Integrated bionic human retina process and in-sensor rc system based on 2d retinomorphic memristor array

• Main Authors: Gong, Yue, Xing, Xuechao, Wang, Xingli, Duan, Ruihuan, Han, Su-Ting, Tay, Beng Kang
• Other Authors: Interdisciplinary Graduate School (IGS)
• Format: Article
• Language: English
• Published: 2024
• Subjects: Engineering; Human retina; In-sensor reservoir computing
• Online Access: https://hdl.handle.net/10356/180141
• Institution: Nanyang Technological University

The rapid development of machine vision has put forward higher requirements for image perception and visual learning systems. However, limited by the von Neumann bottleneck, traditional artificial vision systems rely on complex optical and circuit designs, and the separate distribution of sensors, memory, and computing modules results in high energy consumption and high latency. Memristor-based reservoir computing (RC) system with integrated sensing and memory functions provides a solution to effectively improve the computational efficiency of artificial vision networks. Here, a complete artificial visual recognition system based on in-sensor RC is developed by combining the excellent optoelectronic synaptic properties of 2D WS2 memristor array. Through effectively regulating the conductance of memristors by light or electrical stimulation, image information is converted into high-dimensional information in the reservoir layer, and the classification can be completed by simple training, significantly reducing the training complexity and cost consumption. With this system, an effective recognition of hand-written numbers with an accuracy of 88.3% is demonstrated, and a 100% recognition of traffic signals of different colors is achieved. The proposed in-sensor RC system advances the development and application of artificial vision recognition, paving the way for efficient machine learning and neuromorphic vision systems.