Timing Performance with Broadcom Metal Trench Silicon Photomultipliers

Abstract

Single photon time resolution (SPTR), photon detection efficiency (PDE), and correlated noise rate are important performance characteristics of modern silicon photomultipliers (SiPMs) in consideration of advances in time-of-flight positron emission tomography (TOF-PET). Commercially available Broadcom near-ultra-violet metal-trench (NUV-MT) SiPMs feature metal-filled trench technology to suppress optical crosstalk. We investigated the achievable SPTR and coincidence time resolution (CTR) of NUV-MT SiPMs with various sizes and scintillation crystals, employing low-noise high-frequency readout electronics. The achievable intrinsic SPTRs of 2×2, 4×4, and 6×6 mm² devices were estimated using a picosecond-pulse laser setup. 2- and 4-mm SiPMs were coupled with 2×2×3 mm³ and 3×3×20 mm³ LGSO and BGO crystals to assess achievable CTRs. The intrinsic SPTRs of 2-, 4-, and 6-mm SiPMs biased with 48 V were estimated to be 45, 55, and 137 ps in full-width at half-maximum (FWHM), respectively. The detector comprised a 2-mm SiPM and a 2×2×3 mm³ BGO achieved 111 ps CTR FWHM. The results demonstrated a significant influence of superior SPTR of 2-mm SiPM for the Cherenkov event portion compared to scintillation-based events. The suppressed noise of NUV-MT enabled stable operation at high bias voltage, offering excellent SPTR and PDE for breakthroughs in the timing resolution of TOF-PET detectors.

I. INTRODUCTION

ADVANCEMENTS in time-of-flight positron emission tomography (TOF-PET) have been achieved with continued efforts on improving the photon interaction time resolving power of the radiation detection chain [1], [2], [3], [4]. By measuring the TOF difference of two 511-keV annihilation photons, the origin of the annihilation photons can be estimated in the imaging space with spatial uncertainty determined by the coincidence time resolution (CTR) of the system. Incorporating this information into image reconstruction enables significant signal-to-noise-ratio (SNR) gains, in comparison to a system with no TOF capability [5]. Current state-of-the-art clinical scanners have achieved approximately 200-400 ps CTR, depending upon system design [6], [7], [8], [9], constraining annihilation event origin to 3ñ6 cm regions along system response lines and enabling 2.6 to 3.7 SNR gains (for a 40 cm diameter image space). The latest scanner of Siemens Healthineer was reported to feature a CTR of 178 ps [10]. Next-generation systems aim to push system CTR ≤100 ps, which can more than double the effective system sensitivity. The ultimate goal is to realize systems with CTR which is at the limit of the spatial uncertainty induced by the positron range in tissue from ¹⁸F, at 10 ps, where the TOF-PET technique is fully exploited. In pursuit of these capabilities, the PET instrumentation community aims to push the limits of achievable CTR.

As most of the recent PET scanners employ silicon photomultipliers (SiPMs) for scintillation light detection, the performance of SiPMs plays an important role in achievable CTR. The key metrics of SiPMs for fast-timing applications include single photon time resolution (SPTR), photon detection efficiency (PDE), and uncorrelated and correlated noise rates. Good SPTR leads to improvement in the detector CTR by precise timing of the individual optical photons emitted from radiation interaction and is especially beneficial for prompt emissions such as Cherenkov light [11]. Characterized using high-frequency readout electronics, the SPTRs of recent SiPMs commercially available are reported to be sub-100 picoseconds [11], [12], [13], [14]. Increasing the PDE at the wavelength of relevant optical photon emissions enhances the time resolution of scintillation detectors, proportionally with the inverse square root of the number of detected photons. Both SPTR and PDE improve with increasing bias applied to the SiPM but are limited to an optimum point due to the rapid increase of dark count and optical crosstalk at sufficient overvoltage.

