Insights into post-growth doping and proposals for CdTe:In photovoltaic devices

This paper is motivated by the potential advantages of higher doping and lower contact barriers in CdTe photovoltaic devices that may be realized by using n-type rather than the conventional p-type solar absorber layers. We present post-growth doping trials for indium in thin polycrystalline CdTe films using the diffusion of indium metal with indium chloride. Chemical concentrations of indium up to 1019 cm−3 were achieved and the films were verified as n-type by hard x-ray photoemission. Post-growth chlorine treatment (or InCl3) was found to compensate the n-doping. Trial structures comprising CdS/CdTe:In verified that the doped absorber structures performed as expected both before and after chloride treatment, but it is recognized that this is not an optimum combination. Hence, in order to identify how the advantages of n-type absorbers might be fully realized in future work, we also report simulations of a range of p–n junction combinations with n-CdTe, a number of which have the potential for high V oc.


Introduction
For the last 50 years in the development of p-n junction thin film photovoltaics, the device designs have been dominated by those with p-type absorbers, this being the case for both the commercially successful materials-CdTe and CIGS-and also for emerging materials such as CZTS. In this paper, we focus on an alternative device concept featuring n-type CdTe absorber layers. Devices with n-type absorbers have the potential for high performance arising from favorable material properties. Although we focus on CdTe, the concepts are likely to be translatable to other established and emerging materials systems.
The motivation for our interest in n-type absorbers is illustrated by the rationale for the conventional choice of p-type absorbers in thin film photovoltaics. The selection of p-type is essentially a default choice for the absorber since the window and transparent electrodes must comprise wide-gap semiconductors, and they can only be effectively doped n-type [1][2][3]. This demands that the absorber should be p-type to form a p-n junction with them. This combination is achievable in practice since semiconductors having the mid-range gaps required for solar absorbers can be doped p-type to complete the junction [1]. Therefore, the earliest CdTe solar cells were built in the nominal p-n configuration comprising glass/n-TCO/n-CdS/ p-CdTe/contact. Industrially, the n-type transparent conducting oxide (TCO) is SnO 2 :F (FTO) [4][5][6]. This device design reached a high point of ∼16% PCE [7,8] while the structures that evolved from it (i.e. omitting the CdS and including Cd(Se,Te) in the absorber) have reached 21% [9,10]. All these devices have p-type absorbers. The p-doping has most often been achieved by copper [11][12][13][14][15][16][17][18][19] but alternatives have been explored, including phosphorous [20] and more recently arsenic [21,22], culminating in CdTe:As devices having PCEs > 20% [23].
Despite these successes, the use of p-type CdTe absorber layers present their own challenges related to doping and contacting. There is the potential to sidestep these by using n-type CdTe as follows: Firstly, doping-the efficiency of CdTe devices is limited by their open circuit voltage (V oc ), which is directly related to the achievable doping level [24]. While there is an upper limit for the stable n-or p-type doping density for any semiconductor [3], that limit is lower for p-than it is for n-CdTe. For example, the experimental data from Burst et al [25,26] for CdTe:P shows a compensation crash at p > 10 17 cm −3 . (Marfaing [27] gives a detailed analysis of the phenomena.) On the other hand, stable n-type doping up to n = 10 18 cm −3 has been demonstrated using indium [28], this being close to the degenerate doping density and comfortably exceeding the levels required in PV devices. Secondly, contacting is a well-known issue for p-CdTe [29], and may create a back-contact barrier (from band line ups [30] or Fermi level pinning [31]) which limits the forward bias current of the devices [32]. Strategies, such as using ZnTe:Cu [33], are required to mitigate this barrier by supplying high local p-doping near the contact. Nevertheless, contacting to p-CdTe solar cells is found to be universally problematic, and a recent review identified more than 70 different strategies for making the contacts [29]. Whatever the origin of the difficulties for contacting p-CdTe, they are not an issue with n-type material, for which indium is known to give a low-resistance Ohmic contact [34,35].
We therefore propose that re-designing the CdTe solar cell for an n-type absorber layer will provide options for manipulating the doping level and contacting that are not available from p-type. In particular, there may be opportunities for increasing V oc and in simplifying contacting.
