What is an SiPM and how does it work?

Slawomir Piatek, PhD, Hamamatsu Corporation & New Jersey Institute of Technology
October 7, 2016

Related content: A technical guide to silicon photomultipliers (MPPC)

What is a silicon photomultiplier?
A silicon photomultiplier (SiPM) is a solid-state photodetector that in response to
absorption of a photon can produce a current pulse several tens nanoseconds long
containing 105 to 106 electrons. Therefore, a SiPM has a gain. This gain is comparable to
that of a photomultiplier tube (PMT).
Structure
A SiPM is a pixelated device where each pixel, or a microcell, is a series combination of an
avalanche photodiode (APD) and a quenching resistor (R Q). All of the microcells are
connected in parallel; thus, a SiPM has two prongs: an anode and a cathode. The three
panels in the figure below show, from left to right, a cross section of three microcells in a
possible architecture, top view — the side of impinging light — of the device, and an
equivalent electrical circuit.
Figure

1. This figure depicts a typical structure of a SiPM. It does not correspond to the actual
structure of the Hamamatsu product.
 
In practical SiPMs, the microcells of identical size are arranged in a rectangular pattern.
Depending on the device, the size of a microcell varies from 10 μm to 100 μm and the
number of microcells per device ranges from several hundreds to several tens of
thousands. SiPMs have active areas ranging from 1 mm 2 to 6 mm2 and spectral
sensitivities from UV to IR, peaking in the visible (400 nm - 500 nm).
Operation
A SiPM is externally biased so that the voltage on each APD is above its breakdown
voltage. Thus each APD operates in Geiger mode. The difference between the biasing
voltage and the breakdown voltage is known as overvoltage — the main adjustable
parameter controlling operation of the device. If a SiPM absorbs a photon, the resulting
charge carrier (an electron or hole depending on the structure) can trigger an avalanche in
the gain region (shown as the gray oval in the figure) within the p + - n+ structure. The
avalanche can produce 105 - 106 carriers; this constitutes the gain. The role of the
quenching resistor is to restore the APD back to the Geiger mode. The gain of a SiPM
1
depends linearly on the overvoltage and the value of junction capacitance. Even though a
SiPM is a pixelated device, it is not an image sensor. It does not store charge like a CCD
does. It is an analog device producing a time-varying output signal that is measured in
real time.
Characteristics
There are several key parameters characterizing a SiPM. These are breakdown voltage,
gain versus overvoltage relation, photon detection efficiency, crosstalk probability,
afterpulsing probability, and dark count rate.
Applications
A SiPM most resembles a PMT in its operation and characteristics. Compared to a PMT, a
SiPM requires a lower operating voltage; has immunity to magnetic fields; is sturdier and
more rugged; and is smaller in size. For these and other reasons, the SiPM has begun to
replace PMTs in low-light applications such as positron emission tomography (PET), light
detection and ranging (LIDAR), or radiation detection in high-energy physics.
How does a SiPM work?
A silicon photomultiplier (SiPM), though pixelated, is a photodetector that produces an
analog output signal in real time. The output is a time sequence of waveforms (or current
pulses), which have a discrete distribution of amplitudes: A, 2A, 3A, etc. The histogram of
the amplitudes depends on the intensity and time-characteristics of the incident light. This
note is a closer look into how a single microcell of a SiPM generates a waveform in
response to light.
Operation

Figure 2. A simplified equivalent circuit of a SiPM (two representative microcells) biased
with an external voltage source VBIAS.
 
