IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013 2247

New Developments of Near-UV SiPMs at FBKTiziana Pro, Alessandro Ferri, Alberto Gola, Member, IEEE, Nicola Serra, Alessandro Tarolli, Nicola Zorzi, and

Claudio Piemonte, Member, IEEE

Abstract—A thorough characterization of a novel silicon photo-multiplier technology for near-ultraviolet (near-UV) light detec-tion, called NUV-SiPM, is presented. It features a peak detectionefficiency of more than 30% in the region between 380 and 400nm,mainly limited by the fill factor. An accurate electric field engi-neering allows to have excellent noise properties, with a dark countrate of less than 200 kHz/mm at maximum efficiency and 20 C.In addition, a breakdown voltage uniformity better than 100mV atthe wafer level and a temperature dependence of 25 mV C wereobtained. We coupled a mm device to a mmLYSO scintillator obtaining an energy resolution of 10% FWHMwith 511-keV gamma ray irradiation and a coincidence resolvingtime between two identical detectors of 210-ps FWHM. A detaileddescription of these results is presented in the paper.

Index Terms—Low-level light detection, near ultraviolet light de-tection, silicon photomultipliers.

I. INTRODUCTION

I N the past years, increasing efforts on SiPM technologyhave been made in FBK and many other research insti-

tutes [1]–[3] to improve their performance. There is clear evi-dence that SiPMs are going to replace progressively photo-mul-tiplier tubes (PMTs) in a wide range of applications, includingthe scintillation light readout in medical imaging, high energyphysics and astrophysics [4]–[8]. In particular, there is a verystrong interest in using SiPMs for next-generation MR-compat-ible, time-of-flight PET machines [5]. The foremost advantagesof a SiPM over a PMT are low operating voltage ( 100 V),insensitivity to magnetic fields, compactness, robustness, andhigher photon detection efficiency.Specific applications require specific peak sensitivity. In

positron emission tomography (PET), SiPMs are usuallycoupled to scintillators whose emission spectrum is in theblue-NUV region: LYSO:Ce has a peak emission at 420nm (blue-near-UV) and LaBr :Ce at 380 nm (near-UV) [9].However, new types of scintillators are appearing with lightemission in the green-red region as Ce:GAGG whose peak isat 550 nm. It is then important to enhance the photon detectionefficiency (PDE) in the wavelength region of interest.The first technology developed at FBK and all its subsequent

upgrades had peak sensitivity in the green. Although it was not

Manuscript received December 21, 2012; revised March 21, 2013; acceptedApril 10, 2013. Date of publication May 23, 2013; date of current version June12, 2013. The research leading to these results has partially received fundingfrom the European Union Seventh Framework[FP7/2007-2013] under grantagreement FP7-241711.The authors are with Fondazione Bruno Kessler, Trento 38050, Italy (e-mail:

piemonte@fbk.eu).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TNS.2013.2259505

centered on the LYSO emission spectrum, excellent timing andenergy resolution were reported [10]–[14]. In order to furtherimprove the performance with this scintillator, efforts have beenrecently devoted to the production of a SiPM specifically de-signed for near-UV detection (NUV-SiPM). A first prototypewith a peak PDE of about 30% at 410 nm (50 m cell, 42%of fill factor) was proposed in [15]. However, this technologysuffered from a high dark count rate (DCR) and after-pulsingprobability.At present, according to the literature, some technologies po-

tentially working in the near-UV (350–400 nm) can be found,though in some cases the maximum PDE value is not centeredin this range [16]–[22]. Moreover, a direct comparison of thetechnologies is not straightforward, since the extraction of thedifferent parameters determining the performance can be com-plex and methods are not aligned.With the present work, we introduce a new version of NUV

SiPM, which is specifically conceived for reaching the max-imum efficiency at 380–400 nm. Measurements show a PDEhigher than 30% in this wavelength range, mainly limited by thefill factor (42%), whereas DCR is lower than 200 kHz/mm atthe maximum efficiency at 20 C. Those parameters are amongthe most competitive ones reported in literature. In addition,we have achieved an excellent breakdown voltage uniformityas well as a low temperature dependence.

