Investigation and mitigation of Mott neuronal oscillation fluctuation in spiking neural network
作者机构：School of Integrated CircuitsPeking University Beijing Advanced Innovation Center for Integrated Circuits State Key Laboratory of Information Photonics and Optical CommunicationsBeijing University of Posts and Telecommunications Department of EngineeringUniversity of Cambridge
出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))
年 卷 期：2024年第67卷第2期
页 面：291-301页
核心收录：
学科分类：0808[工学-电气工程] 08[工学] 081104[工学-模式识别与智能系统] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）]
基 金：supported by National Key R&D Program of China (Grant No.2019YFB2205401) National Natural Science Foundation of China (Grant Nos.61834001,62025401,61927901) Beijing Nova Program (Grant No.20220484113) 111 project (Grant No.B10081)
主 题：Mott neuron oscillation fluctuation variation-aware Mott neuronal model conversion-based spiking neural network activation function boundary
摘 要：Mott devices, featuring low hardware cost and high energy efficiency, have been demonstrated as a key oscillatory element in artificial neurons to enable spiking neural networks(SNNs) such as conversionbased SNNs(CSNNs). However, there will be inevitably non-ideal fluctuation in the oscillation behavior,causing the accuracy degradation of networks. In this paper, we investigate the Mott neuronal oscillation fluctuation(NOF) through experiments and modeling. The results show that the NOF phenomenon conforms to Gaussian distribution and originates from thermal fluctuation induced switching voltage variations. We construct a two-layer CSNN for image recognition tasks to study the NOF effect and propose the activation function boundary(AFB) method to strengthen the stability of the network. The results indicate that AFB can improve the accuracy of CSNN by up to 15.5% by tightening output distribution.
同方期刊数据库