Utilizing the powerful combination of molecular-beam epitaxy (MBE) and angle-resolved photoemission spectroscopy (ARPES), we produce and study the effect of different terminating layers on the electronic structure of the metallic delafossite PdCoO2. Attempts to introduce unpaired electrons and synthesize new antiferromagnetic metals akin to the isostructural compound PdCrO2 have been made by replacing cobalt with iron in PdCoO2 films grown by MBE. Using ARPES, we observe similar bulk bands in these PdCoO2 films with Pd-, CoO2-, and FeO2-termination. Nevertheless, Pd- and CoO2-terminated films show a reduced intensity of surface states. Additionally, we are able to epitaxially stabilize PdFexCo1−xO2 films that show an anomaly in the derivative of the electrical resistance with respect to temperature at 20 K, but do not display pronounced magnetic order.
Metallic oxides with the delafossite structure, shown in Fig. 1(a), have drawn significant attention in recent years due to their unique structural and electronic properties. Examples include PtCoO2, which has the highest conductivity per carrier of all materials, and PdCoO2, which has the longest mean free path (exceeding 20 μm at 4 K) among all known metals.1–3 The in-plane electrical conductivity of PdCoO2 at room temperature, which is about four times higher than that of palladium metal itself, has been argued to arise from electron–phonon scattering mainly occurring within a single, closed, highly dispersive band of primarily palladium character at the Fermi level (EF).1,4–8 The large spin-splitting of the surface state arising from the CoO2 termination, in combination with the layered structure of PdCoO2-based heterostructures makes this system ideal to study itinerant surface electrons driven by inversion-symmetry breaking.9 As for the magnetic properties of delafossites, PdCrO2 is the only highly conducting delafossite material that orders magnetically; it orders antiferromagnetically (AFM) at around 37 K. Focusing on the electronic structure, the single band at the Fermi level with palladium character forms a reconstruction driven by the AFM order from the adjacent CrO2 layer.10–14 Comparing AFM PdCrO2 with nonmagnetic PdCoO2, the spins from Cr3+ interacting inside the CrO2 layer with the palladium monolayers on either side of the CrO2 layer play a critical role in the magnetic state of PdCrO2.13
Angle-resolved photoemission spectroscopy (ARPES) is the premiere experimental tool to directly observe electronic structure in quantum materials. The combination of oxide molecular-beam epitaxy (MBE) with ARPES allows us to customize and study the electronic structure of correlated oxides. This setup has enabled understanding of how strain,15,16 thickness,17–19 hetertostructuring,20,21 interfaces,22,23 and terminations24–29 can influence the electronic structure of thin films.
Due to the limited size of delafossite single crystals, the desire to explore the potential of metallic delafossites in electronic30 and spintronic devices,31 together with the exciting opportunities that can be explored in delafossite heterostructures, metallic delafossites are being grown in thin-film form by reactive sputtering,32 pulsed-laser deposition (PLD),30,33–36 and MBE.37,38 Unfortunately, the transport properties of these films (so far) pale in comparison to the single crystals. Differences between the Fermi surface of the so-called Pd-terminated PdCoO2 with CoO2-terminated PdCoO2 have recently been reported for epitaxial films of PdCoO2 grown by PLD.39 The claims of this prior study would be strengthened by improved evidence of surface termination control.
In this work, we describe an improved synthetic strategy for the growth of PdCoO2 films with control of surface termination by MBE. Harnessing the ultra-high vacuum connection between our MBE and ARPES, we then study the electronic structure of Pd-terminated and CoO2-terminated PdCoO2. We find that our PdCoO2 films exhibit similar bulk bands derived from palladium states but weak surface states compared to those in PdCoO2 single crystals. Having succeeded in engineering the surface termination in PdCoO2 over large areas, we then progress to investigate terminations of PdCoO2 where we deliberately add unpaired electrons and study the resulting electronic structure by ARPES. Although we are able to epitaxially stabilize a variety of surface terminations involving iron substituting for cobalt in PdCoO2, we do not see evidence of magnetic order.
