SuperCDMS Public Documentation

HVeV-0VeV Cross Analysis


Abstract This Public Documentation accompanies the study of background from a 3 week long run of a 1g HVeV detector in the Northwestern ADR. During this run, one week of data (roughly 12 hours per day for 7 days) was acquired at 100V bias with a charge resolution of 0.03 electron-hole pairs, and 3 days at 60V with a resolution of 0.05 electron-hole pairs. A 0V control sample was also acquired over multiple days. The comparison of 0V, 60V and 100V data helps us understand the background origin. We show that the pulse shape and energy spectra are compatible between 0V and high voltages, under the assumption of electron-recoil. The resulting limits for DM-nucleus scattering using 0V data is given below, along with plots approved for public release.
These results are published in Phys. Rev. D 105, 112006 – Published 22 June 2022




All results assume a relic DM density of ρDM = 0.3 GeV/cm3. All limits are at 90% confidence level. σSIN is spin-independent dark matter nucleon cross section.

Num. Result Topic Last Updated Download Comments
1 Experimental Setup August 16th, 2019 png, pdf Photograph of the HVeV R2 detector

Picture of the HVeV R2 detector (left) with side RF veto (right) inside of the copper housing.
2 Experimental Setup January 30th, 2020 png, pdf Photograph of the HVeV R2 detector

Picture of the HVeV detector seen from the side
3 Experimental Setup January 30th, 2020 jpg, pdf Photograph of the HVeV R2 detector installed in the Adiabatic Demagnetization Refrigerator (ADR)

Picture of the HVeV detector installed in the ADR at Northwestern University (ArXiv:1903.06517). The ADR has two salt pills, Gadolinium Gallium Garnet (GGG) operating around 300 mK and Ferric Ammonium Alum (FAA) operating around 50 mK. The detector is thermally linked to the FAA salt, with temperature stabilized to 50 mK or 52 mK during the operation. A superconducting magnetic field shield, the Nb can, is mounted on the FAA stage around the detector box. The cold electronic readout device, SQUIDs, are mounted on a thermal stage operated at 1.3 K.
4 Experimental Setup January 30th, 2020 jpg, pdf Photograph of the experimental set-up at the Northwestern University

Photograph of the experimental set-up at the Northwestern University, including the ADR, the computers controlling the cryostat, the data acquisition system.
5 Experimental Setup February 21st, 2020 png, pdf Drawing of the sensor mask

Drawing of the sensor mask used for the HVeV R2 detector. Two channels with similar area are visible and their contacts highlighted with darker squares.
6 Data Acquisition February 24st, 2020 png,pdf Exposure

Summary of total exposure as a function of wall time during the run by data type. This analysis takes into consideration the data acquired at 0V, 60 V and 100 V. Two days of 0 V data were acquired but at two different temperatures, only one of these two days was used for this analysis. The laser data and 55Fe data are not reported in the plot because the exposure was negligible with respect to the other data.
7

Events February 4th, 2021 png, pdf
png, pdf
png, pdf
Example of a burst event at 60 V (top) and of a RF one (middle), and a long decay time event at 0 V (bottom)

The burst events are visible in the high voltage data. They are characterized by a primary pulse followed by one electron-hole pair events that are close in time. The study of the burst events is one of the main objective of this analysis.

The RF events is caused by electro-magnetic interference of various radio devices. Our detector is not fast enough to resolve the GHz scale frequency, so there events appear as wave packet and usually have a square shape.

Some of the 0 V events have longer decay time comparing to the laser pulse template. The event shown here is one of the events in Fig. 26 top panel. More details about how these events are selected are shown in Fig. 25.

