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

T: Fachverband Teilchenphysik

T 41: Trigger+DAQ 1

T 41.7: Vortrag

Dienstag, 5. März 2024, 17:30–17:45, Geb. 30.23: 3/1

Improved Clustering with Graph Neural Networks on FPGAs for the Electromagnetic Calorimeter Trigger at Belle II — •Isabel Haide¹, Torben Ferber¹, and Marc Neu² — ¹Institute of Experimental Particle Physics (ETP), Karlsruhe Institute of Technology (KIT) — ²Institut fuer Technik der Informationsverarbeitung (ITIV), Karlsruhe Institute of Technology (KIT)

For the Belle II experiment, beam background plays a very impactful role, especially on the hardware trigger level. Due to the maximum latency of 1.3 µs the current trigger algorithm for the Belle II electromagnetic calorimeter uses a simple clustering mechanism that, especially in high beam background, identifies a high number of fake clusters and is additionally unable to separate overlapping clusters. As Belle II plans to increase its luminosity by a factor of 40, an update of the trigger algorithm will be necessary. In this talk, we will show the application of a Graph Neural Network in the form of the object condensation algorithm applied on the hardware trigger level of the electromagnetic calorimeter at Belle II. We will show an implementation of the machine learning algorithm on FPGA level, which is necessary to guarantee a fast execution time, and the evaluation on possible Dark Sector models which would be inaccessible with the current trigger algorithm.

Keywords: clustering; graph neural networks; trigger; FPGA; calorimeter