Content uploaded by Piera Maccagnani
Author content
All content in this area was uploaded by Piera Maccagnani
Content may be subject to copyright.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Journal of Modern Optics
Vol. 54, Nos. 2–3, 20 January–15 February 2007, 163–189
Photon counting arrays for astrophysics
F. ZAPPAy, S. TISA*y, S. COVAy, P. MACCAGNANIz, R. SALETTIx,
R. RONCELLAx, F. BARONTIx, D. BONACCINI CALIA{, A. SILBER{,
G. BONANNOkand M. BELLUSOk
yPolitecnico di Milano, Milano, Italy
zIMM-CNR (Microelectronics and Microsystems Institute of the
Italian National Research Council), Bologna, Italy
xUniversity of Pisa, Pisa, Italy
{ESO (European Southern Observatory), Garching, Germany
kINAF – Astrophysics Observatory of Catania, Italy
(Received 9 February 2006; revised 4 April 2006; in final form 7 April 2006)
A compact system for counting and time-tagging single photons is presented,
based on a monolithic array sensor of 60 pixels able to detect single photons,
namely the single-photon avalanche diode array (SPADA). First, the working
principle and performance of the single-photon detector pixel is detailed, with
particular attention paid to monolithic array integration. Then the electronics
needed to quench each pixel after avalanche ignition, namely the active-quenching
circuit (AQC) is discussed, since the features of this quenching electronics
dramatically affect the operating conditions of the detector, hence its actual
performance. The discussion then focuses on integration of the SPADA system
into Astrophysics applications such as adaptive optics, fast-transient imaging and
atmospheric layer sensing. The whole electronics necessary to control SPADA
operating conditions and temperature is also described, together with the
complete opto-mechanics used to focus the telescope pupil onto the detector.
Finally, experimental results are reported.
1. Single-photon detectors
In many fields and in particular in astrophysical observations, a chronic problem is
the photon-starving condition, which becomes severe when images are to be obtained
in short acquisition times (from micro to milliseconds), as happens in hot areas of
astrophysics: optical counterparts of high-energy gamma-ray bursts, study and
interpretation of Supernovae bursts, adaptive optics (AO) [1] for real-time correction
of atmospheric turbulence by means of pulsed lasers. CCDs are inherently unable to
provide accurate measurements of such fast low-intensity transients at high frame
rates, whereas monolithic solid-state arrays of photon counters would be suitable.
While high-sensitivity CCDs are being developed by industry [2], and some
*Corresponding author. Email: tisa@elet.polimi.it
Journal of Modern Optics
ISSN 0950–0340 print/ISSN 1362–3044 online ß2007 Taylor & Francis
http://www.tandf.co.uk/journals
DOI: 10.1080/09500340600742320
Downloaded By: [Tisa,] At: 08:22 27 March 2007
applications have resorted to the use of a set of discrete SPADs [3], the monolithic
array of single-photon detectors presented in this work is state-of-the-art in both the
scientific community and industry.
To respond to single photons, suitable detectors must provide output signals
that are sufficiently high to be individually processed by electronic circuits.
Therefore, only detectors with an internal mechanism that provides a high
multiplication of charge carriers are suitable, namely vacuum tube photomultipliers
(PMTs), solid-state avalanche photodiodes (APDs) and emerging electron-
multiplying CCDs (EM-CCDs) [2]. In PMTs, the photocathodes available for the
visible spectral range provide fairly good quantum efficiency and low noise, whereas
cathodes for the red and near-infrared range have lower quantum efficiency and
must be cooled to reduce the dark-count rate. PMTs are bulky, and so not suitable
for assembly in large arrays, fragile, sensitive to electromagnetic disturbances
and mechanical vibrations, require high supply voltages (2–3 kV) and are costly
devices, particularly the high-performance models. EM-CCDs exploit an internal
multiplication process to achieve sub-electron readout noise, thus being able to
detect single photons [2]. Their quantum efficiency is very high, and they are
inherently suited to imaging applications. However, due to their nature, they cannot
provide frame rates higher than a few kiloframes per second, and cannot be used
in extreme time-resolved measurements.
Semiconductor APDs have the typical advantages of solid state devices (small
size, low bias voltage, low power consumption, ruggedness and reliability, suited to
building integrated systems, etc.). In APDs operating in linear mode, the internal
gain is not sufficient or barely sufficient to detect single photons. Instead, single
photons can be detected efficiently by avalanche diodes operating in Geiger mode,
known as single-photon avalanche diodes (SPADs). Silicon SPADs have been
investigated extensively [4] and are nowadays well known and widely employed;
considerable progress has been made in design and fabrication techniques, and
devices with good characteristics are commercially available [5–8]. In this paper
we present a compact two-dimensional imaging system for counting and time-
tagging single photons, based on a monolithic array of 60 pixels, namely the
single-photon avalanche diode array (SPADA).
1.1 SPAD working principle
Essentially, SPADs are p–n junctions biased at a voltage V
A
above the breakdown
voltage V
B
. At this bias, the electric field is so high (higher than 3 10
5
V/cm) that
a single charge carrier injected in the depletion layer can trigger a self-sustaining
avalanche. The current rises swiftly (nanosecond or subnanosecond rise-time) to a
macroscopic steady level, in the milliampere range. If the primary carrier is
photogenerated, the leading edge of the avalanche pulse marks (with picosecond
jitter) the arrival time of the detected photon. The current continues to flow until
the avalanche is quenched by lowering the bias voltage to or below V
B
: the lower
electric field is no longer able to accelerate the carriers to impact-ionize with
lattice atoms. Then the bias voltage must be restored, in order to be able to detect
164 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
another photon. The circuit that performs such operations is usually referred to as a
quenching circuit [9].
In order to work as a photodetector, a diode must be able to remain biased above
breakdown for a sufficient time, let’s say longer than a few milliseconds. This means
that the generation–recombination phenomenon, which would fire the avalanche,
must be kept very low. Since thermally generated carriers can fire an avalanche, it is
possible to observe output current pulses also when a SPAD is kept in the dark: such
an average counting rate is called the dark-counting rate and is one of the key
parameters in defining detector noise [5].
