GEM-based TPC with CCD Imaging for Directional Dark Matter Detection
N. S. Phan, R. J. Lauer, E. R. Lee, D. Loomba, J. A. J. Matthews, E. H. Miller
GGEM-based TPC with CCD Imaging for Directional Dark Matter Detection
N. S. Phan ∗ , R. J. Lauer, E. R. Lee, D. Loomba, J. A. J. Matthews, E. H. Miller Department of Physics and Astronomy, University of New Mexico, NM 87131, USA
Abstract
The most mature directional dark matter experiments at present all utilize low-pressure gas Time Projection Chamber (TPC) tech-nologies. We discuss some of the challenges for this technology, for which balancing the goal of achieving the best sensitivity withthat of cost e ff ective scale-up requires optimization over a large parameter space. Critical for this are the precision measurementsof the fundamental properties of both electron and nuclear recoil tracks down to the lowest detectable energies. Such measurementsare necessary to provide a benchmark for background discrimination and directional sensitivity that could be used for future opti-mization studies for directional dark matter experiments. In this paper we describe a small, high resolution, high signal-to-noiseGEM-based TPC with a 2D CCD readout designed for this goal. The performance of the detector was characterized using alphaparticles, X-rays, gamma-rays, and neutrons, enabling detailed measurements of electron and nuclear recoil tracks. Stable e ff ectivegas gains of greater than 1 × were obtained in 100 Torr of pure CF by a cascade of three standard CERN GEMs each with a140 µ m pitch. The high signal-to-noise and sub-millimeter spatial resolution of the GEM amplification and CCD readout, togetherwith low di ff usion, allow for excellent background discrimination between electron and nuclear recoils down below ∼
10 keVee( ∼
23 keVr fluorine recoil). Even lower thresholds, necessary for the detection of low mass WIMPs for example, might be achievedby lowering the pressure and utilizing full 3D track reconstruction. These and other paths for improvements are discussed, as arepossible fundamental limitations imposed by the physics of energy loss.
Keywords: dark matter, WIMPs, TPC, GEMs, CF , CCD, directionality, head-tail e ff ect, nuclear recoil, electron recoil
1. Introduction
The nature of dark matter remains one of the most importantunresolved questions in physics. One of the leading candidatesis a class of particles known as weakly interacting massive par-ticles (WIMPs) [1, 2]. They are the target of many ongoingdirect detection experiments, which aim to measure the signalsleft by the elastic scattering of WIMPs with nuclei in the de-tector target material [3]. A review of the many experimentalsearches for dark matter is found in Ref. [4]. Direct detection islimited by several factors, one being the extremely low WIMP-nucleon interaction cross-sections predicted by extensions ofthe Standard Model [5], leading to a requirement of large detec-tor masses. Another is the featureless, exponentially-falling re-coil energy spectrum expected from WIMP interactions, whichencourages low detection thresholds. These issues are com-pounded by the presence of backgrounds whose signals couldmimic those expected from WIMPs. Although powerful tech-niques have been developed to discriminate and shield againsta majority of these backgrounds, the misidentification of back-grounds for signal continues to plague the field [6, 7, 8, 9].For these reasons the definitive proof of discovery in dark mat-ter searches rests on the detection of specific signatures of theWIMP-nucleus interaction arising from the Galactic origin ofthe WIMPs. ∗ Corresponding author
One such textbook signature is the annual modulation in theinteraction rate caused by the seasonal variation in the relativevelocity of the Earth-bound detector with the dark matter halo[10, 11]. The e ff ect is relatively small (a few %), however, andmany known backgrounds also modulate seasonally [12, 13, 14,15, 16]. Although several experiments have observed an excessabove expected backgrounds [17, 18, 19, 20, 21], the results areinconsistent with null results obtained by others [22, 23, 24].A more robust and definitive Galactic signature is the side-real modulation in the direction of the WIMP flux [11]. Due tothe solar system’s motion around the Galaxy, the flux appearsto come from the direction of the constellation Cygnus, but asthe Earth rotates through a sidereal day, the position of Cygnus,and hence, the direction of the incoming flux changes in thedetector’s frame of reference. This signature is detected as amodulation in the mean nuclear recoil track direction, which ispeaked in the direction opposite to Cygnus. Not only is the di-rectional signature more definitive in separating a dark mattersignal from backgrounds, it is also much larger than the an-nual modulation e ff ect due to the strong angular dependence ofthe nuclear recoil direction [11]. There are currently severalunderground experiments that have varying degrees of sensi-tivity to this directional signature, including DRIFT [25, 26],NEWAGE [27], MIMAC [28], and DMTPC [29]. In addition,there are a number of e ff orts performing R&D on directionality[30, 31, 32, 33, 34, 35]. For a thorough review of directionaldark matter detection see Refs. [36] and [37]. Preprint submitted to September 26, 2016 a r X i v : . [ phy s i c s . i n s - d e t ] S e p . Directional Detection Challenges The main challenge for directional detection is that the lowenergy, 10’s of keV (henceforth, keVr), WIMP-induced nuclearrecoil tracks are extremely short in liquids and solids (10’s -100’s nm). Thus, although R&D is underway to develop tech-nologies for solid [31, 32, 35] and high pressure gas targets,[33, 34], most experiments use low pressure gas targets wheredirectionality has been demonstrated [38, 39]. In this low en-ergy regime, the recoiling nucleus will produce only a few hun-dred to a few thousand ionization pairs in the detection medium,with track lengths on the order of a millimeter even at pressuresbelow 100 Torr (0.13 atm). Consequently, a natural choicefor technology, which is currently employed by all gas-baseddirectional experiments, has been the Time Projection Cham-ber (TPC) invented by Nygren [40]. This allows for full 3-dimensional (3D) reconstruction of the recoil track, togetherwith the flexibility to choose gas targets and operating pres-sures over a broad range. With cubic meter scale TPCs, theDRIFT experiment has pioneered the use of this technology fordirectional searches. With detectors that have a demonstrateddirectionality signature down to recoil energies of ∼
40 keVr[38, 41], DRIFT has set competitive limits on spin-dependentinteractions for dark matter [42].Nevertheless, many challenges remain, not least of which isthe scalability of the current generation of directional experi-ments to reach the sensitivity required to test future claims ofdetection by non-directional experiments. To accomplish this,an emphasis on maximizing sensitivity will need balancing withthe scalability, cost, and robustness of the technology. In thiswork we focus on measuring the intrinsic properties of lowenergy electron and nuclear recoil tracks and how they placefundamental limitations on the sensitivity of directional darkmatter searches. The nature of these tracks determine energythresholds for both discrimination and directionality, with thelatter defined as the energy at which the directional signaturebecomes detectable. As we show in this work, these two en-ergy thresholds are not the same, with the onset of direction-ality having a higher energy threshold than discrimination. Itis important to keep this in mind because, all else being equal,the true determinant of sensitivity for directional dark mattersearches is the directionality threshold. For the gas-based TPCdetectors of interest here, the physical processes that a ff ect theoverall sensitivity - for both discrimination and directionality -involve energy loss, straggling, di ff usion, signal generation (gasgain), and readout resolution. These are briefly discussed here.The direction of a recoiling atom in a gas-based TPC is re-constructed from the ionization track produced along its path.For both electrons and nuclear recoils, this track is neverstraight due to multiple scattering (or straggling) with the con-stituents of the gas. This results in a loss of resolution whichvaries with gas pressure, the energies and the masses of the re-coiling particle and gas atoms. For example, a 40 keVr fluo-rine recoil in 100 Torr CF has an average range of 400 µ mbut su ff ers significant range straggling, defined as the standarddeviation of the particle’s final position relative to its initial di-rection, of about 150 µ m and 110 µ m in the longitudinal and lateral directions, respectively [43].Following the generation of the ionization track, the nega-tive charge drifts in a uniform electric field down to the TPCreadout plane, where it undergoes amplification before beingread out by the data acquisition. During the drift, the ioniza-tion that defines the track undergoes additional loss of resolu-tion due to di ff usion, which depends on the drift distance, thestrength of the drift field, and the nature of the drifting chargecarrier and gas. With electron drift, a select few gases such ascarbon tetrafluoride (CF ) have relatively low di ff usion, which,for a drift distance of 20 cm, an electric field of 400 V / cm, anda pressure of 100 Torr, is approximately σ T = µ m (rec-ommended by Ref. [44] based on a fit to the measurements ofRefs. [45, 46, 47]) and σ L = µ m (values in Ref. [48] de-rived from measurements in Ref. [49]) for the transverse andlongitudinal dimensions, respectively.Even lower di ff usion is possible for negative ion driftinggases such as carbon disulfide (CS ), whose molecules havea high electron a ffi nity. These molecules capture the primaryelectrons to form negative ions that drift with a very low veloc-ity and low di ff usion due to thermal coupling with the bulk gas.The use of an electronegative gas to suppress di ff usion with-out a magnetic field was first proposed by Marto ff et al. [50],and measurements of mobility and di ff usion in CS mixtures[50, 51, 52, 53, 54, 55] indeed indicate thermal behavior fordrift fields up to where measurements exist. Using the di ff u-sion temperatures reported in Ref. [55] for CS , the transverseand longitudinal di ff usion widths are 320 µ m and 330 µ m, re-spectively, for a drift length of 20 cm at a field of 1000 V / cm.Besides this clear advantage from a resolution standpoint, de-tector operation in the presence of high voltages and fields isquite stable in low pressure CS (electrical discharges are sup-pressed). However, one of the downsides of CS is its lack ofspin targets (if a spin dependent search is the goal), which ne-cessitates mixing it with a gas such as CF that contains F, atarget with high nuclear spin content.Finally, after the charge arrives at the readout plane it mustundergo avalanche amplification before being read out. Herethe large gas gain, needed for high signal-to-noise, is not eas-ily achieved due to electrical instability at the low gas pressuresrequired for directionality. For example, in Multi-Wire Propor-tional Chambers (MWPCs) used in the DRIFT detectors, typ-ical gas gains are (cid:46) ∼
