基于 IGZO 的尖峰神经网络神经形态系统的仿真与优化

IF 2 3区 工程技术 Q3 ENGINEERING, ELECTRICAL & ELECTRONIC
Junhyeong Park;Yumin Yun;Minji Kim;Soo-Yeon Lee
{"title":"基于 IGZO 的尖峰神经网络神经形态系统的仿真与优化","authors":"Junhyeong Park;Yumin Yun;Minji Kim;Soo-Yeon Lee","doi":"10.1109/JEDS.2024.3373889","DOIUrl":null,"url":null,"abstract":"In this paper, we conducted a simulation of an indium-gallium-zinc oxide (IGZO)-based neuromorphic system and proposed layer-by-layer membrane capacitor (Cmem) optimization for integrate-and-fire (I&F) neuron circuits to minimize the accuracy drop in spiking neural network (SNN). The fabricated synaptic transistor exhibited linear 32 synaptic weights with a large dynamic range \n<inline-formula> <tex-math>$(\\sim 846$ </tex-math></inline-formula>\n), and an n-type-only IGZO I&F neuron circuit was proposed and verified by HSPICE simulation. The network, consisting of three fully connected layers, was evaluated with an offline learning method employing synaptic transistor and I&F circuit models for three datasets: MNIST, Fashion-MNIST, and CIFAR-10. For offline learning, accuracy drop can occur due to information loss caused by overflow or underflow in neurons, which is largely affected by Cmem. To address this problem, we introduced a layer-by-layer \n<inline-formula> <tex-math>${\\mathrm{ C}}_{\\mathrm{ mem}}$ </tex-math></inline-formula>\n optimization method that adjusts appropriate \n<inline-formula> <tex-math>${\\mathrm{ C}}_{\\mathrm{ mem}}$ </tex-math></inline-formula>\n for each layer to minimize the information loss. As a result, high SNN accuracy was achieved for MNIST, Fashion-MNIST, and CIFAR-10 at 98.42%, 89.16%, and 48.06%, respectively. Furthermore, the optimized system showed minimal accuracy degradation under device-to-device variation.","PeriodicalId":13210,"journal":{"name":"IEEE Journal of the Electron Devices Society","volume":null,"pages":null},"PeriodicalIF":2.0000,"publicationDate":"2024-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10461007","citationCount":"0","resultStr":"{\"title\":\"Simulation and Optimization of IGZO-Based Neuromorphic System for Spiking Neural Networks\",\"authors\":\"Junhyeong Park;Yumin Yun;Minji Kim;Soo-Yeon Lee\",\"doi\":\"10.1109/JEDS.2024.3373889\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"In this paper, we conducted a simulation of an indium-gallium-zinc oxide (IGZO)-based neuromorphic system and proposed layer-by-layer membrane capacitor (Cmem) optimization for integrate-and-fire (I&F) neuron circuits to minimize the accuracy drop in spiking neural network (SNN). The fabricated synaptic transistor exhibited linear 32 synaptic weights with a large dynamic range \\n<inline-formula> <tex-math>$(\\\\sim 846$ </tex-math></inline-formula>\\n), and an n-type-only IGZO I&F neuron circuit was proposed and verified by HSPICE simulation. The network, consisting of three fully connected layers, was evaluated with an offline learning method employing synaptic transistor and I&F circuit models for three datasets: MNIST, Fashion-MNIST, and CIFAR-10. For offline learning, accuracy drop can occur due to information loss caused by overflow or underflow in neurons, which is largely affected by Cmem. To address this problem, we introduced a layer-by-layer \\n<inline-formula> <tex-math>${\\\\mathrm{ C}}_{\\\\mathrm{ mem}}$ </tex-math></inline-formula>\\n optimization method that adjusts appropriate \\n<inline-formula> <tex-math>${\\\\mathrm{ C}}_{\\\\mathrm{ mem}}$ </tex-math></inline-formula>\\n for each layer to minimize the information loss. As a result, high SNN accuracy was achieved for MNIST, Fashion-MNIST, and CIFAR-10 at 98.42%, 89.16%, and 48.06%, respectively. Furthermore, the optimized system showed minimal accuracy degradation under device-to-device variation.\",\"PeriodicalId\":13210,\"journal\":{\"name\":\"IEEE Journal of the Electron Devices Society\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":2.0000,\"publicationDate\":\"2024-03-06\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10461007\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"IEEE Journal of the Electron Devices Society\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://ieeexplore.ieee.org/document/10461007/\",\"RegionNum\":3,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"ENGINEERING, ELECTRICAL & ELECTRONIC\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Journal of the Electron Devices Society","FirstCategoryId":"5","ListUrlMain":"https://ieeexplore.ieee.org/document/10461007/","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0

