Erlangen 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 92: Electronics, Trigger, DAQ IV

T 92.3: Talk

Friday, March 20, 2026, 09:30–09:45, KH 00.023

Graph Neural Network based Algorithms for the Belle II Upgrade of the Electromagnetic Calorimeter Trigger on Versal SoCs with integrated AI Engines — •Thomas Lobmaier, Isabel Haide, and Torben Ferber — Institute of Experimental Particle Physics, Karlsruhe Institute of Technology, Karlsruhe, Germany

In order to reach the target design luminosity of Belle II, the instantaneous luminosity has to be increased by adjustments of the SuperKEKB accelerator. Especially with the planned long shutdown II in 2030 a significant gain in luminosity is expected, which directly increases trigger rates.

Planned upgrades for the electromagnetic calorimeter electronics increase the readout granularity, which also allows the subsequent trigger algorithms to access higher granularity inputs. In addition to the associated better energy and position resolutions, this enables the utilization of shower shape information to reconstruct the event more accurately and improves background suppression.

We show a first implementation of a GNN on Versal SoCs with integrated AI Engines, which enables the processing of up to 256 inputs per event. We present the performance on datasets with different input reduction strategies for the Belle II Long Shutdown 2 upgrade.

Keywords: Trigger; Belle II; electromagnetic calorimeter; Graph Neural Network (GNN)