Geiger-mode avalanche photodiodes for Cherenkov detectors

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2010 JINST 5 P01001

(http://iopscience.iop.org/1748-0221/5/01/P01001)

Download details:

IP Address: 141.108.15.81

The article was downloaded on 08/03/2010 at 17:53

Please note that terms and conditions apply.

The Table of Contents and more related content is available

Home Search Collections Journals About Contact us My IOPscience

2010 JINST 5 P01001

PUBLISHED BY IOP PUBLISHING FOR SISSA

RECEIVED: October 23, 2009ACCEPTED: November 18, 2009

PUBLISHED: January 4, 2010

WORKSHOP ON FAST CHERENKOV DETECTORS - PHOTON DETECTION,DIRC DESIGN AND DAQMAY 11-13, 2009, GIESSEN, GERMANY

Geiger-mode avalanche photodiodes for Cherenkovdetectors

D. Renker

Paul Scherrer Institue,5232 Villigen PSI, Switzerland

E-mail: dieter.renker@psi.ch

ABSTRACT: Semiconductor photo sensors have in comparison with other detectors used in thering image Cherenkov (RICH) and in the internally reflected Cherenkov light (DIRC) technique,photomultipier tubes and their derivates, a number of advantages: they have high photon detectionefficiency (∼ 50%), are insensitive to magnetic fields, operate at low voltages and allow a compact,light and robust design. Specially the relatively new Geiger-mode avalanche photodiode (alsocalled silicon photomultiplier) is a promising candidate for a detector of Cherenkov photons. Thestate of the development and the problems of this device will be described.

KEYWORDS: Particle identification methods; Cherenkov detectors; Photon detectors for UV, visi-ble and IR photons (solid-state) (PIN diodes, APDs, Si-PMTs, CCDs, EBCCDs etc)

c© 2010 IOP Publishing Ltd and SISSA doi:10.1088/1748-0221/5/01/P01001

2010 JINST 5 P01001

Contents

1 Introduction 1

2 Properties of Geiger-mode APDs 22.1 Gain 22.2 Temperature and voltage dependence of the gain 32.3 Photon Detection Efficiency (PDE) 32.4 Timing 52.5 Dark counts 62.6 Optical crosstalk 72.7 Afterpulsing 72.8 Pulse shape and recovery time 92.9 Radiation hardness 92.10 More properties 9

3 Summary 11

1 Introduction

The ring image Cherenkov (RICH) technique and the detection of internally reflected Cherenkovlight (DIRC) are powerful tools in high energy physics experiments. State of the art and future sys-tems have to detect few Cherenkov photons in a high multiplicity environment with the best possi-ble efficiency and with excellent time resolution. Photomultiplier tubes (PMT) and their variants,the hybrid PMTs and the microchannel PMTs, are used but they suffer from the strong sensitivityto magnetic fields and therefore force a complicated mechanical design. The transit time spread ofstandard PMTs is 250 ps and more which limits the achievable time resolution.

Insensitive even to very high magnetic fields are the relatively new and very promising Geiger-mode avalanche photodiodes (tested up to 7 Tesla [1]). They are sensitive to single photons withhigh detection efficiency, higher than that of PMTs, and they have a gain in the order of 106 whichallows a direct processing of the signals by the readout electronic circuits without or at most witha simple amplifier. The signal amplitudes are comparable to those of PMTs and therefore somepeople call the devices Silicon-PM.

Geiger-mode Avalanche PhotoDiodes (G-APD) have been invented end of the last millenniumin Russia. Zair Sadygov and Victor Golovin combined many small APDs operated above break-down (APDs like the Single Photon Avalanche Diode, SPAD, investigated in the 1960th [2, 3]) ona silicon chip and connected them all in parallel via individual limiting (quenching) resistors [4, 5].Good progress has been made during recent years and available now are devices from about a dozenof companies and institutes (Amplification Technologies, CPTA/Photonique, FBK-irst, HamamatsuPhotonics, MEPhI/Pulsar, MPI Semiconductor Lab., RMD, SensL, ST-Microelectronics, Zecotek

– 1 –

2010 JINST 5 P01001

Figure 1. Left: oscilloscope picture of the signal from a G-APD (Hamamatsu 1-53-1A-1) recorded withoutan amplifier. Right: the corresponding pulse height spectrum. The horizontal scale is in arbitrary units.

