OAK

First Demonstration of Innovative 3D AND-Type Fully-Parallel Convolution Block with Ultra-High Area-and Energy-Efficiency

Metadata Downloads
Author(s)
Kim, JangsaengIm, JiseongKo, JonghyunLee, SoochangKwon, DongseokShin, WonjunHwang, JoonKoo, Ryun-HanChoi, WooYoungLee, Jong-Ho
Type
Conference Paper
Citation
2023 International Electron Devices Meeting, IEDM 2023
Issued Date
2023-12-09
Abstract
We propose for the first time a novel 3D AND-type flash memory array for a fully-parallel convolution block (FPCB) designed to realize compute-in-memory technology. By leveraging the structural advantages of 3D AND-type cells, the FPCB achieves area- and energy-efficient convolution and fully-connected operations with full memory utilization. In the convolution operation, the FPCB significantly reduces the number of cells and line resistance by a factor of about 10-4 or less (for the ImageNet dataset). Compared to conventional crossbars, the FPCB reduces ~96% and ~74% of area occupancy and energy consumption, respectively. © 2023 IEEE.
Publisher
Institute of Electrical and Electronics Engineers Inc.
Conference Place
US
San Francisco
URI
https://scholar.gist.ac.kr/handle/local/34040
공개 및 라이선스
  • 공개 구분공개
파일 목록
  • 관련 파일이 존재하지 않습니다.

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.