Experimental and theoretical investigation of the formation of two-dimensional Fe silicate on Pd(111)

Two-dimensional (2D) nanosheets, consisting of one to several atomic layers grown on another substrate (e.g., epitaxial metal films),¹ are considered as building blocks for assembling multifunctional materials with well-defined architectures.^2–5 Reducing the material’s thickness down to the atomic scale may induce a wide range of unique electronic, magnetic, optical, and chemical properties, which makes the 2D materials promising candidates for a variety of different applications and technologies.^2,6–9 Most 2D materials are derived from so-called “van der Waals (VDW)” solids,^6,10,11 i.e., layered solids with strong, mostly covalent, intralayer but weak VDW interlayer bonding. If supported on a solid substrate, as is often necessary for applications, these “VDW-type” 2D layers feature weak overlayer-substrate interactions. Inspired by recent successful preparations of well-ordered 2D silica films on Ru(0001),¹² Pt(111),¹³ Pd(111),¹⁴ and Pd(100)¹⁵ single crystal surfaces, SiO₂ bilayers with mirror image planes of corner-sharing SiO₄ tetrahedra have been added to the family of 2D materials.¹⁶ These SiO₂ bilayers have provided a template for preparing 2D aluminosilicates which mimic the interior surfaces of zeolites, industrially important nanoporous aluminosilicate catalysts and sorbents whose relevant surfaces have been difficult to study.¹⁷ Moreover, several families of hypothetical zeolites based on these silicate bilayers have been proposed which raises the possibility of creating new materials with vanishingly low densities, extremely wide channels, and tunable pore diameter and shape.^18,19 Extending this breakthrough to the preparation of well-defined 2D silicates incorporating di- and trivalent cations is an important prerequisite for a deeper understanding of surface reactions in geochemistry and catalysis over silica and related materials. In this paper, it will be shown that 2D Fe silicate can be grown on Pd(111) and that this material can potentially be hydrogenated to facilitate its removal from the growth substrate.

Over the last several years, single sheets of transition metal (TM) silicates incorporating Fe and Ti have been grown on Ru(0001), while 2D Ni silicate has been observed on the surface of Ni-Pd(111) alloys.^20–24 The dioctahedral silicate structures with a 1:1 TM:Si ratio resemble dehydrated kaolinite with a layer of six-membered rings of corner-sharing SiO₄ tetrahedra bound to a layer of TMO₆ or TMO₅ polyhedra.^25,26 interestingly, the dioctahedral Ti and Ni silicate layers have no bulk analogs. Moreover, the inherent lack of inversion symmetry across the layer guarantees a piezoelectric response, and, depending on the TM and the TM:Si ratio, the materials may be ferromagnetic.²⁶ In particular, ferromagnetic order has been observed in bulk iron silicates, including magneto-electric Ba₃NbFe₃Si₂O₁₄ and the 2D layered Fe silicate mineral greenalite.^27–30 Thus, 2D transition metal silicates have the potential to add new functionality to the toolkit of 2D layers that may be used to develop new technologies by stacking complementary atomic layers.^31–33 

There are many open questions that must be addressed before the potential of 2D transition metal silicates may be realized. These include the sensitivity of the growth and the resulting structure to the nature of the substrate, and the ability to weaken the interaction with the substrate sufficiently to release the layer from it. Recently, it was recognized that the substrate can play an important role in the resulting 2D silica and silicate structures.^24,33 Regarding the interaction with the substrate, the dehydrated 2D TM silicate layers are electron deficient and are considered to be stabilized by electron donation from the metal substrate.^21–23 The net result is chemical bonding to the substrate that can make it difficult to remove the 2D layers from the substrate. Hydration or hydrogenation of the 2D TM silicate would appear to be a route to pacify the bonds to the substrate but has not been investigated. This chemistry is also expected to depend on the substrate. To address these questions, we have investigated 2D Fe silicate formation on Pd(111) experimentally and theoretically. The nonmagnetic Pd substrate was chosen to facilitate magnetic property measurements and to weaken the interaction with oxygen compared to Ru(0001), which has been suggested to play an important role in 2D silicate structure formation.³³ Experiments indicate the formation of a well-ordered honeycomb-structured surface layer with the Fe–O–Si bonds expected in a 2D Fe silicate. Meanwhile, density functional theory (DFT)^34,35 identifies the most stable structure in a hydrogen free environment as a 2D Fe₂Si₂O₈ layer atop a chemisorbed O-covered Pd(111) surface. Theory indicates that this layer can readily bind four H atoms which substantially weakens the bond to the substrate, thus offering promise that 2D Fe silicate grown on Pd(111) can be isolated as a single VDW layer.

