Enriched TeO$_2$ bolometers with active particle discrimination: towards the CUPID experiment

We present the performances of two 92% enriched $^{130}$TeO$_2$ crystals operated as thermal bolometers in view of a next generation experiment to search for neutrinoless double beta decay of $^{130}$Te. The crystals, 435 g each, show an energy resolution, evaluated at the 2615 keV $\gamma$-line of $^{208}$Tl, of 6.5 and 4.3 keV FWHM. The only observable internal radioactive contamination arises from $^{238}$U (15 and 8 $\mu$Bq/kg, respectively). The internal activity of the most problematic nuclei for neutrinoless double beta decay, $^{226}$Ra and $^{228}$Th, are both evaluated as $<$3.1 $\mu$Bq/kg for one crystal and $<$2.3 $\mu$Bq/kg for the second. Thanks to the readout of the weak Cherenkov light emitted by $\beta/\gamma$ particles by means of Neganov-Luke bolometric light detectors we were able to perform an event-by-event identification of $\beta/\gamma$ events with a 95% acceptance level, while establishing a rejection factor of 98.21% and 99.99% for $\alpha$ particles.


Introduction
The following three important questions in neutrino physics will be addressed by next generation neutrinoless double beta decay (0ν-DBD) experiments: are neutrinos Majorana particles that differ from antineutrinos only by helicity? Is lepton number conservation violated? What is the neutrino mass-scale? Searches for 0ν-DBD have been carried out for many decades investigating a large variety of nuclei with many different experimental techniques [1]. However the discovery of the atmospheric neutrinos os-* Corresponding author. cillations by Super-Kamiokande as well as those observed in solar neutrinos by the SNO experiment -both awarded the Nobel Prize in 2015 -boosted these searches, and now is an optimum time to launch next generation 0ν-DBD experiments. Recent analyses of all of the atmospheric, solar, and reactor neutrino oscillations [2] indicate that there exist scenarios in which the effective Majorana mass of the electron neutrino could be larger than 0.05 eV. Within the last few years, the most stringent limits on 0ν-DBD came from EXO-200 [3], GERDA [4] and CUORE-0 [5] while, very recently, the KamLAND-Zen experiment [6] set the strongest limit on this decay, using 136 Xe. While recent experimental achievements are impressive, it is difficult to compare results from different isotopes because of the large uncertainties in the nuclear matrix elements. Ultimately, the goal of next the generation 0ν-DBD experiments is to sensitively probe the entire inverted hierarchy region. To reach this sensitivity, the total masses of parent isotopes must be increased using enriched isotopes, and the backgrounds drastically decreased. A ton-scale 76 Ge experiment (with a possible common effort between GERDA and MAJORANA) [7], the full EXO experiment (nEXO [8]), and a possible upgrade of the CUORE experiment (CUPID, CUORE Upgrade with Particle IDentification [9,10]) are all designed to achieve this goal.
The advantage of the bolometric technique proposed for CU-PID is not only the possibility to choose different DBD emitters combined with the capability of having a high resolution detector, but the realization of double-readout detectors in order to perform an active particle discrimination to reject the natural background. CUPID is presently in an R&D phase testing different type of crystals containing most of the interesting DBD emitters ( 82 Se, 100 Mo, 116 Cd, 130 Te). The aim of CUPID, which will use the CUORE infrastructure once that experiment has concluded, is to increase the sensitivity to completely cover the inverted hierarchy region. In order to reach this goal, two major scientific milestones need to be reached: 1. increase the number of active DBD nuclei through development of bolometers made of enriched isotopes (as the experimental volume of the CUORE cryostat is fixed); 2. decrease the present natural radioactive background by two orders of magnitude by rejecting the major source of background for DBD bolometers due to α-particle interactions [11].
The initial idea to decrease the α-background in DBD bolometers was to use scintillating crystals [12] in which the discrimination between e/γ and α/neutron particles can be simply obtained with the additional readout of the scintillation light, through a second -very sensitive -bolometer working as a Light Detector (LD). Rather recently, however, after the observation of a very tiny light signal in a small TeO 2 bolometer [13], it was suggested [14] that particle discrimination could be obtained also in non-scintillating crystal bolometers (like TeO 2 ) by exploiting the Cherenkov light emission. Heavy α particles arising from natural radioactivity have velocities far below the threshold to emit Cherenkov photons in any kind of crystal. In the last four years several tests were performed on large [15][16][17] and very small [18,19] TeO 2 crystal samples coupled with different types of bolometric LDs. The challenge of this method is the detection of the extremely small amount of light emitted by electrons at the 0ν-DBD energy of 130 Te (2.53 MeV) that is of the order of ≈100 eV [20].
In this work we present for the first time the performance of large enriched TeO 2 in which the Cherenkov light is used for particle identification. This work demonstrates that 130 TeO 2 can be a suitable candidate for the CUPID experiment in terms of energy resolution, internal radioactive contaminations and α-background discrimination.