10 dark matter events are needed fordiscovery, versus of order ∼ (10 − ) for experiments with no-HT, or axial, tracking [60, 61, 62, 63, 64]. Studies of the ioniza-2 a) CCD detector m m2 m m2 m m1 0 m m I n d u c t i o n g a pT r a n s f e r g a pT r a n s f e r g a p
D r i f t V o l u m eB K 7 w i n d o w W i r e g r i dG E M 3G E M 2G E M 1C a t h o d e M e s h (b) Detection region
Figure 1: (a) A schematic of the CCD detector showing the relative positions of detector components. The optical system, whichconsists of the CCD and lens, sits outside the vacuum vessel and images only the central 2.8 cm × Fe and
Po sources are used for calibrations. (b) A close up view of the detection volumeand GEM stack showing the dimensions of the important regions.3ion energy loss ( dE / dx ) of low energy recoils indeed predict adecrease with decreasing energy along the direction of the trav-eling recoil [59]. Although a HT asymmetry of this nature hasbeen measured down to about 40 keVr [38], it appears to besmall and after straggling and di ff usion its detection is dimin-ished to the point where the statistical advantage over the axialcase is reduced. Future experimental measurements are neededto determine the lowest energy at which this e ff ect exists as wellas its dependence on the recoiling atom and detection medium.Based on the discussion thus far, it is clear that there is alarge parameter space available for optimizing directional ex-periments to have the highest sensitivity, but with economical,robust and scalable designs. As mentioned at the outset, in thiswork we defer the optimization discussion and focus instead onmeasuring the intrinsic properties of low energy electron andnuclear recoils. For this goal, we made measurements with asmall prototype TPC, which has the lowest di ff usion that a real-istic experiment could achieve, and chose a readout technologywith an emphasis on high resolution and high signal-to-noise.The detector, described in the next section, is based on Micro-patterned Gas Detector (MPGD) technology. For a review ofthe development of MPGD technology, see Ref. [65]. Addi-tionally, refer to Ref. [66] for the webpage of the RD51 Collab-oration whose goal is the advancement of MPGD technologies.In this paper we describe the detector performance, and us-ing data from alpha, X-ray, gamma-ray, and neutron sources,we show how the salient features seen in the electron and nu-clear recoil tracks can be used for background discriminationin dark matter searches. In a second paper, we will describethe directional capabilities of the detector, which will be usedto simulate its sensitivity to the directional signature from darkmatter.
3. Detector Setup
The detector consisted of three standard copper GEMs(Gaseous Electron Multipliers [67]; see Ref. [68] for a review)arranged in a cascade with 2 mm separation between them (Fig-ure 1). The GEMs were manufactured at CERN from 7 × sheets of kapton (50 µ m thick) clad on both sides by copper andmounted on G10 frames. The surface of each sheet was chem-ically etched with biconical holes of diameter of 50 / µ m (in-ner / outer) configured in a hexagonal pattern with 140 µ m pitch.A cathode, placed 1 cm below the GEMs, was fabricated froma 7 × copper mesh made from 140 µ m wires with ∼ µ m pitch. A 1 mm pitch anode wire grid plane made from 20 µ m thick gold plated tungsten wires was located 3 mm abovethe top most GEM (GEM 3), forming the induction gap. Thedetector was housed inside an aluminum vacuum vessel andcalibrated using Fe (5.9 keV X-rays) and
Po (5.3 MeV al-phas) sources, both mounted inside the vacuum vessel. A ro-tary feedthrough was used to individually turn both calibrationsources on or o ff , as needed. Before operating the detector thevacuum vessel was pumped down to < gas. A BK-7 glass win-dow was positioned above the readout grid to allow scintillationlight from the final amplification stage (GEM 3) to be viewed by the CCD. The BK-7 glass material was chosen due to its hightransmittance of CF scintillation, whose optical component ispeaked around 620 nm [69, 70], and its lower cost relative toquartz.A back-illuminated Finger Lakes Instrumentation (FLI)charge-coupled device (CCD) camera (MicroLine ML4710-1-MB) with a 1024 × × µ m square pixels occupy the 18.8 mm diagonal sensor,which has a peak quantum e ffi ciency of 96% at 560 nm. Thecamera could be read out at two speeds, 700 kHz and 2 MHz,with 16-bit digitization, and during data taking the sensor wascooled to the lowest stable operating temperature of − ◦ C bya built-in Peltier cooler. At this temperature, the read-out noisewas measured at ∼
10 e − rms and the dark current was < . − / pix / sec. A fast 58 mm f / . × region of the top most GEM surface. The known pitchof the holes on this surface were used to calibrate the length-scale of the field of view.
4. GEM Gain
A 5.9 keV Fe X-ray source was used to measure the e ff ec-tive gas gain, which includes the loss of electrons in the chargeflow from the detection volume to the collection and readoutsurface. With a W-value (the average energy per ionization) of34 . [71], the primary ionization from the 5.9 keVX-ray creates, on average, 172 electron-ion pairs per conversionevent. To measure the gas gain we used the standard procedureof using a pulser and capacitor to determine the preamplifierpulse height to charge calibration. In our case, we used an OR-TEC 448 research pulse generator and an ORTEC 142IH chargesensitive preamplifier, which comes with a built-in 1 pF capac-itor for this purpose. With the calibration results from this pro-cedure the Fe energy spectrum obtained using a multi-channelanalyzer (MCA) was used to determine the gas gain. All gainmeasurements were made by reading out the signals from thelast GEM electrode (GEM3 in Figure 1) with the preamplifier.We also attempted using the anode wire grid to read out thesignal, as often found in the literature, but found that coronalimited the maximum achievable gain. The presence of the an-ode grid was not superfluous, however, as we found that thesparking probability tended to increase without it. Reference[72] has a discussion on the operation of multiple GEMs with-out the use of an anode board or wire plane for readout.To optimize the GEM voltages, the detector was irradiatedwith alphas for about 1 hour ( ∼
500 alpha interactions) at eachsetting to test for stability. If no sparks occurred during thistime, then the detector was deemed to be stable. The voltageswere then changed and the procedure was repeated until thesetting corresponding to the highest stable gain was found. Ata pressure of 100 Torr, with the biases of GEMs 1 and 2 =
290 V and GEM 3 =
460 V, we obtained an e ff ective gain > × . These settings corresponded to a drift field of 400V / cm, transfer field of 1.45 kV / cm, and induction field of 3154 (mm) Y ( mm )
1 4 8 12 16 20 24 281 4 8 1216202428 σ / σ i m Figure 2: Image of alpha track segments in 100 Torr CF withan e ff ective gas gain of 1 × and 6 × σ im , the rms of the image background. The track segments ( ∼ > σ im above the noise.V / cm. The unbalanced powering scheme with di ff erent biaseson the GEMs was used to circumvent the corona problem weexperienced while operating in balanced power mode at lowpressures (75 Torr and 100 Torr). The disadvantage of such apower scheme with a large fraction of the gain coming fromthe last GEM is an increase in the sparking probability. Indeed,sparking was observed with the Po alpha source (5.3 MeV)turned on when operating at a gain of 2 . × at 75 Torr and3 × at 100 Torr. Although the sparks did not damage theGEMs or other detector components during the test runs (1 hourper voltage setting), they did saturate the CCD, producing anartifact known as residual images or ghost images [73].Ghost images are often associated with front-illuminatedCCDs, rather than back-illuminated CCDs, but were neverthe-less observed in the CCD used in this work. They appeared inframes taken after the initial saturation event and did not fadeuntil many hours afterwards; in general, the relaxation time de-pends on the temperature of the CCD. Even though they canbe identified as spatially non-varying objects across successiveframes, or can be dealt with by flooding the sensor with IRlight, a technique known as RBI flushing, their presence is in-dicative of instability with the potential of damaging the detec-tor. Thus, we adjusted the GEM voltages to find the maximumgain attainable without sparking or corona at 100 Torr, to en-sure stable long-term detector operation. Stability was foundwith GEM 1 =
279 V, GEM 2 =
334 V, and GEM 3 = / cm, transfer fields of 1.40 kV / cm and1.67 kV / cm between GEMs 1 and 2 and GEMs 2 and 3, respec-tively, and induction field of 260 V / cm between GEM 3 and thegrid. With these settings an e ff ective gain of ∼ × was achieved. The excellent signal to noise achieved at this gain isillustrated by the alpha tracks in Figure 2, which show a peaksignal of > σ im above the noise level. Besides its use for gainmeasurements, the charge signal from the preamplifier was notused in the subsequent analysis of the data. As the CCD camerawas operated in non-trigger mode, it would be di ffi cult to corre-late the charge signal with a particular event in the CCD image.Additionally, events with energies above three times the Feenergy would saturate the preamplifier. However, using boththe charge and light signals should further aid discrimination.