2. Avalanche quenching circuits
Once the SPAD is biased above the breakdown voltage and is triggered, current
keeps flowing until the avalanche process is quenched by lowering the bias voltage
to V
B
or below. After a dead-time, the operating voltage must be restored in order to
enable the SPAD to detect another photon. This operation requires suitable
electronics, which has the following tasks:
(1) it senses the leading edge of the avalanche current;
(2) it generates a standard output pulse, synchronous with the onset of current;
(3) it quenches the avalanche by lowering the bias below the breakdown voltage;
(4) it restores the photodiode voltage to the operative level.
This circuit is usually referred to as a quenching circuit. The most commonly used
circuit in studies on Geiger-mode avalanche photodiodes is the passive-quenching
circuit: the avalanche current quenches itself simply by developing a voltage drop
across a high-impedance load (R
L
4100 k). Such a circuit is very simple and can
easily be employed, but sets severe limitations on the maximum admissible photon
counting rate and on detector performance in general [9]. In fact, it was the
introduction of the active-quenching circuit (AQC) concept by S. Cova [10, 11] that
opened the way to practical application of SPADs. Many AQC types have since
been reported, with circuit structure and mounting that evolved from standard NIM
cards [11, 12] to small SMT boards suitable for compact detector modules [6, 13, 14].
2.1 Fully-integrated iAQC
In recent years, work towards monolithic integration of AQCs was started at
Politecnico di Milano. First, the core structure of the AQC was integrated, excluding
only the high-voltage quenching driver [15]. In order to reduce the overall stray
capacitance, improve timing performance and strongly reduce power dissipation,
we designed and fabricated the first fully-integrated active-quenching circuit
(iAQC) [16]. The circuit was the first reported in the literature and is under
European and US patent [17]. For this present work we developed an improved
version of the iAQC, whose microphotograph is shown in the left side of figure 1.
Photon counting arrays for astrophysics 165
Downloaded By: [Tisa,] At: 08:22 27 March 2007
The overall dimensions are about 1 1 mm. In order to enable the integration
of multichannel apparata, like the SPADA system discussed in this paper,
five monolithic iAQC were built onto a single chip.
In the following we report the results of tests made on the iAQC driving thin
SPADs with breakdown voltages ranging from 17 V to 35 V. The applied overvoltage
was in the range from 5 V to 10 V. The very short dead-time is only 30 ns, thus
leading to a maximum saturated counting rate of 30 Mcounts/s. Figure 2 shows the
same measurement with a longer dead-time of about 400 ns, obtained by simply
adjusting the voltage at a control pin. The maximum hold-off time achievable is
500 ns. Figure 3 shows the cathode waveforms and the corresponding output pulses
obtained by illuminating the detector with steady weak light: the detector is triggered
by subsequent single photons with a 200 ns dead-time. Figure 3 proves that the iAQC
can be retriggered immediately after the cathode has been reset to the operating level.
Figure 4 shows gated-mode operation: when the GATE level is high, the detector is
free to operate and the circuit output delivers a standard TTL pulse; when the GATE
is low, the SPAD is kept quenched and no avalanche can be triggered.
3. Astrophysics applications
The SPADA system was mainly aimed at three astrophysical applications that
require ultra-sensitive arrays with a small number of detectors for the visible range:
adaptive optics (AO), fast transient imaging (FTI) and layer sensing (LS). Adaptive
optics was the driving force, and it determined many parameters of the system.
Figure 1. Micrographs of the iAQC (left) and compact board containing five iAQC
channels. (The colour version of this figure is included in the online version of the journal.)
166 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Figure 3. Example of free running operation of the SPAD. (The colour version of this figure
is included in the online version of the journal.)
Figure 2. SPAD cathode waveform and iAQC output pulse, with 200 ns hold-off time.
(The colour version of this figure is included in the online version of the journal.)
Photon counting arrays for astrophysics 167
Downloaded By: [Tisa,] At: 08:22 27 March 2007
The following paragraphs provide a brief description of the applications and the
requirements used to implement the data processing board of the SPADA system.
3.1 Adaptive optics
One of the toughest problems affecting ground-based telescopes is the presence of
the atmosphere, which distorts the spherical wavefront, creating phase errors in the
image-forming ray paths. Even at the best sites, ground-based telescopes observing
at visible wavelengths cannot achieve an angular resolution in the visible better than
telescopes of 10 to 20 cm diameter, because of atmospheric turbulence alone. The
cause is random spatial and temporal wavefront perturbations induced by turbulence
in various layers of the atmosphere; one of the principal reasons for flying the
Hubble Space Telescope was to avoid this image smearing. In addition, image
quality is affected by permanent manufacturing errors and by long timescale
wavefront aberrations introduced by mechanical, thermal, and optical effects in the
telescope, such as defocusing, decentring, or mirror deformations generated by their
supporting devices.
Adaptive optics is the answer to this problem: a deformable mirror is inserted
in the light path of the telescope, and its control signal is based on measurement
of the incoming wavefront, performed by a suitable high-sensitivity detector.
Figure 4. Example of gated-mode operation of the SPAD. The iAQC holds off the
SPAD whenever the Gate signal is low. (The colour version of this figure is included in the
online version of the journal.)
168 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Because of the high bandwidth and the small field to which correction can generally
be applied, adaptive optics employs a small deformable mirror with a diameter of
8 to 20 cm located behind the focus of the telescope at or near an image of the pupil.
The choice of the number of actuators, usually piezoelectric, is a trade off between
degree of correction, use of faint reference sources and available light budget. A large
number of actuators requires a similarly large number of subapertures in the
wavefront sensor.
Two main methods are used to measure the degraded wavefront, the Shack–
Hartmann device [18], which measures the slope of the wavefront from the positions
of the images of the reference star given by each subpupil, and curvature sensing
[18, 19], where the measured intensity in strongly defocused images directly gives the
local curvatures of the wavefront. Correction in the Shack–Hartmann device is made
with individual piezoelectric actuators. Correction in a curvature sensing system is
accomplished with a bimorph adaptive mirror, made of two bonded piezoelectric
plates. With both methods, wavefront sensing is done on a reference star (Natural
Guide Star, NGS), or even on the observed object itself if it is bright enough and
has sufficiently sharp light gradients. The measurement can be performed in the
visible for observation in the infrared, or in the infrared itself. The control system is
generally a dedicated computer that uses the measurements from the wavefront
sensor to calculate the commands to be sent to the actuators of the deformable
mirror. The calculation must be fast (within 0.5 to 1 ms), otherwise the state of the
atmosphere may have changed, thus rendering the wavefront correction inaccurate.
