Development of Dual-Gain SiPM Boards for Extending the Energy Dynamic Range

The region between several hundred keV to several hundred MeV presents one of the most under-explored regions of the astronomical sky [1]. The last major observatory dedicated to the MeV range was the Compton Gamma-Ray Observatory which was de-orbited in the year 2000 [2]. The difficulty associated with MeV gamma-ray instrumentation is largely due to physical phenomena in the MeV region that include low interaction cross sections, high background, and activation of the instrument. Nevertheless, observation in the MeV band will aid in the study of nucleosynthesis in supernova [3], supermassive blackholes [4], and aid in the multi-messenger observation of astrophysical objects [5].

In response, a medium-class Explorers (MIDEX) mission concept named the All-sky Medium-Energy Gamma-ray Observatory-eXplorer (AMEGO-X) is proposed to tackle this observational gap. AMEGO-X will act as a Compton and pair telescope [6]. The telescope consists of two subsystems, a tracker and a calorimeter, and are designed to assemble as four identical towers arranged in a $2\times 2$ array. The tracker, led by NASA’s Goddard Space Flight Center, is composed of layers of silicon complementary metal-oxide-semiconductor (CMOS) monolithic active pixel sensors [7]. It will record the position and energy deposited during a scattering event or track the particles from any ensuing electromagnetic showers. The other subsystem is a thallium-doped CsI (CsI:Tl) calorimeter in development by the U.S. Naval Research Laboratory (NRL). The calorimeter will act as the absorber plane for scattered events and catch the ensuing electromagnetic showers from high-energy gamma rays. Fig. 1 shows a computer-aided design (CAD) drawing of the gamma-ray telescope on board AMEGO-X. The figure also presents an exploded view of a single tower. Fig. 2 shows the CAD model of the CsI calorimeter for a single tower. Its design is hodoscopic in nature such that each layer contains rows of CsI logs. The direction of the logs alternates orthogonally for each layer. Currently, the logs are slated to be $15\times 15\times 388\ \mathrm{mm}^{3}$ in volume with the calorimeter consisting of 24 logs per layer for a total of four layers.

To improve the technological readiness level (TRL) of the calorimeter, we developed a CsI calorimeter R&D prototype for the ComPair balloon-borne telescope [8], which utilizes $16.7\times 16.7\times 100\ \mathrm{mm}^{3}$ CsI:Tl logs read out by a $2\times 2$ array of $6\times 6\ \mathrm{mm}^{2}$ SiPMs [9]. The SiPMs are commercial off-the-shelf (COTS) arrays with a specific model ARRAYJ-60035-4P-PCB by onsemi [10]. The prototype calorimeter, which adopts the design aspects of the Fermi Large Area Telescope’s (LAT) calorimeter, is modular and consists of five layers of CsI:Tl logs arranged in rows of six. The logs are also arranged in a hodoscopic fashion. Read out on each end with silicon photomultipliers (SiPMs) [11] allows for the reconstruction of energy and depth for each interaction. SiPMs present a great many advantages for reading out scintillators when compared to photomultiplier tubes (PMT). Among them are their low power and voltage requirements, insusceptibility to magnetic fields, low weight, and compact form. However, the number of microcells limits their signal in addition to affecting their susceptibility to radiation damage [12].

At the Triangle University Nuclear Laboratory’s (TUNL) High Intensity Gamma-ray Source (HIGS) facility [13], we irradiated the prototype calorimeter system with mono-energetic gamma rays that vary from $2-30\ \mathrm{MeV}$ [9]. Fig. 3 presents the recorded energy calibrated using a $1\ \mathrm{MeV}$ line vs. the known beam energy. The plot reveals the non-linearity in the system which is possibly due to the readout electronics [14], scintillator light-yield non-proportionality [15], and SiPM non-linearity. The SiPM non-linearity may arise from the microcell size as well as the full active area. As reported in [16, 17], the non-linearity can be introduced by the limited number of pixels and active area. The performance of the prototype does not meet our desired upper energy range of $400\ \mathrm{MeV}$. Nevertheless, in principle, such a response can be fitted and corrected but will naturally limit the dynamic range of the system [18].

In response, we developed custom dual-gain SiPMs that utilize two species of SiPMs, one optimized for low-energy ranges and one for high-energy events. The two species have different-sized active areas and might contain a mix of microcell sizes. The concept behind multi-sized SiPMs is inspired by the Fermi LAT CsI calorimeter readout [19] that consists of dual-range PIN diodes to cover the $5\ \mathrm{MeV}-60\ \mathrm{GeV}$ range for a single crystal. Although this work is sponsored to develop readout in support of AMEGO-X, dual-gain SiPMs could be applied to any scintillator-based calorimeters such as ones proposed for use on GECCO [20, 21] or ASTROGAM [22]. This work reports on the development and testing of these custom SiPM to read out CsI:Tl scintillators. Sec. II introduces the custom SiPMs. Sec. III presents their response to gamma rays while Sec. IV reports on the SiPM’s response to high-energy protons. Sec. V discusses the results and shows the spectral unification of the two SiPMs.