One remarkable technology developed to boost SiPM performance is using non-transparent material for the deep trench isolation of SiPM that serves as an optical insulator for crosstalk photons [15], [16]. Confining the optical crosstalk photon within the microcell where it originated can significantly reduce internal crosstalk probability. This enables operating the SiPM at higher overvoltage to take full advantage of high PDE and large SNR for single photon signals (and thereby reduced influence of electronic noise on SPTR) in signal generation with suppressed noise levels. NUV-MT SiPMs, recently commercialized by Broadcom through co-development with Fondazione Bruno Kessler, showed significant improvements in timing characteristics with minimal loss in the fill factor by exploiting the metal trench technology [16].

In this study, we evaluated the timing performance of the commercial, single-pixel NUV-MT SiPMs with different sizes using low-noise high-frequency (LNHF) readout electronics to investigate the achievable timing performance for advanced TOF-PET detector developments. We first measured their SPTRs and estimated the intrinsically achievable SPTR by quantifying the uncertainty contribution from the experimental setup. We also measured CTRs for LGSO and BGO crystals with various sizes using a 511-keV coincidence setup.

II. Materials and methods

A. SiPMs and readout electronics

Three different NUV-MT SiPMs evaluated in this work, which are referred to as, “2-mm” (AFBR-S4N22P014M), “4-mm” (AFBR-S4N44P014M), and “6-mm” (AFBR-S4N66P014M) SiPMs (Fig. 1(a)) throughout the manuscript. Key specifications are highlighted in Table 1 [17], [18], [19]. Because all three types are based on the same microcell structure and metal trench technology, the electro-optical properties such as PDE, dark count rate, and optical crosstalk probability are the same for all three devices. Exact breakdown voltages for the devices were not quantified, but this characteristic is specified at approximately 32.5 V. Meanwhile, the terminal capacitance increases linearly with the active area due to the parallel connection of an increasing number of microcells.

Fig. 1.

Overview of the NUV-MT characterization. (a) The LNHF readout circuit scheme with a photograph of three tested SiPMs with different sizes. (b) SPTR measurement setup. Single photon events were extracted from the amplitude histogram of the timing signals for analysis. (c) CTR measurement setup for different combinations of the SiPMs and crystals. 511-keV photopeak events were extracted from the amplitude histogram of the energy signals.

TABLE I.

Electronic and Optical Specifications of the Tested NUV-MT SiPMs Presented in the Datasheets

	2 mm	4 mm	6 mm
Single device area [mm2]	2.14×2.14	3.84×3.74	6.14×6.14
Active area [mm2]	2.00×2.00	3.72×3.62	6.00×6.00
Microcell pitch [μm]		40
Number of microcells per pixel	2464	8334	22428
Photodetection efficiency* [%]		63
Dark count rate per unit area [kcps/mm2]		125
Optical crosstalk probability** [%]		23
Terminal capacitance [pF]	160	580	1550

^*For wavelength at peak sensitivity (420 nm), biased at 12 V overvoltage

^**Biased 12 V overvoltage

SiPMs were read out with a LNHF front-end circuit designed for fast timing with low power consumption [13]. Anode signals were split between two different signal processing chains: 1) a timing signal chain consisting of a two-stage radiofrequency amplifier cascade and 2) an energy signal chain consisting of a low-power voltage buffer. The timing output signals exhibited a low baseline noise of 1.6 mV in root-mean-square and 38 dB gain at 2 GHz bandwidth.

B. SPTR measurement

The SPTR measurement setup is illustrated in Fig. 1(b). The 408 nm photons produced by the Picosecond-pulsed diode laser (PiLas; Advanced Laser Diode System) passed through a series of focusing iris, three stages of neutral density filters, and a diffuser. The SiPM was placed farther than 50 times the pixel length from the diffuser, ensuring uniform irradiation of the microcells across the pixel area. The SiPM timing signal and a reference trigger from the laser were sampled by an oscilloscope with a sampling rate of 40 GSa/s and a bandwidth of 3.5 GHz (WaverMaster806; LeCroy). For analysis, only the single-photon events (i.e., an avalanche from a single microcell) were selected from pulse amplitude histograms of the number of fired microcells.