II. TECHNOLOGY AND DESIGN

The absorption of near-UV photons in silicon is very superfi-cial. This implies that the dopant profiles must be very shallowto enhance quantum efficiency. In SiPMs, this is not the only as-pect to be considered, since also the avalanche triggering prob-ability plays an important role [1]. For UV light, the carriersstarting the avalanche in an n+/p junction (FBK original tech-nology) are mainly the holes, whose triggering probability islower than for the electrons at any given overvoltage . Onthe contrary, in a p+/n junction, the avalanche is started mainlyby electrons enhancing the efficiency for UV light. Electronsprovide also a much faster triggering probability increase with

: for a given cell structure (size and depletion width), thisallows us to obtain the same PDE at lower overvoltage, whichimplies a lower gain and, thus, a lower correlated noise. Hence,a p+/n junction for UV light provides better performance bothin terms of efficiency and excess noise factor.NUV-SiPMs were fabricated following these considerations

and through an accurate electric field engineering. A varietyof layout configurations were produced, ranging from

mm to mm and with different cell sizes. For sim-plicity, only results from mm will be presented in thispaper, but they are representative of all the other layouts. The

0018-9499/$31.00 © 2013 IEEE

2248 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013

Fig. 1. - plots of mm SiPMs with a cell size of m andm .

attention is focused on the m cell size having a nom-inal fill factor of 42%. In particular cases, also results obtainedwith a m cell size (22% fill factor) will be shown forcomparison.Furthermore, results from another FBK recent technology,

called RGB-SiPM (red-green-blue), with sensitivity peaked inthe visible part of the light spectrum, will be used for compar-ison [2]. This technology has been recently released as an up-grade of the original n+/p technology.

III. EXPERIMENTAL RESULTS

In the following paragraphs, we present a set of measure-ments that gives a complete picture of the device performance.The first one is the reverse and forward current-voltage ( – )measurement in dark condition at controlled temperature. Thesecond test is the pulse analysis, again in dark condition and atcontrolled temperature. This procedure is thoroughly describedin [23]. The third step is the PDE measurement under contin-uous light illumination. Finally, the SiPM is coupled to a LYSOscintillator for both energy and timing resolution tests.

A. I–V Characterization

In Fig. 1, the - of a mm SiPM with a mcell size is shown. For comparison, the current of a device witha m cell size is also plotted.Both devices feature a breakdown voltage, , of 25.65 V.

The postbreakdown current, instead, has a different behavior.The 50- m cell has a parabolic current growth until 30 V, thenit diverges very quickly. As will be shown in the following para-graph, correlated noise is the main cause for this steep increase.On the other hand, the 25- m cell SiPM shows a lower cur-rent, not diverging after breakdown. The lower current is givenby both a smaller gain (smaller cell capacitance) and a lowerprimary noise (due to the lower fill factor). The absence of di-vergence is due to a smaller correlated noise thanks to the lowergain.

values, extracted from - measurements over a popu-lation of 414 SiPM (the number of devices contained in a wafer),

Fig. 2. Histogram of for all the SiPMs contained in a wafer.

are reported in the histogram depicted in Fig. 2. The voltage stepin – measurements is 0.1 V and the bin width is selected ac-cordingly. Notably, falls in a range between 25.5 and 25.9V for all devices. The mean value is 25.65 V with a standarddeviation of only 0.07 V. Excellent breakdown voltage unifor-mity is therefore achieved with this technology. This is a con-siderable benefit for application with a high number of channels,like PET [5].In order to analyze the breakdown voltage temperature de-

pendence, – measurements were repeated between 20 Cand 50 C. In Fig. 3, versus temperature is plotted. In thesame graph, also the dependence of the first NUV tech-nology released in 2011 [15] is shown for comparison. Thelinear fit on the experimental data of the new NUV-SiPM (redsquares) reveals a temperature coefficient of 25.8 mV C. Inthe older technology (blue circles), the coefficient is 27 mV Cup to 20 C and then it increases to 75.4 mV C. The doubleslope is determined by the change with temperature of the de-pletion width at the breakdown voltage. This phenomenon hasbeen studied in [24]. A low temperature coefficient is ex-tremely important to alleviate the requirements on the coolingsystem.Postbreakdown current was analyzed in order to understand

the dark count temperature dependence, similarly to what isdone in [2]. At a constant , the reverse current isproportional to the DCR of the SiPM. In Fig. 4, the Arrhe-nius plot of the SiPM current at 3-V overvoltage is shown.According to the Shockley–Read–Hall theory [25], thermalgeneration is proportional to (where

