Scalable and Energy-Efficient Photonic Neural Networks Through Convolution Compression
This work experimentally demonstrates a convolution compression method to realize scalable and energy-efficient photonic neural networks. Our method achieves a 52-fold reduction in the component count and a 71.5-fold reduction in power consumption compared to conventional approaches, all while maintaining high computational accuracy.
Authors
- Benshan Wang (ORCID: https://orcid.org/0000-0001-9208-1378)
- Chaoran Huang (ORCID: https://orcid.org/0000-0001-8205-8254)
- Yixuan Li (ORCID: https://orcid.org/0000-0002-8000-3417)
- Tengji Xu (ORCID: https://orcid.org/0009-0008-5958-0149)
- Qiarong Xiao (ORCID: https://orcid.org/0009-0009-8900-9426)
- Shaojie Liu (ORCID: https://orcid.org/0009-0008-6369-5651)
Institutions
- Chinese University of Hong Kong (HK)
Publication Details
- Published
- 2024-11-02
- DOI
- https://doi.org/10.1109/acp/ipoc63121.2024.10809912
- Citations
- 1
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.35
Funders
- Innovation and Technology Fund