Weakly interacting massive particles (WIMPs, denoted as χ) have been extensively searched via elastic scattering with normal matter in underground direct detection experiments [1,2] under ultralow background conditions. Liquid noble gas detectors are leading the sensitivities at WIMP mass (mχ) above 10  GeV/c2 [3–5], while solid state detectors are generally used for mχ<10  GeV/c2 [6–13].

With excellent energy resolution and low energy threshold, p-type point contact germanium (pPCGe) detectors have been used and further developed for light WIMP searches by CDEX [7–10]. Located in the China Jinping Underground Laboratory (CJPL) [14], the first generation CDEX-1A (1B) experiments used 1-kg-scale single-element pPCGe cooled by a cold finger since 2010 [8–10]. With an energy threshold of 160 eVee (“eVee” represents electron equivalent energy derived from a charge calibration) and an exposure of 737.1 kg day, CDEX-1B provided improved limits on WIMP-nucleon spin-independent (SI) and spin-dependent (SD) scattering down to mχ of 2  GeV/c2 [10].

Toward a future ton-scale DM experiment, the second generation CDEX experiment with a total detector mass of about 10 kg, called CDEX-10, has used three triple-element pPCGe strings (C10A, B, C) directly immersed in liquid nitrogen (LN2). Compared with cold finger cooling and high-Z material shielding systems, low-Z material shielding, such as with LN2 or liquid argon, provides better control of radiation background. The concept of integrated shielding and cooling, first proposed in the GENIUS project [15], is realized in the GERDA experiment with the lowest background among neutrinoless double beta decay (0νββ) experiments [16] and will be expanded into the next generation LEGEND 0νββ program [17]. CDEX-10 focuses on the arraying technologies and background understanding of the prototype pPCGe detectors developed based on the CDEX-1 technique. The new CDEX-10 array detectors and dedicated data acquisition (DAQ) system started testing and data taking inside a LN2 tank in 2016 at CJPL. C10A was returned to the CANBERRA factory in France for upgrades. Of the remaining six detectors, two had faulty cabling, and two others had a high level of noise. In this Letter, we report the results from a first physics data set of one of the two operational detectors C10B-Ge1, which had the lower threshold.

The stainless steel LN2 tank was located in the polyethylene room with 1 m thick walls at CJPL-I for cooling of the CDEX-10 detectors, which are surrounded by 20 cm thick high-purity oxygen-free copper immersed in LN2 to shield the ambient radioactivities. The shielding configuration of CDEX-10 and the structure of a detector string are shown in Fig. 1.

The DAQ system received signals from the p+ point contact electrode of C10B-Ge1 which were fed into a pulsed reset preamplifier. Five identical output signals of the preamplifier were further processed and digitized. Two of them were distributed into 6  μs (Sp6) and 12  μs (Sp12) shaping amplifiers for a 0–12 keVee energy range. These two channels were used for energy calibration and signal and noise discrimination. The third channel was loaded to a timing amplifier (Tp) to measure the rise time of signals within a 0–12 keVee energy range which can be used for bulk or surface events discrimination. The remaining two were loaded to a shaping amplifier and a timing amplifier with low gains aiming at a high energy range for background understanding. To estimate the dead time of the DAQ system and cut efficiencies uncorrelated with energies, random trigger (RT) events were recorded once every 20 seconds. The output signals of the above amplifiers were digitized by the 14-bit 100-MHz flash analog-to-digital converters. Data taking with C10B-Ge1 was performed from February 26, 2017 to November 7, 2017. The DAQ dead time fraction was measured by RT events to be 4.8%, giving a live time of 112.3 days.

The data analysis follows the procedures described in our earlier work [8–10], starting from the parameters extracted from the digitized pulses. The optimal integrated area of the pulse from Sp12 is selected to define the energy for its excellent energy linearity at the low energy region. Energy calibration was done with the internal cosmogenic x-ray peaks: 10.37 keVee of Ge68 and 8.98 keVee of Zn65, and the zero energy defined by the RT events. Analysis procedures follow those with similar detectors in CDEX-1B [10]. Basic filtering algorithms are first applied to the Sp6,12 and Tp pedestals to reject events with anomalous electronic noise profiles. These cuts are energy independent, and the efficiency is measured to be 97.4% by the survival of RT events, giving rise to a valid data sample of 109.4 days.

The second step is a physics-noise event (PN) cut to discriminate the signals from electronic noises near the energy threshold. The PN cut is based on the relationship between the energy and maximum amplitude of Sp12. The experimental data of a Cs137 source are used to derive the PN cut and the trigger efficiencies. The efficiency curves with 1σ bands are shown in the inset of Fig. 3(a).

Events depositing energy in the n+ surface layer generate a slow rising pulse and an incomplete charge collection due to the weak electric field and severe recombination of electron-hole pairs in this region [18]. Since C10B-Ge1 and CDEX-1B detectors have the same crystal mass, crystal structure, and fabrication procedure, the same dead layer thickness of 0.88±0.12  mm [19] is taken for this analysis. This gives rise to a fiducial mass of 939 g and accordingly a physics data size of 102.8 kg day.

The bulk and surface events (BS) cut is carried out to select bulk events. WIMP candidate events in the bulk of the detector are then separated from the surface events via the rise-time differences of the Tp signals. The rise-times (τ) are measured by fitting the Tp pulse to a hyperbolic tangent function [8–10,18]. The log10(τ) distribution versus measured energy of in situ events is depicted in Fig. 2(a), showing a two-band structure of bulk and surface events well separated above 1.5 keVee. However, at lower energies the bulk and surface events infiltrate into each other, as a result of the electronic noise smearing effect. Multisite events are located off band and of negligible fraction at the keVee-range energy [20].

