FPGA Implementation of MLP, 1D-CNN and TTTratio algorithms for Neutron/Gamma-ray Discrimination using Plastic Scintillator

Pulse shape discrimination algorithms, such as Tailto-Total integral ratio (TTT
ratio
) have been commonly integrated on edge devices for online neutron/gamma discrimination using organic scintillators. These algorithms have a number of limitations, especially with plastic scintillators which have low intrinsic discriminating ability. Machine learning (ML) models have recently been explored as a way to improve discriminating performance. Most of these methods are proposed for liquid and stilbene scintillators and do not address the embedded implementation. The aim of this study is to compare the FPGA implementation of TTT
ratio
algorithm, Multi Layer Perceptron Neural Network (MLP), and 1D Convolution Neural Network (1D-CNN) models that are trained for neutron/gamma-ray discrimination using EJ276 plastic scintillator. Therefore, the comparison between the different methods can be done according to the discrimination performance, latency and resource consumption. The objective is to achieve a latency shorter than the signal duration (500 ns) while using minimal resources.