Reconfigurable Nonvolatile Photodetectors for Brain-Inspired Vision.

Abstract

Machine vision serves as the essential sensorial interface for intelligent systems, yet conventional vision systems based on the von Neumann architecture suffer from significant latency and high power consumption due to the physical separation of sensing and processing units. To address these bottlenecks, neuromorphic vision systems inspired by the efficient, localized processing of the human retina have emerged. As a core paradigm of in-sensor computing, reconfigurable nonvolatile photodetectors (RNVPs) that integrate photodetection, nonvolatile memory, and processing into a single device represent a promising frontier for energy-efficient, real-time artificial intelligence. This review provides a comprehensive overview of RNVPs. It begins by introducing the human visual perception system and the paradigm of bioinspired in-sensor computing architectures. Subsequently, we overview the core concepts of RNVPs and establish the key performance metrics. Recent advances in RNVPs are then discussed in detail according to their working mechanisms, followed by a summary of their potential applications in image processing. To conclude, we highlight current challenges and offer perspectives on the future trajectory of RNVPs for next-generation in-sensor computing.