Driven by the swift advancements of artificial intelligence (AI) and neuromorphic computing technologies, bioinspired electronic devices have emerged as an essential direction for the next-generation intelligent perception and computing architecture. Among them, memristors, as a novel type of two-terminal devices, have become promising candidates for emulating biological synaptic behavior, owing to their tunable resistance states, energy-efficient operation and high integration density. Owing to their excellent optoelectronic characteristics, halide perovskites have recently garnered enormous interest not only in the photovoltaics community but also as insulating layers in two-terminal memristors. In this thesis, a systematic study of the material design, conduction mechanisms and synaptic behavior of perovskite memristors has been conducted, with particular emphasis on the application potential of all-inorganic and organic-inorganic hybrid perovskite memristors in data storage, neuromorphic computing and artificial nociceptive systems. Firstly, this thesis presents an air-stable, all-inorganic RbPbI3 perovskite memristor capable of dual operational modes, exhibiting both non-volatile and volatile resistive switching (RS) by modulating both the electroforming process and the compliance current density. The non-volatile mode is governed by electrochemically induced Ag ion migration, while the volatile mode is primarily governed by halide ion migration driven by the applied electric field across the perovskite film during formation. The device performance is found to be greatly enhanced by introducing a thin MoOX buffer layer at the perovskite/Ag interface. The optimized perovskite memristor not only demonstrates robust bipolar RS behavior but also successfully emulates essential synaptic functionalities. Moreover, the experimental long-term potentiation/depression (LTP/LTD) conductance data are employed to train the neural networks for image classification tasks, achieving classification accuracies of 89.24% (MNIST) and 79.10% (Fashion-MNIST) using the CrossSim simulator. These results document a significant potential of RbPbI3 based perovskite memristors for neuromorphic computing. Secondly, the influence of ion migration on device performance in a typical perovskite memristor is studied. An organic-inorganic hybrid perovskite memristor is developed using Au as the top electrode to eliminate potential Ag-ion diffusion related effects on the device’s RS behavior. The device exhibits excellent threshold switching characteristics and successfully integrates both synaptic functionality and nociceptive behavior within a single physical platform. Based on this, a thermoreceptor is constructed by integrating the perovskite volatile memristor with a thermoelectric module, showcasing a most innovative concept to integrate perovskite devices in artificial sensory systems. Furthermore, the underlying conduction mechanism is identified as interface-type, primarily attributed to ion/vacancy migration within the perovskite layer and the modulation of the Schottky barrier at the perovskite/electrode interface. Lastly, although all-inorganic CsPbBr3 perovskites have attracted considerable attention owing to their excellent stability, the low solubility of cesium bromide in common solvents presents a major challenge for fabricating high-quality, pinhole-free CsPbBr3 films via a conventional one-step solution process for memory device applications. It is demonstrated that the incorporation of a carbohydrazide additive facilitates the formation of uniform, pinhole-free films. The resulting memristor exhibits excellent non-volatile RS performance and reliably emulates various synaptic functionalities. Moreover, associative learning behavior inspired by Pavlov’s classic conditioning experiment is achieved by modulating combined electrical and optical stimuli. Additionally, a valence change mechanism (VCM), driven by the formation and rupture of halide ion/vacancy, is confirmed as the underlying mechanism responsible for the device’s RS behavior. Overall, this thesis correlates device design strategies with material specific conduction mechanisms for perovskite-based memristors. Building upon these findings, tunable dual-mode operation featuring both non-volatile and volatile RS behavior, the integration of synaptic and nociceptive functionalities, and the successful emulation of associative learning have been demonstrated across diverse perovskite material systems. This thesis contributes to establishment of rational design rules for high-performance, multifunctional perovskite memristors and outlines the implementation framework for the development of next-generation brain-inspired computing hardware systems.
Zhiqiang Xie (Thu,) studied this question.