Quantitative x-ray diffraction analysis of strain and interdiffusion in β-Ga₂O₃ superlattices of μ-Fe₂O₃ and β-(Al_xGa_1−x)₂O₃

β-Ga₂O₃ is an ultrawide bandgap (∼4.8 eV) oxide semiconductor that is being intensely studied for applications in high-power devices due to its very high breakdown voltage (E_b = 8 MV/cm).^1,2 β-Ga₂O₃ has a monoclinic structure with a space group of C2/m and lattice parameters a = 12.23, b = 3.04, and c = 5.80 Å.^1,3 High-quality β-Ga₂O₃ epitaxial films with smooth surfaces and uniform doping profiles can be grown using molecular beam epitaxy (MBE).^1,4–8 β-Ga₂O₃ has attracted interest for applications in power electronics and optoelectronics.^9–12 A new monoclinic μ-phase of Fe₂O₃, isomorphic to β-Ga₂O₃, was stabilized using MBE.^13,14 The epitaxial stabilization of the magnetic material μ-Fe₂O₃ paves the way toward realizing practical magneto-electronic devices on β-Ga₂O₃. DFT calculations show that μ-Fe₂O₃ has a bandgap in the visible range of ∼2.2 eV. The smaller bandgap of μ-Fe₂O₃ and the excellent lattice matching to β-Ga₂O₃ makes μ-Fe₂O₃ a very useful low bandgap insert in β-Ga₂O₃ heterostructures.

A material that has a bandgap larger than β-Ga₂O₃ is attractive to form quantum wells with carrier confinement and for optoelectronic applications in the deep ultraviolet.^15,16 Al₂O₃ is an excellent candidate since the corundum (α) phase of Al₂O₃ has a bandgap of 8.82 eV.^15,17 The alloy β-(Al_xGa_1−x)₂O₃ with a monoclinic structure has a bandgap between ∼4.8 and ∼7.2 eV depending on the Al composition.¹⁷ With a composition of ∼80% Al, β-(Al_xGa_1−x)₂O₃ is predicted to have a breakdown voltage of E_b ∼ 16 MV/cm,¹⁸ even higher than β-Ga₂O₃. β-(Al_xGa_1−x)₂O₃ heterostructures have been successfully grown using MBE on (010) β-Ga₂O₃ substrates with Al compositions of up to x = 0.18.^19,20 Heterostructures of μ-Fe₂O₃ and β-(Al_xGa_1−x)₂O₃ with β-Ga₂O₃ are of interest for applications in magneto-electronics, high-power switching applications, and UV optoelectronics.

The highly anisotropic crystal structure of β-Ga₂O₃ poses a challenge to epitaxial growth and characterization since thin film mechanics models typically rely on cubic or hexagonal symmetry.^1,3,21–23 One of the most important quantitative tools for developing epitaxial heterostructures is high-resolution XRD, which when quantitatively fit provides high precision measurements of film thicknesses, strain state, and alloy composition. Although HRXRD has been widely reported in the literature from β-Ga₂O₃ heterostructures, quantitative fits of the XRD data have been lacking. Here, we provide an example of how to adapt a commercial HRXRD simulation software (LEPTOS) to enable quantitative fits of complex heterostructures derived from monoclinic β-Ga₂O₃. The methodology can be adapted for any crystal orientation using a simple unit cell transformation to overcome constraints in the modeling software. The results are validated by comparison to STEM and quantitative EELS chemical maps providing understanding of the issues of interdiffusion in the β-Ga₂O₃ heterostructure family of alloys including Fe and Al Ga₂O₃ alloys for bandgap engineering in UWBG oxides.

The μ-Fe₂O₃ and β-(Al_xGa_1−x)₂O₃ heterostructures used in this study are grown via plasma-assisted molecular beam epitaxy (PAMBE) in a Riber/MBE Control M7 system equipped with Ga and Fe effusion cells and a Veeco oxygen plasma source with a forward power of 300 W (with ≤1 W reflected). The flow of O₂ is continuously monitored and maintained at a growth chamber pressure of 1.5 × 10⁻⁵ Torr. The substrates used for these growths are 5 × 5 × 0.5 mm³ unintentionally doped (UID) or Fe-doped (010) β-Ga₂O₃ single crystals (Tamura Corporation).

