Application of a Novel Machine Learning Approach to SiPM-Based Neutron/Gamma Detection and Discrimination

Silicon photomultipliers (SiPMs) have become a common component of radiation detection systems and have shown promise in distinguishing neutrons and gammas when coupled with detectors sensitive to both particle types. This work investigates the use of a novel machine learning (ML) approach to aid in this discrimination. The proposed ML algorithm performs a regression on a conventionally calculated pulse shape parameter (PSP), producing a more representative PSP while allowing for direct comparison to the traditional pulse shape discrimination (PSD) technique. This work also investigates how the proposed ML-based PSD approach can benefit different scintillator and light-sensor combinations with varying levels of traditional PSD performance, especially those based on SiPMs. A preliminary implementation of the regression method with a Hamamatsu SiPM and stilbene scintillator combination on a data set of mixed gamma and neutron pulses from ²⁵²Cf resulted in an increase in the figure of merit of approximately 100% compared to the traditional PSD technique.