The SPADA system was designed to operate as a curvature wavefront sensor
(CWFS). As we will see, this choice and the need to make it a potential replacement
for the MACAO [20] apparatus already in use at ESO’s Very Large Telescope (VLT)
determined many of its characteristics, especially the geometry of the monolithic
detector. It has also a strong advantage over CCDs if a pulsed laser system is used
thanks to its gating function and parallel readout, which allow faster loop cycles.
Figure 5 depicts the simplified operating principle of a CWFS. By means of a moving
membrane, the focusing plane of the incoming photon flux over the detector is
changed with sinusoidal oscillations. The number of detected photons when such
a plane is both before (counting A) and behind (counting B) the detector surface is
collected. The measurement must be repeated over an integer number of oscillation
periods. Eventually, for each pixel the CWFS provides either the accumulated A
and Bcounts or the curvature signal:
C¼AB
AþB:ð1Þ
The value Cis proportional to the local curvature of the wavefront, associated with
the subaperture encompassed by the pixel in use. The servo drives the system at each
loop step to have a flat wavefront, hence to have C¼0 for all pixels. The acquisition
must be repeated in free running. The CWFS have to accomplish all these tasks:
detect the incoming photon, count their occurrences in the two integration windows
Aand B, provide the curvature signal Cfor all the pixels, control the membrane
that oscillates with a frequency ranging from 1.5 kHz to 3 kHz. On the other hand,
Photon counting arrays for astrophysics 169
Downloaded By: [Tisa,] At: 08:22 27 March 2007
the calculation of the signal to be applied to the deformable mirror to account for the
atmospheric aberration is carried out by a specialized real-time computer.
The timing control logic must alternatively enable the counting of pulses
corresponding to Aor Bsemiperiod. The collections must continue for integer
multiples of the sinusoid period; such multiples must range between 1 and 256.
Therefore, the overall collection period can last a minimum of about 1/3 kHz¼333 ms
to a maximum of 256/1.5 kHz¼171 ms. At the end of the collection interval,
the processing electronics must upload the 60 items of curvature data (Cvalues
for all pixels). Hence, the maximum transfer rate requested is 60 channels
16 bit/333 ms¼3 Mbps. In addition to the curvature signal, it must be possible to
upload also the individual Aand Bcounts every collection period. In conclusion
a maximum 33 Mbps 10 Mbps is required.
3.2 Fast transient imaging
In fast transient imaging (FTI) applications, such as the study of transients of optical
counterparts in high-energy gamma-ray bursts or supernovae bursts, it is important
to acquire the image of the 60 pixel detector for each integration time window,
adjustable from 10 ms to 100 ms. A key parameter in this application is the
maximum attainable frame rate, since short time windows allows finer time-
tagging of the incoming photon flux, with quite enough resolution for many
Timing and
control logic
USB
Processing:
A−B
A+B
Audio
ampl.
Output
GND
Sinusoidal
waveform
Input Wave Front
SPADA
pixel
Moving membrane
Loudspeaker
In focus
Before focus
Behind focus
ABAB
Figure 5. Schematic principle of operation of a curvature wavefront sensor (CWFS).
(The colour version of this figure is included in the online version of the journal.)
170 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
astronomical applications. The image consists of the number of photons detected by
the corresponding pixel and counted by the board during the integration time.
Therefore, every T
W
time period, the 60 collected counts must be uploaded to
a remote computer via a suitable high bandwidth link. The same link must be used
to download settings and parameters to the board. A remote software displays the
images and performs the required post-processing.
The integration time window must be programmable between 10 ms and 100 ms
with 12 bits resolution. Each counter must be 16 bits wide and with saturation.
In this way, at the theoretical maximum counting rate (20 Mcps) allowed by the
iAQCs (at minimum dead-time of 50 ns), the counter will saturate for windows
longer than 3.3 ms. In such a situation, the user can decide to reduce the integration
window, thus improving also the time-tagging performance of the acquisition.
Instead, with T
W
shorter than 1 ms the counter will almost never exceed 14 bits, while
at T
W
¼10 ms 8 bits will be sufficient. Therefore also the uploading to the remote
computer can be adapted to an effective number of bits given by:
neff ¼log2ðTW20 McpsÞ:ð2Þ
Timing and control logic should latch the content of each counter into the
corresponding latch. In this way, no photon will be missed. The content of all 60
latches could be shifted out in serial order. At the shortest T
W
interval, this means a
maximum bit rate of:
Fmax ¼60 pixels 16 bit
10 ms¼96 Mbps:ð3Þ
This is the maximum transfer rate that it has to be sustained during data
uploading. As a consequence, a high bandwidth link is needed to accomplish the
task. However, the effective transfer rate could be reduced if an adaptive algorithm
was adopted, as previously suggested.
3.3 Layer sensing
Natural Guide Star (NGS) adaptive optics suffers from the fact that it is not always
possible to find a suitable NGS close enough to the portion of sky under
investigation. Normally only 1% of the sky has an available NGS for current AO
systems. The most promising way to overcome the lack of sufficiently bright natural
reference stars is the use of artificial reference stars, also referred to as laser guide
stars (LGS). These are patches of light created by the back-scattering of pulsed
laser light by sodium atoms in the high mesosphere, or in the case of Rayleigh LGS,
by molecules and particles located in the low stratosphere. Such an artificial
reference star can be created as close to the astronomical target as desired, and
a wavefront sensor measuring the LGS wavefront is used to correct the atmospheric
aberrations on the target object.
Nevertheless, there are still a number of physical limitations with an LGS.
A problem, focus anisoplanatism, also called the cone effect, became evident
very early on. Because the artificial star is created at a relatively low altitude,
Photon counting arrays for astrophysics 171
Downloaded By: [Tisa,] At: 08:22 27 March 2007
back-scattered light collected by the telescope forms a conical beam, which does not
cross exactly the same turbulence-layer areas as the light coming from the distant
astronomical source. This leads to a phase estimation error. For even larger
telescopes, such as the 100 m OWL under investigation by ESO, a solution to this
problem is mandatory.
