Reconfigurable Resistive Switching in Halide Perovskite-Based Memristors for Neuromorphic Computing

Artificial intelligence has become an integral part of modern society, increasing the demand for energy-efficient hardware systems that integrate memory and computing. However, traditional von Neumann computing architectures struggle to replicate these characteristics due to the inherent separation between memory and processing units, resulting in latency and massive energy consumption. Neuromorphic computing offers a promising solution by mimicking the parallel processing of biological neural networks, and specifically, memristors emerged as a key component for such platforms due to their simple structure for high-density integration. Inorganic halide perovskites emerged as a promising material for memristor-based optoelectronic synapses due to their mixed electronic/ionic conductivity, low activation energy of halide vacancy, and long carrier diffusion length. Here, we demonstrate a reliable approach to tune the resistive switching characteristics in Ag/CsPbBr₃/PEDOT: PSS/ITO memristors by precisely controlling the ionic migration and filament formation. The digital switching exhibits a high ON/OFF ratio of 10³, stable endurance for 500 cycles, and a long retention time of over 4000s with low SET and RESET voltages of 1.2V and -1.8V, respectively. The analog switching is utilized to mimic essential synaptic functions, including short-term plasticity (STP), long-term plasticity (LTP), paired-pulse facilitation (PPF), and spike rate-dependent plasticity (SRDP) with energy consumption as low as 206 nJ/mm² per spiking event. These results highlight the potential of perovskite-based memristors for both data storage and neuromorphic computing applications, paving the way for energy-efficient computing systems.