The Skipper CCD for low-energy threshold particle experiments above ground

Charged Coupled Devices (CCD) with low readout noise and large active volume have been identified among the most promising detector technologies for the low-mass direct dark-matter search experiments, probing electron and nuclear recoils from sub-GeV mass [1, 2, 3, 4, 5]. The recent development of the Skipper-CCD [6, 7, 8] demonstrated the capability to measure ionization events with sub-electron noise extending the reach of this technology to unprecedented low energies. This enabled world leading results for dark matter searches at underground facilities [8]. Experiments based on this technology are planned for the coming years with active masses going from $100$ grams to several kilograms [9, 10], running at underground facilities that allows to rule out most of the cosmic background. At the same time, the low-noise CCD technology has been implemented in low-energy neutrino experiments [11, 12] above surface; implementing Skipper-CCDs is also planned for future developments [13].

CCD sensors to be used in future instruments are part of a significant R&D effort [9, 10, 13]. One of the most important performance requirement to fully exploit the single counting capability of the Skipper-CCD is the control of the background signals below $1$ keV. Rates of single electron depositions larger than expected have been observed in [8], and its possible link to higher-energy particles has been summarized in [14]. Moreover, recent studies on regions in the CCD with low collection efficiency [15] show how large-energy ionization events could produce low-energy signals in the detector. For this reasons, experiments running above ground are more challenging due to the larger rate of atmospheric high-energy particles [16, 17]. Several new low-threshold technologies observe a rapidly increasing rate of background events towards low energies (below 1 keV). This effect is more evident when detectors are not operated underground. Larger background rates result in limiting the reach of the low-threshold capability. The first edition of the EXCESS workshop in 2021 [18] exhibit the effort of the community to overcome this issue.

In [19] it is shown that a good control of the background starting at 15 eV up to 500 eV is a key factor for new physics searches using neutrinos at nuclear reactors. Moreover, other studies such as the silicon energy absorption for photons and neutrons scatterings are critical and require Skipper-CCDs running at the surface level with controlled levels of background. It is fairly known that the long readout time of these sensors does not allow the use of active veto systems. For this reason, passive shield configurations together with event selection criteria based on the spatial resolution of thick fully-depleted CCDs are used to reject undesired signals.

In this article we study the use of Skipper-CCDs as a potential low-threshold technology for experiments above surface. We discuss the main sources of background found on the measurements. We propose a detector operation scheme and selection criteria to reject these contributions. We demonstrate that with the proposed technique, the low energy excess in the background rate seen by other experiments [18] is not present using the Skipper-CCD technology, even down to events producing an ionization of $5$ electron-hole pairs (just electrons from now on). Finally, we compare the background measurements observe by other low threshold technologies already published that have participated in the EXCESS workshop.

Figure 1 (a) shows the experimental setup where some of the main components are labeled. One Skipper-CCD (shown in Fig. 1 (b)) was operated at a temperature of $140$ K using a Sunpower cryocooler [20]. The CCD is glued to a silicon substrate that sits on a Copper tray for mechanical support as well as thermal connectivity. The CCD is placed in an extension of the dewar that fits inside a lead cylinder. A lead cap on top of the sensor (inside the dewar extension) completes the lead shield of two inches of thickness around the device. There was no radio purity selection of materials inside the shield. The CCD was designed by the LBNL Microsystems Laboratory (MSL) [21] and fabricated at Teledyne-DALSA. It has $6144$ columns by $1024$ rows with pixels of $15$ $\mu$m by $15$ $\mu$m with a thickness of $675$ $\mu$m. It is read by four amplifiers, one on each corner, using a Low Threshold Acquisition (LTA) controller [22]. Two quadrants presented larger readout noise and were not used for the analysis in the following sections. The sensor was operated at sub-electron noise by averaging $300$ measurements of the charge in each pixel [7] and with a horizontal binning [23] of $10$ columns.

