Keywords

1 Introduction

Gamma radiation detectors find applications in multiple fields, such as nuclear physics experiments, nuclear medicine, molecular imaging, and homeland security [6, 7, 12, 14]. Often, the novel detector architectures, electronic systems, or data processing techniques introduced in one of these fields also offer potential improvements for the other applications. A substantial push in technological innovation is driven in this context by fundamental research.

This article focuses on presenting a detection module developed for Istituto Nazionale di Fisica Nucleare (INFN) research institute, aiming at spectroscopy and position sensitivity of gamma photon at medium and high energies, while superseding photomultplier tube-based detection modules with silicon photomultipliers (SiPM). The use of SiPMs allows to fully exploit scintillation crystal technologies, e.g. enabling a wide dynamic range (from 1 to 105 photons per cm2) and excellent resolution (2.6% at 662 keV, the photopeak emission energy of 137Cs) when using a co-doped lanthanum bromide crystal. Motivations and architecture of GAMMA detection module are respectively introduced in Sects. 2 and 3, while the GAMMA ASIC is described in detail in Sect. 4. A machine learning-based reconstruction algorithm and the experimental validation of the spectroscopic module are presented in Sect. 5.

2 GAMMA: A Novel Indirect Conversion Detector

Gamma photons are commonly probed in physics experiments, aiming at collecting sets of data that are representative of their energy, the point of interaction with the detection system, or the time of interaction. Conversion of the physical quantity into a probable signal is required during the experiment, and the physical variable is often transformed in an electric signal.

In indirect conversion methods a scintillation material (which, following an interaction with a gamma photon, emits many low-energy photons, typically in the near infra-red, optical or near ultra-violet energy band) is coupled to a photosensor array. The amount of light produced is ideally proportional to the energy deposited by the incident radiation in the detector.

Scintillator-based gamma detectors fall in one of the following categories: detectors with monolithic scintillators, and detectors with pixelated scintillators. A monolithic scintillator consists of a continuous block of scintillator material. In Anger Cameras, a single monolithic scintillator crystal is coupled to an array of photodetectors. These cameras incrementally gained attention in the years: besides requiring smaller fabrication costs per unit area, they also allow to estimate the depth-of-interaction and to improve the spatial resolution beyond pixel size, being the resolution of this solution related instead to the statistical fluctuation occurring in the physical process of light sharing among pixels and to the specific reconstruction algorithms implemented. Secondly, the use of bulkier scintillators implies a higher stopping power, which is a key factor in accelerating the data collection of physics experiment in the range of MeV energies, often resulting in better spectroscopy performances with respect to pixelated crystals, given the higher probability of total energy absorption. Scientific literature on spectroscopy and imaging modules based on monolithic scintillators is extremely rich of applications, and often a test bench for novel technologies of photon sensing and reconstruction algorithms.

It is shown in Fig. 1 a 3D model of the Anger Camera that is introduced in this section, named GAMMA, coupling a 3”\(\times \)3” crystal to a 12\(\times \)12 array of SiPMs (simulation run in ANTS2 environment). The development of front-end readout, system electronics, back-end software, and data processing algorithms for this modern, compact gamma spectrometer was commissioned by the GAMMA experiment, supported by the INFN. Main investigation methods of GAMMA experiment involve the use of gamma spectroscopy: focus of the research activity is the study of shell structure and collective modes, with particular interest towards nuclei produced in exotic beams. Many experiments are based in ion beam facilities, and involve the use of arrays of detectors; the GAMMA experiment is often involved in the R&D phase of innovative detection modules, following with particular interest the latest discoveries in the fields of scintillator materials and readout electronics.