Enriched crystal growth
The 130 TeO 2 crystals used in this work were manufactured starting from enriched 130 Te in the form of metal powder, purchased from JSC Isotope, Russia.
The purity of the enriched material was certified as >99.9875%.
The concentrations of the most troublesome metallic impurities were measured independently by ICP-MS and were found to be below 1 ppm, except for Fe (1.5 ppm), Cu (3.5 ppm) and Al (4.5 ppm). Radioactive 238 U and 232 Th were not observed, with a detection limit of the order of 5 ppt. The isotopic abundances of Tellurium in the powder, as measured by ICP-MS, are given in Table 1 and Table 1 Concentration of the most abundant Tellurium isotopes in the metal used for the production of the crystals in this work. The errors on the measurements are of the order of 0.5% for the first two rows, and of the order of 10% for the other three. compared to the values reported by the vendor and the isotopic concentration of natural Tellurium [21]. The 130 TeO 2 crystals were manufactured by Shanghai Institute of Ceramics of the Chinese Academy of Sciences (SICCAS), P.R. China, following basically the same technology as the one applied for the production of CUORE crystals [22]. Some specific procedures were applied though, in order to reduce the material losses which resulted in a 130 TeO 2 powder synthesis efficiency ≥80%, and a crystal growth efficiency ≥90%, meaning an irrecoverable loss of 28%. 1 As shown in Table 2, the synthesis of the 130 TeO 2 powder acts as a purification process reducing most of the metallic impurities to a concentration below 1 ppm. A dedicated furnace and crucible system were built in order to cope with the relatively small amount of 130 TeO 2 available and a single-growth cycle was applied instead of the double-growth process used for the production of CUORE crystals. This last point was adopted in order to decrease the amount of losses of the enriched material.
Also a dedicated temperature gradient and growth regime were applied in order to minimize any mass transfer between the seed (TeO 2 crystal with natural Te isotopic concentration) and the growing crystal. Preliminary tests were made using low enriched material as marker in order to make sure that the isotopic concentration of the feeding 130 TeO 2 powder remains unchanged in the grown crystal (a detailed description of enriched 130 TeO 2 crystal production will be given in a dedicated article). One single enriched crystal ingot was finally grown. Two crystals were produced out of the single ingot in order to study possible (radioactive) impurities segregation effects during crystal growth. The shapes of the two crystals were fixed by the requirement that the crystals be identical with the maximum total mass yield. Two 36×38×52 mm 3 435 g crystals were cut and processed (shaped, chemical etched and polished) in a dedicated clean room with special precautions aimed at preventing possible radio-contamination of samples. In order to maximize the Cherenkov light output, four of the surfaces were roughly ground while the other two (the hard faces) were polished to optical standards.

Experimental technique
Bolometers are very sensitive calorimeters operated at cryogenic temperatures. These solid-state detectors share with Ge diodes the capability of achieving excellent energy resolution (∼5 keV FWHM from several keV to several MeV) with sizeable active mass devices. Our bolometers apply the calorimetric (source=detector) approach for the detection of rare decays: the source isotope is part of the active mass of the detector. The latter consists of two elements: a single crystal that plays the role of the calorimetric mass, and a sensor that measures the amount of energy converted into heat in the crystal, converting the phonon signal into an electrical one.

