\textbf{Micro} \textbf{Me}sh \textbf{Ga}seous \textbf{S}tructures (Micromegas) are a kind of \textbf{M}icro\textbf{p}attern \textbf{G}aseous \textbf{D}etectors (MPGDs) first introduced in 1996 (Giomataris et al. 1996; Giomataris 1998). The modern Micromegas is the Microbulk Micromegas (Andriamonje et al. 2010).

Interestingly, the name Micromegas is based on the novella Micromégas by Voltaire published in 1752 (Voltaire 1752), an early example of a science fiction story. (Giomataris et al. 1996)

These detectors are – as the name implies – gaseous detectors containing a 'micro mesh'. In the most basic form they are a made of a closed detector volume that is filled with a suitable gas, allowing for ionization (often argon based gas mixtures are used; xenon based detectors for axion helioscopes are in the prototype phase). The volume is split into two different sections, a large drift volume typically \(\mathcal{O}(\text{few }\si{cm})\) and an amplification region, sized \(\mathcal{O}(\SIrange{50}{100}{μm})\). At the top of the volume is a cathode to apply an electric field. Below the mesh is the readout area at the bottom of the volume. In standard Micromegas detectors, strips or pads are used as a readout. The electric field in the drift region is strong enough to avoid recombination of the created electron-ion pairs and to provide reasonably fast drift velocities \(\mathcal{O}(\si{cm.μs⁻¹})\). The amplification gap on the other hand is precisely used to multiply the primary electrons using an avalanche effect. Thus, the electric field reaches values of \(\mathcal{O}(\SI{50}{kV.cm⁻¹})\). These drift and amplification volumes are achieved by an electric field between a cathode and the mesh as well as the mesh and the readout area.

Fig. 1 shows a schematic for the general working principle of such detectors

The Timepix ASIC (Application Specific Integrated Circuit) is a \(256 × 256\) pixel ASIC with each pixel \(\num{55}·\SI{55}{μm^2}\) in size. It is based on the Medipix ASIC, developed for medical imaging applications by the Medipix Collaboration (Llopart et al. 2002). The pixels are distributed over an active area of \(\num{1.41}\times\SI{1.41}{\cm^2}\). Each pixel contains a charge sensitive amplifier, a single threshold discriminator and a \(\SI{14}{bit}\) pseudo random counting logic. It requires use of an external clock, in the range of \(\SIrange{10}{150}{MHz}\), with \(\SI{40}{MHz}\) being the clock frequency used for the applications in this thesis. (Llopart Cudie 2007; Llopart et al. 2007; Llopart and Poikela 2006) A good overview of the Timepix is also given in (Lupberger 2016). A picture of a Timepix ASIC is shown in fig. 2(a).

The Timepix uses a shutter based readout, either with a fixed shutter time or using an external trigger to close a frame. After the shutter is closed in the Timepix, the readout is performed during which the detector is insensitive. Each pixel further can work in four different modes:

hit counting mode / single hit mode

simply counts the number of times the threshold of a pixel was crossed (or whether a pixel was activated once in single hit mode).

\textbf{T}ime \textbf{o}ver \textbf{T}hreshold (ToT)

In the ToT mode the counter of a pixel will count the number of clock cycles that the charge on the pixel exceeds the set threshold, which is set by an \(\SI{8}{bit}\) \textbf{D}igital to \textbf{A}nalog \textbf{C}onverter (DAC) while the shutter is open. ToT is equivalent to the collected charge of each pixel.

\textbf{T}ime \textbf{o}f \textbf{A}rrival (ToA)

The ToA mode records the number of clock cycles from the first time the pixel's threshold is exceeded to the end of the shutter window. Thus, it allows to calculate the time at which the pixel first crossed the threshold.

In the context of this thesis only the ToT mode was used.