For the growths of μ-Fe₂O₃ superlattices, an ∼100-nm thick buffer layer of β-Ga₂O₃ is first grown followed by ten periods of β-Ga₂O₃ spacers and μ-Fe₂O₃ epilayers. The flux of Fe is held constant between samples with a beam equivalent pressure (BEP) of 3.2 × 10⁻⁸ Torr and the Fe deposition time is varied between 30 and 60 s. The substrate temperature T_sub as measured through the thermocouple is varied between 500 and 800 °C. β-Ga₂O₃ spacers are grown at a Ga flux between 4.0 and 1.0 × 10⁻⁷ Torr with a constant deposition time of 400 s. Some of the μ-Fe₂O₃ superlattices investigated in this study are described in Hettiaratchy et al.¹⁴ To investigate how the composition of β-(Al_xGa_1−x)₂O₃ changes with the Al flux, three samples were grown with a 100 nm β-Ga₂O₃ buffer layer and an ∼20–30 nm layer of β-(Al_xGa_1−x)₂O₃ at Al fluxes between 1.2 and 1.7 × 10⁻⁸ Torr. The Ga flux is kept fixed at 1.6 × 10⁻⁷ Torr and T_sub is kept fixed at 725 °C. Superlattices composed of ten periods of β-(Al_xGa_1−x)₂O₃ layers and β-Ga₂O₃ spacers were grown on a ∼40 nm buffer layer of β-Ga₂O₃ at T_sub = 800 °C and T_sub = 725 °C. For the superlattice grown at T_sub = 800 °C, the BEP of Al and Ga is 1.0 × 10⁻⁸ and 8.0 × 10⁻⁸ Torr, respectively. For the superlattice grown at T_sub = 725 °C, the BEP of Al and Ga is 7.0 × 10⁻⁹ and 1.4 × 10⁻⁷ Torr, respectively.

To investigate the structure and composition of superlattices based on β-Ga₂O₃, high-resolution x-ray diffraction (HRXRD) data are fitted with simulations based on dynamical x-ray diffraction. LEPTOS²¹ is a commercially available software package that is used for the analysis of structural and crystallographic parameters of thin film epitaxial heterostructures. LEPTOS uses the recursion matrix formalism for two-wave approximation of dynamical x-ray diffraction theory.²² This algorithm calculates x-ray scattering from thin film layers over the entire range of incidence and exit angles for various experimental scans. Other XRD fitting software packages, including micronova²⁴ and CADEM,²⁵ were also tested, and while they can measure the layer thicknesses, these software packages were currently unable to quantitatively derive the strain state of the μ-Fe₂O₃ superlattices; however, as they are open-source, they could be modified to include these capabilities.

In its default configuration, LEPTOS cannot simulate the XRD patterns of monoclinic structures along the (010) facet. LEPTOS is optimized for the XRD fitting of cubic and hexagonal crystal structures. As a workaround, the monoclinic unit cell of β-Ga₂O₃ is transformed into a triclinic unit cell while swapping the b and c axes. Following this transformation, the former (010) plane of the standard monoclinic cell is equivalent to the (001) plane of the b-c swapped triclinic cell. The triclinic space group [1] P1 has no symmetric atomic positions and all the 20 atomic sites for the transformed unit cell are defined in LEPTOS. This unit cell transformation allows for the fitting of monoclinic crystal structures with LEPTOS. For β-Ga₂O₃, the lattice constants used are a = 12.23, b = 3.04, and c = 5.80 Å.³ Based on DFT calculations, the lattice constants used for strained μ-Fe₂O₃ are a = 12.23, b = 3.16, and c = 5.80 Å,¹³and for relaxed μ-Fe₂O₃, a = 12.42, b = 3.07, c = 5.88 Å.¹³ To fit the XRD data for β-(Al_xGa_1−x)₂O₃ heterostructures, we defined a β-(Al_xGa_1−x)₂O₃ alloy that is fully strained to β-Ga₂O₃ using the triclinic, b–c swapped representations of the unit cells in LEPTOS. From Oshima et al.,²⁶ we assume that the in-plane lattice constants of β-(Al_xGa_1−x)₂O₃ are the same as β-Ga₂O₃ (a = 12.23 and c = 5.80 Å) and fit for the out-of-plane lattice parameter as a function of Al composition using Vegard's law, with b = 2.849 Å for fully strained Al₂O₃ [from Eq. (8a) in Oshima et al.²⁶].

