Superconducting Nb_xTi_1−xN prepared at high deposition rates with plasma-enhanced atomic layer deposition and substrate biasing Open Access

Superconducting thin films are essential in a vast variety of enabling and upcoming quantum technologies. Notable examples include qubits, resonators, and through-silicon vias for quantum processors,^1–5 and single-photon detectors and superconducting quantum interference devices (SQUIDs) for quantum sensors.^6–8 These applications generally require accurate composition control and low impurity contents of the materials to achieve the desired superconducting properties and device performance.^5,8–11 The requirements for film thickness and structural properties depend on the application. For instance, efficient photon detection in the mid- to far-infrared calls for ∼4–10 nm of disordered superconductors,⁸ while qubits typically demand a few tens to ∼300 nm of highly crystalline superconductors.^11–16 

Metal–nitride superconductors have found widespread utility due to their combination of good room-temperature electrical conductivity, high mechanical and chemical (oxidation) resistance, and high critical temperature (T_c) of superconductivity.^17,18 NbTiN especially is a widely used material system as it can combine the high T_c of NbN with the high conductivity and face-centered cubic (fcc) crystal-phase stability of TiN.^19–21 In pure NbN, the fcc phase provides the highest T_c.^17,20 However, in NbN, this phase is only thermodynamically metastable at room temperature.^21,22 The highest T_c of 17–18 K for NbTiN thick films have been achieved for a Nb_0.7Ti_0.3N stoichiometry in the fcc phase,^{17,21,23–26} which is stable at room temperature.^19,20,25 A room-temperature resistivity of 54 μΩ cm has been reported for thick films of this composition.²¹ 

NbTiN thin films are conventionally grown by physical vapor deposition techniques such as reactive magnetron sputtering.^19,27–31 However, interest in atomic layer deposition (ALD) of superconducting NbTiN has risen over the past decade,^32–36 as this technique provides atomic-scale thickness control, composition control, and conformal growth on complex structures. Additionally, plasma-enhanced ALD (PEALD) enables low-temperature processing while maintaining high film quality. Promising results for quantum device demonstrators such as resonators^37,38 and superconducting nanowire single-photon detectors (SNSPDs)^9,39 have been achieved with PEALD of NbN and TiN. However, scalable integration of PEALD in the diverse field of superconducting quantum device fabrication is hampered by low deposition rates of typically ∼5 nm/h^32,40–46 compared to the ∼10–300 nm/min reported for sputtered superconducting NbTiN.^27–30 

In this work, superconducting Nb_xTi_1−xN films of 4–97 nm film thickness are grown by a PEALD supercycle process with RF substrate biasing providing ion-energy control. As shown in previous work, the application of an RF substrate bias during plasma exposure can improve electrical conductivity, composition, and structural properties of the (superconducting) metal nitride.^47–50 The work described in this publication explores Nb_xTi_1−xN films deposited with a range of supercycle values, the so-called NbN/(NbN + TiN) cycle ratios, and substrate bias voltages. First, the self-limitation of the surface reactions is tested by constructing saturation curves as a function of precursor dose and plasma exposure. The deposition rate of the resulting supercycle process is also discussed. Then, the electrical resistivity and T_c are studied as a function of cycle ratio and substrate bias from grounded (0 V) to −149 V. This is accompanied by an investigation of the trends in film composition and structure. Finally, electrical resistivity and T_c of films with thicknesses ranging from 4 to 97 nm are explored.

Nb_xTi_1−xN thin films were prepared by PEALD in the Oxford Instruments PlasmaPro ASP system. It contains a remote plasma source operated at radio frequency (13.56 MHz) into which the power is capacitively coupled using an automated matching unit (Fig. 1). The plasma source design, which has been described in Ref. 51, enables a remote plasma despite the short ∼6 cm distance to the substrate table. The small chamber volume allows for short residence times and the RF (13.56 MHz) substrate bias provides ion-energy control independent from the remote plasma source. By supplying an RF signal with varied amplitudes to the substrate, the voltage over the plasma sheath can be controlled. As a result of the negative average bias voltage, the energy of the ions impacting the film is enhanced.

