Dosimetry and calorimetry performance of a scientific CMOS camera for environmental monitoring
Alexis Aguilar-Arevalo, Xavier Bertou, Carles Canet, Miguel Angel Cruz-Perez, Alexander Deisting, Adriana Dias, Juan Carlos D'Olivo, Francisco Favela-Perez, Estela A. Garces, Adiv Gonzalez Munoz, Jaime Octavio Guerra-Pulido, Javier Mancera-Alejandrez, Daniel Jose Marin-Lambarri, Mauricio Martinez Montero, Jocelyn Monroe, Sean Paling, Simon J. M. Peeters, Paul Scovell, Cenk Turkoglu, Eric Vazquez-Jauregui, Joseph Walding
AArticle
Dosimetry and calorimetry performance of a scientificCMOS camera for environmental monitoring
Alexis Aguilar-Arevalo , Xavier Bertou , Carles Canet , Miguel Angel Cruz-Pérez ,Alexander Deisting *, Adriana Dias , Juan Carlos D’Olivo , Francisco Favela-Pérez , Estela A.Garcés , Adiv González Muñoz , Jaime Octavio Guerra-Pulido , Javier Mancera-Alejandrez ,Daniel José Marín-Lámbarri , Mauricio Martinez Montero , Jocelyn Monroe , Sean Paling ,Simon J. M. Peeters , Paul Scovell , Cenk Türko ˘glu , Eric Vázquez-Jáuregui , JosephWalding Argentina: Centro Atómico Bariloche, CNEA/CONICET/IB, Bariloche;Mexico: Centro de Ciencias de la Atmósfera. Universidad Nacional Autónoma de México, CDMX, 04110 México Facultad de Ingeniería, Universidad Nacional Autónoma de México Instituto de Ciencias Nucleares, Universidad Nacional Autónoma de México, A. P. 70-543, México, D.F. 04510 Instituto de Física, Universidad Nacional Autónoma de México, A. P. 20-364, México D. F. 01000 Programa de Posgrado en Ciencias de la Tierra. Universidad Nacional Autónoma de México, CiudadUniversitaria, 04510 CoyoacánUnited Kingdom: Boulby Underground Laboratory, Boulby Mine, Saltburn-by-the-Sea Department of Physics and Astronomy, University of Sussex, Brighton Royal Holloway, University of London, Egham Hill * Correspondence: [email protected]† Current address: Particle Astrophysics Science And Technology Centre, Warsaw, PolandVersion September 24, 2020 submitted to Sensors
Abstract:
This paper explores the prospect of CMOS devices to assay lead in drinking water, usingcalorimetry. Lead occurs together with traces of radioisotopes, e.g.
Pb, producing γ -emissionswith energies ranging from 10 keV to several 100 keV when they decay; this range is detectable insilicon sensors. In this paper we test a CMOS camera (O XFORD I NSTRUMENTS
Neo 5.5) for its generalperformance as a detector of x-rays and low energy γ -rays and assess its sensitivity relative to theWorld Health Organization upper limit on lead in drinking water. Energies from 6 keV to 60 keV areexamined. The CMOS camera has a linear energy response over this range and its energy resolutionis for the most part slightly better than 2 %. The Neo sCMOS is not sensitive to x-rays with energiesbelow ∼
10 keV. The smallest detectable rate is 40 ± ± ± ± γ -rays, which corresponds to an incident activity of1.0 ± ±
33 Bq). The measured efficiency at this energy is 0.08 ± ± γ - and x-ray detector with sensitivity at the few to 100 ppb level for Pb in a sample.
Keywords:
Lead-210; commercial CMOS cameras; Scientific CMOS sensor; Gamma detection; X-raydetection; Lead in drinking water; Dosimetry; World Health Organisation
Submitted to
Sensors a r X i v : . [ phy s i c s . i n s - d e t ] S e p ersion September 24, 2020 submitted to Sensors
1. Introduction
Ingesting lead can have acute and chronic health effects and it is especially harmful to infants andchildren. There is no safe threshold for the onset of lead’s negative effects on the human condition anddamages are permanent – e.g. a child loses 3 IQ points on average when it consumes as much lead as25 µg kg − body weight per week over a longer period [1]. It is estimated that globally 26 million peoplein low- and middle-income countries are at risk of lead exposure [2] as e.g. people living in rural areasin Mexico [3]. The main source of lead pollution is from improper recycling of lead-acid batteries. Thetonnage of lead in production is rapidly increasing and has grown by more than an order of magnitude inthe past decade [2]. The trend towards clean technologies, such as electric cars, will likely increase thedemand on lead-acid battery recycling, therefore the corresponding pollution can be expected to increase.Lead screening tests in the UK, for example, are usually carried out in both public and private watersupplies [4]. In public water supplies these are done at water treatment facilities, service reservoirs, watersupply points and customer taps in water supply zones. In private water supplies, samples are collected atthe point of use. The laboratories that carry out the analysis of these samples are accredited by the UnitedKingdom Accreditation Service (UKAS) and the Drinking Water Testing Specification (DWTS). In low-and middle-income countries (LMICs), access to such testing is more limited and may be prohibitivelyexpensive. Therefore, a low-cost sensor of lead in drinking water would allow a wider range of peopleaccess to on-demand assay methods. A broader programme of measurements enabled by a low-costtechnology could have important impacts on mitigating lead intake through contaminated water.Most people carry a CMOS sensor in their pocket – the silicon chip in their mobile phones’ cameras.Even lower cost CMOS sensors are available off-the-shelf. These silicon chips are in principle capableof measuring radiation as x-rays, γ -rays, and radiation from β - and α -decays. Radioactive isotopes arefound in trace amounts together with stable isotopes of lead [5]. Detecting the decay radiation of theseisotopes can potentially enable lead detection in food and drinking water. Thus, radio assay methodsbased on a cheap and already common sensor such as a CMOS chip can be of great help to mitigate leadingestion, particularly in LMICs. A major challenge is to detect the small signal from trace contamination.This paper reports on a first step towards developing lead radio assay in CMOS by exploring the potentialof a scientific CMOS, operated to minimize noise. In this paper we qualify how a scientific CMOS, built foroptical light detection, performs as a radiation detector. Previous studies have shown that it is possibleto use CMOS sensors to distinguish between α -decays and other types of radiation, as well as countingevents from different radiation dosages [6]. It has also been shown that these sensors can provide goodspatial resolution, since they allow for a geometrical confinement of the received signal, for either x-rays orfor charged particles [7]. Spacial information can be used to distinguish electronic noise and backgroundradiation from the actual signal of interest, although it is by it self not a pre-requisite for dosimetryapplications. One of the radio-isotopes occurring with stable lead is
Pb, which decays by β − decay ( Q -value of63.5 ± Bi, which de-excites practically immediately while emitting a γ -raywith 46.6 keV energy. This lead isotope can occur in trace amounts with Pb, because it is a precursor ofthe stable lead ( i.e. Pb in this case) in the
U (or
Rn) decay series. Measurements in e.g. [5] show arange of 3.9 × − ppb to 2.4 × − ppb for certain Pb samples. Assaying Pb at trace levels is requiredfor sediment layer dating in geology [8], to assess pollution levels in environmental monitoring [9], and toidentify trace radioactivity in materials for particle physics experiments (e.g. DM searches) [10].The necessary radio-isotope assays for such studies are done with (high purity) germanium detectorsreaching concentrations as low as ∼ × − ppb. Germanium detectors are also used for e.g Pb assays ersion September 24, 2020 submitted to
Sensors of materials for DM experiments [11,12] and many more applications in science and technology. Tomeasure even lower concentrations of
Pb in a sample, ashing of that sample has been applied withsuccess (plants: [9], acrylic: [13]). Heavy (radio-)isotopes are retained in these processes, so the decay-rateper volume is enhanced after ashing. The lowest concentration measured reaches limits of down to1.1 × − g Pb per g of acrylic or 1 × − ppb [11].The measurement sensitivity goal for assay of lead in drinking water is the World Health Organization(WHO) upper limit. The WHO quotes a value of 10 ppb [1] as a guideline value for lead in water,although it states explicitly that there is no safe threshold for lead ingestion. In instances of drinking watercontamination, levels from 100 to 1000 times this value have been measured [14]. For this paper we study a CMOS sensor, which has been designed for imaging optical wavelengths.The paper is structured as follows: in Section 2 we give the technical details of the scientific CMOS sensor– O
XFORD INSTRUMENTS
Neo 5.5 scientific CMOS camera [15] (Neo sCMOS); in Section 3 we outlinethe experimental set-up and our measurement procedures; Section 4 describes the analysis procedure– the steps taken to go from camera exposure frames to the reconstruction of the energy deposited inthe CMOS chip. Results from measurements with radioactive sources and a x-ray tube are presentedin Section 5. We examine the general performance of the Neo sCMOS when exposed to x-rays and γ -rays. The performance analysis includes a study of the calorimetric capabilities of the camera, i.e. its capabilities to measure the energy of incident photons and its energy resolution as function of theincident energy (Sec. 5.1). Furthermore we measure the background rate without any source present,the minimal detectable rate with an
Am source and the camera’s detection efficiency for x-rays and γ -rays (Sec. 5.2). We use the measured efficiencies to make a rough estimate on the sensor thickness inSection 5.2.4. To check our measured results for consistency we construct a simulation of different layersof the Neo sCMOS in Geant4 [16], which is described in Section 5.3 and we compare our measured spectrawith these simulations in order to further understand the effects of the thickness of different layers in thechip. The paper concludes with a discussion of the results (Sec. 6) in which we estimate the sensitivity ofthe CMOS approach to measure lead concentration in water down to the WHO of 10 ppb [1].
