Introduction

Strong electron-electron correlations within narrow d- or f-orbitals underpin a variety of condensed matter phenomena, such as metal-to-insulator transitions (MITs), high-temperature superconductivity, magnetism, and multiferroicity, often observed in correlated electron materials (CEMs)1,2. VO2 is a prototypical example of a CEM with a temperature-dependent metal-to-insulator phase transition. The electronic phase transition in bulk VO2, which occurs at an MIT temperature (TMIT) of ~340 K, is accompanied by a structural phase transition from a metallic rutile phase to an insulating monoclinic phase3,4.

The origin of the MIT in VO2—whether gap-opening is driven by the symmetry-lowering structural transition or by electron-electron correlations—has been widely studied5,6,7,8. In particular, there is widespread interest in the nature of the insulating state and its external control via doping9,10, strain11, oxygen vacancy creation12, hydrogenation13, light-and-pulse-induced modulation14,15, and via electric-fields in a field-effect transistor geometry16,17,18.

For example, n-type doping of VO2 with dopants such as W6+, Mo5+, and Nb5+ was shown to decrease TMIT and stabilize the metallic phase9,10,19. By contrast, p-type doping of VO2 with dopants such as Cr3+, Ga3+, and Al3+ was shown to increase TMIT, thereby stabilizing the insulating phase20,21,22. Similarly, both oxygen vacancy creation and hydrogenation were shown to n-dope VO2 and decrease TMIT12,13,23. Finally, in VO2 thin films, macroscopic tensile strain along the rutile a-axis was also shown to decrease TMIT11,12.

In all these previous approaches, modulation of TMIT was always associated with macroscopic changes to the lattice parameters (due to strain) and/or dopant-induced local structural distortions9,10,12,13,20,23. In such experiments, where both the lattice strain and carrier concentration change, it is challenging and, sometimes, impossible to disentangle the role of carrier concentration changes from the role of lattice strain. For instance, in the case of W-doped VO2, an increase in W-doping concentration increases both the carrier density and lattice strain24.

Other techniques utilizing external stimuli, such as electric-field-induced metallization of VO2 in a field-effect transistor geometry, could, in theory, enable the modulation of its conductivity without inducing macroscopic structural changes. However, previous attempts at electric-field-driven metallization of VO2 were not successful12,16,17,25,26. For example, attempts at modulating the MIT in VO2 in a field effect transistor geometry yielded less than a 1 K change in TMIT27,28. The weak response of TMIT of VO2 to external electric field, even when gated through high-K dielectrics, was attributed to the presence of strong correlations in the insulating VO2 phase16. Further, ionic-liquid gating of VO2, which could enable accessibility to large interfacial electric-fields, led to oxygen vacancy creation and/or hydrogenation of VO212,16. Heterostructures with differing compositions such as for example VO2/WxV1-xO2 based heterostructure thin films showed a larger change in TMIT29. However, these changes are related to elemental doping driven by a chemical potential mismatch of the dopant-ion (W6+ in this specific case). By stark contrast, we propose modulation-doping of VO2 using electronic chemical potential differences at oxide heterostructures.

Modulation- or remote-doping of oxide semiconductors is an alternative method for achieving high dopant carrier densities without inducing local structural distortions30,31,32,33. In modulation-doping, a chemical potential mismatch between a high band gap heavily doped layer (dopant-layer) and a lower band gap undoped layer (channel) leads to charge transfer from the heavily doped dopant-layer to the undoped channel. In general, the dopant layer and the channel are spatially separated by a barrier (or a spacer) layer that kinetically limits the interdiffusion of the dopants from the dopant layer to the channel layer while allowing charge transfer via quantum mechanical tunneling34,35.

Modulation-doping was successfully applied to semiconductors and band-insulating oxides such as ZnO and SrTiO331,33,36,37,38,39. However, modulation-doping of correlated electron insulators has had limited success. For example, Stemmer and colleagues reported modulation-doping of NdNiO3, but this did not lead to any significant changes in TMIT of the nickelate31. Whether modulation-doping could be a generic approach to induce phase transitions in oxides is unclear and several key questions remain unanswered. For example, how do bands evolve in correlated oxides as a function of doping? Can a rigid band model be applied to understand doping in oxides? How sensitive are the ground state properties in correlated oxides to carrier doping?