Fig. 1
figure 1

3D model of a monolithic scintillator with SiPMs readout, and simulated paths of the scintillation light emissions in the cylindrical \(3'' \times 3''\) crystal. The nine tiles of SiPM are arranged in a planar, 12\(\times \)12 pixels array. Simulation run in ANTS2 environment [11]

Full size image

The lanthanum bromide crystals was selected for GAMMA module for its excellent conversion efficiency (allowing energy resolution <3% at the 137Cs emission energy of 662 keV) and fast decay time (16 ns, which potentially enables a theoretical count rate capability higher than 10 MCPS). Lanthanum bromide crystals also have a satisfactory radiation tolerance to high-energy proton irradiation with fluences typical of those in the interplanetary space (LaBr3 was studied for various space missions), and show sufficient radiation hardness when exposed to a high flux of MeV \(\gamma \)-rays.

The GAMMA module is depicted in Fig. 2. In the exploded view, we can identify the main components of the 13 cm\(\times \)13 cm\(\times \)15 cm module: the aluminum enclosure, the scintillation crystal coupled to the SiPMs matrix, and the readout electronics. The front-end collects the charge signals from SiPMs in the low-impedance input stage of the ASIC and provides the conditioned signal to the FPGA-controlled ADCs; the signal is digitized and sent to the central unit.

The lanthanum bromide crystal is coupled to a square array of 144 SiPMs, made up of 9 individual tiles; each tile is a 16 elements square array of 6 mm\(\times \)6 mm SiPM detectors.

Fig. 2
figure 2

(left) The 13 cm\(\times \)13 cm\(\times \)15 cm3 aluminium enclosure encompasses scintillation crystal, the 144 SiPM detectors, front-end electronics and the DAQ. (right) The SiPMs hosted by the Motherboard generate 144 current signals, which are monolithically collected by 9 GAMMA ASICs and digitized by the FPGA-based DAQ on the Powerboard

Full size image

3 Architecture of GAMMA Module

When a gamma photon interacts with the scintillation crystal, photocarriers generated in the SiPMs induce current signal pulses that are processed by the electronics and converted to a digital data. Each SiPM is individually coupled to a low-impedance input node of the front-end GAMMA ASIC [13] via the Motherboard (Fig. 2), in order to convey signal current pulses from SiPM anodes to the analog filter stages while keeping a constant voltage across the detector. If a current pulse is detected from one of the 144 channels, the integration starts for all channels; the readout is therefore monolithic and self-synchronized to particles or gammas interactions in the crystal.

Multiple instances of a 16-channel custom electronic front-ends, the GAMMA ASICs, collect the charge from each of the SiPM anodes individually (for a total of 144 analog channels). Each ASIC channel features a gated integrator stage; each channel has three automatically set gain. The gain is individually selected in each channel by the Adaptive Gain Control (AGC) circuit, the central role of which is discussed in further detail in Sect. 4.2, integrating a support circuitry which predicatively estimates in the analog domain the charge accumulated during the integration period. The voltage dynamic of the gated integrator is therefore optimized on an event-per-event basis and the dynamic range of each channel is extended to 84 dB, which in turn allows to fully exploit the SiPM detectors dynamic range. Superior spectroscopic performance on a wide energy interval also allows to overcome in physics experiments the trade off between energy resolution and observed energy interval, cutting the experiment beamtime.

Fig. 3
figure 3

Block scheme of GAMMA module. The front-end circuit is made up of 16-channel GAMMA ASICs, allowing in the module an individual, low-noise readout of each pixel. The SiPM bias voltage is controlled in closed loop, in order to compensate for gain drifts of the system due to temperature fluctuations

Full size image

The multiplexed voltage outputs of the ASICs are digitized by 13-bit, 5 MSPS, differential input ADCs, one for each ASIC. The converted value is gathered by a Xilinx Artix-7 FPGA, which also serves as a bridge with the custom control application hosted on an external PC.

The module has an analog output proportional to the event energy that, together with the trigger signal coming from the ASICs, can be used to embed the detector in pre-existing facilities that use dedicated multichannel analyzers or digitizer synchronized with the arrival of a gamma photon.

An ARM Cortex-M4 microcontroller collects the temperature data from the SiPM’s temperature sensors and acts on their bias to maintain constant the gain even in the presence of strong temperature variations [8]. The block scheme of the module is shown in Fig. 3.