2. CMOS camera specifications
An ideal silicon sensor for the measurement of x-rays and low energy γ -rays would have a thickconversion region to enhance the probability that photons are absorbed in the silicon, low noise, and ideallyallows the photons to pass unhindered by e.g. an entry windows to the chip. The DAMIC CCDs [17] are agood example for scientific sensors designed for this purpose, as they search for DM by measuring energydeposits (cid:46)
10 keV in their silicon. In this work we explore the potential of commercial CMOS camerasfor dosimetry applications, with the aim of understanding the prospects for CMOS-based sensors in thefield. The O
XFORD INSTRUMENTS
Neo sCMOS camera [15] was chosen because it provides readout noisecomparable with CCDs with relatively more widespread CMOS technology.The Neo sCMOS features a chip with 2560 × × (height × width). The spatial granularity of the pixelsis relevant for the image background characterisation and event classification as discussed in Section 4.1,Section 4.2.3 and Section 4.2.4. Each pixel in the Neo sCMOS’s chip has a typical well depth of 30 × e − (electron) and is equipped with its own micro lens . This micro lens array ensures that light arriving at Since the camera has a glass window before the sensor, we can not measure α particles emitted by the sources tested in Section 5.ersion September 24, 2020 submitted to Sensors
Setting ValueReadout binning 4 × Table 1.
Default settings of Neo sCMOS camera during all measurement runs. Runs with different settingsare explicitly noted. the chip’s surface is focused into the active region of the pixels. The Neo sCMOS has two acquisitionmodes: global shutter, where a full image frame is acquired and rolling shutter, where image data isacquired one row at a time. The detection limit for
Pb we are aiming to reach is small, and thus we choseoperating conditions to minimize the noise. For rolling shutter with a readout frequency of 200 MHz, theNeo sCMOS read noise specification is 1 e − (1.5 e − ) mean (RMS) at − ◦ C. This temperature is enabledby the Neo sCMOS’s cooling system, which allows for cooling the chip to − ◦ C ( − ◦ C) in a roomtemperature environment (with additional water cooling). For comparison, when a global shutter is usedthe mean read noise is 2.3 e − . Groups of hardware pixels (2 ×
2, 3 ×
3, 4 ×
4, 8 ×
8) can be binned together,prior to readout, in order to reduce the overall contribution of readout noise and increase readout speed.These are then read together as one readout pixel . Operating parameters were chosen to minimize the readnoise, to optimize the SNR for the expected small signal. Table 1 lists the operating parameters for theCMOS during the measurements presented in this paper.No information on the cross section of the camera chip, i.e. its different layers and the thickness of theactive silicon, is provided by the supplier. Information on the quantum efficiency is only available forradiation in the wavelength range from 300 nm to 1000 nm and not for short wavelengths (x-ray and γ -rayenergies). Furthermore, it is not known how layers on top of the silicon ( i.e. the micro lenses) affect ameasurement of x-rays / γ -rays. To understand the possible effect of these layers on our results, weperformed Geant4 simulations for the different radiation sources, cf. Section 5.3. We furthermore attemptto assess the thickness of the sensor by comparing our measurements to toy Monte Carlo simulationstreating the camera as only one silicon layer (Sec. 5.2.4).Many of the Neo sCMOS features are not present for cheap, commercial CMOS sensors, especially thevarious features reducing the noise as the chip cooling or the general noise figures. On the other hand,the Neo sCMOS is a sensor optimised for optical wavelengths and the thickness of its conversion layercan be expected to be on the order of a few µm to ∼
10 µm. The micro lense array on top of the actualpixels is an addtional layer as is e.g. a Bayer filter on top of the pixel matrix of a different camera chip.Furthermore incident radiation needs to passs through a window before it can reach the chip as is the casefor commercial sensors enclosed in a housing, like the camera chip of a mobile phone. The camera has allthese properties in common with cheap CMOS cameras.
3. Experimental description
CMOS data for radioactive source measurements are acquired inside a dark box, with dimensionsof 244 cm ×
122 cm ×
122 cm (L × W × H ). The large size of the box allows the distance between theradioactive source and the camera to be increased up to ∼ ersion September 24, 2020 submitted to Sensors (a) (b)Figure 1. (a) Photograph of the source holder and the Neo sCMOS camera in the dark-box. (a) The NeosCMOS in the LD Didactic X-ray apparatus (554 800) – the blue lines are for the water cooling circuit, whichis not necessary for operation in the dark box. understand the camera’s behaviour and to obtain different spectra for calorimetry. In particular, we takebackground data without any sources and also use
Am, Fe, and
Pb sources. Data taking with thex-ray tube is also performed under dark-box conditions, in a different enclosure, shown in Figure 1b anddescribed in Section 3.1.Prior to a data acquisition run the camera is cooled to − ◦ C and the respective source is positioned infront of the camera, with the source supported such that it is aligned on the chip centre, as in Figure 1a.The NEO sCMOS is controlled via a cable, which is fanned out of the dark-box and connects to a customPCIe card hosted in the data acquisition computer. We used the
Andor SOLIS for Imaging software packagefor the data acquisition as well as to set camera’s parameters. The operation settings in Table 1 are chosenfor several reasons: a readout binning of 4 × × × A LD Didactic X-ray apparatus (554 800) [18] is used for the X-ray data taking. Figure 1b shows thecamera inside the apparatus. The door to the compartment with the camera is closed before the datataking and the compartment is sealed light-tight. The x-ray tube and camera develop substantial heat,therefore the camera’s water cooling is used to ensure stable operation at − ◦ C. The rate of x-rays of theapparatus is larger than the rate of any radioactive source we use in this paper, allowing much shorterexposure times than stated in Table 1: 0.004 s and 0.025 s.The anode in the x-ray tube is made of Mo, with its characteristic K α and K β lines at 17.41 keV and19.61 keV, respectively. Data is also acquired with a Cu or Zr filter between the X-ray tube and the camera.The observation of absorption edges adds more energy measurements in addition to the two x-ray lines,which makes these tests valuable for the energy calibration of the sensor. In Section 5.1.1 our results withthe x-ray source are discussed.
4. Data analysis
The camera control software produces files in the FITS (Flexible Image Transport System) [19] format.Each FITS file can contain several frames, i.e.
2D arrays with one Analogue-to-Digital Unit (ADU) ersion September 24, 2020 submitted to
Sensors x [bin nb] y [ b i nnb ] image 10 − (a) BKG x [bin nb] y [ b i nnb ] image 10 − (b) Am Figure 2.
Raw frames (without any correction) recorded when no source of radiation is present, i.e background (BKG), and when the camera is irradiated with an
Am source. A zoom in x and y of Figure(b) is shown in Figure 5a. All images are zoomed on the intensity scale, visible in the colour-bar to the rightof each image. (Which means values larger than 200 are displayed as 200.) measurement for each camera pixel. For example, Figure 2 shows such a frame. After data taking, framesare processed by PYTHON code and CERN ROOT [20] routines. During normal data acquisition conditions(Table 1) we take several run s of N f =
100 frames. The data analysis described below identifies clustersdue to radiation in each frame. Ultimately all clusters found in the frames of several runs are combinedinto one set and further data analysis is done on this set (Sec. 5).
The image processing analysis corrects for the pixel pedestal in two steps, described in Section 4.1.1and Section 4.1.2, then finds clusters of signal pixels, described in Section 4.2, and measures the energywithin each cluster. These clusters are identified by their difference to the remaining pedestal value, afterthe corrections.4.1.1. Column correctionThe first step of the analysis is to correct for the raw image pedestal, which is defined as thebackground ADU measurement in each pixel in the absence of a source. We observe that the ADUvalues of each pixel in a given column are correlated with each other, giving rise to the distinct columns inFigure 2. This correlation pattern is no fixed-pattern-noise as it changes form frame to frame. Thereforewe employ the following approach to correct for it on a single frame basis. First, the mean column value (cid:104) C (cid:105) col ( x , n f ) and its standard deviation σ C col ( x , n f ) for each column is calculated: (cid:104) C (cid:105) col ( x , n f ) = N y N y ∑ y = C ( x , y , n f ) (1) σ C col ( x , n f ) = (cid:118)(cid:117)(cid:117)(cid:116) N y − N y ∑ y = ( C ( x , y , n f ) − (cid:104) C (cid:105) col ( x , n f )) , n f = j , x = k (2) ersion September 24, 2020 submitted to Sensors −
100 0 100 200 300 charge [ADU] − c o un t s [ / A D U ] BKG C BKG C col sub241 Am C Am C col sub (a) − −
100 0 100 200 charge [ADU] − c o un t s [ / A D U ] BKG T Am T (b)Figure 3. Histograms of the (a) raw pixel values C of all pixels in the frame in Figure 2a (BKG C ), Figure 2b( Am C ), and of the C col sub values calculated for the data in these frames using Eqn. (3). (b) Histogram ofthe C col sub values after time-series subtraction ( T , form Eqn. (4)). where C ( x , y , n f ) is the charge (in ADU) measured by a pixel at a given x , y position in frame n f . Thecolumn coordinate and the frame number are fixed ( x = k , n f = j ) while the sum runs over the rowcoordinate ( y = N y ). After a first calculation of the column mean and standard deviation usingEqn. (1) and Eqn. (2), all pixel values C ( k , y , j ) (cid:54)∈ (cid:104) C (cid:105) col ( k , j ) ± · σ C col ( k , j ) are excluded and (cid:104) C (cid:105) col ( k , j ) and σ C col ( k , j ) are calculated again until σ C col ( k , j ) changes less than 0.5 % between two iterations. Thisiterative approach is necessary in order to exclude pixels with a high charge value as e.g. hot pixels –transient high values in a certain x , y position – or pixels with a higher charge value due to a signal inducedby incident radiation. In the zoomed image in Figure 5a some pixels with a high charge value are visiblewith C ≥ C col sub ( x , y , n f ) = C ( x , y , n f ) − (cid:104) C (cid:105) col ( x k , n f ) (3) x k in (cid:104) C (cid:105) indicates that the column mean is the same for all the C ( x , y , n f ) along a column with x = k , i.e. in y direction.The result of this column-pedestal correction procedure is shown in Figure 3a, for the raw data of Figure 2.A raw frame recorded during data taking with no source and with Am has a mean of 129 ±
20 ADUand 130 ±
73 ADU, respectively, where the uncertainty is chosen to be one standard deviation. After thecolumn correction the mean moves to 0 ±
18 ADU and 0 ±
73 ADU, respectively.4.1.2. Time-series analysisAt this stage it is possible that there is still fixed-pattern-noise in the recorded frames, e.g. pixelswhich have, in every frame, a C value elevated over the neighbouring pixel’s values. Such pixels may behot pixels or pixels with charge values of only a few 100 ADU. In order to correct for these we adopt a time-series approach: all charge values C col sub ( x , y , j ) in the n f = j frame in a run, with N f frames in total, ersion September 24, 2020 submitted to Sensors x [bin nb] y [ b i nnb ] image 10 timeSeries − (a) BKG x [bin nb] y [ b i nnb ] image 10 timeSeries − (b) Am Figure 4.
