2009 IEEE Nuclear Science Symposium Conference Record

N25-152

Impact of Dark Counts in Low-light Level Silicon

Photomultiplier Multi-readout Applications
I.F. Castro, A.J. Soares, J.F. Veloso
I3N, Physics Department, University of Aveiro
Campus Universitrio de Santiago 3810-193 Aveiro, Portugal
ifcastro@ua.pt

AbstractThe Silicon Photomultiplier (SiPM) generates noise

due to thermal excitation, producing dark counts. This may be a
critical performance drawback of the SiPM for low light level
applications, despite its otherwise promising features such as
high internal gain and quantum efficiency, low power
consumption and insensitivity to magnetic fields. The dark count
rate is highly dependent on temperature and bias voltage,
increasing with both. Dark count pulses are similar to single
photon interactions and introduce complex challenges in low
light level measurements. In applications where the pulse
integration time is dependent on the decay time of a slow
scintillation counter, the longer the signal integration the higher
the probability of dark counts during the integration period, and
this can be critical for detecting very low number of photons in
that period. For higher light levels, the effects of dark counts can
be reduced by setting an appropriate threshold. This work
presents Monte Carlo simulation techniques to evaluate the
impact of dark counts into position detection algorithms, in low
light level imaging applications where multiple SiPMs are used to
detect position of interaction.

I.

INTRODUCTION

As a solid state device, the Silicon Photomultiplier (SiPM)

generates noise due to thermal excitation, i.e., it produces an
electric signal even in the total absence of light (dark counts).
Since it operates in Geiger mode, these dark counts produce
output signals identical to single photo-electron events which
introduce complex challenges in applications where low light
levels should be detected in simultaneously triggered events.
Depending on the charge integration time required, and on the
dark pulse rate of the SiPM, the dark pulse signal may be
equivalent to multiple photo-electron pulses and, for gated
applications, this may severely impact the usability of this
device for low light level detection. In some applications, the
integration time depends on the decay time of the light source
(e.g. scintillation counter) and it may be required to extend it
to maximise the signal. However, this can be a critical
problem in detecting a very low number of photons. For
higher light levels, the effects of dark counts can be
significantly reduced by setting an appropriate threshold.
The number of output dark pulses measured with a
threshold of 0.5 p.e. is usually defined as the dark count
(number of times that one or more photons are detected). In
some cases, a threshold of 1.5 p.e. in dark count measurements
is used to evaluate optical crosstalk, which consists of the

excitation of neighbour cells due to photon emission during an

avalanche discharge in a given cell [1,2]. Afterpulsing is
another source of noise of SiPM, which arises from charge
carrier trapping and delayed release, due to traps formed in the
breakdown volume of the silicon [3]. In this work, dark counts
refer to all these three contributing sources of dark noise.
II.

A. Setup
Photon-counting spectra of SiPM devices have been
obtained using the Hamamatsu S10362-11-100-U MPPC and
characterized in terms of their background noise. The dark
rates and their variation with temperature and bias voltage were
measured in a dark chamber specially built for this purpose, as
depicted in Fig. 1. An Ortec 710 power supply was used to bias
the SiPM between 69 and 70V. The SiPM output was initially
amplified with a linear amplifier (Canberra 2111) with no
shaping and fed into a peak sensing MCA. This was later
replaced by a custom amplifying circuit (Fig. 2) which was
developed and included inside the dark chamber, providing
similar results. To study the variation of the dark rate with
temperature, a cooling module was set up, consisting of a
thermoelectric Peltier module with a heat-sink and extracting
fan, controlled with a thermocouple.
Measurements were also taken with low light levels. A high
resolution pulse generator (BNC PB-5) was used to illuminate
a 470 nm LED and also to trigger the ADC gate for signal
acquisition from the SiPM. A BC-91A wavelength shifting
optical fiber was used between the LED and the SiPM. SiPMs
are known to be a good choice to detect light emitted from
scintillating fibers [4,5].

This work was supported by project GAMACAM through FEDER and

ADI (Lisbon) programs. I.F. Castro is grateful to portuguese Agncia de
Inovao for funding this project via QREN project 1607.

9781-4244-3962-1/09/$25.00 2009 IEEE

EXPERIMENTAL MEASUREMENTS

1592

Figure 1 Simplified scheme of the experimental setup.

