Study on battery-like synapse device with high linearity and wide dynamic range for a high efficient neuromorphic computing

Abstract

With the recent high demand for "big data processing", bio-inspired synapse devices based on the resistive switching, which mimics the nervous system of the brain, have attracted attention. To emulate the brain's synaptic plasticity, various mechanisms of synapse devices have been studied, but the factors such as non-linearity, asymmetry, and low dynamic range degrade critically the performance of neuromorphic computing. In this study, we proposed a three-terminal architecture of battery-like organic synapse (HBOS) device utilizing acidic hydrogel electrolyte PVA with high linearity (<2.54), wide dynamic range (>450), and analog resistance change characteristics for the application of highly efficient neuromorphic computing with artificial neural network based on a battery mechanism. Based on the understanding of electrical properties and qualitative analysis, we suggested a programming strategy that improves linear (<1) and symmetric synapse characteristics with less energy consumption. Furthermore, with artificial neural network simulations, we validated the suggested technique, which exhibits a value of MNIST maximum recognition rate of real-case (92.68%) highly closed to the ideal case (93.7%).