Content uploaded by Thomas Frach
Author content
All content in this area was uploaded by Thomas Frach on Sep 01, 2015
Content may be subject to copyright.
Available via license: CC BY-NC-ND 3.0
Content may be subject to copyright.
Physics Procedia 37 ( 2012 ) 1546 – 1560
1875-3892 © 2012 Published by Elsevier B.V. Selection and/or peer review under responsibility of the organizing committee for TIPP 11.
doi: 10.1016/j.phpro.2012.03.749
TIPP 2011 – Technology and Instrumentation in Particle Physics 2011
Fully Digital Arrays of Silicon Photomultipliers (dSiPM) – a
Scalable Alternative to Vacuum Photomultiplier Tubes (PMT)
York Haemisch1, Thomas Frach1, Carsten Degenhardt1, Andreas Thon2
1Philips Technologie GmbH, Innovative Technologies, Pauwelsstrasse 17, D-52074 Aachen, Germany
2Philips Research Laboratories, Pauwelsstrasse 17, D-52074 Aachen, Germany
Abstract
Silicon Photomultipliers (SiPMs) have emerged as promising alternative to fast vacuum photomultiplier tubes (PMT).
A fully digital implementation of the Silicon Photomultiplier (dSiPM) has been developed in order to overcome the
deficiencies and limitations of the so far only analog SiPMs (aSiPMs). Our sensor is based on arrays of single photon
avalanche photodiodes (SPADs) integrated in a standard CMOS process. Photons are detected directly by sensing the
voltage at the SPAD anode using a dedicated cell electronics block next to each diode. This block also contains active
quenching and recharge circuits as well as a one bit memory for the selective inhibit of detector cells. A balanced
trigger network is used to propagate the trigger signal from all cells to the integrated time-to-digital converter. In
consequence, photons are detected and counted as digital signals, thus making the sensor less susceptible to
temperature variations and electronic noise. The integration with CMOS logic provides the added benefit of low
power consumption and possible integration of data post-processing directly in the sensor.
In this overview paper, we discuss the sensor architecture together with its characteristics with a focus on scalability
and practicability aspects for applications in medical imaging, high energy- and astrophysics.
© 2011 Elsevier BV. Selection and/or peer-review under responsibility of the organizing committee for
TIPP 2011.
Keywords: SiPM; digital SiPM (dSiPM); single photon detection, timing resolution, energy resolution, Čerenkov, scalability
1. Introduction
Recently, Silicon Photomultipliers (SiPMs) gained interest as a potential candidate to replace
photomultiplier tubes (PMT) in several applications for reasons of ruggedness, compactness and
insensitivity to magnetic fields [1]. Other advantages of solid state detectors in general are their relatively
Available online at www.sciencedirect.com
© 2012 Published by Elsevier B.V. Selection and/or peer review under responsibility of the organizing committee for
TIPP 11. Open access under CC BY-NC-ND license.
Open access under CC BY-NC-ND license.
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1547
low operating voltage, low power consumption and large scale fabrication possibilities. Today, silicon
photomultipliers almost exclusively operate in an analog way. The passively-quenched Geiger-mode cells
of the SiPM are connected in parallel through long interconnects, and the resulting output signal is
therefore the analog sum of the individual currents of all cells. Hereby, the good intrinsic performance of
the SPAD is not fully utilized, as the generated signal is deteriorated by the parasitic capacitances of the
on-chip interconnect, the bond wires and the external load. From a system perspective, large scale
applications of analog silicon photomultipliers imply some design challenges: In systems comprising
several tens of thousands of channels, dedicated readout chips are needed to condition and digitize the
SiPM signals. As the single photon response is still in the mV range, the signals can be easily affected by
electronic noise or unstable baseline due to high dark count levels, thus making single photon detection
difficult. A typical single-channel sensor/readout system for the detection of scintillation light based on
the analog silicon photomultiplier (aSiPM) is shown in fig. 1 (left).