The key papers on the doping and properties of n-type and indium-doped CdTe include: (a) Indium doping [28]: n-doping with In is achievable with close to 100% activation in single crystal CdTe up to ∼10 18 cm 3 after which V Cd compensation sets in, followed by indium precipitation above 10 20 cm −3 . The diffusion coefficient of indium is D = 6.48 × 10 −4 exp(−1.15 eV kT −1 ) cm 2 s −1 at Te-saturation and D = 117 exp(−2.21 eV kT −1 ) cm 2 s −1 at Cd-saturation. Segregation coefficients and the compensation mechanisms for indium in CdTe are comprehensively reviewed in Panchuk and Fochuk [36]. (b) Contacting [34]: low resistance contacts to single crystal n-CdTe may be formed with indium, providing the surface is oxide-free and the indium is thick enough, for example, 500 nm. (c) Minority carrier lifetime and diffusion length: A range of values have been reported for τ h in single crystal samples as follows: single crystal Bridgman-0.5 ns [37]; ion implanted single crystal-5.5 ns [38]; epitaxial MBE [39] and single crystal solvent evaporated 18.6 ns [39]. For polycrystalline CdTe grown under excess Te, the lifetime was measured as 7 ns [40]. The value of τ h may well be a function of the doping density. Indeed, the minority carrier diffusion length L h in single crystals drops in the range 2 × 10 13 -8 × 10 17 cm −3 according to the empirical relation: Published reports of thin film polycrystalline devices having intentionally doped n-CdTe are limited to that from Palekis et al [41] who fabricated glass/TCO/CdS/CdTe:In/ZnTe devices having efficiencies up to 8.8%. The paper reports the effect on devices by varying the II/VI ratio in an elemental vapor delivery system, and of varying the amount of indium supplied to the substrate. No chloride processing was used.
There are also reports of n-CdTe in more complex junction devices: A series of papers from Arizona State University report MBE-growth of a unique ZnTe/CdTe/MgCdTe double heterostructure PV device demonstrating V oc > 1 V, this being higher than for conventional CdTe devices [42][43][44][45][46]. Another work on MBE-grown epitaxial p-ZnTe/n-CdTe reports dark diode behavior, but no PV response [35]. Heterojunctions with n-CdTe have also been attempted on single crystal CdTe wafers coated with p-diamond (0.1%) [47], p-NiO x (0.001%) [48] and MoO x (no power conversion efficiency) [49,50].
CdTe homojunctions have received little attention despite being ubiquitous in the traditional 'n-CdS/p-CdTe' device. It is well-known that the junction in these devices is a shallow homojunction in the CdTe, as demonstrated by electron beam induced current imaging [51]. Nevertheless, they are described as n-CdS/p-CdTe for convenience, and this is sometimes misunderstood [52]. The fabrication of doped homojunctions by MOCVD [22,53] has been reported but with dark diode device results only. The modelling of intentionally doped shallow homojunction options has been reported by Song et al [54,55].
In this work, we report post-growth doping trials using both indium metal and indium chloride with characterization by bulk and surface chemical analysis, near-surface band positions and carrier lifetime measurements. Photoemission measurements confirmed that indium imparted n-conductivity but that chlorine acted to compensate this. Accordingly, junctions based on CdS/CdTe had low efficiencies (∼1%), for indium-only doped CdTe, as would be expected for an n + -n junction. Higher efficiencies were observed for chloride-treated material (∼10%), but given the observed compensation as a result of the chloride treatment, we do not consider this to be an n-CdTe based structure. As a result, we conclude the paper with modelling to evaluate future p-n device designs that have the potential to achieve high V oc .

Materials fabrication and doping
Thin films for the doping studies and for junction fabrication were prepared as follows: CdTe films were deposited by close space sublimation in a custom-built kit (Electro-gas Systems, Manchester). The source material was undoped 5 N CdTe (Alfa Aesar, particle size 0.5-4 mm), and the films were 3 µm thick. CdS was argon ion sputtered from a single source 5 N target (PiKem) in an AJA International Orion system. Au contacts were applied by evaporating 99.99% gold (Goodfellow). Substrates for device fabrication were NSG TEC15, a fluorine doped tin oxide (FTO) coated glass product having a sheet resistance of 15 Ω □ −1 . Where chloride processing was applied to the finished devices, it was done by spraying aqueous MgCl 2 (99+% Alfa Aesar) followed by heating in air in a tube furnace at 410 • C after any doping had been done. Post-growth doping trials were conducted with two strategies, either (a) evaporation of indium metal directly onto the CdTe films (99.999% Sigma Aldrich) and annealing in a tube furnace under argon or (b) InCl 3 spray coating onto CdTe films in either an aqueous or methanolic solution (1 M solutions of 99.999% InCl 3 from Alfa Aesar) followed by annealing in air or under argon. Residues were removed using dilute nitric acid.