First experiments of combining a Micromegas with a Timepix readout were done in 2005 (Campbell et al. 2005) by using classical approaches to place a micromesh on top of the Timepix, at the time still called TimePixGrid. While this worked in principle, it showed a Moiré pattern, due to slight misalignment between the holes of the micromesh and the pixels of the Timepix. Shortly after, an approach based on photolithographic post-processing was developed to perfectly align the Timepix pixels each with a hole of a micromesh (Chefdeville et al. 2006), called the InGrid (integrated grid). The commonly used name for a gaseous detector using an InGrid is GridPix. For an overview of the process to produce InGrids nowadays, see (Scharenberg 2019).

The InGrid consists of a \(\SI{1}{μm}\) thick aluminum grid, resting on small pillars \(\SI{50}{μm}\) above the Timepix. A silicon-nitride \(\ce{Si_x N_y}\) ² layer protects the Timepix from direct exposure to the amplification processes. The main advantage over previous Micromegas technologies of the GridPix is its ability to detect single electrons. As long as the diffusion distance is long enough to avoid multiple electrons entering a single hole of the InGrid, each primary electron produced during the initial ionization event is recorded. Fig. 2(b) shows an image of such an InGrid.

In the course of (Christoph Krieger 2018) a first GridPix based detector for usage at an axion helioscope, CAST, was developed. While the main result was on the coupling constant of the chameleon particle, an axion-electron coupling result was computed in (Schmidt 2016).

The detector consists of a single GridPix in a \(\SI{78}{mm}\) diameter gas volume and a drift distance of \(\SI{3}{cm}\). The detector has a \(\SI{2}{μm}\) thick Mylar (\(\ce{C10 H8 O4}\)) entrance window for X-rays. This detector serves as the foundation the detector used in the course of this thesis was built on. See fig. 3 for an exploded schematic of the detector. Further, fig. 4 shows the achieved background rate of this detector in the center \(\num{5} \times \SI{5}{mm^2}\) region of the detector. The background rate shows the copper Kα line near \(\SI{8}{keV}\), possibly overlaid with a muon contribution as well as the expected argon Kα lines at \(\SI{3}{keV}\). Below \(\SI{2}{keV}\) the background rises the lower the energy becomes, likely due to background- and signal-like events being less geometrically different at low energies (fewer pixels). The average background rate in the range from \(\SIrange{0}{8}{keV}\) is \(\sim\SI{2.9e-05}{keV^{-1}.cm^{-2}.s^{-1}}\).

Generally, the detector follows the same design as the old detector shown in sec. 7.4, mainly so that mounting it inside of the lead shielding and to the vacuum pipes at CAST is possible without significant changes. An exploded view of the full detector can be seen in fig. 5.

At the center of the new detector is the 'septemboard', 7 GridPixes replace the single GridPix on the carrier board, sec. 7.10. Analogue signals induced by the amplified charges on the center GridPix are now read out using a flash ADC (FADC), sec. 7.8. The housing with an inner diameter of \(\SI{78}{mm}\) is again made of acrylic glass, same as in the old detector. The detector entrance window is replaced by a \(\SI{300}{nm}\) \ccsini window (sec. 7.9), which also acts as part of the detector cathode. The copper anode slots in right above the septemboard. The carrier board sits on the intermediate board. Below the intermediate board is a bespoke water cooling made of oxygen-free copper to cool the heat emitted by the additional 6 GridPixes, sec. 7.11. On the underside of the intermediate board is a new small silicone photomultiplier (SiPM), sec. 7.7. Finally, a large veto scintillator is installed above the detector setup at CAST, also sec. 7.7.

During developments multiple septemboards were built and tested. The septemboard used in the final detector is septemboard 'H'. The GridPixes of the final board are listed in tab. 3.

The detector is operated by a Xilinx Virtex-6 \textbf{F}ield \textbf{P}rogrammable \textbf{G}ate \textbf{A}rray (FPGA) in the form of a Virtex-6 ML605 evaluation board. ³ It is connected to the intermediate board via two \textbf{H}igh-\textbf{D}efinition \textbf{M}ultimedia \textbf{I}nterface (HDMI) cables. The Virtex-6 contains the firmware controlling the Timepix ASICs and correlating the scintillator and FADC signals (see appendix sec. 17.1), the Timepix Operating Firmware (TOF). The high voltage (HV) supply both for the septemboard as well as for the scintillators sit inside a VME crate, which also houses the FADC. A USB connection is used to read out and control the FADC and HV supply via the computer running the data acquisition and control software (see sec. 17.2), the Timepix Operating Software (TOS). A schematic of this setup is shown in fig. 6, which leaves out the SiPM and temperature readout.

