An efficient hardware architecture for multilayer spiking neural networks

Abstract

Spiking Neural Network (SNN) is the most recent computational model that can emulate the behaviors of biological neuron system. This paper highlights and discusses an efficient hardware architecture for the hardware SNNs, which includes a layer-level tile architecture (LTA) for the neurons and synapses, and a novel routing architecture (NRA) for the interconnections between the neuron nodes. In addition, a visualization performance monitoring platform is designed, which is used as functional verification and performance monitoring for the SNN hardware system. Experimental results demonstrate that the proposed architecture is feasible and capable of scaling to large hardware multilayer SNNs.