Paper Abstract and Keywords |
Presentation |
2023-04-11 13:20
[Invited Talk]
Crystalline Oxide Semiconductor-based 3D Bank Memory System for Endpoint Artificial Intelligence with Multiple Neural Networks Facilitating Context Switching and Power Gating Yuto Yakubo, Kazuma Furutani, Kouhei Toyotaka, Haruki Katagiri, Masashi Fujita, Munehiro Kozuma, Yoshinori Ando, Yoshiyuki Kurokawa (SEL), Toru Nakura (Fukuoka Univ.), Shunpei Yamazaki (SEL) ICD2023-10 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
We have achieved a small-area, low-power AI chip that enables inference corresponding to multiple neural networks using a three-dimensional stacked oxide semiconductor memory. This chip has two bank memories which are directly stacked on a flip-flop and an AI accelerator utilizing the feature that an oxide semiconductor FET can be embedded in Si CMOS BEOL. An average power of 25.2 μW was achieved by intermittent operation with two types of inference and power gating. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Oxide semiconductor / Machine learning / Memory / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 123, no. 1, ICD2023-10, pp. 18-23, April 2023. |
Paper # |
ICD2023-10 |
Date of Issue |
2023-04-03 (ICD) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
ICD2023-10 |
|