The first general improvement is the addition of two scintillators for veto purposes. While both have slightly different goals, each is there to help with the removal of muon signals or muon induced events (for example X-ray fluorescence) in the detector. Given that cosmic muons (ref. section 6.2) dominate the background by flux, statistically there is a high chance of muons creating X-ray like signatures in the detector. By tagging muons before they interact near the detector, these can be correlated with events seen on the GridPix and thus possibly be vetoed if precise time information is available.

The first scintillator is a large 'paddle' installed above the detector and beamline, aiming to tag a large fraction of cosmic muons traversing in the area around the detector. It has a Canberra 2007 base and the photomultiplier tube (PMT) is a Bicron Corp. 31.49x15.74M2BC408/2-X (first two numbers: dimensions in inches). The full outer dimensions of the scintillator paddle are closer to \(\SI{42}{cm} \times \SI{82}{cm}\). It is the same scintillator paddle used during the Micromegas data taking behind the LLNL telescope prior and after the data taking campaign with the detector described in this thesis.

For this scintillator, muons which traverse through it and the gaseous detector volume are not the main use case. They can be easily identified by the geometric properties of the induced tracks (their zenith angles are relatively small, resulting in track like signatures as the GridPix readout is orthogonal to the zenith angle). There is a small chance however that a muon can ionize an atom of the detector material, which may emit an X-ray upon recombination. One particular source of background can be attributed to the presence of copper whose Kα lines are at \(\sim\SI{8.04}{\keV}\) as well as fluorescence of the argon gas with its Kα lines at \(\sim\SI{2.95}{keV}\) (see. table 1 in sec. 6.1.4 and sec. 6.1.4).

Fig. 7(a) shows a schematic of a side view of the detector chamber with the scintillator paddle on top. When a muon traverses the scintillator, a counter \(t_{\text{Veto}}\) starts on the FPGA. Two different cases are shown. In the extreme, a muon may traverse close to the cathode or close to the anode / readout plane. This changes the time of the total drift time and therefore the time difference between the trigger time \(t_{\text{Veto}}\) and the readout time, which in the readout is precisely the value of the counter on the FPGA. The drift velocity of \(\sim\SI{2}{cm.μs⁻¹}\) and height of the detector chamber (\(\SI{3}{cm}\)) therefore allow to set an upper limit on the maximum time between a veto paddle trigger and the GridPix readout of about \(\SI{1.5}{μs}\). At a clock speed of \(\SI{40}{MHz}\) this corresponds to \(\SI{60}{clock\;cycles}\), a number we will later try to see in the data (see sec. 12.5.1). As the location at which the muon traverses through the detector is random and homogeneous throughout the detection volume, we expect to see a flat distribution up to the maximum possible time and then a sharp drop (equivalent to muons at the cathode).

The second scintillator is a small silicon photomultiplier (SiPM) installed on the underside of the PCB on which the septemboard is installed. This scintillator was calibrated and set up as part of (Schmitz 2017). We are interested in tagging precisely those muons, which enter the detector orthogonally to the readout plane. This implies zenith angles of almost \(\SI{90}{°}\) such that the elongation in the transverse direction of the muon track is small enough to result in a small eccentricity. From the Bethe equation the mean energy loss of muons in the used gas mixture is about \(\SI{8}{\keV}\) along the \(\SI{3}{\cm}\) of drift volume in the detector (see fig. 11 for the energy loss). This coincides with the copper Kα lines and should lead to another source of background in this energy range. Although the muon background will have a much wider energy distribution than the copper lines, which are dominated by the energy resolution of the detector. In a similar manner to the veto paddle we can make an estimate on the typical time scales associated from the time of the scintillator trigger to the GridPix detection. A muon that traverses orthogonally trough the detector can be taken to leave an instant ionization track and trigger the SiPM at the same time (\(\mathcal{O}(\SI{100}{ps}) \ll \SI{25}{ns}\) for one clock cycle). As such, the relevant time scale is the drift time until enough charge has drifted towards the grid as to pass the activation threshold.