Building upon our previous work,38 thin films of PdCoO2 were synthesized by MBE in a Veeco GEN10 MBE system on (001) sapphire substrates. Details of the film growth are provided in the supplementary material. Figure 1(a) shows the shutter timing diagram used to supply fluxes of the individual molecular beams to the substrate to form PdCoO2. After growth, films were cooled down to 300 °C in the same ozone background pressure (around 5 × 10−6 Torr) in which they were grown and transferred under ultra-high vacuum conditions into an adjacent ARPES chamber. The reflection high-energy electron diffraction (RHEED) patterns acquired after deposition and the x-ray diffraction θ–2θ scans indicate the growth of single-phase PdCoO2 films as shown in Figs. 1(b) and 1(c). The structure was characterized by a Panalytical Empyrean x-ray diffractometer utilizing Cu Kα1 radiation. Electrical transport measured by a Quantum Design Physical Property Measurement System (PPMS) using a four-point van der Pauw geometry is shown in Fig. 1(d). The residual resistivity ratio (RRR = ρ300K/ρ4K) of this PdCoO2 sample with a thickness of 52.1 nm is 3.3 in its as-grown state (i.e., no ex-situ postanneal). For comparison, the highest RRR previously achieved by MBE for a film in its as-grown state was 2.2 for a 10 nm thick PdCoO2 film.38 After an ex-situ anneal, films grown by MBE can reach a RRR of around 8 for a 50 nm thick PdCoO2 film.37 While these are the highest reported RRR values for films, PdCoO2 single crystals can exhibit RRR as high as 400.1 The resistivity of epitaxial PdCoO2 films and single crystals are comparable at room temperature; the huge difference in resistivity emerges upon cooling. One likely defect responsible for this difference is the in-plane rotation twins present in all epitaxial delafossite films to date. The presence of 180° in-plane rotation twins in PdCoO2 film grown on (001) Al2O3 substrates manifest in the x-ray ϕ scan and scanning transmission electron microscopy images shown in Ref. 38, as well as the atomic force microscopy images shown in the supplementary material (Fig. S3).
In-situ ARPES measurements are performed to study the effects of termination on the electronic structure of the PdCoO2 films. Our lab-based ARPES system photoexcites electrons with a Scienta omicron VUV 5000 helium discharge lamp using He–I photons at 21.2 eV and He-II photons at 40.4 eV. The emitted electrons are detected with a VG Scienta R4000 electron analyzer. The ARPES is vacuum-connected to the MBE growth chamber via an ultra-high vacuum transfer chamber.
We first compare the Fermi surface of Pd-terminated and CoO2-terminated PdCoO2 films in Figs. 2(a) and 2(b), respectively. The sharp hexagonal pocket centered at Γ [illustrated in red in Figs. 2(a) and 2(b)] is observed in both Pd- and CoO2-terminated PdCoO2 films. Two smaller hexagonal pockets inside the bulk state pocket illustrated in green and blue are observed in the CoO2-terminated PdCoO2 film, in agreement with previous reports of splitting of the CoO2 surface state driven by spin–orbit coupling.40,41 For Pd-terminated PdCoO2, we do not observe pronounced palladium surface states at EF. Below EF, however, there is some spectral weight possibly from the palladium surface state as described in Ref. 41. At He-II photon energy (40.4 eV), we observe stronger spectral weight below EF, as illustrated in Fig. S5 of the supplementary material. This intensity below EF might be related to the palladium surface state, but at this higher energy, still no palladium surface state appears at EF.
In Figs. 3(a) and 3(c), we further compare the dispersion cut along the Γ-K direction of Pd- and CoO2-terminated PdCoO2, respectively. The fitted Fermi velocities (vFs) of the PdCoO2 bulk state of Pd- and CoO2-terminated films are all around 4.5 eV Å, as shown in Table I, in agreement with previous results measured on PdCoO2 single crystals by ARPES.9,13 Despite the invisible palladium surface state at EF, two spin-split surface states from the CoO2 termination show up at EF. These are indicated by blue and green circles in Fig. 3(c). From the momentum dispersion curves (MDCs) comparison of Pd- and CoO2-terminated samples in Fig. 3(e), it is easier to observe the CoO2 surface states at EF (indicated by the blue arrows) and very little of the palladium surface state is observed below EF (indicated by the red arrow). The Fermi velocities of the CoO2 surface states are 0.75 eV Å (blue) and 0.5 eV Å (green), respectively. This is roughly 10% of that of the bulk state, in agreement with the previous study of PdCoO2 single crystals.9,42 Dispersion along the K-M-K direction of Pd- and CoO2-terminated PdCoO2 films are shown in Figs. 3(f) and 3(h). Both terminations of the PdCoO2 films show a split band at the M point 0.75 eV below EF and 1.75 eV below EF, as observed in PdCoO2 single crystals.41 Interestingly, a hole band below EF at the M point driven by the palladium surface state (as shown in Ref. 41) is not seen in our Pd-terminated films.