. . . ). The devices from the miscellaneous producers have quite different properties, even the G-APD versions from one producer. The density of cells (the small individual APDs) ranges from100/mm2 to 40,000/mm2, the gain varies from several 104 to several 106, some devices have a peaksensitivity for blue light (400 to 450 nm) and some in the green/red part of the visible spectrum.When the highest possible photon detection efficiency (PDE) is wanted, necessary for most appli-cations, in particular for RICH and DIRC, the G-APDs have to be operated at high overvoltage withthe consequence that all voltage dependent parameters like the unwanted crosstalk, afterpulses anddark count rates reach high values. This is different from the conduct of PMTs, where the operatingvoltage and by this the gain can be varied over a wide range without or with little influence on thePDE and other parameters.

In the following the main properties of G-APDs will be described together with the solutionsfound by some producers for the problems mentioned above.

Shortly it should be mentioned that avalanche photodiodes (APDs) operated in linear modebelow breakdown are not suited for the detection of single photons. Although they have internalgain (up to few thousands) they still need an amplifier which introduces noise. The smallest de-tectable light flash needs to contain about 20 photons. Very beneficial is the use of APDs in hybridPMTs. The device R10467U produced by Hamamatsu Photonics has a gain 1.5 ·105 and excellenttime resolution. The transit time spread is 50 ps FWHM [6].

2 Properties of Geiger-mode APDs

2.1 Gain

G-APDs produce a standard signal when any of the cells goes to breakdown. The amplitude Ai isproportional to the capacitance of the cell divided by the electron charge times the overvoltage.

Ai ∼C/q(V −Vb) (2.1)

V is the operating bias voltage and Vb is the breakdown voltage.When many cells are fired at the same time, the output is the sum of the standard pulses

A = ∑Ai (2.2)

The gain is typically in the range of 105 to 107. Single photons produce a signal of severalmillivolts on a 50 Ohm load (figure 1). No (or at most a simple) amplifier is needed for single

– 2 –

2010 JINST 5 P01001

Figure 2. Temperature coefficients of a G-APD from Hamamatsu (left) and from Photonique/CPTA (right)as function of the bias voltage. Reprinted from [7].

photon detection. Since there are almost no avalanche fluctuations as in normal APDs, the excessnoise factor is very small and it could eventually be negligible if contributions from optical crosstalkcan be suppressed.

The output signal is proportional to the number of fired cells as long as the number of photonsin a pulse times the photo detection efficiency (PDE) is significant smaller than the number of cells.

Contributions from the Nuclear Counter effect are absolutely negligible — even a heavilyionizing particle passing a cell produces a signal which is not bigger than that of a photon or anoise count.

2.2 Temperature and voltage dependence of the gain

The G-APD signal stability depends mainly on a) temperature changes and b) the stability of theapplied bias.

The breakdown voltage of a silicon diode depends strongly on the temperature because of theinteractions of the carriers with phonons. We can introduce a temperature dependent coefficientkT (V) which is like most parameters of a G-APD a function of the overvoltage V-Vb.

kT (V ) =1A· dA

dT·100% (2.3)

The amplitudes A of the signals of 2 G-APDs from Hamamatsu and Photonique/CPTA havebeen measured at different temperatures and the coefficients derived (figure 2). Reprinted from [7].

Similar we can define a voltage dependent coefficient:

kV (V ) =1A· dA

dV·100% (2.4)

Examples of the voltage dependence are shown in figure 3. Again the amplitude A of the sig-nals of 2 G-APDs from Hamamatsu and Photonique/CPTA have been measured and the coefficientsderived (figure 3) [7].

2.3 Photon Detection Efficiency (PDE)

The PDE is the product of a) the quantum efficiency (QE) of the active area, b) the geometric fillfactor ε (ε = ratio of sensitive to total area) and c) the probability that an incoming photon triggers

– 3 –

2010 JINST 5 P01001

Figure 3. Voltage coefficients kV (V) of a G-APD from Hamamatsu (left) and from Photonique/CPTA (right)depending on the bias voltage at T = 22 oC. Reprinted from [7].

Figure 4. Quantum efficiency of the active area as function of the wavelength for the G-APD 0-50-2 fromHamamatsu with 400 cells/mm2. The error bars denote the systematic measuring error.

a breakdown (Ptrigger).

