All‐Optical Controlled Neuromorphic Vision Sensor for High‐Accuracy Image Recognition

Abstract As a disruptive design in brain‐like computing, event‐driven neuromorphic visual sensors are highly dependent on the synergistic effect of optical and electrical stimulation. However, it is a challenge to achieve bidirectional positive–negative reconfigurable changes of synaptic weights by all‐optical regulation without electrical stimulation. Here, a novel neuromorphic visual sensor based on an indium oxide (In 2 O 3 )/perovskite (PVK) heterojunction is proposed, which achieves precise and reversible regulation of synaptic weights by modulating localized electronic states and interface effects using light. Under UV light, photogenerated electrons in In 2 O 3 are bound by oxygen vacancies, while electron accumulation at the In 2 O 3 /PVK interface and electron–hole recombination together lead to negative photoconductivity. Under red light, the device retains positive photoconductivity. Notably, the device exhibits an ultra‐long persistent photocurrent of 3600 s and 0.8 fJ ultralow energy consumption. In addition, by constructing a 7 × 7 array structure, the imaging, memory, erasing capabilities, and the timing controllability of the novel neuromorphic visual sensor are demonstrated. The artificial vision system obtained is able to recognize images of 150 × 150 pixels and achieve a state‐of‐the‐art accuracy rate of 98.03%. This research provides new ideas for the design of bionic visual devices and paves the way for the development of neuromorphic photonics.