|Fermi Velocity .||Pd .||CoO2 .||FeO2 .||Fe0.17Co0.83O2 .|
|vF (eV Å)||4.72 ± 0.15||4.32 ± 0.25||4.69 ± 0.07||4.43 ± 0.09|
|Fermi Velocity .||Pd .||CoO2 .||FeO2 .||Fe0.17Co0.83O2 .|
|vF (eV Å)||4.72 ± 0.15||4.32 ± 0.25||4.69 ± 0.07||4.43 ± 0.09|
Comparing the electronic structure observed for our epitaxial PdCoO2 thin films to that reported for PdCoO2 single crystals with different terminations, our PdCoO2 films have similar bulk state features to those of PdCoO2 single crystals, but the surface states of our PdCoO2 films are weaker or even disappear at EF. Note that the alternating layers of Pd+ and along the c-axis of PdCoO2 are not charge neutral. Doping of the surface by electrons arising from electronic reconstruction (i.e., no structural surface reconstruction) would generate the surface states observed on bulk single crystals. Ways in which the surfaces of our films differ from the single crystalsprovide routes to different or no surface states. For ARPES measurements of cleaved (001)-oriented PdCoO2 single crystals, the polar surface exposed after cleaving may drive an electronic reconstruction of the surface or alternatively a mixture of termination regions, some of which are terminated by palladium and some of which are terminated by CoO2, to alleviate the polar surface charge. To synthesize PdCoO2 films, we use shuttered MBE growth to provide a full layer of palladium or CoO2 as the terminating surface. This difference in the surface reconstruction structure of epitaxial PdCoO2 films might result in a different electronic reconstruction from that of cleaved PdCoO2 single crystals. We show the reconstruction of our films in low-energy electron diffraction (LEED) with different terminations in Fig. S4 of the supplementary material. Additional differences exist between our PdCoO2 films and PdCoO2 single crystals. For example, PdCoO2 films grown on (001) Al2O3 substrates are known to have a high density of 180° in-plane rotation twins.38 Furthermore, our films might contain oxygen vacancies to neutralize the surface polar effect. These may also play a role in the differences observed in the surface states between PdCoO2 single crystals and our epitaxial films. In particular, the palladium surface state is very reactive; it can be essentially destroyed by temperature cycling of ARPES measurements.42 Prior ARPES results of a PdCoO2 film grown by PLD,39 where the palladium termination is confirmed by the absence of a CoO2 surface state in the electronic structure, does not show the PdCoO2 bulk state. In contrast, our Pd-terminated films show a strong PdCoO2 bulk state without a CoO2 surface state, but are missing the palladium surface state. One possible reason for this is the difference in sample quality, particularly of the sample surface and the ability of MBE to control the termination of the PdCoO2 film. We note that our MBE and ARPES measurements are made immediately following film growth and that the ultra-high vacuum connection between our MBE and ARPES systems enables us to investigate the electronic structure of the pristine growth surface.