PDE = QE · ε ·Ptrigger (2.5)

The QE of the active area can reach 80 to 90% depending on the wavelength. It peaks in arelatively narrow range of wavelengths compared to the QE distribution of a PIN diode becausethe sensitive layer of silicon is very thin. In the case shown in figure 4, the G-APD structure isp-silicon on an n-substrate. The p-layer is 0.5 µm thick on an about 4 µm epitaxial n-layer in theG-APD 0-50-2 from Hamamatsu.

The geometric factor ε needs to be optimized depending on the application. Since somespace is needed between the cells for separation and the individual resistors, the best filling canbe achieved with a small number of big cells. For RICH and DIRC the highest possible PDE iswanted. Since in this case the number of photons is small, big cells are suitable and a geometric fillfactor of 80% or more is possible. Nevertheless, too large a cell size has generally the disadvantageof a low dynamic range and larger dead-time due to the accumulation of ‘dead’ cells caused bythe more frequent noise triggers in the larger depleted volume. The increase in volume of the cellsis eventually limited by the thermally generated electron-hole pairs because the generation rate isproportional to the depleted volume. Therefore, in combination with the recovery time of the cells,given by the capacitance and the value of the individual quenching resistor, the thermal generationsets an upper limit to the volume of the cells.

– 4 –

2010 JINST 5 P01001

Figure 5. Triggering probability for electrons P(e) and for holes P(h) as function of the overvoltage.Reprinted from [8].

The triggering probability depends on the position where the primary electron-hole pair isgenerated and it depends on the overvoltage. Compared to holes, electrons have a better chance insilicon to trigger a breakdown. Therefore, a conversion of the photon in the p-layer has the highestprobability to trigger a breakdown. This has been calculated by W.G. Oldham et al. [8] and verifiedexperimentally (figure 5).

G-APDs with a first layer made of p-doped silicon (the so-called p-on-n structure) will bepreferentially sensitive to blue light, which is absorbed in the first fraction of a micrometer. Anelectron-hole pair is created but only the electron will drift to the high field of the p-n junctionand trigger a breakdown with high probability. Photons with longer wavelengths, in contrast, willmostly be absorbed in the n-layers behind the p-n junction and holes will drift towards the junc-tion (silicon has an indirect bandgap and therefore the absorption length is a steep function of thewavelength). They have a smaller triggering probability and consequently the detection efficiencyfor photons with long wavelengths will be reduced.

The situation is opposite for n-doped layer on top (the n-on-p structure). Photons with shortwavelengths have a reduced PDE and the peak PDE is shifted to longer wavelengths up to ∼ 550to 600 nm.

The PDE depends on the overvoltage (figure 6) and in current G-APDs does not reach a plateau(unity triggering probability) Operation at high overvoltage is favored but, in most cases, a com-promise needs to be found because the dark current and the dark count rate become very high andthe optical crosstalk increases.

When the PDE of a G-APD is measured great care has to be taken to avoid contributions bycrosstalk and afterpulses. Details of the measurement procedure should always be described inpublications.

2.4 Timing

The active layer of silicon in a G-APD is very thin (2 to 4 µm) and the process of the breakdowndevelopment is fast. In addition, the signal amplitude is big because of the high cell capacitance.Therefore, very good timing properties even for single photons can be expected. Fluctuations inthe avalanche development are mainly due to a lateral spreading by diffusion and by the photons

– 5 –

2010 JINST 5 P01001

Figure 6. Voltage dependence of the PDE in relative units measured with the G-APD S10362-33-050 fromHamamatsu.

Figure 7. Time resolution as a function of over-voltage at 400 nm (blue circles) and at 800 nm (red squares).The electronic noise contribution is directly measured (triangles). It is small compared to the device resolu-tion. Reprinted from [11].

emitted in the avalanche [9, 10]. Figure 7 shows a measurement of the time response of a G-APDin the case of single photon triggers [11]. Operation at high overvoltage (high gain) improves thetime resolution.

2.5 Dark counts

A breakdown can be triggered by an incoming photon or by any generation of free carriers in thedepleted layer, which has a thickness of a few microns. The latter produces dark counts with arate of 100 kHz to several MHz per mm2 at 25 oC and with a threshold at half of the one photonamplitude. Two main processes are responsible for dark counts, thermally generated e-h pairs andso-called field-assisted generation of free electrons.

