Multifunctional broadband artificial visual system using all-in-one two-dimensional optoelectronic transistors

Abstract

The bio-inspired artificial neuromorphic visual system, which possesses integrated sensing, memory and computing functionalities, is envisioned to demonstrate great potential in overcoming the challenges of data transmission latency and high energy consumption. Huge efforts have been devoted to developing novel hardware devices to achieve the goal under various working mechanisms, which however yield limited success in emulating the full functionalities of the visual system in a single device with simplified configuration. Here, we report an all-in-one visual platform based on a multifunctional MoS₂ phototransistor array fabricated on the silicon-rich silicon nitride (sr-SiN_x) substrate for in-sensor computing from ultraviolet to near-infrared spectrum. The array exhibits non-volatile optical/electrical programming features through deliberately manipulating the charge storage in the sr-SiN_x dielectric, which are analogous to the learning/forgetting processes in the real human visual system. These characteristics enable the integration of broadband image sensing and pre-processing, dynamic learning and noise filtering, and image recognition in a single device. The array achieves high recognition accuracy of 90.2 % (98.4 %) based on the Fashion MNIST (MNIST) database, suggesting its robust functionalities. These results envision dielectric engineering as a promising approach to realize simplified neuromorphic visual units that integrate all the fundamental functions in the bio-visual system, offering new opportunities for designing innovative neuromorphic hardware.