Optically tunable synaptic transistors based on AlGaN/GaN heterostructure for neuromorphic vision processing
Xiaoqi Li, Huazhen Sun, Mei Ge, Leyang Qian, Xuyang Ge, Xuekun Hong, Weiying Qian, Xiangyang Zhang, Jun‐Ge Liang, Xinyi Shan, Jian Guo, Guofeng Yang
Abstract
Optoelectronic synaptic devices are a promising technology for overcoming the von Neumann bottleneck, meeting the demand from rapidly advancing artificial intelligence for faster, more energy-efficient neuromorphic computing. This study fabricated an optically tunable synaptic transistor based on an AlGaN/GaN heterostructure, which enables the implementation of neuromorphic vision processing. The device exhibits a low dark current in the cutoff region and a high photo-to-dark current ratio of 1.47 × 108, highlighting its excellent photoresponsivity. Under UV illumination, the device demonstrates synaptic behaviors such as excitatory postsynaptic current and paired-pulse facilitation. By tuning the time, power, and number of optical pulses, dynamic transitions from short-term memory to long-term memory are achieved, effectively emulating visual persistent memory. Furthermore, an optically modulated convolutional neural network is implemented, achieving a classification accuracy of 93.86% on the Fashion-MNIST dataset. These results validate the potential of the proposed device for neuromorphic vision processing applications.