Pair wise subtracted frames ( cf.
Sec. 4.1.2). The same frames as in Figure 2 are shown to illustratethis next step in the background rejection procedure. More high intensity points in the
Am frame than inthe BKG frame are visble, when comparing the two plots. are subtracted from their corresponding values in the n f = j + C col sub ( x , y , j + ) . The result are N f − T ( x , y , n f ) given by T ( x , y , n f ) = C col sub ( x , y , n f ) − C col sub ( x , y , n f + ) n f = N f − e.g. radiation from a source, may create negative entries during this procedure, cf Figure 5b. Thiscan be tolerated as long as the source rate is not too high such that transient features occur at the same x , y coordinate in two subsequent frames. Figure 3b shows the effect on the 1D charge distributions. Thetails of the distributions change due to the subtraction of transient pixels with a high charge value. Thisleads to a mean of 0 ±
21 ADU and 0 ±
112 ADU for the time-series corrected frame with no source and
Am, respectively. The standard deviation increases, since there are now more negative pixel values inthe distribution, from the pairwise subtraction of transient features.
The column corrected and time-series subtracted data, following Eqn. (3) and Eqn. (4) are thensearched for clusters. A cluster is defined as one or more spatially adjacent pixels which have a chargevalue larger than the remaining pedestal value.4.2.1. Threshold calculationThe threshold value is constructed by a data driven method: We check all N f − x = m , y = i measures over the course of a run. In the notation introduced before,these values correspond to all T ( x , y , n f ) , where x and y are held constant and n f runs from 0 to N f − Note that in Eq. (4) the n f starts at zero – hence the last index is N f − N f − Sensors
From these charge values a run-averaged pixel pedestal value p ( x , y ) and its standard deviation σ p ( x , y ) are calculated for all pixels at coordinates x , y . p ( x , y ) = (cid:104) T (cid:105) ( x , y ) = N f − N f − ∑ n f = T ( x , y , n f ) (5) σ p ( x , y ) = (cid:118)(cid:117)(cid:117)(cid:116) N f − N f − ∑ n f = ( T ( x , y , n f ) − p ( x , y )) , x = m , y = i (6)This is again an iterative procedure, similar to what is done to calculate the column mean. From subsequentiterations all pixel values T ( x , y , n f ) (cid:54)∈ p ( x , y ) ± · σ p ( x , y ) are rejected when using Eqn. (5) and Eqn. (6)to (re)calculate p ( x , y ) and σ p ( x , y ) . Both values are regarded as final when σ p ( x , y ) changes less than0.5 % between two iterations. While p ( x , y ) is by construction close to zero for column subtracted andtime-series subtracted data, σ p ( x , y ) has a minimal value slightly above 6 ADU and most probable valuebetween 14 ADU and 15 ADU, skewed towards higher values. The camera’s manual states a read noiseRMS value of 1.5 e − and a dark current of 0.015 e − /pixel/s. Combining these values in quadrature, whilsttaking into account the exposure time (10 s) and the readout binning (4 × ∼ − /ADU as specified by the supplier. The modificationof this RMS value by the before described column correction and the pair-wise subtraction have to betaken into account before comparing the RMS to σ p ( x , y ) . While the corrections described in Section 4.1.1lead to a negligible reduction of the RMS, the pairwise subtraction (Sec. 4.1.2) increases the resulting RMSby a factor of √
2. The smallest measured σ p ( x , y ) value of (cid:38) seed and a skirt pixel threshold intensity. The seed is a higher threshold value designed to quickly find the cluster’s largest charge values. The skirt is a lowerthreshold value designed to find potentially dimmer adjacent pixels to the seed pixel associated with thecluster. The following threshold condition is used to discriminate whether a pixel value T ( x , y , n f ) is partof the background or part of the charge deposit of a signal e.g. by radiation incident on the chip: T ( x , y , n f ) > p ( x , y ) + k · σ p ( x , y ) k = k seed ∨ k skirt (7)We distinguish two cases ( k seed or k skirt ) for the multiplier k : First, the factor to find the seed pixel for acluster ( k seed ). After the seed pixel has been found we check in its vicinity for pixels fulfilling Eqn. (7)with k skirt , where k skirt ≤ k seed . All contiguous pixels with charge values larger than the skirt threshold, aswell as the seed pixel, constitute one cluster. For each cluster we store its defining properties such as size,charge, x , y position, frame number, pedestal value and an identification number ( cf. Sec. 5). Figure 5cshows the clusters identified in the previously shown zoomed image of a frame taken with the Am Without time-series subtraction and column correction, the pedestal values and their standard deviation for every pixel shouldallow to discriminate between background and a charge signal, provided the fluctuations of the background are randomlydistributed. However, the column mean of a specific column changes from exposure to exposure, motivating the approachdescribed here.ersion September 24, 2020 submitted to
Sensors
10 of 31
50 100 150 200 x [bin nb] y [ b i nnb ] image 10 (a) raw
50 100 150 200 x [bin nb] y [ b i nnb ] image 10 timeSeries − − − (b) corrections applied
50 100 150 200 x [bin nb] y [ b i nnb ] clusters image 10 timeSeries (c) found clusters Figure 5.
Detail view (a) of the frame in Figure 2b – raw frame without corrections applied – and (b) of theframe in Figure 4b, pair wise subtracted frame. (c) Identified clusters in the zoomed image shown in (b)using k seed =
10 and k skirt = cf . Section 4.2, Eqn. (7). Thescale to the right of the image shows the cluster number. source. The parameter k seed and k skirt are optimised using data obtained without the presence of a radioactivesource (background data), while aiming for a low cluster count by recorded frame and a small clustersize. This optimisation is done on all the clusters found in our set of background run. Clusters in thebackground data will be due to cosmic radiation passing through the chip, due to noise fluctuations andcreated by radiation from natural radio-isotopes. For these sources we expect a cluster size to be small –especially since we use 4 × γ - and x-ray photons should depose their energylocalised and it is not likely that cosmic muons or β particles pass exactly parallel through the chip. With k seed =
10 less than 0.5 clusters per frame are found while a further increase to e.g. k seed =
20 does notresult in a further reduction. The cluster size decreases exponentially with k skirt and approaches a meanof ∼ k seed values. The change is no longer significant for k skirt ≥
3. Therefore, weuse k seed =
10 and k skirt = k seed =
10 and k skirt = • the frame number • the cluster number, which is a counter for all clusters in one frame • the x / y position, i.e the coordinates of the seed pixel • the cluster size , i.e how many pixels make up a cluster • the cluster charge , i.e the integral over all T ( x , y , n f ) in a cluster subtracted by the cluster pedestal • the cluster pedestal , i.e the integral over all the clusters pixels’ p ( x , y ) • the charge of the pixel with the highest charge in the cluster ( maximal charge ). The difference in the bare cluster counts and the shape of the cluster charge spectra ( cf.
Fig. 6) obtained during measurementswith a radioactive source versus measurements without the presence of a source shows that the clusters selected in Figure 5c arenot just noise, but due to the source radiation. In case of the time-series approach p ( x , y ) ∼
0. Without the time-series approach subtracting the cluster pedestal is essentialsince it is different from zero.ersion September 24, 2020 submitted to
Sensors
11 of 31 energy source intensity13.8 keV Np: L α , x-ray17.8 keV Np: L β , x-ray20.8 keV Np: L γ , x-ray26.3 keV Am: γ Am: γ Table 2.
Expected lines in the decay spectrum of
Am based on data from [21,22]. Np x-rays in the energyregion from 11.87 keV to 22.4 keV are expected to make up for another 37 % of intensity [22]. The x-rayenergies are approximate and are composed of several overlapping lines. Therefore no intensities are given,since these require assumptions on the detectors energy resolution.