5. Detector Calibrations
The standard approach to CCD calibration was adoptedwhere each CCD image (or frame) was calibrated using a setof co-averaged flat-field and dark frames. The flat-field frameswere used to correct for vignetting and pixel to pixel variationin sensitivity, and the dark frames corrected for the variable ac-cumulation rate of dark current across pixels. The calibrationwas done by subtracting the co-averaged dark frame from eachimage frame and then dividing the resulting frame by the nor-malized, co-averaged flat-field. Bias frames, which correct forthe electronic bias that is seen as structure in the data frames,were not used since this information is also present in the darkframes. The pedestal due to amplifier bias was removed usingthe overscan region present in each frame.With the CCD being read-noise limited, pixels were binned6 × ∼ × µ m areaof the GEM, which is well-matched to the 140 µ m GEM pitch.The measured read noise was 10 e − rms / pix, and at the − ◦ Coperating temperature of the CCD, the dark current was ∼ e − / pix / sec for 1 × × ∼ − to the total systemnoise.The calibration frames were averaged using an algorithm thatrejects pixels hit by cosmic rays and radioactivity by compar-ing the value of the same pixel across each frame, and rejectingthose above three sigma of the initial average of these pixels.The average value of the pixels was then recalculated exclud-ing the rejected pixels and the same procedure repeated untilconvergence in the average was reached. This procedure wasapplied to all pixels to create master flat and dark calibrationframes. ff usion The detector track reconstruction and angular resolution isintrinsically tied to the di ff usive properties of electrons in thetarget gas. With our CCD readout, only the transverse, or lat-eral, di ff usion could be measured as the longitudinal componentrequired timing resolution from the GEM charge readout thatis well beyond the capability of the ORTEC preamplifier used5 + HV Pulser
Figure 3: A schematic of the pinhole cathode and high voltagepulser setup used to measure transverse di ff usion. Both sur-faces of the cathode are held at a fixed voltage with a high volt-age power supply. The capacitor isolates the pulser from thehigh voltage but allows an impulse to be transferred through.The resistor (1 G Ω ) and high voltage impulse (12 kV) causes ashort duration voltage change only on the bottom surface of thedevice (the top surface is held a fixed voltage by the power sup-ply), initiating a spark in the pinhole and generating ionization.The 10 µ m collimator reduces the transverse extent (perpendic-ular to electric field) of the ionization entering the drift volume(region on top of the collimator) to point-like dimensions.here. To measure the lateral di ff usion, a point source was gener-ated at a specially constructed cathode made from an insulatingsheet sandwiched between two strips of copper tape. The cath-ode had a small hole punctured at its center with a 10 µ m colli-mator placed over the hole on the side facing the drift volume.The electrodes of the cathode were connected to a power sup-ply and a 12 kV high voltage pulse generator; Figure 3 showsa schematic of the cathode and electrical connections. The HVpulser was used to generate a spark inside the hole, which pro-duced ionization that appeared as a point source ‘track’ in thedrift volume after passing through the collimator. We imaged a collection of tracks and measured the spread intheir light profiles. Each of the light profiles was fitted using aGaussian curve, and an average σ tot = ± µ m (stat) wasobtained for the sample of tracks. The main contribution tothe spread is expected from di ff usion of the electron cloud as ittravels the full distance from the cathode to the final GEM stage.This was confirmed using data from Refs. [44, 48] and resultsfrom the MAGBOLTZ program [74], which predict σ di ff = µ m for the di ff usion in our detector. The MAGBOLTZ resultindicates that contributions to σ di ff are divided about equally,when added in quadrature, between the drift gap and the twotransfer gaps between the GEM stages. A secondary source to σ tot is expected from the GEM pitch and the 3-GEM cascade.Attributing this wholly to the di ff erence between σ tot and σ di ff yields 65 ± µ m per GEM when added in quadrature. Thisvalue is consistent with our expectation that the track spreaddue to the GEM pitch is roughly half of its 140 µ m pitch. Weexpect other contributions, such as smearing due to imperfectoptics, to be minor compared to these. Although quite low inour prototype detector, di ff usion could be further reduced if the As a check on this technique for generating point tracks, we also used alphatrack segments (e.g., Figure 2) and found that measurements of their widthsgave results in good agreement.
