Engineering Spiking Neurons Using Threshold Switching Devices for High-Efficient Neuromorphic Computing.

Yanting Ding,Du Xiang,Keji Zhou,Chen Mu,Yajun Zhang,Ming Wang,Lingli Cheng,Pei Chen,Yue Yang,Zhe Yuan,Xumeng Zhang,Qi Liu,Zefeng Zhang,Guangjian Wu

doi:10.3389/fnins.2021.786694

Abstract

Inspired by the human brain, the spike-based neuromorphic system has attracted strong research enthusiasm because of the high energy efficiency and powerful computational capability, in which the spiking neurons and plastic synapses are two fundamental building blocks. Recently, two-terminal threshold switching (TS) devices have been regarded as promising candidates for building spiking neurons in hardware. However, how circuit parameters affect the spiking behavior of TS-based neurons is still an open question. Here, based on a leaky integrate-and-fire (LIF) neuron circuit, we systematically study the effect of both the extrinsic and intrinsic factors of NbOx -based TS neurons on their spiking behaviors. The extrinsic influence factors contain input intensities, connected synaptic weights, and parallel capacitances. To illustrate the effect of intrinsic factors, including the threshold voltage, holding voltage, and high/low resistance states of NbOx devices, we propose an empirical model of the fabricated NbOx devices, fitting well with the experimental results. The results indicate that with enhancing the input intensity, the spiking frequency increases first then decreases after reaching a peak value. Except for the connected synaptic weights, all other parameters can modulate the spiking peak frequency under high enough input intensity. Also, the relationship between energy consumption per spike and frequency of the neuron cell is further studied, leading guidance to design neuron circuits in a system to obtain the lowest energy consumption. At last, to demonstrate the practical applications of TS-based neurons, we construct a spiking neural network (SNN) to control the cart-pole using reinforcement learning, obtaining a reward score up to 450. This work provides valuable guidance on building compact LIF neurons based on TS devices and further bolsters the construction of high-efficiency neuromorphic systems.

Highlights

In the big data era, traditional computing architectures are facing the challenge known as the “Von Neumann bottleneck” due to the separated memory and computing units and struggling on high efficiency to process massive data (Ambrogio et al, 2018; Zidan et al, 2018; Sebastian et al, 2020)
They switch to a low-resistance state (LRS) when the applied bias exceeds a threshold voltage (Vth) and subsequently return to a high-resistance state (HRS) as the voltage drops below a hold voltage (Vhold)
To meet the practical application, we studied the relation between spike energy consumption and the frequency of neuron cells

Summary

Introduction

In the big data era, traditional computing architectures are facing the challenge known as the “Von Neumann bottleneck” due to the separated memory and computing units and struggling on high efficiency to process massive data (Ambrogio et al, 2018; Zidan et al, 2018; Sebastian et al, 2020). TS-based neurons combine a simple TS device with a capacitor or resistor, which are equipped with the characteristic of self-sustained oscillation (Gao et al, 2017; Woo et al, 2019; Wang et al, 2020) Such a neuron circuit allows the design of an inductorfree circuit without needing an additional reset circuit, which has the advantages of low power consumption, nanoscale scalability, and high integration intensity. Construction of a complete performance assessment system for TS-based neurons deserves more attention

Methods

Results

Conclusion