Bayesian inference of high-purity germanium detector impurities based on capacitance measurements and machine-learning accelerated capacitance calculations

Abstract The impurity density in high-purity germanium detectors is crucial to understand and simulate such detectors. However, the information about impurities provided by manufacturer, based on Hall effect measurements, typically limited a few locations comes with large uncertainty. As voltage dependence of capacitance matrix detector strongly depends distribution, measurements can provide path improve knowledge impurities. novel method presented here uses machine-learned surrogate model, trained precise GPU-accelerated calculations, perform full Bayesian inference distribution parameters from measurements. All steps use open-source Julia software packages. Capacitances are calculated SolidStateDetectors.jl , machine learning done Flux.jl performed using BAT.jl . its explained bias-voltage scan an n -type true-coaxial test presented. study indicates that also has radial dependence.