Sub-microsecond Transformers for Jet Tagging on FPGAs

We present the first sub-microsecond transformer implementation on an FPGA achieving competitive performance in 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 makes them prohibitively expensive for real-time applications, such as the Level-1 trigger system of the collider experiments. In this work, we demonstrate the first application of transformers for jet tagging on FPGAs, achieving $\mathcal{O}(100)$ nanosecond latency with superior performance compared to the 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 (Linformer) support to hls4ml, making our work accessible to the broader fast machine learning community. This work advances the next-generation trigger systems for the HL-LHC, enabling the use of transformers for real-time applications in high-energy physics and beyond.