All-optically controlled artificial synaptic device for neural behavior simulation and computer vision

Abstract

The increasing demands for high-speed computation and energy efficiency in artificial intelligence (AI) applications necessitate the development of novel computing paradigms. Neuromorphic computing, which mimics biological neural systems, offers a promising solution to Von Neumann bottleneck by emulating brain-like processing capabilities. As the hardware of neuromorphic computing, the all-optically controlled artificial synaptic device offers a foundational platform for low power consumption and high bandwidth performance. In this work, we introduce an all-optically controlled artificial synaptic device based on an amorphous ZnSiSnO/SnO p-n junction structure. The device exhibits excitatory and inhibitory synaptic behaviors through visible light modulation, demonstrating persistent photoconductivity (PPC) for effective synaptic learning. By simulating biological neural behaviors such as the learning-forgetting-relearning mechanism, pain-pleasure mechanism and noise tolerance, the device achieves diverse functionality for neuromorphic computing. Furthermore, we demonstrate its application in computer vision, achieving edge detection for automatic driving systems and high-performance recognition in artificial neural networks (ANNs) for handwritten and clothing images. The device also enables optical logic operations, offering potential for advanced neuromorphic applications and AI integration.