The SPTR of each SiPM was characterized following the method explained in [20]. The total jitter of the time difference histogram between the correlated SiPM and laser signals (${\sigma }_{\text{total}}$) was assessed by fitting a curve which is a convolution of a Gaussian and an exponential probability to account for the delayed carrier avalanche. ${\sigma }_{\text{total}}$ incorporates intrinsic SPTR (${\text{SPTR}}_{\text{intr}}$) of the SiPM along with laser trigger signal jitter (${\sigma }_{\text{trig}}$), and laser pulse width (${\sigma }_{\text{width}}=24$ ps full width at half maximum (FWHM), measured by the manufacturer), and electronic noise jitter (${\sigma }_{\text{elec}}$). ${\sigma }_{\text{trig}}$, the contribution of the laser trigger-out signal, was measured to be 7 ps FWHM by measuring the jitter between the laser trigger split identically between two channels of the oscilloscope. Detector SPTR, ${\text{SPTR}}_{\text{det}}$, was calculated by removing the jitter components relevant to the laser signals:

$${\text{SPTR}}_{\text{det}}=\sqrt{{\sigma }_{\text{total}}^2-{\sigma }_{\text{trig}}^2-{\sigma }_{\text{width}}^2}$$ (1)

${\sigma }_{\text{elec}}$ corresponds to the contribution of the electronic noise to the timing jitter and was assessed as follows:

$${\sigma }_{\text{elec}}=\frac{{\sigma }_{\text{noise}}}{\text{dV}∕\text{dt}}$$ (2)

where ${\sigma }_{\text{noise}}$ is calculated from the FWHM of the signal baseline and dV/dt is the slope of the rising edge measured at the 40-60% region of the single avalanche signal (see Fig. 2(a)). ${\sigma }_{\text{elec}}$ was measured as a function of the bias voltage (${\mathrm{V}}_{\text{bias}}$). ${\text{SPTR}}_{\text{intr}}$ was calculated by further removing the electronic noise contribution as follows:

$${\text{SPTR}}_{\text{intr}}=\sqrt{{\sigma }_{\text{total}}^2-{\sigma }_{\text{elec}}^2-{\sigma }_{\text{trig}}^2-{\sigma }_{\text{width}}^2}$$ (3)

Fig. 2.

Results obtained from SPTR measurement. (a) Typical single photon waveforms timing output of 2-, 4-, and 6-mm SiPMs biased with 48 V. (b) Electronic noise contribution to the jitter (${\sigma }_{\text{elec}}$), (c) intrinsic SPTRs (solid lines) and detector SPTRs (dashed lines) measured as functions of bias voltage.

C. CTR measurement

CTRs of the detector pairs were measured with the coincidence setup depicted in Fig. 1(c). We coupled 2- and 4-mm SiPMs with 2×2×3 mm³ fast LGSO with cerium doping (OXIDE) and BGO (EPIC) scintillation crystals, while 4-mm SiPMs were additionally tested with 3×3×20 mm³ fast LGSO with cerium doping (OXIDE) and BGO (Shanghai Project Crystal). All crystals in this study were mechanically polished, wrapped in Teflon tape, and were optically coupled to SiPMs using optical grease (BC630, Saint-Gobain). A 5 μCi Ge-68 point positron source was placed between opposing detector pairs. The acquisitions were performed inside a light-tight enclosure maintained at 20°C. Both timing and energy output waveforms were digitized by the oscilloscope where the energy output was used to extract coincident 511-keV photoelectric absorption events for CTR analysis. The timestamp of the event was determined by applying a simple leading-edge discriminator threshold to the timing signal without the necessity of digital baseline correction owing to the short pulse recovery time and low trigger rate.

The time difference distribution of the LGSO detector pair was fitted with a single Gaussian curve to assess CTR in FWHM. For BGO, in contrast, a double Gaussian curve was used for accurate fittings of the fast and slow portions that are likely to involve events of which the time pickoff was made by Cherenkov and scintillation, respectively. The fitting was applied to 1) raw distribution and 2) distribution corrected with the rise time classification method. This method corrects the time skews introduced by different relative contributions of fast Cherenkov and slow scintillation to the slew rates of two 511-keV timing signals [21]. For each detector side, the events were sorted by rise time between the lower (10 mV) and upper (70 mV) thresholds and then were divided into 5 categories with even counts, thereby the coincident events were divided into 5×5 categories. We acquired approximately 1,000 511-keV events per category to ensure sufficient statistics for goodness of curve fitting. The time skew of the events in each category was characterized by the centroid of the time difference distribution and was corrected for the double Gaussian fitting of the entire dataset.

