Monolithic integration of GaN synapse and photodetector for bio-inspired adaptive vision

Abstract

This work addresses critical limitations in conventional artificial vision systems, namely the von Neumann architecture bottleneck and inadequate environmental adaptability, by proposing a bionic vision system with integrated sensing, memory, and computing based on gallium nitride (GaN) optoelectronic synaptic devices. Employing an innovative GaN-on-Si platform and homogeneous integration technology, we achieved monolithic integration of a photodetector (PD) with rapid response and an optoelectronic synaptic transistor with persistent response on a single chip. The front-end PD leverages the wide bandgap properties of GaN to achieve low dark current (∼10−13 A) and high-responsivity detection specifically for ultraviolet optical signals. The back-end synaptic transistor exploits the manipulation of photo-generated carrier trapping/release at the AlGaN/SiO2 interface trap states, inducing a persistent photoconductivity effect. This mechanism emulates key biological synaptic behaviors, including the excitatory postsynaptic current, paired-pulse facilitation, and the dynamic transition from short-term plasticity to long-term plasticity, with an ultra-low energy consumption of 53.5 fJ per synaptic event. Furthermore, the system incorporates a closed-loop optoelectronic feedback mechanism. This enables precise modulation of the conductance state via a gate voltage, adaptively suppressing the response under strong illumination while actively enhancing the gain in low-light conditions, thereby faithfully replicating the dynamic light adaptation behavior of the human eye.