Enriched TeO 2 bolometers
To operate a crystal as a bolometer it must be coupled with a suitable thermometer; in this work we use 3×3×1 mm 3 Neutron Transmutation Doped (NTD) Germanium thermistors [23], thermally coupled to the crystal via nine epoxy glue spots of ∼600 μm diameter and ∼50 μm height. The NTD is a resistive device made of semiconducting material, which converts temperature variations into resistance variations. When the thermistor is biased with a constant current, any resistance variation produces a voltage pulse that constitutes the signal. In addition a ∼300 k resistor, made of a heavily doped meander on a 3.5 mm 3 Silicon chip, is attached to each crystal and acts as a heater to stabilize the gain of the bolometer [24].
The crystal is held by means of four S-shaped PTFE supports mounted on Cu columns (see Fig. 1). These Teflon supports ensure that with the down-cooling of the set-up the crystal is clasped tighter (PTFE thermal contraction is one of the highest among different materials), to minimize heat-noise generated by frictions induced by the acoustical vibration of the cryogenic facility. In order to increase the light collection, the crystal is surrounded laterally and on the bottom part (with no direct thermal contact) by a plastic reflecting sheet (3M Vikuti T M ESR), while the LD is faced to the top part.

Neganov-Luke light detectors
The LDs developed for DBD scintillating bolometers experiments [25] generally consist of Ge wafers coupled to NTD Ge thermistors. The performances of these devices are well satisfactory to read out the scintillation light (few keV) while they are generally insufficient to separate α and β particles on an eventby-event basis in the Region of Interest (RoI) for 0ν-DBD of 130 Te, their RMS baseline (∼100 eV) being comparable with the weak Cherenkov light signal (even if values ∼30-50 eV RMS were recently obtained [26]).
The two LDs used in this work have essentially the same structure and materials, and especially the same temperature sensor, but their signal-to-noise ratio can be significantly improved by exploiting the Neganov-Luke effect [27,28]. This effect is based on the application of an electric field in the light-absorber volume. The work done by the field on the drifting charges (generated by the absorption of scintillation light) is converted into additional heat, which amplifies considerably the thermal signal provided by the NTD Ge thermistor. In our case, the field is generated through a set of concentric Al rings, electrically connected by means of ultrasonic wedge bonding with an alternate pattern. Between the Al electrode and the Ge crystal, a 50 nm-thick layer of hydrogenated amorphous Ge was deposited, in order to reduce possible (dark) current leakage.
This allows the application of a voltage drop (V grid ) between adjacent rings and the production of an electric field parallel to the surface. This ring structure enables increasing the collecting field for a given applied voltage and decreasing the charge trapping probability thanks to the short path length of the charges to the electrodes.
The two Ge LDs (named GeLuke and GeCo in the following) have <100> orientation, a diameter of 44 mm and a thickness of 0.17 mm; the maximum dislocation density (EPD) is <4000 counts/cm 2 and the impurity concentration is certified to be <2×10 10 at/cm 3 . Moreover a 70 nm-thick layer of SiO coating was deposited on one the electrode-equipped face of the GeCo detector in order to increase the light collection [29].
The two LDs were previously tested and characterized above ground at CSNSM in a dilution refrigerator dedicated to the development of luminescent bolometers [30]. In particular the amplification induced by the Neganov-Luke effect as a function of the voltage across the electrodes for a given light pulse was measured. The light was guided to the electrode surface of the Ge absorber through an optical fibre by a room temperature LED. The LED wavelength was 820 nm, in the near infrared, with a corresponding absorption length in Germanium of ∼0.2 μm. An example of the achieved amplification for both LDs is shown in Fig. 2, where we also report the ideal gain 1 + V grid / , where = 1.67 eV is the energy required to produce an electron-hole pair at 820 nm [31]. The comparison between the observed and expected amplification shows that large improvements are possible in the light detector by increasing the fraction of the collected charge. We remark that -for a fixed deposited energy -the amplification for Cherenkov absorption is expected to be lower than the amplification for LED light absorption, since the Cherenkov photons are distributed in the optical and near-ultraviolet frequency range. Therefore, the individual photon energy is higher with respect to LED excitation, decreasing the efficiency in producing electron-hole pairs. In fact, by comparing the LED data in Fig. 2 with the Cherenkov data in Table 5, a gain reduction by a factor ∼1.5 is observed in both LDs in the Cherenkov case.
The LDs are energy-calibrated with 5.9 keV and 6.5 keV photons from a 55 Fe X-ray source. This calibration method however works only when the detector is operated at zero bias on the electrodes, since the Neganov-Luke effect broadens and distorts the X-ray peaks [18]. However, it is possible to select a Cherenkov light signal corresponding to a well-defined γ /β energy measured in the TeO 2 absorber. As explained in Sec. 4.2, the energy associated to this light pulse can be determined at zero Neganov-Luke bias using the X-ray calibration. This energy value can then be used to calibrate the detector response under Neganov-Luke effect and evaluate the amplification for Cherenkov photons.