The dual-gain SiPMs concept relies on fielding two different SiPM species with differently sized active areas and microcells. This concept aims to mitigate the saturation of the SiPMs and electronics. Fig. 4a shows the SiPMs utilized in the ComPair CsI calorimeter prototype [9] while Fig. 4b shows the layout of the custom board with two SiPM species, which we will refer to as ‘dual-gain SiPMs’. The dual-gain SiPMs consist of four $1\times 1\ \mathrm{mm}^{2}$ (‘high-energy’) SiPMs that are placed in the center of the carrier board to cover the high-energy range. They are surrounded by eight (‘low-energy’) SiPMs, each with an active area of $3\times 3\ \mathrm{mm}^{2}$ that will cover the low-energy range. Hereafter, the group of four $1\times 1\ \mathrm{mm}^{2}$ are referred to as as $1\ \mathrm{mm}^{2}$ and eight $3\times 3\ \mathrm{mm}^{2}$ as $9\ \mathrm{mm}^{2}$, based on their active area. Both SiPMs were manufactured by onsemi. Fig. 5 shows a microscope view of the dual-gain SiPM.

We investigated the dual-gain design on $3\times 3\times 6\ \mathrm{cm}^{3}$ CsI chunks. We used the chunks due to their larger detection efficiency as well as serving as a potential geometry for an array-like calorimeter. We also studied CsI logs $(1.67\times 1.67\times 10\ \mathrm{cm}^{3})$ that matched the current calorimeter prototype. While the logs utilize a SiPM on each end (the small faces), the chunks were read out on only one of the $3\times 3\ \mathrm{cm}^{2}$ side. Each scintillator crystal is wrapped with two layers of Tetratex and then warped with electrical tape to maintain light tightness. They were then placed in an additional box to maintain further light tightness. Fig. 6 shows the layout of the scintillators used in the experiment. Each scintillator was placed into a 3D-printed assembly. In the following experiments, the beam’s vector was always out of the page.

Table I summarizes the different custom SiPM versions that were developed for initial testing. Note that these designs are used to demonstrate an initial concept rather than a slated design. For the experimental component of this work, we manufactured two logs and one chunk for a given SiPM version. Table II summarizes the part number of the utilized SiPMs.

The dual-gain setup was tested at the HIGS facility. Due to beam time constraints, we only tested the chunks. The tested energies were $8,15$ and $20\ \mathrm{MeV}$. We also experimented with an $11\ \mathrm{MeV}$ beam. However, during the run, we received unexpected results that were more consistent with a faulty beam. We therefore omit any analysis from the $11\ \mathrm{MeV}$ beam. During the experiment, the high-energy SiPMs of V1 and V2 did not show any response and we therefore conclude the lower threshold on the $1\ \mathrm{mm}^{2}$ SiPMs is higher than $20\ \mathrm{MeV}$. However, V3 displayed an overlap between the two SiPM species.

Fig. 7 plots the different spectra taken at the HIGS facility for the (a) $9\ \mathrm{mm}^{2}$ and (b) $1\ \mathrm{mm}^{2}$ of the V3 SiPMs. When using the $8\ \mathrm{MeV}$ and Th-228 source, which emits a $2.614\ \mathrm{MeV}$ gamma-ray, the full-energy peak is visible in the $9\ \mathrm{mm}^{2}$ SiPMs along with the single and double escape peaks [26]. At higher energies, due to the prevailing physics, those features have disappeared and the spectra have transformed into continua with their endpoint proportional to the maximum deposited energy of the gamma. When analyzing the $1\ \mathrm{mm}^{2}$ SiPMs, we observe that they have a significantly lower ADC value for the same source energy when compared to the $9\ \mathrm{mm}^{2}$. We also do not observe the full-energy peak as seen in the $9\ \mathrm{mm}^{2}$ in the lower energies, but in general, the two groups show the same spectral trends at $8\ \mathrm{MeV}$ and above. Appendix A discusses simulation efforts for this experiment showing similar trends as the experimental data.

Fig. 8 plots the recorded full-energy features as a function of the beam energy. When peaks are visible, a Gaussian fit was applied. When peaks were not visible (i.e. the 15 and 20 MeV runs), a linear fit was applied for the high-energy fall-off to find the endpoint energy. Since the beam flux was low enough to avoid chance coincidences, the highest recorded energy that is part of the spectral distribution most likely originates from a full-energy deposition of the beam gamma rays. We therefore apply a linear fit on the range of 75%-25% of the max peak on the high end of the spectrum. The plot also shows the clear overlap between the two species that ranges between $\sim 2.6$ to $20\ \mathrm{MeV}$ and linearity across the overlap region.

