Low Cost Interconnected Architecture for the Hardware Spiking Neural Networks.

Yuling Luo,Xuemei Ding,Yi Cao,Jim Harkin,Junxiu Liu,Liam Mcdaid,Lei Wan

doi:10.3389/fnins.2018.00857

Abstract

A novel low cost interconnected architecture (LCIA) is proposed in this paper, which is an efficient solution for the neuron interconnections for the hardware spiking neural networks (SNNs). It is based on an all-to-all connection that takes each paired input and output nodes of multi-layer SNNs as the source and destination of connections. The aim is to maintain an efficient routing performance under low hardware overhead. A Networks-on-Chip (NoC) router is proposed as the fundamental component of the LCIA, where an effective scheduler is designed to address the traffic challenge due to irregular spikes. The router can find requests rapidly, make the arbitration decision promptly, and provide equal services to different network traffic requests. Experimental results show that the LCIA can manage the intercommunication of the multi-layer neural networks efficiently and have a low hardware overhead which can maintain the scalability of hardware SNNs.

Highlights

The current understanding from neuroscience research is that the mammalian brain is composed of dense and complex interconnected neurons and exhibits many surprising properties, e.g., pattern recognition, decision making, and so on (Cios and Shields, 1997)
The aim of this paper is to propose the low cost interconnection architecture for the multi-layer spiking neural network (SNN), where the ENAs (Wan et al, 2016) are used as an example for neuron nodes
It has 16 × 2 array of low cost interconnection architecture (LCIA)-based routers as shown by Figure 8B where each router is connected to all the nodes in the previous layer and the local spike event generator (SG), e.g., Figure 8A shows that 16 SGs are attached to the input ports of an router R (Gerstner and Kistler, 2002; Schuman et al, 2017) and one output port is connected to the SC14

Summary

Introduction

The current understanding from neuroscience research is that the mammalian brain is composed of dense and complex interconnected neurons and exhibits many surprising properties, e.g., pattern recognition, decision making, and so on (Cios and Shields, 1997). When the post-synaptic membrane potential of a neuron exceeds a firing threshold value, it fires and generates an output spike to the connected synapses/neurons This leads to a strong computing capability of SNN and the SNN is widely used to solve problems in various fields, e.g., forecasting (Park et al, 1999; Kulkarni and Simon, 2012), image processing (Perrinet, 2008; Charleston-Villalobos et al, 2011), retinal coding (Rast et al, 2008), multi-view pattern recognition (Wu et al, 2008; Wysoski et al, 2008), and so on. These applications generally require an SNN system containing a large number of neurons for the information processing

Objectives

Methods

Results

Conclusion