0.56 eV is half the energy gap, is the Boltzmannconstant, and is the temperature in K). In the fit shown inFig. 4, the activation energy is 0.49 eV, which is very close to

. Thus, in the temperature range from 20 C to 40 C,the main mechanism underneath the e–h pair generation isthermal generation. In the previous NUV release, the sameparameter was found to be 0.31 eV, i.e., with a significanttunneling component. For temperatures higher than 40 C, thediffusion component starts to be dominant. Diffusion mecha-nism follows an trend, thus its decrease

PRO et al.: NEW DEVELOPMENTS OF NEAR-UV SiPMs AT FBK 2249

Fig. 3. Breakdown voltage temperature dependence of a mm SiPM(50- m cell) from the new NUV technology (red squares). Linear regressionline is superposed. A mm SiPM of the first NUV release (blue circles)[15] is also shown.

Fig. 4. Arrhenius plot of the current at 3-V overvoltage for a mm NUVSiPM.

with temperature is steeper than for Shockley–Read–Hallgeneration.Finally, the value of the quenching resistor was calculated

from the forward - . In Fig. 5, it is plotted as a function ofthe temperature. It decreases with temperature, from 1.4 M at20 C to 1 M at 50 C. The trend is almost linear with a

negative coefficient of about 6 C.

B. Pulse Analysis in Dark

As previously mentioned, an automatic procedure, describedin [23], was developed to extract a number of parameters in darkcondition such as gain, primary dark rate and correlated noiseprobability.Considering the gain, two values are defined: , calculated

as the area of single dark pulses divided by the amplifier gainand the electron charge; , calculated as the ratio between thereverse current and the primary dark count rate multiplied by theelementary charge. In case of no correlated noise (no after-pulse,

Fig. 5. Value of the quenching resistor as a function of the temperature.

Fig. 6. Gain pulse and gain current versus overvoltage for twomm SiPMs with 50- m and 25- m cell size, respectively.

AP, nor optical crosstalk, CT), the two values must be identicaland correspond to the number of carriers generated during a celldischarge.In Fig. 6, and for the mm SiPM with a 50- m

cell are plotted as a function of the overvoltage at 20 C. Forcomparison, and for the 25- m cell are also shown. Inthe first case, progressively diverges from at increasing

, reaching a value three times larger at 5 V. This indicatesthat an extra charge is associated to each primary dark event,increasing the reverse current of the device. It can be attributedto CT and AP events. This extra charge can be quantified by theexcess charge factor (ECF): ECF [23].

for the 25- m cell is lower than for the 50- m one sinceits capacitance is smaller. Notably, in this case, the ECF isvery close to one in all the considered voltage range, implyingboth low CT and AP. A low correlated noise, i.e., a low excessnoise factor, is very important in applications such as lowenergy gamma spectroscopy. This result strongly pushes thetechnology towards new developments aimed to increase thefill factor of small cells.Dark count rate was also extracted and compared with the first

NUV version (Fig. 7). Solid lines correspond to primary dark

2250 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013

Fig. 7. Dark count rate as a function of the overvoltage of the new NUV tech-nology (50- m cell size), compared to the first NUV (same cell size). The plotshows also the DCR of a 25- m cell of the newNUV at 20 C. Solid and dashedlines represent the primary and total dark count rate, respectively.

count rate (DCR) at 0.5 photoelectron (p.e.) threshold level.It is the Poisson rate extracted from the exponential fit of thehistogram of the time delay between consecutive dark pulses.Dashed lines represent the total dark rate, DCR at 0.5 p.e. Itincludes DCR, AP and delayed crosstalk (DeCT) [23], which isone of the two components of optical crosstalk. In particular, inDeCT, the photon emitted by the avalanche in one cell is ab-sorbed in the nondepleted region beneath a neighboring one,generating a carrier able to reach by diffusion the active re-gion. The correlated event produced in the second cell is de-layed in time because the diffusion process can take severalnanoseconds. On the contrary, in direct crosstalk (DiCT), thecarrier is generated by the photon in the depletion region ofthe second cell, so the correlated event happens in coincidence.DiCT cannot be evaluated from the time delay histogram.Fig. 7 shows that the DCR is very low: less than 200 kHz at