Analysis based on clusters are done on the full set of clusters found in specific set of data taking runs.Figure 6a and Figure 6b show the cluster charge as well as the maximal charge without additional cutson cluster properties. Cluster charge and maximal charge spectra of the background data peak at a few100 ADU (Fig. 6b) and have a tail towards higher values. Their most probable cluster size is ∼ Am data in Figure 6a. The charge of a clustershould be proportional to the energy deposited by the incident radiation. From
Am-decay energyspectra in the literature there should be 5 prominent lines [21,23] at energies stated in Table 2. The spectrapresented here show four prominent peaks (Fig. 6, first column), which will be discussed in detail inSection 5.1.4.2.4. Sizes of identified clustersThe peaks visible in Figure 6a and Figure 6e sit on a floor which is itself related to the decay radiationof the source. Examining the cluster charge spectrum as a function of the cluster size (Fig. 6c) shows thatthis floor is mainly due to clusters with a size of 1 and 2 pixels. The peaks are furthermore significantlyless prominent for these cluster sizes than e.g. for cluster sizes of 3 pixels and 4 pixels. For cluster sizeslarger than four pixels the peak heights decrease again. It is interesting to note that there is a correlationbetween cluster size and cluster charge, i.e. energy deposited in the sensor. For increasing cluster size theratio of clusters with a large charge value to such with a low charge value increases. Gamma radiationand x-rays are expected to interact in the CMOS sensor and to release their energy locally. Therefore, theobserved cluster sizes are larger than expected, even more so, given the readout binning of 4 ×
4, resultingside length per readout pixel of 4 × =
26 µm each. As stated in Section 3 the exact layout of theactual CMOS is not known – its different layers may lead to a spread of the charge which reaches a few10 µm. Incident radiation can e.g. be absorbed in a non-active layer of the chip and then diffuse towardsthe collection zones. Another possible explanation is that a substantial fraction of the incoming γ energygets transferred to a few δ -electrons which can then travel more than a pixel length in the sensor, whilethey produce further ionisation. It can be excluded that the cluster size gets inflated by pixels accidentallyassigned to the respective cluster. Comparing the spectrum where only the most energetic pixel per clusteris shown (maximal charge) with the cluster charge spectrum shows that the information from the lowerenergy pixels in a cluster is needed to measure a spectrum with distinguishable peaks (Fig. 6a and Fig. 6e).For the analysis of the Neo sCMOS’ calorimetric response we require the cluster size to be larger than In some data taking runs the camera took a short time to reach a stable state. Therefore, as a precaution, we do not use clusters ofthe first five frames of each run. Occasional runs with no-stable camera conditions have in general been rejected.ersion September 24, 2020 submitted to
Sensors
12 of 31 charge [kADU] c o un t s [ /150 A D U ] cluster chargecluster max. charge (a) charge [kADU] c o un t s [ /600 A D U ] cluster chargecluster max. charge (b) charge [kADU] c o un t s [ /150 A D U ] Size = 1Size = 2Size = 3Size = 4Size = 5Size = 6Size > (c) charge [kADU] c o un t s [ /600 A D U ] Size = 1Size = 2Size = 3Size = 4Size = 5Size = 6Size > (d) charge [kADU] c o un t s [ /150 A D U ] cluster chargecluster max. charge (e) charge [kADU] c o un t s [ /600 A D U ] cluster chargecluster max. charge (f)Figure 6. The plots show cluster charge and maximal charge spectra for
Am data in the left column((a), (c), and (e)), and for background data in the right column ((b), (d) and (f)). As energy unit kilo ADU, i.e kADU, is used. The live-time of the camera during the
Am data taking is 2470 s where the camerais radiated with the corresponding source, while the live-time for the background data taking as 2945 s.The first row shows spectra containing all clusters found in the frames of the respective data taking runs,the second row shows cluster charge spectra grouped by cluster size in logarithmic scale while third rowshows all data displayed in the first column, which passed a size > ersion September 24, 2020 submitted to Sensors
13 of 31 >
5. Performance of the CMOS as radiation detector
The camera response to Fe,
Pb and
Am source radiation is examined in order to determine theNeo sCMOS’ calorimetric measurement capabilities. Background data obtained with no source present isused as well for this analysis. Figure 7 shows cluster charge spectra measured for these sources. To createthese, the analysis procedures detailed in Section 4 are applied to the raw frames and a cluster size largerthan two pixels is required for all entries in the plots.The contribution of the source radiation to the spectra has to be disentangled from the contribution ofthe background radiation. To this end, spectra obtained with radioactive sources and the backgroundspectrum are normalised to the same live-time and then the background spectrum is subtracted from thesource spectra. The results are shown in Figure 7b to Figure 7d.
Am:The cleanest spectrum is obtained with the
Am source, which has an activity of 344 ±
17 kBq as ofat the time the measurement. Americium-241 decays via an α -decay to Np. There are many possible α -decays with different Q values from 5000 keV to 5500 keV [22], where the most probable (85 %) decayhas an energy of 5485 keV. These α -decays occur together with γ -ray emission and x-ray emission by the Np atom [22]. Table 2 lists the two γ energies with the largest yield per decay as well as x-ray linesmeasured in Am spectra elsewhere. The CMOS chip of the Neo sCMOS camera is housed behind a glasswindow, with an assumed thickness of 1 mm – therefore the α -particles will not reach the sensor, since therange of α s of this energy is less than 100 µm [24]. The energy deposits measured with the Am sourceare thus for the most part due to γ - and x-rays. Attenuation lengths for different γ - and x-ray energies aregiven in Table 3. Fe:Iron-55 decays via electron capture to Mn [22]. After the decay, the electron shell re-arranges tomatch the levels of Mn and to fill the hole from the electron capture. By doing so, Auger-Meitnerelectrons with an energy of up to 6 keV are released as well as x-rays of 5.9 keV and 6.5 keV. For thesex-ray energies the yield per decay is 16.6 % and 7 %, respectively. Although the source used has a rateof ∼
100 kBq, the background subtracted spectrum in Figure 7c is compatible with zero. For low chargevalues, i.e low energy deposits, the spectrum is more erratic – however, no clear peak can be identified. Forphoton energies ≤
10 keV we estimate a lower limit for the photon absorption in glass with the data from[25], assuming 10 keV photon energy and a glass density of 2.23 g cm − . For a window of 1 mm and 2 mm,at least 97 % and 99.04 % of the x-rays are absorbed in the glass before they reach the chip, respectively.Therefore, the non observation of any clear peak is most likely due to the x-ray absorption in the NeosCMOS window. The uncertainty on the initial source activity is not known, therefore a 5 % error is assumed.ersion September 24, 2020 submitted to
Sensors
14 of 31 charge [kADU] c o un t s [ /150 A D U ] BKG Am Fe Pb (a) Combined plot charge [kADU] c o un t s [ /20 A D U ] (b) Am spectrum, counts scaled charge [kADU] − − − c o un t s [ /150 A D U ] (c) Fe spectrum, counts scaled charge [kADU] c o un t s [ /100 A D U ] (d) Pb spectrum, counts scaled
Figure 7. (a) Combined plot showing an overview of the different data sets acquired with the Neo sCMOS.The live-time of the background (
Am, Fe and
Pb, respectively) measurement is 29 450 s (24 700 s,41 800 s and 30 400 s, respectively). The spectra are shown on a log scale for better visibility since the rates ofthe sources vary as does the observed event rate. (b), (c), (d) shows spectra for the respective sources. Thesespectra have been scaled to a live-time of 47 500 s and are subtracted with the background spectrum scaledto the same live-time. Shaded regions represent the statistical error. For all plots a cluster size > cf .Section 5.1.1. (Note that only the Gaußians are plotted, and not the additionally fitted backgrounds.)Energy Attenuation length/ [ keV ] Range in Silicon γ -/X-ray β −
10 111 µm 1.2 µm15 365 µm 2.5 µm45 7 mm 16 µm Energy Attenuation length/ [ keV ] Range in Silicon γ -/X-ray β −
60 12 mm 28 µm100 20 mm 66 µm1000 59 mm 2 mm
Table 3.
Approximate ranges of γ -rays and β − s (electrons) for typical decay energies in Silicon. For the γ -rays the attenuation length is calculated from the attenuation cross section given in [26] using the densityof silicon-dioxide. The same density is used to calculate the electron range from the CSDA range forelectrons given in [27]. ersion September 24, 2020 submitted to Sensors
15 of 31
Pb:Lead-210 decays via β − decay to Bi as mentioned in Section 1.1. The most probable β − decay(84 %) results in an excited state of Bi, whilst emitting an electron with an average decay energy of4.16 keV. The nucleus de-excites by emitting a γ of 46.5 keV with a 4 % yield per decay. The de-excitationis accompanied by the emission of x-rays from approximately 9 keV to 16 keV with a yield of 22 % perdecay. The second most probable decay (16 %) is a β − decay with an electron mean energy of 16.2 keV to Bi in the ground state [22]. The
Pb source used has a rate of ∼
185 kBq. It holds the lead diluted innitric acid in a small glass vial. It is not likely that any of the low energy β -radiation is detected by theNeo sCMOS, given that the decay electrons have to traverse the liquid, the glass of the vial and of the NeosCMOS before it can be detected by the CMOS chip. Therefore, similarly to the Am source, only thex-rays and γ -rays are measured.The Pb spectrum in Figure 7d contains fewer counts than the
Am spectrum (Fig. 7b). There areseveral factors contributing to this: First, the activity of the
Pb source is a factor of 1.85 lower than theactivity of the
Am source. The latter source has also a significantly smaller extent – compared to theCMOS sensor it can be considered as a point source, while the lead source extends over a vial of more than1 cm length and 0.5 cm diameter. Next, the γ yield for the two sources differs greatly – comparing ∼ ∼
36 % for the
Pb 46.5 keV γ -ray and the Am 59.5 keV γ -ray. In order to establish whether the Amand
Pb spectra are consistent with each other, we first need to establish the overall energy scale andcompare peaks at a known energy directly.5.1.1. Energy response calibrationAll the spectra presented so far are shown with analogue-to-digital units as unit of the deposited energyin the detector. The Neo sCMOS’s manuals consulted during this work do not state a conversion factor fromADU to energy in eV. However, the report [28] specifies a gain of either 0.59 e − /ADU or 0.67 e − /ADUaccording to the supplier. These gain values translate to a conversion factor of either 2.154 eV/ADU or2.446 eV/ADU, respectively, accounting for the W factor in Si of 3.65 eV to create an electron-hole-pair[29]. In order to establish the exact energy scale, the known energies of radioactive sources from literatureare matched to the ADU values at which peaks are observed. All large peaks in the cluster charge spectraare fitted with a Gaußian curve and their mean energy, ε peak , and σ is extracted. Table 4 lists all peaks usedfor this analysis and the result of the fits. The Gaußians are furthermore plotted in Figure 7b, Figure 7dand Figure 8a. The fits are done locally – in the ranges from ε min to ε max as specified in Table 4 – and wherenecessary a polynomial of order one is added to the Gaußian curve to account for the floor due to otherradiation. For the peaks at the high energy end of the Am and
Pb spectrum an error-function is usedinstead of a polynomial.The second, independent, dataset to obtain the energy scale calibration uses measurements where theCMOS is irradiated by an x-ray tube, described in Section 3. For the spectra obtained with the x-ray tubeusing only Gaußian fits with a local background is not sufficient (Fig. 8a): The two characteristic peaks ofthe molybdenum x-ray tube are expected to be located on top of the bremsstrahlung spectrum of the tube.For low x-ray energies the camera has negligible calorimetric capabilities as seen in the measurements withthe Fe source (previous section, Fig. 7c). Hence, the Neo sCMOS should become efficient for x-rays of themolybdenum x-ray tube somewhere after ∼ bremsstrahlung and eventually the molybdenum K α and K β peaks at 17.4 keV and19.6 keV, respectively. Figure 8a shows the spectrum, the fit to the spectrum, and the fit’s components. Anonset of counts is observed at ∼ ersion September 24, 2020 submitted to Sensors
16 of 31
Source Am Pb Mo x-ray tubeRadiation γ x-ray (Np) γ x-ray Zr edgeExpected energy [ keV ] ε min [ kADU ] ε max [ kADU ] ε peak [ kADU ] σ peak [ ADU ]
133 276 95 124 199 239 163 353 n.a. χ / N dof Table 4.
