Zhu Sha, Zhu Ning Hua
Institute of Intelligent Photonics, Nankai University, Tianjin, China.
Light Sci Appl. 2025 Sep 4;14(1):302. doi: 10.1038/s41377-025-01970-3.
The rapidly growing computational demands of artificial intelligence (AI) and complex optimization tasks are increasingly straining conventional electronic architectures, driving the search for novel, energy-efficient processing paradigms. Photonic computing, which harnesses the unique properties of light to perform computation, has emerged as a compelling alternative. This perspective highlights a key advancement: a versatile nonlinear optoelectronic engine based on integrated photodetectors and micro-ring modulators (PD + MRM). This engine enables crucial functionalities like nonlinear activation and signal relay, forming a core building block for monolithic photonic processors. Its application in integrating optical Ising machines for optimization and optical recurrent neural networks (RNNs) for AI has been examined recently. The PD + MRM unit's inherent compactness, efficiency, and on-chip reconfigurable nonlinearity address historical photonic computing challenges, signaling a shift towards more versatile and scalable monolithic photonic processors.
人工智能（AI）快速增长的计算需求以及复杂的优化任务正日益给传统电子架构带来压力，促使人们寻找新颖的、节能的处理范式。利用光的独特特性来执行计算的光子计算已成为一种极具吸引力的替代方案。这一观点突出了一项关键进展：一种基于集成光电探测器和微环调制器（PD + MRM）的通用非线性光电子引擎。该引擎实现了诸如非线性激活和信号中继等关键功能，构成了单片光子处理器的核心构建模块。最近已对其在集成用于优化的光学伊辛机和用于AI的光学循环神经网络（RNN）中的应用进行了研究。PD + MRM单元固有的紧凑性、效率和片上可重构非线性解决了光子计算的历史挑战，标志着向更通用、可扩展的单片光子处理器的转变。