All of the microcells in Figure 2 are in parallel. A microcell is a series combination of an
avalanche photodiode (APD) and a quenching resistor R Q. The circuit modeling the APD
is a parallel combination of a capacitor C J (representing the junction capacitance) with a
series combination of a switch S, voltage source V BD (equal to the breakdown voltage), and
a resistor RS (representing the resistance of the entire APD during a discharge).
In the absence of light and ignoring dark counts, S is open and the voltage on C J is
VBIAS (VBIAS > VBD) — the microcell is in a light-sensitive state and the APD is in Geiger
mode. When the microcell absorbs a photon, an electron-hole pair forms; one of the charge
2
carriers drifts to the avalanche region, where it can initiate an avalanche. At the instant the
avalanche begins, S closes causing C J to be discharging through RS (RS ≪ RQ) with the time
constant RSCJ. The voltage on CJ decreases, lowering the probability of impact ionization.
For optimal RQ, the probability eventually becomes so small that the avalanche quenches.
At this moment, S opens and VBIAS recharges CJ with the time constant RQCJ.

Figure 3. A graph of the current flowing through the terminals of the SiPM during the
process discussed above.
 
In Figure 3, the avalanche begins at t i and quenches at tmax. The leading edge of the pulse
starts at ti and the rise is proportional to [1−exp(−tRsCJ)][1−exp(−tRsCJ)]. The pulse reaches
the maximum value imax≈VBIAS−VBDRQ+RS=△VRQ+RSimax≈VBIAS-
VBDRQ+RS=△VRQ+RS at tmax and then declines as exp(−tRQCJ)exp-tRQCJ. Integrating the
current pulse with respect to time gives the total charge Q that has transferred between
the terminals of the SiPM. Since a single charge carrier caused the transfer, the
gain M=QeM=Qe. A detailed analysis shows that Q=CJ△VQ=CJ△V and,
thus, M=CJ△VeM=CJ△Ve. If two or more microcells "fire" nearly simultaneously in
response to incident light, the output signal is a linear superposition of the pulses, each
like the one depicted in the figure, giving an amplitude 2A, 3A, and so on. In contrast, if
the same pixel absorbs simultaneously two or more photons, the output pulse is identical
to that shown in the figure and the amplitude is A.
Gain of a silicon photomultiplier
An avalanche photodiode in a SiPM operates in Geiger mode. If a charge carrier triggers a
discharge in a microcell, the consequence is a current pulse with the total amount of
charge Q flowing through the terminals of the SiPM. The gain μ is defined as a ratio of Q
and the fundamental charge e = 1.6 x 10 ‐19 C. Typical values of μ are in the range of 10 5 ‐
107. This note is a closer look at the gain.

3
Anatomy of a 1 p.e. waveform

Figure 4. A simplified electrical equivalent circuit of a microcell externally biased with
VBIAS.
 
Figure 4 shows a simplified electrical equivalent circuit of a microcell that is externally
biased with VBIAS. In the absence of light and dark counts, the microcell is in Geiger mode:
the switch S is open, the voltage on the junction capacitance C J is VBIAS, and no current
flows through the quenching resistor R Q or the series resistance RS. At the instant a charge
carrier triggers a discharge (avalanche), the switch S closes and C J begins to discharge
through RS (shown in the figure with the red swirl) causing a voltage drop on R Q and,
thus, a current through the terminals of the SiPM.

Figure 5. Graph of current pulse.

 
Figure 5 shows the current pulse begins at t i. The current rapidly increases until it reaches
the maximum value imax = (VBIAS - VBR)/(RQ + RS) at tmax, which is on the order of ~1 ns. At
tmax, the voltage on the APD drops to approximately V BR, which is not enough to sustain
the discharge. Quenching occurs. The junction capacitance C J begins recharging, causing
the voltage across RQ to be decreasing and, thus, the current through the terminals of the