Energy (ADU)1000 1500 2000 2500 3000 C oun t s Figure 4: An Fe energy spectrum obtained with the CCDcamera and used to calibrate the energy of recoils for one ofthe sub-runs. The energy is shown in ADUs (analog-to-digital-units). The conversion factor is found by taking the ratio ofthe fitted peak of the spectrum to the average energy of an Ferecoil (5.9 keV).transfer gap contribution could be eliminated, or if a negativeion drift gas such as the CS could be employed. With di ff usionscaling as √ L with drift distance, these considerations becomecritical for scale up to large detectors; this is discussed furtherin Section 8.2. Energy calibrations were done using Fe X-ray and
Poalpha sources. The alpha track calibration was made by firstusing SRIM [43] to calculate the Bragg curve of a 5.3 MeV al-pha in 100 Torr CF . We measured the location of the alphasource relative to the drift volume and determined the part ofthe track that would be imaged by the CCD camera. Figure 2shows segments of alpha tracks imaged by the CCD camera atthe maximum stable gain. By comparing the total integratedlight output in the image of the alpha track with the energy cal-culated from the SRIM generated Bragg curve, we obtained thelight to energy conversion factor, ADU / keV α . Since >
99% ofthe energy lost by an alpha particle before its Bragg peak isthrough ionization, we can treat this as keV electron-equivalentenergy (keVee).For an independent calibration method we imaged the elec-tronic recoils from Fe X-ray interactions and obtained an en-ergy spectrum of the scintillation signal; see Figure 4. At ourmaximum stable e ff ective gain, Fe tracks were visible at 6 × Fe in a TPC detector (details provided elsewhere). The twocalibration methods give results that are within 20% of eachother. The small di ff erence could be due to a systematic in de-termining the alpha segment imaged by the CCD camera. In6able 1: Detector Parameters Detector Parameters
CCD and Imaging ParametersPeak QE 96% (560 nm)Pixel Size (1 × × µ m Pixel Binning 6 × µ m / pixImaging Area 2 . × . Read Noise @ 700 kHz 10 e − rmsOperating Temperature − ◦ CDark Current 0.03 e − / s / pixExposure Time 5 secVessel ParametersDetection Volume 2 . × . × . CF Pressure 100 TorrE ff ective Gas Gain 10 E ff ective Transverse Di ff usion 345 ± µ mthe analysis of the data to follow, we will use the energy con-version factor derived from the Fe energy calibration. Lastly,a summary of important detector parameters discussed thus faris included in Table 1. Co and
Cf Data Runs
To study our detector’s response to nuclear recoils, and itsability to distinguish these from gamma backgrounds, we useda
Cf neutron source and a Co gamma source. For the Corun, the source was placed outside the vacuum vessel but insidea lead housing to protect the CCD sensor from direct gamma-ray interactions; there was no lead between the source and theouter vessel wall. The neutron run was conducted in a sim-ilar manner but with the addition of lead bricks between thesource and detector to attenuate the large number of gammasfrom
Cf. In total, about 96 and 36 hours of neutron dataand gamma data were collected, respectively. To evaluate thedetector’s directional sensitivity half of the neutron data wascollected with the neutrons directed in the − x direction, an axislying in the imaging plane, and the other half with the neutronsdirected in the + x direction.For each data taking sequence, or sub-run, the vessel waspumped out and back-filled with fresh CF gas to a pressure of100 ± × − ◦ C, which was monitored by an internalsensor in the camera. An energy calibration was done with an Fe source at the start and end of each sub-run sequence. Thehigh drift speed of electrons in CF made it impossible to trig-ger the CCD (open and shut the shutter) using the charge signalfrom the first GEM stage. Therefore, we operated in non-triggermode with the CCD camera successively taking 5 second expo-sures over a duration of about 12 hours for each sub-run. This corresponded to approximately 9 hours of live time after ac-counting for the CCD readout time. The detector was refilledafter each sub-run to avoid substantial gain degradation due tochanges in gas purity from out-gassing during the data takingsequence. As this and other e ff ects including temperature andcomposition changes from charge avalanching caused the gainto drift over time, we used the average of the values measuredat the end and start of each data sequence for the energy cali-bration constant. In total, eight sub-runs were conducted for theneutron data and three for the gamma data.We analyzed the data using an image analysis algorithm de-veloped with MATLAB and its image processing toolbox. First,images were calibrated and binned 4 × σ im were identified as objects. Allobjects found crossing the image boundaries were rejected. Thebinned image was then up-sampled back to its original size, re-sulting in an index image for the pixel locations of all identifiedobjects. In the remainder of the analysis, all object propertieswere determined from the original, non-software binned image.To exclude hot pixels and CCD events (objects resulting fromdirect interactions of cosmic rays, radioactivity, neutrons, orgamma rays with the CCD sensor), we required objects to con-tain at least four contiguous pixels. Separated pixels belongingto a local grouping of pixels above threshold were connectedback to the primary grouping by morphologically closing theobject using a disk-shaped structuring element with a radius oftwo pixels. In essence, the closing operation, which is a dila-tion follow by an erosion, connected all pixels above thresholdthat lay within the radius of the structure element. Each identi-fied object was fitted with a position and an intensity weightedellipse, which, along with the pixel grouping in the unfitted ob-ject, were used to determine some of its important propertiessuch energy, track length, width, skewness, and energy loss pro-file.
7. Results
The background rate in a detector can vary widely dependingon its size and the materials used in its construction, with eventhe most stringent requirements on radio-purity not eliminat-ing all sources of backgrounds. Consequently, fiducializationand discrimination are of critical importance to dark matter andother rare event searches that require large detection volumes.One important source of backgrounds are gamma-rays and X-rays that can interact inside the detector to produce electronicrecoils. For tracking detectors, the stopping power, dE / dx , pro-vides a powerful tool for discriminating between electronic andnuclear recoils. Electronic recoils have a much lower average dE / dx and, hence, much longer ranges as compared to nuclearrecoils of the same energy, a fact that is evident in the rangeversus energy plots shown in Figures 5 and 6. Given the inabil-ity of our detector to measure the Z-component of the track, therange in these figures is 2D. Of course, to maximize separationbetween the nuclear and electronic recoil bands, a detector withfull 3D tracking capability is desirable (see Section 8.3).7 a) Co data pre-cuts (b) Co data post CCD cuts
Figure 5: (a) The projected range ( R ) vs. energy plot of data from the Co gamma run. The events with short range and lowenergy in the lower left corner of the plot are called CCD events, which result from the direct interaction with the imaging sensor.(b) The same data after analysis cuts are made to remove the CCD events. The events in the short horizontal band extending to 40keVee lie in the nuclear recoil band (see Figure 6 and text). These events are likely due to radon progeny recoils occurring at thecathode or GEM surfaces. (a)
Cf data post CCD cuts (b)
Cf data post selection cuts
Figure 6: (a) The projected range ( R ) vs. energy plot of data from the Cf run after applying the CCD event cuts. The events notpart of the two bands are most likely segments of proton recoils created by neutron interactions with hydrogen-rich materials in thedetector. (b) The same data with nuclear recoil selections cut applied. The lowest energy recoils post-cuts extend to ∼
10 keVee (23keVr). 8 .2. Gamma and Neutron Data
Using the reconstructed tracks passing the track identifica-tion algorithm from the Co gamma run, the 2D range as afunction of energy is shown in Figures 5a and 5b. The hardvertical edge at 2 keVee is the result of a software thresholdset on the energy of detected objects to reduce the number offalse event detections during the initial track finding stage ofthe analysis. The sub-mm events in the lower left region of Fig-ure 5a are the CCD events described in Section 6, which are dueto direct interactions of ionizing radiation with the CCD sensor.As these CCD events su ff er no di ff usion, they tend to have ex-tremely high standard deviations of their pixel values as well asvery high average intensities (total intensity / number of pixels).Therefore, cuts made on these two parameters, in addition totrack size, were used to e ffi ciently remove this class of events.The events in the small branch protruding from the primary ver-tical band at around 1.5 mm are mostly due to detector intrinsicbackgrounds from decays of radon daughters. These are some-times referred to as radon progeny recoils, or RPRs, and occurat the detector surfaces [75, 76]. Events in the primary verti-cal band have low average dE / dx and correspond to electronrecoil events. These are primarily due to Compton scatteringsof the 1.17 and 1.33 MeV gamma rays emitted in the beta de-cay of Co, with a small fraction from ambient and intrinsicelectromagnetic backgrounds in the detector. Altogether, therewere 27 , , Cf neu-tron run are shown in Figures 6a and 6b. Two distinct bands arepresent. The vertical band is the same electronic recoil bandobserved in the gamma run, while the second, near-horizontalband contains the high dE / dx nuclear recoils (both carbonand fluorine with a ratio implied by a GEANT simulation of1:6). The events forming the “haze” between these two pri-mary bands have dE / dx values inconsistent with being due toCompton scatters, and they were absent in the Co data. Their dE / dx is also inconsistent with those of carbon or fluorine re-coils as their lengths far exceed the maximum of these recoilingions at a given energy. Since these events were only seen in theneutron runs, we believe that they are segments of proton re-coils from neutron interactions with hydrogen rich material inthe detector such as the GEM kapton substrate.From Figure 6a, it is evident that even before any selectioncriteria are applied, there is good separation of nuclear recoilsfrom electronic recoils down to low energies. Nevertheless, weused the Co data to develop an algorithm that maximizes therejection of electronic recoils while retaining a high detectione ffi ciency for nuclear recoils. One parameter that gave goodseparation between the two recoil classes is the ratio of the pro-jected Bragg curve peak (peak of the light distribution in keVee)to the track length (major axis of the fitted ellipse in mm). Ahistogram of the natural logarithm of this parameter, defined as η , is shown in Figure 7 for both the Co and