Even when using b–c swapped triclinic unit cells, LEPTOS is unable to model the interface roughness or the strain-relaxation properties of these heterostructures. For the μ-Fe₂O₃ superlattices, the strain-relaxation was modeled by defining an alloy between the strained and relaxed μ-Fe₂O₃ crystal structures as obtained through DFT calculations.¹³ Vegard's law was used to determine the c-lattice constant for the alloy in the triclinic unit cell,

where R is the composition of relaxed μ-Fe₂O₃ in the alloy, $crelaxed=3.07Å$⁠, and $cstrained=3.16Å$⁠. If the lattice constants for both the strained and relaxed structures are known, this framework can be used to quantitatively determine the strain state of a thin film.

Figure 1 shows the simulated ω–2θ XRD patterns for the (020) diffraction condition from a ten-period superlattice structure with varying degrees of relaxation. The superlattice peaks (SL₀, SL_n + 1) shift as R is varied, indicating that this method is a good approximation to model the strain in these superlattices. As R approaches 100% (fully relaxed), the zeroth order superlattice peak SL₀ is blended with the (010) β-Ga₂O₃ substrate peak. LEPTOS simulations of this superlattice structure while varying the roughness of the layers did not show significant changes in the resulting XRD patterns.

XRD is performed on the samples using either a Bede diffractometer in triple axis mode or a Bruker D8 system. Reciprocal space maps (RSMs) in the symmetrical (020) (⁠$ω=30.3449°$⁠, $2θ=60.6898°$⁠) and asymmetric glancing exit (420) diffraction conditions (⁠$ω=61.7846°$⁠, $2θ=69.3916°$⁠) were obtained to investigate the strain-relaxation properties of the superlattices. The ω–2θ XRD patterns of the μ-Fe₂O₃ superlattices are fit using LEPTOS with a Genetic algorithm keeping the layer thicknesses and the composition of the strained/relaxed μ-Fe₂O₃ alloy as free parameters. The structural model used by LEPTOS consists of a β-Ga₂O₃ substrate and ten identical layers of μ-Fe₂O₃ and β-Ga₂O₃ that are constrained to have the same layer thicknesses and lattice constants across the superlattice structure. Instead of fitting the relaxation parameter available in LEPTOS, the composition R of the strained/relaxed μ-Fe₂O₃ alloy is fitted as it can be used as an approximation of the degree of relaxation in the superlattice [Eq. (1)]. The interface roughness is kept fixed at 0 as it showed no measurable impact on the simulated patterns.

The thickness of the μ-Fe₂O₃ layers range from ∼0.8 to ∼1.6 nm (∼5.0–10.2 ml) and the thickness of the β-Ga₂O₃ spacers range from ∼8.1 to ∼33.5 nm (∼52.2–220.4 ml). The composition R of the alloyed μ-Fe₂O₃ layers is ∼14.4%, which corresponds to a μ-Fe₂O₃ lattice parameter of b = 3.147 Å. However, the RSMs of μ-Fe₂O₃ superlattices (Fig. 2) show that the μ-Fe₂O₃ layers are fully strained to the β-Ga₂O₃ lattice. If the μ-Fe₂O₃ layers are fully strained, these measurements instead suggest that the experimentally measured b-lattice parameter of μ-Fe₂O₃ is b = 3.15 Å, which is in good agreement with the DFT prediction of b = 3.16 Å. The slight deviation from the DFT-predicted b-lattice parameter could also arise from off-stoichiometry or Ga/Fe intermixing in the Fe₂O₃ layers.

Figures 2(a) and 2(d) show the ω–2θ XRD patterns for two different μ-Fe₂O₃ superlattices grown at (Φ_Ga = 8.1 × 10⁻⁸ Torr, T_sub = 800 °C) and (Φ_Ga = 4.1 × 10⁻⁸ Torr, T_sub = 500 °C), respectively. The best fits from LEPTOS are shown in red. The fitting procedure provides excellent fits to the observed XRD data. While it is not possible to determine the interface roughness directly through the XRD fitting, atomic force microscopy (AFM) root-mean-square (RMS) measurements can be used as a proxy for the interface roughness. Figure 3 shows 1 × 1 μm² AFM images for these two superlattices. For the superlattice grown at T_sub = 800 °C, the RMS roughness is ∼0.3 nm (∼2 ml) while the superlattice grown at T_sub = 500 °C is ∼3.6× rougher with an RMS roughness of ∼1.1 nm (∼7 ml). This is consistent with the results from Hettiaratchy et al. who found that the RMS roughness of μ-Fe₂O₃ superlattices increases with decreasing T_sub.¹⁴ 

