Neuromorphic vision array based on full-spectrum perovskite materials for object detection in complex environments

Abstract

The physical scaling limitations of CMOS technology and the impending end of Moore’s Law have led to increasing interest in neuromorphic computing approaches to enhance computational performance. Inspired by the parallel computing capabilities of the human brain, neuromorphic devices that integrate memory, computation and perception have been developed to address these challenges. Among various perception simulations, vision is particularly important, as it accounts for over 70 % of sensory input in humans. Full-spectrum technology, as a method capable of capturing rich optical information, has effectively realized the process of simulating visual target detection. Meanwhile, the perovskite-structured material CsPbI₃, with its wide-spectrum absorption, high photoelectric conversion efficiency, and excellent stability, which makes CsPbI₃ an ideal candidate for visual arrays that can span the full spectrum of light. In this study, we propose a full-spectrum perovskite-based visual array for use in unmanned object detection platforms, replacing the conventional perception, memory, and computation units. By extracting parameters from all 64 devices on a single visual array and inputting them into a neural network, we achieved a more realistic and comprehensive simulation of ground target detection and recognition compared to previous methods. The results demonstrate significant improvements in data processing efficiency and performance, suggesting broad potential for application in intelligent devices, industrial automation, and medical equipment.