4 GAMMA ASIC

The GAMMA experiment targets spectroscopy in an energy interval from 20 keV to 20 MeV, and count rates up to 50 kcps. In order to determine the amount of optical photons collected by each SiPM (and correspondingly the range of charge at the input of each ASIC channel), the light emission of the \(3''\times 3''\) crystal, collected by an array of 144 SiPMs of 6 \(\times \) 6 mm2 area each, has been simulated by means of ANTS2 software [11] up to 20 MeV (Fig. 4).

The light distributions extracted by these simulations showed that, considering the minimum SiPM signal at the minimum gamma energy and the maximum signal at the maximum energy, the number of photons collected by each SiPM in the array ranges from few photons to about 104 photons, providing a maximum charge at the ASIC input of 3 nC. Maximization of scintillation light collection is a key factor to enhance resolution, but in order to have a circuit that allows both single-photon resolution and the collection of 104 photons, the dynamic range of each channel has to be larger than 80 dB.

Fig. 4
figure 4

Comparison of equivalent input noise contributions to energy resolution as a function of the incoming photon energy in the 100 keV–20 MeV range, obtained from numerical simulations in ANTS2 software. The adaptive gain control circuit allows for lower noise at low energies and, in turn, to extend the energy interval of the measurement. Fixed gain acquisitions are degraded at low energies by the electronics noise, in the energy interval where the electronics noise results comparable or even higher than the one associated to the collection of photons in the detector (statistical noise)

Full size image

Many ASICs have already been developed for the readout of SiPMs, and a review can be found, for instance, in [5]. Commercially available ASICs are mainly developed for medical applications like PET, with limited DR. Considering our requirements (200 fC equivalent noise charge, 3 nC full scale range) we have decided to develop a new ASIC that is able to provide at the same time a low noise in the lower energy region and high DR capability.

4.1 GAMMA ASIC Top Level Architecture

The block diagram of the ASIC is shown in Fig. 5. The circuit can be connected to 16 SiPMs providing a simultaneous readout of the charge delivered by the detectors with independent AGC in each channel. The outputs of the channels (analog amplitude and digital code corresponding to the used gain for each integration) are multiplexed, converted to fully-differential signal, and routed towards the chip output pads.

Fig. 5
figure 5

GAMMA block diagram. The chip features 16 SiPM inputs and a multiplexed analog output. The Current Conveyor stages collect the signal charge, that is integrated in all filter stages when the input signal overcomes a programmable threshold in the Trigger circuit. The parallel outputs are multiplexed and sent to an external ADC by the fully differential output buffer. The AGC circuit adapts the gain of the filter to the signal amplitude

Full size image

4.2 Adaptive Gain Control circuit

In Fig. 4 the overall statistical noise and the total electronics noise (both extracted from simulations and numerical analysis, assuming that the gain is constant or it is set by the AGC circuit) are reported for the complete energy range. Both contributions are extracted from ANTS2 simulations; if the AGC is active we assume that each ASIC channel has adopted the best gain (that is, the highest gain that does not induce saturation) in the measurement, out of three available settings.

If a constant gain is considered for the amplifier, the simulation showed that the electronics noise is dominant with respect to the statistical noise in the low energy range, affecting the overall resolution up to many hundreds keV; the relation \(\sigma _{stat}/\sigma _{elect}\) \(\le \) 4 only holds down to 600 keV. The constant gain of the readout electronics implies in fact a trade-off for the readout electronics between low-noise and dynamic range capabilities.

We have introduced in the ASIC the AGC circuit, which automatically increases the integration capacitance of the gated integrator stage to reduce the gain for large signals. This technique, already implemented in other ASICs for x-ray detection [2] or astrophysics experiments [3], allows to implement a piece-wise linear characteristic and an overall gain compression.

Because the dynamic range of each individual energy range showing constant gain has to decrease in each interval in order to cope with design requirements (the statistical noise doesn’t increase linearly with the input signal), the author found non beneficial to implement more than 3 possible integrator feedback capacitance settings. A transistor-level schematic description of the gated integrator stage is shown in Fig. 6.

