Hardware-Accelerated GNN Hit Filtering for the Belle II Level-1 Trigger

by Greta Heine

Thursday Nov 27, 2025, 2:00 PM → 3:00 PM Europe/Berlin

Seminarraum EG + zoom (IPE)

Greta Heine will present a real-time hit filtering system based on Graph Neural Networks (GNNs), implemented on FPGAs for the Level-1 trigger of Belle II. The system processes raw data from 14,336 sense wires with a sustained throughput of 32MHz and sub-microsecond latency. It combines GNN inference with static graph-building logic in a latency- and resource-optimized FPGA pipeline. This work demonstrates a scalable, low-latency architecture for detector-level background suppression, enabling efficient real-time data reduction in high-rate collider environments.

251127_ipe-seminar-intro.pdf

IPE_Seminar_GNNHitCleanup_Slides.pdf