Figure 2 Pre-amplifying circuit developed for the Hamamatsu MPPC.

B. Results
The following experimental results were obtained using the
initial amplifying scheme. The dark rates for different bias
voltage values were measured at room temperature and are
shown in Fig. 3, where it is clearly visible the dependence with
bias voltage and with detection threshold, as expected. The
photo-electron pulse-height spectrum for one of the test
conditions (Vb = 69.4V) is shown in Fig. 4. The single electron
peak is clearly visible (second peak from the left), and so are
the 2nd and even 3rd electron peaks, corresponding to multiple
simultaneous dark pulses within the device. Results of the dark
rate variation with temperature, for different threshold values,
are summarized in Fig. 5, where no light source was used. It is
seen that the dark rate (at a threshold of 0.5 p.e.) becomes
approximately half at every 9C.

Figure 5 Dark rate vs. temperature (constant gain).

The ADC spectrum of the time gated signal obtained with the
pulsed LED coupled to an optical fiber is shown in Fig. 6. The
high photo-electron resolution of the SiPM allows a clear
separation of several electron peaks. It can be seen that for very
low light levels, the single electron peak coincides with the
dark pulse shown above.

Figure 6 Low light response spectrum. Vb=69.4V, T=25C.

III.

Figure 3 Dark rate vs. bias voltage, T=25C.

SIMULATION

A. Method
The SiPM charge signal from primary photoelectrons
follows an approximate Poisson distribution. The charge
amplification process and other events such as small
differences between pixels or photon interactions in recharging
pixels, introduce Gaussian spreads in the signals collected [6].
Therefore, the expected SiPM charge spectrum is a Poisson
distribution convoluted with Gaussian distributions, as seen in
Fig. 4. A MATLAB application was developed to simulate
SiPM charge signals, for a given Poisson mean  and Gaussian
standard deviation , ignoring counts below a given threshold
value. The values of  and  were taken from experimental
data, for different conditions of temperature, bias voltage and
charge integration time. The value of  was found to vary
approximately between 0.2 and 0.75 and  between 0.25 and
0.3. As shown in Fig. 7, the simulated spectrum shows a good
agreement with the acquired one, considering that the
simulation ignores the electronic noise.

Figure 4 Dark pulses spectrum. Vb=69.4V, T=25C.

1593

D. Results

Figure 7 a) Acquired spectrum: Vb=68.8V, T=20C. b) Simulated

spectrum:  = 0.6,  = 0.3.

B. Simulated setup and conditions

To test a detection apparatus of 120 per 120 SiPMs
triggered by a common gate, the application simulates 120 per
120 SiPM charge spectra acquired simultaneously, where 5 of
the SiPMs in each direction have a low light signal, i.e., have a
higher Poisson mean than the average for dark noise. The
configuration may consist for example of an inorganic
scintillation crystal coupled to two orthogonal layers of 120
wavelength shifting fibers, each of which is read out by a
SiPM. For the purpose of this study, the light levels reaching
the SiPM were considered to be 10 photons in the central fiber,
6 in the two neighbor fibers and 2 in the next surrounding ones.
The position of the central fiber was defined to be (10, 10) and
5000 counts (events) were simulated. Table I summarizes the
relevant parameters used in the simulation:
TABLE I.

SIMULATED CONDITIONS

Parameter

Simulated values

Number of events

5000
0.3, 0.5, 0.7
0.25
0.5, 1, 2, 3, 4
(10, 10)
10, 6, 2
20, 30, 40, 50%



Threshold
Position of the central fiber
Number of photons in the 5 fibers
Photo detection efficiency (PDE)

C. Position detection algorithms

Three position detection algorithms were used to calculate
the centroid for each count:

(1)

, where qi is the charge signal of the ith SiPM.

, where i is the SiPM with maximum charge signal.

, for the 3 neighbor signals which sum is maximum.

Figure 8 Example of simulation images obtained with the 3 position

detection algorithms, for PDE=50%, =0.3, =0.25, threshold=2.

(2)

(3)

The centroids in x and y directions were used to generate 2D

images, as exemplified in Fig. 8. The image efficiency of
these position detection algorithms, defined as the ratio
between the number of centroids calculated and the total
number of events, was determined for different sets of fibers or
image pixels (Fig. 9) and for different values of PDE,  and
threshold. Results are shown in Figs. 10 and 11. Thresholds
were defined to allow ignoring low signals from certain fibers.