Ionization of a muon is a statistical process, as indicated in fig. 7(b). Depending on the density of the charge cloud for muons orthogonal to readout plane, time to accumulate enough charge to trigger FADC differs. With an average energy deposition of a muon in argon gas of \(\sim\SI{2.67}{keV.cm^{-1}}\), and drift velocity again of \(\SI{2}{cm.μs⁻¹}\) the accumulation time can be estimated. For example assuming an FADC activation threshold of \(\SI{1.5}{keV}\) the necessary charge is accumulated on \(\SI{0.56}{cm}\), which takes about \(\SI{280}{ns} \approx \SI{11}{clock.cycles}\) to accumulate. Different tracks will have deposited different amounts of energy. Therefore, we expect a peak at relatively low clock cycles with a tail up to the same \(\SI{60}{clock.cycles}\) (in case the full \(\SI{3}{cm}\) track needs to be accumulated to activate the FADC).

As the Timepix is read out in a shutter based fashion and typical shutter lengths for low rate experiments are long compared to the rate of cosmic muons, the scintillators introduced in previous section require an external trigger to close the Timepix shutter early if a signal is measured on the Timepix. This is one the main purposes of the \text{f}lash \textbf{a}nalog to \textbf{d}igital \textbf{c}onverter (FADC) that is part of the detector. This is done by decoupling the induced analogue signals from the grid of the center GridPix.

The specific FADC used for the detector is a Caen V1792a. It runs at an internal \(\SI{50}{MHz}\) or \(\SI{100}{MHz}\) clock and utilizes virtual frequency multiplication to achieve sampling rates of \(\SI{1}{GHz}\) or \(\SI{2}{GHz}\), respectively. It has 4 channels, each with a cyclic register of \(\num{2560}\) channels. At an operating clock frequency of \(\SI{1}{GHz}\) that means each channel covers the last \(\sim\SI{2.5}{\micro\second}\) at any time. (CAEN 2010)

The raw signal decoupled from the grid is first fed into an Ortec 142 B pre-amplifier and then feeds into an Ortec 474 shaping amplifier, which integrates and shapes the signal as well as amplifies it. For a detailed introduction to this FADC system, see the thesis of A. Deisting (Deisting 2014) and (Schmidt 2016) for further work integrating it into this detector. In addition see the FADC manual (CAEN 2010) ⁴ for a deep explanation of the working principle of this FADC.

The analogue signal of the center grid is decoupled via a small \(C_{\text{dec}} = \SI{10}{nF}\) capacitor in parallel to the high voltage line. For a schematic of the circuit see fig. 14. When a primary electron traverses through a hole in the grid and is amplified, the back flowing ions induce a small voltage spike on top of the constant high voltage applied to the grid. The parallel capacitor filters out the constant high voltage and only transmits the time varying induced signals. Such signals – the envelope of possibly many primary electrons – are measured by the FADC.

This signal can be used for three distinct things:

1. it may be used as a trigger to close the shutter of the ongoing event. Ideally, we want to only measure a single physical event within one shutter window. A long shutter time can statistically result in multiple events happening, which the FADC trigger helps to alleviate. This allows us to reduce the number of events with multiple physical events and acts as a trigger for the scintillators. This in turn means possible muon induced X-ray fluorescence can be vetoed.
2. By nature of the signal production and drift properties of the primary electrons before they reach the grid, the signal shape can theoretically be used to determine a rough longitudinal shape of the event. The length of the FADC event should be proportional to the size of the primary electron cloud distribution along the 'vertical' detector axis. This potentially allows to differentiate between a muon traversing orthogonally through the readout plane and an X-ray due to their longitudinal shape difference, see sec. 12.5.2.
3. Finally, it provides an independent measure of the collected charge on the center chip. This will prove useful in understanding the detector behavior over time later in sec. 11.2.1.