8 Calibration February 24th, 2021 png, pdf Calibration curve

The data were calibrated using an energy estimator based on the integral of the events. We used laser data in order to calibrate the low energy region. The high energy region was calibrated with 55Fe data acquired at a NTL bias of 40 V, 50 V and 60 V. The black curve is the one used for calibrating the data. The bottom panel shows the residuals between the calibration data points and the calibration curve.
9 Calibration February 4th, 2021 png, pdf 55Fe data used for the calibration

The 55Fe data used in the calibration were acquired with a Neganov-Trofimov-Luke bias at 40 V, 50 V and 60 V.
10
Calibration February 25th, 2021 png, pdf
png, pdf
Laser calibration data before and after the daily correction

The ADR required daily cycles between 4 K and the working temperature (50-52 mK). Daily laser calibration were required to ensure a correct calibration. These two plots show all the laser data acquired with a NTL bias of 100 V before and after the correction between days.
11 Live-time cuts February 25th, 2021 png, pdf
png, pdf
Mean baseline and fridge temperature live-time cut on 60 V data

The data passing these two cuts are represented with a 2D histogram and the data failing the cut are represented in gray.
12 Live-time cut February 10th, 2021 png, pdf
png, pdf
120 Hz phase of the events as a function of time.

120 Hz phase of the events as a function of time: diagonal overpopulated clusters are caused by the periodic 60 Hz and 120 Hz noise. The orange error bars in the bottom plot highlight the average and the circular standard deviation for 3-second intervals.
13 Live-time cut February 10th, 2021 png, pdf Live time cut used to remove periodic noise

The distribution of intervals between adjacent events shows pronounced peaks with a period corresponding to 60Hz noise, as well as broader peaks corresponding to 120 Hz. The two histograms show the data before and after the live-time cut used to remove periodic noise
14 Live-time cuts February 5th, 2021 png, pdf Summary of live-time cuts for 100 V, 60 V and 0 V data.

The fridge temperature cut and the mean baseline cut were applied to all the data (0 V, 60 V and 100 V). The 60 Hz cut was applied only to the 0 V data.
15 Data quality cut February 5th, 2021 png, pdf
png, pdf
png, pdf
Average pulse width cut for 100 V (top), 60 V(middle) and 0V (bottom)

This quality cut was developed to remove RF noise from the data. The data are represented with a 2D histogram and the gray area shows the data out of the region of interest (ROI). The red area represents the area that would be removed by the cut. This cut was applied only on the 0 V data because it mainly affects the data out of the ROI for the HV data.


16 Data quality cuts February 5th, 2021 png, pdf
png, pdf
png, pdf
Time-domain χ2 cut for 100 V (top), 60 V(middle) and 0V (bottom)

This quality cut was developed to trigger artifacts from the data. The data are represented with a 2D histogram and the gray area shows the data out of the region of interest (ROI). The red area represents the area that would be removed by the cut. This cut was applied only on the 0 V data because it affects the data out of the ROI for the HV data.
17 Efficiency February 5th, 2021 png, pdf Cut efficiency of 0 V data for quality cuts that are used for comparison between 0V and HV data

The quality cuts are applied only on the 0 V data and therefore the efficiency was evaluated only for the 0 V data. In order to take into account the presence of both laser-like events (so fast events) and long-tail events, the efficiency was evaluated both with the laser data and the simulations. The shaded region are the uncertainty band of the efficiency estimations. An envelope of the two efficiencies was considered in the analysis.
18 Data February 5th, 2021 png, pdf
png, pdf
png, pdf
Energy specta after applying quality cuts for 100 V (top), 60 V(middle) and 0V (bottom)

These three spectra show the effect of the cuts for the 0 V, 60 V and 100 V data. The gray area represents the energy region that was not considered in this analysis. The χ2 and average pulse width cut were not applied in the final analysis for the 60 V and 100 V data because they are effective only out of the region of interest (ROI).
19 Data February 8st, 2021 png, pdf Example of dt calculation

The dt parameter is calculated as the time distance between two consecutive pulses. The identification of the pulse is done with a gaussian derivative filter. This plot shows the pulse identification and relative distance between the pulses in the case of a burst event.
20 Data February 8st, 2021 png, pdf dt distribution

