Retina-inspired in-sensor broadband image preprocessing for accurate recognition via the flexophototronic effect

•3DSS process can achieve permanent polarization, integration, and practical application •The enhanced photosensors have realized broadband perception and high photoresponsivity •The light-dosage-tunable synaptic characteristics have been simulated by the sensors •An artificial visual system can perform preprocessing tasks for accurate recognition Neuromorphic visual systems are promising candidates for emulating vision perception and recognition tasks, especially for two-dimensional (2D) semiconductor-based imaging sensors due to their tunable electrical/optical properties and compatibility for heterogeneous integration. The device integration of sensing-memory-processing units and accurate recognition for image sensors become critical obstacles to simulating visual systems; thus, developing more advanced neuromorphic structures is required. Here, we demonstrate a retina-inspired 2D molybdenum disulfide (MoS2)-based vision sensor capable of both broadband perception and preprocessing functions by exploiting the flexophototronic effect. Three-dimensional stress stabilization (3DSS) is consciously introduced into the flexophototronic sensor to achieve broadband sensing and high photoresponsivity, resulting in distinctive UV information extraction for image preprocessing. Furthermore, the artificial neural network can remarkably enhance the recognition accuracy from 35% to 90% due to preprocessing validity. This work lays the foundation for developing emerging flexophototronics for artificial visual systems endowed with real-time interaction and efficient execution. Neuromorphic visual systems are promising candidates for emulating vision perception and recognition tasks, especially for two-dimensional (2D) semiconductor-based imaging sensors due to their tunable electrical/optical properties and compatibility for heterogeneous integration. The device integration of sensing-memory-processing units and accurate recognition for image sensors become critical obstacles to simulating visual systems; thus, developing more advanced neuromorphic structures is required. Here, we demonstrate a retina-inspired 2D molybdenum disulfide (MoS2)-based vision sensor capable of both broadband perception and preprocessing functions by exploiting the flexophototronic effect. Three-dimensional stress stabilization (3DSS) is consciously introduced into the flexophototronic sensor to achieve broadband sensing and high photoresponsivity, resulting in distinctive UV information extraction for image preprocessing. Furthermore, the artificial neural network can remarkably enhance the recognition accuracy from 35% to 90% due to preprocessing validity. This work lays the foundation for developing emerging flexophototronics for artificial visual systems endowed with real-time interaction and efficient execution.