Event classification close to the noise threshold for HPGe detectors with machine learning

Coherent elastic neutrino nucleus scattering (CEvNS) describes the interaction of the neutrino with a nucleus as a whole. The interaction is coherent for neutrino energies below about 50 MeV as provided for example by the Spallation Neutron Source (SNS) at Oak Ridge National Laboratory. The COHERENT experiment located there measures CEvNS with a multitude of detector technologies with includes Ge-Mini, an array of eight high-purity germanium spectrometers (HPGe).

The signature of the interaction is a keV to sub-keV nuclear recoil; thus low noise thresholds in the keV ionization energy range and a successful rejection of noise events are crucial. For the first CEvNS detection on Ge achieved last year an analysis threshold of 1.5 keV ionization energy was used. In my talk, I will present a machine learning tool to quickly identify noise events, extract information to potentially find their origin, and further reduce the analysis threshold. Moreover, I will outline a trigger algorithm design to use machine learning classification to avoid triggering on these noise events while still reliably identifying signals.