With the goal of introducing magnetic order into the surface of PdCoO2, we attempt to replace the cobalt in the CoO2 surface termination with a different transition metal. The low spin state (S = 0) of the d6 electron configuration of Co3+ in octahedral coordination underlies the lack of magnetic order in PdCoO2. In contrast, PdCrO2 is known to order antiferromagnetically at 37 K due to the unpaired spins (S = 3/2) arising from Cr3+ with its d3 electron configuration.10,11 Considering what other transition metals are stable in the 3+ oxidation state under the highly oxidizing conditions needed to stabilize PdCoO2 led us to attempt to substitute Fe3+ for Co3+. Other known iron-containing delafossites, such as the semiconductors AgFeO2 and CuFeO2 with bandgap larger than 1 eV, are known to exhibit magnetic phase transitions.43–46 Combining magnetic FeO2 with conducting palladium might create a new metallic delafossite with interesting magnetic properties. Note that PdFeO2 is neither a known compound nor has it been suggested to form in the delafossite structure by prior crystal chemistry based suggestions of delafossites47,48 nor first-principles suggestions for new delafossites.49,50
Our attempts to terminate PdCoO2 with a monolayer of FeO2 were successful. To do so, we used epitaxial stabilization, a process in which lattice misfit strain energies and interfacial energies are exploited to favor a desired metastable phase over the equilibrium phase.51–56 ARPES measurements reveal the FeO2-terminated PdCoO2 film to have a similar bulk band to PdCoO2, but there is no extra surface state at the Fermi surface nor a reconstruction driven by AFM order like in PdCrO2. In addition, we synthesized Pd(Co, Fe)O2 films containing a solid solution of iron and cobalt in each CoO2 layer of the Pd(Co,Fe)O2 film. Electrical transport measurements on a series of 13 nm thick PdFexCo1−xO2 films with varying x (0 0.2) all show a drop in resistance at low temperature. Other than the dip shown in dρ/dT at low temperature, which is different from other known delafossites, no pronounced magnetic order is observed by magnetic susceptibility measurement down to 3 K. This behavior is in contrast to what is seen for PdCrO2. Turning to the electronic structure of the PdFexCo1−xO2 films revealed by ARPES, a similar bulk band to PdCoO2 is observed. No reconstruction appears at the Fermi surface nor is any temperature dependence of the electronic structure of PdFexCo1−xO2 seen.
We added a full monolayer of FeO2 to a Pd-terminated PdCoO2 film in an attempt to vary the termination of PdCoO2 by introducing unpaired electrons (spin) from Fe3+. As ARPES is a surface sensitive measurement, if the unpaired electrons from Fe3+ with its d5 configuration in the surface FeO2 layer interact within the FeO2 layer and with the adjacent palladium layer like the CrO2 layer does in the AFM metal PdCrO2,13 we expect to see distinct features arise in the Fermi surface of the FeO2-terminated PdCoO2 film. The well crystallized FeO2 termination determined by RHEED is shown in Fig. 5(a). Unfortunately, no difference is observed in the ARPES other than the similar PdCoO2 bulk state appearing at the Fermi surface in the FeO2-terminated PdCoO2 film, as shown in Figs. 2(c) and 2(g). No reconstruction of the Fermi surface was seen at low temperature. The bulk band was also free of any temperature-dependent feature when we analyzed the MDCs in the Γ-K direction shown in Fig. 4(e). Thus, our epitaxial FeO2 layer on the surface of PdCoO2 film does not appear to create a spin interaction with the underlying palladium layer. On the other hand, PdFeO2 is a metastable phase and we can only stabilize one formula unit of PdFeO2. It is possible that the FeO2 termination is insulating due to oxygen vacancies to neutralize its otherwise polar surface and preventing it from contributing to the electronic structure. The photoemission intensity data we collect include the film beneath the FeO2 layer, which is PdCoO2 itself.
In addition to replacing the entire CoO2 monolayer with an FeO2 monolayer, we also investigated the partial replacement of cobalt with iron hoping that the presence of iron in multiple layers of the Pd(Co,Fe)O2 structure would enhance the chance of spin interaction between the Fe3+ and the adjacent palladium layer. In Figs. 2(d) and 2(h), one can observe that the band of the PdFe0.17Co0.83O2 film still has similar features to the bulk state of PdCoO2, but with significant noise. Meanwhile, similar to the FeO2-terminated PdCoO2 film, no temperature dependence of the bulk band of Pd(Co,Fe)O2 is seen at EF [see Figs. 4(c)–4(f)]. We compared the Fermi velocity (vF) of the PdFe0.17Co0.83O2 film with the other three different terminations of PdCoO2 films in Table I, and within experimental error, they all have the same vF value as that of PdCoO2 single crystals.41 Thus, the partial replacement of cobalt by iron in each CoO2 layer does not bring any ordered spin interaction between the (Co,Fe)O2 layers nor do we observe any evidence of interaction with the electrons of Pd(Co,Fe)O2, which contribute to the bulk band of PdCoO2. The noise observed in the Fermi surface of the PdFe0.17Co0.83O2 film might come from the increased disorder accompanying the replacement of cobalt by iron. With the d5 configuration of Fe3+ and the d6 configuration of Co3+, the Fe0.17Co0.83O2 layer should, in principle, have a d5.83 configuration resulting in unfilled d electron bands. Nonetheless, no new conducting band is seen at EF as shown in Fig. 4. On the other hand, a hole doping scenario in which 0.17 electrons move from the palladium layer to the Fe0.17Co0.83O2 layer would leave the Fe0.17Co0.83O2 layer in an insulating state. Such a scenario would require hole doping of 0.17 holes on the palladium layer. A comparison of the momentum at EF (kF) shown in Fig. S7 of the supplementary material, reveals no pronounced difference between the kFs of a PdCoO2 film and a Fe0.17Co0.83O2 film.