Experimental set-up
The detector, sketched in Fig. 1, was enclosed in a copper shield. A 880 nm-wavelength LED (Osram, SFH485P) was enclosed in the copper shield in order to neutralize ionized impurities that can build up due to charge trapping inside the LD. The LED was used to flash the detectors every 2-4 days depending whether a calibration or a background measurement was performed. The entire set-up was mounted in the CUPID R&D cryostat at Laboratori Nazionali del Gran Sasso. The cryostat consists of a 3 He/ 4 He -wet -dilution refrigerator (Oxford TL 200). The detector was cooled down to a temperature of ∼12 mK. A complete description of the set-up and the electronics can be found in [32][33][34].
The thermistors of the detectors are biased with a constant voltage through large (few G ) load resistors [35], resulting in a constant current operation. The resistance variations, generated by the temperature rise, are converted into voltage pulses read across the resistive sensors. The heat and light voltages are then amplified, filtered by a 6-pole Bessel-filter (with a cut-off frequency of 8 Hz for the 130 TeO 2 crystals and 120 Hz for the LDs) and finally fed into a NI PXI-6284 18-bits ADC.
The sampling rate of the ADC is 2 kHz for the 130 TeO 2 crystals and 4 kHz for the LDs. All triggers are software generated: when a trigger fires, for the main bolometer and the LD, waveforms 2.5 s and 0.25 s long are recorded. Moreover, when the trigger of the 130 TeO 2 crystal fires, the corresponding waveform from its LD is always recorded, irrespectively of its trigger. The amplitude and the shape of the voltage pulse is then determined by the off-line analysis. The pulse amplitude of the thermal signals are estimated by means of the Optimum Filter (OF) technique [36,37], that maximizes the signal-to-noise ratio in a way that improves the energy resolution and lowers the threshold of the detector. The amplitude of the light signals, however, is evaluated from the filtered waveforms at a fixed time delay with respect to the 130 TeO 2 bolometer, as described in detail in [38].
The amplitude of the acquired heat spectrum is energycalibrated using several γ -ray peaks arising from the calibration with 228 Th and, if present, known high-energy α-lines. The energy scale is linearised with a second order polynomial function in log(V ) with zero intercept, where V is the heat pulse ampli- tude. The LDs, on the contrary, are calibrated thanks to the 5.9 and 6.5 keV lines produced by the permanent 55 Fe X-ray sources faced to the detectors (see Fig. 1). The data sets analysed here consist of a first background campaign, in which only the two 130 TeO 2 crystals were acquired, and several calibration runs to optimize the α vs. β/γ rejection power by changing the working conditions of the LDs and cryostat operation.