The experiments were performed using a UHV system equipped with a high-speed variable temperature scanning tunneling microscope,³⁶ a double-pass cylindrical mirror analyzer for Auger electron spectroscopy (AES) measurements, and reverse-view low-energy electron diffraction (LEED) optics, as well as the usual facilities for sample manipulation, surface cleaning, and thin film evaporation. The sample temperature was measured using a type K thermocouple fixed to the sample surface and was cross-checked with an infrared pyrometer. All scanning tunneling microscopy (STM) images were recorded at room temperature in constant current mode (Pt/Ir tips made by the cut-and-pull method) at positive sample biases and with feedback current set between 0.05 and 1.0 nA during scanning. Iron atoms were evaporated from a commercial electron beam evaporator (McAllister Technical Services), while the silicon oxide was deposited via sublimation of SiO granules from an effusion cell (DCA Instruments) at 1300 K. The evaporation flux (0.15–1.2 ng s⁻¹ cm⁻²) was monitored with a water-cooled quartz crystal microbalance (Inficon). The evaporated amount of Fe and silicon oxide is given in monolayers (ML) of Fe atoms and SiO units, whereby 1 Ml is defined as 4.11 × 10¹⁴ SiO/cm², a coverage sufficient to form a sheet of corner-sharing SiO₄ tetrahedra.

Core-level x-ray photoelectron spectra (XPS) were recorded after transferring the sample through air to a PHI VersaProbe II XPS system with a monochromatic Al Kα x-ray source. A Shirley background was subtracted from all core-level spectra. The fitting was performed using a mixed Lorentzian–Gaussian line shape with the Gaussian percentage held at 25% which was found to consistently give the best fits. One complication with studying oxygen on Pd substrates with XPS is the partial overlap of the O 1s and Pd 3p_1/2 peaks. For the O 1s spectra shown below, the contribution of the Pd 3p_1/2 peak was subtracted by also analyzing the Pd 3d_5/2 peak and assuming a fixed Pd 3p_1/2 to Pd 3d_5/2 peak integral ratio (based on the bare surface spectrum) and that the Pd support peak was unaffected by the 2D Fe silicate overlayer and adsorption at the interface.

Polarization modulation reflection-absorption infrared spectroscopy (RAIRS) data were recorded using a Thermo Fisher Nicolet iS50 FTIR spectrometer with the incident light at a grazing angle of 81° in a dry nitrogen environment. Following the ex situ measurements, the sample was returned to the UHV system where the sample was grown and initially characterized; and after heating at 925 K in 2 × 10⁻⁶ Torr O₂, no observable changes in the film composition or structure could be detected by AES, LEED, or STM.

All DFT calculations were performed using the plane-wave Vienna ab initio Simulation Package (VASP),^37,38 using the Perdew–Burke–Ernzerhof exchange-correlation functional with D3 semiempirical vdW correction (PBE + D3).^39,40 A Hubbard-U value of 5.3 eV was used on all the d-orbitals of Fe atoms; and this value is known to provide very accurate formation enthalpies in binary iron-oxides,⁴¹ hence it can be reasonable for surface interactions. We used a kinetic energy cutoff of 520 eV, periodic boundary conditions, projector-augmented wave (PAW) pseudopotentials,⁴² and dipole correction.⁴³ A Gaussian smearing of 0.2 eV was used to better capture the metallic nature of the system together with a dense reciprocal grid density of 6 × 6 × 1. Geometry relaxations were performed until the forces on all atoms were smaller than 10⁻² eV/Å with an electronic energy convergence of 10⁻⁵ eV. Each simulation cell was constructed from a 2 × 2 supercell of the Pd(111) surface primitive cell and a vacuum separation between slabs of more than 15 Å. Four layers of Pd atoms were used as the substrate, with the bottom layer atomic positions fixed, but all the remaining atoms geometry-optimized. A lattice parameter of 3.92 Å was obtained, as opposed to the experimental value of 3.89 Å.⁴⁴ This bulk cell was used to generate the substrate, and no further cell optimization was performed. All the calculations were carried out with spin-polarization due to inclusion of Fe. All Fe atoms were initialized in the high-spin ferromagnetic configuration.