room temperature and at maximum overvoltage for the 50- mcell. It is more than one order of magnitude lower than the pre-vious NUV production [15]. This was achieved thanks to bothan optimized electric field shape and a new gettering procedure.The DCR scales by a factor of 16 from 20 to 20 C, halvingevery 10 C.The total rate increases faster with the overvoltage because

it includes also AP and DeCT, whose probability increase withovervoltage. At the maximum overvoltage, DCR is about threetimes larger than DCR. The ratio DCR DCR is close to the ex-cess charge factor, meaning that the AP and DeCT representsthe largest fraction of the correlated noise. It has to be notedthat, in this technology, DeCT is not separated from AP in thetime delay histogram probably because the exponential decaytime constants of the two phenomena are similar. This behavioris different fromwhat is reported in [23], which refers to an n+/pjunction, where the two contributions were clearly distinguish-able. For this reason, the correlated noise determined by AP andDeCT will be considered as a single class of events named de-layed correlated pulses. The probability is calculated as the ratioof the number of delayed correlated events to the total one. Fig. 8shows this parameter as a function of the overvoltage. It is about

Fig. 8. Delayed correlated pulse probability as function of overvoltage.

Fig. 9. DCR versus threshold level at different over-voltages of the NUV tech-nology.

50% at 4-V overvoltage, where the PDE has almost reachedits maximum for NUV light (see next paragraph). The dashedline in the plot is a second-order polynomial fit serving as aneye-guide.In Fig. 7, also DCR and DCR for a mm SiPM with a

25- m cell are plotted. DCR is lower compared with the 50- mcell since the fill factor is lower: in the range 2–5 it is below70 kHz. Moreover, it is evident that the delayed correlated eventcontribution is negligible, even at 8 , as it was expectedfrom the low ECF.DiCT was extracted from the versus threshold plot,

shown in Fig. 9, for the 50- m cell at three different values of. The threshold value was normalized to the amplitude of

the single cell pulse. The DiCT probability, calculated as theratio between the number of events between 1.5 and 2.5 andthose between 1.5 and 0.5 p.e. threshold [23], is 1% at 2.35

and 7% at 4.35 , which is much smaller comparedwith delayed correlated event probability.It is also interesting to note that the NUV-SiPM DiCT proba-

bility is almost three times smaller compared with the RGB one.The main reason is that the optical crosstalk events in the first

PRO et al.: NEW DEVELOPMENTS OF NEAR-UV SiPMs AT FBK 2251

Fig. 10. Photon detection efficiency as a function of the wavelength for NUV(at three different ) and RGB technologies (at 4.5 ). Devices haveexactly the same cell layout: 50- m size and 42% fill factor.

technology are mainly determined by holes, which have a lowertriggering probability.

C. Photo-Detection Efficiency Measurement

PDE was measured under continuous low-level light illumi-nation. The measurement was performed between 360 and 900nm with a step of 10 nm using a monochromator. Our currentsetup does not allow the PDE characterization at lower wave-lengths to cover the whole NUV spectrum. The light intensitywas determined with a calibrated photodiode. Neutral filterswith known attenuation factors were used to reduce the numberof photons impinging on the device under test. In order to facili-tate the measurement, getting rid of optical crosstalk, single-cellSiPMs having exactly the same size and layout of the devicesreported in the paper were used. In order to suppress also APcontributions, the photon count rate was not calculated as thetotal rate at 0.5 p.e., but it was extracted from the Poisson fitof the pulse arrival time distribution (similarly to the procedureused in dark pulse analysis). DCR was subtracted from the lightcount rate.In Fig. 10, the PDE as a function of the wavelength, at dif-

ferent overvoltages, is shown for a 50- m cell. For comparison,the PDE of a RGB-SiPM biased at 4.5 is also plotted. Thetwo devices have the same layout: 50- m cell and a 42% fillfactor. In this current version, the fill factor is mainly limitedby the use of a mask aligner lithography, which has a relativelylarge minimum feature size.The peak PDE of the NUV-SiPM is reached at 390 nm with

a maximum value of 32% at 4.5 V overvoltage. Given the 42%fill factor, the product of the quantum efficiency times the trig-gering probability is about 0.76, which is an excellent result fornear-UV light. Above 450 nm, the PDE rapidly decreases be-cause photons aremainly absorbed beyond the high-field region,where hole triggering starts to be dominant. The RGB-SiPMtechnology, on the other hand, peaks in the 450–650-nm range,reaching a maximum value of 31% at 570 nm [2]. Oscillationson PDE curves are determined by the interference in the stackof dielectric layers on top of the silicon substrate. They can