The table lists the fit results (peak position ε peak and standard deviation σ ) of Gaußian peak fitsin Figure 7b, Figure 7d, and Figure 8a as well as the absorption edge read off from Figure 8b. (Since theproperties of the absorption edge are not determined by a fit, the corresponding values in the “Zr edge”column are labelled n.a. for non applicable.). The expected energies have been extracted from [22] and the Zrabsorption edge form [26]. charge [kADU] c o un t s [ /150 A D U ] No filterFull fitBrems.Mo K α Mo K β (a) charge [kADU] c o un t s [ /150 A D U ] No filterCu filterZr filter (b)Figure 8. (a) Spectrum recorded with the molybdenum x-ray tube ( cf.
Sec. 3.1) as well as fitted curvesto establish the position of the K α and K β peak. (b) The same data as in (a) is shown together with datarecorded when a Zr or a Cu filter is placed between the x-ray tube with Mo target and the Neo sCMOS.For this plot all data has been normalised to a live-time of 50 ms and a cluster size > ersion September 24, 2020 submitted to Sensors
17 of 31 charge [kADU] e n e r g y [ k e V ] (a) ε peak [keV] . . . . . σ p e a k / ε p e a k Am Mo, x-ray
Am Mo, x-ray Am Am Pb Am (b)Figure 9. (a) Comparison between the expected and the measured peak and edge energies ( ε peak ) in Table 4.One σ peak of the peak is used as uncertainty for ε peak and the red line through the points is a fit without anadditional axis intercept. (b) Measured energy resolution ( σ peak / ε peak ) as function of the measured peakposition. The boxes indicate to what spectrum a given peak belongs to. is no clear expected functional shape for the bremsstrahlung contribution, we model it as the minimalfunctional addition ( Brems ( ε ) ) needed so Brems ( ε ) + Gauß ( ε ) K α + Gauß ( ε ) K β fits the data well. Brems ( ε ) = p · exp (cid:32) − (cid:18) ε − p σ (cid:19) (cid:33) + p · (cid:16) − erf (cid:16) p · ( ε − p ) (cid:17) (cid:17) + p · ε + p (8) Gauß ( ε ) K j = p j · exp − ε − ε j peak σ j j = α ∨ β (9)In these equations ε is the cluster charge (or energy deposited in the chip) in ADU. The parametrisation(8) for the bremsstrahlung contribution yields the lowest χ / N dof of 4.62 for the total fit of Brems ( ε ) + Gauß ( ε ) K α + Gauß ( ε ) K β to the data, whilst all fit parameters are free. The extracted parameters of the two K lines ( ε peak , σ peak ) are listed in Table 4.Using a Cu or a Zr foil to filter the molybdenum x-rays results in the spectra shown in Figure 8b. Theabsorption edges of those two elements for energies higher than ∼ (cid:46) ∼ ± ± ersion September 24, 2020 submitted to Sensors
18 of 31 fit χ / N dof is 1.64 while using a function with an intercept results in a χ / N dof of 0.56, an intercept of-0.36 ± ± ± χ / N dof in the latter case indicates over-fitting.The measured conversion factors matches well with the higher of the two gain values discussed before, i.e. σ peak divided by the peak positions ε peak (Tab. 4). Asthe uncertainty on σ the uncertainty of the fit is used while σ peak itself is used as the uncertainty of thepeak position ε peak . The uncertainty on the energy resolution, ∆ (cid:16) σ peak / ε peak (cid:17) includes both of thesecontributions. For the most part the resolution is better than 2 %. Outliers from this trend are the twomolybdenum x-ray lines and the Np, L γ line (at ∼
20 keV). The two x-ray lines are extracted from amore complicated fit with the worst χ / N dof and the uncertainty on their σ peak values is likely to beunderestimated.The energy resolution is determined by several factors. The full containment of all electrons producedduring the photon conversion and their subsequent readout will play an important role. Furthermoretheir can be pixel-to-pixel variations of each pixels’ amplifier gain. An increasing trend in cluster sizewith increasing energy has been observed, as stated in Section 4.2.4, but the good linearity of the eV toADU relation suggests that all electrons produced by a photon interaction are read out. The pixel-to-pixelamplifier gain variations for the Neo sCMOS are not known. Typical values are in the few % range – e.g. [30] shows a variation of about (cid:46) ∼ In this section we quantify the minimum detectable radioactivity using the Neo sCMOS, and measurethe efficiency of the sensor as a detector for γ - and x-rays. Both are done using the Am source, since thissource has a well suited activity and an energy spectrum with clear peaks.5.2.1. Geometric acceptance of the experimental set-upThe fraction of the
Am activity detected by the sensor depends on the source-detector distance.The geometric acceptance ( (cid:101) G ) is calculated assuming the Am emits radiation as a point source, as (cid:101) G = A spherical cap A sphere A camera A (cid:35) camera plane = r ( r in cm ) (10)The calculation of this expression exploits the sphere into which the source emits radiation and itsintersection with the plane of the camera chip. Therefore, A sphere is surface area of that sphere, A spherical cap is the surface area of the base of the spherical cone covering the camera chip, A (cid:35) camera plane is the ersion September 24, 2020 submitted to Sensors
19 of 31 distance [cm] − − − − g e o m e t r i c a cce p t a n ce [ % ] Figure 10.
Geometric acceptance of the experimental set-up with the Neo sCMOS: Shown is the analyticalfunction for a point source in Eqn. (10) and values from a toy Monte Carlo resembling the actual sourcegeometry. corresponding surface area of an otherwise similar cone with a flat base, A camera is the surface area of thecamera and r is the camera to source distance. Figure 10 compares the analytical estimate of Eqn. (10) witha toy Monte Carlo simulation, using the actual source geometry, showing good consistency.5.2.2. Minimum detectable radioactivityData is acquired for 3800 s each at distances from 1.8 cm up to 258.6 cm between source and sensor. The detected spectra, the inferred incident activity using the geometric acceptance from Eqn. (10) andthe known source activity are shown in Figure 11. At a distance of 258.6 ± ± γ -rayenergies emitted by the Am source, as opposed to the full source activity. These are obtained bymultiplying the incident activity for all
Am decays by the respective peak yields which are 2.40 ± ± Am source and the uncertainty of the emission probabilities. In the case of the toy Monte Carlosimulation, the statistical uncertainty on the counts obtained is used.In order to estimate the minimum detectable radioactivity and eventually the Neo sCMOS’ detectionefficiency, the detected rates in Figure 11b are determined as rate = (cid:82) spectrum d timetime (cid:20) countss (cid:21) . (11)The measured rate without any source present is calculated as well using the spectrum in Figure 6f. During30 400 s of data taking 619 clusters with a size > ± σ of the integration limits around the 26.3 keV and the59.5 keV γ -peak energies (Tab. 4) the background rate measured with no source present is 1.1 ± From now on we omit “between camera chip and radioactive source” when referring to the distance between these twocomponents.ersion September 24, 2020 submitted to
Sensors
20 of 31 distance [cm] − − i n c i d e n t a c t i v i t y [ c o un t s / s ] Act., full (MC)Act., 26 . . (a) distance [cm] − − − − d e t ec t e d r a t e [ c o un t s / s ] Full.Peak, 26 . . · σ BKG R . · σ BKG R . · σ (b) energy [keV] c o un t [ /365 e V ] .
75 cm2 .
55 cm4 . . . (c) energy [keV] c o un t [ /365 e V ] . . . . . (d)Figure 11. (a) Simulated incident activity on the camera chip for different camera-to-source distances for thefull source activity, and for the fractional activities corresponding to the 26.3 keV and 59.5 keV γ -lines. (b)The measured rates are calculated using Eqn. (11) and integrating over the full Am spectra in Figures (c)and (d), or integrating only over the energy region of the 26.3 keV and 59.5 keV peaks in the same spectra.The measured background rate is established by integrating over the full spectrum recorded in absence ofany source and in ± σ peak windows around the mentioned peak energies. The shaded areas in the left halfof the plot below the lines of the respective background rates show the five and 1.28 standard deviation ( σ )regions around the background rate in yellow and green, respectively. (c) and (d) Measured Am spectraat different distances with a live time of 3800 s. The data has to pass the cluster size > e.g. Fig. 6f) and is normalised to the same live-time. ersion September 24, 2020 submitted to
