Charged Particle Track Finding and Fitting with Neural Networks on Heterogeneous Computing Systems

Particle tracking, the most computationally demanding step in event reconstruction, is key to dealing with the challenge of higher pile-up for the High-Luminosity LHC (HL-LHC) upgrades of ATLAS. We present a neural network (NN) approach to track finding and fitting to be used in the ATLAS track trigger pipeline. Our algorithm is designed for potential implementation on heterogenous acceleration hardware such as FPGA or GPU so that it can meet the high-throughput requirements of track finding, while maintaining room for precision measurements downstream.