It has been shown that the background and calibration sources data share the common bulk or surface rise-time distribution probability density function (PDF) [21]. The ratio method has been developed accordingly to address the BS discrimination problem in pPCGe [10,21]. In this analysis, the inputs of the ratio method include the background data and three calibration samples (Cs137, Co60, Cd109), while Cd109 is a pure surface source. Considering that the low-energy gammas from the Cd109 source can hardly penetrate the n+ surface layer, their rise-time distribution can describe the surface PDF. Four boundary parameters related to the approximately “pure” bulk and surface regions are depicted in Fig. 2(a). Two outside boundaries [log10(τ)=b0 and log10(τ)=s1] are derived by fitting the best normalization interval of each energy bin of 500 eVee from 160 eVee on, based on the selection principles of making the statistics as significant as possible while the rise-time distributions of those events remain as consistent as possible. As depicted in Figs. 2(b) and 2(c) and Figs. 2(d) and 2(e), the comparisons of the rise-time distributions of those samples at 0.16–0.66 keVee and 1.66–2.16 keVee demonstrate that they share common rise-time distribution PDFs when normalized to the “pure” bulk and surface regions.

There are extremely-fast events (EFEs) with a faster rise time in the bulk band due to better rise-time resolution in C10B-Ge1 than CDEX-1A and CDEX-1B [9,10]. It has been verified with simulations using realistic detector electric field that these EFEs mainly originate from the region in the vicinity of the p+ point electrode. An additional convincing evidence is that x rays from Cu are observed only in the EFEs band at 8 keV energy; they can only enter the active area through the passivated surface layer near p+ point. Unfortunately, EFEs can only be distinguished clearly from the bulk band above sub-keVee, while the differentiation is not possible at a low energy region due to the smearing from electronic noise. A cut [log10(τ)<b0] was used to remove the EFEs, together with an extremely-slow events cut [log10(τ)>s1] [21] to remove those events which are seriously attenuated by the n+ surface layer. Both kinds of events are included to bulk and surface counts after the B and S correction procedures [21].

The corrected bulk or surface counts (Br/Sr) can be derived by integrating the optimized PDFs which are derived by numerically minimizing the χ2 of Eq. (7) in Ref. [21]. The reconstructed Cs137 and Co60 spectra are consistent with geant4 [22] simulations. The Br of the background from C10B-Ge1 with the main contributions of errors at the first bin of 0.16–0.26 keVee and a typical high energy of 1.96–2.06 keVee are shown in Table I. The systematic errors mainly come from the choices of b0, b1, s0, and s1, the errors of which are estimated by varying the more “pure” bulk and surface regions of Fig. 2(a). Further details of the BS analysis and uncertainties derivations can be found in Ref. [21].

The spectra after different event-selection cuts are shown in Fig. 3(a). The physics analysis threshold is 160 eVee. From the spectra in Fig. 3(a), characteristic K-shell x-ray peaks from internal cosmogenic radionuclides like Ge68,71, Ga68, Zn65, Co57, Fe55, Mn54, and V49 can be identified. In addition, x-ray peaks from Cu and Zn isotopes excited by high energy γ rays are observed in the extremely-fast events region of the background spectrum. Their intensities are derived by best fit from the spectrum [8–10]. The contributions of L- or M-shell x-ray peaks are derived from corresponding K-shell peaks and subtracted from the Br spectrum, shown in Fig. 3(b) [23]. A minimum-χ2 analysis [8] is applied to the residual spectrum, using two free parameters characterizing the flat background and the possible χ-N SI cross section (σχNSI). The best-fit spectrum at mχ=5  GeV/c2, where σχNSI=(−0.61±4.3)×10−42  cm2 at χ2/DOF=11.2/22 (p value =0.97), is also depicted in Fig. 3(c). A standard WIMP galactic halo assumption and conventional astrophysical models [24] are used to describe WIMP-induced interactions, with the local WIMP density of 0.3  GeV/cm3, the Maxwellian velocity distribution of v0=220  km/s, and the escape velocity of vesc=544  km/s. The quenching factor in Ge is calculated by the trim software package [20,25–27] with a 10% systematic error adopted for the analysis [9].

Upper limits are derived following standard procedures [9,28]. The exclusion plots of SI and SD at a 90% confidence level (C.L.) are depicted in Figs. 4(a) and 4(b), respectively, with several selected benchmark direct search experiments superimposed [3,4,9–13,29–32]. The most stringent accelerator bounds on SI from the LHC experiments [33–37] are more constraining in SI—with σχNSI<10-48  cm2 for mχ∼5  GeV—than the scale displayed in Fig. 4(a). They are, however, extremely sensitive to particle physics models and the choice of parameters. The LHC results are derived with χ-proton cross sections and hence unrelated to the SD constraints on χ-neutron cross sections. This study achieves the lowest threshold and background among the various CDEX data set to date, which brings forth almost an order of magnitude improvement over our previous bounds [9,10]. New regions on SI for mχ at 4–5  GeV/c2 are probed and excluded. The CDEX-10 detector array will be installed in a new large LN2 cryotank with a volume of about 1700  m3 at Hall-C of CJPL-II [14] by the end of 2018, where shielding from ambient radioactivity is provided by the 6 m-thick LN2. The projected parameter space to be probed with a reduced background comparable to the best achieved in germanium detectors [38] is also shown in Fig. 4.