A multifunctional memristor for neuromorphic vision and secure voice encryption.

Abstract

Neuromorphic vision and secure voice encryption are crucial for next-generation intelligent systems, enabling efficient visual perception and confidential acoustic information processing beyond conventional computing architectures. However, the integration of reliable resistive switching, synaptic plasticity, neural-network inference, logic computing, and secure signal processing in a memristive device remains challenging. In this work, an Ag/CeOx/FTO memristor was fabricated by magnetron sputtering and developed as a multifunctional hardware platform for neuromorphic visual computing and secure voice encryption. The device exhibits stable bipolar resistive switching, excellent endurance, reliable retention, 20 distinguishable conductance states, and diverse synaptic functions, including excitatory postsynaptic current, paired-pulse facilitation, and long-term potentiation/depression. Electrical analysis and theoretical fitting reveal that the switching behavior originates from the synergistic modulation of Ag-ions migration and oxygen-vacancy dynamics. Benefiting from controllable conductance evolution, the device successfully emulates leaky integrate-and-fire neuronal activity, the Ebbinghaus forgetting law, and brain-inspired forgetting processes, while a conductance-mapped convolutional neural network achieves 96.07% accuracy in PathMNIST histopathological image classification. Moreover, XOR, XNOR, half-adder, and full-adder logic operations are implemented to realize key-sequence-dependent voice encryption/decryption, highlighting the broad potential of memristors for low-power neuromorphic vision, intelligent medical diagnosis, logic computing, and secure acoustic information processing.