The working principle of how the FADC and the scintillators can be used together to remove certain types of background, by correlating events in the scintillators, the FADC and the GridPix, is shown in fig. 15.

Next up, a major limitation of the previous detector was its limited combined efficiency below \(\SI{2}{keV}\), due to its \(\SI{2}{μm}\) Mylar window. Therefore, the next improvement for the new detector is an ultra-thin silicon nitride \(\ce{Si_3 N_4}\) window of \(\SI{300}{nm}\) thickness and \(\SI{14}{mm}\) diameter, developed by Norcada ⁵. A strongback support structure consisting of 4 lines of \(\SI{200}{μm}\) thick and \(\SI{500}{μm}\) wide \(\ce{Si_3 N_4}\), helps to support a pressure difference of up to \(\SI{1.5}{bar}\). On the outer side a \(\SI{20}{nm}\) thin layer of aluminum is coated to allow the window to be part of the detector cathode. The strongback occludes about \(\SI{17}{\%}\) of the full window area. In reality it is slightly more, as the strongbacks become somewhat wider towards the edges. In the centermost region they are straight and in the center \(\num{5} \times \SI{5}{mm²}\) area, they occlude \(\SI{22.2}{\%}\).

Fig. 16(a) shows the idealized strongback structure without a widening towards the edges of the window. Fig. 16(b) shows an image of one such window under testing conditions in the laboratory, as it withstands a pressure difference of \(\SI{1.5}{bar}\).

As the main purpose is the increase of transmission at low energies, fig. 17 shows the transmission of the mylar window of the old detector and the new \(\ce{Si_3 N_4}\) window in the energy range below \(\SI{3}{keV}\). The \(\ce{Si_3 N_4}\) window shows a significant increase in transmission below \(\SI{2}{keV}\), which is very important for the sensitivity in solar axion-electron and chameleon searches, which both peak near \(\SI{1}{keV}\) in their solar flux. The window alone significantly increases the signal to noise ratio of these physics searches.

The main motivation for extending the readout area from a single chip to a 7 chip readout is to reduce background towards the outer sides of the chip, in particular in the corners. Against common intuition however, it also plays a role for events which have cluster centers near the center of the readout. The latter is, because diffusion can produce quite large clusters even at low energies. In particular in lower energy events, tracks may have gaps in them large enough to avoid being detected as a single cluster for standard radii in cluster searching algorithms. This is particularly of interest as different searches produce an 'image' at different positions and sizes on the detector. While the center chip is large enough to fully cover the image for essentially all models, it may not be in the regions of lowest background. Hence, improvements to larger areas are needed.

The septemboard is implemented in such a way to optimize the loss of active area due to bonding requirements and general manufacturing realities. As the Timepix ASIC is a \(\SI{16.1}{mm}\) by \(\SI{14.1}{mm}\) large chip (the bonding area adding \(\SI{2}{mm}\) on one side), the upper two rows are installed such that they are inverted to another. The bonding area is above the upper row and below the center row. The bottom row again has its bonding area below. This way the top two rows are as close together as realistically possible, with a decent gap on the order of \(\SI{2}{mm}\) between the middle and bottom row. Any gap is potentially problematic as it implies loss of signal in that area, complicating the possible reconstruction methods. The layout can be seen in fig. 20 in the next section.

All 7 GridPix are connected in a daisy-chained way. This means that in particular for data readout, all chips are read out in serial order. The dead time for readouts therefore is approximately 7 times the readout time of a single Timepix. A single Timepix has a readout time of \(\sim\SI{25}{ms}\) at a clock frequency of \(\SI{40}{MHz}\) (the frequency used for this detector). This leads to an expected readout time of the full septemboard of \(\SI{175}{ms}\). ⁶ Such a long readout time leads to a strong restriction of the possible applications for such a detector. Fortunately, for the use cases in a very low rate experiment such as CAST, long shutter times are possible, mitigating the effect on the fractional dead time to a large extent.