Fig. 11. Photon detection efficiency as a function of overvoltage for NUV tech-nology and RGB technology at 380 nm.

be eliminated or modified by changing the thicknesses of thelayers.The two SiPM technologies are complementary with respect

to the peak sensitivity wavelength, so that the most suitable onecan be selected according to the specific application.The PDE as a function of the overvoltage at 380 nm is plotted

in Fig. 11, for the two technologies. As far as the NUV-SiPM isconcerned, the PDE at 380 nm increases quite rapidly, accordingto the electron triggering probability, and approaches saturationat 4-V overvoltage. On the contrary, the RGB-SiPM has a muchlower PDE at 380 nm, slowly increasing with , since it isproportional to the hole triggering probability.A fair comparison between technologies cannot take into ac-

count only a single parameter. Depending on the application,the best achievable performance will be a compromise betweenPDE at the wavelength of interest, primary noise and corre-lated noise. For this reason, plots showing their interdependenceare very important. In Fig. 12, the primary dark noise, DCR, isplotted as a function of the PDE at 380 nm. The NUV reachesmuch higher PDE than the RGB SiPM and, for any given PDEvalue, it has a much lower noise.Considering the signal-correlated noise, CT and AP deterio-

rate the photon number resolution of the detector, with respectto the pure Poisson statistics, by the excess noise factor (ENF).This can be the dominant noise source in many photon countingapplications, including -ray spectroscopy, in which SiPMs areused for the readout of the scintillation photons. If CT is theonly source of correlated noise, the ENF is equal to the ECF[26]. For the AP, on the other hand, the ENF is always smallerthan the ECF and the two values are equal at low gain only.Therefore, at a first approximation, the ECF can be consideredan upper bound for the ENF and can give an estimate for thetotal amount of correlated noise.The ECF as a function of the PDE at 380 nm is also shown in

Fig. 12. Again, we can see that the NUV technology has a veryfavorable behavior in the NUV region, having an ECF below1.5 almost up to the maximum efficiency.This plot shows that the ECF, and thus the correlated noise,

starts diverging when the PDE has almost reached saturation.

2252 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 3, JUNE 2013

Fig. 12. Primary dark count rate and excess charge factor versus photon detec-tion efficiency at 380-nm wavelength for NUV and RGB technologies (cell sizeof 50 m). C.

Fig. 13. Energy spectrum obtained from a mm SiPM coupled to amm LYSO crystal irradiated with a Na source.

To this end, the 50- m size can be considered an upper limitbecause for larger cells (with large gain) the correlated noisewould diverge before reaching the maximum triggering proba-bility.

D. Energy and Timing Resolution With Scintillator

Finally, a mm SiPM was coupled to a mmLYSO scintillator and irradiated with gamma rays from a Nasource. The energy spectrum was extracted from the integral ofSiPM signals in response to the incident radiation. In Fig. 13, theenergy spectrum is plotted in the range 200 keV–1.5 MeV, at 5V overvoltage. The 511 keV and 1.27 MeV peaks are clearlyvisible. The plot was obtained correcting the SiPM signal forthe nonlinearity due to the limited number of cells compared tothe number of photons. The energy scale is reconstructed fromthe position of the two photopeaks corresponding to the twogamma rays emitted by the Na source. The 511-keV peak isfitted with two Gaussian curves, one related to the photopeak,and the other, centered at lower energy, related to the Lutetiumescape peak (see as an example [27]).

Fig. 14. Energy resolution for a NUV-SiPM, coupled to a LYSO scintillator at20 C.

Fig. 15. Coincidence resolving time of two NUV-SiPMs, each one coupled toa mm LYSO scintillator. Temperature is varied between 20 Cand 20 C.

