Sub-microsecond Transformers for Jet Tagging on FPGAs
Lauri Laatu 1 , Chang Sun 2 , Arianna Cox 1 , Abhijith Gandrakota 3 , Benedikt Maier 1 , Jennifer Ngadiuba 3 , Zhiqiang Que 1 , Wayne Luk 1 , Maria Spiropulu 2 , and Alexander Tapper 1 1 Imperial College London, United Kingdom 2 California Institute of Technology, USA 3 Fermilab, USA · 2025
Read Sub-microsecond Transformers for Jet Tagging on FPGAs by Lauri Laatu 1 , Chang Sun 2 , Arianna Cox 1 , Abhijith Gandrakota 3 , Benedikt Maier 1 , Jennifer Ngadiuba 3 , Zhiqiang Que 1 , Wayne Luk 1 , Maria Spiropulu 2 , and Alexander Tapper 1 1 Imperial College London, United Kingdom 2 California Institute of Technology, USA 3 Fermilab, USA (2025) 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-2510-24784.txt.
Description
Lauri Laatu leads a team with nine co-authors who put a transformer model on a fast chip for use at a giant particle collider. Such models tag jets, sprays of parts from a hit, very well, but they are too slow for the snap choice stage of an event. Their chip run breaks the one microsecond mark, a first for this model type.
Fine cuts to number size and a math trick let the whole model fit on one chip. The team also added support for key model parts to a known open tool, so peers can reuse it. Scores matched top past tools, at a span of about a hundred billionths of a second.
Catalog record
| CREATOR | Lauri Laatu 1 , Chang Sun 2 , Arianna Cox 1 , Abhijith Gandrakota 3 , Benedikt Maier 1 , Jennifer Ngadiuba 3 , Zhiqiang Que 1 , Wayne Luk 1 , Maria Spiropulu 2 , and Alexander Tapper 1 1 Imperial College London, United Kingdom 2 California Institute of Technology, USA 3 Fermilab, USA |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 13 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-24784