2019 Fiscal Year Annual Research Report

撮像素子とアナログCNN回路の集積化により画像認識のエネルギーを1/1000倍に

Research Project

Project/Area Number	19H02188
Research Institution	The University of Tokyo
Principal Investigator	高宮真東京大学, 生産技術研究所, 教授 (20419261)
Project Period (FY)	2019-04-01 – 2022-03-31
Keywords	画像認識 / 畳み込みニューラルネットワーク / 撮像素子 / エネルギー
Outline of Annual Research Achievements	Recently, convolutional neural network (CNN) is actively used for image recognition while still has the bottleneck of energy efficiency due to the big amount of access data among the interfaces of imager, the memory and GPU. So far, several approaches including bit-precision-reduction of filter weights, networks-pruning and near/in-memory computation show the possibilities of low-energy consumption. In this work, we propose a near-pixel binary convolution engine to approach high energy-efficiency in convolutional layers by applying binary-input binary-weight CNN operations near pixels instead of long transmission signal lines.
Current Status of Research Progress	Current Status of Research Progress 2: Research has progressed on the whole more than it was originally planned. Reason 当初計画通りに研究を実行した。
Strategy for Future Research Activity	LSIチップ試作と評価を行う。