Introduction

The presence of dark matter (DM) in the universe is supported by many observations based on gravitational effects1,2,3 but its real nature remains unknown. Dark matter may consist of an undiscovered elementary particle4. A leading candidate is a Weakly Interacting Massive Particle (WIMP), with a mass ranging from 10 GeV c−2 to few TeV c−2. This range is extensively searched for via elastic scattering off atomic nuclei—later called nuclear recoils (NR). These searches often use underground Time Projection chambers (TPC)5,6,7,8,9. The lack of concrete evidence of direct DM detection so far motivates the search for lighter WIMPs, below 10 GeV c−2, and for light DM candidates interacting with shell electrons—later called electron recoils (ER)—which may subsequently produce sufficiently large ionization signals10.

DarkSide-20k (DS-20k) is the next generation of liquid argon (LAr) dual-phase TPCs, presently in construction at INFN Laboratori Nazionali del Gran Sasso in Italy. It is expected to start taking data in 2027. It is primarily designed to perform a nearly instrumental background-free search for high mass (>10 GeV c−2) WIMPs. DS-20k aims for <0.1 background events with an exposure of 200 ton ⋅ year. The detection mechanism relies on the combined observation of the scintillation light (S1) and ionization (S2) signals. DarkSide-50 (DS-50), a first generation LAr dual-phase TPC, already demonstrated the capability of the technology and obtained world best sensitivities to light DM particles using only the S2 signal, when the S1 signal is no longer observable11,12. The analysis was based solely on the number of electrons in S2 (\({N}_{{e}^{-}}\)) as discriminating variable. An update based on the capability to measure the liquid argon ionization yield for low energy electron and nuclear recoils down to  ~180 eVer and  ~500 eVnr13, respectively, was one of the key ingredients to obtain world best limits for WIMPs in the mass range [1.2, 3.6] GeV c−214. A factor of 10 improvement with respect to the previously published limit of Ref. 11 was achieved. The analysis was also used to place limits on DM–nucleon interaction via the Migdal effect15 and on sub-GeV c−2 DM–electron scattering16. The stability of the electroluminescence yield has been measured to be better than 0.5% over almost three years17. Based on these successes of DS-50, this article presents the low-mass DM sensitivity prospects for the DS-20k detector. DS-20k will increase the low-radioactivity LAr volume by about a factor of 1,000 with respect to DS-50. It will also significantly improve the radio-purity of the components surrounding the active volume. A detector specifically designed for the investigation of light dark matter using LAr and assuming further isotopic depletion via cryogenic distillation has also been recently studied18.

Results and discussion

The DS-20k detector

The DS-20k TPC is filled with a 49.7 ton active mass of argon extracted from underground CO2 wells, hence called Underground Argon (UAr). The TPC is shaped as a prism with an octagonal base, with a vertical drift length of 348 cm and an octagonal inscribed circle diameter of 350 cm. The active volume is immersed in a uniform electric field generated by applying a voltage potential of 73.4 kV between the anode and the cathode made of transparent acrylic (PMMA) coated with a conductive material (Clevios™). The corresponding maximum drift time for the ionization electrons is 3.7 ms. A set of 200 μm wide stainless steel wires spaced by 3 mm located 3 mm below the liquid level is used to define a high field extraction region for drifting electrons. A multiplication region filled with gaseous argon, 7 mm thick between the liquid level and the anode, allows the S2 electroluminescence signal to develop.

To ensure the best possible collection of both scintillation and ionization-induced photons, the inner TPC walls are covered with reflectors. All the inner surfaces are coated with tetraphenyl butadiene to shift the native argon UV light to visible wavelength for which photo-detection efficiency is maximal. Two planes of cryogenic Silicon Photo Multipliers (SiPMs) covering the top and the bottom faces of the TPC detect the light signals. The  ~200 k SiPMs are gathered in 5 × 5 cm2 arrays, called Photo Detector Modules (PDMs)19. PDMs include the front-end electronics20 and are arranged in 2112 readout channels. The S2 yield is about 25 photo-electrons per ionization electron extracted in the gas pocket. If neutrons scatter in the TPC and produce a WIMP-like signal, they are likely to be captured in the 15 cm thick acrylic TPC walls or endcaps that enclose the SiPM planes, with the latter possibly further doped with Gadolinium. Neutrons captured in this way release γ-rays which are detected in the TPC and/or the 32 ton UAr veto surrounding it, which is read by 480 photo detector channels. The TPC and this UAr veto are housed in a stainless steel (SS) vessel. This SS vessel is immersed in a bath of 650 tons of atmospheric argon (AAr), acting as a shield and an outer veto detector for muons and associated products. The AAr is contained in a ProtoDUNE-like membrane cryostat21. DS-20k adopts a triggerless data acquisition system ensuring to save all SiPM signals with close to 100% efficiency for single photo-electrons. Figure 1 shows a cross-section of DS-20k.