Materials characterization
X-ray diffraction was carried out using monochromated Cu-Kα 1 radiation in θ-2θ mode on a Rigaku Smart Lab system. X-ray photoelectron spectroscopy (XPS) of the surface chemistry of the films was undertaken at Harwell XPS. Calibration standards for the quantitative secondary ion mass spectroscopy (SIMS) measurement of indium in CdTe were prepared by implanting 115 In into single crystal CdTe wafers to a peak density of 10 19 cm −3 at the Surrey Ion Beam Centre, UK. SIMS profiles were measured on a Cameca IMS 7f instrument by LSA Ltd.
Hot probe measurements were attempted on the films, with one contact being heated with a soldering iron to give a temperature gradient of 40 • C cm −1 . The currents were measured using a Keithley 2400 source meter on the 105 µA range and n-type InSb wafer was used as a reference. Hall measurements were attempted using a high sensitivity Semilab PDL Hall system at the University of Loughborough.
Hard x-ray photoemission spectroscopy (HAXPES) was performed on a beamline i09 at the Diamond Light Source synchrotron at Didcot, UK. Measurements were made using beam energies of 1.09, 2.2 and 6.6 keV in order to achieve a range of sampling depths. The core levels of Cd, Te, O and In were used to gain chemical information while the positions of the Fermi levels in the band gap were estimated by measuring the valence band (VB) edge relative to the Fermi level of a gold calibration specimen. Further details of the methods are reported by Hobson et al [56]. Both In-metal diffused and doped/MgCl 2 -treated samples were measured, these being freshly prepared in the normal laboratory environment and transferred to the synchrotron in a vacuum suitcase system.
Carrier lifetime measurements were performed using an integrated Horiba time-correlated single photon counting system comprised of a DeltaDiode DD-650L 650 nm 100 MHz diode laser with a DeltaDiode controller and DeltaHub timing units. The excitation was performed through the glass to avoid any artefacts from free surface recombination. Time resolved photoluminescence (TRPL) signals were detected using a Horiba PPD-900 Si CCD-based picosecond photon detector mounted on a Horiba Jobin Yvon iHR320 fully automated spectrometer utilizing a 900 lines mm −1 diffraction grating. The decay signal for TRPL was selectively monitored at the peaks of near-gap CdTe emission. Data fitting was done using Horiba DAS-6 software. Standard PL spectra were measured using the 650 nm pulsed diode laser as an excitation source and collected using a Horiba Syncerity Si CCD detector.

Devices and device characterization
Trial structures based on a glass/FTO/n-CdS/n-CdTe/Au were fabricated using all of the different doping and chloride treatment protocols above. Current voltage (J-V) curves were measured using a Keithley 2400 source meter and when required, AAA AM1.5 illumination was provided at 1000 W m −2 by a TS Space Systems solar simulator. External quantum efficiency (EQE) measurements were made under white light bias using a Bentham PVE300 system. Capacitance-voltage (CV) measurements were used to determine carrier concentrations in the devices. Measurements were performed using a Solartron 1260 frequency response analyser with SmartLab software used to apply a 100 kHz frequency 30 mV AC perturbation voltage with the DC bias voltage sweep being from −1 to +1 V.

Simulations
Device simulations were performed with SCAPS Version 3.3.07, which is a one-dimensional solar cell simulation application developed at the Department of Electronics and Information Systems of the University of Gent, Belgium [57,58]. This software is designed for thin film PV simulations and is based on the Poisson equation and the continuity equations for electrons and holes. A fulsome description of the simulation parameters used is given in the supplementary information.