Sensors
21 of 31 incident activity measured rate5 · σ
90 % CL 5 · σ
90 % CLfull spectrum 7 ± ± ± ± ± ± ± ± ±
33 Bq 57 ±
33 Bq 1.5 ± ± Table 5.
The lowest incident activities and the corresponding measured rates obtained with the NeosCMOS. See Section 5.2.2 and Figure 11b. Values for two cases – either a 5 σ or a 90 % Confidence Level (CL)condition – are quoted. (0.10 ± i.e. σ , around the measured background rate, wherewe use the background uncertainty as standard deviation, i.e. σ . Overall, our results are for the most casethe same when when using 5 σ or a 90 % CL criterion to differentiate between the measured backgroundrate and the source rate. Table 5 shows all result for both cases, in the following we refer to the resultsobtained with the 5 σ criterion. The detected rate measured for a distance of 128.5 cm is the first to becompatible with the background rate added with five standard deviations (27 ± ± ± ± Am γ -lines allows to establish whether thislimit can be improved taking calorimetric information into account. For the 26.3 keV line a falling trend forthe detected rate is observed over the full distance range (Fig. 11b). Comparing to the rate measured whenno source is present the point at 64.5 cm is already compatible within five standard deviations. For 32.5 cmdistance the detected activity is 4 ± ± γ line, the 32.5 cm point is already compatible with the background rate for this energywindow within five standard deviations. The detected rate at 16.5 cm of 1.5 ± ±
33 Bq, is thus the minimal detected rate, different from the background rate inthis energy window.5.2.3. Detection efficiency of the Neo sCMOSA comparison between Figure 11a and Figure 11b allows to calculate both the intrinsic and absoluteefficiency of the detector. These are defined as (cid:101) intrinsic = number of particles recordednumber of particles incident on the detector (12) (cid:101) absolute = number of particles recordednumber of particles emitted by source = (cid:101) intrinsic · (cid:101) G , (13)where (cid:101) G is the geometric acceptance, which is given by Eqn. (10). The ratio of the recorded rate (Fig. 11b)and the incident activity (Figure 11a) yield the intrinsic efficiency of the Neo sCMOS camera for γ -raysof an Am source. This is shown in Figure 12a for the 26.3 keV and 59.5 keV peaks as a function of the Thus count rates higher than 21.41 mHz, 1.4 mHz and 0.17 mHz for the full background rate, the background rate in the 26.3 keVpeak window, and the background rate in the 59.5 keV peak window, respectively, are larger than the background at 90 % CL.ersion September 24, 2020 submitted to
Sensors
22 of 31 distance [cm] − − − i n t r i n s i ce ffi c i e n c y [ % ] . . . . (a) distance [cm] − − − − − a b s o l u t ee ffi c i e n c y [ % ] . . (b)Figure 12. (a) Intrinsic and (b) absolute efficiency for the 26.3 keV and the 59.5 keV peaks. The bands in (a)correspond to the result of fitting a constant to the data as discussed in Section 5.2.3 with its error bars. Thetwo full sample points are extracted from data in Figure 7b in the same manner as all the other points areextracted from the data in Figure 11. camera-to-source distance. For completeness, the absolute efficiencies for these two γ -rays are shown inFigure 12b.By definition (cid:101) intrinsic has to be independent of the distance. The results of a constant fit to the data areshown in Figure 12a. However, a weak distance dependence of (cid:101) intrinsic is observed; this is likely causedby a slight discrepancy between the calculated geometric acceptance and the actual source geometry.However, all points are within error-bars compatible with a constant as expected from Eqn. (12). Theabsolute efficiency’s dependence on the distance in Figure 12b is expected, due its dependence on thegeometric efficiency (Eqn. (13)). The corresponding intrinsic efficiencies for the 26.3 keV and the 59.5 keVpeaks are 0.08 ± ± γ -rayis about a factor of 80 higher than the efficiency to detect a γ -ray with an energy of 59.5 keV. To check thisratio for consistency the material composition and thickness of the sensor would need to be known, whichis however not the case. Figure 13b in the next section shows photon absorption efficiencies calculated fromthe attenuation coefficients in [31]. Depending on the CMOS thickness, the difference in photo-absorptionfor 59.5 keV and 26.3 keV can easily reach a factor 80. However, the efficiency at the 59.5 keV γ -line appearslower than expected from photo-absorption cross sections in Si.5.2.4. CMOS sensor thicknessBased on the efficiencies determined in the previous section one can estimate the thickness of theCMOS chip used in the Neo sCMOS camera. A toy Monte Carlo (MC) simulation of the Am spectrumis used together with the photon attenuation coefficient from [31] to calculate, for each photon, theenergy-dependent photon absorption probability in silicon. This step is repeated for different siliconthicknesses. For each thickness we create a spectrum of the photons absorbed in Si – these photons arethe ones which would be measured by a chip as in the Neo sCMOS. The ratio of the unattenuated
Amtoy MC spectrum divided by the spectra of absorbed photons is constructed. Finally, these ratios arecompared to the ratio of the measured count rate at the γ peak energies divided by the incident activityfor the respective γ line.For the toy MC, all possible γ -rays and x-rays of Am decays as listed in [32] are taken into account. ersion September 24, 2020 submitted to
Sensors
23 of 31
20 40 60 energy [keV] c o un t s [ /33 e V ] µ m2 µ m3 µ m4 µ m5 µ mNo att, × / (a)
20 40 60 energy [keV] . . . . e ffi c i e n c y [ % ] µ m2 µ m3 µ m4 µ m Am data (b)Figure 13. (a) Am γ -ray and x-ray simulation for 1 M decays. The spectrum with no material effectstaken into account ( No att. ) is scaled by a factor of 1/1000 to improve the readability of the plot. The otherspectra show the photons which are absorbed by a silicon layer of given thickness. (b) Photon absorptionefficiencies in silicon for different silicon layer thickness as well as the measured intrinsic efficiency of theNeo sCMOS at the two Am γ -ray energies. These are used to create a probability density function for photons emitted during
Am decays. Each γ - and x-ray line is represented by a Gaußian peak where the amplitude is proportional to the yield perdecay. The ε peak is set to the respective x-ray or γ -ray energy and the σ peak is set to 2 % of the peak energyto approximate the measurement for the Neo sCMOS in Section 5.1.2. Creating 1 M decays from thisprobability density function results in the No att. cluster charge spectrum in Figure 13a. This unattenuatedspectrum does not take material effects into account.Figure 13a shows also spectra of photons absorbed in silicon layers with a thickness from 1 µm to 5 µm.Toy MC spectra for absorbed photons and the measured spectrum ( e.g.
Fig. 7b) are similar. The orderingof the different peaks’ height is the same except for the fact that the largest measured peak is at 18 keV,while the peak with the most counts in the simulation is the peak at 14 keV. In Section 5.1.1 the detectionefficiency is observed to drop at lower energies – most notably making it impossible to detect a peak below ∼
10 keV, cf . Figure 7c and Figure 8. The efficiency turn-on responsible for this behaviour may also affectthe peak height of the 14 keV peak and lead to the non-observation of the small peak visible at ∼
11 keV inFigure 13a.Figure 13b shows photon absorption efficiency curves calculated from the coefficients in [31] for 1 µm, 2 µmand 4 µm silicon layer thickness. Furthermore, the intrinsic efficiency values calculated in Section 5.2.3are shown (Fig. 12a). The two points follow roughly the trend expected by photon absorption in silicon,although the efficiency measured for the 59.5 keV peak is lower than expected from any absorptionefficiency curve. The actual peak height of any γ -line is given by the absorption probability as well as bythe capability of the active material to contain the full energy deposit. The latter is not contained in the toyMC simulation and this may explain that the 59.5 keV does not fit with the displayed curves.The 26.3 keVfavours a silicon layer thickness of 2 µm. This scale of a few µm is on the same order of magnitude asthe pixel dimension and seems reasonable for commercial CMOS chips, with a typical silicon-dioxidelayer thickness of less than 10 µm [6,7]. Since the silicon thickness is not the only contribution to the NeosCMOS intrinsic efficiency, the estimate in this section can only be seen as a lower limit to the actual sensorthickness. ersion September 24, 2020 submitted to Sensors
24 of 31
To extend the limited knowledge of the sensor geometry beyond the estimations based on the toy MCsimulations, a Geant4 MC simulation study is carried out to estimate the impact of the window, the microlens array and sensor thickness on the γ - and x-ray absorption.Geant4 [16,33,34] version 10.5 patch 1 is used to simulate the primary particles and the productionof the resulting secondary particles, for the tracking of all particles through the detector geometry,and to assess the energy deposition in the sensitive detector parts. This work employs physics listswhich follow mainly the Shielding physics list of the above mentioned Geant4 version with one physicsmodification; namely G4EmStandardPhysics_option4. G4EmStandardPhysics_option4 is used instead ofG4EmStandardPhysics, because the former is more accurate for low-energy electromagnetic interactions.5.3.1. Detector geometry and simulated particlesThe sensor is modelled as a silicon CMOS layer, behind a glass layer to represent the entrance window,behind an acrylic layer to represent the micro lens array. For this simulation, the material of the window ischosen to be SiO . The camera specifications indicate an organic material for the micro lens. Thus, for thematerial of this volume element, acrylic C H O is chosen. For the thicknesses of each volume element,values in the range of O ( − ) are used. This study varies the thicknesses of the silicon and glasswindow. As discussed in Section 5.1, Pb decays either by a combined γ - and β -decay (84 ± γ -ray has an energy of 46.5 keV and an average β energy of 4.2 keV, or by a pure β -decay (16 ± β energy of 16.2 keV. When simulating electrons of this energy impinging on the detectorwith a 200 µm glass window and 5 µm of Si thickness, there are zero hits on the active region out of 10 simulated events. Since the energy of the other β is even lower, the signature Pb γ -ray of 46.5 keV is themain focus of this simulation work whereas β -particles are not included.The simulated particle source is located 14.06 mm from the outermost layer of the surface, and firesmono-energetic γ - and x-rays directly at the detector.To construct the energy variable used to compare with data, the simulated ionisation energy deposition inthe Si layer is recorded for each incident particle, and then smeared according to the energy resolutionfunction σ peak ε peak = p + p √ ε fit (14)fitted to the data in Figure 9b.5.3.2. Analysis of the simulated spectraA set of different thicknesses for the silicon layer and the glass window are simulated for comparisonwith the data. The list of thicknesses simulated ranges from 2 − − • The total number of events registered in the silicon layer increases with Si thickness which is causedby more particles being absorbed by a thicker Si layer. • The number of events in the photo-peak and
Compton continuum increases and decreases, respectively,as the silicon thickness increases. This is because the fraction of events for which the incident photons’energy is fully contained in the Si rises as the thickness increases. ersion September 24, 2020 submitted to
Sensors
25 of 31
Silicon thickness Glass window Full absorption η (cid:2) − (cid:3) [ µm ] thickness [ µm ] efficiency (cid:2) − % (cid:3) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Table 6.