130 TeO 2 performances
The crystals were operated at a temperature of ∼13 mK. The most important parameters of the bolometers are shown in Table 3. In Fig. 3 we present the calibration spectra of the two 130 TeO 2 crystals: the energy resolution close to the 0ν-DBD RoI of 130 Te, evaluated with the 2615 keV line of 208 Tl, is 4.3 and 6.5 keV FWHM, respectively. These values are fully compatible with the ones obtained in the CUORE-0 experiment [40], performed with natural TeO 2 crystals.
Background measurements were performed to assess the internal contaminations of the enriched crystals. This, together with the energy resolution, was the most important issue for this work, since previous experience with enriched TeO 2 crystals did not give satisfactory results for either energy resolution [41] or for internal background [42]. Since, after the run was ended, the analysis gave only limits on the most worrisome internal contaminants, we decided to use also the 228 Th calibration runs in order to increase our sensitivity in the high energy α-region. The final spectra of the α-region of the two 130 TeO 2 crystals is presented in Fig. 4. The only peaks appearing in the spectrum correspond to a clear signal due to 210 Po and to a tiny contamination of 238 U. 210 Po is a very well-known contamination of TeO 2 crystals and is present (both internally and on the surface) in all the TeO 2 crystals produced for CUORE [43]. Due to the relatively short decay time, this isotope does not represent a problem for a DBD search. 238 U, on the contrary, was not observed in the CUORE natural crystals, with a detection limit of 5×10 −14 g/g (corresponding to 0.6 μBq/kg). No other contaminations are visible in the spectra. The results are presented in Table 4. To obtain the limits, we defined the signal as the number of events falling in the energy region within   internal contaminations. This search, performed within a 4 τ decay time interval -with respect to each specific decay -, did not give evidences of any events.

Light detector performance
Most of the calibrations with the 228 Th source were performed to study and optimize the Neganov-Luke amplified LDs. The first step, the energy calibration of the light signal, was a 45 h long measurement with the 228 Th source, with V grid set to zero. Despite the fact that the RMS baseline at V grid = 0 is of the same order of the signal, the high statistical value of the data and the use of the light-synchronization [38] allowed the mean energy of the Cherenkov light from the 2615 keV γ -rays to be accurately converted to an energy value using the 55 Fe X-ray calibration.  The obtained values for the absolute Cherenkov light signals are 153±4 eV and 160±5 eV for 130 TeO 2 -1 and 130 TeO 2 -2, respectively. These values are in good agreement with the ones obtained with natural TeO 2 crystals of similar size [20,17]. This monochromatic light signal, independent from any parameter, is then used to calibrate the gain of the detectors once a V grid is applied and the 55 Fe peaks are no longer clearly identifiable (due to the presence of V grid ). Several calibrations were performed with different values of V grid in order to evaluate the best signal-to-noise ratio of the LDs. The best compromise between gain and noise was found at V grid = 25 V for GeLuke and V grid = 55 V for GeCo. The main parameters of the LDs are shown in Table 5.

