Multiwavelength plasmonic optoelectronic memristor for color image classification via physical reservoir

Abstract

Optoelectronic memristors have tremendous potential for constructing highly efficient neuromorphic visual systems because of their integration of sensing, storage, and computing. Here, a multiwavelength-modulated plasmonic optoelectronic memristor is developed using Au/WO3 nanocomposite. Benefitting from the size-dependent localized surface plasmon resonance absorption property, the distinction of the R (680 nm), G (550 nm), and B (450 nm) light response can be utilized in a single device. Based on this, multiwavelength synaptic plasticity is mimicked, including excitatory postsynaptic current, paired-pulse facilitation, and spike-number-dependent plasticity. Furthermore, 4-bit reservoirs with RGB inputs are implemented for constructing an RGB reservoir computing system. Consequently, considerable accuracy (98.3%) for classification of color images is implanted via the RGB input mode, demonstrating the advantage of color information for complicated neuromorphic tasks. This work demonstrates the potential of the plasmonic optoelectronic memristors for high-performance neuromorphic vision systems.