Particle discrimination in a NaI crystal using the COSINUS remote TES design

The COSINUS direct dark matter experiment situated at Laboratori Nazionali del Gran Sasso in Italy is set to investigate the nature of the annually modulating signal detected by the DAMA/LIBRA experiment. COSINUS has already demonstrated that sodium iodide crystals can be operated at mK temperature as cryogenic scintillating calorimeters using transition edge sensors, despite the complication of handling a hygroscopic and low melting point material. With results from a new COSINUS prototype, we show that particle discrimination on an event-by-event basis in NaI is feasible using the dual-channel readout of both phonons and scintillation light. The detector was mounted in the novel remoTES design and operated in an above-ground facility for 9.06 g$\cdot$d of exposure. With a 3.7 g NaI crystal, e$^-$/$\gamma$ events could be clearly distinguished from nuclear recoils down to the nuclear recoil energy threshold of 15 keV.


I. INTRODUCTION
In the field of direct dark matter searches nullresults have been reported by most experiments [1] with the notable exception of DAMA/LIBRA [2].DAMA/LIBRA measures scintillation light created by particle interactions in NaI target crystals at room temperature.An annual modulation of the recorded event rate has been observed for many years, which is consistent with dark matter particle interactions, but incompatible with results from other direct searches in this interpretation [1].The origin of the signal remains unclear.Several experiments have set out to study this phenomenon using the same target material, and strong constraints on the modulation amplitude have been reported by COSINE-100 [3] and ANAIS [4], which follow a similar detection principle as DAMA/LIBRA.Among the NaI experiments, COSINUS (Cryogenic Observatory for SIgnals seen in Next generation Underground Searches) will be the only one to feature a direct measurement of the nuclear recoil energy per event.This is possible through the use of transition edge sensors (TES) which are coupled to the NaI target crystals to provide another channel in addition to the scintillation light.NaI poses certain difficulties when operated in this calorimetric approach, such as hygroscopicity and a low melting point [5].A solution to this problem is the re-moTES design, where the TES sensor is deposited on a separate wafer, which is then coupled to the absorber crystal using an Au-wire and pad [6].The first results of this coupling scheme for detectors with Si and TeO 2 absorbers were described in Ref. [6].We demonstrate in this work that the same principle is applicable to NaI crystals, and present results from the first NaI-remoTES detector.In particular, we show that discrimination between e − /γ events and nuclear recoils on an eventby-event basis is possible in NaI, which constitutes a milestone for COSINUS.

II. DETECTOR MODULE
The detector module consists of a remoTES phonon detector (cf.[6]) shown in Fig. 1 and 2a and a silicon-on-sapphire (SOS) wafer as light detector (cf.Fig. 3).The absorber is a (10×10×10) mm 3 NaI-crystal with a mass of 3.7 g, manufactured by the Shanghai Institute of Ceramics (SICCAS).An Au-foil, cut from an ingot to a thickness of 1 µm and an area of 4 mm 2 , was glued on the absorber with EPO-TEK 301-2, a two component low out-gassing epoxy resin [7].The residual resistivity ratio (RRR) of the Au-foil is about 22.The Au-foil was coupled to the TES wafer with two Au-wire bonds with a diameter of 17 µm each and lengths of 6.7 mm and 10.3 mm.A zoomed picture of the remoTES coupling to the absorber is shown in Fig. 2b.An ohmic heater, fabricated on a (3×3) mm 2 silicon pad with a thickness of 1 mm, was glued with EPO-TEK 301-2 on the surface of the NaI-absorber, and was used to inject heater pulses into the crystal.A 55 Fe X-ray source with an activity of 3 mBq was taped to the copper holder facing the NaI-absorber, and irradiated it from the side as indicated in Fig. 2a.The wafer is a (10×20×1) mm 3 Al 2 O 3 crystal, with an evaporated W-TES of (100×407) µm 2 in area and a thickness of 156 nm with two aluminum bonding pads for the connection to the bias circuit.The Auwires from the Au-foil are bonded on the Au-bridge which overlaps with the W-film (see Fig. 1).Another Au-stripe is used as a thermal link connecting the TES to the thermal bath; its resistance is about 82.3 Ω at room temperature.An Au-film with an area of (200×150) µm 2 and a thickness of 100 nm, deposited on the wafer surface, is used to inject heater pulses and thus monitor the temperature of the TES.
The light detector is a (20×20×0.4)mm 3 SOS wafer, equipped with a (284×423) µm 2 W-TES of 200 nm thickness, which has two (526×1027) µm 2 phonon collectors (Al/W bilayers) of 1 µm thickness [8].It is mounted on the lid of the copper holder, which is used to protect the NaI from humid air.The light detector is irradiated with a second 55 Fe calibration source of similar activity as the one shining on the absorber.A picture of the light de- The TES is deposited on a wafer, which is separated from the absorber crystal.The coupling between the absorber and the TES consists of an Au-pad glued on the absorber surface and connected to the TES by two Au-wire bonds [6].
tector (Fig. 3a) and an enlarged view of the wire bonding of its TES (Fig. 3b) are shown in Fig. 3.
In the following, we refer to the NaI-remoTES as the phonon channel and the light detector as the light channel, interchangeably.The description of the detector components is summarised in Tab.I.