Attempts to overcome the cone effect are under study at different institutes [21],
which use atmosphere’s layers sensing and pulsed laser beams. The sensor timing is
extremely important, in a sort of two-dimensional LIDAR technique. The key
features are that a parallel (or nearly parallel) laser beam is projected from the
full primary aperture and that sensing takes place on the upward path. The
methods therefore rely on an observable modulation of the scattered intensity by
turbulence-induced phase distortions during upward propagation of the laser beam.
The SPADA system can be used to detect the small number of photons that are
back-scattered by the different layers.
Concerning the functions required of the SPADA system, this application is a
mixture of the two previous ones. The system should wait for a synchronization pulse
at 7 kHz repetition rate, that signals the firing of the laser pulse; the detector is then
gated-off for the subsequent 2.5 ms in order to mask the fluorescence of the optics,
due to the powerful laser pulse; then the SPAD is gated on and the board records
10 samples lasting 13 ms each for each SPADA pixel, corresponding to 10 atmosphere
layers, during the upward path of the laser beam. The acquisition is repeated a
user-definable number of times. Eventually, the cumulative acquisition of the 10 sets
of 60 pixels each is uploaded to the remote computer at the end of the acquisition,
when the measurement will be restarted. At the same time, the board generates
a sinusoidal signal, with the same specifications as listed for the AO application.
This signal drives a pupil plane membrane mirror, of the type used in curvature
systems such as MACAO systems. The sinusoidal motion of the membrane mirror
creates for one half of the period a progressive conjugation of the sensor with
different atmospheric heights above the telescope primary. It does so by acting
as a gentle negative lens which collimates the layer conjugate. By adjusting the
membrane mirror period and phase parameters, it is possible to obtain perfect
tracking of the pulsed laser beam conjugate while propagating upward. Only sensors
like SPADA can grab the images of the layers with the timing required.
The maximum communication bit rate for this application is almost equivalent
to that required for FTI, setting at about 74 Mbps.
4. System integration
Figure 6 shows the architecture of the whole SPADA system. It is composed of
the SPADA silicon chip mounted on a micro-machined holder, the matched
optomechanics for focusing and alignment, a detection electronic board based
on the iAQCs, a metallic housing, a custom data processing electronic board,
and the control software. The complete system during bench testing is
shown in figure 7. The data pocessing board acquires counts from the SPADA
172 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Figure 7. Rack containing the complete data processing electronics and power supplies
(on the right) and the detection board (on the left). (The colour version of this figure is
included in the online version of the journal.)
Figure 6. Architecture of the SPADA system. (The colour version of this figure is included in
the online version of the journal.)
Photon counting arrays for astrophysics 173
Downloaded By: [Tisa,] At: 08:22 27 March 2007
sensor (via the detection board), generates the sync signals and sinusoid for AO and
LS applications, and handles hardware and software gates and interlocks. The board
counts photon signals at each of the 60 inputs, generates also the required user-
programmable time windows for the different applications and uploads processed
data to a remote computer through a firewire connection that is able to sustain the
required maximum data rate. Given the tasks to be performed, the board needs some
dedicated logic for pixel counting, together with a processor for data processing and
data transfer. The choice was to build a dedicated data acquisition system based
on an Orsys board mod. C6713, which is equipped with a TMS320C6713 DSP and
a Virtex-II FPGA [22].
4.1 SPADA detector
The SPADA sensor is made of 60 SPAD elements arranged in a circular concentric
geometry, as shown in figure 8. This disposition is the same as that used for the
MACAO system, and is optimum for the 60 subaperture adaptive optics apparatus
designed by ESO. The inner circle has a diameter of 2.53 mm, whereas the outer has a
diameter of 14.49 mm. This implied a square shape for the SPADA detector, with
dimensions of 18 18 mm, in order to accommodate the bonding pads. The 60 pixels
are at the focus loci of small spherelets, placed at the field focus of a 60 subaperture
segmented lenslet array, about 16 mm in diameter, that ensures that the entire
telescope’s pupil is projected onto the 60 pixels.
Figure 8. Photograph of the SPADA chip. Dimensions 18 mm 18 mm. (The colour version
of this figure is included in the online version of the journal.)
174 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
In order to maximize the chance of having a system with better attainable
performance during the prototyping phase, the SPADA chip contains four sets of
60 pixels, with diameters of 20 mm (active area of about 300 mm
2
), 35 mm (962 mm
2
),
50 mm (2,000 mm
2
), and 75 mm (4,400 mm
2
), as shown in figure 9. The input optics is
aligned and focused onto one set of SPADs (i.e. one single pixel for each element)
and only those 60 pixels are bonded to the holder and reach the detection board.
The set to use was selected after characterization of SPADA performance and yield,
and for a good compromise between diameter and dark-counting rate the choice
was to use the 50 mm diameter set. The alignment is performed with the help of three
alignment marks, such as the one visible on the right side of Figure 9.
The SPADA chip also includes an integrated thermoresistor that is used by the
temperature controller of the detection board to provide closed-loop regulation of
the operating temperature of the detector. This assures constant detection efficiency
over the whole measurement, and permits one to obtain a lower dark-counting rate
if required by the application. It should be noticed that detector cooling is not
necessary to guarantee correct behaviour, but it is a simple, viable method to further
enhance it. This is not true for CCDs, which usually require strong cooling to attain
reasonable noise level, or for thick SPADs that require cooling to avoid faults, due to
the much higher power dissipation. Figure 8 shows a photograph of the SPADA
sensor with all the aforementioned features highlighted.
4.2 Detection electronics
The detection electronics was designed to work under two operating conditions:
together with the developed data processing board (as already shown in figure 6)
or connected and one to one compatible with the MACAO system developed by
ESO. A schematic block diagram of the detection electronics is shown in figure 10.
Figure 9. A pixel of the array, showing the four different diameter SPADs. Each detector
has its own cathode, whereas anodes are shared. On the right side is shown one of the three
marks that permit alignment of the optics with the chosen set of detectors. (The colour version
of this figure is included in the online version of the journal.)
