CoGeNT: A search for low-mass dark matter using $p$-type point contact germanium detectors

CoGeNT employs $p$-type point-contact (PPC) germanium detectors to search for weakly interacting massive particles (WIMPs). By virtue of its low-energy threshold and ability to reject surface backgrounds, this type of device allows an emphasis on low-mass dark matter candidates ($m_{\chi }\sim 10\mathrm{GeV}/c^2$). We report on the characteristics of the PPC detector presently taking data at the Soudan Underground Laboratory, elaborating on aspects of shielding, data acquisition, instrumental stability, data analysis, and background estimation. A detailed background model is used to investigate the low-energy excess of events previously reported and to assess the possibility of temporal modulations in the low-energy event rate. Extensive simulations of all presently known backgrounds do not provide a viable background explanation for the excess of low-energy events in the CoGeNT data or the previously observed temporal variation in the event rate. Also reported for the first time is a determination of the surface (slow pulse rise time) event contamination in the data as a function of energy. We conclude that the CoGeNT detector technology is well suited to search for the annual modulation signature expected from dark matter particle interactions in the region of WIMP mass and coupling favored by the DAMA/LIBRA results.

Figure 1

Partially disassembled shield of the CoGeNT detector at SUL, showing the cylindrical OFHC end cap and innermost 5 cm of ancient 0.02 Bq ${}^{210}\mathrm{Pb}/\mathrm{kg}$ lead, characteristically oxidized following etching. The preamplifier is visible at the top right (black box). A minimum of 7 cm of lead thickness shields the detector from the naturally occurring radioactivity in the preamplifier’s electronic components.

Figure 2

Layout of the complete shield for the CoGeNT detector. The outermost component is a layer of recycled HDPE, used to moderate neutrons. Next toward the interior, a 1 in thick layer of borated polyethylene captures moderated neutrons. Three layers of lead are indicated by the three different inner shaded regions. The outermost lead is composed of stock bricks, not chemically etched, the middle layer is chemically etched and cleaned, and the innermost layer consists of ultralow background ancient lead. An automated liquid nitrogen transfer system refills the detector Dewar every 48 h, maintaining the germanium crystal at a near constant temperature. See text for a full description of these components.

Figure 3

Schematic of the data acquisition system for the CoGeNT detector at SUL (see text).

Figure 4

Example digitized traces from the six CoGeNT DAQ read-out channels, corresponding to an event with energy $\sim 2.5\mathrm{keVee}$. Preamplifier traces are DC-offset at the Phillips Scientific 777 amplifier to allow for rise time measurements of pulses in the range 0–12 keVee, following offline wavelet denoising [4] (not yet applied to these traces).

Figure 5

Neutron scattering measurements of the low-energy quenching factor for nuclear recoils in germanium, compared to Lindhard theory predictions. CoGeNT adopts the expression relating ionization and recoil energy $E_i(\mathrm{keVee})=0.2\times E_r^{1.12}(\mathrm{keVr})$, valid for the range $0.2\mathrm{keVr}<E_r<10\mathrm{keVr}$, and essentially indistinguishable from the Lindhard case plotted.

Figure 6

Characterization of surface structure on the external $n+$ contact of a PPC (see text). The two free sigmoid parameters are fit via comparison of calibration data and Monte Carlo simulation. Energy depositions taking place in the transition layer near its boundary with the dead layer lead to large signal rise times, i.e., slow pulses. On the opposite side of the transition layer, rise times progressively approach the small values typical of a fast (bulk) event.

Figure 7

Trigger efficiency vs energy equivalent for 10 Hz tailed electronic pulses generated with a 814FP CANBERRA pulser. Inset: gain shift stability monitored through the centroid of a Gaussian fit to the 10.3 keV cosmogenic peak. The count rate under this peak decayed from roughly 500 to 150 events per month over the time span plotted.

Figure 8

Daily average electronic noise and trigger threshold in the CoGeNT PPC at SUL. The small jump in electronic noise postfire has a negligible effect on the detector threshold. It is the result of either temperature cycling of the crystal (leading to known processes capable of altering the detector leakage current, minimally in this case) or a displacement of cables during emergency postfire interventions. The fluctuations in trigger threshold agree well with expectations based on manufacturer specifications for the ORTEC 672 shaping amplifier and NI PCI-5102 digitizers, and the observed $\pm 1°\mathrm{C}$ environmental temperature changes measured at SUL.

