A Spiking Visual Neuron for Depth Perceptual Systems

Abstract The biological visual system encodes information into spikes and processes them parallelly by the neural network, which enables the perception with high throughput of visual information processing at an energy budget of a few watts. The parallelism and efficiency of bio-visual system motivates electronic implementation of this biological computing paradigm, which is challenged by the lack of bionic devices, such as spiking neurons that can mimic its biological counterpart. Here, we present a highly bio-realistic spiking visual neuron based on an Ag/TaOX/ITO memristor. Such spiking visual neuron collects visual information by a photodetector, encodes them into action potentials through the memristive spiking encoder, and interprets them for recognition tasks based on a network of neuromorphic transistors. The firing spikes generated by the memristive spiking encoders have a frequency range of 1-200 Hz and sub-micro watts power consumption, very close to the biological counterparts. Furthermore, a spiking visual system is demonstrated, replicating the distance-dependent response and eye fatigue of biological visual systems. The mimicked depth perception shows a recognition improvement by adapting to sights at different distance. Our design presents a fundamental building block for energy-efficient and biologically plausible artificial visual systems.