Figure 1: Scintillation light detector systems based on the analog (left) and digital silicon photomultipliers (right).
Similar functionality can be realized in a single chip according to the scheme shown in fig. 1 (right).
Here, the SPADs are integrated with conventional CMOS circuits on the same substrate. Each SPAD has
its own readout circuit, which also provides means for active quenching and recharging of the SPAD. A
one bit memory cell integrated next to the SPAD can be used to selectively enable or disable the
respective diode. Each cell, composed of the SPAD itself and the corresponding electronics block, is
connected to the time-to-digital converter via a configurable, balanced trigger network. A separate
synchronous bus is used to connect each cell to the photon counters to determine the number of detected
photons. Eventually, correction look-up tables and other data post-processing could be implemented on
the same chip.
Integration of additional logic next to the light sensing element can unlock the true potential of this
new light detector in applications requiring very high time resolution while detecting only few photons
per event. For example, the position of the photon can be detected and stored to the accuracy of the SPAD
size, typically several tens of microns. Additionally, this information can be used to correct for the
propagation delay non-uniformity of the trigger network and so to improve time resolution. The above
figure illustrates also one characteristic of the new concept: by integrating part of the readout on-chip, the
sensor becomes an application specific integrated circuit (ASIC). In this particular example, the digital
Silicon Photomultiplier is intended to be used as a detector for scintillation light. A modified concept is
needed to make the sensor also a good candidate for e.g. Čerenkov light detectors [2]. Extensions of the
current digital Silicon Photomultiplier architecture to this effect will be presented below.
2. Digital Sensor Architecture & Operation (from [2])
Like in its analog counterparts, digital silicon photomultiplier pixels consist of arrays of Geiger-mode
microcells (SPADs), each capable of detecting single photons. Contrary to the analog SiPM, however,
each cell is capable of detecting and storing exactly one photon. Upon the detection of a photon, the
1548 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
avalanche is actively quenched using a dedicated transistor, and a different transistor is used to quickly
recharge the diode back to its sensitive state. SPAD breakdown results in an immediate voltage change of
approximately the excess voltage at the anode. Upon reaching the inverter threshold, the anode voltage is
Figure 2: Close-up of a single cell (SPAD) and simplified schematics of electronics.
forced to the breakdown voltage level by closing the quenching transistor, thereby stopping the current
flow through the diode. The combination of the diode capacitance and the quenching transistor feedback
ensures proper storage of the information. The quenching transistor is disconnected and the diode is reset
to sensitive state by a separate recharge transistor (fig. 2).
Each cell provides a fast asynchronous trigger signal and a slower synchronous data output signal. The
trigger signal of each cell is connected to a balanced, low skew trigger network connected to the on-chip
time-to-digital converter (TDC). The trigger network can be configured to start the TDC at the detection
of the first photon or, alternatively, higher photon thresholds. In case of first photon trigger, the trigger
network is equivalent to a balanced, distributed “all cells”-to-1 OR-gate. The discrimination of higher
photon thresholds is described in [6]. The data output signals of all cells in a column are connected to the
same data line and the data is applied to the data line using a row output enable signal. Then, the acquired
data can be read out by selecting individual rows of the array one after the other and reading the data lines
at the periphery of the pixel.
To simplify the design of the chip, a pixel is composed of four identical subpixels, each consisting of
an array of 64 x 32 cells (demonstrator test chip). The cell size is 52 x 30 μm² each resulting in a 50% fill
factor including the cell electronics. The total pixel size of 3840 x 3328 μm² therefore contains a sensitive
area of 6403016 μm². One diode has been omitted in each subpixel to make space for the trigger logic at
the center of the pixel, thus resulting in 2047 cells per subpixel. Each subpixel has its periphery logic on
two adjacent sides to control the access to the cell inhibit memory and to provide the means to readout the
Figure 3: (a) Microphotograph of the demonstrator (b) Acquisition sequence including typical
chip DLDK8. The size is 4.555x5.253 mm². timing values.
