An Optoelectronic Artificial Synapse Based on CuIn0.7Ga0.3Se2/ Al-doped ZnO p-n Heterojunction for Bioinspired Neuromorphic Computing

Abstract

The traditional von Neumann architecture continues to limit the development of artificial intelligence. Memristors have become one of the most promising devices for breaking through the traditional von Neumann architecture. In this work, an optoelectronic synapse based on the CuIn0.7Ga0.3Se2 (CIGS)/ Al-doped ZnO (AZO) p-n heterojunction is prepared by radio-frequency (RF) magnetron sputtering. And the Au/CIGS/AZO/ITO p-n heterojunction artificial synapse has been utilized to simulate various synaptic behaviors as well as the learning-forgetting-relearning process of the human brain. Furthermore, employing a convolutional neural network (CNN) architecture with an enhanced stochastic gradient descent algorithm, the recognition accuracy for the MNIST and Fashion-MNIST datasets is achieved at 97.36% and 83%, respectively, demonstrating the potential application of Au/CIGS/AZO/ITO p-n heterojunction artificial synapse in neuromorphic computing and providing a feasible method for the development of high-performance optoelectronic devices based on CIGS/AZO p-n heterojunctions.