Particle identification
In Fig. 5 (top) we show the Light vs. Energy scatter plot obtained with the two enriched crystals. There is a clear separation between the β/γ induced events and the α-events (mainly due to 210 Po). In order to evaluate the β/γ vs. α Discrimination Power (DP), we select the light signals belonging to the 2615 keV 208 Tl Table 5 Main parameters of the two LDs bolometers. The gain is evaluated as the amplitude of the Cherenkov light signal induced by the absorption of a 2615 keV γ quanta in the 130 TeO 2 at V grid = 25 V (GeLuke) and V grid = 55 V (GeCo), divided by the corresponding value obtained at V grid = 0 V. The last column, instead, represents the baseline noise obtained by applying the V grid voltage. The fact that RMS/RMS V grid < Gain clearly indicates the presence of excess noise induced by charge injection due to biasing the detector.
Using the fit values shown in Fig. 5, we get a DP = 2.65 for 130 TeO 2 -1 (GeLuke) and DP = 3.5 for the 130 TeO 2 -2 (GeCo), respectively. Choosing an acceptance level of 95% for the e/γ signal, we get α rejection factors of 98.21% and 99.99%, respectively. In Table 6 we compare this result with the ones obtained with other kind LD's and TeO 2 crystals of different sizes. For the sake of completeness, two important remarks should be made. The first regards the fact that to correctly evaluate the efficiency of a particle ID, the two classes of events should have the same energy, while in this case the light-signal distributions we compared were generated by 2615 keV γ s and 5407 keV αs.
However, since the α-particles do not emit photons (the mean energy of the two light distributions -as shown in Fig. 5 -is very Fig. 6. Typical Cherenkov light signals read out by the GeCo LD. Top: signals corresponding to ≈200 eV. Bottom: signals corresponding to ≈80 eV. The different colours simply distinguish between a pulse that can be eye-recognized (black) and a pulse that is "confused" within micro-phonic noise (blue). All signals are aligned at the maximum reconstructed amplitude of the pulse, 80 ms. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 6
A comparison of the DPs recently obtained with different types of LDs and crystals of rather different size. The LDs marked with * did not use NL amplification. We point out that, due to the relatively short absorption length of the Cherenkov light in TeO 2 [44], the amount of photons escaping a large crystal can be smaller with respect to the one escaping from a small crystal. At the same time, increasing the area (size) of a bolometric LD (to match a larger crystal) could worsen the energy threshold.  Fig. 5) is fully compatible with the RMS noise of the LDs, as given in Table 5.
The second point is that the Q ββ value of 130 Te is at 2528 keV, slightly less than 2615 keV. Since the Cherenkov light, at first level, is proportional to the energy, this would correspond to a slightly lower DP in the RoI. However, simulations show [44] that the Cherenkov light generated by two electrons (i.e. the 0ν-DBD sig- 2 In fact a small deviation from zero could be induced by a very small cross talk between light and heat channels or by a very weak scintillation light emitted by the TeO 2 crystal in addition to Cherenkov light, as the results in Ref. [13] seem to suggest. In any case, this behaviour slightly underestimates the real DP at lower energies. nal) sharing 2528 keV is statistically larger with respect to the one generated by a γ of the same energy. In fact this last effect overtakes the first one so that, actually, the evaluated DP at 2615 keV slightly underestimates the one at Q ββ for the 0ν-DBD detection.
Moreover, to have a more clear picture of the extremely small amplitude of the acquired Cherenkov light signals, in Fig. 6 we plot four pulses corresponding to two different energies. These pulses come from the tails (left and right) of the 208 Tl γ light distribution of Fig. 5 (bottom right). Thanks to the Optimum Filter technique [36,37] and, especially, to the trigger synchronization [38] signals that cannot be auto-triggered, give rise to a very powerful α vs. β/γ event-by-event identification.
Finally we would like to point out that we strongly believe that the energy resolution of our LDs can be further enhanced in two independent ways: i) by improving the intrinsic energy resolution of the "unbiased" detectors and ii) by decreasing the noise induced by charge injection due to biasing the detector. Actually the latest detectors we developed show a RMS median resolution <45 eV [26], while it is already demonstrated by the CDMSlite experiment [45] that it is possible to run NL amplified Ge detectors with a bias of the order of 70 V without any worsening of the baseline noise.

Conclusions
In this work we clearly demonstrated for the first time the possibility to operate large enriched 130 TeO 2 crystals to search for DBD with the bolometric technique. The obtained energy resolution is compatible with those obtained with natural crystals that will soon be operated in the CUORE experiment. The internal radioactive contaminations show only a very small contribution from 238 U. The corresponding secular equilibrium of the decay chain is broken and, finally, the internal activity of the most problematic nuclei for 0ν-DBD, both 226 Ra and 228 Th are evaluated as <3.1 (2.3) μBq/kg, respectively. Both the enriched crystals have the 226 Ra contamination limit that is within the specification of the CUORE crystals (<3.7 μBq/kg), while with respect to 228 Th, the obtained limits are very close to the specification (<1.2 μBq/kg). These values represent a very good starting point even if strong efforts are needed for a further reduction of the internal contaminations. Furthermore we finally demonstrated that the weak Cherenkov signal can be read with standard thermistor-based bolometers and used to discriminate the α induced background with a rejection factor more than 100, the goal of the CUPID project.