4
SiPM to be also decreasing exponentially with the characteristic time τ = R QCJ. For typical
values of RQ and CJ, τ ≈ 10 ns.
The amount of charge in the pulse is approximately Q = imax⋅τ. The gain is then:
Equation 1
μ=Qe=imaxτe=1e⋅(VBIAS−VBR)(RQ+RS)⋅RQCJμ=Qe=imaxτe=1e·VBIAS-VBRRQ+RS·RQCJ
But RQ ≫ RS, thus
Equation 2
μ≈(VBIAS−VBR)CJe=ΔV⋅CJeμ≈VBIAS-VBRCJe=∆V·CJe
where ΔV≡ VBIAS ‐ VBR is known as "overvoltage." The equation for the gain shows that μ
varies linearly with ΔV and C J and, somewhat surprisingly, it does not depend on R Q. The
value of CJ is fixed by the specific architecture of the SiPM; thus, ΔV is the adjustable
parameter that controls μ. There is no explicit dependence of μ on temperature; however,
because VBR does depend on temperature, so does μ for a fixed V BIAS. To eliminate the
temperature dependence of the gain, the SiPM should be operated so that ΔV is constant.
In the linear relationship between the gain and overvoltage shown by Figure 6 for three
SiPMs, note that for a given ΔV, the gain increases with the size of a microcell and, thus,
with CJ.

Learn more about how temperature affects SiPM performance.

What is the photon detection efficiency (PDE) of a SiPM?
Photon detection efficiency (ξ) is a probability that a silicon photomultiplier (SiPM)
produces an output signal in response to an incident photon. It is a function of

5
overvoltage ΔV and wavelength λ of the incident light, and can be expressed as a product
ξ(ΔV,λ) = ƒ · η · PG. In the equation, ƒ is a geometrical fill factor, η quantum efficiency, and
PG probability of Geiger discharge. Photon detection efficiency is a key characteristic of a
SiPM; below is its more detailed discussion.
Geometrical fill factor
A SiPM is a pixelated device. In a given SiPM, all of the pixels, or microcells, are square, of
the same size, and arranged in a rectangular (tiled) pattern. The total area occupied by the
microcells is known as an active area. Depending on the design of a SiPM, the active areas
range from 1 x 1 mm 2 to 6 x 6 mm2. Because certain regions of a microcell have no
photosensitivity, only some fraction of the active area is sensitive to light. This fraction is
known as the geometrical fill factor. The values of ƒ range from ∼30% to ∼80%, with the
larger value for the larger size of a microcell.
Quantum efficiency
This parameter is the probability that an incident photon produces a charge carrier
(electron or hole) capable of triggering Geiger discharge in the avalanche section of the
depletion region. The layered structure of a SiPM can be optimized through the Beers-
Lambert law so that the absorption of light occurs in the depletion layer for the desired
range of wavelengths. An electric field in the depletion region separates the photo-
generated electron-hole pair and injects one of the charge carriers into the avalanche
region. Outside of the depletion region, the electric field is weak and the photo-generated
electron-hole pair is likely to recombine. The above discussion implies that η is a function
of wavelength.
Probability of Geiger discharge

Figure7. SamplesofplotsshowinghowξdependsonΔVandλ. 

6
Once in the avalanche region, the injected charge carrier gains kinetic energy from the
electric field and if the kinetic energy exceeds the ionization threshold, may be able to
impact-ionize silicon atoms, thus triggering an avalanche. The likelihood of this scenario
depends on three factors: 1) the strength of the electric field, which is controlled by ΔV; 2)
the width of the avalanche region, which also depends on ΔV and 3) collisional cross
section of charge carriers with phonons and the specific history of these interactions. If ΔV
< 0, the likelihood of Geiger discharge is exceedingly small. As ΔV increases above 0, the
kinetic energy of a charge carrier can be above the ionization threshold energy and,
moreover, the width of the avalanche region widens, making the likelihood increasingly
higher. Inelastic scattering of charge carriers with phonons increases with ΔV, which
limits steeply the likelihood of Geiger discharge rising with ΔV. The figures are samples of
plots showing how ξ depends on ΔV and λ.
How to measure breakdown voltage in a SiPM
The construction of a silicon photomultiplier (SiPM) consists of microcells (pixels)
connected in parallel. Each microcell is a series combination of an avalanche photodiode
(APD) and a quenching resistor. In a normal application, an external voltage source
VBIAS biases the SiPM so that each APD operates in Geiger mode: the voltage on the APD,
equal to VBIAS, is larger than its breakdown voltage, V BD. The difference VBIAS − VBD is
known as overvoltage ΔV, which is one of the most important parameters affecting the
operation of a SiPM. Thus, knowing the value of V BD is crucial because it determines the
required value of VBIAS for the desired ΔV. VBD is the key characteristic of a SiPM.
Measuring VBD
Several techniques of measuring VBD exist. This note describes one that relies on the
reverse-bias I-V characteristic. Figure 8 shows the experimental setup. The SiPM is in
darkness, inside of a temperature-controlled "black box." It is reverse-biased with a
voltage source VS and the picoammeter A measures the current.