III. Results

A. SPTR measurement

Fig. 2(a) shows typical single photon pulse waveforms of the tested SiPM pixels produced by the LNHF circuit with 48 V of ${\mathrm{V}}_{\text{bias}}$ applied. The pulse of 2-mm SiPM featured a sharp rising edge and short decay achieving 1-ns FWHM width, owing to the improved signal bandwidth. Both the slew rate and amplitude degraded for larger SiPM areas as a result of the increasing terminal capacitance.

Fig. 2(b) shows ${\sigma }_{\text{elec}}$ of the single photon signal of the tested SiPMs. The high slew rate of the rising edge (dV/dt) of 2-mm SiPM, as presented in the waveform, led to robustness to the influence of baseline noise on the time pickoff uncertainties. Fig. 2(c) shows SPTR as a function of ${\mathrm{V}}_{\text{bias}}$. The estimated intrinsic SPTRs of the 2-, 4-, and 6- mm pixels at ${\mathrm{V}}_{\text{bias}}=48\mathrm{V}$ were 45 ± 1 ps, 55 ± 1 ps, and 137 ± 4 ps, respectively. The error bar at the data point indicates the 95% confidence level of the curve fitting. The 2-mm pixel showed excellent SPTR without deterioration at ${\mathrm{V}}_{\text{bias}}$ up to 54 V thanks to the low dark count rate and crosstalk probability. The 4-mm pixel was also operable up to 54 V but showed a slight degradation in SPTR and significant fluctuation of the signal baseline above 50 V. Detector SPTRs, which reflect the combined effect of the SiPM and the electronics, were estimated to be 47 ± 1 ps, 61 ± 1 ps, and 172 ± 3 ps for 2-, 4- and 6-mm devices, respectively. The minor contribution of ${\sigma }_{\text{elec}}$ to detector SPTR of the 2-mm SiPM implies that the high slew rate benefitted from minimized electronic noise and fast response of the LNHF electronics.

SPTR of the 6-mm pixel at < 40 V ${\mathrm{V}}_{\text{bias}}$ was not measurable due to its very small signal amplitude, where the time pickoff was impacted by the baseline noise. Even at high ${\mathrm{V}}_{\text{bias}}$ up to 48 V, the 6-mm pixel performed significantly worse SPTR and ${\sigma }_{\text{elec}}$ compared to 2- and 4-mm pixels because of its large terminal capacitance and increased signal transit time skew between microcells.

B. CTR measurement

Fig. 3 and Table 2 summarize the CTR results of 2- and 4-mm SiPMs coupled with 2×2×3 mm³ and 3×3×20 mm³ crystals at bias voltage of 48 V. Fig. 3(a) shows the CTRs of LGSO detectors for different time pickoff thresholds up to 100 mV. When coupled with 2×2×3 mm³ LGSO, the 2-mm SiPM exhibited slightly worse CTR than 4-mm despite its better SPTR, due to scintillation light loss at the edges of the active area from one-to-one crystal-to-sensor coupling. An approximately 200-μm thick cover window surrounding the photosensitive area allows a chance for the optical photons at the edge regions to exit the SiPM through the side surface, resulting in a decrease in photon collection of the active area. This effect is more considerable for small devices where the relative area of the edge region is large. Furthermore, a limitation in the perfect alignment of the SiPM and crystal to the active area was inevitable for the 2-mm SiPM. In contrast, the active area of 4-mm SiPM covered the entire cross-sectional area of all the tested crystals, where a more complete collection of scintillation light is expected. The detector consisting of a 4-mm SiPM and a 3-mm and 20-mm LGSOs achieved superior CTRs of 67 ± 1 ps and 102 ± 1 ps, respectively.

Fig. 3.

