Reconfigurable Computing Challenge: Transformer for Jet Tagging on Versal AI Engines

2026-06-16

AI Summary

Deploying high-performance Transformer models in the low-latency, resource-constrained trigger systems of CERN’s Large Hadron Collider presents significant challenges. This work proposes a reusable software framework that abstracts quantized, integer-only Transformers into composable building blocks tailored for the AMD Versal AI Engine, enabling efficient mapping of dense layers and multi-head attention mechanisms. The framework supports automatic generation of Vitis graph code from high-level Python models, facilitating the initial deployment of a Jet Tagging Transformer on the AI Engine. By releasing this extensible framework as open-source, the project fosters deeper integration between high-energy physics and reconfigurable computing.

Abstract

Transformer-based models achieve strong performance for jet tagging at the CERN LHC, but deploying them in low-latency, resource-constrained trigger systems is challenging. We present an initial implementation of a quantized, integer-only transformer for jet tagging on the AMD Versal AI Engine (AIE), mapping dense and multi-head attention (MHA) layers to AIE tiles. The main contribution is a reusable software framework that represents transformer layers as composable AIE building blocks and automatically generates the corresponding Vitis graph code from a high-level Python model description. This framework provides a foundation for future research and is released as open-source software at https://github.com/KastnerRG/particle_transformer_aie.

Problem

• Reconfigurable Computing
• Transformer
• Jet Tagging
• Low-latency Deployment
• Resource-constrained Systems

Innovation

• Reconfigurable Computing
• Transformer
• Jet Tagging
• Versal AI Engine
• Quantized Neural Networks