The Nb_xTi_1−xN films were grown on Si(100) wafers with a diameter of 150 mm and a native oxide surface through a supercycle process as detailed in Fig. 2. One supercycle consists of m NbN and n TiN cycles. The chosen combinations are m:n = 0:1, 1:4, 2:3, 1:1, 3:2, 4:1, and 1:0. The supercycle is repeated p times until the desired film thickness is reached. The cycle ratio NbN/(NbN + TiN) is given by m/(m + n). The processes were separately optimized for low electrical resistivity of the resulting NbN and TiN films. The used precursors were Nb[N(CH₃)₂]₃[NC(CH₃)₃] (TBTDEN) and Ti[N(CH₃)₂]₄ (TDMAT), which were, respectively, heated to 100 and 60 °C and supplied to the chamber with a 40 sccm Ar bubbling flow. The gas flow through the plasma source consisted of a mixture of Ar-H₂-N₂ gas and the plasma was generated with a source power of 250 and 200 W for the NbN and TiN processes, respectively. The table temperature was kept at 320 °C, which has been chosen for obtaining films with low resistivity. The source and reactor walls were kept at 200 °C. Using these process conditions, 50 nm Nb_xTi_1−xN films were prepared with a −90 V substrate bias, again selected for low resistivity, with NbN/(NbN + TiN) cycle ratios ranging from 0 to 1. The dependence on substrate bias was explored for 30 nm Nb_0.5Ti_0.5N films grown using a NbN/(NbN + TiN) = 0.5 cycle ratio. A more detailed outline of the process conditions is provided in the supplementary material.

Film thicknesses are determined by x-ray reflectivity (XRR) supplemented by ex situ spectroscopic ellipsometry (SE). To fit the SE data, the dielectric function was modeled in the 0.74–5.1 eV range using one Drude and two Lorentz oscillators (see supplementary material). The free-carrier absorption is parameterized by the Drude oscillator in the SE model, such that the optical resistivity can be obtained.⁵² In addition, the lateral electrical sheet resistance is measured by four-point probe (FPP). The electrical resistivity is then calculated using the thickness determined by XRR. The T_c is determined by FPP resistance measurements and SQUID measurements of the magnetic susceptibility in a temperature sweep from 2.4 K to room temperature. X-ray photoelectron spectroscopy (XPS) depth profiling and Rutherford backscattering spectrometry (RBS) are used to study film composition. The structural properties are studied through x-ray diffraction (XRD) and (scanning) transmission electron microscopy [(S)TEM] imaging. These techniques are complemented by XRR for the mass density, wafer curvature measurements to quantify the residual stress, and atomic force microscopy (AFM) for the surface roughness. Moreover, STEM was combined with energy-dispersive x-ray spectroscopy (EDX) to identify the elemental distribution. These methods, including associated modeling procedures, are further elaborated in the supplementary material.

Saturated growth is confirmed for both the TiN [Figs. 3(a) and 3(b)] and NbN [Figs. 3(c) and 3(d)] processes. A growth per cycle (GPC) of 1.2 ± 0.1 Å and 0.7 ± 0.1 Å is found for TiN and NbN, respectively. The GPC values are in good agreement with those reported in the literature. In other PEALD works using the TDMAT and TBTDEN precursors, the TiN GPC shows a wide variation between 0.4 and 2.0 Å,^{42–46,53–57} while the NbN GPC remains between 0.5 and 0.7 Å.^{40,41,58–60} Combining the TiN and NbN processes in a supercycle process (as shown in Fig. 2), the resulting NbTiN GPC is defined by dividing the experimentally determined growth per supercycle (GPSC) by the number of TiN and NbN cycles (m + n) in a supercycle. As expected from the lower GPC of NbN compared to TiN, the NbTiN GPC gradually decreases for increasing NbN/(NbN + TiN) cycle ratio (see supplementary material). A 0.5 cycle ratio results in a GPC of 1.0–1.2 Å depending on the substrate bias used.

By dividing the GPC by the cycle time (Fig. 2), a high deposition rate between 58 ± 2 nm/h (TiN) and 25 ± 2 nm/h (NbN) is found (Fig. 4). Compared to PEALD of NbTiN,^32,33 TiN,^{42–46,53–57} and NbN^{40,41,58–60} reported in the literature the deposition rates in this work are exceptionally high despite similar GPC values. This is due to the short cycle times; a 0.5 cycle ratio corresponds to a 16.55 s duration for the supercycle. In Nb_xTi_1−xN PEALD literature, the 10–40 s plasma exposures^32,40–46 that are used to prepare conductive films are generally the main cause for low deposition rates. Increased deposition rates are beneficial for more widespread integration of PEALD in superconducting quantum device fabrication, especially for devices requiring film thicknesses beyond 100 nm.