Two different kind of measurements were performed:

• •

  DATA-SET A: Interactions in the active region. $3.21$ days of data were collected from the active region of the sensor in continuous readout mode [8].

• •

  DATA-SET B: Interactions collected in the serial register (also called horizontal register). It consists of charge packets collected only by the serial register, not coming from the active region, from here on called “serial-register events”¹¹1Later it will be shown that these depositions dominate the low energy spectrum in the output data. This charge is generated in the inactive silicon surrounding the active volume on the edges of the die. The charge can diffuse from the undepleted region and eventually reach the collection region of the pixels in the serial register. For more information about the sensor electrostatic see [24].. To make an isolated measurement of this source of events, the vertical lines of the sensor were left in idle mode while clocking only the control signals from the serial and output stage. In other words, the sensor is read out as in DATA-SET A but without moving the charge from the active area to the serial register. 2.7 days were collected in this mode.

An absolute energy calibration of the sensor is performed using the electron counting capability. A histogram of the pixel values from the active region (DATA-SET A) is produced as shown in Fig. 2. Each peak correspond to a discretized number of electrons in the pixel after the calibration and is fitted using a Gaussian distribution whose mean value is used to build a look-up table (digital unit vs. electrons) to calibrate the sensor up to around $700$ $e^{-}$. To calibrate the sensor at $2146$ $e^{-}$, single pixel X-ray events with energy of 8.048 keV produced by the fluorescence of the surrounding copper material are also used. We assume an average energy deposition per collected electron of 3.75 eVee [25].

The readout noise of the sensor, evaluated as the standard deviation of the values of the empty pixels ($0$ $e^{-}$ peak in Fig. 2), is 0.165 $e^{-}$ and 0.167 $e^{-}$ for the two quadrants in use. The average single electron rate per pixel measured are $0.01$ $e^{-}$/pix and $0.009$ $e^{-}$/pix after binning in each quadrant, with a contribution from the exposure dependent charge production [26] of $0.018$ $e^{-}$/pix/day (pixels before binning). The diffusion charge transport of carriers in the bulk of the sensor was measured and it turned out to be similar to the previously published in [8].

Figure 3 shows the shape and energy distributions of the events analyzed. Left plots show the event size in the $y$-direction (rows of the image), $\sigma_{y}$ (determined as the standard deviation of the charge distribution in the pixels of the event) as a function of energy for three different data sets: the two previously discussed and a third set generated from a simulation. The plots on the right show the relationship between the size ($\sigma_{x}$) in $x$-direction (columns of the image) and $\sigma_{y}$.

Figure 3(a) and (b) correspond to simulations of particle energy depositions in the active region (bulk) of the sensor. Spatially and energetically uniformly distributed events ($30000$ events) of charge ionization from $5$ to $800$ electrons were simulated in the bulk [27] using the charge transport calibration curve in [8]. Since the charge of ten columns is collapsed into one due to the horizontal binning, there is little information left concerning the shape of the event in the $x$-direction of the output images as shown in Fig. 3 (b). Some events present $\sigma_{x}>0$. They take place on the edge of the binned regions and have information shared between two binned pixels and their probability of occurrence is low.

From Fig. 3(a) we can see that the $\sigma_{y}$ distribution is similar for all energies. Most of the events have a $\sigma_{y}$ larger than $0.4$ pixels (see color scale) with a maximum of about $1.1$ pixels due to the maximum expected lateral diffusion.

Figure 3(c) and (d) shows the distribution of events measured in the serial register (DATA-SET B). A dark-current signal following a Poisson distribution with mean equal to $0.01$ $e^{-}$/pixel (after binning, as expected in DATA-SET A) was added to emulate both contributions together. In contrast to the events in the bulk, these ones show a $\sigma_{y}$ distribution close to $0$. Since the charge is only collected in one row (the serial register), serial-register events have practically no spatial information in the $y$-direction. They extend on the $x$-direction projecting on the $\sigma_{x}$ axis. Low-energy events may appear with an increased size in the $y$-direction due to dark counts in neighboring pixels paired in the same cluster.