Figures 2(b), 2(e), 2(c), and 2(f) show the (020) and (420) RSMs for these two superlattices, respectively. Superlattice satellite peaks up to the sixth order are visible in the RSM data for the superlattice grown at T_sub = 800 °C. In the symmetrical (020) RSMs, the reciprocal lattice points (RLPs) for the superlattice peaks are vertically aligned, indicating a negligible lattice tilt within the heterostructures. The asymmetric (420) RSMs show that the RLPs for the superlattice peaks are vertically aligned to the (010) β-Ga₂O₃ substrate. This indicates that μ-Fe₂O₃ and β-Ga₂O₃ spacers in the superlattice are fully strained to the (010) β-Ga₂O₃ substrate and the superlattice growths are coherent and of good quality. The RSMs for the two superlattices differ in the amount of spread in the RLPs. The RLPs for the superlattice grown at T_sub = 800 °C are narrow, indicating few defects in the superlattice. In contrast, the RLPs for the superlattice grown at T_sub = 500 °C are significantly more spread out and indicate a greater number of defects in the superlattice structure. This is consistent with the differences in the surface roughness measurements from the AFM data. μ-Fe₂O₃ superlattices grown at T_sub > 700 °C are more coherent and of better quality than those grown at colder substrate temperatures.

Using the layer thicknesses determined from the LEPTOS fits, the μ-Fe₂O₃ growth rate can be determined as a function of T_sub. The average growth rate of μ-Fe₂O₃ as measured through XRD is ∼2.7 nm/min at T_sub = 500 °C and ∼2.5 nm/min at T_sub = 700 °C. At T_sub = 800 °C, the growth rate drops to ∼1.6 nm/min. Figure 4 shows the growth rate of μ-Fe₂O₃ and β-Ga₂O₃ normalized by the BEP. The reported BEPs are calibrated for N₂, so for a better comparison, the BEP of the Fe and Ga cells must be normalized by the ratio of masses $m(Fe/Ga)m(N2)$⁠.

The normalized growth rate of μ-Fe₂O₃ is generally larger than that of β-Ga₂O₃, which may indicate that Fe atoms have a higher sticking coefficient than Ga atoms. The growth rate of μ-Fe₂O₃ seemingly drops at T_sub = 800 °C. The thermal decomposition of Fe₂O₃ happens at much higher temperatures of about 1470–1570 °C,²⁷ so it is unlikely to cause reduced growth rates. However, Fe can diffuse substantially at interfaces and the rate of Fe diffusion is known to increase with temperature in various material systems.^28–30 Increased interdiffusion of Fe into β-Ga₂O₃ spacers at higher T_sub can explain the reduced growth rates. This will reduce the apparent thicknesses of μ-Fe₂O₃ layers as probed by XRD and lead to a reduction in calculated XRD growth rates. The calibration of μ-Fe₂O₃ growth rates using superlattices with very thin μ-Fe₂O₃ layers (∼few ML) is challenging. It may be possible to improve the μ-Fe₂O₃ growth rates by growing heterostructures with thicker μ-Fe₂O₃ layers.

This method of fitting the XRD patterns of monoclinic heterostructures with LEPTOS can also be applied to the monoclinic β-(Al_xGa_1−x)₂O₃ heterostructures. The ω–2θ XRD patterns of β-(Al_xGa_1−x)₂O₃ samples with varying compositions were fit with LEPTOS keeping the thickness of the β-(Al_xGa_1−x)₂O₃ layer and its composition x as free parameters. The β-(Al_xGa_1−x)₂O₃ layers are assumed to be fully strained when fitting the XRD data. β-(Al_xGa_1−x)₂O₃ layers with thickness up to ∼60 nm were found to be fully strained to the β-Ga₂O₃ lattice,¹⁹ and since the thickness of the β-(Al_xGa_1−x)₂O₃ layers investigated in this study is substantially smaller than ∼60 nm, that the β-(Al_xGa_1−x)₂O₃ layers are fully strained is a valid assumption.