Photon counting arrays for astrophysics 175
Downloaded By: [Tisa,] At: 08:22 27 March 2007
The board is built around 60 integrated active quenching circuits (iAQC), one for
each pixel of the SPADA sensor, and outputs 60 differential signals through two
SCSI-like connectors (68 pins each). As we have already seen, the iAQCs are
fabricated as integrated chips containing five iAQCs each. For easier replacement in
the case of a faulty iAQC, each chip is assembled on a compact printed circuit board
as shown in figure 1, on the right, with dimensions 1.5 2.5 cm. The connectors on
the left side of the board are mated to those from the SPADA, thus resulting in the
shortest possible path between SPADs and iAQCs. The connectors on the right
side mate with connectors on the detection board, providing power supplies and
signal from and to the iAQCs. A total of 12 iAQC boards are needed to drive
the 60 SPADA pixels, and they are arranged along a square, three for each side of
the SPADA holder, as shown in figure 11.
iAQC Diff.out
iAQC
iAQC
iAQC
SCSI connector
68 pin
1
2
3
60
Peltier
Controller
Common
Anode
Regulator
µC
Interlock
RS232
USB
Gate
FT232BM
MAX232
Buffer
gate
−Vlow
T meas
T set
From SPADA sensorTo SPADA sensor
To Data-Procesing Electronics
From Data-Procesing
Electronics
To remote
computer
Buffer
gate
Power supplies
+5V
+12V
−24V
Gnd
+12V Peltier
Gnd Peltier
Interlock
Thermores
Peltier
+5V
+12V
−24V
+12V Peltier
Diff.out
Diff.out
Diff.out
SCSI connector
68 pin
Figure 10. Schematic block diagram of the detection board. (The colour version of this
figure is included in the online version of the journal.)
176 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
All iAQCs have a common Gate input, for enabling/disabling the SPADA sensor
(for instance for the layer sensing application). The hold-off duration of all iAQCs
is adjustable from about 10 ns to 500 ns, through a common setting controlled by
the microcontroller. The detection board includes a temperature controller that
drives the Peltier cooler based on the reading of the SPADA integrated
thermoresistor. The temperatures of both side of the Peltier stage are also monitored
for diagnostics and to prevent hazard conditions. The Peltier is powered by a
dedicated þ12 V supply, due to its high power demand. An 8 bit microcontroller
manages all settings, diagnostics, and communication from the detection electronics
to either the data processing board or the MACAO equipment. Commands and
readings (temperature, overvoltage, hold-off duration) are transferred via RS-232
or USB interface.
In order to provide access for the cold finger, the detection board has a 4 4cm
hole in the centre, with the iAQC stamps mounted vertically between the detector
and the board. Figure 12 shows the board with the different subsections highlighted.
The board has a diameter of 20 cm and it is built on a six-layer-PCB, with two plane
layers for ground and power supplies.
4.2.1 iAQC interfacing and settings. For compatibility reasons with the existing
MACAO system, the 60 TTL outputs from the iAQCs are fed to a set of quad-
differential RS-485 buffers, model DS26C31, mounted in close proximity to the
iAQC socket (see figure 12), thus improving the noise immunity of signals. The 120
lines are then connected to two high-density HD68 SCSI connectors.
The common Gate of all iAQCs enabling the switch-off of all 60 pixels can be
controlled either by the microcontroller or externally through a SMC connector and
an optocoupler PC357NT, thanks to a wired-OR interconnection: whenever one of
the two signals is high, the detector is switched off. The microcontroller can do this
Figure 11. Complete detection board assembled with the SPADA detector. (The colour
version of this figure is included in the online version of the journal.)
Photon counting arrays for astrophysics 177
Downloaded By: [Tisa,] At: 08:22 27 March 2007
on an user request provided through the serial connection, or as the result of a failure
in the board.
The common hold-off setting is carried out through the use of a digital
potentiometer AD5220, with 128 taps. It is controlled by the microcontroller
through a three-wire connection that permits one to increment or decrement the
current value of the DigPot. The desired setting can be saved in the microcontroller’s
EEPROM so that the hold-off is automatically adjusted at power-on of the system.
The negative supply, common to all SPADs anodes, is adjusted using a linear
voltage regulator LM337. This regulator was preferred to a switching regulator in
order to have no ripple on the supply and to avoid any disturbance to the critical
iAQC section caused by the commutation of the power MOSFETS. In addition, the
low current requirement of this supply (less than 120 mA), results in moderate power
dissipation of the regulator even with the largest voltage drop applied. The stage
achieves a 22.8 V to 13.8 V range for the SPADA anode. Since the positive iAQC
supply is fixed at þ12 V, this corresponds to an overvoltage of up to 8 V for a SPAD
with V
BD
¼26.8 V at room temperature, such as those in the SPADA detector.
4.2.2 Temperature control loop. SPAD performance and in particular the dark-
counting rate can be greatly enhanced by moderate cooling (about 10 to 20C).
However, since the breakdown voltage is temperature dependent, it is crucial that the
temperature remains constant regardless of any external stimulus: a change in
breakdown at a fixed applied bias would result in a change in the overvoltage and
thus a change in the detection efficiency. For this reason the detection board includes
a temperature control loop.
The temperature controller is composed of an integrated thermoresistor, a signal
reading and amplification stage, a software PID controller that runs in the
PIC microcontroller, a DC–DC Peltier driver and a Peltier stage. The energy
dissipated within a SPAD with 25 V breakdown and a 5 V overvoltage is
Figure 12. Photographs of the two sides of the detection board, with the different
subsections highlighted. (The colour version of this figure is included in the online version of
the journal.)
178 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
W¼VIt
ON
¼30 V 15 mA 5ns¼2.25 nJ per avalanche pulse of about 5 ns
duration. At the maximum counting rate of 30 Mcps this corresponds to a power
dissipation of 68 mW for each SPAD and 4 W for the whole array operating
at saturation. This is the heat that the Peltier must be able to remove, to which all
thermal losses should be added. The chosen Peltier stage, model CP1.0-71-05 from
Melcor, meets these requirements, being a single-stage TEC with a maximum heat
removal capability of 18 W, a maximum Tof 75C and dimensions 23 23 mm.
The system design included the possibility to use a cold finger with a flowing coolant
to easily remove both the heat produced by the SPADA and that produced by the
Peltier itself.
