Photoelectric Reservoir Computing Based on TiOx Memristor for Analog Signal Processing

Abstract

The bioinspired computing system aims to enhance the ability to handle complex tasks in an efficient, low-cost, and parallel processing as manner of neuron and neural network. Memristors are ideal components for achieving this goal. We have developed a memristor with an Au/TiO_x/ Indium tin oxide (ITO) structure, showing highly sensitive to light stimuli and self-rectifying switching memory. These features enable our memristor with synaptic plasticity such as short-term plasticity (STP), long-term plasticity (LTP), paired-pulse facilitation (PPF), spike-timing-dependent plasticity (STDP) and so on. The photoconductance weight can be precisely regulated through the variety of light pulse parameters including the light intensity, stimuli frequency, pulse number, pule width, suggesting that this TiO_x optoelectronic memristor can execute complex intelligent task by giving different light dosage. We have designed two systems, an electrocardiogram diagnosis and digital recognition, to demonstrate the capability of the memristor that as real physical node to implement reservoir computing, indicating that our memristor has rich intermediate states to efficiently execute the intelligent tasks. This work lays a significant foundation on optoelectronic memristor-based edge computing.