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1549
data and recharge the diodes. An integrated time-to-digital converter is connected to the trigger block at
the pixel center via the master trigger line. A microphotograph of the technology demonstrator test chip is
shown in Fig. 3a. The TDC is located on top of the SPAD array and has a block size of 96 x 3300 μm².
All measurements presented in this paper were obtained using a sensor with 25 V breakdown voltage. The
actual bias voltage of the sensor equals the breakdown voltage plus the 3.3 V excess voltage.
A typical data acquisition sequence is shown in Fig. 3b. This sequence is tailored to the acquisition of
scintillation pulses [4] and has to be modified to reduce the sensor dead time when used for other
applications with lower light fluxes (e.g. Čerenkov light detection). The timings shown in the figure are
either defined by the sensor architecture and SPAD properties (readout and recharge phase) or
programmable (validation and collection) to allow the user to adapt the acquisition sequence to the decay
time of various scintillators.
The pixel state machine starts in the state READY with all diodes charged above their breakdown
voltage and recharge transistors open. A trigger signal starts the acquisition sequence, forcing the pixel
controller to change the state to VALIDATE. The pixel controller stays in this state for a user-defined
time of 5 to 40 ns. After the validation hold-off timer expired, the validation signal is checked to
determine if the event is indeed a real light pulse or dark count. In case of a dark count generated trigger,
the pixel state machine changes to RESET to quickly reset the pixel and change back to the READY state
in preparation for the next event.
In case of a real scintillator pulse, the validation threshold is reached and the state machine changes to
the state COLLECT. While in this state, the pixel waits for the scintillator pulse to decay. The photons
impinging on the sensor are detected and stored in the cells for later readout. After the expiration of the
timer, the pixel state machine switches to the state READOUT. In this state, each line of the sensor is
selected separately and the number of photons detected in the line is added to the photon accumulator.
Finally, the pixel controller goes to the RESET state for global pixel recharge and TDC reset, and then
back to the READY state. A more detailed description of the sensor can be found in [6].
3. Digital Sensor Performance of current sensor
A. Intrinsic Timing & Triggers (from [2])
The new sensor is designed to operate on 200 MHz reference clock, which can be supplied either via
single-ended (LVCMOS) or differential (LVDS) inputs. Also, the electrical test input (SYNC) used to test
and calibrate the TDCs can be supplied either in single-ended or differential configuration independent of
selected clock configuration, to provide maximum flexibility for the system builder. When using
differential inputs for clock and sync, the single-ended IO pads can be internally re-configured to provide
additional external trigger input and internal trigger output. Using these signals, multiple sensors can be
connected to trigger groups, so one sensor is capable to start acquisition in one or more of its neighbors.
The acquired event data is transferred via two 100 MHz serial data links to the readout infrastructure. The
transfer is independent of the acquisition and no additional dead time is introduced. Even for the
minimum acquisition time, the sensor is capable of outputting data faster than acquiring new data.
The timings of the individual phases of the acquisition sequence (see figure 3b) can be programmed
using the JTAG interface, which is also used to program the cell inhibit memory allowing to switch of
cells with high dark count levels. The integrated time-to-digital converters have a typical bin width of
24 ps under nominal operating conditions. Fig. 4a shows the distribution of the bin width values for one
particular sensor. Two time-to-digital converters running with complementary 100 MHz clock have been
1550 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
Figure 4: (a) Bin width histogram of the two time-to- (b) Cross-correlation histogram of the two
digital converters. complementary time-to-digital converter time
stamps for randomly generated events.
integrated to provide at least one valid time stamp for every event. To counteract the meta-stability close
to the reference clock edge, the TDC design has been modified and a 500 ps wide exclusion window has
been added. As both TDCs are running 180° out of phase, both TDCs deliver a valid time stamp for 90%
of all events, which can be used to further improve the timing accuracy.