4.2.3 Component assembly. As already outlined, the SPADA system is composed
of two assemblies, one containing the data processing board and the power supplies
for the system, and the other hosting all the detection electronics and the fore-optics.
Figure 13 shows a cross-section of the housing. The upper part hosts the fore-optics
and ends with the coupling flange that permits connection to the telescope.
The central part protects the SPADA detector mounted on the rectified holder
and the detection electronics. Finally, the back plate holds all connectors needed and
the attachment for the cold-finger coolant. The entire housing is vacuum proof,
since during operation it will be filled with dry N
2
to avoid any condensation when
the detector is operated below 0C. The choice to fill the housing with dry N
2
and
not only the detector package was made on account of the difficulties in assuring
a tight seal and also in consideration of the small distance between the detector and
the optics.
The fore-optics is derived from that used in the MACAO system [20]. It is
composed of a custom 60 subaperture lenslet array (made by Heptagon) and 60 ball
lenses. The 60 subaperture lenslet array is made of two back-to-back keystones shape
apertures in order to realize the 45 mm focal length needed with less sag on each
lenslet. The 60 ball lenses are placed in the focal plane of the lenslet array, and they
are used to focus the light into the SPADs’ active areas, from which they are at
a distance of only 60 mm. In order to assure perfect alignment of these optical
elements with the detector, they are mounted on precisely micro-machined holders,
whose relative positions have been chosen after having measured the effective
placement of the SPADA on its holder. Figure 14 shows the plate that hosts the
60 ball lenses, together with the clamp that keeps the lenses in position. It should be
noted that, as is visible in figure 13, all the optics are fixed onto the rectified plate
to which the SPADA is glued (light grey shapes). This ensures that any manipulation
of the external case (dark grey shapes) or of the detection board will not affect the
optics alignment.
The complete detection electronics, assembled and ready for insertion into the
enclosures, is shown in figure 15. The exterior wall of the enclosures can be removed
without having to dismount any part of the electronics, thus allowing easy
troubleshooting. The overall dimensions of the detection head are about 30 cm in
height and 20 cm in diameter.
Photon counting arrays for astrophysics 179
Downloaded By: [Tisa,] At: 08:22 27 March 2007
N2 Inlet N2 Outlet
Cold finger
Detection board
Lenslet array Ball lenses
Ball lenses
holder
SPADA
detector
Entrance window
Clamp
Rectified plate
Connectors
O-Ring
Figure 13. Cross-section of the complete SPADA detection head assembly. During
operation it is filled with dry nitrogen to avoid condensation. Sealing is guaranteed by the
use of O-rings and vacuum-tight connectors. (The colour version of this figure is included in
the online version of the journal.)
Figure 14. On the right, the rectified plate that holds the 60 ball lenses and ensures perfect
alignment of the SPADA detector with the fore-optics. On the left, the clamp that keeps
the ball lenses in position, just 60 mm above the detector. (The colour version of this figure is
included in the online version of the journal.)
180 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
5. Experimental results
The performance of the SPADA detectors was investigated through many tests on
several different silicon wafers. In particular, breakdown uniformity over the same
chip, dark counts, photon detection efficiency and optical crosstalk were measured.
These characteristics resulted to be reasonably uniform over different SPADA. All
the results reported are for the 50 mm pixels, which proved to be a good compromise
between diameter of the active area and noise.
The breakdown voltages for a typical SPADA chip at room temperature show
a total spread of less than 0.5 V. As we will see in a while, this spreading is truly
irrelevant for dark counting rate, though it has a minor effect on the detection
efficiency uniformity. Figure 16 shows the measured dark counts for a typical
SPADA detector at three overvoltages and at room temperature. For easier reading,
pixels were ordered for increasing dark counting rate at 5 V overvoltage. It can be
noted that all pixels are below 40 kcps at 5 V overvoltage, with a third of them
being below 1000 cps. Dark counts are expected to decrease by about 15 times
at a temperature of 10C, on the basis of dark count measurements at low
temperature performed on a single SPAD from the same wafer. As is visible in
figure 17 (left and centre), it was not possible to find any correlation between dark
counts and breakdown voltages. Photon detection efficiency (PDE) was measured
in the wavelength range from 400 nm to 1000 nm, using a calibrated monochromator
Figure 15. Complete assembled detection head, composed of the detection board, the
60 iAQCs and the SPADA detector mounted in its holder. (The colour version of this figure is
included in the online version of the journal.)
Photon counting arrays for astrophysics 181
Downloaded By: [Tisa,] At: 08:22 27 March 2007
and an integrating sphere to achieve uniform optical power over the entire SPADA
detector. There is a broad peak ranging from 400 nm to 700 nm, higher than 30%
even at 5 V overvoltage, and also a not negligible few per cent in the near-infrared
range. An even better sensitivity can be obtained through the use of an antireflection
coating, as shown in figure 18. Figure 17 (left and right) shows the breakdown
voltages of a typical SPADA chip, and the photon detection efficiency for the
same chip when biased 5 V above the average breakdown. Pixels with lower
breakdown, and thus higher effective overvoltage, have higher PDE, and vice versa.
This indicates that spreading in the PDE is mainly due to the non-uniform biasing of
the detectors in the SPADA system, which uses a common anode, and that the
process variations are truly negligible in terms of PDE uniformity over a chip.
Silicon p–n junctions emit photons when operated in the avalanche regime [23].
In a monolithic array of detectors, photons emitted from a SPAD can trigger an
avalanche in another detector, thus causing correlations among array elements.
For this reason, we measured the optical crosstalk with high accuracy using the
coincidence detector shown in figure 19. This is implemented using an external
FPGA, thus allowing the simultaneous monitoring of all 60 channels of the
SPADA system. In order to measure the crosstalk between two detectors (SPAD
A
and SPAD
B
), the output pulses of their associate iAQCs are first reshaped,
extending the corresponding pulse widths to T
A
and T
B
, and then they are
sent to an AND gate. When both SPAD
A
and SPAD
B
detect an event within
a time interval T¼T
A
þT
B
, the AND gate produces a coincidence pulse.