Further characterization of the PdFexCo1−xO2 films is shown in Fig. 5. The maximum percentage of iron that we are able to incorporate into epitaxial PdFexCo1−xO2 films while retaining a single phase is x = 20%. RHEED of a single-phase, 13 nm thick PdFe0.2Co0.8O2 film is shown in Fig. 5(a). The fringes in the x-ray diffraction θ–2θ scans of the PdFexCo1−xO2 films indicate the high structural quality of these films. Electrical transport measurement on the PdFexCo1−xO2 films is shown in Fig. 5(d). Note that the PdCoO2 film shown in this comparison has a thickness of 13 nm, a quarter of the thickness of the PdCoO2 film in Fig. 1(d). The upturn in resistivity of the pure PdCoO2 film (0% Fe) seen in Fig. 5(d) below 20 K likely originates from localization in the ultrathin film. As more cobalt is replaced by iron, the absolute resistivity of the iron-doped PdCoO2 film keeps increasing. Interestingly, instead of showing an upturn at low temperature like is seen in the pure PdCoO2 film, the Fe-doped PdCoO2 films show a drop at low temperature in electrical resistivity. Moreover, as the iron content (x) of the PdFexCo1−xO2 film is increased, a more pronounced drop in resistivity is seen. Derivatives of the resistivity as a function of temperature of these PdFexCo1−xO2 films are shown in Fig. 5(e). Strikingly, a dip at about 20 K is observed in the temperature derivatives of all iron-doped PdCoO2 films, which is opposite to the dρ/dT in PdCrO2 where a peak shows up at TN driven by AFM order.10 The amplitude of the dip observed for PdFexCo1−xO2 increases with larger iron concentration.
A comparison of the temperature dependence of the Hall coefficient (RH) between a PdCoO2 film and a PdFe0.17Co0.83O2 film is shown in Fig. 6(a). The RH measurements are consistent with electrons acting as the carriers in both PdCoO2 and PdFe0.17Co0.83O2 films. The magnitude of RH in the PdCoO2 film is in agreement with prior reports from PdCoO2 single crystals.57 In contrast to Ref. 39, we do not observe an anomalous Hall effect in our PdCoO2 films at low temperature. Hall resistivities of the PdCoO2 film and the PdFe0.17Co0.83O2 film are shown in Fig. S9 of the supplementary material. The PdFe0.17Co0.83O2 film exhibits a larger RH than does the pure PdCoO2 film, which could be a result of carriers being trapped by iron-induced structural defects. For the PdCoO2 film, the temperature dependence of RH at low temperature becomes flat while for the iron-doped PdCoO2 film the RH starts increasing below 20 K, which is the same temperature at which the change in dρ/dT is observed in Fig. 5. One possibility for the observed resistivity anomaly at low temperature is that it is driven by iron disorder, since it is independent of the iron concentration. One scenario explaining why the RH difference between of PdCoO2 film and the PdFe0.17Co0.83O2 film does not reflect on the band structure is that the electrons from iron do not interact with the electrons from palladium. Instead, iron clusters just trap the electrons from the PdCoO2. Within this scenario, it is possible that iron disorder clusters in PdFe0.17Co0.83O2 films are revealed by AFM in the supplementary material (Fig. S3). In Figs. 6(c) and 6(d), the magnetoresistance of the same PdCoO2 film and PdFe0.17Co0.83O2 film shows distinct magnetic dependences. The overall scale of magnetoresistance in the PdCoO2 film is much smaller than that observed in PdCoO2 single crystals.58 The temperature dependence of the magnetoresistance of the PdFe0.17Co0.83O2 film shows weak-localization-like behavior, which may arise from the magnetic disorder resulting from the addition of iron. The temperature dependence of the magnetic susceptibility of the PdFe0.17Co0.83O2 thin film shows no transition or difference between the zero-field-cooled (ZFC) and field-cooled (FC) curves as shown in the supplementary material (Fig. S10). The observed behavior is in contrast to the splitting that is expected when AFM order is observed, such as in PdCrO2.11 Thus, the replacement of cobalt by iron does not appear to result in any spin order. Instead, only signs of magnetic disorder are seen.