The cross-correlation between the time stamps of the two TDCs is shown in Fig. 4b. This two-
dimensional histogram is generated by using the time stamps of random events. The histogram can be
used to generate the look-up table to correct for the TDC non-linearity.
The sensors employ a trigger network to provide a low-skew signal path from each micro-cell to the
time-to-digital converters. It also provides some configuration options meant to increase the trigger
threshold to higher photon levels to minimize system dead-time due to dark count related triggers. The
basic function of the trigger network has been explained in [7] with the comment that the trigger
thresholds except for the first photon are statistical thresholds. This behavior is now shown in the
following figure 5. The graphs show the individual probability of the n-th photon to generate the trigger,
based on an analytical model of the trigger network.
Figure 5: (a) The trigger generation probability for the (b) Total trigger generation probability for the
n-th photon as a function of the selected number of detected photons as a function of
trigger level. the selected trigger level.
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1551
As shown in fig. 5a, for the ≥4 photon trigger network configuration, the 6th photon has the highest
individual probability to generate the actual trigger signal. The actual trigger probability, as displayed in
fig. 5b, is given by the sum of the individual probabilities over the number of detected photons. Fig. 5b
indicates that for the ≥4 photon trigger configuration, at least 7 photons have to be detected in order to
reach 50% trigger probability level.
The validation logic is based on the same principle. The only difference is the number of regions the
pixel is partitioned to, namely 32 in the current sensor configuration. This, together with the variable
validation interval, allows one to set a higher energy threshold to discriminate real events from dark count
events to reduce the system dead-time [2].
An important goal of a sensor development aiming at high timing accuracy for e.g. Time-of-Flight
Positron-Emission-Tomography (TOF-PET) applications is to achieve good timing resolution across
larger sensor arrays. We have been testing the intrinsic timing resolution (jitter) of arrays with 8 x 8 pixels
by means of electronic and picosecond laser pulses (24 ps FWHM), as depicted in fig. 6. The time stamp
differences between the two sensors are shown in the graphs on the bottom. The full-width-half-max
(FWHM) of the distribution indicates the timing jitter to be 44 ps for the electronic measurement, which
is mainly due to the TDC’s and the clock distribution network, and 59 ps for the laser pulses, which adds
the contribution of the Geiger-mode cells. For these measurements the sensors were triggered on the first
photon.
Figure 6: Timing jitter of dSiPM sensor arrays (8x8 pixel) as measured with electronic trigger (left) and picoseconds
laser pulses (right). The measurement setup is shown on top.
B. Photon Detection Efficiency
In order to be able to trigger on the first photon arriving, a high Photon Detection Efficiency (PDE) is
desirable. PDE is primarily influenced by the thickness of the active layer, the Quantum Efficiency (QE)
of the diodes, the fill factor of the sensor surface and the reflectance of the surface layers.
Fig. 7 depicts the PDE of the DLDK8 demonstrator chip as shown in fig. 3 as a function of the
incident wavelength emitted from a calibrated light source. The peak value of about 30% is achieved at
430 nm. The spectrum already points to a number of improvement possibilities as discussed later. The
steep decline towards dark blue and UV on the left is attributable to surface reflection processes and could
probably be shifted further to lower wavelengths for example by well designed anti-reflective coating
(ARC).
1552 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
Figure 7: Photon Detection Efficiency of the demonstrator chip DLDK8 as shown in fig. 3 (red curve) as a function
of the light wavelength. The measurement was performed on one cell at a time so that optical crosstalk is
excluded. For comparison the emission spectra of 2 scinitillators for PET are shown (blue and violet
curves).