Distribution of the dt parameter (see Fig. 19 for the definition) for all the events in the HV run. The orange fit shows the poissonian behavior of the distribution for large dt. Smaller dt values have an excess that deviates from the poissonian distribution. The red region is zoomed in the inset panel. The blue area represents the trace length used in the analysis: the ratio between the poissonian-distributed events and the total number of events is of the order of 1:100. We expect that the majority of pile-up events in a 5.4 ms trace is due to burst events.
21 Data February 25st, 2021 png, pdf Energy spectra of 100 V burst and non-burst event

For the burst event, the reconstructed energy is close to the energy of the primary pulse.
22 Data February 25st, 2021 png, pdf Time distribution of secondary pulses in the burst event

This plot represents the distribution of the time of the secondary pulses with respect to the primary pulse.

23
Data February 25th, 2021 png, pdf
png, pdf
Energy spectra of burst and non-burst event for 100 V (top) and 60 V (bottom)

These two plots consider only the burst events at 60 V and 100 V. Every pulse in the burst events is fitted and two histograms are considered: the distribution of the primary events and of the secondary events. The energy of the secondary events is quantized in 1, 2, 3 electron-hole pairs. The 2 and 3 electron-hole pair peaks are compatible with pile-up of the 1 electron-hole pair events.
24
Data February 25th, 2021 png, pdf
png, pdf
Reconstructed energy of secondary pulses in burst and non-burst event for 100 V (top) and 60 V (bottom)

Energy distribution of both the secondary events in a burst and single electron-hole lekeage events, zoomed around the energy of 1 electron-hole pair. The total area of the histograms are normalized to 1.
25

Data selection February 25th, 2021 png, pdf
png, pdf
Event selection for 0 V and 60 V data used in Fig. 26

Time-domain χ2 as a function of the energy in the case of the 0V and 60 V. The black-grey colored are 2-D histogram of all the events after quality cuts. The data highlighted with different colors are selected and reported in Fig. 26. Selection creteria for each group of events:
  • Pink selection on 0 V data:
    energy range: 25-100 eV and time-domain χ2 < 2
  • Orange selection on 0 V data:
    energy: > 25 eV and slope > 0
  • Green selection on 0 V data:
    energy range: 100-800 eV and time-domain χ2 > 2
  • Red selection on 60 V data:
    energy range: 675-2700 eV and number of pile-up pulses in a trace > 0
  • Blue selection on 60 V data:
    energy range: 2200-2700 eV and number of pile-up pulses in a trace = 0
26 26.(1) Green events in Fig. 25: 100 eV < energy<800 eV and time-domain χ2 > 2:

26.(2) Orange events in Fig. 25: 25 eV < energy and slope > 0:

26.(3) Pink, red and blue events in Fig. 25:

26.(4) Pink, red and blue events in Fig. 25:

Traces February 5th, 2021 png, pdf
png, pdf
png, pdf
png, pdf
(Average) Pulse shape of selected events in Fig. 25
  • Panel 1: Green events in Fig. 25. Traces are plotted with transparency to show the overall longer fall time comparing to laser pulse shape.
  • Panel 2: Orange events in Fig. 25. Those events have a second/slow pulse that has ms-scale fall time. Note that the time scale is different comparing to Panel 1.
  • Panel 3,4: Pink, red and blue events in Fig. 25.
    The black curve is the composed by the average of laser events used as the average pulse for the optimum filter. The blue events correspond to non-burst 60 V events that start to be saturated. The pink events are 0 V events with a low χ2 and the red events correspond to burst events at 60 V.
27 Data April 9th, 2021 png, pdf
png, pdf
Example of slow pulses triggered with a slow pulse template

The data is processed with trigger that uses the averaged slow pulse in Fig. 26 panel 2 as trigger template. The top panel contains pulses from 100 V data, the bottom panel contains pulses from 0 V data.
28 Data and simulations February 25th, 2021 png, pdf
png, pdf
png, pdf
Comparison of the 0 V data and simulations

