Adaptable photonic artificial neurons for attention-based object identification

Abstract

The human visual system, reflected by its unique attention-dependent object recognition, holds remarkable potential. However, emulating the capabilities of the human visual system using typical two-terminal photodetectors is challenging. This is because these devices, which respond directly to optical and/or electrical stimuli, do not possess the adaptive and selective features inherent in bio-neuronal systems. In this study, we present a report of adaptable, attention-dependent object recognition utilizing gallium oxide-based photodetectors. The device demonstrates a pronounced hysteresis loop opening with an on/off ratio of less than 10 while maintaining ultra-low dark current levels below 10−10 A in its current-voltage curves. Notably, the on/off ratio can dynamically adjust to multilevel, > 102, and even rise to 103 under ultraviolet illumination. The observed results are attributed to the charge trapping and detrapping processes, as evidenced by detailed photocurrent mapping. Moreover, when subjected to simultaneous optical and electrical stimuli, the current first rises and subsequently falls post-peak, exhibiting neuron-like dynamics. These dynamics are subsequently used to emulate attention-based object identification by designing 3 × 3 array. Our findings present proof-of-concept photodetectors that emulate bio-neuronal object recognition systems, suggesting considerable potential for breakthroughs in areas like surveillance, remote sensing, medical imaging, and machine vision.