Cf data. Thedistribution of the former has mainly one population whereasthe latter, which contains both electronic and nuclear recoils,has two. The one with the larger log η corresponds to nuclearrecoils, as also confirmed by noting that they lie in the nuclear recoil band in the R versus energy plots (Figures 6a). Based onthese distributions, we defined the gamma cut of log η < . η histogram of the Co data, shown in Figure 7b, clearly shows the 65 events inquestion forming a distinct population, which overlaps that ofthe nuclear recoil events from the
Cf data in Figure 7a. Thatthese events are nuclear recoils is further corroborated by thefact that they lie in the nuclear recoil band in the R versusenergy plane, as shown in Figure 8a. Finally, the evidence thatthese events are mostly, if not all, due to RPRs comes from theirenergy distribution. For the 56 lower energy events this distri-bution has a mean of 26 keVee with a cut-o ff at 38 keVee, con-sistent with the ionization energy distribution from the short-lived lead isotopes in the radon chain [77].After excluding the 65 RPR and associated alpha events, the Co dataset contains 25,696 events, which we identify as elec-tronic recoils. Applying the gamma cut described above re-moves all of these, resulting in the detector’s gamma rejection at ≤ . × − . This rejection level is achieved at a pressure of 100Torr with two dimensional (2D) track reconstruction, and maybe improved further with full 3D reconstruction (Section 8.3)and / or by operating at lower pressure, where tracks are longerand better resolved. Additionally, more sophisticated analysisalgorithms should give better results.The resulting energy spectrum from the Cf run after apply-ing the analysis cuts to remove CCD events and electron recoilsis shown in Figure 8. Conversion from the measured energyin keVee to nuclear recoil energy in keVr is based on the fluo-rine quenching factors from Ref. [78]. The spectrum rises from20 keVr to 35 keVr, where it peaks, indicating that maximumnuclear recoil e ffi ciency has been reached. Thus, the e ff ectivediscrimination threshold of this detector is approximately at 10keVee ( ∼
23 keVr), see Figure 8a and inset in Figure 8b. At 100Torr, this is the lowest discrimination threshold of any direc-tional detector to date. Nevertheless, our ∼
10 keVee discrimi-nation threshold is significantly above the detection threshold,which we estimate to be 2 keVee for a di ff used, point-like eventbased on our Fe calibration data (see Figure 4). Our direc-tional threshold is higher yet, ∼ × the discrimination thresh-old, with the reason being that nuclear recoil tracks are shorterdue to their higher dE / dx , and therefore become unresolved athigher energies; further details are discussed in a separate pa-per on directionality. The importance of a low energy threshold,for both discrimination and directionality, is that it provides one9 og η C oun t s −3 −2 −1 0 1 2 305001000150020002500300035004000 Cf Co (a) Discrimination parameter histogram log η C oun t s −1 −0.5 0 0.5 1 1.5 205101520253035404550 Co (b) Co data RPR separation
Figure 7: (a) A histogram of the discrimination parameter, log η , defined as the ratio of the Bragg curve peak to track length, for the Co (red) and
Cf (blue) runs. There are two distributions in the
Cf data representing the nuclear and electronic recoils whileonly one prominent one in the Co data which contains the electronic recoils. (b) A histogram of the log η parameter for the Codata zoomed in to see the small distribution of RPR events which overlaps the nuclear recoil peak in the
Cf data. The verticaldotted line at 0.50 is the value of the cut set on this parameter used for discrimination. (a) Co and
Cf data post selection cuts
Energy (keVr) C oun t s C oun t s
15 25 35 450200400600 (b) Recoil Energy Spectrum
Figure 8: (a) The R vs energy plot for events passing the analysis cuts for nuclear recoils from both the Co and
Cf datasets.The events remaining from the Co data lie in the nuclear recoil band and have an energy distribution consistent with RPRs, asdiscussed in the text. (b) The measured energy spectrum of carbon and fluorine recoils from
Cf neutrons after analysis cuts wereapplied to remove the electronic recoils. The spectrum, which is peaked around 35 keVr, was derived assuming the quenchingfactors from [78] and that all recoils were fluorine. 10ath towards increasing the sensitivity of directional dark mat-ter detectors. In fact, it is critical for a low mass WIMP searchas the recoil energy spectrum is shifted towards lower energies.Finally, sample recoil images from the Co and
Cf runs areshown in Figures 9, 10, and 11.
8. Discussion
The break down of discrimination in our detector below ∼ ∼
23 keVr) is due to a number of e ff ects that lead to theconvergence of electron and nuclear recoil tracks in the rangeversus energy parameter space. These are due to physical ef-fects, such as di ff usion and energy-loss processes, as well asdetector limitations. We discuss these below in Sections 8.1and 8.2, and describe possible ways to circumvent them to im-prove discrimination. The discrimination at low energies is a ff ected first and fore-most by di ff usion; as tracks fall below the resolution limit,range versus energy no longer works as a discriminant. Evenif di ff usion were suppressed, however, energy-loss processesa ff ecting both electrons and nuclear recoils could pose funda-mental limits to discrimination.For electrons the dominant e ff ects are the well known energy-loss fluctuations and straggling (e.g., Figures 9b, 9c, and 10),which give rise to a large spread in their range. These e ff ectsbecoming stronger at lower energies, pushing the short-tracktail of the electron distribution below the di ff usion limit. There,these electron recoils merge with the nuclear recoil population(Figures 5 and 6) which, with their much larger dE / dx , are al-ready unresolved at ∼
20 keVee ( ∼
40 keVr). In addition, theprobability for large angle scattering at low energies increasesfor electrons, producing a trajectory that is almost di ff usive innature. As a result the energy is deposited into smaller unre-solved regions of space, which, together with projection of thetrack to 2D, systematically biases the dE / dx upward towardsthat of nuclear recoils. In Figure 6 events of a given energya ff ected in this manner have their R underestimated and dropdown into the nuclear recoil band.For nuclear recoils the opposite trend occurs, wherebyenergy-loss in the ionization channel (the detectable dE / dx ) de-creases as the ions slow down, with other energy-loss channelsmaking up the di ff erence. Both theoretical (Ref. [78]) and ex-perimental (Ref. [79]) studies of the ratio of ionization to to-tal energy-loss (the quenching factor) indicate values less than0.25 for E <
10 keVr in a variety of gases and their mixtures.This e ff ect underestimates the energy of events of a given tracklength in Figure 6, pushing them leftward into the electronicrecoil band.Thus, the detected dE / dx for both classes of recoils con-verges at low energies, potentially posing a fundamental limiton discrimination using the range versus energy technique. Theenergy where this occurs cannot be determined from our datawhere, as mentioned above, the limitations on discrimination are due to di ff usion. Progress toward this goal will require re-solving tracks below our E <
10 keVee threshold, for exam-ple by lowering the gas pressure to lengthen tracks. From apractical perspective, directional gas TPCs have been shownto operate down to 20 Torr ([80, 81]), and with Thick GEMs(THGEMs) good gas gain has been demonstrated down to 0.5Torr in certain gases [57, 58]. So, measurements in the 10- 40 Torr pressure range could feasibly map out the possibleparameter space for discrimination below 10 keVee, and willlikely lower the threshold as well. Exploring gas mixtures withlower straggling and energy loss fluctuations should also be at-tempted. All such e ff orts will be most critical for low massWIMP searches where energy thresholds <
10 keVr (note thisis recoil energy, not ionization energy) are desired. Dependingon the degree of quenching, the detected energy in this regimecould be as low as a few keVee, where achieving both discrim-ination and directionality could be extremely challenging.
