Broadband sensory networks with locally stored responsivities for neuromorphic machine vision

• DOI: 10.1126/sciadv.adi5104
• Publication Date: 2023-09-15T17:58:22Z
• Authors: Guo-Xin Zhang; Zhi-Cheng Zhang; Xu-Dong Chen; Lixing Kang; Yuan Li; Fu-Dong Wang; Lei Shi; Ke Shi; Zhi-Bo Liu; Jian-Guo Tian; Tong-Bu Lu; Jin Zhang
• Keywords: Neuromorphic engineering; Machine Vision

As the most promising candidates for implementation of in-sensor computing, retinomorphic vision sensors can constitute built-in neural networks and directly implement multiply-and-accumulation operations using responsivities as weights. However, existing mainly use a sustained gate bias to maintain responsivity due its volatile nature. Here, we propose an ion-induced localized-field strategy develop with nonvolatile tunable in both positive negative regimes construct broadband reconfigurable sensory network locally stored weights convolutional processing spectral range 400 1800 nanometers. In addition this device in-memory computing benefiting from conductance, complete neuromorphic visual system involving front-end back-end architectures has been constructed, executing supervised unsupervised learning tasks demonstrations. This work paves way development high-speed low-power machine time-critical data-intensive applications.