2.1 Design of Model Systems for VO₂ Films with Two Crystal Orientations

To investigate the influence of the crystal orientation of the VO₂ film in the oxygen ionic transport out from the film, we explored an appropriate model system by selecting representative crystal orientations expected to exhibit the most distinct differences in ionic transport (Figure 1). Ion movement in anisotropic tetragonal VO₂ based on the rutile structure is most preferred toward [001] direction, along which empty channels are aligned, as shown in Figure 1a.^{[14, 17, 19]} Thus, as depicted in Figure 1b, we prepared model samples of the anti-channeling (100)- and channeling (001)-VO₂ films with the [001] channels parallel (left) and perpendicular (right) to the surface, respectively. To induce oxygen ionic transport from an oxide, another oxide with an identical crystal structure but different oxygen chemical potential was epitaxially grown on the oxide.^{[10-13, 23]} In VO₂, Park et al. and Lu et al. reported that when TiO₂ is epitaxially grown on VO₂ film under an oxygen-deficient environment, stoichiometric TiO₂ can be grown by absorbing the oxygen ions from VO₂, consequently transferring the oxygen vacancies to VO₂ layer.^{[13, 23]} Based on previous reports, we suggest the TiO₂ top layers on the (100)- and (001)-VO₂ films as model systems.

Utilizing atomic structure models and first-principles calculations (Figure 1c,d), we predicted the differences in oxygen migration behaviors in each VO₂ film model sample. Oxygen migration occurs via oxygen vacancies in the lattice due to an oxygen-deficient environment during the top TiO₂ layer's growth. We considered oxygen migration between planes oriented toward the film surface. In (100)-VO₂, oxygen migrates to the same position on the subsequent (100) plane through the two nearest oxygen sites, as depicted in Figure 1c. In contrast, in (001)-VO₂, oxygen moves through the four nearest oxygen sites to reach an identical position on the next (001) plane, doubling the path compared to (100)-VO₂, as depicted in Figure 1d. This difference in migration paths is closely linked to the channel orientation. Oxygen vacancies travel along the edges of the VO₆ octahedral chain, i.e., channel walls. The (100)-VO₂ allows movement along only two channel walls perpendicular to the surface direction, as shown at the bottom of Figure 1c, while (001)-VO₂ involves all four channel walls, as shown at the bottom of Figure 1d. Therefore, channeling (001)-VO₂ facilitates oxygen migration with a higher probability than anti-channeling (100)-VO₂. Additionally, we compared the oxygen migration barriers in each direction of VO₂ thin films using first-principles calculations, as shown in Figure 1e. The VO₂ structure was optimized considering its form as a thin film on TiO₂ substrates (i.e., clamping on TiO₂ substrates). The oxygen migration barrier in channeling (001)-VO₂ is significantly lower at 0.48 compared to 0.69 eV in anti-channeling (100)-VO₂. A lower migration barrier implies a substantially higher probability of migration. Therefore, given the crystallographic anisotropy of VO₂, oxygen migration from VO₂ to the top TiO₂ layer is notably more efficient in channeling (001)-VO₂ than anti-channeling (100)-VO₂.

2.2 Epitaxial Growth of VO₂ Films and Top TiO₂ Layers for Oxygen Transport

To realize the model samples as mentioned in Figure 2, we prepared 13-nm-thick VO₂ films on single-crystalline rutile (100)- and (001)-TiO₂ substrates by pulsed laser deposition (PLD). The VO₂ thickness of 13 nm is the optimal condition to prevent microstructural defects by strain relaxation^[31-33] or interdiffusion from substrates^{[33, 34]} that deteriorate VO₂ characteristics, allowing us to focus on the effect of the crystallographic channel inside VO₂ on oxygen transport. In the symmetrical X-ray diffraction (XRD) (Figure 2a,b), the TiO₂ substrate peaks (denoted by closed arrow) are sharp and intense, while the VO₂ film peaks (denoted by open arrow) are broadened due to their finite thickness.^[35] The VO₂ films exhibit a (200) peak with a TiO₂ (200) peak in Figure 2a, and a (002) peak with a TiO₂ (002) peak in Figure 2b, indicating that the intended (100)- and (001)-VO₂ films were easily fabricated by changing the surface orientation of the TiO₂ substrate. The layer fringes around the VO₂ peak in both two films demonstrate that the VO₂ films have very sharp interfaces and a uniform thickness of ≈13 nm, as estimated from the fringe period.