Figure 5(a) shows that Nb_xTi_1−xN films prepared with a −90 V bias conduct well at room temperature. For a 50 nm film thickness, an electrical resistivity of 154 ± 11 μΩ cm is found for TiN, which gradually increases to 255 ± 17 μΩ cm for NbN. This approximately linear increase is also observed in sputtering literature.^21,63 Similarly, T_c increases from 3.0 ± 0.5 K for TiN to 13.0 ± 0.5 K for NbN [Fig. 5(b)]. As illustrated by the trend line in Fig. 5(b), the literature discussing NbTiN growth using reactive nitrogen diffusion^20,23,25 and sputtering techniques^17,26 consistently finds NbTiN has a higher T_c than NbN near the Nb_0.7Ti_0.3N composition. In the literature, Ti is reported to reduce the number of nitrogen vacancies and stabilize the cubic phase in NbTiN,^19–21 thus enhancing T_c when present in small amounts. In this work, XPS measurements indeed indicate that the addition of Ti reduces the deviation from stoichiometric composition [Fig. 6(b)]. However, all grown films in this work, including NbN, have a polycrystalline fcc structure (see supplementary material). The growth of fcc NbN could explain the absence of a maximum T_c near Nb_0.7Ti_0.3N. For PEALD of NbTiN both the presence³³ and absence³² of this maximum have been observed.

The obtained room-temperature resistivity and T_c values are in the expected range for thin, polycrystalline films. In PEALD literature, room-temperature resistivities of 80–1000 and 100–1600 μΩ cm, and critical temperatures of 2–5 K and 10–14 K have been obtained for TiN and NbN, respectively, for similar film thicknesses.^{37,39–41,43,44,46,53–56,58,60} Hence, the developed supercycle process allows for good electrical properties of the films with short cycle times.

To study the role of energetic ion bombardment in the growth of conductive Nb_xTi_1−xN films, 30 nm Nb_xTi_1−xN films are grown using a 0.5 cycle ratio with a grounded (0 V) substrate and substrate bias ranging from −25 to −149 V. Figures 5(c) and 5(d) show the room-temperature resistivity and T_c as a function of substrate bias. Both the electrical and optical resistivity have been determined [Fig. 5(c)]. The electrical (inter-grain) resistivity improves significantly from 497 ± 45 μΩ cm for the grounded substrate to 184 ± 19 μΩ cm when a moderate −81 V substrate bias is applied. A decrease in resistivity is similarly observed for the optical (intra-grain) resistivity, suggesting grain quality is improved through biasing. The inter-grain resistivity contains contributions from both the intra-grain resistivity and grain boundary scattering.⁵² The difference between electrical and optical resistivity is comparatively small, which is indicative of high-quality grain boundaries. The difference appears slightly larger for a low bias voltage, which could be due to less dense grain boundary regions with some oxygen present. The importance of accurate ion-energy control is emphasized by the slight increase in resistivity for high bias voltages. The onset of resistivity degradation at approximately −100 V can be understood from ion-induced damage as observed in previous work.^47,48

Contrary to the resistivity, the T_c of the 30 nm Nb_xTi_1−xN films does not show strong variations with substrate bias [Fig. 5(d)]. Instead, the highest T_c of 7.2 ± 0.4 K is attained for a grounded substrate, whereas a biased substrate results in a quite similar T_c of 5.5–6.5 K. In addition to T_c, the FPP low-temperature resistance measurements also allow determination of the residual resistance ratio (RRR), defined as $R(T=130 K)/R(T=20 K)$ in this work, and the superconducting transition width ΔT, defined as $ΔT=T(R=0.9*RT=20 K)−T(R=0.1*RT=20 K).$³⁹ The RRR provides an indication of the film crystallinity, purity, and nature of conduction.⁶⁴ An RRR ranging from 0.96 to 1.0 is found independent of substrate bias, indicating high film quality and metallic conduction above T_c. Moreover, a low transition width ΔT of approximately 0.5 K is found, which suggests good film homogeneity.