Harmonic vibrational frequencies at γ were calculated using the frozen phonon method using central differences with 0.015 Å displacements along cartesian directions. Zone-center phonon modes were obtained from the eigenvalues of the dynamical matrix. Since the experimental RAIRS method is only sensitive to changes in the dipole moment perpendicular to the substrate surface. Therefore, simulated RAIRS intensities were obtained from the changes in the square of the dipole moments along the z axis at each phonon mode.⁴⁵ The vibrational frequencies were scaled with a factor of 1.0341 derived from a comparison between the theoretical and experimental vibrational frequencies of α-quartz.¹² 

Gibbs free energies of hydrogenation reactions were obtained using a reference chemical potential for H₂ gas using the relation as $Fe2Si2O9/Pd(111)+n2H2→Fe2Si2O9H4/Pd(111)+ΔG$⁠, where $ΔG$ is the Gibbs free energy of reaction. Using the PBE + D3 functional, H₂ gas has the total energy of $EH20,DFT=−6.774eV$ in VASP, using the H PAW pseudopotential from VASP. Using the formula of $μHref(T)=1/2(EH20,DFT−TSH2exp(T))$⁠, where $SH2exp(T)=130.680J/mol×K$⁠,⁴⁶ we obtain a reference $μHref(T)$ of −3.571 at 298 K. Because the contributions of vibrational entropies of surface complexes and reference solids are typically similar in magnitude to the computational uncertainties,⁴⁷ the entropic contributions to film/substrate complexes are assumed to be zero.

The Fe silicate was prepared by first depositing 1 Ml SiO, then depositing enough Fe to obtain a 1:1 Fe:Si ratio, both in UHV. The sample was then annealed in 2–5 × 10⁻⁶ Torr O₂ to successively higher temperatures until a well-ordered LEED pattern was obtained. We found that order emerged near 900 K and that the patterns did not significantly change past 950 K. With the assistance of STM and LEED, it was found that the Fe silicate on Pd formed under these conditions is a highly crystalline 2D overlayer with a low-defect density. Figures 1 and 2 reveal details of the Fe silicate structure on Pd(111). The pristine Pd(111) surface shows a characteristic LEED pattern of the clean surface with a hexagonal unit cell as indicated in Fig. 1(a). The LEED pattern of the Fe silicate/Pd(111) surface in Fig. 1(b) features sharp reflections confirming an overall long-range order. The pattern consists of six main diffraction spots surrounded by rosettes of satellite spots which are characteristic of a moiré structure commonly observed for relaxed oxide layers with a lattice mismatch to the support.⁴⁸ On the other hand, the unit cell of the oxide overlayer is rotated by 30° with respect to that of the Pd(111) surface [see Figs. 1(a) and 1(b)], this behavior is similar to previously investigated Fe silicate on Ru(0001).²² Figure 1(c) shows a topographic constant current large-scale STM image of Fe silicate bilayer on Pd, in which the Pd(111) surface is completely covered by a well-ordered and atomically flat wetting oxide layer. It is notable that the STM image presents flat terraces separated by ≈2.25 Å high steps corresponding to monoatomic Pd(111) steps underneath the Fe silicate layer. The terraces exhibit a moiré pattern with a ≈46 Å periodicity, consistent with an incommensurate overlayer. The schematic of the moiré pattern [Fig. 1(c), inset] reproduces the experimental findings nicely.

The periodic lattice visible in the STM image [Fig. 2(a)] is not the atomic Fe silicate lattice but, as indicated above, a moiré superlattice of much larger periodicity. Figure 2(b) presents a zoomed in STM image, where the atomic structure of the Fe silicate (lattice parameter a = 5.35 ± 0.1 Å) is clearly observed and the moiré pattern is also apparent. The lattice constant is within the range expected for layered silicates. The higher resolution image in Fig. 2(c) shows a honeycomb structure consistent with the atomic structure of the top layer of six-membered rings of corner-sharing SiO₄ tetrahedra, as well as prior results for 2D Fe, Ti, and Ni silicates.^21–23 As mentioned above, the moiré pattern results from the superposition of the Pd(111) substrate (a = 2.75 Å) and oxide layer (a = 5.35 ± 0.1 Å) lattices. Despite the large 4.6% lattice mismatch, the Fe silicate layer is nearly defect-free and extends over a large surface area.