7
Figure 8. Diagram of experimental setup for measuring VBD inside a temperature-
controlled "black box."
 
Figure 9 shows a possible I-V characteristic around V BD.1 For VS << VBD (not shown), the
(dark) current I monotonically increases with V S. The dark current has two contributions:
1) the bulk current due to thermally generated charge carriers, primarily in the depletion
region, and 2) the surface current due to defects at the Si-SiO 2 interface. Both contributions
increase with VS. As VS approaches VBD, the carriers generated in the bulk begin to have
enough energy to impact ionize Si atoms in the avalanche section of the depletion region
— an APD has a gain larger than 1. The current now increases more rapidly with each
voltage step, reaching the highest rate of increase when V S = VBD. For VS > VBD, the APDs
operate in Geiger mode, with the gain linearly proportional to ΔV. In addition, crosstalk
and afterpulsing contribute to the net current. The current increases with the voltage but
at the smaller rate than when VS was approaching VBD. However, as VS increases further,
another region develops on the I-V characteristic where the current sharply increases with
VS. Here, the density of charge carriers is so high that the quenching resistors cannot
restore the APDs back to Geiger mode and the SiPM becomes an Ohmic device. Plotting
d(log I)/dVS versus VS reveals the voltages at which the maximum rate of the current
increase with voltage occurs; these voltages correspond to the peaks on this plot.

8
Figure 9. Plot of a possible I-V characteristic around VBD.
 
Figure 10 shows such a plot for the I-V characteristics listed in this scenario. The first peak
(at the lower voltage) exists at V BD. This is how VBD can be measured. The precision of the
measurement depends on the voltage step used while performing the sweep. A smaller
step yields more accurate determination of V BD. In practice, one would first determine the
approximate value of VBD with coarse voltage steps. A finer sweep around V BD would
follow. Repeating the finer sweeps and averaging the results would further improve the
accuracy of the measured VBD.

Figure 10. Plot of d(log I)/dVs versus Vs for the I-V characteristics plot in Figure 9.
S/N in a continuous wave operation of a SiPM
Analogously to a photomultiplier tube, a SiPM can operate in two distinct modes:
continuous wave and photon counting. In the former, the output pulses are neither
detected nor counted individually; instead, an analog output current is measured. In the
latter, the individual pulses are either counted (digital photon counting) or integrated to
yield the charge released in Geiger discharge (analog photon counting). The choice of the
detection technique depends, among other factors, on the frequency and duration of the
output pulses. If the pulses are crowded and overlapping, the continuous wave mode is

