On-Chip infrared image denoising using silicon-based memristors

Memristors provide a compelling platform for integrating sensing, memory, and computation at the device level. Here, we present a silicon-compatible Ag/native-SiOx/p++-Si memristive device capable of performing near-sensor infrared image denoising and nonvolatile storage through its intrinsic resistive switching dynamics. By mapping pixel intensities to voltage stimuli, the device selectively suppresses low-amplitude noise components while preserving meaningful signal features, leveraging its analog conductance modulation and switching threshold characteristics. Simultaneously, its nonvolatile memory property enables the denoised image to be stored directly during the processing operation. The denoising and storage functionality is experimentally validated using simulated and real infrared images, including astronomical data, demonstrating effective noise suppression and signal retention under low signal-to-noise conditions. This work demonstrates how a functional memristive interface can physically achieve near-sensor-level signal processing, offering a promising pathway for edge vision systems, intelligent sensing applications, and advances beyond the von Neumann architecture.