Linearly Tunable Optoelectronic Synaptic Transistor for In-Sensor Reservoir Computing

Conventional machine vision systems face challenges, such as data latency and energy inefficiency in cognitive tasks, primarily due to the discrete architecture of their sensing, memory, and processing components. Herein, we propose an in-sensor reservoir computing (RC) system based on yttrium oxide (YOx) electrolyte-gated optoelectronic synaptic transistors (OSTs). The device features a simple structure, low-voltage operation, well-separated photocurrent responses, and nonlinear decay characteristics, enabling direct sensing, memory, and processing of visual signals. Results demonstrate that the in-sensor RC system based on the proposed device achieves a remarkable accuracy rate of 93.42% on the MNIST dataset. Furthermore, the device exhibits various optical plasticity and utilizes its gate-tunable characteristic to perform basic addition and multiplication operations and emulate the photoadaptive behaviors observed in biological visual systems. This work not only demonstrates the potential of optoelectric synaptic transistors for in-sensor neuromorphic vision systems but also presents a novel approach to develop efficient and cost-effective machine vision systems.