Fig. 1: Cross-section of the DS-20k detector.
Fig. 1: Cross-section of the DS-20k detector.
Full size image

The Time Projection Chamber is at the center, with its acrylic walls in green and electrodes in pink. The stainless steel vessel surrounding it is shown in gray, immersed in the ProtoDUNE-like cryostat.

Selection

The analysis relies on the strategy successfully put in place for DS-5014 and adapted to DS-20k. WIMPs are expected to scatter only once in the LAr volume of the TPC. As each particle interaction is associated with a S2 pulse, events with a single S2 pulse are selected. Pulses are required to be isolated from any other S2 pulse preceding or following by >3.7 ms—which corresponds to the maximum electron drift time. It is assumed that two close S2 pulses can be distinguished if they are separated in time by >2 μs (2 mm along the drift direction), as achieved in DS-50. Relaxing this assumption produces a negligible impact on the expected sensitivity. Anomalously low S2 could come from α-background. It is assumed that S2 signals can be identified without introducing significant inefficiencies, and that surface α-background that may produce low energy S2 can be efficiently suppressed, as in DS-50. Before any selection procedure, the total rate of ER-events from electron and γ-ray backgrounds from radioactive decays is estimated to be 80 Hz (0.0016 Hz kg−1 of UAr) in the TPC, to be compared with 1.5 Hz in DS-50 (0.03 Hz kg−1 of UAr). The accidental coincidences within the 3.7 ms isolation window between the S2 pulses induced by this background and by the signal result in an effective livetime of 51%. Unlike the high mass WIMP search analysis, it is not possible to use the pulse shape discrimination of the S1 signal9,22 to remove the ER background. Instead, the selection aims at mitigating γ-rays and X-rays from radioactive isotopes of the detector components surrounding the UAr active volume (including PDMs). They are efficiently suppressed by a fiducialization based on the S2 pattern in the transverse direction of the TPC, removing an outer 30 cm thick octagonal shell. The position reconstruction resolution along this direction is estimated to be better than 3 cm. As the present analysis is based solely on S2, the drift time, computed as the time difference between S1 and S2, is not available and no fiducialization is performed along the electron drift direction. This procedure leads to a UAr fiducial mass of 34.2 tons, hence an exposure of 17.4 ton ⋅ year for one year of data, taking into the account the effective livetime.

Sensitivity results

Using the signal and background models which are thoroughly described in the Methods section, DS-20k prospects for 90% C.L. exclusion limits are derived from a binned profile-likelihood fit implemented in the RooFit package23. The likelihood is built as the product of Poissonian terms, one for each of the considered bins. Systematic uncertainties that are believed to be reachable at the time of data taking are accounted for by introducing the nuisance parameters shown in Table 1. They are classified as amplitude or shape systematics, the latter accounting for uncertainties on 39Ar and 85Kr β-decays14 and on “spurious” electrons (SE) modelling, as well as for spectral distortions from the ionization response.

Table 1 List of systematic uncertainties included in the binned profile likelihood
Full size table

Figure 2 shows the pre-fit \({N}_{{e}^{-}}\) distribution for the background model. The 39Ar component is dominating for \({N}_{{e}^{-}}\ge 4\). The SE contribution dominates in the \(2\le {N}_{{e}^{-}}\le 3\) range and is a factor  ~ 18 below 39Ar at \({N}_{{e}^{-}}=4\). The contribution from the PDMs is dominating the external γ-ray background. The neutrino background from coherent elastic scattering off nucleus (CEνNS) and elastic scattering (ν-ES) off argon electrons is two to four orders of magnitude below the 39Ar background. A typical signal, corresponding to a 2 GeV c−2 WIMP mass with a cross-section of 3 × 10−44 cm2, is superimposed for illustration.

