Researchers from Soochow University, Northwestern Polytechnical University, Yangzhou University and King Saud University have developed a broadband neuromorphic visual sensor capable of stable, all-day operation across both visible and near-infrared (Vis–NIR) spectral regimes. Inspired by the scotopic (low-light) vision of the cat eye, the device integrates optical sensing, temporal encoding, and synaptic plasticity within a single architecture, enabling efficient spatiotemporal perception under dynamically varying illumination.
The device is based on a vertically coupled heterostructure combining a narrow-bandgap FASn0.5Pb0.5I3 perovskite absorber with defect-engineered SnS2. This design leverages sulfur vacancy-mediated non-equilibrium carrier transport and trap-state dynamics to emulate retina-like temporal integration. Photogenerated carriers are captured and gradually released by defect states, creating a time-dependent transport process that naturally reproduces synaptic plasticity and temporal weighting without the need for complex external circuitry.
As a result, the sensor exhibits stable broadband synaptic responses spanning the visible to near-infrared range, addressing a key limitation of conventional photonic synapses that typically operate only within narrow spectral windows. The inclusion of near-infrared sensitivity is particularly important for night-time and low-visibility conditions, where reduced scattering and stronger penetration enable more reliable signal acquisition compared to visible light alone.
A defining feature of the device is its ultralow energy consumption of just 0.15 fJ per event, significantly lower than many previously reported near-infrared neuromorphic systems that rely on voltage-driven operation. This efficiency arises from the intrinsic coupling between material properties and carrier dynamics, allowing sensing and computation to occur simultaneously within the device.
Beyond device-level performance, the researchers demonstrate system-level functionality by directly encoding continuous video streams using the intrinsic temporal dynamics of the sensor, eliminating the need for external buffering. By integrating a lightweight optical flow framework, they constructed an all-day dynamic tracking system capable of robust performance across varying illumination conditions.
The system achieves high tracking accuracy in both daytime and nighttime scenarios. During the day, vehicle and pedestrian tracking accuracies reach 94.94% and 94.71%, respectively. At night, performance further improves to 97.46% for vehicles and 95.23% for pedestrians. The study also highlights the relationship between tracking accuracy and the time accumulation constant of different targets, emphasizing the importance of temporal integration in dynamic perception tasks.
Overall, this work presents a material-driven approach to neuromorphic vision, where defect-regulated carrier dynamics in a perovskite/SnS2 heterostructure enable unified sensing, memory, and processing. By extending synaptic functionality into the near-infrared and demonstrating efficient real-time tracking, the study outlines a promising pathway toward energy-efficient, all-day intelligent vision systems operating in complex, real-world environments.