CTRs of the 2- and 4-mm SiPMs coupled to (a) LGSO and (b-c) BGO crystals with sweeping the leading-edge time pickoff threshold biased with 48 V. (b) shows FWHM of double Gaussian curve fitted into the raw time difference distribution, while (c) shows that with rise time classification correction applied.

Table II.

CTRs in FWHM of 2- and 4-mm SiPMs coupled with different crystals at optimal time pickoff thresholds (${\mathrm{V}}_{\text{BIAS}}=48\mathrm{V}$).

SiPM size	Crystal size	LGSO	BGO
			Raw distribution	Rise time classification
2 mm	2×2×3 mm3	73 ± 1 ps	145 ± 1 ps	111 ± 2 ps
4 mm	2×2×3 mm3	67 ± 1 ps	168 ± 1 ps	137 ± 2 ps
4 mm	3×3×20 mm3	102 ± 1 ps	252 ± 2 ps	216 ± 1 ps

CTRs of the BGO detectors assessed without and with rise time classification are shown in Figs. 3(b) and 3(c), respectively. Overall, the rise time classification effectively improved CTR by correcting the asymmetry in time difference caused by different portions of Cherenkov detection for each category. In contrast to the case of LGSO, the improved SPTR of 2-mm SiPMs relative to 4-mm SiPMs translates to improved CTR, by narrowing the fast component comprised of fast Cherenkov detections (Fig. 4). Since the average number of detected Cherenkov photons is <2 due to the small yield of Cherenkov photons in BGO which is only ~18 for 511-keV events [22], the timing uncertainty is not impacted by the imperfect coupling of 2-mm SiPMs in the same manner as exhibited in the LGSO measurements. Rather, for 2-mm SiPMs, the fraction of coincidence events where one or more Cherenkov photons were detected in both detectors (indicated as ‘Fast fraction’ in Fig 4) decreased because of the lowered chance of light arriving at the active area by spreading through the protection window at the edge region of the SiPM (edge escape). On the other hand, FWHM for the slow distribution of scintillation-based events was worse, which can also be explained by the aforementioned light loss effect. These results highlight SPTR as a key factor for the precise timing of Cherenkov photons, while that of scintillation is predominantly affected by photon statistics.

Fig. 4.

Time difference histograms of BGO detector measurement fitted with double Gaussian curves (${\mathrm{V}}_{\text{bias}}=48\mathrm{V}$). The upper and lower rows show the raw histograms and corrected histograms with rise time classification, respectively. The first three values in each subplot indicate FWHMs of the fitted curves for total, fast, and slow distributions, while the last value indicates the fraction of the fast distribution among the total.

The measured CTRs of 2×2×3 mm³ crystals as a function of bias voltage are shown in Fig. 5. As expected from the SPTR measurement results, degradation in CTR was not noticeable up to 50 V owing to the highly suppressed optical crosstalk of NUV-MT. A bias voltage higher than 50 V was not applicable due to the large contributions of external crosstalk photons to the baseline fluctuation, inhibiting accurate measurement of the timestamp with simple leading edge time pickoff of 511-keV event signals.

Fig. 5.

CTRs of 2×2×3 mm³ (a) LGSO and (b) BGO crystals coupled to 2- and 4-mm SiPMs at different bias voltages. The legends “Raw” and “RT” correspond to raw distribution and rise time classification, respectively.

IV. Discussion

The dependency of the NUV-MT SiPM properties on the pixel size investigated in this study demonstrated the benefits of small-area pixels in terms of timing performance. The 2-mm SiPM exhibited a fast single photon response with LNHF electronics, owing to its small terminal capacitance. This provides very low electronic noise influence and excellent intrinsic SPTR of nearly 45 ps. In addition, small pixels benefit from a small transit skew in signal track lengths of microcells across the pixel area [23]. A recent study with a 1×1 mm² research SiPM has also shown that employing techniques to overcome intrinsic limitations in achievable SPTR due to interactions near the edge of microcells can enable SPTR as low as 28 ps FWHM [14]. Regarding the CTR measurements, imperfect light collection of 2-mm SiPMs from the 2×2×3 mm³ crystals was a limitation for investigating the pure impact of the SPTR of SiPMs on scintillation-based events. Indeed, a leading edge discriminator for scintillation signal time pickoff is governed by the combined contributions of SPTR and PDE [24] and SPTR is not the driving factor for CTR for this detector configuration. On the other hand, timing uncertainty from Cherenkov detection is strongly influenced by achievable SPTR due to prompt emission and extremely low yield [11]. Enabled by superior SPTR, the 2-mm SiPMs achieved 111 ps CTR FWHM with 2×2×3 mm³ BGO crystals when biased with 48 V. The CTRs of the detectors with 4-mm SiPMs measured in this study were comparable to those reported from other high-frequency electronic configurations [25]. Since the measured SPTR for the 6-mm SiPMs was much worse than that achieved with the 2- and 4-mm devices, we did not measure CTR with the 6-mm SiPMs, as our aim is to present the current timing performance limits with the technology.

