Fully Phase-Change Artificial Neural Networks Based on Phase-Change Memory With Reconfigurable Activation Functions

IF 4.6 2区工程技术 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC

IEEE Electron Device Letters Pub Date : 2026-10-01 Epub Date: 2026-08-14 DOI:10.1109/LED.2026.3722897

Ruizhe Zhao;Xinbao Chen;Hanyang Chen;Hao Shu;Wenzhuo Xu;Shuliang Zhang;Xin Li;Tao Wang;Hao Tong;Xiangshui Miao

{"title":"Fully Phase-Change Artificial Neural Networks Based on Phase-Change Memory With Reconfigurable Activation Functions","authors":"Ruizhe Zhao;Xinbao Chen;Hanyang Chen;Hao Shu;Wenzhuo Xu;Shuliang Zhang;Xin Li;Tao Wang;Hao Tong;Xiangshui Miao","doi":"10.1109/LED.2026.3722897","DOIUrl":null,"url":null,"abstract":"We demonstrate an all-phase-change artificial neural network in which phase-change memory (PCM) devices simultaneously realize synaptic plasticity and reconfigurable activation-function artificial neurons. Precise control of the voltage-amplitude step size during the SET and RESET processes enables gradual conductance modulation, thereby realizing synaptic long-term potentiation (LTP) and long-term depression (LTD). More importantly, threshold switching behavior induced by larger pulse widths and larger voltage steps allows the realization of reconfigurable activation functions, including ReLU, sigmoid, and softplus. System-level simulations on the MNIST dataset demonstrate recognition accuracy exceeding 98%, comparable to those achieved using ideal software-defined activation functions, thereby validating the robustness and reliability of the all-PCM neural network. These results highlight the strong potential of PCM devices for highly integrated and multifunctional neuromorphic computing systems.","PeriodicalId":13198,"journal":{"name":"IEEE Electron Device Letters","volume":"47 10","pages":"2043-2046"},"PeriodicalIF":4.6000,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Electron Device Letters","FirstCategoryId":"5","ListUrlMain":"https://ieeexplore.ieee.org/document/11655861/","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/8/14 0:00:00","PubModel":"Epub","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}

引用次数: 0

Abstract

We demonstrate an all-phase-change artificial neural network in which phase-change memory (PCM) devices simultaneously realize synaptic plasticity and reconfigurable activation-function artificial neurons. Precise control of the voltage-amplitude step size during the SET and RESET processes enables gradual conductance modulation, thereby realizing synaptic long-term potentiation (LTP) and long-term depression (LTD). More importantly, threshold switching behavior induced by larger pulse widths and larger voltage steps allows the realization of reconfigurable activation functions, including ReLU, sigmoid, and softplus. System-level simulations on the MNIST dataset demonstrate recognition accuracy exceeding 98%, comparable to those achieved using ideal software-defined activation functions, thereby validating the robustness and reliability of the all-PCM neural network. These results highlight the strong potential of PCM devices for highly integrated and multifunctional neuromorphic computing systems.

查看原文本刊更多论文

基于可重构激活函数相变记忆的全相变人工神经网络

我们展示了一个全相变人工神经网络，其中相变记忆（PCM）器件同时实现突触可塑性和可重构激活功能的人工神经元。在SET和RESET过程中精确控制电压幅值步长可以实现逐渐的电导调制，从而实现突触长期增强（LTP）和长期抑制（LTD）。更重要的是，更大的脉冲宽度和更大的电压步长引起的阈值开关行为允许实现可重构的激活功能，包括ReLU， sigmoid和softplus。在MNIST数据集上的系统级仿真表明，识别精度超过98%，与使用理想的软件定义激活函数所达到的精度相当，从而验证了全pcm神经网络的鲁棒性和可靠性。这些结果突出了PCM设备在高度集成和多功能神经形态计算系统中的强大潜力。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

IEEE Electron Device Letters 工程技术-工程：电子与电气

CiteScore

8.20

自引率

10.20%

发文量

551

审稿时长

1.4 months

期刊介绍： IEEE Electron Device Letters publishes original and significant contributions relating to the theory, modeling, 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, nanoelectronics, optoelectronics, photovoltaics, power ICs and micro-sensors.