To study the high-energy performance of the custom SiPMs, we conducted high-energy proton beam experiments at the Crocker Nuclear Laboratory at the University of California$-$Davis [27]. The beam delivered $64\ \mathrm{MeV}$ protons and was collimated using a $0.61\ \mathrm{mm}$ diameter pinhole. We then added various aluminum attenuators throughout the experiment to reduce the beam’s energy. Therefore, the beam energy ranged from $27$ to $60\ \mathrm{MeV}$, which we estimated using SRIM [28]. Fig. 9 shows the experimental setup containing both the logs and chunks.

This work demonstrates the dual-gain SiPMs concept. The most crucial observation was to check that the energy ranges of the two species overlapped, ensuring continuous coverage over the entire energy range of interest. The results from HIGS (Fig. 8) show the clear overlap in ranges with data points ranging from $2.6$ to $20\ \mathrm{MeV}$ for the V3 SiPMs. It is in this region that we will declare the turnover point between the $9\ \mathrm{mm}^{2}$ and $1\ \mathrm{mm}^{2}$ SiPM group.

The goal of the experiment at the Crocker Nuclear Laboratory was to verify the overlap region and to study the high-energy response of the SiPMs. Protons were chosen as they can easily deposit more energy into the log than high-energy gammas. With the different proton energies, we were able to plot the response and estimate that the higher end of the energy range is around $400\ \mathrm{MeV}$ for the logs and chunks alike when linearly extrapolating with the acquired data. The upper energy limit is derived using the ROSSPAD’s 12-bit ADC range with the assumption that the SiPM/readout electronics maintain linearity throughout the energy range and that they do not saturate. In reality, both the SiPM and the ROSSPAD do experience non-linearity and therefore the true upper limit might be lower than $400\ \mathrm{MeV}$. In addition, we did not have the opportunity to sample higher energies to study the non-linearity of the $1\ \mathrm{mm}^{2}$ as well as its upper threshold. Therefore, with the aforementioned assumptions, the combination of the two SiPMs with the ROSSPAD may result in an energy range from $250\ \mathrm{keV}$ to $400\ \mathrm{MeV}$.

We note that the linearity that is experienced by each group might differ, due to the limited number of pixels and the active area as is commonly described with (3):

where $N_{\mathrm{cells\ fired}}$ is the number of cells that fired and is proportional to the signal, $N_{\mathrm{pixel}}$ is the total number of pixels while $N_{\mathrm{photon}}$ is the total number of photons in a pulse. $\epsilon$ is the photon detection efficiency. However, this model does not include non-linear light yield from the scintillator, noise in the SiPM, and non-linearity in the readout electronics.

Fig. 16 plots unified data between the two species using both the gamma and proton data where we chose $20\ \mathrm{MeV}$ as the transition energy. The energy calibration is derived from a linear correction applied to each SiPM group. This plot displays the dual-gain concept in which $9\ \mathrm{mm}^{2}$ SiPMs are sensitive to “low-energy” events while the $1\ \mathrm{mm}^{2}$ SiPMs take over for “high-energy” events.

During the log scanning effort in Sec. IV-B, we also irradiated the SiPMs with high-energy protons. In doing so, we have observed no overflow triggers in the $1\ \mathrm{mm}^{2}$ SiPMs or events associated with proton-SiPM interactions. There, the protons entered along the plane of the SiPM. The negligible effect is most likely due to the small active region in the SiPM to which the proton deposits minimal energy. In addition, the SiPM is resistant to avalanche effects from one microcell to another in standard operations [30].

This initial investigation demonstrates the responses of dual-gain SiPMs utilized to read out the scintillator. The low-energy SiPMs can cover the ‘low’ energy range and the high-energy SiPMs will accommodate the ‘higher’ energy range. While the high-energy SiPMs have a measured up to 60 MeV with protons data. According to our calculation, and assuming the SiPMs and readout system do not experience non-linearity, the dual-combination might provide a dynamic energy range from $250\ \mathrm{keV}$ to $400\ \mathrm{MeV}$ using the SiPHRA’s 12-bit ADC. Dedicated tests on SiPMs response and photoelectron yields are needed to confirm this hypothesis. In the future, we plan to build a flight-like calorimeter that is intended to serve as a single tower for the AMEGO-X mission concept [6]. We are also investigating the usage of custom dual-gain SiPMs manufactured by Fondazione Bruno Kessler (FBK).

The authors thank Alexander A. Moiseev for the opportunity to join his HIGS beam test. We are also grateful to Michael Backfish, the staff at the Crocker Nuclear Laboratory, and the accelerator physics group at High Intensity Gamma-ray Source (HIGS). We thank Ms. Emily Kong for simulation support in this project.

Much appreciation is given to the anonymous referees for the most valuable comments which have improved this manuscript.