Fig. 14 shows the energy resolution measured at differentovervoltages at 20 C. It is found to be 10% FWHM in a verywide range of bias conditions, namely between 2 and 5 V. Itis worth noting that similar results were found with the RGB-SiPM technology. In both cases, the resolution approaches thelimit of the LYSO scintillator.We also measured the timing properties of this technology

with scintillator. The methodology is thoroughly described in[11]–[13]. The coincidence resolving time is measured betweentwo identical detectors, each one composed of a mm SiPMcoupled to a mm LYSO scintillator. The radioactivesource is placed between the two in head-on configuration. Thepick-off time of each SiPM is determined with the leading edgediscrimination method after a signal baseline compensation toget rid of the SiPM noise [12]. Fig. 15 shows the coincidenceresolving time as a function of the overvoltage. Similarly to theenergy resolution, the best value is reached at about 5 V. No-tably, the timing performance does not depend on the tempera-ture since the dark rate of this technology is low. The minimumcoincidence resolving time is about 210-ps FWHM. This value

PRO et al.: NEW DEVELOPMENTS OF NEAR-UV SiPMs AT FBK 2253

is excellent but slightly worse than the 175 ps measured on aRGB-SiPM with the same geometry. We are still investigatingthe reason of this difference.

IV. CONCLUSION

A novel SiPM technology designed for near-UV light hasbeen developed at FBK (NUV-SiPM) and its in depth character-ization is presented in this paper. A peak PDE of more than 30%is reached at 390 nm on a m cell, with a 42% fill factor.PDE is higher than 25% in the range between 360 and 450 nm,thus making the detector suitable for near-UV detection. Verylow DCR, around 200 kHz/mm at 20 C, is shown at the men-tioned PDE, which is a remarkable improvement compared toprevious technologies and it is very competitive with respect toother commercial devices. The excess charge factor, accountingfor the correlated noise, reaches a value around 1.5 at maximumPDE, and it is mainly determined by after-pulses.The same measurements repeated on a SiPM featuring a

m cell size showed a remarkably lower correlatednoise. The excess charge factor is close to unity up to 7 Vovervoltage. At this bias level, the triggering probability forelectrons, and consequently the PDE in the near-UV region, hasalready reached its maximum. This is a very important advan-tage of tiny cells. The limit of the actual technology combinedwith small cell size is the poor fill factor (22% on a mcell). FBK is strongly committed to the development of a newcell border structure allowing high fill factor ( 50%) with verysmall cells (down to m ). The first significant resultsin this direction have already been obtained in an upgrade ofthe RGB-SiPM technology.

REFERENCES

[1] C. Piemonte, “A new silicon photomultiplier structure for blue lightdetection,” Nucl. Instrum. Methods A, vol. 568, pp. 224–232, 2006.

[2] N. Serra et al., “Characterization of new FBK SiPM technology forvisible light detection,” J. Instrument., Apr. 2013, to be published.

[3] Y. Musienko, “Advances in multipixel Geiger-mode avalanche photo-diodes (silicon photomultipliers),” Nucl. Instrum. Methods A, vol. 598,pp. 213–216, 2009.

[4] V. D. Kovaltchouk, G. J. Lolos, Z. Papandreou, and K. Wolbaum,“Comparison of a silicon photomultiplier to a traditional vacuum pho-tomultiplier,” Nucl. Instrum. Methods A, vol. 610, pp. 150–153, 2009.

[5] E. Roncali and S. R. Cherry, “Application of silicon photomultipliersto positron emission tomography,” Ann. Biomed. Eng., vol. 39, no. 4,pp. 1358–1377, April 2011.

[6] E. Garutti, “Overview on calorimetry,” Nucl. Instrum. Methods A, vol.628, pp. 31–39, 2011.

[7] P. Buzhan, B. Dolgoshein, A. Ilyin, V. Kaplin, S. Klemin, R. Mir-zoyan, E. Popova, and M. Teshima, “The cross-talk problem in SiPMsand their use as light sensors for imaging atmospheric Cherenkov tele-scopes,” Nucl. Instrum. Methods A, vol. 610, pp. 131–134, 2009.

[8] E. Grigorieva, A. Akindinova,M. Breitenmoser, S. Buono, E. Charbon,C. Niclass, I. Desforges, and R. Rocca, “Silicon photomultipliers andtheir bio-medical applications,” Nucl. Instrum. Methods A, vol. 571,pp. 130–133, 2007.