Corresponding full absorption efficiencies and η -values of 46.5 keV γ -rays for silicon layers ofdifferent thicknesses. The information regarding the number of events in the photo-peak can be used to characterise the thicknessof the silicon layer in conjunction with the amount of events in the Compton continuum. The ratio η isused as a variable to compare the simulated templates with data: η = ε peak + σ peak (cid:82) ε peak − σ peak N events d N ε peak − σ peak (cid:82) N events d N (15)where ε peak stands again for the energy of the photo-peak, σ peak for the peak’s standard deviation and N events for the number of events. The exact value of the photo-peak and the energy resolution of the detectorneed to be known, to precisely define the integration limits in (15). We take ε peak to be 46.539 ± σ peak from the energy resolution for this energy using Eqn. (14) to be 0.595 keV. η -valuesfor different silicon layer and window thickness calculated from the simulation results can be found inTable 6.5.3.3. Comparison of simulation with the experimental dataThe thickness of the Neo sCMOS active layer, assuming Si as material, is estimated by comparingthe spectra in Figure 14 and the photopeak ratio η with the data. When this variable is calculated for thedata (black line in Fig. 14), it is found to be η data = ( ± ) × − . Comparing η data with the valuescalculated from simulated templates ( cf. Tab. 6), it can be seen clearly that a combination of 4 µm siliconlayer and 1000 µm glass window comes closest to this value with η = ( ± ) × − .As another check, one can investigate the spectral shape of simulation with respect to experimental data.Above 30 keV, the spectral shape of the data are closest to the spectrum of the 2 µm Si thickness simulation.Below that, the shape matches of the measured spectrum lies between the spectrum simulated for a 3 µmand 4 µm thick Si layer. All templates except for the 4 µm and 5 µm ones show a discontinuity before thephoto-peak as the energies of the particles increase, which is an expected result due to thinner siliconlayers being less efficient in stopping γ -rays and containing the particles energy. ersion September 24, 2020 submitted to Sensors
26 of 31 energy [keV] c o un t [ /365 e V ] µ m3 µ m4 µ m 5 µ m Pb data (a) energy [keV] c o un t [ /365 e V ] µ m3 µ m4 µ m 5 µ m Pb data (b) energy [keV] c o un t [ /365 e V ] µ m3 µ m4 µ m 5 µ m Pb data (c)Figure 14.
Simulated spectra of 46.5 keV γ -rays shot towards the detector with varying active silicon layerthicknesses along with constant micro lens thickness of 4 µm with following glass window thicknesses: (a)200 µm, (b) 1000 µm and (c) 2000 µm. Photopeak of each individual simulated spectrum is normalised tothe photopeak of experimental data ( Pb). The γ -rays’ origin from a point source centred on the detectorover the window, cf . Section 5.3.1. ersion September 24, 2020 submitted to Sensors
27 of 31 η -values and the spectral shape between the data and the Geant4 simulationsuggest a silicon thickness between 2 µm and 4 µm for a window of 1000 µm, where the comparison to η data places the thickness at the high end of this range. Table 6 shows that for this geometry only ( ± ) × − % of the γ -rays of Pb are fully absorbed in the silicon layer, i.e. contribute to the photo-peak.In Section 5.2.3 the intrinsic efficiencies for the 26.3 keV and 59.5 keV
Am-peak are determined tobe ( ± ) × − % and ( ± ) × − %, respectively. The efficiency for the 59.5 keV γ -line iscompatible with the value simulated here for 46.5 keV γ -rays. Given the fact that the measured value is at ahigher energy than the 46.54 keV of the simulated γ -rays, the Geant4 simulation most likely underestimatesthe Neo sCMOS detection efficiency at 46.5 keV slightly or the actual Si thickness is larger than 4 µm. Thisis, because the efficiency to detect 46.54 keV photons is expected to be larger than the one for higher energyphotons.Both the Geant4 simulation of Pb γ -rays and the toy MC simulation of Am photons place the siliconlayer thickness in the range between 2 µm and 4 µm, cf.
Section 5.2.4, Figure 12b. While the full Geant4simulation gives more parameters to compare between the data and the simulation, the toy MC simulationis significantly faster as it runs in an instant. The agreement between the two gives confidence to use theless sophisticated toy MC simulation in instances where a full fledged Geant4 simulation is not easilyaccessible.
6. Summary and discussion
An O
XFORD I NSTRUMENTS
Neo 5.5 scientific CMOS [15] camera is examined as a detector for photonsin the x-ray and low γ -ray energy regime. This camera is designed to image photons of optical wavelengths.The analysis (Sec. 4) of camera images identifies clusters (Sec. 4.2) – contiguous pixels with high chargevalues (unit: ADU) – which are due to energy deposits of radiation impinging on the camera chip.Requiring the cluster size to be larger than 2 pixel allows to sufficiently reject most of the clusters due tobackground radiation. We note a trend towards larger cluster sizes for increasing photon energy (Sec. 4.2.4).The relation of the cluster charge in ADU to eV is measured to be 2.467 ± ± ∼
10 keV, which is most likely due to the photonabsorption in glass for energies ≤
10 keV ( cf . the comments on Fe in Sec. 5.1).The rate of background events detected without the presence of any source is measured to be 20.4 ± Am source we reduce the activityincident on the camera chip. The lowest detectable rate measured 5 σ above background is 40 ± ± γ -peaks, the minimal detectable rate is 4 ± ± ± ±
33 Bq(Sec. 5.2.2), respectively.Comparing the measured rates and incident source activity allows to determine the intrinsic efficiency ofthe Neo sCMOS camera at the two Am γ -lines. They are found to be 0.08 ± ± ersion September 24, 2020 submitted to Sensors
28 of 31 simulations (Sec. 5.3) show that for thin silicon layers the fraction of an absorbed photon’s energy containedin the silicon decreases with increasing energy of the photon. Thus, lowering the fraction of counts in thephotopeak and the detection efficiency measured at the energy of the photopeak. The increasing volume,in which a photon deposes its energy, for increasing photon energy fits the measurement of larger clustersizes for high energy γ -lines (Sec. 4.2.4). The Geant4 simulations as well as the toy MC simulations indicatea thickness in the order of 2 µm to 4 µm for Neo sCMOS silicon layer. The ratio of the radioactive isotopes in Pb to stable Pb isotopes is required to estimate the capability of theNeo sCMOS to detect lead in drinking water. In [9] the Pb to stable lead ratio is used to monitor thelead intake of plants. They measure 96 ± − for Pb/Pb in rainwater in London, correspondingto a ratio of 34 ppb of
Pb/Pb (molar mass of lead: 207.2 g/mol, half-life of lead: 22.3 yr). However, theorigin of the stable lead and
Pb are not necessarily the same. In [5] several Pb samples of differentage – from ancient lead to recently produced lead – are analysed for their
Pb,
Th and
U contents.They find
Pb/Pb ratios from 0.09 Bq kg − to 68.7 Bq kg − (3.9 × − ppb to 2.4 × − ppb) in theirlead samples.In the following we will estimate the incident activity on the Neo sCMOS, produced by 10 ppb lead inwater. Doing so the Pb/Pb ratio of [9] will be used, since it results from a measurement of Pb in water.However, the results in [5] have to be kept in mind as caveat.With the
Pb/Pb ratio of 34 ppb, the WHO limit of 10 ppb [1] of lead in drinking water translates to afraction of 3.4 × − parts Pb to one part of water. One gram of water contains at this ratio 9.9 × Pb atoms, which initially decay at a rate of 0.96 mBq ( R decay Pb ). A sample of water containing lead couldbe placed on the camera’s window – in 1.75 cm distance from the silicon chip. Considering 1 g of wateras point source, the incident rate ( R incident Pb ) is only 0.001 mHz after taking the geometric acceptance inEqn. (10) into account as well as the fact that the γ -yield for the 45 keV γ is only 4 %: R incident Pb = R decay Pb · (cid:101) G ( ) · (cid:110) Pb γ yield (cid:111) = × − Bq · ± · = ( ± ) × − Bq (16) R incident Pb has to be compared to the measured value of the incident Am source activity of 1.0 ± ±
33 Bq at the 26.3 keV and 59.5 keV lines, respectively (Sec. 5.2.3). With ∼
46 keV the Pb γ -lineis located between these two energies. Thus, the sensitivity of the Neo sCMOS is a factor of 10 to 10 toolow to detect the decay radiation of trace amounts of Pb occurring with 10 ppb lead in 1 g of water. Thefraction is even lower, given that the above calculations assume a point source and 1 g of water measures a1 cm .Independent from the actual Pb/Pb ratio for commercial lead or lead in the water one can estimate thefraction of
Pb per mass ( e.g.
Pb/Pb or
Pb per gram of water) which corresponds to a R incident Pb asthe measured, minimal incident rate. Considering the same parameters as for Eqn. (16) and using the two Am γ -line energies we estimate a sensitivity between 0.3 ± ±
11 ppb
Pb per mass.This sensitivity is by itself not enough to reach the WHO limit, but makes the Neo sCMOS a competitiveradiation detector.Two more points need to be mentioned: the first estimate depends highly on the
Pb/Pb ratio; and,water can be concentrated by boiling it, potentially leaving heavy metals behind. In case of ashing ofplants and acrylic [9,13], it has been found that heavy metals stay behind after the process. Furthermore,
Pb is not the only trace isotope potentially occurring together with stable lead. In case the radio-isotope ersion September 24, 2020 submitted to
Sensors
29 of 31 to stable lead ratio is larger than the
Pb/Pb in [9], the γ yield of these radio-isotope decays is largerthan the one of Pb – the same is true when the water’s volume has been reduced.
7. Outlook and conclusion
Based on the results above, we are currently examining how well
Pb is retained when water isconcentrated by volume reducing it through boiling off. Furthermore, while our results do not allow toconclude that we can measure
Pb in low enough concentrations as needed to detect 10 ppb of lead, ourstudy has shown that a CMOS sensor optimised for optical wavelengths is well suited as γ - and x-raydetector for low energies in the range from ∼
10 keV to ∼
60 keV. Follow up studies will establish theperformance of commercial CMOS sensors and whether they are suitable for radio assay of materials.
