A Heterosynapse-Inspired Photodetector for Spatiotemporal Feature Fusion

The increasing amount of data puts enormous pressure on data center. Therefore, the edge visual system is required to extract and process information more efficiently. Despite the development of neuromorphic visual systems, it has remained a challenge to process spatiotemporal information in the hardware-implemented visual system. Here, a heterosynapse-inspired photodetector is proposed for the intelligent edge visual system to process spatiotemporal information. The persistent photoconductivity (PPC) effect enables the photodetector to process temporal information, and the heterosynapse-inspired structure allows the visual system based on this photodetector to deal with spatial patterns. The proposed bionic photodetector provides a strategy for spatiotemporal feature fusion, which enables the target with spatiotemporal features recognized by the classifier composed of voltage comparators. Moreover, it is demonstrated that the visual system based on this design is able to conduct spatiotemporal pattern recognition and motion direction recognition tasks. This design within pixels paves a novel route for the intelligent edge visual system to extract, fuse, and recognize spatiotemporal information.