Optimizing the energy threshold of light detectors coupled to luminescent bolometers

Bolometers have proven to be good detectors for the search of neutrinoless double beta decay. By operating at cryogenic temperatures, they feature excellent energy resolution and low background. The detection of the possible light emitted when particles interact in the bolometer is a promising method to lower the background of the experiments. The different amount of light emitted in beta/gamma and alpha interactions, whether due to scintillation or Cerenkov emission, allows to discriminate the two interaction types. Because of the cryogenic environment, light detectors are often bolometers. In this work we present a software algorithm to lower the energy threshold of bolometric light detectors coupled to luminescent bolometers. The application to data from Ge light detectors coupled to ZnMoO4 and TeO2 bolometers shows that the energy threshold can be lowered substantially, increasing the discrimination power when the amount of emitted light is small.

💡 Research Summary

The paper addresses a critical limitation in the use of bolometric light detectors (LDs) for neutrinoless double‑beta decay (0νDBD) experiments: the relatively high energy threshold of the LDs, which hampers the discrimination between β/γ and α interactions when the emitted light is very small. The authors develop a software‑based algorithm that lowers this threshold by exploiting the fixed time relationship between the heat signal (HD) of the main bolometer and the light signal recorded by the LD.

Experimental data were collected at the Gran Sasso National Laboratory using two different scintillating or Cerenkov‑emitting crystals: a 29.9 g ZnMoO₄ crystal and a 116.7 g TeO₂ crystal, each coupled to a germanium LD (36 mm × 1 mm for ZnMoO₄, 66 mm × 1 mm for TeO₂). Both bolometers were equipped with NTD‑Ge thermistors and operated at ~13 mK. The LDs were calibrated with a 55Fe source (5.9 keV and 6.5 keV X‑rays). Signals were digitized at 2 kHz, filtered with a six‑pole Bessel anti‑aliasing filter, and then processed offline with an optimum filter that maximizes signal‑to‑noise ratio.

In the conventional analysis, the amplitude of the LD signal is taken as the maximum of the filtered waveform. This works well for high‑energy events where the light yield is well above the noise, but fails for low‑energy events (below ~300 keV) because the maximum is dominated by random noise fluctuations. Consequently, the light‑versus‑heat scatter plot shows a flat “pedestal” at ~250 eV, erasing the distinction between β/γ and α populations.

The new method proceeds as follows: (1) select a clean set of high‑energy β/γ events where both HD and LD signals are clearly visible; (2) measure the time difference Δt between the HD maximum and the LD maximum for each event; (3) determine a robust estimator of Δt (the mode of the distribution) to avoid outliers; (4) for every event, regardless of its energy, take the LD amplitude not from its own maximum but from the filtered LD waveform at a fixed offset Δt after the HD maximum. This effectively samples the LD signal at the expected arrival time, even when the signal is buried in noise.

Applying this to the ZnMoO₄ data, the authors find Δt ≈ ‑8.96 ms (LD responds faster than HD). The revised light‑energy distribution for low‑energy β/γ events (E < 20 keV) becomes Gaussian, with a 50 % containment threshold of 19 eV, compared with 249 eV using the traditional method. The 90 % containment improves from 382 eV to 125 eV, demonstrating a dramatic reduction of the effective energy threshold. The method also yields negative light‑energy values for some events, which is physically acceptable because baseline temperature fluctuations can be positive or negative in a thermal detector.

For the TeO₂ crystal, which emits only a tiny amount of Cerenkov light, the conventional maximum‑search fails entirely: the LD waveform shows no discernible peak at any energy. Nevertheless, using the same Δt derived from ZnMoO₄ (or a similar procedure) the authors are able to extract a statistically meaningful LD amplitude, marking the first observation of Cerenkov light in this configuration. This opens the possibility of α‑background rejection in pure TeO₂ bolometers, a crucial step for future CUORE‑like experiments.

The algorithm is straightforward to implement in existing data‑processing pipelines, requires no hardware modifications, and adds negligible computational overhead. It is robust against outliers because the mode of the Δt distribution is used, and the small variation of Δt with energy has been shown to have an insignificant impact on the final LD amplitude.

In summary, the paper presents a practical, software‑only solution that lowers the LD energy threshold from a few hundred electronvolts to a few tens of electronvolts. This improvement restores the ability to discriminate β/γ from α events even when the scintillation or Cerenkov yield is extremely low, thereby enhancing background suppression in next‑generation 0νDBD searches. The technique is especially valuable for crystals with modest scintillation yields (e.g., ZnMoO₄) or those relying solely on Cerenkov emission (e.g., TeO₂), and it can be readily adopted by other bolometric experiments seeking to push their sensitivity to the inverted‑hierarchy region of neutrino masses.