摘要

本文对基于铟镓锌氧化物(IGZO)的神经形态系统进行了仿真,并提出了逐层膜电容(Cmem)优化积分发射(I&F)神经元电路的方法,以最大限度地降低尖峰神经网络(SNN)的精度下降。所制造的突触晶体管表现出 32 个突触权重的线性,且具有较大的动态范围$(\sim 846$ ),同时还提出了一种纯 n 型 IGZO I&F 神经元电路,并通过 HSPICE 仿真进行了验证。该网络由三个全连接层组成,采用离线学习方法,利用突触晶体管和 I&F 电路模型对三个数据集进行了评估:MNIST、Fashion-MNIST 和 CIFAR-10。对于离线学习,神经元的溢出或下溢会造成信息丢失,从而导致准确率下降,而这在很大程度上会受到 Cmem 的影响。为了解决这个问题,我们引入了一种逐层${/mathrm{ C}}_{mathrm{ mem}}$ 优化方法,为每一层调整适当的${/mathrm{ C}}_{mathrm{ mem}}$ 以最小化信息丢失。因此,MNIST、Fashion-MNIST 和 CIFAR-10 的 SNN 准确率分别达到了 98.42%、89.16% 和 48.06%。此外,优化后的系统在设备间变化的情况下显示出最小的准确率下降。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Simulation and Optimization of IGZO-Based Neuromorphic System for Spiking Neural Networks
In this paper, we conducted a simulation of an indium-gallium-zinc oxide (IGZO)-based neuromorphic system and proposed layer-by-layer membrane capacitor (Cmem) optimization for integrate-and-fire (I&F) neuron circuits to minimize the accuracy drop in spiking neural network (SNN). The fabricated synaptic transistor exhibited linear 32 synaptic weights with a large dynamic range $(\sim 846$ ), and an n-type-only IGZO I&F neuron circuit was proposed and verified by HSPICE simulation. The network, consisting of three fully connected layers, was evaluated with an offline learning method employing synaptic transistor and I&F circuit models for three datasets: MNIST, Fashion-MNIST, and CIFAR-10. For offline learning, accuracy drop can occur due to information loss caused by overflow or underflow in neurons, which is largely affected by Cmem. To address this problem, we introduced a layer-by-layer ${\mathrm{ C}}_{\mathrm{ mem}}$ optimization method that adjusts appropriate ${\mathrm{ C}}_{\mathrm{ mem}}$ for each layer to minimize the information loss. As a result, high SNN accuracy was achieved for MNIST, Fashion-MNIST, and CIFAR-10 at 98.42%, 89.16%, and 48.06%, respectively. Furthermore, the optimized system showed minimal accuracy degradation under device-to-device variation.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
IEEE Journal of the Electron Devices Society
IEEE Journal of the Electron Devices Society Biochemistry, Genetics and Molecular Biology-Biotechnology
CiteScore
5.20
自引率
4.30%
发文量
124
审稿时长
9 weeks
期刊介绍: The IEEE Journal of the Electron Devices Society (J-EDS) is an open-access, fully electronic scientific journal publishing papers ranging from fundamental to applied research that are scientifically rigorous and relevant to electron devices. The J-EDS publishes original and significant contributions relating to the theory, modelling, design, performance, and reliability of electron and ion integrated circuit devices and interconnects, involving insulators, metals, organic materials, micro-plasmas, semiconductors, quantum-effect structures, vacuum devices, and emerging materials with applications in bioelectronics, biomedical electronics, computation, communications, displays, microelectromechanics, imaging, micro-actuators, nanodevices, optoelectronics, photovoltaics, power IC''s, and micro-sensors. Tutorial and review papers on these subjects are, also, published. And, occasionally special issues with a collection of papers on particular areas in more depth and breadth are, also, published. J-EDS publishes all papers that are judged to be technically valid and original.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信