In summary, we have synthesized high-quality PdCoO2 films by MBE and harnessed the atomic layer control afforded by MBE to tune the termination of these films to study the resulting electronic structure by ARPES. On comparing the Pd-terminated and CoO2-terminated PdCoO2 films with those of PdCoO2 single crystals, though the resistivity of our PdCoO2 films are far higher than that of single crystals at low temperature, we find the PdCoO2 bulk states in our films show features similar to those of PdCoO2 single crystals, while the palladium surface state and CoO2 surface state are not as strong as those of the PdCoO2 single crystals. This difference might arise due to different electronic reconstructions. We also studied FeO2-terminated PdCoO2 films and find that the only remaining PdCoO2 bulk state in the electronic structure is similar to that of PdCoO2. In addition, we have successfully synthesized PdFexCo1−xO2 films. From the electric transport measurements, the addition of iron further increases the resistivity of PdCoO2 films at room temperature. Meanwhile, we see different behavior at low temperature compared to pure PdCoO2 films, but no magnetic ordering akin to what happens in PdCrO2 is seen in our PdFexCo1−xO2 films. The electronic structure of a PdFe0.17Co0.83O2 film shows a bulk state similar to that seen in pure PdCoO2 films. Although we do not induce new spin order in delafossite films by replacing cobalt by iron in PdFexCo1−xO2 with x up to 0.2 or a pure FeO2 terminating monolayer, our work invites further exploration of ways in which the electronic structure of delafossites can be perturbed by exploiting the ability of MBE to control atomic layering in combination with ARPES to measure its impact.
See the supplementary material for the growth methods of the PdFexCo1−xO2 (0 ≤ x ≤ 0.2) films and additional characterization by AFM, LEED, ARPES, and XPS, as well as the results of Hall effect and magnetization measurements.
The authors thank Alfred Zong, Juan Jiang, Yu He, and Liguo Ma for their insightful discussions. This paper was primarily supported by the U.S. Department of Energy, office of Basic Sciences, Division of Materials Science and Engineering under Award No. DE-SC0002334. This paper made use of the Cornell Center for Materials Research shared facilities, which are supported through the NSF Materials Research Science and Engineering Centers Program (Grant No. DMR-1719875). This paper also made use of the Cornell Energy Systems Institute Shared Facilities partly sponsored by the NSF (Grant No. MRI DMR-1338010) and the Kavli Institute at Cornell. Substrate preparation was performed, in part, at the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure, which is supported by the NSF (Grant No. NNCI-2025233). C.T.P. acknowledges support from Air Force Office of Scientific Research Grant No. FA9550-21-1-0168 and National Science Foundation Grant No. DMR-2104427. P.D.C.K. gratefully acknowledges support from the European Research Council (through the QUESTDO project, Grant No. 714193). Q.X. acknowledges support from the REU Site: Summer Research Program at PARADIM (Grant No. DMR-2150446). The authors thank Sean C. Palmer for his assistance with substrate preparation.
Conflict of Interest
The authors have no conflicts to disclose.
Qi Song: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Writing – original draft (equal). Brendan D. Faeth: Data curation (supporting). Hanjong Paik: Data curation (supporting). Phil D. C. King: Writing – review & editing (supporting). Kyle M. Shen: Resources (equal). Darrell G. Schlom: Conceptualization (equal); Data curation (supporting); Funding acquisition (lead); Project administration (equal); Resources (equal); Writing – review & editing (equal). Jiaxin Sun: Data curation (equal); Writing – review & editing (supporting). Christopher T. Parzyck: Formal analysis (supporting); Writing – review & editing (equal). Ludi Miao: Data curation (equal); Writing – review & editing (supporting). Qing Xu: Data curation (supporting). Felix V.E. Hensling: Data curation (supporting). Matthew R. Barone: Data curation (supporting); Formal analysis (supporting); Writing – review & editing (supporting). Cheng Hu: Data curation (supporting). Jinkwon Kim: Data curation (supporting).
The data that support the findings of this study are available from the corresponding author upon reasonable request.