To understand the role of ion bombardment in the growth of Nb_xTi_1−xN thin films with good electrical properties, the composition and structural properties of the films are studied.

The film composition is investigated by XPS and RBS measurements for a cycle ratio NbN/(NbN + TiN) ranging from 0 to 1 using a −90 V substrate bias [Figs. 6(a)–6(c)] and as a function of substrate bias for a 0.5 cycle ratio [Figs. 6(d)–6(f)]. Through depth profiling in XPS, the thin oxidized surface layer is identified and excluded from the analysis. Within the film, the composition is determined after a 3 s Ar⁺ sputter time, which is then averaged over 120 subsequent sputter steps. In contrast, the RBS atomic content is determined for the complete film thickness including interfaces. Figure 6(a) shows the gradual change in measured Nb:Ti ratio with cycle ratio, following the rule of mixtures, with reasonable agreement between XPS and RBS measurements. This finding agrees with the gradual increase in room-temperature resistivity and T_c as seen in Figs. 5(a) and 5(b).

From XPS measurements, it is also observed that the Nb_xTi_1−xN films, except for NbN, are near-stoichiometric in (N + C) to (Nb + Ti) ratio with a small contribution from carbon incorporated from the precursor ligands in the form of C bonded to metal [Figs. 6(b) and 6(e)]. As the film surface, containing adventitious carbon, and the nitrogen-rich substrate interface (see EDX in supplementary material) are not distinguishable from the Nb_xTi_1−xN films in RBS, RBS data are not used to evaluate the stoichiometry. XPS shows the NbN film is substoichiometric with (N + C):Nb = 0.72 ± 0.08. Substoichiometric phases are known to be stable for cubic NbN.^65–68 The C1s binding energy spectrum is used to identify the binding configurations of C (see also the supplementary material), isolating the contribution from C bonded to metal atoms from the other C bonds to determine the stoichiometry. Hence, the grown material can be referred to as a carbonitride. As the C contribution is small, the films are denoted as Nb_xTi_1−xN. While C incorporation degrades TiN conductivity, NbC has a higher conductivity than NbN.¹⁷ Moreover, NbC has a lower thick-film T_c of 11 K compared to the 17 K for NbN,¹⁷ whereas superconductivity has not been experimentally observed for TiC.^17,25,69 The presence of C thus has a slight flattening effect on the dependence of resistivity on cycle ratio [Fig. 5(a)] and somewhat lowers the T_c [Figs. 5(b) and 5(d)].

Figure 6(c) shows a low O content without dependence on the cycle ratio. While XPS yields a 2–5 at. % O content, RBS gives 1 at. %. In XPS depth profiling it is observed that the film slightly oxidizes during the measurement after each sputter step, leading to increased O content (see supplementary material). For this reason, the O content from RBS is adopted for the film composition. Because RBS measurements include the native oxide interfacial layer and the oxidized surface, the O content in the Nb_xTi_1−xN film is assumed to be virtually negligible.

Interestingly, XPS and RBS measurements reveal substrate biasing does not result in a notable change in composition [Figs. 6(d)–6(f)]. A constant Nb/(Nb + Ti) of approximately 0.5 is found and a low ∼1 at. % O content is observed from RBS measurements across the range of explored bias voltages. Thus, the trend in electrical resistivity [Fig. 5(c)] cannot be explained by compositional variations. Possibly, the short cycle time of the process allows for negligible O impurities without the need for enhanced ion energies. In other works, it has been observed that longer cycle times lead to increased O incorporation from O-containing species in the reactor background.^55,70 Lowering of the O content required a high ion-energy dose in the form of a long plasma exposure or substrate-biased plasma.^47–49 

The constant Nb_0.5Ti_0.5N stoichiometry found as a function of substrate bias could be a part of the explanation for the fairly small variation in T_c [Fig. 5(d)]. For ALD Nb_xTi_1−xN films, it has been experimentally determined that a decrease below x = 0.7 in Nb_xTi_1−xN introduces the bosonic mechanism of superconductivity suppression.³⁴ This means that as the sheet resistance increases, T_c decreases only weakly. Consequently, for the Nb_0.5Ti_0.5N composition the trend in T_c is expected to deviate from the trend in resistivity. To further understand how energetic ion bombardment influences the resistivity and T_c of the thin Nb_0.5Ti_0.5N films, the structural properties are studied (Fig. 7).