The RAIRS spectrum in Fig. 3(a) shows a sharp single peak centered at 1050 cm⁻¹. This frequency is characteristic of the stretching vibration of the Si–O–TM interlayer linkage, a mode that is common to all 2D TM silicate layers.^20–24 Notably, the spectrum does not show any vibrational absorption around 1300 cm⁻¹, which is a benchmark for the collinear Si–O–Si bonds in 2D silica bilayers.^16,24 Additionally, the bulk quartzlike vibrational stretching modes between 1100 and 1260 cm⁻¹ were absent.⁴⁹ Thus, the vibrational absorption around 1000 cm⁻¹ is usually assumed to be experimental proof of the metal incorporation into SiO_X films to form TM silicate layers.²⁴ The conclusion that we have formed 2D Fe silicate is reinforced by the comparison of the experimental spectrum with the computed spectrum for the favored structure comprised an Fe₂Si₂O₈ layer with oxygen chemisorbed at the interface; and this structure is detailed in Sec. III. Note that the predicted peak below 700 cm⁻¹ is not seen in the experiment because the polarization modulator used in the experimental setup has a low-energy cutoff just below 900 cm⁻¹.

Further information on the structural and electronic properties of the Fe–O–Si 2D phase can be obtained from photoelectron spectra. Figures 4(a)–4(c) display XPS of the Fe 2p, Si 2p, and O 1s core levels, respectively. In Fig. 4(a), the Fe 2p core-level spectrum displays a characteristic Fe 2p line shape, which can be fitted with four core-level components, 2p_3/2–2p_1/2 doublet (separated by a spin–orbit splitting of 13.1 eV) plus two peaks attributable to satellite peaks in Fe oxides.^50,51 The Fe 2p_3/2 core-level emission peak occurs at a binding energy of 711.4 eV and the satellite peak at ≈7 eV higher energy. The position of the main peaks and the energy difference between the main peak and the satellite peaks depend on the oxidation state with intensity of the satellite peaks also dependent on the crystal structure.^52,53 The main 2p_3/2 line appears near 711 eV in Fe³⁺ compounds with the satellite feature 7–8 eV higher in energy, while the same peak appears closer to 709 eV in Fe²⁺ compounds with the satellite peak 6 eV or less higher in energy.⁵³ Thus, the observed 2p line shape and binding energies indicate that all of the Fe in the layer is 3+. The Si 2p XPS spectrum [see Fig. 4(b)] exhibits a peak due to the closely spaced 2p_3/2 and 2p_1/2 levels at 103.5 eV. This value is typical for the 4+ oxidation state of Si in SiO₂ in which each Si atom is surrounded by four O atoms in SiO₄ tetrahedra.⁵⁴ Taken together, the analysis of the Fe 2p and Si 2p core-level spectra show that the Fe/Si ratio is ≈1. In addition, the take-off angle-dependent XPS data [Fig. 4(d)] show the relative changes in the Fe 2p and Si 2p intensities in moving from normal emission (θ = 0°) to the more surface sensitive 60° grazing emission angle. The increase (decrease) of the Si 2p (Fe 2p) signals with the emission angle is a clear indication that Si atoms are involved in the surface layer of the Fe–O–Si phase, whereas the Fe atoms are buried near the metal-oxide interface. The O 1s core-level spectrum [Fig. 4(c)] appears broad. Under the experimental conditions, exposure to atmospheric air, oxygen is expected to adsorb at the interface between 2D silica and silicates on Ru and Pd surfaces.^22,54,55 Therefore, the O 1s spectrum was fit to two spectral components at 530 and 531.4 eV binding energy with the position of the lower energy peak fixed at the value seen for oxygen adsorbed at the interface between 2D silica and Pd(111) [note that the O 1s binding energy of the Pd-(2 × 2)O surface is at 529.9 eV].⁵⁶ The quality of the fit with two peaks validates the presence of at least two inequivalent O atoms in the structure.