Besides lowering the pressure and optimizing gas mixtures,improvements in the detector itself could also lead to better dis-crimination and directionality. The three detector parametersthat we believe play a critical role for this are signal-to-noise,resolution, and tracking dimensionality (discussed in Section8.3). We restrict our discussion to an optical detector of thetype used here, but many of the ideas apply to charge readoutdetectors as well.A benefit of signal-to-noise, especially where discrimina-tion is concerned, is that it enables full mapping of electrontracks that have large energy-loss fluctuations of the type seenin Figures 9b, 9c, and 10. These tracks show regions with highenergy-loss interspersed with barely discernible regions of lowenergy-loss. A detector with low signal-to-noise would detectonly the peak regions, which, due to their systematically higher dE / dx , would look like nuclear recoil tracks. This is illustratedin Figure 12, where we show three images of an event from the Cf run, each with a di ff erent signal-to-noise level. Figure 12bis the original calibrated CCD image obtained by our detector,showing a track with a high density segment and a suggestionof a very faint tail. In Figure 12a, we artificially added about50% more noise to the image in software. In this image thelong faint tail of the recoil is lost in the noise, leaving just thebright high density region that could easily be misidentified asa low energy nuclear recoil track. Finally, in Figure 12c theoriginal image has been processed using a Gaussian noise re-duction filter. The long faint tail is now clearly visible as is itsconnection to the brighter segment, leaving little doubt that thisis an electron track.The two obvious paths to achieving high signal-to-noise areto lower the noise and / or increase the signal. The former ap-proach would require a reduction in the CCD camera systemnoise, which is usually dominated by the read noise for shortexposures, but the tradeo ff is slower readout speed or morecostly multi-node readouts. In the direction of increasing sig-nal, there are many approaches that could be taken. The firstis boosting the CCD sensor quantum e ffi ciency, which, for theback-illuminated CCD used here, is already highly optimized.11 (mm) Y ( mm ) (a) 9 keVee electron recoil X (mm) Y ( mm )
15 16 17 18 19 20 21 223456789 (b) 13 keVee electron recoil
X (mm) Y ( mm )
14 15 16 17 18 19 20 211819202122232425 (c) 21 keVee electron recoil
X (mm) Y ( mm ) (d) >
30 keVee electron recoils
Figure 9: (a-d) Electronic recoils of di ff erent energies from the Cf and Co runs. The images have been contrast adjustedto enhance visualization. The magenta contours trace out the track boundaries and are included as a visualization aid and tohelp illustrate the straggling of low energy recoils and the clumpy ionization deposition. (d) Two high energy electronic recoilscontaining smaller delta ray tracks emerging perpendicular to the primary electronic recoil track. These image boundary crossingtracks were rejected from the analysis and were found only by visually scanning events by eye.12 (mm) Y ( mm ) I n t en s i t y ( k e V ee ) (a) Electron Recoil X (mm) Y ( mm ) I n t en s i t y ( k e V ee ) (b) Projected Bragg Curve Figure 10: (a) An image of a 24 keVee electronic recoil in 100 Torr CF . The magenta curve traces out the perimeter of the trackand helps in visualizing the straggling of the recoil. (b) The Bragg curve of the recoil, obtained from the projection of the trackalong the major axis of the fitted ellipse, shows the large energy fluctuations that are also clearly apparent in the CCD image. Adetector with a lower signal to noise ratio would only see the brightest region(s) of the track and possibly misidentifying them as anuclear recoil(s).Secondly, one can increase signal through better light collectionwith a more e ffi cient optical system (faster lens), and / or a setupthat allows for more light collection by decreasing the distancebetween the GEM and lens. However, the latter approach re-quires a sacrifice of imaging area whereas the former requiresa potentially uneconomical and sophisticated custom lens de-sign. Although both are potential drawbacks for scale-up tolarge detector volumes, these approaches should be consideredif cheaper CCDs or other technologies become available in thefuture.Another approach is to increase the light output by select-ing gas mixtures with higher avalanche photon yield, definedas the number of photons per secondary electron released dur-ing amplification, or by increasing the absolute gas gain. Al-though we have achieved very high gas gain in our detectorwith the triple-GEM stack, the light yield of CF , albeit one ofthe better scintillating gases, could be improved further. Forexample, the addition of Ar at high concentrations has beenshown to increase the photon yield of pure CF [70, 82] from ∼ ∼ ff ects have been noted at high gains [84], where both the gainand photon yield are charge density dependent. This could havea deleterious e ff ect on both energy and directional sense deter-mination.Detector resolution, also critical for both directionality and discrimination, is governed by various design and operationchoices such as the readout pitch, gas mixture, pressure and dif-fusion. How pressure can be used to vary track lengths and howthe choice of gas can e ff ect fluctuations and straggling werebriefly discussed above, so we focus on the other two factorshere. Di ff usion can be reduced by limiting the maximum driftdistance and by making a judicious choice of gas mixture. Al-though CF exhibits relatively good di ff usive characteristics foran electron drift gas, negative ion gases such as CS , whichdrift in the thermal regime, provide the lowest di ff usion possi-ble without employing magnetic fields (see Section 2). The lowdi ff usion in our small detector, σ ∼ σ ∼ over a 50cm drift in the DRIFT detector. Thus, the di ff usion achievedin our detector is a reasonable goal for a large scale directionalexperiment, and any meaningful reduction would likely requireother techniques.In principle, the readout pitch of the detector should be fineenough to extract the maximum information possible with thegiven di ff usion. The e ff ective pitch in our detector, which isdue to a combination of the GEM pitch, 140 µ m, and the CCDbinned pixels, 165 µ m, was a little less than half the sigma dueto di ff usion, σ ∼ .
35 mm. This allowed tracks to be mea-sured with a su ffi cient number of independent samples to ex-tract features, such as energy-loss fluctuations and asymmetryin ionization, important for discrimination and directionality.A good example demonstrating this for discrimination isfound by comparing the images of the electronic recoil and nu-clear recoil shown in Figures 9a and 11a, respectively. Both13 (mm) Y ( mm )
18 19 20 2118192021 (a) 28 keVr ( ∼
13 keVee) nuclear recoil
X (mm) Y ( mm ) (b) 53 keVr ( ∼
28 keVee) nuclear recoil
X (mm) Y ( mm ) (c) 104 keVr ( ∼
66 keVee) nuclear recoil
X (mm) Y ( mm )
12 13 14 15 16 17 18 191819202122232425 (d) 214 keVr ( ∼
160 keVee) nuclear recoil
Figure 11: (a-d) Nuclear recoils from the
Cf runs at various kinetic energies in 100 Torr CF using the Hitachi quenching factorsand assuming all the nuclear recoils are fluorine atoms. The images have been contrast adjusted to enhance visualization and havedi ff erent X / Y scales. The directionality and asymmetry in the energy deposition often referred to as the head-tail signature becomeapparent for the highest three energy recoils. In all images, the average neutron direction is from left to right. Also note how thetrack in (a) deposits roughly the same amount of energy as the electronic recoils in Figures 9a and 9b but appear in a much tighterarea with a higher intensity peak pixel value. A lower signal-to-noise and lower resolution detector could fail to di ff erentiate thetrack in (a) from that in Figure 9a. 14 mm (a) Noise Added (b) Original (c) Filtered Original Figure 12: An ∼
16 keVee event from the
Cf run shown with di ff erent signal-to-noise levels. The middle panel (b) shows theoriginal calibrated CCD image, and the left panel (a) shows the same but with 50% higher noise added in software. The right panel(c) is the original image with a Gaussian noise reduction filter applied. See text for details.have similar detected energy but the electronic recoil looksmore disperse with larger fluctuations, and the nuclear recoilmore concentrated and smoothly distributed. These di ff erencesare consistent with the energy-loss processes discussed above,and could be used in more sophisticated algorithms to improvediscrimination and directionality. The finer pitch also opens thedoor to deconvolution techniques, such as those used in astron-omy (for example, see [85]), which could be applied to achievebetter resolution. The detector parameter that is arguably most critical for gooddiscrimination and directionality is the number of indepen-dently measured track components. Although full 3D track re-construction is preferred, any benefit it brings to discriminationor the directional sensitivity must be justified relative to the costincrease or added design and operational complexity. Here westudy the improvement in discrimination from 1D to 2D to 3Dand, except for brief remarks below, postpone the discussion ondirectionality for a separate paper.We begin by studying the di ff erence in discrimination powerbetween 1D and 2D. For this we took our 2D data from the Co and