Energetic ion bombardment has been shown to significantly affect thin-film structure in previous PEALD work^47–49 and in physical vapor deposition literature.⁷¹  Figure 7 shows how the Nb_0.5Ti_0.5N films are affected by biasing in terms of crystallinity, disorder, density, and residual stress. Gonio [Fig. 7(a)] and grazing incidence (see supplementary material) XRD measurements show that all studied films have a polycrystalline fcc structure. Compared to a grounded substrate, a low −49 V bias yields an enhanced (111) peak signal. For moderate (−81 V) to high (−149 V) bias, the (111) peak signal decreases and widens, which is indicative of more disordered films. Similarly, cross-sectional HR-TEM images [Fig. 7(b)] show a columnar structure for two different bias powers, with more distinct columns for the low −49 V bias and more disordered, interrupted columns for the moderate −90 V bias. For both films, grains can be identified that span the film thickness. Combining STEM with EDX mapping (see supplementary material) confirms homogeneous mixing of Ti and Nb.

Remarkably, the HAADF-STEM imaging mode reveals directional density fluctuations. The −49 V bias results in distinct vertical regions of lower density, while the film prepared with −90 V bias has less distinct, horizontal fluctuations. The vertical regions for −49 V bias could be the manifestation of a lower material density at the vertical grain boundaries. For the horizontal fluctuations for −90 V bias, grains are observed to continue across the fluctuations in density, suggesting that disorder and reduced density may be present locally in these lower-density planes.

The higher lateral resistivity found for the film prepared with −49 V bias compared to higher bias powers [Fig. 5(c)] can be understood from the distinct vertical grain boundaries that may result in enhanced grain boundary scattering. The latter could explain the different trend observed in T_c [Fig. 5(d)]. Whereas the electrical resistivity is heightened by grain boundaries as observed at low bias voltages, the T_c is not degraded, presuming that no insulating oxides are present at these grain boundaries.⁷² The slight drop in T_c when a substrate bias is applied could be a result of a decrease in crystallinity^34,73,74 and enhanced interfacial roughness,^75–78 which increases slightly up to 1 nm in the explored substrate bias range (see supplementary material).

The trend in resistivity [Fig. 5(c)] is also consistent with the trend in mass density as measured by XRR [Fig. 7(c)]. Values between 5.6 ± 0.4 and 6.2 ± 0.3 g/cm³ typical for polycrystalline thin NbTiN films^32,79 are found. The mass density shows ion-induced densification followed by relaxation for highly energetic ion bombardment. A trend comparable to that of the mass density is also observed in the residual stress [Fig. 7(c)], where a moderate substrate bias induces a significant −3 ± 1 GPa compressive stress, which relaxes for higher ion energies.⁴⁸ The shift in (111) peak position in the x-ray diffractograms of Fig. 7(a) reveals an increase in the out-of-plane lattice parameter through biasing, in agreement with the in-plane residual compressive stress. Equivalent to structural disorder, lattice distortions can generally result in a (marginal) reduction in T_c.^80–82 Hence, the dependence of resistivity and T_c on substrate bias can be understood through the composition and structural properties of the Nb_0.5Ti_0.5N films.

To further elucidate the electrical properties of the Nb_xTi_1−xN films, films were grown with 4–97 nm thicknesses. This was done using a NbN/(NbN + TiN) = 0.5 cycle ratio and −090 V bias (Fig. 8). The 97 nm film has been grown with a different 0.8 cycle ratio to reduce the residual stress (see supplementary material), which is highest around the −90 V bias [Fig. 7(c)], and prevent delamination of the film. Enabling more Ti-rich Nb_xTi_1−xN films beyond 100 nm thickness requires the optimization of biased PEALD processes for low residual stress while maintaining desirable electrical properties. This is outside the scope of the current work. The electrical room-temperature resistivity [Fig. 8(a)] of the 4–45 nm films decreases with thickness according to the Fuchs–Sondheimer model for polycrystalline films (see supplementary material),⁸³ which assumes diffuse surface scattering in addition to bulk scattering. The model yields a bulk room-temperature resistivity of 216 ± 15 μΩ cm and electron mean free path of 4.0 ± 0.4 nm. This resistivity is somewhat higher than the 35 μΩ cm reported in the literature for the Nb_0.5Ti_0.5N composition.²¹ 