C. Dark Count Measurements& Temperature Sensitivity
The design of the sensor as described in chapter 2 allows a control of the dark count rate (DCR) at the
individual SPAD level. By subsequently switching on and off each individual SPAD cell in a completely
dark environment, dark count maps and histograms like the one shown in fig. 8 (a) can be generated. As
becomes clear from the histogram and the cumulative logarithmic plot in fig. 8 there is an over
proportional contribution of a few cells to the DCR of the sensor. Since the digital architecture allows to
Figure 8: (a) Histogram of dark counts of cells of a (b) DCR as a function of the number of active cells.
subpixel taken at room temperature.
address each cell individually the DCR can be lowered by orders of magnitude by switching of the “bad”
cells. This feature also contributes to the much reduced temperature sensitivity (compared to analog
SiPMs) of the sensor as is shown in figure 9. Typical values of active cells are ranging between 90-98%.
The measurement was carried out with a pulsed picosecond laser (24 ps FWHM). The photopeak changes
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1553
by 0.33%/°C, which is an order of magnitude lower than with analog SiPMs. The PDE drift can be
compensated for by adapting the bias voltage. The TDC and trigger network drifts cause a change of the
time of 15.3 ps/°C. TDC offsets can be periodically re-calibrated using the SYNC-input of the sensor.
Figure 9: (a) PDE drift as function of the temperature and (b) TDC drift as function of temperature
measured with picosecond laser pulses (24 ps FWHM) (no bias adjusted).
D. Performance with Scintillators (from [3])
Fig. 10 shows a schematic of the measurement setup that was used for coincidence measurements with
scintillator crystals and arrays, similar to fig. 6. The dSiPM arrays are mounted on printed circuit boards
containing an FPGA that provides the sensors with a 200 MHz reference clock and configures the sensors
via their serial interface. One board contains a 200 MHz master oscillator that is distributed via an
Ethernet cable to the second board. Scintillator arrays containing 8 x 8 LYSO crystals with 4 mm x 4 mm
pitch and 22 mm length were optically coupled to the detector arrays, leading to a 1:1 coupling between
the crystals and the dSiPMs. The individual scintillator crystals were separated by reflecting foil.
Figure 10: Schematic of the measurement setup. LYSO scintillator arrays are coupled to arrays of digital Silicon
Photomultipliers. Readout boards with FPGAs are used to provide the dSiPMs with a 200 MHz reference
clock and to configure and control the sensors. The data is transferred to a PC via a USB interface.
Recording all the timestamp differences using the above described LYSO scintillator arrays coupled to
the dSiPM arrays generates a histogram presenting the sum of all individual timestamp differences
recorded for the individual pixels. Fitting a Gaussian distribution to the data yields the coincidence timing
resolution (CTR) of the setup, as shown in fig. 11b, of 283 ps (FWHM). In order to prove that the timing
axis is calibrated correctly, two 22Na sources, 8.3 cm apart, were put between the two arrays. Two peaks
in the timestamp difference histogram were recorded, separated by 570 ps or 8.6 cm, which confirms the
correct calibration of the timing axis. The measurements were performed at 10°C.
1554 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
Figure 11: (a) Timing resolution as a function of the pixel position and (b) histogram of time stamps for the
measurement as shown in fig. 10 using DPC3200-22-44 sensors with 8x8 pixels at 10°C.
Another important parameter of a scintillation detector is its energy resolution. The light detector as an
important part of the detection chain adds to the intrinsic energy resolution of the scintillation crystal used
[7, 8].
Figure 12: (a) Energy resolution as a function of the pixel position and (b) energy resolution summed over all pixels
of the sensor for the setup according to fig. 10 using DPC3200-22-44 sensor arrays at 10°C.
Energy resolution is obtained by summing up the number of events in the photopeak per pixel. Prior
to determining the FWHM of the photopeak, the spectrum is corrected for saturation. Since the number of
active cells, N, and the number of triggered cells, k, is known, the number of photons, p, can be calculated
according to ൌെܰכሺͳെ
ேሻ. Fig. 12b shows the summed energy spectrum over all pixels and
yields a global energy resolution for the configuration described above of 10.4% (FWHM).
