光神经形态计算研究进展与展望(特邀)

Progress and Prospects of Photonic Neuromorphic Computing(Invited)

脑科学与类脑研究是国际必争战略性前沿。人工智能与深度学习的飞速发展对算力提出了迫切需求。而传统的冯诺依曼架构,由于存算分离导致功耗墙和内存墙,摩尔定律也逐渐放缓。光神经拟态计算充分融合高速光通信、光互连、光集成、硅基光电子与神经拟态计算的特点,具有超高速、大带宽、多维度等优势,在高性能计算、人工智能领域有广阔的应用前景,是突破后摩尔时代传统微电子计算极限极具竞争力的方案。本文回顾了国内外主要研究团队在光神经元、光突触、光神经网络的理论、算法及器件方面的工作,并提出了展望。

Brain science and brain-like research have become the strategic frontier of international competition.The rapid development of artifical intelligence and deep learning has put forward an urgent demand for the computing capacities.In the traditional von Neumann architecture,the physical separation between memory and computing units results in power consumption wall and memory wall problems.Besides,Moore's law is gradually slowing down.Photonic neuromorphic computing,which fully combines the characteristics of high-speed optical communication,optical interconnection,optical integration,silicon-based optoelectronics and neuromorphic computing,has the advantages of ultra-high speed,large bandwidth and multi-dimension.It has wide application prospects in the fields of high-performance computing and artificial intelligence.Furthermore,it is a highly competitive solution that breaks through the limits of traditional microelectronics computing in the post-Moore era.This article reviews the work of the main research teams at home and abroad on the theory,algorithms,and devices of photonic neurons,synapses,and neural networks,and puts forward a prospect.