Guihai Yu,Zheng Zhou,Ruiqi Chen,Jiaqi Li,Peng Huang,Jinfeng Kang,Xiaoyan Liu
出处
期刊:IEEE Electron Device Letters [Institute of Electrical and Electronics Engineers] 日期:2023-04-01卷期号:44 (4): 670-673被引量:2
标识
DOI:10.1109/led.2023.3248076
摘要
Two type CMOS-compatible complementary phototransistors (CPTs) with NMOS or PMOS transistor are proposed in this work for in-sensor vector-matrix multiplication (VMM). The CPT is fabricated by the 22nm FDSOI process technology and composed of a MOSFET with a p-doped well under the buried oxide (BOX). The well serves for photoelectron generation and affects the on-state current of the MOSFET by the photoelectrons. The bidirectional and tunable photoresponsivity characteristics as well as the optical-electrical signed multiplication operation (average error < 2.5%) are experimentally demonstrated in the pair of CPTs. The in-sensor VMM is verified by typical artificial neural networks using the CPT array as the image-input layer. Simulated results show slight inference accuracy losses (< 1%) in LeNet3 for MNIST and VGG11/VGG16 for CIFAR-10 recognition.