Event-driven retinomorphic photodiode with bio-plausible temporal dynamics

Machine vision is indispensable in Industry 4.0 and autonomous driving, enabling the perception and reaction necessary to navigate dynamic environments. Current machine vision sensors, including frame-based and event-based types, often fall short due to their limited temporal dynamics compared with the human retina, hindering their overall performance and adaptability. In this work, we present an event-driven retinomorphic photodiode (RPD) that mimics the retina's layered structure and signal pathway. The RPD achieves this by vertically integrating an organic donor-acceptor heterojunction, an ion reservoir with a porous web-like morphology, and a Schottky junction into a single diode through controlled layer-by-layer fabrication and precise nanostructure modulation. Each component replicates a key retinal process, and their spontaneous interaction results in environment-adaptive dynamics. This design yields a dynamic range exceeding 200 dB, substantially reduces noise and data redundancy, and allows for high-density integration. We demonstrate that these improvements enable high-quality machine vision, even under extreme lighting conditions. Our work demonstrates a bottom-up approach to retinomorphic sensors, propelling the development of robust and responsive machine vision systems adaptable to complex and dynamic lighting environments.