Figures 5(a)–5(c) show the XRD data and the LEPTOS fits for β-(Al_xGa_1−x)₂O₃ sample grown with Al fluxes of 1.7 × 10⁻⁸, 1.5 × 10⁻⁸, and 1.3 × 10⁻⁸ Torr, respectively. The fits to the XRD data are excellent. The β-(Al_xGa_1−x)₂O₃ layer thicknesses are ∼24.7, ∼26.3, and ∼27.2 nm, and the Al compositions are $x=19.36%$⁠, $x=16.38%$⁠, and $x=13.74%$ at 1.7 × 10⁻⁸, 1.5 × 10⁻⁸, and 1.3 × 10⁻⁸ Torr, respectively. The Al composition x can also be estimated from a numerical fit to the separation $Δθ020$ of the (020) β-(Al_xGa_1−x)₂O₃ peak from the β-Ga₂O₃ substrate peak $θ020$⁠.²⁶ This relation is given as

Using Eq. (2), the Al compositions are x = 18.73%, x = 15.91%, and x = 12.78% at 1.7 × 10⁻⁸, 1.5 × 10⁻⁸, and 1.3 × 10⁻⁸ Torr, respectively. The Al compositions from the LEPTOS agree very well with those obtained from Eq. (2). The LEPTOS compositions are only ∼0.5%–1.0% smaller than those obtained from the analytical method. Oshima et al. validated their analytical estimate of the composition with atomic probe tomography (APT) and found that the APT results were consistent with the estimates given 2%–3% uncertainties of the APT measurements.²⁶ Additionally, the Al and Ga beam flux ratios were found to impact the derived Al compositions. For beam flux ratios $ϕAl/(ϕGa+ϕAl)$ of 0.061 and 0.104, the Al composition measured from APT were 12.0% and 16.8%, respectively.²⁶ The beam flux ratio for the growth with an Al flux of 1.7 × 10⁻⁸ Torr is $ϕAl/(ϕGa+ϕAl)=0.096$⁠, and a simple interpolation of the APT results gives an Al composition of ∼15.9% for this growth. The Al composition of ∼19.4% obtained from LEPTOS is ∼3.5% larger than the interpolated APT results of Oshima et al.²⁶ Considering the uncertainties in the analytical estimate, the Al compositions derived from the LEPTOS agree well with the estimates from Oshima et al.²⁶ 

The XRD patterns of β-(Al_xGa_1−x)₂O₃ superlattices were also fit using this method. The thickness of the β-(Al_xGa_1−x)₂O₃ layers and their composition were fit along with the thickness of the β-Ga₂O₃ spacers. Figures 6(a) and 6(d) show the XRD data and fits for two β-(Al_xGa_1−x)₂O₃ superlattices grown at T_sub = 800 and T_sub = 725 °C, respectively. For the superlattice grown at T_sub = 725 °C, the β-Ga₂O₃ layers are ∼31.5 nm thick and the β-(Al_xGa_1−x)₂O₃ layers are ∼4.6 nm thick with a composition of $x∼4.55%$⁠. For the superlattice grown at T_sub = 800 °C, the β-Ga₂O₃ layers are ∼13.5 nm thick and the β-(Al_xGa_1−x)₂O₃ layers are ∼2.0 nm thick with a composition of $x∼23.0%$⁠. The β-Ga₂O₃ growth rates from the XRD fits are as expected for the Ga beam flux and T_sub. The growth rate of β-(Al_xGa_1−x)₂O₃ decreases with increasing Al flux. Figures 6(b), 6(c), 6(e), and 6(f) show the (020) and (420) RSMs for the two β-(Al_xGa_1−x)₂O₃ superlattices grown at T_sub = 800 and T_sub = 725 °C. Superlattice peaks up to the fourth order are visible in the RSMs. Both the β-(Al_xGa_1−x)₂O₃ superlattices have no tilt and are fully strained and coherent. For the superlattice grown at T_sub = 725 °C, the SL₀ RLP is blended with that of the β-Ga₂O₃ substrate. The RLPs for the superlattice grown at T_sub = 725 °C are more spread out than those for the superlattice grown at T_sub = 800 °C, indicating that the superlattice quality improves at hotter growths.

