Accelerated Charged Particle Tracking with Graph Neural Networks on FPGAs

TL;DR

This paper presents FPGA implementations of graph neural network algorithms for charged particle tracking, achieving significant speedups over CPU methods, with potential applications in high-energy physics experiments like CERN.

Contribution

It introduces two FPGA-based designs for particle tracking using GNNs, leveraging OpenCL and hls4ml frameworks, and compares their performance on benchmark datasets.

Findings

Significant speedup over CPU-based tracking algorithms

Feasible FPGA implementation for real-time particle tracking

Potential application in CERN's Level-1 trigger system

Abstract

We develop and study FPGA implementations of algorithms for charged particle tracking based on graph neural networks. The two complementary FPGA designs are based on OpenCL, a framework for writing programs that execute across heterogeneous platforms, and hls4ml, a high-level-synthesis-based compiler for neural network to firmware conversion. We evaluate and compare the resource usage, latency, and tracking performance of our implementations based on a benchmark dataset. We find a considerable speedup over CPU-based execution is possible, potentially enabling such algorithms to be used effectively in future computing workflows and the FPGA-based Level-1 trigger at the CERN Large Hadron Collider.