Metal Oxide Nano-Interface Boosting the Deep Ultraviolet Adjustable Noise-Filtering In-Sensor Computing.

Abstract

Long-afterglow light-emitting devices (LALEDs), which combine the capabilities of sensing, memory, processing, and display integration, allow for the simultaneous implementation of optical and electrical in-sensor computing. However, the inherently low conductivity of conventional deep-ultraviolet (DUV) responsive materials hinders their use in the DUV-responsive LALEDs (DUV-LALEDs). Herein, we demonstrate that sol-gel-fractured indium-magnesium oxide (InMgO) concurrently exhibits ideal nano-interface for DUV photon absorption and semiconductor crystal for efficient charge transport via hopping, enabling to reach high mobility (0.6 cm² V^-1 s^-1), excellent memory dynamic range (70 dB), and responsivity (523.7 A/W). The InMgO-based DUV-LALEDs display an electrical and optical post-synaptic output when irradiated with DUV light. Moreover, hardware-level noise-filtering processes to the pre-synaptic weight are revealed in the DUV-LALEDs, emerging as the inhibition of light emission due to the insufficient post-synaptic charge injection from the channel layer. Consequently, an adjustable noise-filtering in-sensor computing is successfully achieved by modulating the channel length. By taking advantage of the DUV-LALED multifunctional nature, fusion-node reservoir computing networks are employed to accomplish multi-dimensional recognition tasks, displaying a high recognition accuracy of 99%. These findings demonstrate that the joint materials and devices optimization is a powerful strategy for fabricating cost-efficient DUV analytical chips.