The iron silicate on the Pd(111) system was studied using density functional theory (DFT) to characterize its structure and identify possible hydrogenation sites that can affect its interaction with the substrate. The experiments indicate that Fe silicate is incommensurate with Pd(111) with a ≈46 Å coincidence cell that would be prohibitively large to model with DFT. Hence, we used a pseudoepitaxial structure with a 2 × 2 superstructure of the Pd(111) primitive unit cell, as has previously been used to model incommensurate 2D Fe silicate on Ru(0001) and 2D Ni silicate on Ni-Pd(111).^22,23 Although the experiments indicate a 30° rotation of the overlayer, the 2 × 2 superstructure without rotation gives a closer lattice match in a modest-sized unit cell, and the incommensuration indicates that the bonding to the substrate is not strongly sensitive to the registry with the substrate. Results described below confirm the latter point. Nevertheless, the pseudoepitaxial structure would mean there is a finite amount of artificial strain involved, ≈4.6% tensile. We expect that this will influence the hydrogen binding and exfoliation energies. Our future work will use the equation of states in vacuum to understand how much this factor affects the reported energies.

Because the experimental conditions nominally involve only Fe, Si, O, and Pd, we start by considering the fully dehydrogenated dioctahedral silicate Fe₂Si₂O₉ atop Pd(111). We find that one of the oxygens prefers to transfer to the Pd substrate to form a chemisorbed O atom occupying a threefold hollow site, similar to O on bare Pd(111); and this leaves an Fe₂Si₂O₈ layer atop an oxygen-covered surface. This finding is similar to previous observations for Fe silicate on Ru(0001).²² In keeping with prior convention, we refer to this system as Fe₂Si₂O₈ O/Pd(111).²² Figures 5(a) and 6(a) show the optimized geometry for this system which was obtained by sliding the Fe silicate layer over the substrate to ensure that the most energetically favorable docking between the layer and the substrate was found. With a 3 × 3 lateral slide over the substrate, the energy varied by only 0.25 eV per formula unit, indicating that the registry has only a small effect. This result instills confidence that the analysis of the more compact unit cell without the 30° rotation remains relevant to the experimental system. Consistent with experiment, the network of six-membered rings of corner-sharing SiO₄ tetrahedra face the vacuum with Fe closer to the Pd substrate. The chemisorbed O can also account for a lower binding energy component of the O 1s XPS as detailed above.64,65 Lastly, note that Fig. 6(a) shows Fe–O–Pd bonds; and as discussed in Sec. I, this can be associated with charge transfer to the silicate layer to compensate the charge deficiency induced by removing the hydrogens that are normally found in these layered silicates.

Motivated by both the need to consider the most common impurities in UHV systems, H₂ and H₂O (in addition to CO), and the potential to passivate the silicate layer bonds to the Pd surface, we investigated hydrogenation of the Fe₂Si₂O₈∙O/Pd(111) system. Previously, we showed that oxygens bound to Si are not energetically favorable sites for binding hydrogen for isolated TM silicate layers.²⁶ A total of five O atoms are bonded to Si atoms, leaving three potential H binding sites within the Fe₂Si₂O₈ layer. The adsorbed oxygen was also considered as a potential hydrogen binding site, bringing the total to four possible hydrogenation sites. To study hydrogenation, we first introduced a single hydrogen and investigated its binding energies on these four possible sites. With a single hydrogen, the preferred binding site is Fe–O–Fe bridging oxygens, as pictured in Figs. 5(b) and 6(b); and isolated layers behave similarly. Subsequent H atoms attach to Fe–O–Pd bridging oxygens but with unequal energies [Figs. 5(c)–5(e) and 6(c)–6(e)]. In contrast, the H binding energies for these three sites are identical for isolated layers.²⁶ 

