Fully optically controlled Li-ion-mediated artificial vision reflection arc system

Abstract

Artificial vision systems have been extensively reported for processing unstructured data, performing pattern recognition, and performing visual simulations. Photonic synapses have become the core of most artificial vision systems due to their advantages such as low energy consumption and ultra-fast signal transmission. However, their integration is difficult, the integration difficulty of photon synapses is higher than that of crossbar arrays. Moreover, due to the hybrid drive of light and external power supply, low energy consumption and weight update under full optical control cannot be achieved. Here, we designed a fully light-controlled multifunctional artificial vision reflex arc system. The solar cell and oscillator generate presynaptic voltage signals that are then transmitted to a lithium-ion-based memristor for the stimulation of different wavelengths and light intensities. The results show that our artificial vision system can distinguish wavelength and light intensity information and has good weight update symmetry and repeatability. Finally, the deformation experiment of the shape memory alloy demonstrated that the system can be used for controlling artificial muscles under different light signals.