版权所有:内蒙古大学图书馆 技术提供:维普资讯• 智图
内蒙古自治区呼和浩特市赛罕区大学西街235号 邮编: 010021
作者机构:Tsinghua Univ Sch Integrated Circuits Beijing Adv Innovat Ctr Integrated Circuits Beijing 100084 Peoples R China Tsinghua Univ Beijing Natl Res Ctr Informat Sci & Technol Beijing 100084 Peoples R China
出 版 物:《IEEE ELECTRON DEVICE LETTERS》 (IEEE Electron Device Lett)
年 卷 期:2025年第46卷第2期
页 面:199-202页
核心收录:
基 金:Scientific and Technological Innovation (STI) 2030-Major Project [.2022ZD0210200] National Natural Science Foundation of China [92264201, 62304122, 52250418]
主 题:Programming Ions In-memory computing Neural networks Materials reliability Electrodes Computer architecture Common Information Model (computing) Transmission electron microscopy Tin Resistive random-access memory (RRAM) relaxation reverse forming computing-in-memory (CIM)
摘 要:This study presents a novel reverse forming (RevF) operation scheme to enhance the multi-bit programming reliability of analog resistive random-access memory (RRAM), which is essential for advancing computing-in-memory (CIM) technology. The proposed RevF method uses a series of progressively intensified RESET operations following the initial forming process. The RevF scheme significantly strengthens the unstable conductive filaments (CF) at the bottom of the HfO2 resistive switching layer (RSL) and reduces excessive oxygen ions in the TaOx thermal enhanced layer (TEL). Array-level experimental evaluations on a fully integrated 128Kb RRAM chip showcase a 3-fold reduction in the post-programming conductance instability (i.e., relaxation effect), as quantified by the relative deviation (RD), and a 10(6)-fold enhancement in the retention time. To further verify the effectiveness and advantage of the RevF scheme, a classical denoising diffusion probabilistic model (DDPM) is implemented, yielding significantly enhanced image quality compared to the conventional operation scheme.