A Photomemristor With Temporal Dynamics for In-Sensor Reservoir Computing
Bingqi Cai, Tianyu Wang, Chen Wang, Qingqing Sun, David Wei Zhang, Lin Chen
Abstract
In-sensor reservoir computing has recently gained considerable attention due to its highly efficient training process and advanced integration of sensing, storage, and processing functionalities. These advancements greatly enhance the machine vision capabilities by reducing data latency and energy overheads. However, the development of a highly efficient and low-cost in-sensor reservoir computing system remains a challenging task, primarily due to the lack of suitable materials and processes. In this letter, we present a simple ITO/NiOx/Au two-terminal photomemristor fabricated using the full physical vapor deposition (PVD) technique at room temperature without further treatment. This photomemristor leverages light-triggered dynamics to map input signals into a high-dimensional space and extract hidden information. As a proof of concept, we demonstrate an in-sensor reservoir computing system based on the photomemristor. Experimental results indicate that the system exhibits an impressive accuracy of 90.88% for image classification task and a low normalized root mean squared error (NRMSE) of 0.0082 for time-series prediction task. This work has complemented the wide spectrum of applications of NiOx in in-sensor neuromorphic computing.