III. DATA TAKING
The measurement was carried out in an aboveground wet dilution refrigerator at the Max Planck Institute for Physics in Munich.The cryostat is equipped with four superconducting quantum interference devices (SQUIDs) from the APS company [9] and continuously read out with a 16-bit analog-digital converter at a sampling frequency  II.

IV. DATA ANALYSIS
For the three datasets discussed above, the continuous stream from the phonon channel is triggered in software using an optimal filter trigger (cf.Ref. [10] and Ref. [11]).The light channel is read out in parallel.The filter is created from a parametric description of the NaI channel pulse shape based on Ref. [12] and a noise power spectrum obtained from randomly drawn empty noise traces.The typical pulse shape of absorber recoils for the NaI channel is shown in Fig. 4a and features a very long decay time.A baseline energy resolution of the phonon channel of 2.07±0.06keV is determined by superimposing the pulse shape onto a set of ran-domly drawn empty baselines and reconstructing these artificial events.The resulting amplitude distribution is illustrated in Fig. 4b.The baseline resolution of the SOS light detector was determined to be 2.02±0.05keV ee (electron-equivalent) using the same technique.The datasets were triggered with a threshold of 10 keV, where no noise triggers were observed.The energy scale for the aforementioned results comes from a fit to the peaks visible in the 57 Co dataset; this is illustrated in Fig. 5. Peaks from the 55 Fe source could not be observed, as their energies of 5.9 keV and 6.4 keV are below the energy threshold.The optimum filter amplitude is used as an energy estimator.In the energy range from threshold up to ∼500 keV, the detector response is in good approximation linear.All datasets were cleaned by applying a set of quality cuts.Severely unstable detector operation intervals are removed by monitoring the reconstructed amplitude of injected heater pulses over time.Single voltage spike events are removed by cutting on the ratio of the numerical derivative of the pulse to the baseline RMS.The RMS from the optimal filter reconstruction and the RMS from an additional truncated standard event fit reconstruction (cf.e.g.Ref. [13]) are used to remove pile-up events and events from particle in-   teractions in the sapphire wafer.The effect of each quality cut is assessed by simulating pulses with a flat energy spectrum on the set of randomly drawn detector baselines and studying the fraction of surviving events as a function of the simulated energy.We find that the detector threshold of 10 keV is only nominal, i.e. no simulated signal event survives down to this energy.This is due to varying noise conditions in the above-ground setup, very long pulse decay time, and the presence of particle recoils in the wafer, which require strong quality cuts to be discarded.An analysis threshold of 15 keV is used in the following, where the signal survival probability is still about 5 %.Above this threshold, no wafer-induced events or noise events are observed in the background dataset.