The third group in Fig 3(e) and (f) corresponds to measurements in the active region of the sensor (DATA-SET A). The plots show components of the previously mentioned contribution: events from the bulk (as in the simulations) and from the serial register. The firsts have sizes below $1.1$ pixels in $y$-direction and smaller than $0.5$ pixel in $x$-direction, as expected from the charge transport calibration and event simulations, while the latter populates both plots for small $\sigma_{y}$.

As shown in the previous analysis, the proposed horizontal binning provides the advantage of orthogonalizing the size information of both sources of events, and therefore a more pure selection of bulk events can be made. It also accelerates the total readout time needed to read the pixel charge while staying in the serial register, and thus it reduces the exposure time to this source of spurious events.

Figure 4(a) shows the measured differential spectra of events in the active region with $5$ or more electrons of ionized charge (from DATA-SET A). Columns of the CCD that presented high single electron rate (hot columns [23]) were eliminated from the analysis at an early stage. The remaining active mass of the sensor in use is $0.675$ grams. The gray plot depicts all events reconstructed from the output images showing a monotonic increase of the rate towards low energy. The peak at around $8$ keV corresponds to the fluorescence X-rays from the copper in the mechanical package of the sensor. To study the impact of the results obtained in the previous section a selection criteria based on the event shape was performed. Figure 5 shows the one dimensional distribution of the size, in $x$ and $y$ directions, for the events simulated in the active region of the sensor. Although events in the bulk can show a size in $\sigma_{x}$ up to 0.5 pixels (Fig. 5(a)), the majority of them (72%) have a width smaller than 0.1 pixels. A constraint of $\sigma_{x}<0.3$ is used to provide strong rejection to the serial-register and dark-current events. At the same time, only sizes in the y-direction compatible with depositions in the bulk are selected: $0.3<\sigma_{y}<1$. The minimum limit is to prevent counting serial-register events. The spectrum of events after the selection cuts is also shown in blue in Fig. 4(a). Most of the low energy excess was removed because its contribution was coming from the serial register. The average flux of events with equivalent deposited energy from 5 $e^{-}$ to 2000 $e^{-}$ (7.5keV) is $12$ kdru. The plot does not show any significant increase of events towards low energies. The spectrum is normalized by the active mass of the sensor, but not by the efficiency of the selection criteria.

To evaluate the efficiency of the selection criteria, simulated events from interactions spatially and energetically uniformly distributed in the active volume of the sensor were added to the real images. The diffusion transport model of carriers were also used in this simulation. The percentage of reconstructed events passing the selection criteria are shown in Fig. 4(b). Each energy bin is $50$ eV wide and the first bin starts at $15$ eV (equivalent to 5 electrons included). The efficiency is almost flat for all energies showing that the final shape of the spectra is not modified by the selection criteria. The red curve shows a linear fit of the points.

The results in this work compared to other current technologies for low threshold experiment are shown in Fig. 4(c). The plot was produced using the online tool provided by organizers of the EXCESS workshop [18]. Only data-sets from measurements above ground with published results are compared in the plot. Table 1 summarizes the references to the articles were the detector and data analysis information is presented. All spectra are normalized by efficiency. Our results do not show the rapidly increasing rate of events observed for the other technologies for energies below 1 keV. Moreover, for energies below 500 eV where the neutrino signal from nuclear reactors is expected, our sensor shows the lowest background rate demonstrating the potential of the Skipper-CCD technology for this application and other low-energy measurements.

Experimental results using a single-electron resolution Skipper-CCD running above ground have been presented. We demonstrated the potential of this technology for reactor neutrinos and other low-energy particle interaction experiments. The main background source of events has been identified and a technique to decouple its contribution has been presented. Our final result for events with energies of $5$ ionized electron-hole pairs was compared to other state-of-the-art low-energy-threshold technologies showing that the exponentially increasing rate of events observed is not present in our data as shown in Fig. 4(c).