Chemical effects and indium incorporation in post-growth doped CdTe:In
Diffusion of In metal into CdTe surfaces did not leave any residue. However, use of either methanolic or aqueous InCl 3 left a white coating. Figures 1(a) and SI1 show the XPS survey spectra which have strong oxygen signals along with weaker signatures from indium and chlorine. The residue was sufficiently thick to allow XRD (figures 1(b) and SI2) and the principal peaks corresponded with those of InOCl (JCPDS 11-0510) [59]. This was considered to arise from the reaction [60]: This reaction is known to occur above 280 • C, while further heat treatment at 400 • C > T > 900 • C for extended periods is known to yield In 2 O 3 as follows: 3InOCl → In 2 O 3 + InCl 3 . Rinsing in 1 M HCl dissolved the InOCl and returned the XRD of the coated surfaces back to the standard pattern for CdTe ( figure 1(b), bottom panel, JCPDS 15-0770). Hence the use of InCl 3 in both methanolic and aqueous solutions yielded InOCl-this being expected since methanol contains sufficient residual water to supply the reaction above.
Quantitative SIMS was used to check the incorporation of indium into the films.

Carrier type and hard x-ray photoemission determination of Fermi level positions
Determination of the carrier type in polycrystalline CdTe films using transport methods is known to be difficult due to the adverse influence of the grain boundary fields. Nevertheless, we attempted both hot-probe and Hall measurements. While we were able to confirm n-type conductivity using the hot probe on large-grained (>5 mm) fragments of bulk CdTe:In, no reliable data could be obtained from any of the thin films prepared for this work. Similarly, high sensitivity Hall measurements made using both conventional and parallel dipole line systems failed to give a conclusive indication of the conductivity type.
Instead, we utilised HAXPES to determine the Fermi level positions which could then be used to infer the majority carrier type directly. Whereas conventional laboratory XPS is capable of Fermi level determination, we used a synchrotron hard x-ray source here to obtain depth-sensitive information: surface measurements may be compromised by surface band bending or else the surface chemistry. In this work, we used x-ray energies of 1.09, 2.2 and 6.6 keV at the Diamond Light Source i09 beamline in order to identify trends as a function of depth. Figure 3 shows the variation in the apparent band positions for the two samples measured, i.e. those which had been (i) post-growth doped by diffusion of indium metal and (ii) doped and then treated with MgCl 2 . The depth scale was obtained by estimating the escape depths of the photoelectrons as being twice the inelastic mean free paths (IMFP) using the IMFP data from Shinotsuka et al [61] and using the average of the photoelectron energies for Cd and Te 3d 5/2 , and 3d 3/2 emissions for each energy (see figure SI3).
The trend line for the doped-only band positions in figure 3 shows an unexpected peak-such behaviour would be unphysical as a solution to Poisson's equation for a homogeneous semiconductor terminated by a charged surface. Hence, we considered the possible origins of this anomaly. First, we ruled out sample charging effects by monitoring the invariance of the C1s and core-level peaks. Secondly, we looked into systematic errors in the VB-edge fitting procedure. The data in the figure comes from applying a linear fit to the VB edge, with an example being shown in figure SI4. The uncertainty was <0.1 eV, which is less than the   difference between samples. Moreover, these results were comparable to those obtained from fitting the calculated VB DOS to the experimental spectra. Thirdly, and most tellingly, we estimated the near-surface oxide formation using the HAXPES core-level peaks and using photoelectron cross sections from Yeh and Lindau [62] (see figure SI5). For the doped sample, the ratio [O]/[Cd + Te] was 0.72 at the near-surface, falling to 0.29 and 0.23 for harder x-rays, while for the doped and chloride-treated sample it was 1.69, falling to 1.0 and 0.93 deeper in the sample. Recognising that these results represent convolutions over the escape depths (as opposed to true depth profiles), it remains clear in both cases that the near-surface measurements (1.09 keV) represent a more oxidised region than the deeper measurements (2.2 and 6.6 keV). We have therefore interpreted figure 3 using the data from these two deeper measurements only.
In both the doped and doped/chloride samples, the Fermi level measured with the highest energy x-rays is above mid-gap, indicating that they all have n-type characteristics. The In doped-only sample has the highest Fermi level position (0 eV in figure 3), this being consistent with n-doping. On the other hand, chloride treatment acts to depress the Fermi level, indicating that it acts to compensate the n-doping. In both cases, there is upward band bending towards the surface (disregarding the near surface data point for the oxidised region as explained above). Therefore, while all these photoemission measurements are surface-based, and so do not represent the true bulk materials behaviour, the trend is clear, and it is reasonable to conclude that the indium doping imparts n-type conductivity (indeed, this is borne out by the behaviour of the junction test structures reported in section 3.3). Figure 4 shows the room temperature PL spectra and decay curves obtained from CdTe films which were undoped (i.e. as grown) and post-growth doped with indium metal. All samples showed broad PL peaks which are consistent with the material being polycrystalline rather than single crystal. The strongest peaks were centered on ∼820 nm (1.51 eV) which is close to the room temperature bandgap. This energy is consistent with the near-gap emission seen in, for example, high quality MBE-grown CdTe:In samples, although it is sharper at low temperatures [63].

