Adaptive tunneling photodiodes enable visual recognition in high-contrast scenes

Accurate visual sensing in extreme high-contrast lighting environments is critical for emerging intelligent systems such as autonomous vehicles, smart surveillance, and robotics. These systems rely heavily on high dynamic range (HDR) imaging to capture scenes with wide illumination variations. However, existing HDR solutions primarily depend on computational algorithms or hardware-intensive techniques such as multiexposure fusion or mechanical modulation, which suffers from latency, motion artifacts, high power consumption, and limited adaptability. Here, we present a hardware-centric approach that realizes HDR functionality directly into the photodetectors of a camera image sensor. Our design leverages an engineered tunneling mechanism to achieve bias-controllable, continuously tunable dynamic range covering up to 150 dB. In real-world high-contrast autonomous driving scenarios, this sensor-level HDR capability enables robust object classification confidence score exceeding 91% across the entire exposure scene. By shifting HDR processing to the sensor front-end, our approach substantially reduces the computational load, enabling system-level efficiency, speed, energy consumption, and recognition reliability. This prototype paves the way for the development of next-generation, compact, on-chip visual preprocessing hardware for intelligent vision systems.