Light-Bound Transformers: Hardware-Anchored Robustness for Silicon-Photonic Computer Vision Systems
Xuming Chen ∗,1 , Deniz Najafi ∗,2 , Chengwei Zhou 1 , Pietro Mercati 3 , Arman Roohi 4 Mohsen Imani 5 , Mahdi Nikdast 6 , Shaahin Angizi 2 , and Gourav Datta 1 1 Case Western Reserve University, USA 2 New Jersey Institute of Technology, USA 3 Intel Corporation, USA 4 University of Illinois Chicago, USA 5 University of California, Irvine, USA 6 Colorado State University, USA ∗ Equal Contributions · 2026
Read Light-Bound Transformers: Hardware-Anchored Robustness for Silicon-Photonic Computer Vision Systems by Xuming Chen ∗,1 , Deniz Najafi ∗,2 , Chengwei Zhou 1 , Pietro Mercati 3 , Arman Roohi 4 Mohsen Imani 5 , Mahdi Nikdast 6 , Shaahin Angizi 2 , and Gourav Datta 1 1 Case Western Reserve University, USA 2 New Jersey Institute of Technology, USA 3 Intel Corporation, USA 4 University of Illinois Chicago, USA 5 University of California, Irvine, USA 6 Colorado State University, USA ∗ Equal Contributions (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-04330.txt.
Description
Chen and colleagues build a hardware-aware training pipeline for vision transformers on silicon-photonic near-sensor accelerators. They first measure noise in microring-resonator banks, covering fabrication variation, thermal drift and amplitude noise, then turn those readings into variance proxies. Their Chance-Constrained Training enforces logit margins to limit attention rank flips, joined by a noise-aware LayerNorm that steadies feature statistics without touching the optical schedule. Hardware tests with photonic banks restore near-clean accuracy under realistic noise, with no on-site learning or extra optical operations.
Catalog record
| CREATOR | Xuming Chen ∗,1 , Deniz Najafi ∗,2 , Chengwei Zhou 1 , Pietro Mercati 3 , Arman Roohi 4 Mohsen Imani 5 , Mahdi Nikdast 6 , Shaahin Angizi 2 , and Gourav Datta 1 1 Case Western Reserve University, USA 2 New Jersey Institute of Technology, USA 3 Intel Corporation, USA 4 University of Illinois Chicago, USA 5 University of California, Irvine, USA 6 Colorado State University, USA ∗ Equal Contributions |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38976 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.04330