A spatiotemporal-specific artificial neuron based on In2Se3 ferroelectric memristor for adaptable information processing

Abstract

Neuromorphic computing provides a remarkably efficient and adaptable alternative to traditional computing architectures by embodying the impressive power efficiency and parallel processing capabilities of the human brain. However, the prevailing focus on integrate-and-fire mode in current artificial neurons fails to fully acknowledge the nuanced multifunctionality and adaptive characteristics, especially the temporally variable operating modes and spatial heterogeneity present in natural neurons. Here we report a spatiotemporal-specific artificial neuron implemented with a ferroelectric planar memristor, by engineering the inherent in-plane ferroelectricity of α-In2Se3 and the extensive regulation capability of the co-planar multi-electrodes. With enhanced information processing capabilities, the artificial neuron facilitates adjustable reservoir computing and reconfigurable 16 types of logic-gate operations, ultimately achieving precise speech recognition with an accuracy approaching 100%. Our work clearly demonstrates the benefits of spatiotemporal specificity in artificial neurons, and contributes to the advancement of more realistic neuromorphic computing systems. (Figure presented.).