Emulating Germanium Detector Pulse Shape Parameters with Machine Learning

LEGEND searches for the rare lepton number violating process of Neutrinoless Double Beta Decay (0νββ) in ⁷⁶Ge using High-Purity Germanium (HPGe) detectors. Background rejection is critical for the potential discovery of 0νββ. In particular, Pulse Shape Discrimination (PSD) is used to distinguish signal and background events based on their event topologies in the HPGe detectors. Accurately modeling the performance of these parameters is important for precise background modeling in LEGEND-200 and in setting radiopurity requirements in LEGEND-1000, but first-principles Pulse Shape Simulation (PSS) in HPGe is computationally demanding. This talk presents an alternative machine learning-based approach using an Implicit Quantile Network (IQN) to emulate the PSD parameter used for multi-site gamma event rejection. Using first-principles PSS, we successfully train the IQN model to accurately predict the distribution of this PSD parameter from particle interaction simulations. This network can serve as a fast and accurate emulator for multi-site event rejection, and can be used for background modeling.