Cf runs and reduced them to 1D. We have definedthe X(Y) component of the track length as R cos θ ( R sin θ ),where θ is the reconstructed angle of the track in the X-Y plane.Of course, this artificially extends the 1D track length downto zero, whereas the di ff usion in a real 1D detector would im-pose a minimum. Nevertheless, this e ff ect, which is apparent inFigures 13a and 13b, does not change the relative comparisonwe wish to make here. In addition, to better gauge the sepa-ration between the electronic and nuclear recoil bands, the 65events associated with RPRs in the Co were removed fromthis dataset.In Figures 13a and 13b the X and Y, 1D components of thetracks are plotted as a function of energy, respectively. As onewould expect, the electronic recoil bands in these two figures are very similar because their recoil directions are distributedmore or less isotropically in the imaging plane. The nuclearrecoil band, although smeared out greatly in both figures, ismarginally denser in Figure 13a because the neutrons were di-rected along the X-axis. For comparison, an overlay of the 2D, R versus E data from both runs is also shown in Figure 13c.The results show significantly better discrimination with 2Dtracks versus 1D. In 2D, discrimination is achieved down to ∼ Corun have strayed into the nuclear recoil band out to energies of ∼ ∼
65 keVr). This e ff ectively puts the discrim-ination energy threshold of the 2D data at a factor ∼ ∼ · c − scattering o ff fluorine through spin-independent inter-action. Perhaps cuts made on other track parameters could beused to reduce this gap, but it is unlikely that 1D discriminationwould improve to extend the threshold much below 30 keVee(55 keVr).Next, with the aid of simulations we explored the potentialdi ff erence in discrimination capability between a 2D vs. 3Ddetector. The simulation program SRIM(CASINO) [43]([86])was used to simulate nuclear(electronic) recoil tracks with anisotropic distribution in 3D and with the same energy distribu-tion as our calibration data. Each simulated track was thenprojected onto the image plane with the pixellization, noise,and signal adjusted to match those of our CCD detector. Thedi ff used and signal-to-noise adjusted projected track from eachimage plane (XY, XZ, YZ) was analyzed using the same imageanalysis algorithm as the one used for our neutron and gammadata (Section 6).The results of these simulations are shown in Figures 14a and14b. As expected, the nuclear recoil band for the case of 3D We note that the nuclear recoil simulations do not take into account sec-ondary recoils. a) X Component (b) Y Component (c) R Length
Figure 13: (a) and (b) Plots of the X and Y, 1D components of the track length vs. energy for the Co run overlaid on top of the
Cf run. These show the approximate level of discrimination that a 1D detector would achieve. (c) The 2D projected track lengthvs. energy data from the same two data runs reproduced for comparison (from Figures 5 and 6). For all 3 panels the RPRs fromthe Co have been removed to show the true separation between the electronic recoil and the nuclear recoil bands. In the 2D datathere is separation of the bands down to about 10 keVee (23 keVr), but in the 1D data (a-b), events from the electronic recoil bandsare leaking into the nuclear recoil region up to energies >
35 keVee (62 keVr). (a) 2D Reconstruction (b) 3D Reconstruction
Figure 14: Simulation of range vs. energy for fluorine and electron recoils in 100 Torr CF for 2D (a) and 3D (b) track reconstruc-tions. In the 2D reconstruction (a), events from the electron band leak into the nuclear band up to energies of ∼ ∼ ff ects of straggling and energy loss fluctua-tions (discussed in Section 8.1), resulting in large scatter in bothrange and energy. However, in the region where the two bandsintersect, the 3D electron events are more tightly distributedthan in 2D, yielding better separation from the nuclear recoils.This results in about a ∼
35% lower discrimination threshold,which, not surprisingly, is not as large as the di ff erence seen be-tween the 1D and 2D data. Nevertheless, when combined withlower, ∼ ff erence between 2D and 3D whenperfect HT sense recognition is assumed [60, 61, 63]. If othervariables are included then even 1D appears competitive [64].This would seem to suggest that multi-dimensional tracking issomething desired but not absolutely necessary. There are twocaveats to this, however. The first is that the assumption ofperfect HT sense recognition is unrealistic, and we argue thathigher dimensionality is needed even to approach this goal. Thesecond is that, in the case of low pressure TPCs, discriminationpower and directional sensitivity are coupled to the tracking di-mensionality of the detector. As the primary discriminant isthe stopping power, dE / dx , robust discrimination requires highquality measurements of both energy and track range. The lat-ter, as we have shown here, is best accomplished with a 3D de-tector. A more extensive discussion on the relationship betweentracking dimensionality and directional sensitivity is reservedfor a separate paper.
9. Conclusion and Prospects
In this work we have described a small high resolution, highsignal-to-noise GEM-based TPC with a 2D CCD readout. Thedetector was designed to make detailed studies of low energyelectron and nuclear recoil tracks for the purpose of directionaldark matter searches. Detector performance was characterizedusing alpha particles from
Po, X-rays from Fe, gamma-rays from Co, and ∼ MeV neutrons from
Cf. Stable gasgains upward of 10 were achieved in 100 Torr of pure CF with a triple-GEM cascade, resulting in a very high signal-to-noise. This, together with an e ff ective 165 µ m track samplingand low di ff usion, σ ∼ Co and
Cf data we also studied discrimina-tion between electronic and nuclear recoils. Using the standardrange versus energy technique, relatively simple selection crite-ria were used to demonstrate excellent discrimination down to ∼
10 keVee, or ∼
23 keVr recoil energy. This result, the best todate at 100 Torr, was especially aided by the high spatial reso-lution and signal-to-noise of the detector. Without the latter, the large energy-loss fluctuations su ff ered by low energy electronswould cause only the peak intensity regions of the tracks to bedetected. Such tracks would be reconstructed with their dE / dx and track lengths systematically too high and too low, respec-tively, resulting in these events being misidentified as nuclearrecoils. That both high spatial resolution and high signal-to-noise are necessary for good discrimination is an important re-sult of this work.Pushing the discrimination threshold to even lower energiesis an important future goal, especially critical for directionallow mass WIMP searches. Our ∼
10 keVee threshold is due to acombination of di ff usion and electron straggling, which resultsin the merger of the electron and nuclear recoil populations inthe R versus E parameter space. Two paths around this are tolower the gas pressure to lengthen tracks, and full 3D track re-construction. The former would allow mapping the parameterspace for discrimination below our 10 keVee threshold and findany fundamental limit if it exists. The influence of track dimen-sionality on discrimination was also investigated and, using ourdata, we found that the 1D threshold is a factor ∼ ∼
35% lower discrimination threshold in 3D. Inaddition to these two strategies, better analysis techniques thattake full advantage of the di ff erence seen between electron andnuclear recoil tracks (e.g., compare Figures 9a and 11a) shouldbe investigated in the future.Finally, the data obtained in this work can also be used tocharacterize the directionality of the nuclear recoil tracks. Withtheir higher dE / dx , we find that these tracks become unresolvedat energies around ∼
20 keVee ( ∼
40 keVr), resulting in a direc-tionality threshold that is a factor ∼ Acknowledgements
This material is based upon work supported by the NSF underGrant Nos. 0548208, 1103420, and 1407773.
References [1] K. Griest and M. Kamionkowski, Phys. Rep. 333-334 (2000) 167-182.[2] G. Jungman, M. Kamionkowski, and K. Griest, Phys. Rep. 267 (1996)195-373.[3] M. W. Goodman and E. Witten, Phys. Rev. D 31, 3059 (1985).[4] R. J. Gaitskell, Annu. Rev. Nucl. Part. Sci. 2004. 54:31559.[5] H. Baer, C. Balazs, A. Belyaev, and J. OFarrill, JCAP 0309:007 (2003).[6] G. Angloher et al., Eur. Phys. J. C 74, 3184 (2014).[7] M. Kuzniak, M. Boulay and T. Pollmann, Astropart. Phys. 36, 77 (2012).[8] H. Davis, C. McCabe and C. Boehm, JCAP 1408, 014 (2014).[9] J. H. Davis, Int. J. Mod. Phys. A, 30, 1530038 (2015).[10] A. K. Drukier, K. Freese, and D. N. Spergel, Phys. Rev. D, Vol. 33, No.12, (1986).[11] D. N. Spergel, Phys. Rev. D 37, 1353 (1988).
