Modular Neural Tile Architecture for Compact Embedded Hardware Spiking Neural Network

Biologically-inspired packet switched Network on Chip (NoC) basedhardware Spiking Neural Network (SNN) architectures have been proposed asan embedded computing platform for classi cation, estimation and control ap-plications. Storage of large synaptic connectivity (SNN topology) information inSNNs require large distributed on-chip memory, which poses serious challenges forcompact hardware implementation of such architectures. Based on the structuredneural organisation observed in human brain, a Modular Neural Networks (MNN)design strategy partitions complex application tasks into smaller subtasks execut-ing on distinct neural network modules, and integrates intermediate outputs inhigher level functions.This paper proposes a hardware Modular Neural Tile (MNT) architecture thatreduces the SNN topology memory requirement of NoC-based hardware SNNs byusing a combination of xed and con gurable synaptic connections. The proposedMNT contains a 16:16 fully-connected feed-forward SNN structure and integratesin a mesh topology NoC communication infrastructure. The SNN topology memoryrequirement is 50% of the monolithic NoC-based hardware SNN implementation.The paper also presents a lookup table based SNN topology memory allocationtechnique, which further increases the memory utilisation eciency. Overall thearea requirement of the architecture is reduced by an average of 66% for practicalSNN application topologies.The paper presents micro-architecture details of the proposed MNT and digitalneuron circuit. The proposed architecture has been validated on a Xilinx Virtex-6 FPGA and synthesised using 65nm low-power CMOS technology. The evolvablecapability of the proposed MNT and its suitability for executing subtasks within aMNN execution architecture is demonstrated by successfully evolving benchmarkSNN application tasks representing classi cation and non-linear control functions.The paper addresses hardware modular SNN design and implementation chal-lenges and contributes to the development of a compact hardware modular SNNarchitecture suitable for embedded applications.