In-Sensor Noise Reduction and Reservoir Computing System Using ZnO Optoelectronic Memristors for Artificial Vision Applications
Liang Wang, Le Zhang, Shuai‐Bin Hua, Anran Chen, Qiuyun Fu, Xin Guo
Abstract
Rapid advancements in artificial intelligence (AI) and the Internet of Things (IoT) demand more efficient data processing than conventional von Neumann architectures offer. In-sensor reservoir computing (RC) addresses this by enabling data processing directly within sensors. Optoelectronic memristors, capable of responding to both electrical and optical inputs, have emerged as a promising solution. We present electronic neurons and opto-synapses made of Pt/Ag/ZnO/Pt/Ti memristors, demonstrating stable threshold switching (with cumulative probability variations of 5.06% for V th ) and neuron functions (such as spike encoding and LIF behavior) under electrical stimuli, as well as light-tunable synaptic behaviors (including PPF and STM). This enables the device to perform image sensing and noise reduction. Moreover, we propose an in-sensor noise reduction and RC system that emulates the human vision system, achieving high-precision classification (99.33%) of noisy images. This system offers cost-effective training and efficient processing of optical stimuli, opening innovative avenues for edge computing and machine vision applications.