24 August 2021 in IEEE Transactions on Nuclear Science
Gain amplitude modulation multichannel ASIC (GAMMA) is a 16-channel application specified integrated circuit (ASIC), designed in 0.35- $\mu \text{m}$ technology to read the charge provided by silicon photomultipliers (SiPMs) on a wide amplitude range. This chip was designed to satisfy the demanding requirements of the INFN the GAMMA project, where a 3” $\times3$ ” large lanthanum bromide crystal is to be read by SiPMs, aiming at state-of-the-art energy resolution. Because the energy range of the application is 100 keV–20 MeV, signals in a wide dynamic range (DR) have to be processed by the ASIC. This wide DR requirement raised the necessity of an adaptive gain change solution in order to guarantee a high signal-to-noise ratio together with a large full-scale range. While the ASIC input stage features a programmable static gain, the analog filtering channel exploits a self-triggered gated integrator stage with an automatically adjusted gain for each detected input current pulse; one of the following values is selected: 10, 2, or 0.83 [mV/pC]. The gated integrator gain is predictively adjusted before the circuit reaches saturation due to a time-gated threshold mechanism and provides negligible charge loss when the gain is switched (< 20 fC). A preintegration feature is also implemented in order to get rid of nonlinearity due to delays in the integration phase that could take place in a multi-ASIC system. The ASIC uses 5- and 3.3-V bias voltages, the overall power consumption being 280 mW. The ASIC was tested with a $12\times12$ SiPMs matrix coupled with a 3 inch LaBr 3 , providing good spectroscopic performances (2.6% at the 137 Cs photopeak emission energy of 662 keV), and was also tested irradiating the SiPMs with fast, bright NUV light pulses, in order to explore the ASIC wide range coverage while simulating the scintillation light emission of a lanthanum bromide crystal.
8 October 2020 in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
The enhancement of a compact gamma-ray detection module for spectroscopy and imaging with machine learning for directional sensitivity is presented. In particular this development is targeted towards drone-based localization of radioactive sources in the environment. The unit is composed of a cylindrical monolithic scintillator crystal (3” × 3” LaBr3(Ce3++Sr2+)), read by an array of 144 solid-state SiPM detectors whose signals are conditioned by an integrated front-end. In addition to state-ofthe- art energy resolution (2.6% at 662 keV) and sub-centimeter spatial resolution in the reconstruction of the photon interaction point projected on the base, this portable unit enables the 2D angular localization of gamma sources on a plane parallel to the detectors array as a function of the reconstructed interaction point distribution, thanks to a decision tree. The classifier is compared with other techniques (k-NN, PCA) and optimized with 1000 splits. It runs in a 32-bit ARM micro-controller for real-time operation with a processing time of 2.75 μs per event, compatible with high gamma-ray counting rate (100 kcps) operation. Despite the absence of a collimator, classification is correct within ±30° for a single photon. Angular resolution of 0.5° and accuracy better than 2° are experimentally demonstrated (along with 0.8W power consumption and 3 kg weight), showing potential for identification of sources in the field.
31 October 2023 in Nanophotonics
In this paper, we combine a Content-Addressable Memory (CAM) encoding scheme previously proposed for analog electronic CAMs (E-CAMs) with optical multiplexing techniques to create two new photonic CAM architectures—wavelength-division multiplexing (WDM) optical ternary CAM (O-TCAM) and time-division multiplexing (TDM) O-TCAM. As an example, we show how these two O-TCAM schemes can be implemented by performing minor modifications in microring-based silicon photonic (SiPh) circuits originally optimized for exascale interconnects. Here, our SiPh O-TCAM designs include not only the actual search engine, but also the transmitter circuits. For the first time, we experimentally demonstrate O-TCAM functionality in SiPh up to and we prove in simulation feasibility for speeds up to 10 Gbps, 10 times faster than typical E-TCAMs at the expense of higher energy consumption per symbol of our O-TCAM Search Engine circuits than the corresponding E-TCAMs. Finally, we identify which hardware and architecture modifications are required to improve the O-CAM’s energy efficiency towards the level of E-CAMs.
25 February 2022 in IEEE Transactions on Radiation and Plasma Medical Sciences
We present a gamma-ray detection module for quantifying the boron neutron capture events that occur in Boron Neutron Capture Therapy (BNCT) and Neutron Capture Enhanced Particle Therapy (NCEPT). The goal of the module is to differentiate between the background prompt gamma peaks and the 478 keV neutron capture photopeak, in order to estimate the dose delivered to the patient. It is a compact module, coupling a large array of 64 SiPMs with a 2"×2" cylindrical LaBr3(Ce+Sr) scintillator crystal (73 ph/keV light yield, 25 ns decay time). The electronic front-end ASIC features low-noise processing of photodetector signals, while SiPMs pixellation and individual readout allow for position sensitivity in the crystal, although position estimation is not the object of this work. The module experimental characterization shows excellent energy resolution (2.7% FWHM at 662 keV), that allows to discriminate the neutron capture photons at 478 keV from the annihilation photons at 511 keV. The module features also an anti-coincidence circuit that provides a mechanism to distinguish and reject scintillation events created within specific temporal windows, thus enhancing the signal-to-background ratio of the spectrometer.
31 August 2022 in Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment
14 November 2024 in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
Structured, or tabular, data is the most common format in data science. While deep learning models have proven formidable in learning from unstructured data such as images or speech, they are less accurate than simpler approaches when learning from tabular data. In contrast, modern tree-based Machine Learning (ML) models shine in extracting relevant information from structured data. An essential requirement in data science is to reduce model inference latency in cases where, for example, models are used in a closed loop with simulation to accelerate scientific discovery. However, the hardware acceleration community has mostly focused on deep neural networks and largely ignored other forms of machine learning. Previous work has described the use of an analog content addressable memory (CAM) component for efficiently mapping random forests. In this work, we develop an analog-digital architecture that implements a novel increased precision analog CAM and a programmable chip for inference of state-of-the-art tree-based ML models, such as XGBoost, CatBoost, and others. Thanks to hardware-aware training, X-TIME reaches state-of-the-art accuracy and 119× higher throughput at 9740× lower latency with > 150× improved energy efficiency compared with a state-of-the-art GPU for models with up to 4096 trees and depth of 8, with a 19W peak power consumption.