[220] B. Razavi: "Design of Analog CMOS Integrated Circuits." New York, USA, McGraw-
Hill 2001.
[221] A. Hastings: "The Art of Analog Layout." Prentice Hall, Upper Saddle River, NJ, 2000.
[222] Chi-Hung Lin, K. Bult: "A 10-b, 500-Msample/s CMOS DAC in 0.6 mm2." IEEE J. Solid-
State Circuits, vol. 33, no. 12, 1998, p. 1948−1958.
[223] J. Bastos, A.M. Maraques, M.S.J. Steyaert, W. Sansen: "A 12-bit intrinsic accuracy high-
speed CMOS DAC." IEEE J. Solid-State Circuits, vol. 33, no. 12, 1998, p. 1959−1969.
[224] C. Bastiaansen, D. Groeneveld, H. Schouwenaars, H. Termeer: "A 10-b 40-MHz 0.8-μm
CMOS current-output D/A converter." IEEE J. Solid-State Circuits, vol. 26, no. 7, 1991,
p. 917−921.
[225] G. Plas, J. Vandenbussche, W. Sansen, M. Steyaert, G. Gielen: "A 14-bit intrinsic accuracy
Q2 random walk CMOS DAC." IEEE J. Solid-State Circuits, vol. 34, no. 12, 1999,
p. 1708−1718
[226] M. McNutt, S. LeMarquis, J. Dunkley: "Systematic capacitance matching errors and cor-
rective layout procedures." IEEE J. Solid-State Circuits, vol. 29, no. 5, 1994, p. 611−616.
[227] P.G. Drennan, M.L. Kniffin, D.R. Locascio: "Implications of proximity effects for analog
design." Proc. of IEEE Custom Integrated Circuits Conference, 2006.
[228] H. Tuinhout, M. Pelgrom, R. Penning de Vries, M. Vertegret: "Effects on metal coverage
on MOSFET matching." Technical Digest IEDM’96, 1996, p. 735−739.
[229] H. Tuinhout, M. Vertreget: "Test structures for investigation of metal coverage effects on
MOSFET matching." Proceedings IEEE Int. Conference on Microelectronics Test Struc-
tures ICMTS’97, 1997, p. 179−183.
[230] Y. Unno: "High-density low-mass hybrid and associated technologies." Proceedings of 6th
Workshop on Electronics for LHC Experiments, CERN 2000−010, p.66−76, Cracow, Po-
land, September 2000.
[231] Q. Zhang, J.J. Liou, J.R. McMacken, J. Thomson, P. Layman: "SPICE modeling and quick
estimation of MOSFET mismatch based on BSIM3 model and parametric." IEEE J. Solid-
State Circuits, vol. 36, no. 10, 2001, p. 1592−1595.
[232] W. Kuźmicz: "Introduction to design for manufacturability: random variation, defects,
faults, yield and cost." Proceedings of the 10th International Conference Mixed Design of
Integrated Circuits and Systems. MIXDES 2003, Łódź, Poland, June 2003, p. 119−139.
[233] A. Vasilescu (INPE Bucharest) and G. Lindstroem (University of Hamburg), Displacement
damage in silicon, on-line compilation. http://sesam.desy.de/members/gunnar/Si-
dfuncs.html.
[234] W. Dąbrowski: "Radiation damage in Si detectors and front-end electronics." Nucl. Phys.
Proc. Suppl. vol. 44, 1995, p. 463−467.
[235] F.W. Sexton, J.R. Schwank: "Correlation of radiation effects in transistors and integrated
circuits." IEEE Trans. Nucl. Sci., vol. 32, no. 6, 1985, p. 3975−3981.
[236] E. Chesi, et al.: "Performance of a 128 channel analog front end chip for read-out of Si strip
detector modules for LHC experiments." IEEE Trans. Nucl. Sci., vol. 47, 2000,
p. 1434−1441.
[237] R. Szczygieł: "Projektowanie bloków cyfrowych układów ASIC do odczytu detektorów
pozycyjnych eksperymentu ATLAS pod katem odporności radiacyjnej." PhD Thesis, 2005
Technical University of Łódź, Poland.
