Analysis of Leaky Integrate-and-Fire Neuron Using Al2O3/Si3N4 Gate Insulator Stack for Spiking Neural Network

Abstract

Leaky integrate-and-fire (LIF)-based spiking neural networks (SNNs) are analyzed using a field-effect transistor (FET)-type neuron device with a charge trap insulator stack (Al₂O₃/Si₃N₄). By using both the memory functionality and the poor retention characteristic of the device, we successfully implemented the LIF function. SPICE modeling of the device and LIF circuit demonstrated that the large membrane capacitor in a neuron circuit could be replaced, which promises a smaller area and lower energy consumption (∼0.3 pJ/spike). Based on the measured properties, spiking neural networks are simulated to find the optimal leaky constant while maintaining a low operational voltage.