Figure 9

Diurnal stability of CoGeNT at SUL. Periods of human presence at SUL are $\sim 7\mathrm{am}–5\mathrm{pm}$.

Figure 10

Steps in data selection through the UC analysis pipeline: (a) All data including microphonics-intensive periods of LN2 Dewar filling. (b) Following removal of LN2 transfer periods and ensuing 10 min (boiling in the Dewar lasts a few minutes). No correlated excess of events is observed to extend beyond this 10 min cut. (c) Following application of cuts intended to remove anomalous electronic pulses (see text). The boundaries for a final cut using the $\mathrm{CH}0/\mathrm{CH}1$ amplitude method in [24] are shown as horizontal lines. These boundaries are selected to minimize the effect of this cut for both radiation-induced and pulser events, with the exception of a distinct family of residual microphonic events visible as a diagonal band in this panel. (d) Fast electronic pulser events prior to any cuts (only the $\mathrm{CH}0/\mathrm{CH}1$ amplitude criterion is seen to minimally affect these).

Figure 11

Distribution of time span between consecutive events passing microphonic cuts (see text). A small deviation from a Poisson distribution is observed at $t<12\mathrm{s}$. A large fraction of events in the first bin corresponds to the decay of cosmogenic ${}^{73}\mathrm{As}$, involving a short-lived ($t_{1/2}=0.5\mathrm{s}$) excited state [4, 12].

Figure 12

Gray-scale plot showing the distribution of rise time vs energy for events passing all other cuts, collected over a 27 month live period for the detector at SUL. Fast bulk events appear highly concentrated around a $\sim 325\mathrm{ns}$ rise time, their distribution becoming progressively slower toward zero energy by the effect of electronic noise in preamplifier traces (Sec. 4b), already visibly affecting the cosmogenic peaks around 1.3 keV. The dotted red line corresponds to the 90% acceptance boundary for fast electronic pulse events, used for rise time cuts in [4, 5].

Figure 13

Event-by-event comparison of energy estimators from UC and UW data analysis pipelines (442 day data set, [5]). The top panel shows that the two energy estimators are very well correlated. The bottom panel indicates that the maximum difference between energy estimators is $<4\%$ above the analysis threshold of 0.5 keV.

Figure 14

Event-by-event comparison of rise time estimators from UC and UW data analysis pipelines (442 day data set [5]) in the region 0.5–3.0 keVee. The top panel (a) shows the correlation between the two estimators. The fractional difference between the two estimators is shown in (b). The two rise time estimators are fairly well correlated, with a disagreement in the classification as fast (bulk) or slow (surface) for only 11% of the events in the 0.5–3.0 keVee analysis region. In the region of 0.5–1.0 keVee this disagreement affects 16% of the events.

Figure 15

Comparison of energy spectra and overall event rate from the UC and UW analysis pipelines. Panel (a) shows the similar energy spectra obtained following independent data selection and rise time cuts. Panel (b) displays the daily rates in the region 0.5–3.0 keVee for events passing all cuts.

Figure 16

Comparison between irreducible monthly rates in two different energy regions, for the UC (black) and UW (circles) analysis pipelines. The correction for low-energy cosmogenics present in these regions [5] is applied and calculated independently for each pipeline.

Figure 17

True coincidences between muon veto and PPC appear as an excess above spurious coincidences, displaying the typical delay by a few $\mu \mathrm{s}$ characteristic of fast neutron straggling. See text for a discussion on the comparison of their rate with that predicted by simulations. Inset: fraction of events removed by a muon veto cut (see text).

Figure 18

Effect of the application of a veto coincidence cut on the monthly irreducible event rate (see text). White circles incorporate this cut following all other data cuts, as opposed to the inset of Fig. 17, where it is applied directly on uncut data. This leads to minor differences in the obtained reduction in event rate. The energy range for this figure is 0.5–3.0 keVee.

