Speaker
Description
Suppressing α background in HPGe detectors for next-generation 0νββ searches
A. Biondi$^1$, G. Zuzel$^1$
1) M. Smoluchowski Institute of Physics, Jagiellonian University, Krakow, Poland
High-purity germanium detectors enriched in $^{76}\mathrm{Ge}$ are among the most sensitive technologies for neutrinoless double beta decay searches, thanks to their excellent energy resolution, low intrinsic radioactivity, and use as both source and detector. In experiments such as LEGEND, pulse shape discrimination is essential to suppress background multi-site events. Surface $\alpha$ contamination is a more challenging background: $\alpha$ emitters as $^{210}\mathrm{Po}$ may produce events near the region of interest if located on the thin $p+$ detector contact. However, the number of $\alpha$ events expected in low-background experiments is too small to train dedicated classifiers, while still being relevant for the final background budget.
In this contribution, I will present a study of $\alpha$-event rejection in a BEGe-type high-purity germanium detector using pulse shape discrimination. The goal is to test whether classifiers trained on $\gamma$ calibration data can reject surface $\alpha$ events without using any dedicated training procedure.
To test this approach, the $p+$ surface of a point-contact semi-planar germanium detector was exposed to $^{209}\mathrm{Po}$ and $^{210}\mathrm{Po}$ sources deposited on thin gold foils. Two dedicated measurement campaigns were performed, yielding $1.36\times10^{5}$ and $1.87\times10^{6}$ $\alpha$ events, respectively. The classifiers were trained using selected single-site and multi-site dominated regions of a $^{228}\mathrm{Th}$ calibration spectrum. Two machine-learning methods were investigated: a multilayer perceptron and a projective likelihood classifier, with the standard $A/E$ method used as a benchmark.
Using these dedicated datasets, the response of the classifiers was evaluated. Both machine-learning methods efficiently separate single-site and multi-site $\gamma$ events while strongly reducing the $\alpha$ component. The multilayer perceptron provides the best overall performance, with a signal-like event survival greater than 80%, a background-like event survival below 20%, and an $\alpha$-rejection factor exceeding $2.71\times10^{4}$. These results demonstrate that robust pulse shape discrimination against both $\gamma$ and surface $\alpha$ events can be achieved using training information derived solely from $\gamma$-ray events.