References 185
[238] S.C. Sun, J.D. Plummer: "Electron mobility in inversion and accumulation layers on ther-
mally oxided silicon surfaces." IEEE Trans. Electron Devices, vol. 27, no. 8, 1980,
p. 1497−1507.
[239] S. Dimitrijev, N. Stojadinovic: "Analysis of CMOS transistor instabilities." Solid State-
Electronics, vol. 30, 1987, p. 991−1003.
[240] G.F. Galloway, R. D. Schrimpf: "MOS device degradation due to total dose ionizing radia-
tion in the natural space environment. A review." Microelectronics Journal, vol. 21, 1990,
p. 67−81.
[241] G. Anelli: "Conception et Characterisation de Circuit Integres Restistant aux Radiations
pour les Detecteur de Particule du LHC en Technologies CMOS Submicronique Pro-
fondes." PhD Thesis, Institut National Polytechnique de Grenoble, 2000.
[242] G. Anelli, et al.: "Total dose behaviour of submicron and deep submicron CMOS technolo-
gies." Proc. of the Third Workshop on Electronics for LHC Experiments, 1997, London,
England, p. 139−143.
[243] J.V. Osborn, R.C. Lacoe, D.C. Mayer, G. Yabiku: "Total dose hardness of three commer-
cial CMOS microelectronics foundries." IEEE Trans. Nucl. Sci., vol. 45 no. 3, 1998,
p. 1458−1463.
[244] F. W. Sexton: "Measurement of single event phenomena in devices and ICs." Proc. IEEE
NSREC short Course, New Orleans, LA, 1992, p. III−1−55.
[245] A. Giraldo, A. Paccagnella, A. Minzoni: "Aspect ratio calculation in n-channel MOSFETs
gate-enclosed layout." Solid-State Electronics, vol. 44, p. 981−989.
[246] J. E. Vinson: "Circuit reliability of memory cells with SEU protection." IEEE Trans. Nucl.
Sci., vol. 39,no. 6, 1992, p. 1671−1678.
[247] S. Bonacini, K. Kloukinas, A. Marchioro: "Development of SEU-robust, radiation-tolerant
and industry-compatible programmable logic components." Journal of Instrumentation
2007 JINST 2 P09009, http://iopscience.iop.org/1748-0221/2/09/P09009.
[248] T. Akesson, et al.: "Implementation of the DTMROC-S ASIC for ATLAS TRT detector in
0.25 μm CMOS technology." IEEE NSS Conference Record, Nov. 2002, Norfolk, Vir-
ginia, USA, p. 549−553.
[249] E. Heijne, et al.: "LHC1: A semiconductor pixel detector readout chip with internal, tun-
able delay providing a binary pattern of selected events." Nucl. Instr. and Meth., vol. A383,
1996, p. 55−63.
[250] J.-R. Lutz, et al.: "Electrical characterization of ALICE128C: a low-power CMOS ASIC
for the readout of silicon strip detectors." Proceedings of the Fourth Workshop on Electron-
ics for the LHC Experiments, Rome, September, 1998, CERN/LHCC/98−36.
[251] W. Dąbrowski, et. al.: "Design and performance of the ABCD chip for the binary readout
of silicon strip detectors in the ATLAS Semiconductor Tracker." IEEE Trans. Nucl. Sci.,
vol. 47,no. 6, 2000, p. 1843−1850.
[252] P. Schmitt, et al.: "Performance of a CMOS mixed analogue-digital circuit (APVD) for the
silicon tracker in CMS", Proceedings of the Fourth Workshop on Electronics for the LHC
Experiments, Rome, September, 1998, CERN/LHCC/98−36.
[253] M. Feuerstack-Raible: "Overview of microstrip readout chips." Nucl. Instr. and Meth. A,
447, 2000, p. 35−43.
[254] M. Overdick, et al.: "Status of direct conversion detectors for medical imaging with
X-rays." IEEE Trans. Nucl. Sci., vol. 56, no. 4, 2009, p. 1800−1809.
[255] S. O. Rice: "Mathematical analysis of random noise." Bell System Tech. J., 24, 1945,
p. 46−156.