Both the coincidence timing resolution (CTR) as well as the energy resolution are depending on the
geometry of the scintillation crystals used which determines the light collection properties, light travel
paths etc. Typically small and long crystals with high aspect ratio deliver worse results than shorter and
larger ones. We have been measuring a CTR of 153 ps (FWHM) for LYSO crystals of 3x3x5 mm³, values
of down to 120 ps (FWHM) for Calcium doped LSO of the same geometry have been reported using the
digital sensor arrays [8,9]. The timing resolution is mainly affected by the aspect ratio of the crystals
which determines the travel length of the light affecting the arrival time of the first photon.
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1555
E. Čerenkov Light Detection (from [7])
We successfully tested the technology demonstrator as a sensor for Čerenkov light. The test setup
shown in Fig. 13 was developed by the team of Prof. Dueren at Giessen University, Germany and tested
at Philips Digital Photon Counting in Aachen using a picosecond-pulsed laser beam [4].
For the beam test, the Y-shaped Čerenkov radiator made of acrylic glass was coupled via
approximately 1 mm air gap to two DLD8K demonstrator chips described in [6] connected in
coincidence. The measurements were carried out using 120 GeV protons from the SPS test beam at
CERN. The test setup was operated in first-photon trigger mode and all events have been validated. The
setup temperature of 3°C was controlled by a thermoelectric cooler and 2% of the diodes with the highest
dark counts have been disabled to further reduce the dark count rate of the sensors. The measured dark
Figure 13: (a) Čerenkov light detector setup. The Y-shaped (b) Coincidence time resolution of the Čerenkov
radiator is coupled via a small air gap to two light detector setup in (a).
digital SiPMs connected in coincidence.
count rates for both sensors were in the range of 500 kHz. Additionally, an external gate signal was used
to reduce coincidences due to the low beam duty cycle of only 16%, to reduce the data file size.
Measurement times ranged from few minutes to several hours. Coincidence resolving time of σ = 86 ps
was measured for both sensors (fig. 13), resulting in σ = 61 ps per sensor. As the skew in the trigger
network is currently the major contributor to the time resolution for single photons, re-balancing and fine-
tuning of the trigger network will make it possible to improve the time resolution of the sensor to
30 ps - 40 ps at the single-photon level.
4. Optimization Possibilities towards Single Photon Detection (from [2])
The current versions of our digital Silicon Photomultiplier have been developed primarily for
scintillation light detection. This resulted in an acquisition sequence which is not optimal for single
photon detection where the dark count events cannot be separated from true events. As a consequence we
observe a relatively large dead time of the sensor. The acquisition sequence could be optimized to reduce
that dead time under the assumption that at most one photon is impinging on the sensor per event. The
saving comes from skipping the readout phase of the sequence. This can be made optional to allow the
same sensor to be used for both scintillation and Čerenkov light detection. Fig. 14 shows the difference in
the state machine diagrams. Furthermore, reduction of the dark count rate further helps to reduce the dead
time.
1556 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
Figure 14: Original and modified acquisition sequence.
Second area of optimization is certainly the photon detection efficiency. A new improved version of
the digital Silicon Photomultiplier has been developed and tested (fig. 15a). The new sensor has 59.4 μm
x 64 μm cells with 78% fill factor. The peak photon detection efficiency (PDE) can reach more than 50%
in single diodes but strong interferences due to the presence of nitride-oxide layers in the chip
interconnect stack lead to a significantly lower average PDE (Fig. 15b). However, future optimizations of
the optical interface will enable higher average PDE and improved sensitivity below 400 nm.
Figure 15: (a) Layout of new dSiPM with improved (b) Typical photon detection efficiency of the new sensor.
area efficiency. The red line indicates the moving average over ± 20 nm.
