An all-in-one multi-mode photodetector for ultrafast adaptive vision in extreme-to-obscured conditions with intelligence-augmented classification

Abstract

Traditional photodetectors are often limited to specific tasks, such as continuous illumination sensing or event detection, relying on distinct architectures for functionalities like static and dynamic pattern recognition. Achieving seamless integration of temporal processing, event-driven adaptability, and static pattern recognition within a unified architecture remains a significant challenge. These limitations are further intensified by environmental variations, such as fluctuating lighting and obscured conditions, which undermine their reliability in real-world applications. Here, we present a reconfigurable and adaptive single-pixel photodetector capable of transitioning seamlessly between photodetector, synaptic, and retinomorphic modes by altering its operating conditions. Leveraging quantum-well-inspired charge trapping mechanisms, the device achieves ultrafast transient detection, cumulative signal integration, and robust adaptability, mimicking key functionalities of biological vision. Experimental results, supported by simulations, demonstrate real-time adaptive vision, multi-object tracking, and enhanced pattern recognition under diverse environments, ranging from intense sunlight to obscured conditions. The device transcends conventional sensing by enabling object classification through machine learning, achieving over 94 % accuracy using multidimensional metrics, even for objects with similar shapes and sizes. This multifunctional photodetection platform addresses critical challenges in next-generation sensing technologies by combining adaptability, high-speed response, and intelligent classification, paving the way for transformative applications in autonomous vision, neuromorphic computing, and intelligent imaging systems.