Subsequently, ≈7-nm-thick TiO₂ layers were grown on the as-grown (100)- and (001)-VO₂ films under a sufficiently low pO₂ ≈8 mTorr to cause the oxygen transport from VO₂ to TiO₂ as depicted in Figure 2c.^{[13, 23]} During this interfacial reaction step, the oxygen vacancies should be inevitably generated in VO₂ near the interface region, and thus the oxygen ions inside the VO₂ film diffuse to the interface region due to the gradient of the oxygen concentration. Interestingly, the c lattice parameter of VO₂ increased by ≈0.6% in the case of TiO₂ on the (001)-VO₂ film, which indicates the favorable formation of oxygen vacancies in VO₂, because the lattice expansion is one of the representative phenomena caused by the oxygen vacancies in VO₂.^{[13, 17, 21]}

The crystal orientation and crystallinity of the top TiO₂ on the (100)- and (001)-VO₂ films are shown in Figure 2c,d. The high-angle annular dark-field (HAADF)-STEM images demonstrate the perfect crystallinity in both TiO₂ on the (100)- and (001)-VO₂ films, as confirmed by the well-aligned Ti atom columns in their TiO₂. It is quite intriguing that both TiO₂ layers are epitaxially grown well with identical structures and orientations of their rutile TiO₂ substrates because the TiO₂ layers were grown at 150 °C, which is considerably lower than the conventional growth temperature of 400–500 °C.^{[19, 36]} These epitaxial growths of 7-nm-thick TiO₂ at a low temperature indicate that both (100)- and (001)-VO₂ films supply oxygen ions to the top TiO₂, as hinted by the previous reports.^{[13, 23]} However, considering the different lattice expansion results in the two VO₂ films in XRD after TiO₂ growth, there is a difference in the degree of oxygen transport between the (100)- and (001)- VO₂ films.

2.3 Local Lattice Strain Analysis on the VO₂ Films using STEM

To investigate the lattice expansion inside the (100)- and (001)-VO₂ films in more detail, we visualized the local lattice strain maps obtained by the geometric phase analysis (GPA; HREM Research Inc., Japan) of HAADF-STEM images (Figure 3).^{[37, 38]} Here, only the lattice strains along the out-of-plane direction (z) are mentioned because the in-plane lattice parameters negligibly change after the oxygen transport owing to the tight constraint by interaction with their substrates (Figure S1, Supporting Information). The lattice spacing along the z-direction of the VO₂ films was calculated with respect to the TiO₂ substrates and lattice mismatch expressed by the lattice strain (i.e., ɛ_zz = (z − z_substrate)/z_substrate),^[38] as shown in Figure 3c,d. Unlike the (001)-VO₂ film, where the lattice strain is expressed by the green color in the entire VO₂ film, the (100)-VO₂ film exhibits a bimodal distribution, indicated by the red and green colors.

To quantitatively confirm the local lattice expansion, the values of ɛ_zz were extracted and profiled, as shown in Figure 3e,f. The actual strain can be elucidated by comparing the measured ɛ_zz (red dotted lines in e and f) and reference ɛ_zz (green dotted lines in e and f). The reference ɛ_zz was calculated by the lattice spacing of bulk VO₂ using the Poisson's ratio (Supplementary 1). In (100)-VO₂ film, the measured ɛ_zz at the lower part of VO₂ is consistent with the reference ɛ_z, whereas the measured ɛ_zz at the upper part is higher (0.035) than that of the reference ɛ_zz (−0.017) indicated by the red arrow in Figure 3e, which corresponds to the lattice expansion of ≈5.3%. Alternatively, in the (001)-VO₂ film, the measured ɛ_zz is homogeneous and higher (−0.035) than that of the reference ɛ_zz (−0.043), implying the lattice expansion of ≈0.8% in whole areas of the (001)-VO₂ film. As the lattice expansion is mainly caused by the oxygen vacancies, the results of the strain analysis indicate that the oxygen vacancies in the (100)-VO₂ film are preferentially formed at the upper region of VO₂ near the interface_, whereas the oxygen vacancies in the (001)-VO₂ film are formed across the entire regions of VO₂. The different distribution of the oxygen vacancies in the (100)- and (001)-VO₂ films is consistent with the first-principles calculations shown in Figure 1, in which the oxygen ions are predicted to migrate more efficiently in the (001)-VO₂ film than in the (100)-VO₂ film.

