Research On Self-powered Artificial Vision System

The human senses can sense and respond to a large number of stimuli from the complex environment.80%of the perceived external stimuli are the light information obtained through the visual system,and the visual system can also process the perception and memory learning process in parallel to identify and process the external environmental information,which facilitates the development of electronic devices to realize bionic vision systems.Given the ability of computing systems based on the von Neumann architecture to store large amounts of data quickly,it is widely used in the field of artificial intelligence.However,the separation of processor and memory can limit the processing speed of computers.Existing computational methods are unable to learn or process complex signals as biological vision systems do.In contrast,data processing and storage in memristors are parallel,simulating the signal transmission and information processing functions of human brain.Therefore,artificial vision system based on memristor has attracted extensive attention in the field of artificial intelligence and bionics because of its potential to perform complex tasks such as image recognition,target detection,learning and memory.The reported bionic vision system integrating three terminal synaptic transistor and optical sensor realizes lossless data reading because of its advantage of separation of reading and writing.However,the transistor has three ports,which leads to its complex preparation process and integrated circuit design Moreover,due to the drive of external power supply,the synaptic weight update under full optical control cannot be realized.ere,we designed a self powered fully optically controlled artificial vision reflex arc system,which can simulate the basic functions of synapses only by light irradiation,and achieved high classification accuracy in neural network training.Firstly,in order to realize the recognition and memory function of the artificial vision system for the optical signal,the monocrystalline silicon solar cell,memristor and digital multimeter are connected to form an artificial vision system to simulate the recognition and memory function of the human eye for the optical signal.The electrical power output from the solar cell illuminated by light is also used to drive the system to achieve self-powered visual signal processing.The voltage of photoelectric conversion will be unstable,so high stability memristor is needed to increase the stability of the test.A digital Ag/Si O₂/Ta₂O₅/Pt memristor device with high and low resistive states using Ta₂O₅ as dielectric layer was prepared.We tested the discrimination and memory ability of the device for different optical signals,and verified the feasibility of the system.Secondly,the simulation of synaptic function by visual bionic system is explored.In the artificial vision system,a self-made ring oscillator is connected to convert the dc voltage signal generated by solar cells under illumination into pulse signal,so as to simulate the neural pulse signal in neural transmission.We fabricated simulated Pt/Ta₂O₅/W devices with multi-value conductance states as artificial synapses and connected it in the system.Artificial vision systems mimic the basic functions of synapses,and the distinction between wavelength and light intensity is realized,which proves the reliability of visual simulation ability of visual system.Finally,a memristor with better performance was prepared to optimize the simulation of synaptic function by the visual bionic system.Based on the principle of electrochromism,two terminal W/Li⁺/WO_x/W devices were prepared with WO_x as conductive channel.Experimental studies show that the optimized visual system can not only simulate the function of neural morphology,but also has excellent stability in the cycle process of long-term enhancement(LTP)/long-term inhibition(LDP)of postsynaptic current under continuous light irradiation.The continuously changing conductance of the device is mapped into the neural network for synaptic weight training,and a high classification accuracy is obtained.A complete artificial vision reflex arc system is constructed by connecting the artificial muscle to simulate the motion output.