光神经拟态系统能以比生物大脑快几百万至十亿倍的运行速度模拟神经拟态算法,优于电神经拟态硬件系统,且其可胜任比传统光计算更复杂的计算任务。光神经拟态计算探索超快光脉冲信号的自适应性、稳健性和快速性,能够避免传统数字光计算的芯片集成及模拟光计算的噪声积累等问题。本文报道了光子神经拟态信息处理的发展历程,并从光子神经元,光脉冲学习算法以及可集成光学神经拟态网络框架等方面介绍了光神经拟态计算的关键理论和技术。阐述了光神经拟态计算研究的必要性及存在的问题,展望了其潜在的应用前景。 Optical neuromimetic systems can simulate neuromimetic algorithms at speeds of several to one billion times faster than biological ones, outperform electrical neuromorphic hardware systems and are capable of performing more complex computational tasks than traditional optics. Optical neuromorphology calculation explores the adaptability, robustness and speediness of ultrafast optical pulse signals, and avoids the problems of traditional digital light chip integration and noise accumulation in analog light calculation. This paper reports the development of photon neuromimetic information processing and introduces the key theories and techniques of photon neuromimetry from the aspects of photon neurons, optical pulse learning algorithms and the integrated optical neuromorphic network framework. The necessity and existing problems of light neuromorphism calculation are expounded, and its potential application prospects are prospected.