An Optoelectronic SnP2S6 Memristive Neuron for Biologically Inspired Multimodal Neuromorphic Information Processing

Abstract Neuromorphic systems capable of integrating multiple sensory modalities offer a promising route toward energy-efficient information processing that more closely emulates biological neural networks. Here, we demonstrate a multifunctional ITO/SnP2S6/Ti memristive neuron that combines reproducible volatile resistive switching and integrate-and-fire dynamics under electrical-pulse stimulation with wavelength-selective photoresponses and short-term optoelectronic synaptic plasticity. Leveraging these experimentally measured device characteristics, we construct a multimodal neuromorphic recognition framework that integrates volatile memristive neurons with programmable memristive synapses. Inspired by the parallel processing of visual features in the primate visual system, structural and color information are independently encoded and processed through separate electrical and optical pathways before being integrated for classification. Importantly, inference is implemented using experimentally measured neuronal firing statistics and programmed synaptic conductance states, rather than idealized device models, thereby establishing a direct link between device characteristics and system-level functionality. As a proof-of-concept, the framework is evaluated on a fruit-recognition task using a subset of the Fruits-360 image data set, achieving a classification accuracy of 94.9% through multimodal fusion and substantially outperforming recognition based on either modality alone. These results establish SnP2S6 memristive devices as a versatile platform for integrating electrical and optical neuronal functionalities and highlight their potential for hardware-realistic multimodal in-sensor neuromorphic computing.