Fig. 2: Pre-fit DS-20k \({N}_{{e}^{-}}\) spectra distribution.
Fig. 2: Pre-fit DS-20k $${N}_{{e}^{-}}$$ N e − spectra distribution.
Full size image

The corresponding ER and NR energy scales are indicated at the top. The pre-fit background model (red line) is shown with its uncertainties (shaded area). Contributions from all the components of the background are also shown. A typical signal model is superimposed for illustration (green dotted line), assuming a 2 GeV c−2 WIMP mass with a cross-section of 3 × 10−44 cm2 and quenching fluctuations.

In the following, given the unknowns on the SE background before data taking, two different fit strategies are envisaged: the first one (conservative) is reproducing the DS-50 approach and uses the \({N}_{{e}^{-}}\) range from 4 to 170, while the second one (ultimate) assumes a good understanding of the spurious electrons in DS-20k and uses the total fit range from 2 to 170. Nuisance parameters affecting 39Ar, PDMs, TPC and spurious electrons (only in the ultimate fit case for the latter) are strongly constrained by the fit, since they are related to the dominant backgrounds. For the same reason, a strong (anti)-correlation exists between these amplitude nuisance parameters and the one associated to the exposure. As 39Ar is the dominant background, its spectral shape will need to be computed with higher precision in order to interpret any possible future excess in DS-20k as a signal.

The DS-20k expected median 90% C.L. upper limits on spin-independent WIMP–nucleon cross-section (σSI), computed with the CLs prescription24 and for the two fit strategies, are shown in Fig. 3 for 1 year of data. They are compared to the published 90% C.L. limits from DS-5014 and from other experiments7,25,26,27,28,29,30,31. An improvement in sensitivity by up to a factor 40 over DS-50 is achieved using the conservative fit. This increases to a factor 170 at 1.2 GeV c−2 WIMP mass with the ultimate fit, reaching a σSI of 1 × 10−43 cm2. Overall, a sensitivity to σSI below 1 × 10−42 cm2 is achieved for WIMP masses above 800 MeV c−2, covering a large uncharted phase space with 1 year of data. The dominant systematic uncertainties come from the main background (39Ar, SE) modelling and from the detector response (ER ionization yield). The sensitivity would be ~60% better if systematics were neglected. Results without quenching fluctuation on NR are shown in Supplementary Figs. 1 and 2.

Fig. 3: Expected DS-20k 90% C.L. exclusion limits for spin-independent WIMP–nucleon cross-section (σSI).
Fig. 3: Expected DS-20k 90% C.L. exclusion limits for spin-independent WIMP–nucleon cross-section (σSI).
Full size image

They are shown as bold red lines, assuming quenching fluctuations (QF) (dotted: fit from \({N}_{{e}^{-}}\) = 4, dashed: fit from \({N}_{{e}^{-}}\) = 2). One year of data is assumed. They are compared to the published 90% C.L. limits from DS-5014 and from other experiments7,25,26,27,28,29,30,31, with currently excluded parameter space shaded in light gray, as well as claimed discovery from Ref. 49. The neutrino fog in LAr with index n = 232 is also shown. A local dark matter density of 0.3 GeV c−2 cm−3 is assumed.

The sensitivity scales with the square root of the exposure. With 10 years exposure, the sensitivity will improve by a factor 3 compared to 1 year, whatever the WIMP mass: the neutrino fog in LAr with index n = 232 could be reached for WIMP masses around 5 GeV c−2 (Supplementary Fig. 3).

To assess the robustness of these expectations, variations of detector response model and background activities with respect to the nominal assumptions have been considered. The electron lifetime, single electron response and x − y resolutions and 39Ar activity have been varied by a factor 2 with respect to the nominal assumptions. The main impact comes from the 39Ar activity, assumed to be the same as DS-50, and from the single electron response resolution. If the latter is degraded by a factor 2 with respect to expectations, then the sensitivity would degrade at most by a factor 2 using the ultimate fit strategy, mainly affecting WIMP masses around 1 GeV c−2. A significant improvement in sensitivity is expected if the UAr extraction plant would further reduce the contamination of 39Ar. For a factor 2 lower contamination, the sensitivity would improve up to a factor 1.8 (Supplementary Fig. 4).

The sensitivities to other models described in Refs. 15 and 16 have been evaluated. They are shown in Fig. 4 for 1 year of data, using both conservative and ultimate fit approaches. In both cases and for all models, significant improvements in sensitivity over DS-50 is found over the whole mass ranges, resulting in sensitivity to uncharted phase spaces.

