数十年来,人们一直在探索基于光的计算体系,期望通过光的特性,突破电子计算机的局限,从而提高计算速度和降低能耗。然而,传统光计算由于缺乏有效的逻辑、存储、互连单元,以及合理的应用场景,光计算止步于概念研究阶段。在当今大数据时代的推动下,信息通量及用量呈爆炸性增长。具有高集成度、大带宽、低成本、低能耗特征的硅基光电子技术应运而生,日趋成熟,并且验证了光电融合这一信息技术发展的趋势。光计算也随之向光电计算转变。分析近年来硅基光电子技术在光电计算方面的应用和发展,如在人工神经网络、非多项式时间复杂度难题的启发式算法、光电模拟计算、集成光电量子处理器和神经拟态计算等,重点论述了硅基光电子技术在促进光电计算的信息连接、数据处理和实用化演进等方面的优势,提出了硅基光电计算概念及初级系统,以突破传统电子技术或光子技术在计算方面的不足及其在人工智能等高性能计算领域中的限制。

Scientists have been exploring the optical computing system for decades, in hoping that it will have faster speed and lower energy consumption to break the limitations of the traditional electronic computing system. However, due to the ineffective optical logic units, slow electronic interconnects, and lack of optical memory, the idea of the optical computing did not go too far and was eventually abandoned. In the era of big data, the flow and usage of information have been increased exponentially, which provides a unique opportunity for the rapid development of the silicon photonic technology. The heterogeneous integration of electronics and photonics on the same silicon substrate, the Silicon Photonics, ensures a wide-bandwidth, high-speed, low-cost, low energy-consumption, and application-oriented optoelectronic computing platform. Here, we review the recent efforts of computing using silicon photonic technology, including artificial neural network accelerators, heuristic solvers of nondeterministic polynomial problems, analog calculations, quantum logic processors, and neuromorphic photonics, and propose a computing structure based on silicon photonic platform to take the advantages of high speed optical parallel processing and high capacity optical interconnections at the same time. We expect the Silicon Photonics will be the best technology to solve the computing challenges and problems in the electronical or optical domain.