Furthermore, the skew in the trigger network currently limits the single photon time resolution. To
achieve optimal performance, the trigger network needs to be manually tuned to minimize its skew. This
has already been done in simulation and needs to be verified in silicon. Nevertheless, even a perfectly
balanced trigger network will suffer from process variations. Therefore, further improvement in the time
resolution could be achieved if the position of the triggered diode was known. As there is at most one
photon detected per pixel, this can be realized using a scheme shown in fig. 16.
Every cell is connected to a horizontal and a vertical data line. When a cell detects a photon, the position
of the cell becomes visible at the periphery of the array as a signal on these two lines. The trigger signal
can be used to latch-in the data lines into a register, which can be evaluated to calculate the diode position
as a binary value. As this position is directly linked to the skew map, the skew map can be used to
completely correct for the actual trigger network skew of the sensor.
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1557
Figure 16: Potential future sensor architecture with position readout.
Optical crosstalk in the sensor and random coincidence of the photon with a dark count will make the
position information ambiguous. Normally, this situation can be easily detected, as more than one line
becomes active, and the event could be discarded. However, as optical crosstalk occurs most likely
between neighboring cells the approximate position could still be estimated if two neighboring lines
become active. This information could be used to reduce the loss of events due to optical crosstalk.
5. Scalability of Digital Light Sensors
From the beginning it has been our goal to develop not only a new kind of light sensor but a
(disruptive) technology that could be scaled up to be useful in many applications such as medical
imaging, high energy- or astrophysics detection or analytical instrumentation, basically everywhere today
vacuum photomultiplier tubes (PMTs) are still used. In order to achieve this goal we went from one pixel
to several 1000 pixels (Feb. 2012) within a relatively short time frame. The number of pixels operational
Figure 17: Evolution of the Philips dSiPM technology: the number of pixels doubled every 3 months.
1558 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
basically doubled every 3 months, as can be seen in fig.17. This rapid increase is a direct consequence of
the full utilization of the early signal digitization, the high level of integration and the complete absence
of analog electronics. Pre-processing and configuration can be done at sensor level, sensors can be
combined into larger units (modules) with intermittent data concentration and clean up steps and such
modules can form larger detector areas such as planar cameras or rings. All the necessary electronics can
be built from digital, off-the-shelf-components, thus reducing the overall costs of such detectors.
A very important feature of scalable technology is its energy consumption. A DPC digital sensor
consumes on average 1.1 W of power whereas a detector module as shown in fig.17 consumes about
10 W at high rates, including electronics. Table 1 provides an overview of the digital light sensor arrays
that are currently being produced and investigated. They fundamentally differ in the number of cells
(SPADs) per pixel and cell size, thus in fill factor, PDE and dynamic range. However, they have the same
“outer” geometry in terms of their size, the pixel matrix and pixel pitch. Also, they can be operated with
the same interface boards, firmware and software.
Table 1: Digital Light Sensor Arrays produced and investigated
Feature
DPC 3200-22-44
DPC 6400-22-44
Sensor array size (mm²)
Sensor Matrix
Pixel size (mm²)
32.6 x 32.6
8 x 8 pixel
3.328 x 3.840
Pixel pitch (mm), all directions
4
Cells (SPADs) per pixel
Cell size (μm²)
Fill factor (%)
PDE (%) @ 430 nm
3200
59.4 x 64
54
~ 50
6396
59.4 x 32
78
~ 30
Figure 18: Philips DPC digital light sensors front- and back side (left) and integrated electronics and interface
structure (right).
Fig. 18 illustrates the layout of the sensors with their electronic components on the back and the
corresponding interface structure. The FPGA ensures the proper clock distribution, allows data collection
and concentration, TDC linearization, saturation- and skew correction. The flash memory contains the
FPGA firmware, the configuration files and inhibit-memory maps that allow to reduce the DCR or to
generate patterns of active/inactive cells on the sensor. The built-on temperature sensor enables to record
York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560 1559
the sensor temperature at all times. The sensors are four-side buttable so that larger areas of detector can
be built with minimal gaps. MR compatible versions of these sensors are under development.