9
appropriate. If the pulses are distinguishable, then photon counting — analog or digital —
is preferable.
Continuous mode operation
If the individual output current waveforms have a significant overlap, the continuous
wave operation of a SiPM is warranted. Suppose a monochromatic light with average
power P illuminates the active area of a SiPM. In response, the average signal current is:
Equation 3
Is=μPλehcξIs=μPλehcξ
where μ and ξ are the gain and photon detection efficiency of the SiPM, respectively; λ is
wavelength of the incident light; e is the fundamental charge; h is Planck's constant; and c
is the speed of light. The total output current, I out, is a superposition of dark current, I d,
signal current, IS, and the "excess current" due to crosstalk and afterpulsing. In the first-
order approximation, the excess current is (P ct + Pap)IS, where Pct and Pap are the
probabilities of crosstalk and afterpulsing, respectively. Thus, I out = Id + IS + (Pct+Pap)IS. This
equation assumes that Id already includes the contributions from crosstalk and
afterpulsing.
For the measured values of Iout and Id, the photo-generated current is:
Equation 4
Is=Iout−Id(1+Pct+Pap)Is=Iout-Id1+Pct+Pap
Equating equations 3 and 4 and solving for P yields for the incident light power:
Equation 5
P=hcλe1ξ1μ(Iout−Id1+Pct+Pap)P=hcλe1ξ1μIout-Id1+Pct+Pap
Signal-to-noise ratio
If IS is measured under the conditions that the measurement bandwidth is B (in Hz) and
the excess noise figure is F, then the signal-to-noise ratio is given by
Equation 6
S/N=Is√i2d+i2sSN=Isid2+is2
where id = √(2eμBFId) and iS = √(2eμBFIS) are expressions for the shot noise of dark current
and signal current, respectively, and IS is given by Equation 4. Note that the expressions
for the dark current and photocurrent shot noise contain only one power of μ because
both the dark current and the photocurrent are already amplified by μ.
Excess noise figure
The excess noise figure F must be known for the above equation for S/N to be useful. The
theoretical understanding of F is still lacking because numerous factors may contribute to
F in a complex way. It is likely that F is a function of overvoltage, input light level,
crosstalk probability, afterpulsing probability, microcell-to-microcell gain variations, and
stochastic nature of the gain mechanism in a single microcell.
Optical crosstalk in a SiPM
Optical crosstalk occurs when a primary discharge (avalanche) in a microcell triggers
secondary discharges in one or more adjacent microcells. The secondary discharge may be

10
 nearly simultaneous with the primary (direct or prompt crosstalk) or delayed by several
10's of ns (delayed crosstalk). Optical crosstalk is an example of a correlated noise: it can
be present only if a primary discharge is present. The primary discharge can be due to 1)
absorption of a photon, 2) thermal generation of a charge carrier in the multiplication
region, 3) injection of a charge carrier, thermally generated outside of the avalanche
region, into the avalanche region, or 4) crosstalk-induced secondary discharge becoming
the primary discharge for subsequent crosstalk events. If not corrected for, crosstalk
makes the output signal higher than that implied by the amount of the incident light.
Mechanism
The figure below depicts the mechanism for the prompt (P-CT), delayed (D-CT), and no
(No-CT) crosstalk. The primary avalanche in the middle pixel creates three representative
photons. One of them moves directly to the avalanche region of the microcell on the right
and triggers a simultaneous secondary avalanche there. This is a direct or prompt
crosstalk (P-CT). The other photon creates a charge carrier in the vicinity of the avalanche
region of the microcell on the left. The charge carrier diffuses to the avalanche region,
triggering a secondary avalanche that is delayed with respect to the primary. This is a
delayed crosstalk (D-CT). The third photon leaves the SiPM; no crosstalk occurs (No-CT).
The majority of photons produced by the primary discharge does not produce crosstalk.
Figure

11. This diagram depicts the mechanism for the prompt (P-CT), delayed (D-CT), and no
(No-CT) crosstalk. (It also shows a typical structure of a SiPM, but it does not correspond
to the actual structure of the Hamamatsu product.)
Dependence
The probability of optical crosstalk depends on several factors: 1) the size of a microcell, 2)
the layered architecture of a SiPM, and 3) the difference between the bias voltage and the
breakdown voltage, or the overvoltage. For a given SiPM, the first two factors are fixed;
therefore, the overvoltage is the parameter that primarily controls the probability. The
probability increases with increasing overvoltage.
Consequences
1. If a SiPM is used to measure the incident photon flux, whether it is by photon
counting or continuous wave operation, crosstalk causes an appearance of a higher
incident flux.

