Dual-mode α-FA-based perovskite memristors with volatile and nonvolatile switching for neuromorphic computing and handwritten <?A3B2 pi6?>digit recognition

Halide perovskite memristors, known for their ion mobility, have emerged as strong candidates for computational units in next-generation memory and neuromorphic computing systems. Nevertheless, most memristors are limited to operating in a single mode, either resistive switching or threshold switching. In this work, we overcome this limitation by developing dual-mode α-formamidinium lead triiodide (α-FAPbI3) perovskite memristors with switchable volatile/nonvolatile states, enabled by engineered SnO2 electron transport layers (ETLs). Through molecular interface optimization using 3-(N,N′-dimethylmyristylammonio) propanesulfonate (Z14) and 4,4′-(1,10-phenanthroline-3,8-diyl)bis(N,N′-bis(4-methoxyphen-yl)aniline) (PNL), we achieved exceptional device stability. Volatile devices exhibited >500 switching cycles, while nonvolatile devices surpassed 1000 cycles, both maintaining a high on/off ratio (~103). Beyond memory applications, these devices successfully emulated biological functionalities. The volatile mode replicated four key nociceptor characteristics (threshold, relaxation, sensitization, and no adaptation), while the nonvolatile mode demonstrated advanced synaptic plasticity, including paired-pulse facilitation (PPF) and spike-timing-dependent plasticity (STDP). Capitalizing on this dual-mode synergy, we constructed a spiking neural network (SNN) for handwritten digit recognition, achieving a 93% accuracy rate—a significant milestone for perovskite-based neuromorphic systems. This study not only provides a material-level strategy for multifunctional memristor design but also bridges the gap between biological sensing and artificial intelligence, paving the way for adaptive neuromorphic hardware.