Silicon Photomultipliers, A New Device For Low
Light Level Photon Detection
Hans-Günther Moser
Max-Planck-Institut für Physik (Werner Heisenberg Institut),
Föhringer Ring 6, 80805 München, Germany
MPI Halbleiterlabor, Otto-Hahn-Ring 6, 81739 München, Germany
Abstract. Silicon Photomultipliers (SiPM) are novel detectors for low level light detection based
on arrays of avalanche photodiodes operating in Geiger mode. Offering good characteristics (fast
response, high gain, photon counting capability, insensitivity to magnetic fields, low voltage
operation) they have the potential to replace classical photomultipliers (PMT) in many
applications. Drawbacks are dark rate and optical cross talk. Though their quantum efficiency is
already comparable or better than that of bialkali PMT it is still limited by the structures on the
light sensitive front surface. A new concept, presently developed at the Max-Planck
semiconductor laboratory, allows boosting the efficiency to almost 100%. Using a fully depleted
substrate the light enters through the unstructured backside. A drift diode structure collects the
electrons on a small “point like” avalanche structure for multiplication. Engineering the thin
entrance window at the backside using antireflective layers a high efficiency can be achieved in
a wide wavelength range (300–1000nm). The paper will summarize the status of front
illuminated SiPMs and report on the development of the backside illuminated devices.
Keywords: Photon Detection, Avalanche Photon Counters, Backside Illumination
PACS: 29.40.Wk
INTRODUCTION
The Silicon Photomultiplier (SiPM) is a novel type of photon detector which
allows for compact, robust, low cost devices with photon counting capability, high
quantum efficiency, large gain and fast response (<100 ps). In contrast to photo tubes
SiPMs are insensitive to magnetic fields. The development started about a decade ago,
mainly in Russia [1,2,3]. When the first prototypes became available they triggered
large interest for many applications, like calorimeters, air-shower Cerenkov
telescopes, medical applications etc.
A SiPM is an array of avalanche photodiodes operating in limited Geiger mode.
Figure 1 shows a cross section of an individual pixel and the corresponding electrical
field. Each pixel has an individual quenching resistor and the outputs of all pixels are
connected together. The SiPM is operated at a voltage above the breakdown voltage. If
a photon converts in the sensitive (=depleted) volume of such a pixel, the electron or
hole drifts in the high field region and triggers the avalanche multiplication, which
discharges the diode capacitance Cpixel. The multiplication process stops if the voltage
drops below the breakdown voltage. Hence per pixel a charge of Q=Cpixel (Uop –
Ubreakdown) is released. The quenching resistor R limits the current recharging the diode
which is needed not to reinitialize the avalanche process. The pixels is active after a
recovery time τ ~ Cpixel R.
Hence a single cell is a binary device giving the same signal if one or more photons
convert. In a matrix the signals of all cells are added. For low photon flux, when the
probability for a single cell to be hit very low, the output is simply proportional to the
number of photons. However, for higher photon fluxes some cells might be hit by
more than one photon leading to a nonlinear response and finally to saturation. In a
crude approximation the dynamic range is given by the number of cells.
E(V/cm)
n+
Al
Resist
Al
1000000
High Field (Avalanche)
Region
100000
z
p+
p-epi layer
p-substrate
10000
1000
100
Drift Region
10
1
0
2
4
6
z
FIGURE 1. Cross section through a SiPM single cell and configuration of the electric field. The high
field region (E ~ 5x105 V/cm) is built up in the highly doped n+p+ junction.
FEATURES OF SILICON PHOTOMULTIPLIERS
Dark Rate
Due to thermally generated currents SiPMs have a substantial dark rate, a typical
value is about 1 MHz for a 1 mm2 device operated at room temperature. This causes a
problem especially for large area devices. Dark rates increase with operation voltage
and therefore with gain. The leakage current has a strong temperature dependence of:
I (T ) = I (T0 )
⎛ Eg ⎞
T2
⎟⎟
exp⎜⎜ −
2
2
kT
T0
⎝
⎠
⎛ Eg
exp⎜⎜ −
⎝ 2kT0
⎞
⎟⎟
⎠
(1)
With Eg = 1.1 eV (band gap in Silicon), k is the Bolzmann constant and (T0) T the
(reference) temperature in 0K. Lowering the temperature by ~7oK reduces the dark rate
be a factor of two.
Since leakage currents depend largely on the quality of the fabrication process, careful
processing will be one tool to produce devices with low dark rates.
