摘要
In light of recent achievements in optical computing and machine learning, we consider the conditions under which all-optical computing may surpass electronic and optoelectronic computing in terms of energy efficiency and scalability. When considering the performance of a system as a whole, the cost of memory access and data acquisition is likely to be one of the main efficiency bottlenecks not only for electronic, but also for optoelectronic and all-optical devices. However, we predict that all-optical devices will be at an advantage in the case of inference in large neural network models, and the advantage will be particularly large in the case of generative models. We also consider the limitations of all-optical neural networks, including footprint, strength of nonlinearity, optical signal degradation, limited precision of computations, and quantum noise.
学科与关键词
- Optical neural networks: progress and challengesFu, Tingzhao; Zhang, Jianfa; Sun, Run; Huang, Yuyao; Xu, Wei; Yang, Sigang; Zhu, Zhihong; Chen, Hongwei*Light-Science & Applications, 2024, 13(1): 263.被引 1
- Light in AI: Toward Efficient Neurocomputing With Optical Neural Networks-A TutorialGu, Jiaqi*; Feng, Chenghao; Zhu, Hanqing; Chen, Ray T.; Pan, David Z.IEEE Transactions on Circuits and Systems II: Express Briefs , 2022, 69(6): 2581-2585.被引 11
- Energy-Efficient Neural Network Inference with Microcavity Exciton PolaritonsMatuszewski, M.*; Opala, A.; Mirek, R.; Furman, M.; Krol, M.; Tyszka, K.; Liew, T. C. H.; Ballarini, D.; Sanvitto, D.; Szczytko, J.; Pietka, B.Physical Review Applied, 2021, 16(2): 024045.被引 16
