Reconfigurable optoelectronic memristive architecture based on doped nanowire array for in-memory parallel perception and computation

Abstract Advanced hardware functional integration for emergent computing paradigms facilitates the potential optimization of computational redundancy in artificial intelligence. However, the device design for parallel perception and in–memory computing remains challenging. To develop a state–of–the–art integrated functional memory, this study demonstrates a reconfigurable optoelectronic memristive architecture (ROMA) based on a doped nanowire array for in situ parallel perception and in–sensor computation. The memristor based on In2S3–XAsX exhibits favorable photoconductive retention and reconfigurable optoelectronic modulation which originates from vacancy engineering induced by doping modulation. The memristive performance of In2S3–XAsX can be tuned by controlling the doping dose. A monolithically integrated array demonstrates an improvement of the discriminative state by more than two orders of magnitude with a double sampling of the output signal, and achieves recognition and encoding of 12–bit binary optical signals on a single column. The nanowire memristive architecture provides an efficient hardware foundation for highly–parallel and efficiently distributed computational paradigms.