光学神经网络实现与应用研究

Research on Implementation and Application of Optical Neural Networks

信息技术和人工智能的发展带来了数据的指数增长和对算力的爆发式需求。集成电路技术进步所能提供算力的速度已经远落后于人工智能算力需求的增长速度。同时传统电子计算系统由于冯诺伊曼架构的局限,难以满足对计算速度和功耗的要求。光学计算系统是有效解决传统电子计算系统算力、速度和功耗难题的可行方案之一。光学神经网络是光学计算的一个重要方面,通过光学硬件搭建的光学神经网络可直接在物理上实现卷积、微分、积分等数学运算。由于光学神经网络具有高并行能力、高带宽、高速度、低功耗等优势,能够有效解决人工智能的算力焦虑和功耗制约,在图像处理、语音识别等广泛领域有重要应用价值。

The rapid advancements in information technology and artificial intelligence(AI)have fueled exponential data growth and an unprecedented demand for computational capabilities.However,the pace of computational power enhancement through integrated circuit technology advancements has struggled to keep up with the soaring needs of AI.Furthermore,traditional electronic computing systems,constrained by the Von Neumann Architecture,struggle to meet the stringent requirements of speed and power consumption.Optical computing systems emerge as promising solutions,addressing the computational limitations and power challenges faced by their electronic counterparts.At the heart of optical computing lie optical neural networks,realized through optical hardware,which inherently facilitate mathematical operations such as convolution,differentiation,and integration in a physical manner.Leveraging their inherent advantages of high parallelism,wide bandwidth,blazing speeds,and minimal power consumption,optical neural networks offer a viable path to alleviate the computational and power constraints hindering AI.Consequently,they hold significant potential for applications spanning image recognition,edge detection,voice recognition,and beyond.