Vacancy-ordered double-perovskite-based memristors for image processing and pattern recognition

Abstract

High-performance memristors have emerged as efficient hardware for integrating noisy image recognition and noise reduction. Herein, we report a fast-switching memristor featuring tens of nanoseconds switching time fabricated using a vacancy-ordered double perovskite, Cs₂TiBr₆ nanocrystals. The spatially ordered vacancies in the double perovskite facilitate the predictable formation and rupture of conductive filaments, which are explored through a comprehensive simulation using the finite element analysis physical model. These unique microscopic features suppress random conducting filament growth and enhance bromine vacancy diffusion, boosting memristor switching speed. A further study of synapse-like behaviors reveals that Cs₂TiBr₆-based memristors exhibit high robustness and reproducibility. We further developed the crossbar-array memristors as artificial neural networks for image denoising and classification, achieving a 10% increase in recognition accuracy for pre-denoised images over non-denoised samples. Our work highlights the potential of intrinsic vacancy-ordered memristive materials for advancing efficient, real-time, robust visual recognition.