In-Sensor Reservoir\nComputing Based on Self-Rectifying\nTiO<sub><i>x</i></sub> Photosynapse for Image Recognition\nand Speech Signal Processing

A homojunction memristor with a double layer of resistive\nfunction\nwas prepared by the sol–gel method. Electrical measurements\nshow that the F-doped SnO<sub>2</sub> (FTO)/TiO<sub><i>x</i></sub>/Au memristor exhibits typical volatility, endowing the device\nwith excellent synaptic properties. By analyzing the voltammetry characteristics\nof the FTO/TiO<sub><i>x</i></sub>/Au memristor, it is suggested\nthat the titanium oxide homomode memristor exhibits self-rectifying\nand analog switching characteristics, which can provide inherent advantages\nfor large-scale array integration for high-efficiency bioinspired\ncomputing as well as in-sensing artificial vision computing. For its\nelectric switchings, this photoelectronic memristor holds a low operating\npower consumption, dual-pulse dissimilation learning rules that can\nfaithfully simulate the multifunction of synapses, and learning–forgetting–relearning\nrules. In terms of optical properties, our memristor exhibits excellent\npositive photoconductance memory effect (PPM), and the conductance\nstates are sensitive to the light intensity and frequency, significantly\npresenting synaptic habituation characteristics that are fully controlled.\nMore importantly, integrating the light and electric stimuli control,\nthe conductance weight can be highly linear update, largely accelerating\ntraining processing for the reservoir computing (RC). Recognition\nof image and speech signal processing as the artificial intelligence\ntask is demonstrated using our photoelectronic TiO<sub><i>x</i></sub> memristor, showing over 90% accuracy after several epoch training.\nThis in-sensor reservoir computing based on self-rectifying TiO<sub><i>x</i></sub> photosynapse gives a brand-new horizon on\nneuromorphic vision computing in future edge computing.