6. Summary
Fully integrated digital arrays of silicon photomultipliers (dSiPM) have been produced and
characterized. Their physical performance parameters have been obtained and their functionality has been
tested and optimized. Further optimization is possible in the direction of single photon counting by
improving the timing and trigger networks and increasing the PDE. Respective developments are
underway.
It has been shown that this technology is fully scalable, thus providing an elementary pre-requisite for its
industrial application. The number of pixels could be doubled every 3 months and most recently a full
system (a demonstrator ring) has been built and is currently being tested at the Philips DPC labs in
Aachen, Germany.
Based on the results obtained so far the technology developed will have the potential to replace long-
time used vacuum photomultiplier tubes in many applications, especially where accurate timing, very low
light levels, low power consumption, compact size and magnetic fields are the challenges.
Acknowledgements
The authors thank their colleagues Andre Salomon, Peter Michael Dueppenbecker, Torsten Solf and
Volkmar Schulz from Philips Research Germany for their experimental support and the always
constructive discussions.
References
[1] D. Renker and E. Lorenz, "Advances in solid state photon detectors," Journal of Instrumentation, vol. 4, no. 4, p. 404, 2009.
[2] T. Frach “Optimization of the Digital Silicon Photomultiplier for Cherenkov Light Detection“, JINST 7, C01112, 2012
[3] C. Degenhardt, B. Zwans, T. Frach, and R. de Gruyter, "Arrays of Digital Silicon Photomultipliers – Intrinsic Performance
and Application to Scintillator Readout”, NSS-MIC Conference Record, 2011.
[4] A. Hayrapetyan et al., "New digital SiPMs from Philips: Applications and first tests," Detector Workshop of the Helmholtz
Alliance "Physics at the Terascale", Heidelberg University, Oct. 2010.
[5] C. Degenhardt et al., "The Digital Silicon Photomultiplier - A novel sensor for the detection of scintillation light”, NSS-MIC
Conference Record, 2009.
[6] T. Frach et al., "The Digital Silicon Photomultiplier - Principle of operation and intrinsic detector performance," NSS-MIC
Conference Record, 2009.
[7] T. Frach, G. Prescher, C. Degenhardt, B. Zwans, "The Digital Silicon Photomultiplier - System architecture and performance
evaluation", NSS-MIC Conference Record, 2010.
[8] C.M. Pepin, P. Bérard, A-L. Perrot, C. Pépin, D. Houde, R. Lecomte, C. L. Melcher, and H. Dautet, Transactions on
Nuclear Science, Vol. 51, No. 3, p. 789-795, 2004
[9] W. Chewpraditkul, L. Swiderski, M. Moszynski, T. Szczesniak, A. Syntfeld-Kazuch, C. Wanarak, and P. Limsuwan,
“Scintillation Properties of LuAG:Ce, YAG:Ce and LYSO:Ce Crystals for Gamma-Ray Detection”, IEEE
Transactions on Nuclear Science, Vol. 56, No. 6, December 2009
1560 York Haemisch et al. / Physics Procedia 37 ( 2012 ) 1546 – 1560
[10] D. R. Schaart, H. T. van Dam, G. J. van der Lei, S. Seifert, “The Digital SiPM: Initial Evaluation of a New Photosensor for
Time-of-Flight PET,” 2011 IEEE Nuclear Science Symposium and Medical Imaging Conference, Valencia, Spain,
October 23-29, 2011, abstr. MIC15. S-137.
[11] D.R. Schaart, H.T. van Dam, G. J. van der Lei, S. Seifert, “Initial Evaluation of Digital Silicon Photomultipliers for Time-
of-Flight PET,” 2011 World Molecular Imaging Congress (WMIC 2011), San Diego, CA, September 7-10, 2011.