Author Contributions:
Conceptualization, A. Deisting and J. Monroe; methodology, A. Deisting and A. Dias and J.Monroe and J. Walding; software, A. Deisting and A. Dias and C. Türko ˘glu; validation, A. Deisting and A. Dias and C.Türko ˘glu; formal analysis, A. Deisting and A. Dias and C. Türko ˘glu; investigation, A. Deisting; resources, J. Monroe;data curation, A. Deisting and A. Dias; writing–original draft preparation, A. Deisting and A. Dias and C. Türko ˘glu;writing–review and editing, A. Aguilar-Arevalo and X. Bertou and C. Canet and M. A. Cruz-Pérez and A. Deistingand A. Dias and J. C. D’Olivo and F. Favela-Pérez and E. A. Garcés and A. González Muñoz and J. O. Guerra-Pulidoand J. Mancera-Alejandrez and D. J. Marín-Lámbarri and M. Martinez Montero and J. Monroe and S. Paling and S. J.M. Peeters and P. Scovell and C. Türko ˘glu and E. Vázquez-Jáuregui and J. Walding; visualization, A. Deisting and A.Dias and C. Türko ˘glu; supervision, J. Monroe and S. J. M. Peeters and J. Walding; project administration, J. Monroeand E. Vázquez-Jáuregui; funding acquisition, X. Bertou and J. Monroe and E. Vázquez-Jáuregui. All authors haveread and agreed to the published version of the manuscript.
Funding:
This research was funded by STFC Global Challenges Research Fund (Foundation Awards, GrantST/R002908/1), by STFC Grant ST/T506382/1, by DGAPA UNAM grants PAPIIT-IT100420 and PAPIIT-IN108020,and by CONACyT grants CB-240666 and A1-S-8960.
Acknowledgments:
The Authors wish to thank Ian Murray, Royal Holloway, University of London for his technicalsupport.
Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; inthe collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish theresults.
References
1. Geneva: World Health Organization. Guidelines for drinking-water quality, 4 th edition, incorporating the 1 st addendum. WHO Library Cataloguing-in-Publication Data . Licence: CC BY-NC-SA 3.0 IGO.2. Pure Earth. Fact Sheet: Lead. 2019. [accessed 04-Feburary-2019].3. Caravanos, J.; Dowling, R.; Téllez-Rojo, M.M.; Cantoral, A.; Kobrosly, R.; Estrada, D.; Orjuela, M.; Gualtero, S.;Ericson, B.; Rivera, A.; Fuller, R. Blood Lead Levels in Mexico and Pediatric Burden of Disease Implications.
Annals of Global Health , , 269 – 277. doi:https://doi.org/10.1016/j.aogh.2014.08.002.4. Drinking water and health: a guide for public and environmental health professionals and for those in thewater industry in Northern Ireland. 2020. [accessed 14-April-2020].5. Orrell, J.L.; others. Assay methods for U, Th, and
Pb in lead and calibration of
Bi bremsstrahlungemission from lead.
J. Radioanal. Nucl. Chem. , , 1271–1281, [arXiv:physics.ins-det/1512.06494].doi:10.1007/s10967-016-4732-6.6. Pérez, M.; Lipovetzky, J.; Haro, M.S.; Sidelnik, I.; Blostein, J.J.; Bessia, F.A.; Berisso, M.G. Particle detectionand classification using commercial off the shelf CMOS image sensors. Nuclear Instruments and Methods inPhysics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment , , 171 – 180.doi:https://doi.org/10.1016/j.nima.2016.04.072.7. Servoli, L.; Biagetti, D.; Passeri, D.; Gattuso, E.S. Characterization of standard CMOS pixel imagers as ionizingradiation detectors. Journal of Instrumentation , , P07003–P07003. doi:10.1088/1748-0221/5/07/p07003. ersion September 24, 2020 submitted to Sensors
30 of 31
8. Appleby, P. Three decades of dating recent sediments by fallout radionuclides: a review.
The Holocene , , 83–93. doi:10.1177/0959683607085598.9. Yanga, H.; Appleby, P.G. Use of lead-210 as a novel tracer for lead (Pb) sources in plants. Scientific reports , , 21707. doi:https://dx.doi.org/10.1038/srep21707.10. Amaudruz, P.A.; Baldwin, M.; Batygov, M.; Beltran, B.; Bina, C.; Bishop, D.; Bonatt, J.; Boorman, G.; Boulay,M.; Broerman, B.; Bromwich, T.; Bueno, J.; Burghardt, P.; Butcher, A.; Cai, B.; Chan, S.; Chen, M.; Chouinard,R.; Churchwell, S.; Cleveland, B.; Cranshaw, D.; Dering, K.; DiGioseffo, J.; Dittmeier, S.; Duncan, F.; Dunford,M.; Erlandson, A.; Fatemighomi, N.; Florian, S.; Flower, A.; Ford, R.; Gagnon, R.; Giampa, P.; Golovko, V.;Gorel, P.; Gornea, R.; Grace, E.; Graham, K.; Grant, D.; Gulyev, E.; Hall, A.; Hallin, A.; Hamstra, M.; Harvey, P.;Hearns, C.; Jillings, C.; Kamaev, O.; Kemp, A.; Ku´zniak, M.; Langrock, S.; Zia, F.L.; Lehnert, B.; Li, O.; Lidgard,J.; Liimatainen, P.; Lim, C.; Lindner, T.; Linn, Y.; Liu, S.; Majewski, P.; Mathew, R.; McDonald, A.; McElroy,T.; McFarlane, K.; McGinn, T.; McLaughlin, J.; Mead, S.; Mehdiyev, R.; Mielnichuk, C.; Monroe, J.; Muir, A.;Nadeau, P.; Nantais, C.; Ng, C.; Noble, A.; O’Dwyer, E.; Ohlmann, C.; Olchanski, K.; Olsen, K.; Ouellet, C.;Pasuthip, P.; Peeters, S.; Pollmann, T.; Rand, E.; Rau, W.; Rethmeier, C.; Retière, F.; Seeburn, N.; Shaw, B.;Singhrao, K.; Skensved, P.; Smith, B.; Smith, N.; Sonley, T.; Soukup, J.; Stainforth, R.; Stone, C.; Strickland, V.;Sur, B.; Tang, J.; Taylor, J.; Veloce, L.; Vázquez-Jáuregui, E.; Walding, J.; Ward, M.; Westerdale, S.; White, R.;Woolsey, E.; Zielinski, J. Design and construction of the DEAP-3600 dark matter detector. Astroparticle Physics , , 1 – 23. doi:https://doi.org/10.1016/j.astropartphys.2018.09.006.11. Nantais, C.M. Radiopurity measurement of acrylic for the DEAP-3600 dark matter experiment. Master’s thesis,Queen’s University, Kingston, Ontario, Canada, 2014. Master Thesis.12. Widorski, M. Development of a Detection System to Measure Low Levels of 210Pb using β / γ Coincidence.Master’s thesis, Department of Physics, Royal Holloway University of London, Egham (UK), 2013. Masterthesis.13. Milton, G.; Kramer, S.; Deal, R.; Earle, E. Ultra trace analysis of acrylic for 232Th and 238U daughters.
AppliedRadiation and Isotopes , , 539 – 547. doi:https://doi.org/10.1016/0969-8043(94)90195-3.14. The Virginia Tech Research Team. http://flintwaterstudy.org, 2017. [accessed 11-September-2020].15. Oxford instruments. Neo 5.5 sCMOS Specifications. 2020. [accessed 16-March-2020].16. Agostinelli, S.; Allison, J.; Amako, K.; et al.. Geant4—a simulation toolkit. Nuclear Instruments and Methods inPhysics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment , , 250 – 303.doi:https://doi.org/10.1016/S0168-9002(03)01368-8.17. Aguilar-Arevalo, A.; others. Search for low-mass WIMPs in a 0.6 kg day exposure of the DAMIC experiment atSNOLAB. Phys. Rev. D , , 082006, [arXiv:astro-ph.CO/1607.07410]. doi:10.1103/PhysRevD.94.082006.18. LD Didactic GmbH. Instruction sheet 554 800. 2020. [accessed 24-April-2020].19. The FITS Support Office at NASA/GSFC. Flexible Image Transport System. 2020. [accessed 20-March-2020].20. Brun, R.; Rademakers, F. ROOT: An object oriented data analysis framework. Nucl. Instrum. Meth. A , , 81–86. doi:10.1016/S0168-9002(97)00048-X.21. Chong, C.S.; Ibrahim, S.E.; Ahmad, S.M.K.; Abdul, A.T. Gamma ray spectrum of Am 241 in a backscatteringgeometry using a high purity germanium detector, 1997.22. International Atomic Energy Agency – Nuclear data services. Live Chart of Nuclides. 2020. [accessed26-March-2020].23. Demir, D.; Ero ˘glu, M.; Tur¸sucu, A. Studying of characteristics of the HPGe detector for radioactivitymeasurements. Journal of Instrumentation , , P10027–P10027. doi:10.1088/1748-0221/8/10/p10027.24. NIST, Physical Measurement Laboratory. Stopping-Power & Range Tables for Electrons, Protons, and HeliumIons. 2019. [accessed 20-January-2019].25. Hubbell, J.H.; Seltzer, S.M. Tables of X-Ray Mass Attenuation Coefficients and Mass Energy-AbsorptionCoefficients 1 keV to 20 MeV for Elements Z = 1 to 92 and 48 Additional Substances of Dosimetric Interest. NISTIR 5632 . ersion September 24, 2020 submitted to Sensors
31 of 31
26. Berger, M.; Hubbell, J.; Seltzer, S.; Chang, J.; Coursey, J.; Sukumar, R.; Zucker, D.; .; Olsen,K. XCOM: Photon Cross Sections Database.
NIST Standard Reference Database 8 (XGAM) .doi:https://dx.doi.org/10.18434/T48G6X.27. Berger, M.; Coursey, J.; Zucker, M.; Chang, J. Stopping-Power & Range Tables for Electrons, Protons, andHelium Ions.
NIST Standard Reference Database 124 . doi:https://dx.doi.org/10.18434/T4NC7P.28. Druit, G. Dark Matter Time Projection Chamber (DMTPC) Optical TPC for low energy particle tracking, 2013.29. Kolanoski, H.; Wermes, N.
Teilchendetektoren ; Springer, 2016.30. Gow, R.D.; Renshaw, D.; Findlater, K.; Grant, L.; McLeod, S.J.; Hart, J.; Nicol, R.L. A Comprehensive Tool forModeling CMOS Image-Sensor-Noise Performance.