Time-domain χ2 as a function of the energy. The plots compare the 0 V data (in blue) with the simulation of the burst events. The burst event simulation are divided in three cases: nominal number of secondary events, half and double number of secondary events.
29


Data February 5th, 2021 png, pdf
png, pdf
png, pdf
Comparison of 100 V, 60 V and 0V data

The 60 V and 100 V data are divided by the NTL gain assuming three different εeff (infinite, 7.6 eV and 6 eV). εeff is the effective charge pair creation energy and is equal to εeh/yield. εeh is the average charge pair creation energy for electron recoil in silicon. We use 3.8 eV/eh. For each εeff, the corresponding yield value y is shown in the parenthesis.

The shadded region (25-100 eVr) is where the comparison is focused on. At energy > 100 eVr, the spectrum is dominated by Compton scattering; at energy < 25 eVr, the 0 V data is affected by increasing noise trigger.
30 Quality cuts February 8st, 2021 png, pdf
png, pdf
Response matrices of four simulations at εeff = 4 eV.

The response matrices convert the energy estimator used in the HV data — which is based on the pulse intergal — to the optimum filter that is used in the 0V data. The matrices are built using the simulations assuming 4 cases: (1) nominal number of secondary pulses; (2) double of secondary pulses; (3) half of the secondary pulse; (4) no secondary pulses. Red line is an identity line, indicating the case in which the OF energy and the MF energy are identical.
The bottom pannel shows only the "nominal" setting of secondary pulses in the simulation.
31
Data February 8st, 2021 png, pdf
png, pdf
png, pdf
png, pdf
Comparison of spectra between 0 V and high voltage data (60 V and 100 V)

If we assume that the burst events visible in the HV data are also present in the 0V data, we can compare the two spectra and check the agreement between the two. The gray area represents the region in which the comparison is done. The 0 V data are fit with three exponents and a constant plateau to simplify the comparison. These two sets of plots assume the following values of εeff: 4 eV and 5 eV.
32 Quality cuts February 1st, 2021 png, pdf
png, pdf
Frequency-domain and time-domain χ2 cuts used for the limit setting analysis

Given the low treashold reached by this detector, we decided to set a dark matter limit with the 0 V data. We developed two χ2 cuts for this analysis in order to exclude the burst events. The laser data at 60 V were used to develop the cut, in order to avoid a bias by cutting directly on the 0V science data. The maximum of the χ2 distribution for each peak is highlighted with a yellow point. The red points indicate a +3 σ distance from the maximum of the χ2 distribution. The red points are fit with a constant line in red which is used as the cut in the 0 V data.
33 Cut Efficiency January 1st, 2021 png, pdf Cut Efficiency for the 0 V data

The cut efficiency was calculated with 60 V laser data. The top panel repesents the laser data before and after the two χ2 cuts. The ratio between the two spectra is reported in the bottom panel. The efficiency data are fit with a constant line in red. The gray band around the fit line represents the systematic uncertainty on the efficiency.
34 Trigger efficiency February 10th, 2021 png, pdf Trigger efficiency from simulation

The red line corresponds to a 50% trigger efficiency and is the trigger threshold used to set the nuclear recoil dark-matter limit.
35 Data February 10th, 2021 png, pdf 0 V spectrum before and after the χ2 cuts used in the limit setting

The events removed by the χ2 cuts are the RF-induced events and events with a long-decay time.
36 Data February 10th, 2021 png, pdf Optimum Interval limit setting illustration

The energy intervals chosen by the OI to set the limit on are shown as colored vertical bands for several dark matter masses and the corresponding signal models are shown scaled to the resulting cross section limits.
37 Limit February 11th, 2021 png, pdf
png, pdf
0VeV spin-independent dark matter nuclear recoil cross-section limit

Top panel: 0VeV spin-independent dark-matter nuclear cross section limit. Bottom panel: 0VeV spin-independent dark-matter nuclear cross section limit including the upper bound due to atmospheric shielding


References for Fig. 36: