Grain-Size-Controlled Resistive Switching Memories Enabling Domain-Specific Functionality for Real-Time Video Signal Processing.

Conventional neuromorphic platforms often rely on heterogeneous device integration to deliver multiple functionalities, a strategy that increases system complexity and hinders scalability. Here, we present a platform based on a grain-size-controlled resistive switching memory (RSM) array incorporating a Sn-halide perovskite thin film. A photo-thermochemical process produces laterally varying grain sizes and therefore spatially graded grain-boundary densities across the array. This intentionally introduced structural heterogeneity produces domain-dependent volatile threshold-switching behavior and short-term neural dynamics, since local grain-boundary density controls conductive-filament formation and ionic transport. As a result, time-dependent processing primitives-nonlinear conductance modulation, relaxation dynamics, integrate-and-fire responses, and signal separability-arise intrinsically and differentially across spatial domains, with their characteristics finely tunable via grain-boundary density. We validate the approach in a real-time spatiotemporal signal-processing system for autonomous-driving tasks, illustrating that grain-boundary engineering provides a scalable, fabrication-friendly route to embed diverse temporal functions within a single RSM array.