Fig. 18 shows a heatmap of all cluster centers during roughly \(\SI{2000}{h}\) of background data after passing these clusters through a likelihood based cut method aiming to filter out non X-ray like clusters (details of this follow later in sec. 12.1). It is clearly visible that the further a cluster center is towards the chip edges, and especially the corners, the more likely it is to be considered an X-ray like cluster. This has an easy geometric explanation. Consider a perfect track traversing over the whole chip. In this case it is very eccentric. Move the same track such that its center is in one of the corners and rotate it by \(\SI{45}{°}\) and suddenly the majority of the track won't be detected on the chip anymore. Instead something roughly circular remains visible, 'fooling' the likelihood method. For a schematic illustrating this, see fig. 19.

The septemboard therefore is expected to significantly reduce the background over the whole center chip, with the biggest effect in the regions with the most amount of background.

During development of the septemboard one particular set of problems manifested. While testing a prototype board with 5 active GridPix in a gaseous detector, the readout was plagued by excessive noise problems. The detector exhibited a large number of frames with more than \(\num{4096}\) active pixels (the limit for a zero suppressed readout) and common pixel values of \(\num{11810}\) indicating overrun ToT counters. On an occupancy (sum of all active pixels) of the individual chips, it is quite visible the data is clearly not due to cosmic background. Fig. 20 shows such an occupancy with the color scale topping out at the \(80^{\text{th}}\) percentile of the counts for each chip individually. The chip in the bottom left shows a large number of sparks (overlapping half ellipses pointing downwards) at the top end. Especially the center chip in the top row shows highly structured activity, which is in contrast to the expectation of a homogeneous occupancy for a normal background run. In addition, on all chips some level of general noise on certain pixels is visible (some being clearly more active than others resulting in a scatter of 'points').

The intermediate board and carrier board used during these tests were the first boards equipped with two PT1000 temperature sensors. One on the bottom side of the carrier board and another on the intermediate board. Each is read out using a MAX31685 micro controllers. Both of which are communicated with via a MCP2210 USB-to-SPI micro controllers over a single USB port on the intermediate board. The single MCP2210 communicates with both temperature sensors via the Serial Peripheral Interface (SPI) (see sec. 17.2.4 for more information about the temperature logging and readout). In the run shown in fig. 20 the temperature sensors were not functional yet, as the readout software was not written. The required logic was added to the Timepix Operating Software (TOS), the readout software of the detector, motivated by this noise activity to monitor the temperature before and during a data taking period. The temperature on the carrier board indicated temperatures of \(\sim\SI{75}{\celsius}\) in background runs similar to the one of fig. 20. One way to get a measure for the noise-like activity seen on the detector is to look at the rate of active pixels over time. With values well above numbers expected due to background, excess temperature seemed a possible cause for the issues. As no proper cooling mechanism was available, a regular desk fan was placed pointing at the detector when it was run without any kind of shielding. This saw the temperature under the carrier board drop from \(\SI{76}{\celsius}\) down to \(\SI{68}{\celsius}\). As a result the majority of noise disappeared as can be seen in fig. 21(b) with the temperature curve during the full run in fig. 21(a). ^{7 , 8}

The features visible in the occupancy plots are thus likely multiple different artifacts due to too high temperatures. A mixture of real sparks (bottom left chip in fig. 20) and possible instabilities that possibly affect voltages for the pixels (and thus change the thresholds of each pixel). As the temperature is measured on the bottom side of the carrier board, temperatures in the amplification region are likely higher.

Following this a bespoke water cooling was designed by T. Schiffer, made from oxygen-free copper with \(\SI{3}{mm}\) channels for water to circulate through the copper body. (Schiffer 2024) The body has the same diameter as the intermediate board and is installed right below. The water circulation is handled by an off-the-shelf pump and radiator from Alphacool ⁹ intended for water cooling setups for desktop computers. The pump manages a water flow rate of about \(\SI{0.3}{\liter\per\minute}\) through the \(\SI{3}{mm}\) channels in the copper. In common operation the temperatures on the carrier board are between \(\SIrange{45}{50}{\celsius}\) and noise free operation is possible.