Figure 19

Simulated preamplifier pulses with an initial rise time of 325 ns, representing ideal fast (bulk) events, are convoluted with electronic noise and treated with the same wavelet denoising and rise time measurement algorithms applied to real events. This electronic noise is grafted directly from pretrigger preamplifier traces taken from real detector events, leading to perfect modeling of the noise frequency spectrum. The resulting rise time distributions are represented by red curves, labeled by their energy equivalent. The same is repeated for typical slow (surface) pulses with a rise time of $2\mu \mathrm{s}$, generating the blue curves. Each simulation contains 35 k events. These simulations provide a qualitative understanding of the behavior observed in Fig. 12.

Figure 20

Example rise time distributions for events falling within discrete energy bins, from a 27 month exposure of the CoGeNT detector at SUL. These are fitted by two log-normal distributions with free parameters, corresponding to slow rise time surface events (blue) and fast rise time bulk events (red). Small vertical arrows point at the location of the 90% C.L. fast signal acceptance boundary dictated by electronic pulser calibrations (dotted red line in Fig. 12). A contamination of the events passing this cut by unrejected surface events progresses as energy decreases (see text).

Figure 21

Fraction of events passing the 90% fast signal acceptance cut (pulser cut, dotted red line in Fig. 12) identified as true bulk events via the analysis discussed in Sec. 4b. Alternatively defined, its complement is the fraction of events passing the pulser cut that are in actuality misidentified surface events (see Fig. 20). The dotted line is a fit with functional form $1-e^{-a·E(\mathrm{keVee})}$, with $a=1.21\pm 0.11$. Error bars are extracted from the uncertainties in fits like those exemplified in Fig. 20.

Figure 22

Steps in the treatment of a low-energy CoGeNT spectrum. (a) Spectrum following data selection cuts (Sec. 4), including 90% fast signal acceptance cuts from pulser calibrations (dotted red line in Fig. 12) [4, 5]. This spectrum is nominally composed by a majority of bulk events. Overimposed is the combined trigger and background cut efficiency. This efficiency is derived from high-statistics pulser runs (Figs. 7 and 10), resulting in a negligible associated uncertainty. (b) Spectrum following this trigger plus cut efficiency correction. Overimposed is the residual surface event correction. This correction and its associated uncertainty can be found in Fig. 21. (c) Spectrum following this surface event contamination correction. Overimposed is the predicted cosmogenic background contribution, reduced by 10% as in [5]. The modest uncertainties associated with this prediction, dominated by present knowledge of L/K shell electron capture ratios, are discussed in [5]. (d) Irreducible spectrum of bulk events, now devoid of surface and cosmogenic contaminations [28, 56]. Overimposed is the expected signal from a $m_{\chi }=8.2\mathrm{GeV}/c^2$, ${\sigma }_{SI}=2.2\times {10}^{-41}{\mathrm{cm}}^2$ WIMP, corresponding to the best fit to a possible nuclear recoil excess in CDMS germanium detector data [60]. A bumplike feature around 0.95 keVee is absent in the alternative UW analysis shown in Fig. 15 and is therefore likely merely a fluctuation.

Figure 23

Irreducible spectrum of bulk events (points) showing cumulative uncertainties from the corrective steps discussed in Fig. 22. The simulated total background spectrum from Sec. 5 is shown as a histogram, scaled to the larger exposure in this figure, and corrected for the combined trigger and background cut efficiency.

Figure 24

The 90% C.L. WIMP limits extracted from the irreducible bulk event spectrum in Fig. 23, placed in the context of other low-threshold detectors. A Maxwellian galactic halo is assumed, with local parameters $v_0=220\mathrm{km}/\mathrm{s}$, $v_{\mathrm{esc}}=550\mathrm{km}/\mathrm{s}$, $\rho =0.3\mathrm{GeV}/c^2{\mathrm{cm}}^3$. A ROI (red solid 90% C.L., red dashed 99% C.L.) can be extracted if a WIMP origin is assigned to the rise in the spectrum. This ROI includes the cumulative uncertainties shown in Fig. 23 and allows for a flat background component, independent of energy, in addition to a WIMP signal. Deviations from this background model, or changes in the choice of rise time cuts, can minimally displace and/or enlarge this ROI. The reader is referred to [57] for a discussion on astrophysical uncertainties also not included here (see also [5, 56]). This ROI partially overlaps with another one, not shown here for clarity, extracted from a possible excess of low-energy nuclear recoils in CDMS germanium data [60]. A best fit to that possible excess is shown in the bottom panel of Fig. 22. Recent low-mass WIMP limits from CDMS-Ge [61], EDELWEISS [62], TEXONO [63], MALBEK [64], and CDMS-Si [65] are indicated. A blue asterisk indicates the centroid within a large ROI generated by an excess of three nuclear-recoils in CDMS silicon detector data [65].