186 References
[256] R. Turchetta, Y. Hu, Y. Zinzius, C. Colledani, A. Loge: "A low-power, low-noise mixed
mode VLSI ASIC for infinite dynamic range imaging applications." Report LEPSI−9806,
LEPSI, IN2P3/CNRS−ULP, Strasburg, France,
[257] C. Colledani, et al.: "Castor 1.0, a VLSI analog-digital circuit for pixel imaging applica-
tions." Nucl. Instr. and Meth., A395, 1997, p. 435−442.
[258] P. Maj: "Zintegrowany wielokanałowy system pomiarowy do detekcji niskoener-
getycznego promieniowania X o dużym natężeniu." PhD Thesis, AGH University of Sci-
ence and Technology, Cracow, Poland 2008.
[259] E. Beuville, T. Collins, I. Kipnis, D. Nygren, E. Oltman: "An efficient digital X-ray imag-
ing system with high spatial resolution and robust energy sensitivity." IEEE NSS 1995, San
Francisco, USA, p. 1602−1606.
[260] E. Beuville, et al.: "High resolution X-ray imaging using a silicon strip detector." IEEE
Trans. Nucl. Sci., vol. 45, no. 6, 1998, p. 1269−1273.
[261] P. Gryboś, W. Dąbrowski: "Development of fully integrated readout system for high count
rate position-sensitive measurements of X-rays using silicon strip detectors." IEEE Trans.
Nucl. Sci., vol. 48, no. 3, 2001, p. 466−472.
[262] E.F. Tsakas, et al.: "Low noise high-speed X-ray readout IC for imaging applications."
Nucl. Instr and. Meth., vol. A469, 2001, p. 106−115.
[263] A. Studen, et al.: "Development of silicon pad detectors and readout electronics for
a Compton camera." Nucl. Instr and. Meth., vol. A501, 2003, p. 273−279.
[264] B. Schmitt, et al.: "Mythen detector system." Nucl. Instr. Meth. A, vol. 501, 2003,
p. 267−272.
[265] B. Krieger, I. Kipnis, B. A. Ludewigt: "XPS: A multi-channel preamplifier-shaper IC for
X-ray spectroscopy," IEEE Trans. Nucl. Sci., vol. 45, no. 3, 1998, p. 732−734.
[266] B. Krieger, et al.: "XPS: An 8 × 8 pixel IC for X-ray spectroscopy." IEEE Trans. Nucl.
Sci., vol. 48, no. 3, 2001, p. 493−498.
[267] G. De Geronimo, P. O'Connor, A. Kandasamy, J. Grosholz: "Advanced readout ASICs for
multielement CZT detectors." SPIE Annual Mtg., Seattle, WA. 8-9 July 2002, Proc. SPIE
vol. 4784, p. 105−118.
[268] G. De Geronimo, et al.: "Development of a high-rate high-resolution detector for EXAFS
experiments." IEEE Trans. Nucl. Sci., vol. 50, no. 4, 2003, p. 885−891.
[269] D. Moraes, J. Kaplon, N. Nygard, et al.: "CERN_DxCTA counting mode chip." Nucl. Instr
and. Meth., vol. A591, 2003, p. 167−170.
[270] E. Kraft: "Counting and integrating microelectronics development for direct conversion
X-ray imaging." PhD Thesis, Universitat Bonn, 2007.
[271] E. Kraft, et al.: "Counting and integrating readout for direct conversion X-ray imaging:
concept, realization and first prototype measurements." IEEE Trans. Nucl. Sci., vol. 54, no.
2, 2007, p. 383−390.
[272] J. Fink, et al.: "Comparison of pixelated CdZnTe, CdTe and Si sensors with the simultane-
ously counting and integrating CIX chip." IEEE Trans. Nucl. Sci., vol. 56, no. 6, 2009,
p. 3819−3827.
[273] Y. Hu, G. Deptuch, R. Turchetta, C. Guo: "A low-noise, low-power CMOS SOI readout
front-end for silicon detectors leakage current compensation with capability." IEEE Trans.
Circuits and Systems I, vol. 48. no. 8, 2001, p. 1022−1030.