2.4 Local STEM-EELS Analysis on the VO₂ Films and TiO₂ Layers

We verified the role of oxygen vacancies in the lattice expansion of the VO₂ films. To confirm the distribution of oxygen vacancies in (100)- and (001)-VO₂ films, we measured V L_3,2- and O K-edge EELS by line scanning through the cross-sectional positions of the VO₂ films, as shown in Figure 4a,b. In the 3d transition metal oxides, the L-edge of the transition metal and K-edge of oxygen obtained by EELS are known to be sensitive to oxygen vacancies, which change the chemical bonding states and reduce the valence states of the multivalent transition metals (i.e., from V⁴⁺ to V³⁺ in VO₂).^[39-41]

In the EELS results from (100)-VO₂ (Figure 4a), different features are shown at the upper and middle parts compared to the lower part of the VO₂ film. First, compared to the lower part, the V L_3,2-edges of the upper and middle parts of VO₂ shifted to the left by ≈0.4 eV (dotted lines in a, from 516.4 eV at the lower part to 516.0 eV at the upper and middle parts in the L₃-edges), indicating the weaker binding energy with the introduction of oxygen vacancies.^{[39, 42]} The increases of the L₃/L₂ intensity ratio in the upper and middle parts (open arrows “▽” in a, from 0.98 in the lower part to 1.02 in the upper and middle parts) depict the phenomena induced by the decreased oxidation state of V⁴⁺ to V³⁺ due to the oxygen vacancies.^[42-44] Based on the V–O hybridization in VO₂, the O K-edge spectra can also describe the changes in the oxidation state of the V ions.^{[43, 45]} In the O K-edge, divided into the first t_2g peak and second e_g peak by crystal field splitting, the first t_2g peak (closed arrows) at the upper and middle parts of VO₂ is suppressed, compared to that of the lower part, which implies the decreased oxidation state of vanadium owing to the oxygen vacancies.^{[42, 43, 45]} Therefore, in the (100)-VO₂ film, oxygen vacancies are distributed in the upper half region, which proves that oxygen transport to the top TiO₂ occurs only in the top local region of the (100)-VO₂ film, where oxygen moves across the [001] channel which is parallel to the surface. In contrast, the (001)-VO₂ film exhibits homogenous characteristics in its EELS spectra over the entire film region, as shown in Figure 4b. All V L_3,2-edges in the entire area shifted to the left by ≈0.5 eV (dotted lines in b, 515.9 eV in L₃-edge) with an increase in the L₃/L₂ intensity ratios (1.02–1.03) compared to the lower part of the (100)-VO₂ film. Moreover, the first t_2g peak of the O K-edges are suppressed at the entire VO₂ area. For a detailed investigation of the non-stoichiometries in the VO_2-x films after oxygen transport, the contents of oxygen vacancies were estimated to be x ≈ 0.002 and 0.13 in the lower and upper regions of (100)-VO₂ film, respectively, and x ≈ 0.16 and 0.17 in the lower and upper regions of (001)-VO₂ film, respectively. These estimates were obtained using a model-based quantification method on the EELS spectra (Figure S3, Supporting Information). These results indicate that oxygen vacancies were uniformly formed across the entire (001)-VO₂ film and in greater quantities compared to the (100)-VO₂ film, where vacancies were formed only in the upper region.

Therefore, in the (001)-VO₂ film, where oxygens travel along the [001] channel which is perpendicular to the surface, oxygen vacancies are distributed throughout the entire VO₂ film. The enhanced depth of the oxygen-deficient region in the (001)-VO₂ film than that in the (100)-VO₂ film confirms that the oxygen transport to the top TiO₂ is facilitated along the [001] channel direction by the relatively favorable oxygen vacancy migration.

We also measured Ti L_3,2-edge and O K-edge EELS on the top TiO₂ layers of the (100)- and (001)-VO₂ films to verify the different oxygen supplies from their bottom VO₂, as shown in Figure 4c. In the top TiO₂ on (001)-VO₂ film (bottom of c), the Ti L_3,2-edges and O K-edges are clearly split into two peaks (t_2g and e_g) due to the crystal field splitting of Ti 3d from the surrounding oxygens, and the energy splitting between the t_2g and e_g peaks in the Ti L₃-edge (dotted lines in c) is 2.5 eV, which is consistent with the stoichiometric TiO₂ reported by Grunes et al.^[46] These results indicates that the TiO₂ layer has a good stoichiometry with the Ti⁴⁺ ions despite the low pO₂ of ≈8 mTorr owing to the sufficient oxygen transport from the entire region of the bottom VO₂ film. In contrast, for the TiO₂ on the (100)-VO₂ film (top of c), the t_2g–e_g splits in Ti L_3,2- and O K-edges almost disappeared, which directly confirms that the Ti³⁺ ions are ascribed to the oxygen vacancies.^{[47, 48]} Additionally, XPS analysis, which is suitable for characterizing the chemical state of surface layers (up to a depth of ≈5 nm), also confirmed that the contribution of Ti³⁺ spectra is higher in (100)-TiO₂/VO₂ than in (001)-TiO₂/VO₂ (Figure S4, Supporting Information), similar to the EELS results. The TiO₂ on the (100)-VO₂ film exhibited good crystallinity in the HAADF-STEM images, as shown in Figure 2a, however, in reality, the (100) top TiO₂ is nonstoichiometric owing to the oxygen vacancies because the (100)-VO₂ film could not provide sufficient oxygen ions from the upper half of the VO₂ film. Therefore, we directly determined the oxygen vacancies distribution in the local region of the VO₂ film and top TiO₂ through lattice strain and EELS analysis, which visualizes the differential oxygen transport between the (100)- and (001)-VO₂ films. As predicted in Figure 1, oxygen transport was more effective when the [001] channel was perpendicular to the surface.