Figure 25

MCNP-Polimi neutron simulations compared with an early spectrum from the CoGeNT detector at SUL (see text).

Figure 26

Fraction of external $(\mu ,n)$ cavern neutrons giving rise to energy depositions in the 0.5–3.0 keVee energy window of the CoGeNT detector at SUL, as a function of incident neutron energy, derived from a Monte Carlo simulation (open circles). Also shown, using the right-hand scale, is the emission energy spectrum for these neutrons, taken from [33] (histogram).

Figure 27

Deposited energy spectra from all known backgrounds in the CoGeNT detector at SUL, compared to the 442 days of data in [5]. An unidentified low-energy excess and L-shell EC cosmogenic contributions are visible [4, 5]. The corrections in Fig. 22 reduce this excess by $\sim 30\%$ at 0.5 keVee. The blue band represents the sum of muon-induced backgrounds (Sec. 5a1), the green hatched band is a conservative upper limit to the background from cosmogenic ${}^3\mathrm{H}$ (Sec. 5b), and the red band is from $(\alpha ,n)$ natural radioactivity in cavern walls (Sec. 5a2). The solid line represents the background distribution from the ${}^{238}\mathrm{U}$ and ${}^{232}\mathrm{Th}$ chains as well as ${}^{40}\mathrm{K}$ contamination in the front-end resistors, estimated in Sec. 5d2. The dashed line is the sum of all background contributions. Contributions from bremsstrahlung from ${}^{210}\mathrm{Pb}$ in the inner lead shield (Sec. 2a) and radioactivity from cryostat parts (Sec. 5d1) are found to contribute negligibly.

Figure 28

Negligible upper-limit to the contribution from cosmogenic activity in the near-threshold energy region of the CoGeNT detector at SUL (see text).

Figure 29

Counts per 30 day bins from the 0.5–3.0 keVee CoGeNT energy window (black dots) compared to the MINOS radon data at SUL (dashed line), averaged over the period 2007–2011, exhibiting a peak on August 28 [34, 35]. The solid curve represents a sinusoidal fit to CoGeNT data. An analysis by the MINOS Collaboration finds a three-sigma inconsistency between the phase of their measured seasonal modulation in radon concentration at SUL and CoGeNT data [26].

Figure 30

Similar to Fig. 27, with expanded ranges: energy spectrum of the simulated ${}^{238}\mathrm{U}$, ${}^{232}\mathrm{Th}$, and ${}^{40}\mathrm{K}$ resistor background (dotted line) compared to CoGeNT data (solid line). In the energy range displayed the estimated resistor backgrounds are by far dominant. The resistor background spectrum is for metal film resistors, the same used in the CoGeNT front end. Also shown are other background contributions and their sum. Contributions from ${}^{210}\mathrm{Pb}$ bremsstrahlung and radioactivity in PTFE and OFHC cryostat parts are comparatively negligible.

Figure 31

Existing CoGeNT data in the range up to 300 keV, with possible weak ${}^{212}\mathrm{Pb}$ (238 keV) and ${}^{214}\mathrm{Pb}$ (295 keV) gamma lines indicated by arrows. The extrapolated energy scale can only be considered approximate. The energy binning corresponds to the approximate FWHM resolution for these two lines. See text for a discussion on a possible origin for these putative lines in the front-end resistors. Notoriously absent are a ${}^{210}\mathrm{Pb}$ peak at 46.5 keV and excess lead x rays, a result of the radiopurity of the inner lead layers in the shield (Sec. 2a) and detector surfaces (Sec. 5c).