Photonic parallel convolutional processing based on spectrum slicing on silicon photonics platform
We have demonstrated a photonic CNN system employing spectrum slicing on a broadband light source and testing with MNIST handwritten characters dataset. The standard deviation of the photonic multiply-accumulate operations is 0.0417, resulting in the bit precision of 4-bit, with a recognition accuracy reaching 94.4%.
Authors
- Yung‐Jr Hung (ORCID: https://orcid.org/0000-0002-2132-2135)
- Chua‐Chin Wang (ORCID: https://orcid.org/0000-0002-2426-2879)
- Pao-Chia Shih
- Po-Hsiang Huang
- Chewn-Pu Jou
- Z. Chang
Institutions
- National Sun Yat-sen University (TW)
- Taiwan Semiconductor Manufacturing Company (Taiwan) (TW)
Publication Details
- Published
- 2025-04-14
- DOI
- https://doi.org/10.1109/siphotonics64386.2025.10985669
- Primary Topic
- Photonic and Optical Devices
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Synopsys