mlsanityData to 7 Oct 2026

Sub-microsecond Transformers for Jet Tagging on FPGAs

arXiv 2510.24784 ↗First posted 26 Oct 2025physics.ins-det cs.LG cs.PFcited by 6

Not on the map. Why?

Abstract (arXiv)

We present the first sub-microsecond transformer implementation on an FPGA achieving competitive performance for state-of-the-art high-energy physics benchmarks. Transformers have shown exceptional performance on multiple tasks in modern machine learning applications, including jet tagging at the CERN Large Hadron Collider (LHC). However, their computational complexity prohibits use in real-time applications, such as the hardware trigger system of the collider experiments up until now. In this work, we demonstrate the first application of transformers for jet tagging on FPGAs, achieving O(100) nanosecond latency with superior performance compared to alternative baseline models. We leverage high-granularity quantization and distributed arithmetic optimization to fit the entire transformer model on a single FPGA, achieving the required throughput and latency. Furthermore, we add multi-head attention and linear attention support to hls4ml, making our work accessible to the broader fast machine learning community. This work advances the next-generation trigger systems for the High Luminosity LHC, enabling the use of transformers for real-time applications in high-energy physics and beyond.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Builds on 16 of its 35 references that are in the map. First posted on arXiv 26 Oct 2025.

Builds on

16 of 35 references.

References · 16

  1. 2017Attention Is All You Needcited by 59,932
  2. 2020Linformer: Self-Attention with Linear Complexitycited by 1,462
  3. 2022Particle Transformer for Jet Taggingcited by 95
  4. 2020Automatic heterogeneous quantization of deep neural networks for low-latency inference on the edge for particle detectorscited by 78
  5. 2021hls4ml: An Open-Source Codesign Workflow to Empower Scientific Low-Power Machine Learning Devicescited by 61
  6. 2019AutoQ: Automated Kernel-Wise Neural Network Quantizationcited by 60
  7. 2020Compressing deep neural networks on FPGAs to binary and ternary precision with HLS4MLcited by 42
  8. 2024Lorentz-Equivariant Geometric Algebra Transformers for High-Energy Physicscited by 36
  9. 2023Explainable Equivariant Neural Networks for Particle Physics: PELICANcited by 22
  10. 2022LL-GNN: Low Latency Graph Neural Networks on FPGAs for High Energy Physicscited by 12
  11. 2024Ultrafast jet classification on FPGAs for the HL-LHCcited by 12
  12. 2025Fast Jet Tagging with MLP-Mixers on FPGAscited by 9
4 more
  1. 2025da4ml: Distributed Arithmetic for Real-time Neural Networks on FPGAscited by 8
  2. 2024Ultra Fast Transformers on FPGAs for Particle Physics Experimentscited by 6
  3. 2023Fast Neural Network Inference on FPGAs for Triggering on Long-Lived Particles at Colliderscited by 5
  4. 2025Neural Architecture Codesign for Fast Physics Applicationscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 1; 2026: 5.

6 papers.

  1. 2025hls4ml: A Flexible, Open-Source Platform for Deep Learning Acceleration on Reconfigurable Hardwarecited by 8
  2. 2026JetFormer: A Scalable and Efficient Transformer for Jet Tagging from Offline Analysis to FPGA Triggerscited by 3
  3. 2026Towards Tensor Network Models for Low-Latency Jet Tagging on FPGAscited by 2
  4. 2026HGQ-LUT: Fast LUT-Aware Training and Efficient Architectures for DNN Inferencecited by 1
  5. 2026Reconfigurable Computing Challenge: Transformer for Jet Tagging on Versal AI Enginescited by 0
  6. 2026Machine learning for the LHC physics program: a 2025-2026 stocktakecited by 0