In conclusion, this work has demonstrated a fast PEALD process for superconducting Nb_xTi_1−xN films on the PlasmaPro ASP system with substrate bias. The short cycle times of the supercycle process lead to high deposition rates of ∼30–60 nm/h. The substrate bias has a notable influence on film properties, which could be interesting for various quantum applications. Low O impurity contents, independent of NbN/(NbN + TiN) cycle ratio and substrate bias, and accurate composition control were confirmed. Substrate biasing improved electrical resistivity through enhanced structural properties while maintaining a good T_c of 6–7 K. The influence of substrate biasing on resistivity and T_c was explained by composition, grain boundaries, disorder, mass density, film stress, and surface roughness. Superconducting transitions were observed at 3.5 ± 0.7 K up to 7.1 ± 0.5 K for Nb_0.5Ti_0.5N films of thicknesses ranging from 4 to 45 nm, and at 10.2 ± 0.4 K for a Nb_0.8Ti_0.2N film of 97 nm thickness. The tunability through accurate ion-energy control and the high deposition rate through reactor design help establish PEALD of superconducting nitrides as an enabling technique for a wide range of quantum devices.

See the supplementary material for the PEALD process details, elaboration on characterization techniques, film thicknesses, GPC as a function of cycle ratio and substrate bias, RRR and ΔT as a function of substrate bias and film thickness, O1s and C1s peak data from XPS, EDX elemental mappings, grazing incidence XRD measurements of Nb_0.5Ti_0.5N films grown with various cycle ratios and substrate bias voltages, roughness as a function of substrate bias from AFM measurements, and residual stress as a function of cycle ratio.

This work has been carried out within the Open Technology Program with Project No. 19438, which is financed by the Netherlands Organization for Scientific Research (NWO). The authors gratefully acknowledge Arpita Saha (OIPT) for additional film depositions; Dr. Wim Arnold Bik (Detect99) for performing the RBS measurements; the Henry Royce Institute at the University of Sheffield for performing the SQUID measurements; Caspar van Bommel, Janneke Zeebregts, Barathi Krishnamoorthy for technical support at TU/e; and Cristian van Helvoirt (TU/e) for FIB preparation of TEM samples. Solliance and the Dutch province of Noord-Brabant are acknowledged for funding the TEM facility. The authors are thankful to Dr. Guillaume Krieger, Dr. Nicholas Chittock, Arthur de Jong, Sanne Deijkers, Renée van Limpt, and Boris de Bruin for insightful discussion and Dr. Roel Theeuwes for providing the schematic of Fig. 1.

Robert Hadfield acknowledges support from the UK Engineering & Physical Sciences Research Council (EPSRC—Project Nos. EP/W032627/1, EP/S026428/1, and EP/T00097X/1) and the UK Science and Technologies Facilities Council (STFC—Project No. ST/T005920/1). Nidhi Choudhary thanks the University of Glasgow for support through the Mary Gibb postgraduate research scholarship and access to the James Watt Nanofabrication Centre.

The authors have no conflicts to disclose.

Silke A. Peeters: Conceptualization (equal); Formal analysis (equal); Methodology (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Lisa E.W. H. M. Nelissen: Conceptualization (supporting); Formal analysis (equal); Methodology (equal); Visualization (equal). Dmytro Besprozvannyy: Formal analysis (supporting); Methodology (equal). Nidhi Choudhary: Formal analysis (supporting); Methodology (equal). Ciaran T. Lennon: Formal analysis (supporting); Methodology (equal); Writing – review & editing (supporting). Marcel A. Verheijen: Formal analysis (supporting); Methodology (equal); Writing – review & editing (equal). Michael Powell: Conceptualization (equal); Methodology (supporting). Louise Bailey: Conceptualization (equal); Methodology (supporting). Robert H. Hadfield: Supervision (equal); Writing – review & editing (supporting). W. M. M. (Erwin) Kessels: Conceptualization (equal); Funding acquisition (equal); Supervision (equal); Writing – review & editing (equal). Harm C. M. Knoops: Conceptualization (equal); Funding acquisition (equal); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplementary material.