For the HL-LHC upgrade of the ATLAS TDAQ system, a heterogeneous computing farm deploying GPUs and/or FPGAs is considered to be used for the Event Filter system, together with the use of modern machine learning algorithms such as Graph Neural Networks (GNNs) to solve computationally complex tasks within that system. In this study, the development of a GNN based track finding pipeline on FPGAs for the ATLAS Inner Tracker is presented as part of the Event Filter system. Each step in the GNN-based tracking pipeline is explored: graph construction, edge classification using an interaction network, and segmentation of the graph into track candidates. Optimizations of the GNN approach are investigated to minimize FPGA resource utilization and maximize throughput while maintaining high track reconstruction efficiency and low fake rates required for the ATLAS Event Filter tracking system. These optimizations include model hyperparameter tuning, model pruning, quantization-aware training, and sequential processing of sub-graphs across the detector.
S. J. Dittmeier (Wed,) studied this question.