The Fe–O–Fe bridging oxygen is 3.80 Å from its nearest Pd atom; and thus, this first H atom is not expected to significantly weaken the adhesion to the substrate. Therefore, increasing numbers of hydrogens were introduced to this structure up to n = 4 to study the Fe₂Si₂O₉H_n/Pd(111) system. We find that adding more hydrogen modifies the shortest Pd–O bond length, d_Pd–O, found in the structure; and d_Pd–O is then used as a metric to compare the exfoliation potential of different structures. In Fe₂Si₂O₈ ∙ O/Pd(111), the adsorbed oxygen has an average d_Pd–O of 1.98 Å. This oxygen has three Pd nearest neighbors, two of which are nearly equidistant. The bond length closely matches the 1.99 Å reported for O on bare Pd(111).⁵⁷ In the Fe₂Si₂O₉H/Pd(111) structure, in Fig. 6(b) two important changes can be seen upon hydrogenation. The adsorbed oxygen moves from its original fcc hollow site to a bridging site between two Pd atoms and effectively joins the Fe silicate film. In its new position, the shortest d_Pd–O is 2.01 Å. Hence, O/Pd(111) is no longer used to identify the substrate, and the system is now denoted as Fe₂Si₂O₈H/Pd(111). Although d_Pd–O did not significantly increase, the coordination of the oxygen to the substrate was substantially modified. Overall, we find that d_Pd–O increases monotonically as n increases from 0 to 4, going from 1.98, 2.01, 2.04, 2.07, and finally to 2.24 Å. The substantial increase in going from n = 3 to n = 4 is interesting since n = 4 corresponds to Fe₂Si₂O₉H₄ which puts both Fe atoms in the 3+ oxidation state and matches the kaolinite stoichiometry. In Fig. 7, we show the binding energy curve of Fe₂Si₂O₈H₄, where the maximum binding energy is found to be $46meV/Å2$⁠. According to Mounet et al., exfoliation energies between $35and130meV/Å2$ are considered to be “potentially exfoliable.”⁵⁸ Thus, the results suggest that complete hydrogenation is required to weaken the substrate interaction and enter the VDW exfoliation regime.

To assess how difficult it would be to hydrogenate the 2D Fe silicate grown on Pd(111), we calculated the Gibbs free energy of the hydrogenation reactions (ΔG) at room temperature assuming a H₂ gas reservoir. In Fig. 8, ΔG is plotted as a function of hydrogen chemical potential (μ_H) which is proportional to the logarithm of the H₂ partial pressure. Notably, the fully hydrogenated state is strongly favored at atmospheric pressure (μ_H = 0) and remains so well into the vacuum range. Thus, hydrogenation is feasible under mild conditions.

A single 2D layer of Fe silicate with a 1:1 Fe:Si ratio was prepared by the controlled solid state reaction of SiO and Fe under UHV conditions. Characterization by complementary techniques sensitive to the composition, oxidation state, bonding, and reciprocal and real space structure all indicate a single, well-ordered phase with Fe in the 3+ oxidation state. Both LEED and STM reveal that the layer is incommensurate with the substrate and, thus, relaxes to its favored lattice constant, 5.35 ± 0.1 Å, in the range expected for layered transition metal silicates. In the presence of only Fe, Si, and O above the Pd surface, theory identified an Fe₂Si₂O₈ sitting atop above an oxygen-covered Pd surface as the lowest energy structure. This layer is composed of a sheet of six-membered rings of corner-sharing SiO₄ tetrahedra bound to a sheet of fivefold coordinated Fe atoms. Consistent with experimental observations, the corner-sharing SiO₄ tetrahedra face the vacuum while the FeO₅ polyhedra bond to the Pd surface. The effect of hydrogen on the structure and the bonding of the 2D layer to the substrate were investigated theoretically. The results indicate that even a single H atom is sufficient to favor shifting the oxygen adsorbed on the Pd surface to the silicate layer, creating Fe₂Si₂O₉H with the Fe octahedrally coordinated to six oxygens. At room temperature and H₂ pressures extending into the vacuum regime, Fe₂Si₂O₉H₄/Pd(111) with a substantially expanded minimum O–Pd distance is favored. The silicate layer adopts the single-layer kaolinite structure with Fe in a formal 3+ oxidation state. The combined experimental and theoretical results demonstrate that a well-defined 2D single-layer transition metal silicate can be straightforwardly prepared and that hydrogenation can be a viable route to passivate the layer’s bonds to the substrate to allow its isolation.

The authors acknowledge the contribution of Min Li of the Yale West Campus Materials Characterization Core to carrying out the XPS measurements. The Yale Center for Research Computing is thanked for its support of the computing infrastructure. This work also used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation (Grant No. ACI-1548562), by using computer time on the Comet and Expanse supercomputers as enabled by XSEDE allocation MCA08X007. This work was supported by the US Army Research Office under Grant No. W911NF-19-1-0371.

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