摘要
当前物联网、云计算等产生的海量非结构化数据,极大提高了对数据处理算力和能效的需求。神经形态计算借鉴生物大脑的信息处理方式,以神经元与神经突触为基本单元,从互联架构与信息处理模式等方面模拟生物神经系统,能够实现实时、超低功耗信息处理,成为大数据时代计算技术发展的前沿热点。其中,光子神经形态计算技术是在光域上进行神经形态计算数据处理的技术,既能够充分发挥光子高速传输、低功耗、高并行度的优势,又能够避免光电和电光转换带来的额外时间功耗开销,具有很大的研究和应用价值。近年来,相变材料作为一种具有高折射率对比度和非易失特性的光学材料,可在光、电、热等激励作用下进行折射率的连续调节,为非易失光子神经形态计算提供了一种可行的解决方案,成为当前的研究热点。本文首先介绍了光子神经形态计算的基本原理和实现方法,在此基础上讨论了相变材料用于光子神经形态计算的原理。其次,针对不同类型的实现途径,研究了不同相变材料的特点和选型办法,综合分析了当前应用较多的两类相变材料以及各类光突触器件和阵列集成应用。最后对基于相变材料的光子神经形态计算技术的发展进行了展望。
The large amount of unstructured data generated by the Internet of Things and cloud computing has recently increased the demand for data computing power and energy efficiency.Referencing the information processing method of the biological brain with neuron and synapse as basic units,neuromorphic computing can simulate the biological nervous system from the aspects of interconnection architecture and information processing mode,and realize ultra-low power processing of realtime information,which have become the forefront of the development of computing technology in the big data era.The processing of computational data in the optical domain makes photonic neuromorphic computing research important owing to its high application potential.On the one hand,photonic neuromorphic computing can take advantage of high-speed transmission,low power consumption,and high parallelism of photons.On the other hand,it can also prevent photoelectric and electro-optic conversion,thus,reducing additional time and power consumption.In recent years,phase-change materials(PCM),as a kind of optical material with high refractive index contrast and non-volatile property,whose refractive rate can be continuously adjusted under the driving of optical,electrical,and thermal excitations,have provided a feasible solution for non-volatile photonic neuromorphic computing and have become the current research hotspot.In this paper,we first introduce the basic principle and implementation method of photonic neuromorphic computing.Subsequently,we discuss the principle of utilizing phase-change materials in photonic neuromorphic computing.According to the unique characteristics of phase-change materials selected in different implementation schemes,two kinds of phase-change materials and different applications of optical synapse devices and integrated arrays are then summarized.Finally,we prospect the development of photonic neuromorphic computing techniques based on phase-change materials.
作者
王金融
宋兵
徐晖
张亨宇
孙振源
李清江
Wang Jinrong;Song Bing;Xu Hui;Zhang Hengyu;Sun Zhenyuan;Li Qingjiang(College of Electronic Science and Technology,National University of Defense Technology,Changsha 410073,Hunan,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2023年第21期101-116,共16页
Laser & Optoelectronics Progress
基金
国家自然科学基金(62104256,61974164,62074166,61804181,62004219,62004220)。
关键词
材料
相变材料
神经形态计算
光神经网络
materials
phase-change material
neuromorphic computing
optical neural network