Fig. 6
figure 6

Implemented gated integrator filter with adaptive gain control block. The feedback resistance Rf is connected until the start of the integration phase. The S &H switch is open until the end of the integration phase, then it is closed (the two complementary switches are opened) and the integrator holds the charge to be read. At the end of the read phase, the feedback capacitance is discharged by switch RES

Full size image

4.3 Current Buffer Input Stage

The input stage requirements strongly depend on the characteristics of the SiPM detectors. A typical circuit model for SiPM NUV-HD detector that was adopted in our simulation phase is described in [1, 5, 9]. Key parameters of such model are the SiPM detectors reset time, the micro-cell capacitance, and the ASIC input impedance, the values of which also define the minimum time constant of the input current signal [5]. We considered that in potential applications of our ASIC (e.g., in compact spectrometers for radiation monitoring, or in measurements with a much larger number of channels), it could be useful to merge several SiPMs (NSiPM) to a single readout channel. In this case, the input capacitance scales linearly with the number of detectors connected. In particular, a large capacitive load at the input stage may impose severe challenges to the frequency response and even stability issues when a classical negative-feedback regulated cascode configuration, adopted in several commercial ASICs, is considered [10]. The current-buffer Temes architecture allowed for the merging, in the experimental measurements described in [4], of few tens of SiPMs for a total input capacitance up to 70 nF, while providing an input impedance sufficiently low for the integration to last few microseconds.

Fig. 7
figure 7

(left) Superposition of the decay spectrum from 133Ba,137Cs,60Co and AmBe radioactive sources and spectral lines generated by a calibrated laser-based test setup, measured with GAMMA module. (right) Experimental position sensitivity characterization setup and 3D plot of the validation dataset confusion matrix for the Decision Tree algorithm embedded in the FPGA. Each True Class row corresponds to a different position of the collimated 137Cs source. Values on the diagonal correspond to the correct classification rates

Full size image

5 Spectroscopy and Position Sensitivity Experimental Measurements

The spectroscopy capability of the system was tested using 133Ba,137Cs,60Co and AmBe radioactive sources and using a custom laser-based setup, in order to simulate the scintillation light of LaBr3 generated by the interaction with gamma photons at energies higher than the ones available.

The superposed spectra are shown in Fig. 7. The experimental measurements show a 2.7% energy resolution at 662 keV that monotonically decreases at higher energies, reaching the value of 1.0% at 9.0 MeV. The energy resolution in the measured energy interval is well fit by a curve proportional to \(1/\sqrt{E}\). The non-linearity error of the measurement is 0.6%, and is limited by detector non-linearity at higher energies. Maximum measured count-rate is 80 kCps and the overall power consumption of the module is 6 W.

Position sensitivity was tested scanning the scintillation crystal with \(\gamma \)-rays emitted from a 137Cs source, collimated on the orthogonal direction to the SiPM detectors plane in a 1 mm large pinhole. The scintillation crystal is irradiated from the top side on a square grid that has 1.5 cm pitch. Collected data are labelled using the scan positions, which are known, and are used as training datasets in classification algorithms.

Training a decision tree algorithm and embedding the classifier in the FPGA-based DAQ, from 80 to 90% (depending on the real beam position) of the events 2D reconstructed position are classified with an error smaller than 2 cm, computed from a validation dataset acquired using the same procedure of the training dataset (Fig. 7).

6 Conclusion

GAMMA module is a compact spectrometer designed for nuclear science experiments (13 cm\(\times \)13 cm\(\times \)15 cm, 6 W power consumption), which successfully exploits the codoped LaBr3 resolution capability (2.7% at 662 keV, and 1% at 9.0 MeV) and SiPM linearity (0.6% error at 9 MeV). Pixellation of the detector matrix allows for coarse position of interaction sensitivity in the scintillation crystal.

The module that features nine GAMMA ASICs will soon be used in beamline facilities in order to validate its energy resolution performances using higher energy gamma-ray sources, and to validate the relativistic Doppler effect correction with machine learning algorithms.