Fig. 4: Expected DS-20k upper limits at 90% C.L. for various signal models.
Fig. 4: Expected DS-20k upper limits at 90% C.L. for various signal models.
Full size image

The x-axis shows the mass of the candidate while the y-axis shows the model parameter. One year of data is assumed. These results (bold red lines, dotted: fit from \({N}_{{e}^{-}}\) = 4, dashed: fit from \({N}_{{e}^{-}}\) = 2) are compared to the published 90% C.L. limits from DS-5015,16, other experiments7,25,26,27,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67 and astrophysical constraints68,69,70,71, with currently excluded parameter space shaded in light gray. a Spin-independent WIMP-nucleon cross-section when considering the Migdal effect and with quenching fluctuations (QF) for the NR signal. b, c Light dark matter cross-section for an heavy (b) and light (c) mediator, respectively. Thick lines show cross-sections giving the relic DM abundance through freeze-in72,73 or freeze-out74 production mechanisms. d Axion-electron coupling strength gAe. e Kinetic mixing parameter κ between the photon and dark photon. f Mixing angle ∣Ue4∣2 between sterile neutrino and an active neutrino state. For the latter, the authors of Ref. 16, who contributed to this work as well, identified an error in evaluating the limit, which is corrected here. The indirect detection limits set by the NuSTAR experiment71, which looks for anomalous X-ray lines from radiative sterile neutrino DM decays, extends downwards to ∣Ue4∣2 = 10−13 at 20 keV c−2. A local dark matter density of 0.3 GeV c−2 cm−3 is assumed.

Conclusions

The dual-phase liquid argon time projection chamber is presently one of the leading technologies to search for light galactic dark matter particles with masses below 10 GeV c−2, as demonstrated by the DS-50 experiment with 50 kg of underground liquid argon. Based on this success, a prospect analysis was carried out for the upcoming DS-20k experiment which will have a 1,000 times larger LAr volume target. Already with 1 year of data, a sensitivity improvement to DM–matter interaction cross-sections by at least one order of magnitude with respect to DS-50 is expected for a wide range of DM models: WIMP with or without the Migdal effect in the MeV c−2 to GeV c−2 mass range as well as Light Dark Matter, axion-like particles, sterile neutrinos and dark photons in the keV c−2 or sub-keV c−2 mass range. A sensitivity to WIMP–nucleon interaction cross-sections below 1 × 10−42 cm2 is achievable for WIMP masses above 800 MeV c−2. With 10 years exposure, the neutrino fog in LAr with index n = 2 can be reached for WIMP masses around 5 GeV c−2. Further improvements in the sensitivity of DS-20k could be achieved through dedicated measurements aimed at quantifying the fluctuations in quenching for nuclear recoils or identifying the processes underlying the generation of spurious electrons, thereby enabling their effective suppression.

Methods

Detector response model

The number of electrons in S2 is derived from the energy deposited by a single scatter event in the UAr using the ER and NR ionization yields measured by DS-5013. The intrinsic fluctuations to ER signals are modelled with an empirical fudge factor implemented as a Fano factor measured by DS-5013. Fluctuations from the ionization quenching effect in NR are not known and two models, assuming no fluctuation (NQ) or binomial fluctuations between detectable and undetectable quanta (QF), are considered14. Unless explicitly stated, QF is assumed in this article. A Geant4-based Monte Carlo package33 is used for an accurate simulation of light production, propagation and detection for background and signal events. Additionally, effects due to SiPM and electronics noise are simulated on the basis of pre-production SiPM module performance, resulting in a 23% single electron response resolution. Electron losses during the drift due to attachment to impurities are taken into account assuming a 16 ms lifetime, as measured in DS-5033.

Background model

The β-decay background intrinsic to LAr is fully dominated by the radioactive isotopes 39Ar and 85Kr present in the active volume of UAr. Extracting the argon from underground (same source as DS-50) significantly reduces their contamination with respect to AAr8. Their activities are assumed to be 0.73 mBq kg−1 and 1.9 × 10−2 mBq kg−1, respectively. The former has the same level as in DS-50 while the latter is reduced by a factor 100 compared to the one of DS-50, thanks to a new multiple distillation column system that has been added at the UAr extraction plant. The argon and krypton spectral shapes are based on calculations of atomic exchange and screening effects, validated on measured 63Ni and 241Pu spectra with a 200 eV threshold34,35. Below this value, a linearly increasing uncertainty on the corresponding corrections is assumed, reaching 25% at 0 eV14. Further systematics on the spectral shape originate from the uncertainty on the Q-value (1% for 39Ar and 0.4% for 85Kr).

