Ultra-Low Power In-Sensor Computing β-Ga₂O₃ Ultraviolet Optoelectronic Synaptic Devices

Abstract

Deep ultraviolet (DUV) photodetection typically struggles with significant noise and low contrast due to radiation and atmospheric interference. Integrating image enhancement and preprocessing functionalities often necessitates complex circuitry. To address these issues, this study introduces a $\beta $ -Ga2O3-based optoelectronic neuromorphic device utilizing pulsed light stimulation, designed to emulate brain-like integrated sensing and computing capabilities. By increasing the TEGa flow rate during the growth process, extra oxygen vacancies (V $_{\mathrm {o}}$ ) were introduced into $\beta $ -Ga2O3, enabling the device to mimic critical biological synapse traits such as short-term plasticity and the learning-forgetting-relearning cycle, essential for dynamic data processing. These synaptic features allow the device to perform effective visual preprocessing, which significantly improves image recognition accuracy. Specifically, with added noise standard deviations of 0.2, 0.3, and 0.4, preprocessing resulted in recognition accuracy increases of 19.4%, 54.7%, and 161.7%, respectively. Importantly, the Vo-rich composition resulted in reduced photocurrent and ultra-low energy consumption (25 fJ) approaches of biological synapses. This device exhibits only 0.1% of the energy consuming compared to similar Ga2O3 synaptic devices through normalization comparison. These improvements highlight the device’s capability to significantly enhance DUV image quality and usability, offering valuable insights for the development of integrated sensing and computing Ga2O3 devices.