Crosstalk
The hot carriers (electrons and holes) in an avalanche emit light, about 1 photon per
105 electrons [4], see Fig. 2. These photons may travel to another pixel of the matrix
and initiate an avalanche breakdown there. As a result the average number of pixel
fired per incident photon is larger than one and the dark count distribution does not
follow Poisson statistics. Practically this leads to non-linearities, additional noise, and
calibration problems. The amount of cross talk depends on many parameters, like
pixel size, dead space between the active area and gain. One way to reduce cross talk
is to optically insulate the pixels from each other by trenches etched around each pixel.
A reduction by a factor of ~100 has been achieved [5].
FIGURE 2. Light emitted by avalanche breakdowns in a SiPM. Picture taken using a Hamamatsu
Phemos 1000 emission microscope. SiPM from MEPhI with 42 μm pitch. The cells fire due to dark
currents, the light is integrated over several seconds.
Quantum Efficiency
At sufficiently high electric fields the efficiency for an electron or hole to trigger an
avalanche breakdown is almost 100%. Hence it should be possible to produce SiPMs
with a Quantum Efficiency (QE) close to 100%. However, available devices reach at
most 35% QE. The limiting factors are:
-
-
Surface transmission: A fraction of the light is reflected at the surface of the
entrance window or absorbed in the undepleted surface layer. To achieve high
efficiency this layer should be as thin as possible. The reflection losses can be
reduced applying special antireflective coatings.
Fill Factor: The major limitation of the present front illuminated devices is the
fact that a substantial area of the pixel is inactive: space is needed for guard
rings separating pixels electrically, bias resistors and Al-traces for electrical
contacts. Furthermore the high field regions of each pixel have to be separated
from the neighbors using guard rings and gaps. Therefore the resulting fill
factor is always considerably lower than 100% and can be as low as 16% for
small pixels (Fig. 3). It generally becomes better for larger pixel sizes,
however, such devices may develop other problems (e.g. small dynamic range
because of limited number of pixels). Backside illuminated devices, as
discussed in the last section do not have this problem.
FIGURE 3. Measurement of the Fill Factor. Left figure: a light spot of ~3 μm diameter (visible in the
centre of the image) is scanned across the SiPM surface, while the SiPM response is measured. Right
Figure: Response map of a single cell. The active area corresponds to 16% of the cell area [6]. Note that
the sensitive area corresponds exactly to the light emitting area of Fig. 2 (same SiPM measured).
Sensitivity for Blue and UV Light
The peak sensitivity of SiPMs is in the range of 450-600 nm. Especially at short
wavelengths their sensitivity drops considerably which limits the performance for the
detection of Cerenkov Light or blue/UV scintillator light. The wavelength dependence
of the SiPM sensitivity can be understood from the light absorption curve of Silicon
(Fig. 4). Three regions can be identified:
-
-
λ > 450 nm: the light converts deep in the silicon below the high field region.
Due to the field configuration in p-type devices (to date standard) electrons
drift back into the high field region. Electrons trigger avalanches efficiently;
hence in this wavelength region the efficiency is high.
350 nm < λ < 400 nm: the light is absorbed in the first microns of the device.
Here holes drift into the high filed region and trigger the avalanche. Since
holes have (at the same field) a lower probability to trigger an avalanche (Fig.
5), the QE is reduced in this wavelengths region.
-
λ < 350 nm. The light is absorbed in a thin surface layer. The electron/holes
tend to recombine before they diffuse into the drift and finally high field
region. In this region the QE drops sharply.
10000
Absorption length ( μ m)
1000
light absorption in Silicon
100
10
Electrons
trigger
1000000
1
100000
10000
Holes
trigger
0.1
1000
100
0.01
10
holes
electrones
1
Thin entrance window
0
2
4
6
0.001
250
450
650
850
1050
Wavelength (nm)
FIGURE 4. Light absorption in Silicon. The insert indicates the field configuration in a SiPM (see Fig.
1) and the drift direction of electrons and holes.
Several methods exist to increase the sensitivity for short wavelength: Operation at
very high voltage increases the avalanche probability for holes. However, in practice
the maximal operation voltage is limited due to the increase of cross talk and dark rate.
Clearly, entrance windows have to be made as thin as possible. An interesting
approach is the inverted structure: the n-substrate is used instead of p-type and the
doping of the high field junction is reversed. Hence the role of electrons and holes are
reversed resulting in an avalanche trigger by electrons for short wavelengths.
Overall Optimization
As described above the performance of a SiPM depends on many parameters and it
is difficult to optimize all simultaneously. For instance, large QE needs large pixels
with small insensitive gaps, a feature which necessarily enhances cross talk. Or,
enhancing the operation voltage leads to higher (enhanced avalanche probability) but
also increases cross talk and dark rate. It is therefore important to design a SiPM
carefully for a given application and most likely compromises must be made.