12] K. Blum, arXiv:1110.0857.[13] J. P. Ralston, arXiv:1006.5255.[14] D. DAngelo (Borexino Collaboration), arXiv:1109.3901.[15] E. Fernandez-Martinez and R. Mahbubani, J. Cosmol. Astropart. Phys. 07(2012) 029.[16] J. H. Davis, Phys. Rev. Lett. 113, 081302 (2014).[17] R. Bernabei et al., Eur. Phys. J. C 56, 333 (2008).[18] R. Bernabei, P. Belli, F. Cappella, V. Caracciolo, S. Castellano et al., Eur.Phys. J. C 73 (2013) 2648.[19] G. Angloher et al., Eur. Phys. J. C 72, 1971 (2012).[20] C. E. Aalseth et al., Phys. Rev. D 88, 012002 (2013).[21] R. Agnese et al., Phys. Rev. Lett. 111, 251301 (2013).[22] E. Aprile et al., Phys. Rev. Lett. 109, 181301 (2012).[23] R. Agnese et al., Phys. Rev. Lett. 112, 241302 (2014).[24] D.S. Akerib et al., Phys. Rev. Lett. 116, 161301 (2016).[25] D. P. Snowden-I ff t, C. J. Marto ff , and J. M. Burwell, Phys. Rev. D 61,101301 (2000).[26] G. J. Alner et al., Nucl. Instr. Meth. Phys. Res. A 535 (2004) 644655.[27] K. Miuchi et al., Phys. Lett. B 654 (2007) 58.[28] D. Santos et al., J. Phys. Conf. Ser. 65 (2007) 012012.[29] D. Dujmic et al., Nucl. Instr. Meth. A 584 (2008) 327.[30] S.E. Vahsen et al., The Directional Dark Matter Detector (D3), EAS Pub-lications Series, 53, pp 43-50, (2012).[31] T. Naka et al., Nucl. Instr. Meth. A 718 (2013) 51952.[32] N. DAmbrosio et al., JINST 9 C01043 (2014).[33] D. R. Nygren, J. Phys.: Conf. Ser. 460 012006 (2013).[34] V. M. Gehman et al., JINST 8 C10001 (2013).[35] F. Cappella et al., Eur. Phys. J. C (2013) 73:2276.[36] S. Ahlen et al., Int. J. Mod. Phys. A 25:1-51 (2010).[37] F. Mayet et al., Physics Reports 627 (2016) 149.[38] S. Burgos et al., Astropart. Phys. 31 (2009) 261266.[39] D. Dujmic et al., Nucl. Instr. Meth. A 584 (2008) 327.[40] D. Nygren, PEP-198-1975.[41] S. Burgos et al., Nucl. Instr. Meth. A 600 (2009) 417-423.[42] J. B. R. Battat et al., Phys. Dark Univ. 9-10 (2014) 1-7, arXiv:1410.7821.[43] J. F. Ziegler, J. P. Biersack, and U. Littmark, “The Stopping and Range ofIons in Solids” (Pergamon Press, New York, 1985), .[44] L.G. Christophorou, J. K. Oltho ff , and M. V. V. S. Rao, J. Phys. ChemRef. Data 25, 1341 (1996).[45] M. G. Curtis, I. C. Walker, and K. J. Mathieson, J. Phys. D 21, 1271(1988).[46] M. S. Naidu and A. N. Prasad, J. Phys. D 5, 983 (1972).[47] C. S. Lakshminarasimha, J. Lucas, and D. A. Price, Proc. IEE 120, 1044(1973).[48] L.G. Christophorou and J. K. Oltho ff , J. Phys. Chem Ref. Data 28, 967(1999).[49] Y. Hayashi and Y. Nakamura, Proceedings of the International Confer-ence on Atomic and Molecular Data and their Applications, NIST SpecialPublication 926 , edited by W. L. Wiese and P. J. Mohr (U.S. Departmentof Commerce , Gaithersburg, MD, 1998); p.248.[50] C. J. Marto ff et al., Nucl. Instr. Meth. A 440 (2000) 355-359.[51] T. Ohnuki, D. P. Snowden-I ff t, and C. J. Marto ff , Nucl. Instr. Meth. A 463(2001) 142148.[52] C. J. Marto ff et al., Nucl. Instr. Meth. A 555 (2005) 5558.[53] K. Puskin and D. P. Snowden-I ff t, Nucl. Instr. Meth. A 606 (2009)569577.[54] M.P. Dion, S. Son, S. D. Hunter, and G. A. de Nolfo, Nucl. Instr. Meth. A648 (2011) 186191.[55] D. P. Snowden-I ff t and J.-L. Gauvreau, Rev. Sci. Instr. 84, 053304 (2013).[56] J. B. R. Battat et al., JINST 9 P11004 (2014).[57] C. Shalem, R. Chechik, A. Breskin, K. Michaeli, Nucl. Instr. Meth. A 558(2006) 475489.[58] C.K. Shalem, R. Chechik, A. Breskin, K. Michaeli, N. Ben-Haim, Nucl.Instr. Meth. A 558 (2006) 468474.[59] A. Hitachi, J. Phys.: Conf. Ser. 65, 012013 (2007).[60] B. Morgan, A. M. Green, and N. J. C. Spooner, Phys. Rev. D 71, 103507(2005).[61] B. Morgan and A. M. Green, Phys. Rev. D 72, 123501 (2005).[62] A. M. Green and B. Morgan, Astropart. Phys 27, 142149 (2007).[63] C. J. Copi, L. M. Krauss, D. Simmons-Du ffi n, and S. R. Stroiney, Phys. Rev. D 75, 023514 (2007).[64] J. Billard, Phys. Rev. D 91, 023513 (2015).[65] F. Sauli and A. Sharma, Annu. Rev. Nucl. Part. Sci. 1999. 49:34188.[66] RD51 Collaboration, http://rd51-public.web.cern.ch/RD51-Public/ .[67] F. Sauli, Nucl. Instr. Meth. Phys. Res. A, 1997, vol. 386, p. 531.[68] A. F. Buzulutskov, Instr. Exp. Tech., 2007, Vol. 50, No. 3, pp. 287.[69] A. Morozov, M. M. F. R. Fraga, L. Pereira, L. M. S. Margato, S. T. G.Fetal, B. Guerard, G. Manzin, F. A. F. Fraga, Nucl. Instr. Meth. Phys.Res. B 268 (2010) 14561459.[70] A. Kaboth, et al., Nucl. Instr. Meth. A 592 (2008) 63.[71] G.F. Reinking, L. G. Christophorou, and S. R. Hunter, J. Appl. Phys. 60,499 (1986).[72] A. Bondar, et al., Nucl. Instr. Meth. A 535 (2004) 299-302.[73] J. R. Janesick, “Scientific Charge-coupled Devices” (Society of Photo Op-tical Press, 2001).[74] S. F. Biagi, Nucl. Instr. Meth. A 421 (1999) 234-240[75] J. Brack et al., Physics Procedia 61 (2015) 130137.[76] J. B. R. Battat et al., Nucl. Instr. Meth. Phys. Res. A 794 (2015) 3346.[77] Q. Ri ff ard et al., arXiv:1504.05865v1.[78] A. Hitachi, Rad. Phys. Chem. 77 (2008) 13111317.[79] O. Guillaudin, J. Billard, G. Bosson, et al., EAS Publ. Ser. 53, 119 (2012),arXiv:1110.2042.[80] K. N. Buckland, M. J. Lehner, G. E. Masek, and M. Mojaver, Phys. Rev.Lett. 73, 1067 (1994).[81] K. N. Buckland, M. J. Lehner, G. E. Masek, IEEE Transactions on 44.1(1997) 6-13.[82] A. Pansky, A. Breskin, A. Buzulutskov, R. Chechik, V. Elkind, J. Vavra,Nucl. Instr. Meth. Phys. Res. A 354 (1995) 262-269.[83] M.M. Fraga, F.A.F. Fraga, S.T.G. Fetal, L.M.S. Margato, R. Ferreira Mar-ques, A.J.P.L. Policarpo, Presented at the Beaune 2002 Conference onNew Development in Photodetection, Nucl. Instr. and Meth. A 504 (2003)88.[84] A. Kozlov, I. Ravinovich, L. Shekhtman, Z. Fraenkel, M. Inuzuka, I. Tser-ruya, Nucl. Instr. Meth. Phys. Res. A 523 (2004) 345354.[85] P. Magain, F. Courbin, and S. Sohy, Astrophys. Journ., 494 (1998) 472-477.[86] H. Demers et al., Scanning 2011 May ; 33(3): 135146.ard et al., arXiv:1504.05865v1.[78] A. Hitachi, Rad. Phys. Chem. 77 (2008) 13111317.[79] O. Guillaudin, J. Billard, G. Bosson, et al., EAS Publ. Ser. 53, 119 (2012),arXiv:1110.2042.[80] K. N. Buckland, M. J. Lehner, G. E. Masek, and M. Mojaver, Phys. Rev.Lett. 73, 1067 (1994).[81] K. N. Buckland, M. J. Lehner, G. E. Masek, IEEE Transactions on 44.1(1997) 6-13.[82] A. Pansky, A. Breskin, A. Buzulutskov, R. Chechik, V. Elkind, J. Vavra,Nucl. Instr. Meth. Phys. Res. A 354 (1995) 262-269.[83] M.M. Fraga, F.A.F. Fraga, S.T.G. Fetal, L.M.S. Margato, R. Ferreira Mar-ques, A.J.P.L. Policarpo, Presented at the Beaune 2002 Conference onNew Development in Photodetection, Nucl. Instr. and Meth. A 504 (2003)88.[84] A. Kozlov, I. Ravinovich, L. Shekhtman, Z. Fraenkel, M. Inuzuka, I. Tser-ruya, Nucl. Instr. Meth. Phys. Res. A 523 (2004) 345354.[85] P. Magain, F. Courbin, and S. Sohy, Astrophys. Journ., 494 (1998) 472-477.[86] H. Demers et al., Scanning 2011 May ; 33(3): 135146.