Time resolved PL measurement of minority carrier lifetime
The decay curves and their time constants are shown in figure 4 and table 1 respectively. For all samples, the time dependent PL decay was fitted to a two-component exponential: where A is a background offset (fitting parameter), the process subscripted '1' is charge separation and that subscripted '2' represents minority carrier recombination. For these samples, the initial decay in the PL intensity in the first one or two nanoseconds is dominated by the fast component τ 1 (0.21-0.54 ns) while thereafter the curve shape matches that for the slower, bulk recombination, lifetime, τ 2 , which falls in the range 1.5-2.5 ns. As expected, these values for polycrystalline material are all somewhat less than those reported for single crystal n-CdTe (5-20 ns) [37][38][39][40].

CdS/CdTe:In junctions
To gain insight into how CdTe:In films performed in the context of a well-understood junction, we incorporated CdTe:In into 'traditional' CdS/CdTe junction designs. For example, it would be expected that n-type behavior in the CdTe would give a very weak junction when combined with CdS, which is n-type. We also examined the effects of chloride processing, for which the outcomes on CdTe:In in devices are unknown. Figure 5 shows the evolution of the CdS/CdTe:In solar cell performance parameters with the post-growth doping temperature in the range 200 • C-400 • C for 20 min diffusion time, plus an undoped control. (Shunt and series resistances are shown in figure SI6.) Although there is a step up in V oc from ∼0.2 to ∼0.4 V for the doped films compared to the undoped ones, all the performance parameters are low, and the average efficiency did not exceed 0.1%. The highest performing device achieved 0.24%, that being for doping at 400 • C. The reason for this low performance is explored in figure SI7 which shows the EQE responses for the same devices. All have a very low EQE response with a peak at the long wavelength (near-gap) end of the spectrum. This is characteristic of a buried junction which accounts for the very low photocurrents achieved.

Junctions having CdTe post-growth doped with indium metal
Chloride treatment of indium-doped devices had the effect of increasing all performance parameters significantly: figures 6(a)-(d) show the working parameters of devices that had been first post-growth indium metal doped and then MgCl 2 treated at 410 • C for 20-60 min. There has been an increase of PCE from <0.5 to >10%. This increase is reflected in the change in the EQE shape from the buried junction seen for indium-only devices (figure SI7) to the shallow junction shape seen in figure 6(e). Moreover, the EQE shape for the chloride-treated devices very closely resembles that for a conventional CdS/CdTe device having copper doping and chloride processing, i.e. having a shoulder for wavelengths <500 nm due to parasitic absorption in the CdS and high collection between 600 nm and the bandgap cutoff. Repeat runs of the best processing conditions gave a peak performing cell having PCE = 10.3%, V oc = 0.74 V, J sc = 25 mA cm −2 and FF = 55%, R shunt = 985 Ω and R series = 11.2 Ω as shown in figure 6(f). However, since the CdTe:In is known to be compensated by chlorine (section 3.1) this performance cannot be attributed to n-doping-a point which we discuss further in section 4. 3 We conducted similar trials of CdS/CdTe junctions with InCl 3 processing in the temperature range 200 • C-400 • C. A temperature of 200 • C gave a buried junction EQE response as shown in figure SI8, with shallow junctions forming for higher temperatures. Hence temperatures in the range 390 • C-420 • C were explored in greater detail, with the device performance parameters being shown in figures 7(a)-(d) and their EQEs in figure SI9. The PV performance parameters did not change greatly in this temperature range but there was a peak in the C-V carrier concentration of ∼6 × 10 14 cm −3 for processing at 400 • C, which was twice that measured for the MgCl 2 control sample figure 7(e). (The C-V data is shown in figure SI10.) The highest performance for this series was a PCE of 9.6%, as shown in figure 7(f). Again, compensation of the n-doping by chlorine has generated a response similar to that of a traditional CdS/CdTe device and is discussed in section 4.