Thermally generated free carriers can be reduced by cooling. There is a factor 2 reduction ofthe thermally generated dark counts every 8 oC drop in temperature.

Field-assisted generation without the help of a phonon (trap-assisted tunneling [12, 13]) is,compared to the thermal generation, a relatively small effect. It can only be reduced by operatingthe G-APDs at a smaller electric field, thereby lowering the gain and reducing the PDE.

– 6 –

2010 JINST 5 P01001

Figure 8. Optical crosstalk for 1× 1 mm2 G-APD produced by MEPHI/Pulsar, measured as the pulseheight distribution: no suppression (a); with suppression of the optical crosstalk (b) by grooves. Reprintedfrom [17].

The dark counts can be influenced by the G-APD production process aiming to minimize thenumber of generation-recombination centers (GR center), the impurities and crystal defects, whichgive rise to the Shockley-Read-Hall process.

2.6 Optical crosstalk

In an avalanche breakdown, there are on average 3 photons emitted per 105 carriers crossing thep-n junction with photon energies higher than 1.14 eV, the bandgap of silicon [14]. When thesephotons travel to a neighboring cell they can trigger a breakdown there, like any external photon.This effect is called optical crosstalk. Particularly critical are photons in the spectral range of some850 to 1100 nm [15] because higher energy photons are practically all absorbed within the samecell and infrared photons with wavelengths longer than 1100 nm travel over long distances withoutbeing absorbed.

The optical crosstalk acts like shower fluctuations in an APD. It is a stochastic process andintroduces an excess noise factor F as in a normal APD or in a PMT. Neglecting saturation effectsand contributions from afterpulses and dark counts F can be approximated:

F ≈ 1+ pct (2.6)

The probability pct is defined by the rate of dark count events with crosstalk (threshold 1.5fired cells) divided by the total dark count rate (threshold 0.5 fired cells).

With a dedicated design, which has an additional junction and with grooves between the cells,which act as an optical isolation, the optical crosstalk can be reduced (figure 8) [16–18].

Operation at relatively low gain is advantageous to reduce optical crosstalk (figure 9), albeitwith the disadvantage of reducing the PDE significantly.

The count rate falls dramatically when increasing the threshold of the readout electronics.Each increase of the threshold by the equivalent of the 1 photo-electron amplitude reduces thenoise count rate by about one order of magnitude when the crosstalk probability is ∼ 10%.

2.7 Afterpulsing

In the silicon volume where a breakdown happened a plasma with high temperatures (a few 1000 oC)is formed and deep-lying traps in the silicon are filled. Carrier trapping and delayed release causes

– 7 –

2010 JINST 5 P01001

Figure 9. Measurement of the rate of dark count events with crosstalk (threshold 1.5 fired cells) in percent ofthe total rate (threshold 0.5 fired cells) at different bias voltages for Hamamatsu S10362-33-050C. Reprintedfrom [19].

Figure 10. The probability for a delayed release of carriers as function of time after a breakdown event. Thelevel of dark counts is indicated by the dashed line. Reprinted from [19].

Figure 11. The probability for the occurrence of afterpulses at 5 different temperatures as function of theovervoltage. Reprinted from [20].

afterpulses during a period of several 100 nanoseconds after a breakdown. The afterpulse proba-bility of the device Hamamatsu S10362-33-050C has been measured by counting dark counts in agate with fixed width but variable delay (figure 10). Two components have been found with a 50 nsand 140 ns time constant, respectively [19].

The rate of both components of the afterpulses with long and short time constants increaseswith the overvoltage and cannot be reduced by lowering the temperature (figure 11) [20].

– 8 –

2010 JINST 5 P01001

2.8 Pulse shape and recovery time

The signal rise time is determined by the resistance of the silicon in the breakdown channel, thespace charge, the resistance of neutral regions and the parasitic capacitance of the whole deviceincluding the Al-lines which connect all cells in parallel, which is 2 orders of magnitude higherthan the capacitance of one single cell. For a 3× 3 mm2 G-APD with 50× 50 µm2 cells fromHamamatsu, the parasitic capacitance adds up to some 10 pF while the capacitance of a single cellis ∼ 90 fF.