External γ-ray and X-ray backgrounds come from materials used to build the whole inner detector system. Materials are carefully selected for low levels of radioactivity and their activities are measured in an extensive material screening campaign based on the combination of different radio assay techniques. As a summary, Table 2 lists the expected activity of each considered radio-contaminant of the inner detector materials that may lead to γ-ray and X-ray background. Each isotope is simulated uniformly in the component material and decaying particles are tracked over the DS-20k geometry. Thanks to the use of SiPMs instead of classical photomultipliers, the use of PMMA walls for the TPC and the minimization of the amount of passive material, the γ-ray background level is expected to be reduced with respect to DS-50 by a factor 2.5 per surface area orthogonal to the electron drift direction. This validates the extreme care taken to consider radio-pure materials in the design of the DS-20k inner detector.

Table 2 Estimation of DS-20k material activities for each radio-contaminant, in Bq, for external γ-ray and X-ray background sources
Full size table

“Spurious” electrons (SE), whose origin might be ionization electrons trapped by impurities and released later, are a major component of the background at low number of electrons (\({N}_{{e}^{-}} < 4\)). The ionization spectrum model is built by fitting DS-50 data in this \({N}_{{e}^{-}}\) range, assuming the same spectrum in DS-20k, with a Poisson distribution convolved with a Gaussian accounting for the single electron response. The origin of the spurious electron signals is assumed to be completely explained by the impurity mechanism in UAr, with the same level of impurities as in DS-50. The expected rate in DS-20k is extrapolated from the DS-50 rate as in Ref. 18. This background will need to be thoroughly characterized once real DS-20k data is available. The systematic uncertainty on the spectral shape derived from the fit to DS-50 data is assigned to the DS-20k SE modelling.

Other backgrounds to the DM search coming from the interaction of neutrinos via coherent elastic scattering off nucleus (CEνNS)36, recently observed with argon37 and xenon38,39 nuclei, have been considered. The study includes radiative corrections40 and an accurate parametrization of the nuclear structure41. The main contribution impacting this analysis comes from neutrinos from solar 8B which deposit <10 keV in UAr. Elastic scattering (ν-ES) off argon electrons42 have also been considered, surpassing CEνNS for \({N}_{{e}^{-}}\) > 30. In both cases, neutrino fluxes are normalized according to Ref. 43. Finally, the rate of NRs from radiogenic and cosmogenic neutrons is expected to be over four orders of magnitude lower than the ER one, and therefore not considered in this analysis.

Signal models

The signal models are derived assuming the standard isothermal WIMP halo model with an escape velocity of 544 km s−1, a local standard of rest velocity of 238 km s−1, and a local dark matter density of 0.3 GeV c−2 cm−343. WIMPs are assumed to elastically scatter off nucleons. Atomic effects predicted by Migdal44 could add an extra emission of electrons to a fraction of nuclear recoils, increasing the sensitivity to low mass WIMPs. As experimental efforts to confirm the existence of such effects in nuclear scattering are still underway45,46,47, WIMP signals with and without them, modelled as in Ref. 15, are considered in this article.

Other dark matter candidates interacting with electrons and producing a S2 signal are possible. First, fermion or scalar boson light dark matter particles with a mass below 1 GeV c−2 could interact with bound electrons via a vector mediator. The interaction and cross-section depend on the momentum-transfer q via a form factor FDM and two benchmark models are considered: a heavy mediator with FDM = 1 and a light mediator with \({F}_{{{{\rm{DM}}}}}={(\alpha {m}_{e}/q)}^{2}\) where α is the fine structure constant and me the mass of the electron. Other possibilities are the absorption of axion-like particles—coupled to electron via gAe, and vector-boson like dark photon – mixing with photon via the κ parameter —by argon shell electrons. Finally sterile neutrinos, mixing with an active neutrino state by an angle ∣Ue4∣2, could inelastically scatter off a bound electron. All these models are described in details in Ref. 16.