For the applications at CAST, the detector is filled with \(\ce{Ar}\) / \(\ce{iC_4 H_{10}}\) : \(\SI{97.7}{\%} / \SI{2.3}{\%}\) gas. Combined with its \SI{300}{nm} \(\ce{Si_3 N_4}\) window, the combined detection efficiency can be computed, if the \(\SI{20}{nm}\) \(\ce{Al}\) coating for the detector cathode is included by computing the product of the different efficiencies. The efficiency of the window and coating are the transmissions of X-rays at different energies for each material \(t_i\). For the gas, the absorption probability of the gas \(a_i\) is needed. As such

\[ ε_{\text{tot}} = t_{\ce{Si_3 N_4}} · t_{\ce{Al}} · a_{\ce{Ar} / \ce{iC_4H_{10}}} \]

describes the full detector efficiency assuming the parts of the detector, which are not obstructed by the window strongbacks. For a statistical measure of detection efficiency the occlusion of the window needs to be taken into account. Because it is position (and thus area) dependent, the need to include it is decided on a case by case basis. For the absorption of a gas mixture, we can use Dalton's law and compute the absorption of the individual gases according to their mole fractions (their percentage as indicated by the gas mixture) and then compute it for each partial pressure

\[ a_i = \text{Absorption}(P_{\text{total}} · f_i) \]

where \(P_{\text{total}}\) is the total pressure of the gas mixture (in this case \(\SI{1050}{mbar}\)) and \(f_i\) is the fraction of the gas \(i\). 'Absorption' simply refers to the generic function computing the absorption for a gas at a given pressure (see sec. 6.3.1 and sec. 6.1.1).

The full combined efficiency as presented here is shown in fig. 23. Different aspects dominate the combined efficiency (purple line) in different energy ranges. At energies above \(\SI{5}{keV}\) the probability of X-rays to not generate a photoelectron within the \(\SI{3}{cm}\) of drift distance becomes the major factor for a loss in efficiency. This means the combined efficiency at \(\SI{10}{keV}\) is slightly below \(\SI{30}{\%}\). The best combined efficiency of about \(\SI{95}{\%}\) is reached at about \(\SI{3.75}{keV}\) where both the absorption is likely and the energy is high enough to transmit well through the window. The argon \(K 1s\) absorption edge is clearly visible at around \(\SI{3.2}{keV}\). At energies below the mean free path of X-rays is significantly longer as the \(K 1s\) absorption is a significant factor in the possible generation of a photoelectron. The window leads to a similar, but inverse, effect namely due to the \(K 1s\) line of \(\ce{Si}\) at around \(\SI{1.84}{keV}\). Because transmission is desired through the window material, the efficiency increases once we go below that energy. Finally, the nitrogen \(K 1s\) line also contributes to an increase in efficiency once we cross below about \(\SI{400}{eV}\). The average efficiencies in the energy ranges between \(\SIrange{0}{3}{keV}\) and \(\SIrange{0}{10}{keV}\) are \(\SI{73.42}{\%}\) and \(\SI{67.84}{\%}\), respectively.

The improvement in efficiency at energies below \(\SI{3}{keV}\) in comparison to the mylar window used in the 2014/15 detector (see sec. 7.9) leads to a significant improvement in possible signal detection at those energies, which is especially important for searches with peak fluxes around \(\SIrange{1}{2}{keV}\) as is the case for the axion-electron coupling or a possible chameleon coupling.

The data acquisition software used for the Septemboard detector, the Timepix Operating Software (TOS), is not of direct importance for the thesis. But a longer section about it can be found in appendix 17. It includes discussions about how the software works internally, how it is used, what the data format produced looks like and its configuration files. Further, it goes over the temperature readout functionality and finally presents the software used for detector monitoring, in the form of an event display used at CAST.