The impact of internal crosstalk suppression was represented as stable operation over a wide range of ${\mathrm{V}}_{\text{bias}}$, as also demonstrated in [16]. The previous version of SiPM, NUV-HD, showed current divergence at excess bias of ~10 V due to rapid growth in crosstalk probability, hence stopping the improvement in CTR at an early point when increasing ${\mathrm{V}}_{\text{bias}}$. However, the 2-mm NUV-MT SiPM, in particular, exhibited a steady enhancement of SPTR towards the excess bias of ~21.5 V (${\mathrm{V}}_{\text{bias}}=54\mathrm{V}$). The achievable CTRs measured with devices in this study also show improved performance, as compared with those reported for NUV-HD SiPMs in the literature that also implemented high-frequency readout electronics, including 113 ps for 3×3×20 mm³ fast LGSO [13], 234 ps for FWHM for 3×3×15 mm³ BGO [26], and 136 and 265 ps FWHM for 2×2×3 mm³ and 3×3×20 mm³ BGO crystals, respectively [14]. Although other differences in characteristics between NUV-HD and NUV-MT may be relevant as well, the results imply that new metal trench technology allows high PDE and high signal-to-noise ratio for fast timing to be exploited by alleviating the limitation introduced by correlated noise. Aside from the metal trench technology, other strategies for pushing the limit of the intrinsic SPTR of the SiPMs targeting are also under development, such as masking out the low-electric-field regions of the microcells to enhance the effective SPTR [14], [27], which would be studied in the future for performance comparison.

The excellent SPTRs of SiPMs measured in this study provide a strong motivation to aim for the full realization of Cherenkov photons from BGO. Assuming that a hypothetical detection system is capable of reading out the individual photon timestamps in a “digital” manner and identifying the first photon arrival accurately, the impact of SPTR is further emphasized to push the CTR of BGO detectors [11]. Time pickoff from the leading edge of the “analog” signal which contains a mixture of scintillation and Cherenkov light also gains from improvements in SPTR, but the impact is mitigated by the influence of slower scintillation light on time pickoff. Moreover, it is reported that suppressing prompt emissions with higher travel dispersion is advantageous for improving CTR [28], [29], which implies that pileup of late impinging Cherenkov photons in the rising edge also degrades CTR. Therefore, a drastic improvement in the achievable time resolution of Cherenkov radiator-based detectors is conceivable by employing the high-performance SiPM and readout electronics investigated in this study, along with strategic extraction of the first Cherenkov photon, which is the subject of future studies [30].

V. Conclusion

We characterized the timing performance of NUV-MT SiPMs with different sizes utilizing low-noise high-frequency front-end electronics. The estimated intrinsic SPTRs at a bias voltage of 48 V were 45 ps, 55 ps, and 137 ps FWHM for 2-, 4-, and 6-mm SiPMs, respectively, indicating the advantage of a small active area on signal bandwidth. The CTR measurements with various crystal couplings showed that the CTRs of BGO detectors significantly improved with the 2-mm SiPM compared to the 4-mm SiPM (111 ps vs. 137 ps for 2×2×3 mm³), while LGSO detectors showed relatively small dependence on SPTR. The study emphasized the benefit of good SPTR for Cherenkov-based timing estimators, suggesting the potential for breakthroughs in time resolution of prompt photon-emitting detectors.