11
2. Crosstalk is the main contributor to the excess noise figure.
3. Crosstalk effectively limits the practical setting for the gain.
Measuring crosstalk probability in a SiPM
A Geiger discharge in a microcell of a SiPM emits isotropically about a few tens of
photons that have enough energy to create an electron-hole pair. Most of these photons
either escape the SiPM or create pairs far from the avalanche region. However, there is a
finite probability that a photon creates a pair in, or close to, an avalanche region of a
nearby microcell. One of the charge carriers from the pair may trigger a secondary
avalanche — this avalanche is known as crosstalk. It is also possible that not only one but
two, or more, photons emitted by the primary discharge trigger secondary discharges. The
probability of crosstalk for a given SiPM is a function of overvoltage ΔV. This note
describes a method of measuring crosstalk probability PCT.

Output waveforms
Fig

ure 12. An oscilloscope trace with the display set to persistence mode.
In the absence of crosstalk, the output waveform resembles the one labeled 1 p.e. in the
figure above. If the primary avalanche is accompanied by one secondary crosstalk
12
discharge, the resulting waveform, labeled 2 p.e. in the figure, is a superposition of two 1
p.e. waveforms. If it is accompanied by two, a 3 p.e. waveform results, and so on. The
figure is an oscilloscope trace with the display set to persistence mode. The SiPM is in
darkness; the 1 p.e. waveforms are due to dark counts. The color gradient, from warmer
(red) to cooler (blue), is proportional to the "density" of waveforms, that is, the frequency
of occurrence. The most common events are 1 p.e., followed by 2 p.e., 3 p.e., etc. All of the
waveforms that are different from 1 p.e. are due to crosstalk.
Measuring crosstalk probability
The primary waveforms (1 p.e.) are due to dark counts; whereas, 2 p.e., 3 p.e., etc. are due
to crosstalk. Let N1p.e., N2p.e., N3p.e. be the number of 1 p.e., 2 p.e., 3 p.e., etc. waveforms
counted within some observation time τ. Therefore, the dark count rate is (N 1p.e. + N2p.e. +
N3p.e. +... )/τ, whereas the probability of crosstalk is P CT = (N2p.e. + N3p.e. + N4p.e. + ...)/(N1p.e. +
N2p.e. + N3p.e. + ...).
A relatively simple procedure to measure this probability is to count the number of
waveforms with the setup shown in Figure 13.

Figure 13. Diagram of a simple setup to measure probability of crosstalk.

The SiPM operates in the dark. The wide-bandwidth amplifier G converts the output
current signal to a voltage signal and amplifies the signal's strength. The output of the
amplifier is an input to a digital counter such as Tektronix FCA3100. The counter counts a
signal if its amplitude exceeds a certain user-set level (lower discrimination threshold).
The experimental data consist of the number of counts per unit time as a function of
threshold level. Plotting this data produces the characteristic "staircase" shown in Figure
14.

13
 Figure 14. Plot of

experimental data producing characteristic "staircase."

The vertical axis plots the count rate (frequency) and the horizontal axis the threshold
level. The frequency corresponding to the top red line is due to all detectable waveforms,
whereas that for the red line below the top is due to crosstalk waveforms. Referring to the
first frequency as ν0.5p.e. and the other as ν1.5p.e., the probability of crosstalk is PCT =
ν1.5p.e./ν0.5p.e..
References
1 Adopted from "Study of the Silicon Photomultipliers and Their Applications in Positron
Emission Tomography," PhD Dissertation by Chen Xu.
About the author
Slawomir S. Piatek, PhD, is a senior university lecturer of physics at New Jersey Institute
of Technology. In his role as scientific consultant at Hamamatsu Corporation, he has
developed a photonics training program for engineers and is involved in popularizing
SiPM as a novel photodetector by writing and lecturing about the device.

14