10
100
1000
10000
100000
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
55
57
59
Pixel
Dark counts (cps)
VOV =7 V
VOV =6 V
VOV =5 V
−10 °C
RT
Figure 16. Dark counting rate at room temperature (upper set) for all the 50 mm pixels of a
typical SPADA detector at three different overvoltages. The pixels are ordered for increasing
dark counts. Expected values at 10C (lower set), based on known temperature dependence
of single SPAD from the same wafer is also reported. (The colour version of this figure is
included in the online version of the journal.)
182 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Figure 17. Breakdown voltage uniformity of a typical SPADA chip (left). Dark counts at 5 V overvoltage (centre), showing no correlation
with breakdown voltage. Photon detection efficiency (right) at 5 V overvoltage, showing slight correlation with breakdown voltage. (The colour
version of this figure is included in the online version of the journal.)
Photon counting arrays for astrophysics 183
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Provided that Tis larger than the duration of the avalanche current pulse, all
optical crosstalk events generate a coincidence pulse. Three counters accumulate the
pulses generated by SPAD
A
and SPAD
B
and the coincidence pulses for a given
measurement time T. The total counts N
A
(N
B
) includes both the actual counts
generated by SPAD
A
(SPAD
B
), and the counts caused by crosstalk from SPAD
B
(SPAD
A
).
The total coincidence counts Cincludes all the optical crosstalk events, but also
uncorrelated dark counts occurring within the time interval T. The average number
of these uncorrelated events in the measurement time Tis given by
Cuc ¼NANBT
T:ð4Þ
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
400 500 600 700 800
Wavelen
g
th
(
nm
)
DE
VOV = 7 V AR
VOV = 5 V AR
VOV =5 V
VOV =7 V
Figure 18. Photon detection efficiency (PDE) for a 50 mm pixel of a typical SPADA detector
at different overvoltages, with an optimum anti-reflection coating, made using 78 nm of SiO
2
and 50 nm of Si
3
N
4
. Dashed lines represent the PDE with no specific anti-reflection coating.
(The colour version of this figure is included in the online version of the journal.)
−Vpol
Pulse shaper
Pulse shaper
Counter
Counter
Counter
iAQC
iAQC
N
A
NB
C
∆TA
∆TB
Figure 19. Block diagram of the coincidence detector implemented using an FPGA.
(The colour version of this figure is included in the online version of the journal.)
184 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Therefore, the actual number of crosstalk events in the same time Tis given by:
CC¼CCuc ¼CNANBT
T:ð5Þ
The optical crosstalk probability P
C
between SPAD
A
and SPAD
B
can therefore
be calculated as follows:
PC¼CC
NAþNBCCffiCNANBT=T
NAþNB
:ð6Þ
The accuracy of the optical crosstalk measurement can be evaluated by assuming
that the distribution of the coincidence counts is Poissonian. In this case, the
standard deviation of the coincidence counts would be equal to
ffiffiffiffi
C
p. The accuracy
in the estimation of the optical crosstalk events C
c
is therefore given by
CC
CC¼3
CC¼3
ffiffiffiffi
C
p
CNANBT=T:ð7Þ
By performing long measurements (T20 min) we were able to attain accuracy
better than 5%.
Figure 20 shows the optical crosstalk probability as a function of distance
between 50 mm elements for a SPADA detector, measured at an overvoltage of 5 V.
A worst-case crosstalk probability of 4 10
4
was measured between two SPADs
spaced at 1.5 mm. Strategies for further reducing the crosstalk probability are
currently under investigation.
Photon detection efficiency can be increased by operating the SPADA at a higher
overvoltage. One may thus think that the higher the overvoltage, the better the
10−3
10−4
10−5
10−6
0 5000 10000 15000
Distance (mm)
Crosstalk probability
Figure 20. Measured crosstalk probability for a SPADA detector as a function of the
distance between 50 mm pixels. (The colour version of this figure is included in the online
version of the journal.)
Photon counting arrays for astrophysics 185
Downloaded By: [Tisa,] At: 08:22 27 March 2007
performance of the system. However, this is not true, since the dark counts and
hence the intrinsic noise of the detector also increases with the overvoltage,
thus impairing the SNR. Moreover, the noise due to the statistical nature of the
incident radiation must be accounted for, and the relative weight of the two noise
source (detector and radiation) on the SNR depends on the duration of the
measurement, following the equation
SNR ¼nsTM
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
nsTMþnDTM
p:ð8Þ
In order to determine the optimum overvoltage value, a diagram such as the one
reported in figure 21 is useful. It represents the minimum signal photon rate needed
to give SNR¼1 as a function of the integration time, and thus it indicates the
maximum sensitivity of the detector for a given operating condition. Figure 21 shows
curves at three different overvoltages for a typical SPADA detector at room
temperature and at 550 nm, both for the best pixel (high PDE, low dark count) and
for the worst pixel (low PDE, high dark count). The sensitivity limit imposed by
the radiation statistic is also reported, since this cannot be exceeded by any detector.
For each pixel, an optimistic curve corresponding to PDE¼100% and to a noise
corresponding to 7 V overvoltage is reported, in order to provide an upper limit to
10−610−510−410−310−210−1
106
105
104
103
Worst SPAD
Corner
Radiation noise limit
Best SPAD
VOV =7 V
VOV =6 V
VOV =5 V
Integration time TM (s)
Signal photon rate (cps)
Figure 21. Minimum signal photon rate needed to attain a SNR ¼1, as a function of the
integration time, for the best and the worst 50 mm pixels of a typical SPADA chip, at room
temperature and at 550 nm. For each pixel, the curves for three different overvoltages are
reported, together with the one that would correspond to optimistic behaviour with
PDE ¼100% and the same dark counts as 7 V overvoltage. As a reference, the limit
imposed by the radiation intrinsic noise is reported. (The colour version of this figure is
included in the online version of the journal.)
186 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
the increase in sensitivity that one could ideally achieve. It is possible make the
following points:
(a) The curves have two regions, one dominated by detector noise (right-hand
side) and proportional to T1=2
M, the other by radiation noise (left-hand side)
proportional to T1
M.
(b) The corner between the two regions is inversely proportional to the dark
count, so it moves towards longer integration times as the dark count
decreases.
(c) Increasing the overvoltage is effective in increasing sensitivity only if we
are working in the left side (noise radiation limited). Otherwise the
corresponding increase in dark count impairs the advantage of higher PDE.
