No Cover Image

Journal article 590 views 32 downloads

Hardware Spiking Neural Networks with Pair-Based STDP Using Stochastic Computing

Junxiu Liu, Yanhu Wang, Yuling Luo Orcid Logo, Shunsheng Zhang, Dong Jiang, Yifan Hua, Sheng Qin, Scott Yang Orcid Logo

Neural Processing Letters, Volume: 55, Issue: 6, Pages: 7155 - 7173

Swansea University Author: Scott Yang Orcid Logo

Check full text

DOI (Published version): 10.1007/s11063-023-11255-8

Abstract

Spiking Neural Networks (SNNs) can closely mimic the biological neural network systems. Recently, the SNNs have been developed in hardware circuits to emulate the time encoding and information-processing aspects of the human brain in real-time. However, the hardware SNN systems are suffering from la...

Full description

Published in: Neural Processing Letters
ISSN: 1370-4621 1573-773X
Published: Springer Science and Business Media LLC 2023
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa63218
Tags: Add Tag
No Tags, Be the first to tag this record!
Abstract: Spiking Neural Networks (SNNs) can closely mimic the biological neural network systems. Recently, the SNNs have been developed in hardware circuits to emulate the time encoding and information-processing aspects of the human brain in real-time. However, the hardware SNN systems are suffering from large hardware resource consumption due to the high complexity of computational units. In this paper, a novel hardware SNN system based on stochastic computing is proposed to address this problem. Pair-based spiking-timing-dependent plasticity, coupled with integrate-and-fire neurons are employed to design the SNN. Stochastic computing can simplify the computational components of multipliers, adders, and subtractors in conventional hardware SNNs, hence reduce the hardware resource cost. Experimental results show that compared with the state-of-the-art approaches the proposed SNN system reduces the resource consumption by 58.0% (especially registers by ≥ 65.6%). In the meantime, the maximum normalized root mean square error between the proposed hardware and others is only 0.0097, which can maintain the behaviours of SNN. This work provides a beneficial alternative to the large-scale hardware SNN implementations.
Keywords: Spiking neural networks, Pair-based spiking-timing-dependent plasticity, integrate-and-fire neurons, stochastic computing
College: Faculty of Science and Engineering
Funders: This research is supported by the National Natural Science Foundation of China under Grant 61976063, the Guangxi Natural Science Foundation under Grant 2022GXNSFFA035028, research fund of Guangxi Normal University under Grant 2021JC006, the AI+Education research project of Guangxi Humanities Society Science Development Research Center under Grant ZXZJ202205.
Issue: 6
Start Page: 7155
End Page: 7173

Similar Items