P. Tseng, Yu-Hsuan Lin, F. Lee, Tian-Cig Bo, Yung-Chun Li, Ming-Hsiu Lee, K. Hsieh, Keh-Chung Wang, Chih-Yuan Lu
{"title":"In-Memory Approximate Computing Architecture Based on 3D-NAND Flash Memories","authors":"P. Tseng, Yu-Hsuan Lin, F. Lee, Tian-Cig Bo, Yung-Chun Li, Ming-Hsiu Lee, K. Hsieh, Keh-Chung Wang, Chih-Yuan Lu","doi":"10.1109/vlsitechnologyandcir46769.2022.9830405","DOIUrl":null,"url":null,"abstract":"A high performance 3D-NAND-flash based approximate computing architecture is proposed to execute in-memory similarity computation. This approximate-computing chip features fuzzy in-memory search (IMS) function with ultra-high parallelism at full-block scale in just one read cycle. The system architecture from the IMS unit cell/string/array configuration to the novel approximate comparison scheme are discussed in detail. Practical issues including Vt distribution, retention loss, and read disturbance are evaluated. We also introduce a novel IMS group-encoding scheme, which can significantly increase the content density under the same string length. Face recognition with VGGFace2 dataset is demonstrated with high accuracy and good tolerability on reliability degradation.","PeriodicalId":332454,"journal":{"name":"2022 IEEE Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2022-06-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 IEEE Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/vlsitechnologyandcir46769.2022.9830405","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
A high performance 3D-NAND-flash based approximate computing architecture is proposed to execute in-memory similarity computation. This approximate-computing chip features fuzzy in-memory search (IMS) function with ultra-high parallelism at full-block scale in just one read cycle. The system architecture from the IMS unit cell/string/array configuration to the novel approximate comparison scheme are discussed in detail. Practical issues including Vt distribution, retention loss, and read disturbance are evaluated. We also introduce a novel IMS group-encoding scheme, which can significantly increase the content density under the same string length. Face recognition with VGGFace2 dataset is demonstrated with high accuracy and good tolerability on reliability degradation.