Wenshuo Yue, Kai Wu, Zhiyuan Li, Juchen Zhou, Zeyu Wang, Teng Zhang, Yuxiang Yang, Lintao Ye, Yongqin Wu, Weihai Bu, Shaozhi Wang, Xiaodong He, Xiaobing Yan, Yaoyu Tao, Bonan Yan, Ru Huang, Yuchao Yang
{"title":"Physical unclonable in-memory computing for simultaneous protecting private data and deep learning models","authors":"Wenshuo Yue, Kai Wu, Zhiyuan Li, Juchen Zhou, Zeyu Wang, Teng Zhang, Yuxiang Yang, Lintao Ye, Yongqin Wu, Weihai Bu, Shaozhi Wang, Xiaodong He, Xiaobing Yan, Yaoyu Tao, Bonan Yan, Ru Huang, Yuchao Yang","doi":"10.1038/s41467-025-56412-w","DOIUrl":null,"url":null,"abstract":"<p>Compute-in-memory based on resistive random-access memory has emerged as a promising technology for accelerating neural networks on edge devices. It can reduce frequent data transfers and improve energy efficiency. However, the nonvolatile nature of resistive memory raises concerns that stored weights can be easily extracted during computation. To address this challenge, we propose RePACK, a threefold data protection scheme that safeguards neural network input, weight, and structural information. It utilizes a bipartite-sort coding scheme to store data with a fully on-chip physical unclonable function. Experimental results demonstrate the effectiveness of increasing enumeration complexity to 5.77 × 10<sup>75</sup> for a 128-column compute-in-memory core. We further implement and evaluate a RePACK computing system on a 40 nm resistive memory compute-in-memory chip. This work represents a step towards developing safe, robust, and efficient edge neural network accelerators. It potentially serves as the hardware infrastructure for edge devices in federated learning or other systems.</p>","PeriodicalId":19066,"journal":{"name":"Nature Communications","volume":"13 1","pages":""},"PeriodicalIF":14.7000,"publicationDate":"2025-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Nature Communications","FirstCategoryId":"103","ListUrlMain":"https://doi.org/10.1038/s41467-025-56412-w","RegionNum":1,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
Physical unclonable in-memory computing for simultaneous protecting private data and deep learning models
Compute-in-memory based on resistive random-access memory has emerged as a promising technology for accelerating neural networks on edge devices. It can reduce frequent data transfers and improve energy efficiency. However, the nonvolatile nature of resistive memory raises concerns that stored weights can be easily extracted during computation. To address this challenge, we propose RePACK, a threefold data protection scheme that safeguards neural network input, weight, and structural information. It utilizes a bipartite-sort coding scheme to store data with a fully on-chip physical unclonable function. Experimental results demonstrate the effectiveness of increasing enumeration complexity to 5.77 × 1075 for a 128-column compute-in-memory core. We further implement and evaluate a RePACK computing system on a 40 nm resistive memory compute-in-memory chip. This work represents a step towards developing safe, robust, and efficient edge neural network accelerators. It potentially serves as the hardware infrastructure for edge devices in federated learning or other systems.
期刊介绍:
Nature Communications, an open-access journal, publishes high-quality research spanning all areas of the natural sciences. Papers featured in the journal showcase significant advances relevant to specialists in each respective field. With a 2-year impact factor of 16.6 (2022) and a median time of 8 days from submission to the first editorial decision, Nature Communications is committed to rapid dissemination of research findings. As a multidisciplinary journal, it welcomes contributions from biological, health, physical, chemical, Earth, social, mathematical, applied, and engineering sciences, aiming to highlight important breakthroughs within each domain.