Pulse Shape Discrimination of Neutrons and Gamma Rays Using Neuromorphic Models

This study introduces two brain-inspired neural networks—the random-coupled neural network (RCNN) and the continuous-coupled neural network (CCNN)—for pulse shape discrimination (PSD). Leveraging their unique biological interpretability and inherent temporal structure, RCNN and CCNN can capture the dynamic characteristics of pulse shapes through neural activity, enabling effective PSD without a formal training phase. This approach significantly reduces computational resource consumption compared to conventional second-generation neural networks. We evaluated the models using multiple sub-datasets derived from a time-of-flight (TOF) dataset of a PuBe source, where the dataset was decomposed into several sub-datasets with different energy thresholds. These energy thresholds were determined based on TOF measurements, where neutron energy can be calculated from their velocity, which is derived from the time it takes for neutrons to travel from the source to the detector. Experimental results demonstrate that both RCNN and CCNN achieve high discrimination efficacy and accuracy across all energy thresholds, outperforming traditional discrimination methods. Specifically, RCNN achieved the highest discrimination accuracy of 95.29% at a 2-MeV energy threshold, while CCNN excelled at 3-, 4-, and 5-MeV thresholds with accuracies of 95.87%, 95.96%, and 94.00%, respectively. Both models consistently outperformed traditional methods such as zero-crossing (28.80%–36.63%), charge comparison (88.94%–94.24%), frequency gradient analysis (87.21%–91.62%), and simplified digital charge collection (92.80%–94.67%). This paper also discusses the future development trajectories and limiting factors of brain-like neural networks in the context of PSD. The dataset and the RCNN/CCNN-based discrimination algorithms are publicly available on GitHub at https://github.com/HaoranLiu507/NeuroPulseDiscriminator.