V. DISCRIMINATION OF NUCLEAR RECOIL BANDS
The light yield (LY) is defined as the ratio of the energy measured in the light channel and the energy measured in the phonon channel for each event.It enables the discrimination of different recoil event classes.In Fig. 6 the LY is shown as a function of the phonon channel energy for the background and the neutron calibration datasets, respectively.We use a parametrization of the recoil bands and of the spectra which contribute to them, described in detail in [14].For events in the electron band, the LY is normalized to 1. Two quenched bands are expected for Na and I nuclear recoils due to the different masses of the nuclei.We observe that these cannot be reliably separated by the likelihood fit, which yields a high correlation between parameters for the two bands.Therefore, we conservatively estimate the parameters of the e − /γ band from the background dataset, and compare them to the observed events in the neutron calibration dataset.We will address a modification of the band description and separation of the individual nuclear recoil bands in a future publication.Figure 6 shows the result of the likelihood bandfit to the background dataset (left panel) and to the neutron calibration FIG.4: (a) Normalized standard events in the light detector (dashed, dark-blue curve) and in the NaI-remoTES detector (solid, water-green curve).
(b) Reconstructed pulse height (truncated fit) distribution for artificial pulses, which are obtained by superimposing an averaged signal pulse onto empty traces for the phonon channel.dataset (right panel), using 90 % boundaries for the e − /γ band.It can be clearly seen that a new population of events appears below the e − /γ band, far outside the 2 σ boundary.The additional band is quenched in LY to ∼ 0.5 at energies above 100 keV, and the LY appears to be decreasing as the recoil energy approaches the threshold.In Ref. [15], a similar decline of the quenching factor in NaI at lower energies was reported for measurements at room temperature and with Tl-doped crystals.Above ca. 100 keV, the quenching factor observed with our prototype is even higher than in Ref. [15].At increasing energy, a downward tilt of the e − /γ band is visible, which is due to increasing nonlinearity in the light detector response.

VI. CONCLUSIONS
This measurement marks the first proof of eventby-event particle discrimination in a cryogenic NaI detector.We operated a COSINUS prototype with a remoTES sensor, which displayed a baseline resolution of 2 keV despite suboptimal, above-ground conditions.It was calibrated with a 57 Co γ source and analysed with a nuclear recoil threshold of 15 keV.Particle discrimination was verified with neutrons from an AmBe source.In Ref. [6], the re-moTES design was suggested as an improved readout for delicate absorber materials, which are for example hygroscopic or feature a low melting point.This work shows that the design is indeed suitable for NaI absorbers.The next step in the COSINUS detector optimization strategy in the direction of achieving an energy threshold of 1 keV is an underground measurement with a similar detector, in order to assess its performance in a low-background environment.This measurement has already been performed, and will be subject of a future publication.

FIG. 1 :
FIG.1: Schematic of the remoTES detector.The TES is deposited on a wafer, which is separated from the absorber crystal.The coupling between the absorber and the TES consists of an Au-pad glued on the absorber surface and connected to the TES by two Au-wire bonds[6].
FIG. 2: (a) The remoTES wafer mounted in the copper holder.The sapphire wafer is placed on sapphire balls and held by a bronze clamp.A 55 Fe source is taped on a copper piece facing the absorber for the purpose of energy calibration.(b) Microscopic view of the Au-pad glued on the NaI-crystal and the wire bonding to the remoTES.

FIG. 3 :
FIG. 3: (a) SOS light detector mounted in a copper holder.(b) Microscopic view of the electrical connections of the light detector TES and of its separate ohmic heater.

FIG. 5 :
FIG.5: Energy spectrum from the 57 Co γ calibration measurement.Different components are fitted using a parametric description.The fit includes gaussian peaks at 136 keV and 122 keV on top of a linearly decreasing background, as well as a γ escape peak due to I K-α, expected at around 89 keV.The parametric fit also considers 'shoulders' on the left side of each peak, which originate from 57 Co γs depositing a fraction of their energy in parts of the setup before reaching the detector.

TABLE I :
Dimensions of all components of phonon detector and light detector.

TABLE II :
Measuring times, exposures, event rates and calibration sources for the three datasets.