The interdiffusion of Al atoms in β-(Al_xGa_1−x)₂O₃ heterostructures has previously been observed.^31,32 STEM data of an Al₂O₃/β-Ga₂O₃ interface showed that the interface was abrupt with a transitional region of ∼2 nm, and energy dispersive x-ray spectroscopy (EDS) found that the Al atoms are diffused within β-Ga₂O₃.³² Studies of Fe interdiffusion in Fe/GaAs, Fe/Si, and Fe/Al systems have shown considerable intermixing.^28–30 Secondary ion mass spectrometry (SIMS) results showed that Fe originating from a Fe-doped β-Ga₂O₃ substrate diffused up to ∼200 nm into a β-Ga₂O₃ buffer layer grown at T_sub = 700 °C.³³ 

To investigate the interdiffusion of Fe atoms in the μ-Fe₂O₃ superlattices, we obtain scanning transmission electron microscopy (STEM) high-angle annular dark field (HAADF) and electron energy loss spectroscopy (EELS) data of a superlattice grown at T_sub = 700 °C and Φ_Ga = 6 × 10⁻⁸ Torr with ∼1.1 nm μ-Fe₂O₃ layers and ∼15.9 nm β-Ga₂O₃ spacers. The EELS data were acquired using a K2 detector on a Titan3™ G2 60–300 S/TEM at 300 kV with a pixel time of 0.25 s. At the interface between a μ-Fe₂O₃ layer and a β-Ga₂O₃ spacer shown in Fig. 7(a), Fe atoms diffuse substantially in the (010) growth direction, resulting in a dimmer contrast. Figures 7(b) and 7(c) show EELS data of the Fe-L edge signals near this interface. The concentration of Fe atoms near the interface is asymmetric indicating the asymmetric diffusion of Fe atoms into the β-Ga₂O₃ spacer layers. The Fe-L edge signal spans ∼8 nm, whereas the XRD thickness of the μ-Fe₂O₃ layer is ∼1.1 nm, suggesting the presence of Fe atoms in ∼6.9 nm of β-Ga₂O₃. The ordered interface phase of (Fe_octGa_tet)₂O₃ with Fe occupying octahedral sites and Ga occupying tetrahedral sites is predicted to be ferromagnetic from DFT calculations.¹³ Hettiaratchy et al. found that increasing the interface roughness resulted in an increased ferromagnetic response from the superlattices, which supports the DFT prediction.¹⁴ Additionally, the diffusion of Fe atoms in the β-Ga₂O₃ spacer layers produces a paramagnetic component in the observed magnetic response of the superlattices.^13,14

Superlattices of μ-Fe₂O₃ or β-(Al_xGa_1−x)₂O₃ and β-Ga₂O₃ layers are grown on (010) β-Ga₂O₃ substrates using PAMBE. A method to adapt commercial dynamical x-ray diffraction software to quantitatively fit the HRXRD patterns of monoclinic heterostructures is presented. The layer thicknesses, strain-relaxation properties, and alloy compositions of the β-(Al_xGa_1−x)₂O₃ and μ-Fe₂O₃ heterostructures were obtained by fitting XRD data. The RSMs in the symmetric (020) and asymmetric (420) diffraction conditions for μ-Fe₂O₃ and β-(Al_xGa_1−x)₂O₃ superlattices indicate coherent growths that are fully strained to the (010) β-Ga₂O₃ lattice. Superlattices grown at hotter substrate temperatures result in fewer crystal defects and better coherency. The growth rate of μ-Fe₂O₃ is ∼2.6 nm/min at T_sub = 700 °C and drops to ∼1.6 nm/min at T_sub = 800 °C due to increased Fe interdiffusion at hotter substrate temperatures. STEM and EELS data of an interface in a μ-Fe₂O₃ superlattice grown at T_sub = 700 °C confirm that Fe atoms diffuse significantly into β-Ga₂O₃ spacers.

This work was supported by the Center for Emergent Materials at The Ohio State University, an NSF MRSEC (No. DMR-1420451) and by the Army Research Office MURI (No. W911NF-14-1-0016). B.W. acknowledges support from the Presidential Fellowship of the Ohio State University. Electron microscopy experiments were supported by the Center for Electron Microscopy and Analysis at the Ohio State University

The authors have no conflicts to disclose.

Elline C. Hettiaratchy: Formal analysis (lead); Investigation (lead); Writing – original draft (lead); Writing – review & editing (lead). Binbin Wang: Investigation (supporting). Ashok Dheenan: Investigation (supporting). Joe McGlone: Investigation (supporting). Nidhin Kurian Kalarickal: Investigation (supporting). Núria Bagués: Investigation (supporting). Steven Ringel: Project administration (supporting); Supervision (supporting). David W. McComb: Supervision (supporting). Siddharth Rajan: Investigation (supporting); Project administration (supporting); Supervision (supporting). Roberto C. Myers: Conceptualization (lead); Funding acquisition (lead); Investigation (supporting); Project administration (lead); Supervision (lead); Writing – original draft (supporting); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.