Proceedings of MATSUS Spring 2025 Conference (MATSUSSpring25)
DOI: https://doi.org/10.29363/nanoge.matsusspring.2025.260
Publication date: 16th December 2024
Halide perovskite memristors exhibit excellent properties for neuromorphic computing including analog resistive switching, endurance, and low power requirements. However, the reproducibility and stability are limiting factors for the practical application of these devices. Recently, our research group demonstrated that introducing an interfacial buffer layer between the metal contact and the perovskite active layer can improve the stability of the halide perovskite memristors and even modify the activation mechanism [1, 2]. Nevertheless, the mechanisms through which these devices work remain unclear, and a deeper understanding is required to achieve a rational optimization.
In this work, we present the fabrication of highly reproducible memristors using Ag as active electrode, a highly stable perovskite formulation (MAPbBr3), and a buffer layer containing oxidized silver (AgI). These memristors show high reproducibility in fabrication and stability (>104 cycles of pulse trains) [3]. This highly reliable system enabled an in-depth study of the mechanisms operating in perovskite memristors, which provided insights into the nature of the dual volatile and nonvolatile responses. Short-duration pulses at the activation voltage lead to a two-state volatile response due to the formation of an ionic double layer close to the contacts, which returns to the initial state once the bias ceases. In contrast, long-duration pulses lead to a gradual increase in current and a nonvolatile response caused by the appearance of a chemical inductor. The observed multi-state nonvolatile regime is related to the formation of Ag+ conductive filaments and provides suitable conditions for analog computing. Overall, we provide a clear understanding of the nature of these two operating regimes in halide perovskite memristors and show the tools to investigate them in other systems.
The authors acknowledge the funding provided by the project NEUROVISIONM (Generalitat Valenciana, code: MFA/2022/055).