Modelling future concepts for n-CdTe devices
In order to realize the full potential of n-CdTe in devices we recognize that it will be necessary to deploy it in proper p-n junctions rather than the CdS/CdTe:In structures used in this work to gain initial insights. We therefore conclude this paper with a survey of potential device architectures and some SCAPS modelling of specific designs. There are four basic alternatives, as shown in figure 8. They are (a) superstrate/n-uppermost, (b) superstrate/p-uppermost, (c) substrate/n-uppermost and (d) substrate/p-uppermost. Configurations (a) and (c), for which the n-type absorber is uppermost, have the advantage that light entering the junction does not have to first pass through a p-type semiconductor for which there will be optical losses.
We therefore trialled simulations of a wide range of device structures having the 'superstrate' configuration on FTO coated glass and with the light entering the n-CdTe before the p-type partner layer. Partner layers included i-and p-ZnTe, p-aSi, p-P3HT and p-CuI and also n-ZnO. We recognize that to make realistic models, it would be necessary to conduct extensive experiments to determine the bulk and interface trapping characteristics. Since that was out of the scope of this study, we instead included mid-gap traps in the model for all materials and having densities similar to those reported in the literature. Examples include: P3HT, 10 13   selection of structures for which SCAPS modelling [57,58] indicates V oc > 1 V. For example, figure 9 shows the SCAPS-generated band diagram for glass/FTO/n-ZnO/n-CdTe/p-ZnTe/P3HT/Au and for which PCE = 24.6%; V oc = 1.1 V; J sc = 30.2 mA·cm −2 and FF = 74.2%. This exercise gives some pointers to the structures for n-CdTe devices having the potential for high voltages and which may be worthy of future experimental investigation.

Discussion
We expected that n-doping of thin film polycrystalline CdTe would be feasible given the reports of it for single crystals [28,64]. However, direct measurements using Hall and hot probe methods failed to confirm this and we speculate that this was due to grain boundary potentials. Nevertheless, HAXPES determinations of the Fermi level were able to identify n-type behavior for CdTe:In, with upward band bending near the  surface. However, C-V measurements of CdS/CdTe:In test structures indicated carrier concentrations of approximately 6 × 10 14 cm −3 , while the chemical concentration was estimated as 10 19 cm −3 . The low carrier activation is most likely due to compensation from over-doping, as described by Watson and Shaw [28]. Such compensation may also be responsible for the relatively low minority carrier lifetimes observed (∼2.5 ns) although there are relatively few studies of lifetimes in n-type CdTe with which to compare.
Introduction of chlorides via MgCl 2 acted to depress the Fermi level indicating that chlorine acts to compensate the n-doping from indium. We know of no other specific study of co-doping or compensation in CdTe with both In and Cl. However, whereas other halogens are n-dopants, chlorine is known to have complex behaviour in CdTe and is used for compensating high resistivity CdTe for detectors [65]. In CdTe solar cells, it has been linked to 'type conversion' from n-to p-type [4,5] to impart PV performance in junctions with n-CdTe [66]. (Grain boundary segregation of chlorine is also considered to impart grain boundary electrical passivation [67].) The CdS/CdTe:In test structures made here acted to confirm that the CdTe:In performed as might be expected from the above in the context of junctions. Indeed, the weak junction response for CdS/CdTe:In was consistent with there being an n-n junction. Compensation of the CdTe:In with chlorine acted to strengthen the junction, but only to the level of a non-intentionally doped CdTe structure with chlorine treatment.
Hence, in order to realize the full potential of devices having n-CdTe absorbers, it will be necessary to make junctions having uncompensated n-CdTe with p-type partner layers. While SCAPS simulations (figures 8 and 9) have identified some possible candidate structures for which V oc may exceed 1 V, we recognize the limitations of modelling in this respect. In particular, practical realization of devices having n-CdTe will have to overcome the issue of grain boundary passivation, namely that while chlorine effects passivation it would also compensate the n-doping. Further experimental investigations will be required to explore the device architecture and performance opportunities for n-CdTe.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// datacat.liverpool.ac.uk/id/eprint/1677.