Time-multiplexed Weight Sharing of Photonic Neural Networks
Silicon photonic neural networks often rely on thermal tuning of weights, limiting scaling for deep learning tasks. We demonstrate a time-domain weight sharing method that can reduce reconfigurations and energy by 6.6x without compromising accuracy.
Authors
- Paul R. Prucnal
- Eli A. Doris (ORCID: https://orcid.org/0000-0003-4015-3303)
- Bhavin J. Shastri (ORCID: https://orcid.org/0000-0001-5040-8248)
- Yusuf O. Jimoh
- Jiawei Zhang (ORCID: https://orcid.org/0000-0003-0740-0851)
- Weipeng Zhang (ORCID: https://orcid.org/0000-0002-6208-2051)
Institutions
- Princeton University (US)
- Queen's University (CA)
Publication Details
- Published
- 2025-01-01
- DOI
- https://doi.org/10.1364/cleo_si.2025.ss169_5
- Primary Topic
- Neural Networks and Reservoir Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.00