2.5 Differential tuning of Metal–Insulator Transitions in VO₂ Films

To confirm the tuning of the MI transition properties of the (100)- and (001)-VO₂ films after oxygen transport, we measured the temperature-dependent sheet resistance (R_s vs T) of the VO₂ films before and after the growth of the top TiO₂ layers, as shown in Figure 5a,b. In TiO₂/VO₂ films, the top TiO₂ thin film has a negligible effect on the measurement of sheet resistance of VO₂ film (Figure S5, Supporting Information) due to its much higher resistivity^{[19, 49]} and high leakage current^{[50, 51]} in its very thin thickness of 7 nm. Both the as-grown (100)- and (001)-VO₂ films exhibit large resistance variations across three and four orders of magnitude, even at different transition temperatures owing to their strain from their substrates (Figures S1 and S2, Supporting Information).^{[13, 14, 16, 52]} After growing the top TiO₂ layers, different MI transition behaviors are noted on the (100)- and (001)-VO₂ films (red arrows in a and b). In particular, the R_s at the insulating state decreased by an order in the (100)-VO₂ film, whereas it decreased by three orders of magnitude in the (001)-VO₂ film. Additionally, the transition temperature, based on the heating curve, decreases by 9 K (from 334 to 325 K) in the (100)-VO₂ film, whereas it significantly decreases by 104 K (from 294 to 190 K) in the (001)-VO₂ film. The decrease in the resistance of the insulating state and transition temperature of the VO₂ films are ascribed to the oxygen vacancy that stabilized VO₂ in the metallic–rutile phase.^{[16, 20, 21]} Oxygen vacancies introduce extra electrons and destabilize the V-V dimerization of the insulating monoclinic phase, leading to the transition into the metallic rutile phase. We observed that the oxygen-deficient upper half region of the (100)-VO₂ film was stabilized to the rutile structure, whereas the lower region has a monoclinic structure, as shown in Figure S6 (Supporting Information). This suggests the different changes of the MI transition in the (100)- and (001)-VO₂ films owing to the different contents and distribution of the oxygen vacancies in VO₂, as confirmed by the previous GPA and EELS analyses illustrated in Figure 5c,d. The MI transition of the (100)-VO₂ is minimally suppressed because the local upper part of the film is metalized, whereas that of (001)-VO₂ film is fully suppressed because the entire region of the film is metalized.

As other well-known possible factors for the changes in the MI transition properties, we also considered elastic strain changes and Ti doping after top TiO₂ growth. However, RSM results confirmed that the in-plane strain state remained unchanged before and after TiO₂ growth due to the rigid substrate (Figure S1, Supporting Information). Additionally, Ti doping would increase the transition temperature and resistance, which is the opposite of our observed changes.^[53] Therefore, these factors are minor in the metallization of VO₂ films after oxygen transport. Additionally, we considered the possibility of Magnéli phases V_nO_2n-1, which possess a lower oxygen stoichiometric ratio than VO₂, as a cause for metallization. Given the estimated oxygen vacancy content in our VO₂ films, there is a possibility of a phase transition to V_nO_2n-1 (5 ≤ n ≤ 8). However, in our study, the temperature and duration during the oxygen vacancy formation were very low at 150 °C and less than 10 min, respectively, which kinetically hinders the structural transformation to Magnéli phases.^{[20, 54]} Z. Zhang et al.^[20] annealed the insulating monoclinic VO₂ at low temperatures of 450 °C with low pO₂ and demonstrated that the annealed VO₂ transformed into oxygen-deficient VO_2-δ with δ ≈ 0.2, but did not transform into Magnéli phases despite having very similar stoichiometry to our cases. This is because the structural transition to Magnéli phases requires sufficient time and high temperature to order oxygen vacancies over long ranges. Therefore, we believe that the most reasonable factor causing changes in the electrical properties of both VO₂ films is the formation of oxygen vacancies and their stabilization of the metallic rutile phase as we proposed. The differences in the distribution of these vacancies led to the different changes in the MI transition in the (100)- and (001)-VO₂ films.