The recharging of the cells defines the signal fall time, which is the same as the recovery time.In some devices, the polysilicon quenching resistor lies on top of the junction area and forms adirect capacitive coupling between the resistor itself and the diode. These devices show a shortpeak with a duration of 2 to 3 ns followed by a slow tail due to the recharging of the cell [21].

2.9 Radiation hardness

Massive particles (electrons, protons, neutrons etc.) can displace Si atoms from their lattice positionand cause defects which act as generation/recombination centres for free carriers. Irradiation withthese particles therefore causes an increase of the dark count rate. Figure 12 shows the increasemeasured with G-APDs from different producers after an irradiation with 1010 protons/cm2 withan energy of 82 MeV (equivalent to a fluence of 2 ·1010 neutrons/cm2, NIEL factor difference∼ 2).The devices have different properties including the area, the geometric fill factors and crosstalkprobabilities. In order to show values that can be compared ∆N* (∆N normalized to an area of1 mm2) is plotted versus PDE* which is the PDE corrected for the crosstalk which would fake toohigh PDE values. Assuming that only the defects created in the sensitive part of the area contributeto the increase of the dark counts, the PDE* was chosen as horizontal axis because one of thedetermining factors of the PDE* is the geometric fill factor, the ratio of sensitive to total area. Inthis way the different properties of the devices are taken into account.

No change, which is not compatible with the measurement errors, has been reported for thePDE, the breakdown voltage, the gain, the crosstalk probability, the recovery time constant and thevalue of the quenching resistors.

The afterpulse probability is increased by irradiation with massive particles because moregeneration/recombination centres are generated.

An irradiation with X- and γ-rays (e.g. γ-rays from a 60Co source) causes defects in the bulkonly with very low probability but there is a serious effect on the surface at the Si/SiO2 interface(breakup of the SiO2 bonds). With a proper design of the width of the isolation on the surface thesensitivity to γ-radiation can be eliminated.

2.10 More properties

There are more features which are not mentioned yet:

• G-APDs work at low bias voltage (∼ 50 V),

• have low power consumption (< 50 µW/mm2),

• are compact, rugged and show no aging,

– 9 –

2010 JINST 5 P01001

Figure 12. Increase of the dark count rate after irradiation with 1010 protons/cm2 normalized to an area of1 mm2 as function of PDE* which is the measured PDE corrected for the crosstalk. Reprinted from [22].

• tolerate accidental illumination,

• could be cheap because they are produced in a standard MOS process.

The G-APDs from the different producers show the significant progress achieved over the last2 to 3 years. The main difference between the devices is the type of the substrate (n- or p-doped)which defines the region of the wavelength (blue or green) with the peak PDE. An attempt tosummarise the current status is given in the list:

• High PDE of more than 40% for green (CPTA/Photonique) and blue light (Hamamatsu,Zecotek).

• Reduction of the dark count rate at room temperature down to 300 kHz/mm2 (Hamamastu,Zecotek).

• Low cross-talk <1–3% (CPTA/Photonique, STMicroelectronics).

• Low temperature coefficient of ∼ 0.3%/oC (CPTA/Photonique).

• Fast timing ∼ 50 ps (RMS) for single photons (all).

• Large dynamic range with 15 000–40 000 pixels/mm2 (Zecotek).

• Large area of 3× 3 mm2 and more (CPTA/Photonique, Hamamatsu, FBK, SensL, Zecotek,STMicroelectronics. . . ).

There are two very interesting devices developed at the semiconductor laboratory of the Max-Planck-Institute in Munich. The so-called SiMPI is a G-APD with bulk integrated quenching re-sistors which has a high geometric factor of ∼ 75% because there is no resistor on the surface. Itallows engineering of the entrance window and since there are no lateral high field regions on thesurface, the radiation hardness could be improved [23]. The back illuminated G-APD is a combi-nation of a drift diode and a G-APD which has a geometric factor of 100%, allows direct couplingof the entrance window to a scintillator and bump bonding to a readout chip is possible [24].

– 10 –

2010 JINST 5 P01001

3 Summary

When single photons need to be detected with very short time resolution, a domain where up tonow only photomultiplier tubes could be used, the Geiger-mode APDs promise to perform verywell. Their high PDE makes them in some applications even superior to PMTs. In addition theycould allow the construction of cost effective detectors because they potentially are cheap due tothe standard MOS production process and because they need only simple electronic circuits, noshielding, little space and have low power consumption.