1594

Figure 9 Sets of fibers (1 mm2 image pixels)

Figure 11 Algorithm efficiency vs. threshold, for a PDE of 30% (same

legend as Fig. 10).

Figure 10 Algorithm efficiency vs. threshold, for a PDE of 50%.

The reverse analysis was carried out, calculating the radius

around the central fiber within which 76% of events fall, as
shown in Fig. 12. This percentage level was chosen because it
represents the theoretical percentage for 2.35 of a Gaussian
distribution, i.e., the FWHM. Most of the times, and especially
for algorithms B and C, the centroids distribution is narrower
than a Gaussian one, so this value gives us an overestimated
approximation of the FWHM, i.e., of the spatial resolution.

1595

dark noise average. This happens because these algorithms rely

on maximum values.
Further work is currently under way for developing a small
prototype of the proposed configuration, where all SiPMs will
have a simultaneous readout triggered by a common gate based
on the scintillation signals obtained by a PMT, thus reducing
the impact of dark counts. A point image can then be created
and compared with the simulation results.
Preliminary measurements of the scintillation light retained in a
fiber, made both with a PMT and a SiPM, clearly show that
this external trigger is indispensable with the SiPM (Fig.
13).

Figure 12 Radius where 76% of events above threshold fall inside

(PDE=50%). The same legend as above applies.

CONCLUSIONS AND FURTHER WORK

An application was developed to simulate a configuration
where 240 SiPMs are used to readout wavelength shifting
optical fibers coupled to a scintillation crystal. In this kind of
configuration, the light levels at the end of the fiber are
expected to be in the order of only a few photons for gamma
rays in the range of 80 to 200 keV, so the SiPM is a very
interesting readout device due to its excellent photon counting
capability. On the other hand, the SiPM has high dark counts,
so it is of great importance to discriminate this low light signal
from the random dark noise signal.
From the experimental measurements, it is evident that the
stability of bias voltage and temperature is crucial for good
signal stability and dark rate control of each individual SiPM.
Cooling the SiPM significantly reduces the dark counts
approximately by half at every 9C, but on the other hand it
increases afterpulsing. Lowering the bias voltage reduces noise,
but with loss of gain and PDE [3], which is critical for lowlight level applications.
The simulation allowed us to evaluate the impact of the
simultaneously acquired dark counts into position detection
algorithms and the results show that an image with good
resolution can be obtained, despite the very low light levels. Of
the three algorithms tested, algorithm C shows the best results
for low light level centroiding, indicating that a sub-milimetric
spatial resolution is possible, as suggested by the results in Fig.
12. Algorithm A leads to high efficiencies but only for
thresholds over 3 and it is more vulnerable to changes in dark
noise. On the contrary, algorithms B and C show better
performance for lower threshold values, independently of the

Figure 13 Spectra of scintillation light from 241Am retained in one fiber,

compared to noise, with no external trigger applied.

ACKNOWLEDGMENT
I.F. Castro is grateful to the IEEE 2009 NSS-MIC General and
Scholarship Chairs for the Trainee Grant attributed.
REFERENCES
[1]
[2]

[3]
[4]

[5]
[6]

1596

A.G. Stewart et al., Performance of 1-mm2 Silicon Photomultiplier,

IEEE J. Quant. Electron., vol. 44, n. 2, 2008.
K. Yamamoto et al., Development of Multi-Pixel Photon Counter
(MPPC), IEEE Nucl. Sci. Symp. Conf. Rec., vol. 2, pp. 1094-1097,
2006.
D. Renker and E. Lorenz, Advances in solid state photon detectors,
J.Inst. 4, 4004-4052, 2009.
B.W. Baumbaugh et al., Studies of SiPM and Scintillation Plates with
Waveshifter Fiber and SiPM Readout, presented at the IEEE Nuclear
Science Symposium, Orlando, FL, 2009.
M. Yokoyama et al., Application of Hamamatsu MPPCs to T2K
neutrino detectors, Nucl.Instrum. Meth. A, 610, pp. 128-130, 2009.
P. Finocchiaro et al., Characterization of a Novel 100-Channel Silicon
Photomultiplier Part II: Charge and Time, IEEE Trans. Elect. Dev.,
vol. 55, n. 10, pp. 2765-2763, 2008.