From these considerations, it is possible to conclude that working at high
overvoltage is useful only at short integration times, when the radiation noise
overwhelms detector noise. However, short integration times require strong signal to
achieve a good SNR. If the source is too weak to allow short measurement times, the
best approach to increase sensitivity, rather than raising the overvoltage, is to cool
down the detector to reduce the dark count (thus shifting to the right the region
dominated by radiation noise) and if possible applying an anti-reflection coating
to boost the PDE to the maximum.
Finally, in order to demonstrate the fast imaging capability of the system,
we coupled the detector to a standard camera objective to focus a real-world image
on the SPADA. Then, we positioned the detector in front of a CRT monitor,
completely dark except for some lines at low luminosity. Figure 22 shows the counts
Figure 22. Example of fast transient acquisition. The detector was placed in front of a dark
CRT, with only one horizontal line highlighted. The graph reports the counts collected by
a pixel of the detector during 50 ms integration windows. The refresh of the CRT at 60 Hz
and the exponential decay of the phosphorous are clearly visible. (The colour version of this
figure is included in the online version of the journal.)
Photon counting arrays for astrophysics 187
Downloaded By: [Tisa,] At: 08:22 27 March 2007
accumulated in 50 ms integration windows by a pixel of the detector, when only a
horizontal line is highlighted on the monitor. The fast frame capability of the system
catches without any problem not only the refresh rate of the monitor (16.6 ms,
corresponding to 60 Hz), but also the exponential decay of the phosphorous,
showing a time constant of a few milliseconds. Figure 23 shows ten frames separated
by 500 ms of the entire detector, when only a circular ring centred on the SPADA
is highlighted on the monitor. It is possible to see how, from one frame to the other,
the electron beam moves along the screen to paint the circle.
5. Conclusions
We have discussed the design and fabrication of an imaging system based on
monolithic arrays of single-photon avalanche diodes to be employed in Astrophysics
applications with fast frame rates. We have detailed the design and realization of
both the detection and the data processing boards, and we have discussed the
integration of the whole system into real astrophysical and adaptive optics
applications. The SPADA imager shows state-of-the-art performance, achieving
single-photon sensitivity in the visible range, together with low noise, high frame rate
(up to 20 kframes/s) and parallel readout. Moreover, the diodes are extremely
robust and work at low voltages, making them much more suitable for field and
space instrument applications than existing photon counting systems. On-field
performance within adaptive optics control loops is currently under investigation.
For future high-sensitivity imaging detectors, SPAD arrays with a larger number of
elements and subapertures, with square and linear layouts, are under development.
500 µs
Figure 23. Example of fast transient imaging. The SPADA detector was placed in front of
a dark CRT, with only one circular line highlighted. The figure shows ten frames of the
SPADA image, acquired with a 50 ms integration window and frames separated by 0.5 ms from
each other. The motion of the electron beam is clearly visible as it paints the circle on the
CRT with a refresh frequency of 60 Hz. (The colour version of this figure is included in the
online version of the journal.)
188 F. Zappa et al.
Downloaded By: [Tisa,] At: 08:22 27 March 2007
Acknowledgments
This work was supported by the Italian Ministry of University and Research
(MIUR-PRIN project No. 2002021224). The authors wish to thank Mr Sergio Masci
for the excellent work done in the assembly of the detection head.
References
[1] R.K. Tyson, Principles of Adaptive Optics (Academic Press, San Diego, 1991).
[2] M. Downing, N. Hubin, M. Kasper, et al., in Proceedings of SDW2005, Taormina, 2005.
[3] D. Bonaccini, F. Rigaut, A. Glindemann, et al., Proc. SPIE 3353 224 (1998).
[4] S. Cova, M. Ghioni, A. Lotito, et al., J. Mod. Opt. 51 1267 (2004).
[5] M. Ghioni and G. Ripamonti, Rev. Sci. Instrum. 62 163 (1991).
[6] Photon counting detector module, datasheet, Micro Photon Devices, Bolzano, Italy.
Available at http://www.micro-photon-devices.com.
[7] H. Dautet, P. Deschamps, B. Dion, et al., Appl. Opt. 32 3894 (1993).
[8] SPCM-AQ single-photon counting module data sheet, Perkin Elmer Optoelectronics.
Available at: http://opto.perkinelmer.com.
[9] S. Cova, M. Ghioni, A. Lacaita, et al., Appl. Opt. 35 1956 (1996).
[10] P. Antognetti, S. Cova and A. Longoni, in Proceedings 2nd Ispra Nuclear Electronics
Symposium, Stresa May 20–23, 1975, pp. 453–456, Euratom Publication EUR 537e
(1975).
[11] S. Cova, A. Longoni and A. Andreoni, Rev. Sci. Instrum. 52 408 (1981).
[12] R.G. Brown, R. Jones, J.G. Rarity and K.D. Ridley, Appl. Opt. 26 2383 (1987).
[13] M. Ghioni, S. Cova, F. Zappa and C. Samori, Rev. Sci. Instrum. 67 3440 (1996).
[14] H. Dautet, P. Deschampes, B. Dion, et al., Appl. Opt. 32 3894 (1993).
[15] F. Zappa, M. Ghioni, S. Cova, et al.,IEEE Trans.Instrum.Meas.496 1167 (2000).
[16] F. Zappa, A. C. Giudice, M. Ghioni and S. Cova, in Proceedings of ESSCIRC 2002,
Firenze, Italy.
[17] F. Zappa, S. Cova and M. Ghioni, US Patent Appl. No. 09/797,974, filed 5 March 2001;
European Patent Appl. No. 01200852.2-2217, filed 6 March 2001.
[18] F. Rigaut, B.L. Ellerbroek and M.J. Northcott, Appl. Opt. 36 13 (1997).
[19] F. Roddier, Appl. Opt. 23 1223 (1988).
[20] R. Arsenault, J. Alonso, H. Bonnet, et al., Proc.SPIE 4839 174 (2003).
[21] D. Bonaccini Calia, R.M. Myers, et al.,Proc.SPIE 5490 1315 (2004).
[22] http://www.orsys.de/327c6713.htm.
[23] A.L. Lacaita, F. Zappa, S. Bigliardi and M. Manfredi, IEEE Trans. Electron Dev.40
577 (1993).
Photon counting arrays for astrophysics 189