When the highest possible PDE and the best time resolution are wanted the G-APDs have tobe operated at high overvoltage with the consequence that all voltage dependent parameters like theunwanted crosstalk, afterpulses and dark count rates reach high values. In addition the currents inthe G-APDs can become unacceptably large in high rate environments leading to a self-heating ofthe G-APDs which are very sensitive to temperature changes. A re-engineering might be necessary(e.g. reduction of cell capacitance) when the gain and by this the current have to be reduced.

References

[1] A. Kolb et al., PET block detector readout approaches using G-APDs, IEEE Nucl. Sci. Symp. Conf.Rec. (2008).

[2] R.J. McIntyre, Theory of microplasma instability in silicon, J. Appl. Phys. 32 (1961) 983.

[3] R.H. Haitz, Model for the electrical behaviour of a microplasma, J. Appl. Phys. 35 (1964) 1370.

[4] Z. Sadygov, Avalanche Detector, Russian Agency for Patents and Trademarks, Patent No. RU2102820 (1998).

[5] V. Golovin, Avalanche Photodetector, Russian Agency for Patents and Trademarks, Patent No. RU2142175 (1999).

[6] http://sales.hamamatsu.com/assets/pdf/parts R/HPD-R10467U 200810.pdf.

[7] Y. Musienko et al., Tests and performance of multipixel Geiger mode APD, PoS(PD07)012.

[8] W.G. Oldham et al., Triggering phenomena in avalanche diodes, IEEE Trans. Electron Dev. 19(1972) 1056.

[9] A. Lacaita et al., Observation of avalanche propagation by multiplication assisted diffusion in p-njunctions, Appl. Phys. Lett. 57 (1990) 489.

[10] A. Lacaita et al., Photon-assisted avalanche spreading in reach-through photodiodes, Appl. Phys.Lett. 62 (1993) 606.

[11] G. Collazuol et al., Single photon timing resolution and detection efficiency of the IRST siliconphoto-multipliers, Nucl. Instrum. Meth. A 581 (2007) 461.

[12] G.A.M. Hurkx et al., A new recombination model for device simulation including tunnelling, IEEETrans. Electron Dev. 39 (1992) 331.

[13] G.A.M. Hurkx et al., A new analytical diode model including tunneling and avalanche breakdown,IEEE Trans. Electron Dev. 39 (1992) 2090.

[14] A. Lacaita et al., On the bremsstrahlung origin of hot-carrier-induced photons in silicon devices,IEEE Trans. Electron Dev. 40 (1993) 577.

– 11 –

2010 JINST 5 P01001

[15] N. Otte, On the Efficiency of Photon Emission during electrical Breakdown in Silicon, Nucl. Instrum.Meth. A 610 (2009) 105.

[16] N. Basharuli et al., Registration of charged particles by scintillating fibers coupled with micro-cell SiAPD, Advanced Technology and Particle Physics, World Scientific, Singapore (2002).

[17] P. Buzhan et al., Large area silicon photomultipliers: performance and applications, Nucl. Instrum.Meth. A 567 (2006) 78.

[18] W.J. Kindt, Geiger Mode Avalanche Photodiode Arrays, Thesis, Delft University Press, DelftNetherlands (1999).

[19] Th. Kraehenbuehl, G-APD arrays and their use in axial PET modules, Diploma Thesis, ETH Zurich,Zurich Switzerland (2008).

[20] F. Retiere et al., Characterization of Multi Pixel Photon Counters for T2K Near Detector, Nucl.Instrum. Meth. A 610 (2009) 378.

[21] C. Piemonte et al., Characterization of the first prototypes of silicon photomultiplier fabricated atITC-irst, IEEE Trans. Nucl. Sci. 54 (2007) 236.

[22] Y Musienko et al., Study of radiation damage induced by 82 MeV protons on multi-pixel Geiger-modeavalanche photodiodes, Nucl. Instrum. Meth. A 610 (2009) 87.

[23] J. Ninkovic et al., SiMPl — Novel high QE photosensor, Nucl. Instrum. Meth. A 610 (2009) 142.

[24] J. Ninkovic et al., The avalanche drift diode — A back illuminated drift silicon photomultiplier, Nucl.Instrum. Meth. A 580 (2007) 1013.

– 12 –