A Reconfigurable Neuromorphic Vision Sensor with Dual-Mode Capability for High-Quality Static and Dynamic Imaging

Advanced artificial vision systems (AVSs) require both high-fidelity spatial imaging and efficient temporal information processing to perceive static and dynamic visual information. However, integrating these two capabilities within a compact sensing platform remains challenging. Here, we develop a reconfigurable two-terminal vision sensor enabled by asymmetric defect engineering at metal contact/semiconductor interface. Essentially, by spatially introducing an amorphous-GaON (a-GaON)/GaN heterointerface into the GaN conduction channel, the sensor possesses two distinct functionalities with the capability of switching between high-gain photodetection and neuromorphic temporal encoding mode under negative or positive bias condition. This functional reconfigurability arises from bias-modulated heterointerface-trap-assisted electron injection and bulk-defect-mediated persistent photoconductivity at the two different contacts of device, respectively. Consequently, under positive bias, the device achieves high-gain photodetection with a low dark current below 1 pA, a maximum responsivity of 1.7×10 4 A/W, and a photo-to-dark current ratio of 3.3×10 10 , enabling high-contrast imaging using an 8×8 device array. Under negative bias, the device exhibits programmable photosynaptic plasticity, including paired-pulse facilitation (PPF) with a PPF index of 183%. In the end, an in-sensor reservoir computing framework is demonstrated, which encodes temporal information into reservoir states achieving 97.2% accuracy in recognizing 16 motion categories at different directions and speeds. This work establishes an asymmetric defect engineering approach at metal/semiconductor contact for integrated high-performance optical sensing and in-sensor temporal computing in the pursuit of efficient and compact AVSs of future.