[274] S. Szczygieł: "Krummenacher feedback analysis for high-count-rate semiconductor pixel
detector readout." 17th International Conference Mixed Design of Integrated Circuits and
Systems. MIXDES 2010, Wrocław, Poland, June 2010.
References 187
[275] E. Beuville, et al.: "A 16 × 16 pixel array detector for protein crystallography." Nucl. Instr.
and Meth. A395, 1997, p. 429−434.
[276] P. Datte, et al.: "Status of the digital pixel array detector for protein crystallography." Nucl.
Instr. and Meth. A421, 1999, p. 576−590.
[277] P. Fischer, et al.: "A counting pixel readout chip for imaging applications." Nucl. Instr. and
Meth. A405, 1998, p. 53−59.
[278] P. Fischer, et al.: "Single photon X-ray imaging with Si- and CdTe-sensors."
HEP−EX−0209040, Talk given at the 3rd int. workshop on radiation imaging detectors,
Orosei, Italy, 2002.
[279] M. Locker, et al.: "Single photon counting X-ray imaging with Si and CdTe single chip
pixel detectors and multichip pixel modules." IEEE Trans. Nucl. Sci., vol. 51, no. 4, 2004,
p. 1717−1723.
[280] M. Lindner, l. Blanquart, P. Fischer, H. Kruger, N. Wermes: "Medical X-ray imaging with
energy windowing." Nucl. Instr. and Meth. A465, 2001, p. 229−234.
[281] M. Campbell, E.H. Heijne, G. Meddeler, E. Pernigotti, W. Snoeys: "A readout chip for
a 64 × 64 pixel matrix with 15-bit single photon counting." IEEE Trans. Nucl. Sci.,
vol. 45, no. 3, 1998, p. 751−753.
[282] http://medipix.web.cern.ch/MEDIPIX/
[283] Z. Vykydal, et al.: "The RELAXd project: Development of four-side tilable photon-
counting imagers." Nucl. Instr. and Meth. A591, 2008, p. 241−244.
[284] X. Llopart, R. Ballabriga, M. Campbell, L. Tlustos, W. Wong: "Timepix, a 65k program-
mable pixel readout chip for arrival time, energy and/or photon counting measurements."
Nucl. Instr. and Meth. A581, 2007, p. 485−494.
[285] R. Ballabriga, M. Campbell, E.H.M. Heijne, X. Llopart, L. Tlustos: "The Medipix3 proto-
type, a pixel readout chip working in single photon counting mode with improved spectro-
metric performance." IEEE Trans. Nucl. Sci., vol. 54, no. 5, 2007, p. 1824−1829.
[286] R. Ballabriga, et al.: "Characterisation of the Medipix3 pixel readout chip." IEEE NSS
2010, Orlando, USA.
[287] http://pilatus.web.psi.ch
[288] P. Kraft, et al.: "Characterisation and calibration of Pilatus detectors." IEEE Trans. Nucl.
Sci., vol. 56, no. 3, 2009, p. 758−764.
[289] P. Pangaud, et al.: "XPAD3: A new photon counting chip for X-ray CT-scanner." Nucl.
Instr. and Meth. A571, 2007, p. 321−324.
[290] P. Delpierre, et al.: "PIXSCAN: Pixel detector CT-scanner for small animal imaging."
Nucl. Instr. and Meth. A571, 2007, p. 425−428.
[291] R. Dinapoli, et al.: "A new family of pixel detectors for high frame rate X-ray applications"
Nucl. Instr. and Meth. A617, 2010, p. 384−386.
[292] R. Szczygieł, P. Gryboś, P. Maj: "A pixel readout chip designed in 90 nm CMOS process
for high count rate imaging systems." IEEE NSS 2010, Orlando, USA, p. 1103−1106.
[293] G. Deptuch, et al.: "Vertically integrated circuits at Fermilab." IEEE NSS 2009, Orlando,
USA, p. 1907−1915.
[294] G. Deptuch, (P. Gryboś), el al.: "VIPIC - subreticule J." 3D-IC Consortium Meeting,
Marselles, France, 18-19 March, 2010.
[295] P. Fischer: "First implementation of the MEPHISTO binary readout architecture for strip
detectors." Nucl. Instr. and Meth. A461, 2001, p. 499−504.