Logical Perception of Artificial Vision Based on Nonlinear Neuromorphic Responses of Optoelectronic Synapses

Abstract

Artificial optoelectronic synapses have emerged as a promising technology for in-memory neuromorphic computing of artificial visual perception. To improve perception capability, logical functions of optoelectronic synapses, in which the photocurrent threshold is used to determine the output, have been reported. However, to implement the logical functions into neuromorphic computing of artificial visual perception, the relationship between two input images must be tested using a fixed threshold of recognition accuracy. Herein, artificial optoelectronic synapses based on MoS2 thin films and carbon quantum dots are fabricated. Due to photogenerated carrier transfer, the integration of quantum dots extends the memory of photocurrent responses. Voltage-modulated plasticity, paired-pulse facilitation, and high-efficiency learning ability of the synapses have been demonstrated. For neuromorphic computing, handwritten digital images are simulated using optoelectronic responses and input into an artificial neural network for recognition. With the increase in photocurrent, the recognition accuracy is enhanced quickly first and then saturates. The nonlinear relationship between the photocurrents and accuracy values enables the synapses to conduct logical operations. A fixed accuracy threshold can be utilized for “AND”, “OR”, and “XOR” operations. Moreover, the operations have a high tolerance, since the accuracy threshold can be set within a broad range. The results demonstrate attractive bioinspired logical behaviors in high-capability information processing, opening up potential applications of artificial visual systems in unmanned vehicles, robotics, and cyborgs.