SMuK 2023 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 33: DAQ NN/ML – GRID I

T 33.1: Vortrag

Dienstag, 21. März 2023, 17:00–17:15, HSZ/0301

Track reconstruction with Graph Neural Networks on FPGAs for the ATLAS Event Filter at the HL-LHC — Sebastian Dittmeier and •Sachin Gupta — Physikalisches Institut, Universität Heidelberg

The High-Luminosity LHC (HL-LHC) will enhance the potential to discover new physics with the ATLAS experiment beyond its reach at the LHC. To cope with the increased pile-up foreseen during the HL-LHC, major upgrades to the ATLAS detector and trigger system are required. The trigger system will consist of a hardware-based trigger and an online server farm, called the Event Filter (EF), with track reconstruction capabilities. For the EF, a heterogeneous computing farm consisting of CPUs and potentially GPUs and/or FPGAs is under study, together with the use of modern machine learning algorithms such as Graph Neural Networks (GNNs).

GNNs are a powerful class of geometric deep learning methods for modelling spatial dependencies via message passing over graphs. They are well-suited for track reconstruction tasks by learning on an expressive structured graph representation of hit data. A considerable speed-up over CPU-based execution is possible on FPGAs.

In this talk, a study of track reconstruction for the ATLAS EF system at HL-LHC using GNNs on FPGAs is presented. The main focus is set on model size minimization using quantization aware training, as resource utilization is a key aspect in the application of GNNs on FPGAs.