[9] C. W. E. Eijk, “Radiation detector developments in medical applica-tions: Inorganic scintillators in positron emission tomography,” Radiat.Protect. Dosimetry, vol. 129, no. 1–3, pp. 13–21, 2008.

[10] C. Piemonte, A. Gola, A. Picciotto, N. Serra, A. Tarolli, and N. Zorzi,“Timing performance of large area SiPMs coupled to LYSO using noisecompensationmethods,” in IEEENucl. Sci. Symp. Conf. Rec., 2011, pp.59–63, Paper N4-5.

[11] A. Gola, C. Piemonte, and A. Tarolli, “The DLED algorithm fortiming measurements on large area SiPMs coupled to scintillators,”IEEE Trans. Nucl. Sci, vol. 59, no. 2, pp. 358–365, Apr. 2012.

[12] A. Gola, A. Tarolli, and C. Piemonte, “Analog circuit for timing mea-surements with large area SiPMs coupled to LYSO crystals,” IEEETrans. Nucl. Sci., vol. 60, no. 2, pp. 1296–1302, 2013.

[13] A. Gola, A. Tarolli, and C. Piemonte, “SiPM cross-talk amplificationdue to scintillator crystal: Effects on timing performance,” presentedat the IEEE Nucl. Sci. Symp. Conf., 2012, Paper N1-203.

[14] R. I. Wiener, S. Surti, C. Piemonte, and J. S. Karp, “Timing and en-ergy characteristics of LaBr Ce and scintillators read by FBKSiPMs,” in IEEE Nucl. Sci. Symp. Conf. Rec., pp. 4013–4019, PaperMIC20-7.

[15] A. Ferri, A. Gola, C. Piemonte, T. Pro, N. Serra, A. Tarolli, and N.Zorzi, “First results on NUV SiPMs at FBK,” Nucl. Instrum. MethodsA, 10.1016/j.nima.2012.10.021, to be published.

[16] A. Vacheret et al., “Characterization and simulation of the responseof multi-pixel photon counters to low light levels,” Nucl. Instrum.Methods A, vol. 656, pp. 69–83, 2011.

[17] Hamamatsu [Online]. Available: www.hamamatsu.com[18] SensL [Online]. Available: http://sensl.com/products/silicon-photo-

multipliers/bseries/[19] SiPM PM3350 STD Trench Datasheet rev02.07.2012 [Online]. Avail-

able: www.ketek.net[20] B. Dolgoshein et al., “Large are UV SiPMs with extremely low cross-

talk,” Nucl. Instrum. Methods A, vol. 695, pp. 40–43, 2011.[21] M. Mazzillo et al., “Electro-optical performance of p-on-n and n-on-p

silicon photomultipliers,” IEEE Trans. Electron Devices, vol. 59, no.12, pp. 3419–3425, Dec. 2012.

[22] P. Berard et al., “Performance measurement for a new low dark countUV-SiPM,” in IEEE Nucl. Sci. Symp. Conf. Rec., pp. 544–547, PaperN12-3.

[23] C. Piemonte, A. Ferri, A. Gola, A. Picciotto, T. Pro, N. Serra, A. Tarolli,and N. Zorzi, “Development of an automatic procedure for the charac-terization of silicon photomultipliers,” in IEEE Nucl. Sci. Symp. Conf.Rec., 2012, Paper N1-206.

[24] N. Serra, G. Giacomini, A. Piazza, C. Piemonte, A. Tarolli, and N.Zorzi, “Experimental and TCAD study of breakdown voltage temper-ature behavior in n+/p SiPMs,” IEEE Trans. Nucl. Sci., vol. 58, no. 3,pp. 1233–1240, Jun. 2011.

[25] S. M. Sze, Physics of Semiconductor Devices. New York, NY, USA:Wiley-Interscience, 1969.

[26] S. Vinogradov, “Analytical models of probability distribution and ex-cess noise factor of solid state photomultiplier signals with crosstalk,”Nucl. Instrum. Methods A, vol. 695, pp. 247–251, 2012.

[27] M. Grodzicka et al., “2 2 MPPC arrays in gamma spectrometry withCsI(Tl), LSO:Ce(Ca), LaBr3, BGO,” presented at the IEEE NuclearScience Symp. Conf., 2011, Paper NP5.S-150.