Leveraging Dual Resistive Switching in Quasi-2D Perovskite Memristors for Integrated Non-volatile Memory, Synaptic Emulation, and Reservoir Computing.

Abstract

The increasing computational demands of artificial intelligence (AI) algorithms are exceeding the capabilities of conventional computing architectures, creating a strong need for novel materials and paradigms. Memristors that integrate diverse resistive switching (RS) behaviors provide a promising avenue for developing novel computing architectures. In this study, we achieve the coexistence of volatile and nonvolatile RS behaviors in quasi-2D perovskite memristor (Q-2DPM). The Q-2DPM exhibits competitive performance as a nonvolatile memory. Multiple synaptic functions have been successfully simulated on Q-2DPM, such as excitatory postsynaptic currents, paired-pulse facilitation, and long-term potentiation/depression. Furthermore, artificial neural networks using Q-2DPM synapses achieve high accuracy in MNIST image classification tasks. The Q-2DPM's inherent characteristics suitable for reservoir computing are also demonstrated through its application in a pulse-stream-based digital classification experiment, showcasing its impressive performance. The elucidation of the dual RS mechanisms within Q-2DPM provides fresh insights into memristor RS behavior and underscores the potential of achieving diverse computational units through a single device. This work paves the way for the implementation of physical neuromorphic hardware architectures and the advancement of sophisticated computational primitives, offering a significant step toward the next generation of computing technologies.