1
0.9
0.8
Efficiency
0.7
0.6
Electrones
0.5
Holes
0.4
0.3
0.2
0.1
0
200000
300000
400000
500000
600000
700000
Field (V/cm)
FIGURE 5. Efficiency for a single electron or hole to initiate an avalanche breakdown. Qualitative
Monte Carlo calculation based on ionization coefficients from [7].
EXISTING DEVICES
Various prototypes have been produced by several companies and research
institutes (CPTA, Moscow; MEPhI/Pulsar, Moscow; Dubna/Micron, Moscow;
Hamamatsu, Japan; and Sensl, Irland). Sizes range from 0.5x0.5 mm2 to 5x5 mm2,
with 100 – 10000 pixels. Typical pixel sizes are 25x25 μm2 to 100x100 μm2. To date
no device seems to be commercially available, only prototypes have been distributed
for evaluation, but this should change in the near future.
DEVELOPMENT OF BACKSIDE ILLUMINATED DEVICES
Driven by the need for photon detectors with very high QE for use in the MAGIC
air Cerenkov telescope [8] MPI Munich started to develop SiPM with backside
illumination [9]. The basic structure of a single cell is shown in Fig. 6: Light enters
through the unstructured backside and photons convert in the fully depleted silicon
bulk. Assisted by field shaping drift rings the electrons drift in an electric field into a
small high field region where the avalanche multiplication occurs. While the cell may
have a size of 100 μm, the avalanche region can be as small as a few micron. The
advantages of this approach are:
-
-
The backside is unstructured and can be made very thin. This leads to 100%
fill factor and high conversion efficiency even at short wavelengths. In
addition it is also possible to coat the backside with antireflective layers
enhancing the QE further [10].
The avalanche is always triggered by electrons giving high and uniform QE.
-
Since the avalanche region can be kept small resulting in small capacitance and
low gain (which must still be kept high enough to have detectable signals).
These can help to reduce cross talk and recovery time.
However, these advantages are counterbalanced by some disadvantages:
-
-
-
The whole silicon volume must be depleted, which leads to larger dark currents
and rates. Possible solutions are: thinning of the device form nominal 450 μm
to about 50 μm, using processes with inherent low dark currents, and cooling.
The large sensitive volume may increase the cross talk. It is not yet known if
this overcompensates the reduction expected due to the small capacitance.
Again, thinning should help to reduce the crosstalk.
The additional drift of the electrons through the bulk towards the high field
regions increases time jitter. From simulations [11] it is expected that the time
resolution can still be kept below 2ns.
deep p
n+ contact
FIGURE 6. Schematic cross section of a back illuminated SiPM cell [9]. The high field region is
between the deep p implantation and the n+ contact (which is connected to the quenching resistor).
Such devices have been designed in the MPI Semiconductor laboratory and
production of first structures is ongoing. Since such devices need a more
complicated production process (double sided processing) their price will be
higher. Nevertheless they will be very interesting for photon hungry applications
where the need for high QE outweighs other requirements.
CONCLUSIONS
Since several years SiPMs have been produced and proved to be suitable photon
detectors with a performance comparable or even better than photomultiplier tubes.
Noteworthy are single photon resolution, good efficiency, insensitivity to magnetic
fields and (probably) low costs. This makes them especially interesting for large scale
applications like in calorimeters. Drawbacks are the high dark rate, cross talk and poor
UV/blue sensitivity. Ongoing R&D may improve the devices further. A new concept
using back illumination has the potential to boost the quantum efficiency close to
100%, which will be interesting for applications where extremely high QE is needed.
ACKNOWLEDGMENTS
I would like to thank B. Dolgoshein and E. Popova for discussions and providing
material. I am also indebted to my colleagues from the MPI Semiconductor laboratory
for preparing this paper.
REFERENCES
1. Z.Y. Sadygov, et al., IEEE Trans. Nucl. Sci, NS-43 (3), p. 1009 (2005)
2. Z.Y. Sadygov, et al. Proc SPIE 1621, p 158 (1991)
3. B. Dolgoshein, Nucl.Instrum.Meth.A504, p. 48-52 (2003)
4. A. Lacaita et al., IEEE Trans. Electron Devices, ED-40, No 3, p. 577 (1993)
5. B. Dolgoshein, private communication.
6. M. Laatiaoui, Master Thesis, TU Munich, Germany (2005)
7. R. van Overstraeten and H. De Man, Solid State Electron (GB) vol 13, p. 583 (1970)
8. J. A Barrio et al., Design Report on MAGIC, MPI-PhE/98-5 (1998)
9. G. Lutz et al., IEEE Trans. Nucl. Sci. 52, p. 1156 (2005)
10. R. Hartmann et al., Proc. SPIE 5903, N1-N9 (2005)
11. C. Merck et al, Proceedings Beaune05, to be published in NIM A.