In this work, we address some of these open questions using the MIT in VO2 as a model system. We propose a modulation-doped heterostructure to n-dope VO2 without inducing any structural distortions. Commonly in modulation-doping, an epitaxial structure is grown with a spacer, and the dopant layers are epitaxially matched to the semiconducting channel layer. Note that both the spacer and dopant layers must be insulating with a bandgap that is higher than that of the channel layer. However, the only stable rutile oxide that is insulating with a compatible band mismatch that allows modulation-doping of VO2 is TiO2. The other rutile oxides such as CrO2, RuO2, and IrO2 are metallic and therefore not compatible as dopant layers40,41.

As an additional challenge, oxygen lattice continuity in epitaxial structures might also lead to oxygen vacancy diffusion across the layers23,42,43. We note that oxygen vacancy formation, which was shown to affect the MIT in VO2, is commonly observed in transition metal oxides12,44,45. Thus, in order to prevent oxygen vacancy diffusion across the wide band-gap spacer layer as well as to circumvent the lack of lattice-matching insulating rutile oxides, we have gone away from epitaxially-matched modulation-doped heterostructures. Instead, we propose an amorphous LaAlO3 (LAO) layer (with a reported electronic band gap of ~5.6 eV)46 as the spacer layer. Since LAO has a low oxygen vacancy-diffusivity47, we use an amorphous oxygen-deficient TiO2-x as the dopant layer. Stoichiometric TiO2 has a bandgap of ~3 eV48 and TiO2-x is n-type conducting. Using TiO2-x instead of a conventionally doped TiO2 such as Nb-doped TiO2, significantly simplified heterostructure deposition. Furthermore, this approach avoids the interdiffusion of metallic dopants such as Nb in Nb-doped TiO2 and the associated unintentional doping of VO2.

The modulation-doped structure for all samples used in this work is comprised of a VO2 channel layer, a 2 nm thick LAO spacer layer, and a 3 nm thick TiO2-x dopant layer, as shown schematically in Fig. 1a. All heterostructures were capped with a 1 nm thick LAO layer to prevent dopant passivation from atmospheric impurities as well as oxidation of the TiO2-x dopant layer. Fermi level alignment across the structure is expected to lead to an electron accumulation region at the LAO/VO2 interface. Expected band-alignments for this type-I heterojunction before and after heterostructure formation are shown in Fig. 1b.

Fig. 1: Structure of Modulation-doped VO2 heterostructures.
figure 1

a Schematic diagram of the heterostructures used in this work. The thickness of VO2 is varied while the thicknesses of all the other layers are as mentioned in the schematic. b Schematic energy band diagram for a VO2/LAO/TiO2-x heterostructure before and after the junction formation. Electron accumulation is expected based on the known band offsets. The color intensities are chosen to be proportional to expected electron densities for better visualization. EC, EV, and EF indicate the conduction band edge, valence band edge, and Fermi level, respectively. c High-resolution cross-sectional high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image showing abrupt interfaces between TiO2 substrate and VO2 film and VO2 film and the amorphous LAO spacer layer. d Elemental mapping using energy dispersive x-ray spectroscopy (EDS) showing the various layers in the heterostructure. Note that the scales of c and d are different.

Full size image

Results

To experimentally realize the proposed modulation-doped structure, all samples were grown using pulsed laser deposition (PLD) on single-crystalline TiO2 (001) substrates. VO2 was deposited at 425 °C, while all the other amorphous layers were deposited at room temperature (see methods section for details). It is important to note that room-temperature deposition of the spacer, dopant and capping layers also minimizes any interfacial interdiffusion. A cross-sectional scanning transmission electron microscopy (STEM) image (Fig. 1c) and the associated energy dispersive spectroscopy (EDS) elemental maps (Fig. 1d) show abrupt high-quality interfaces between the TiO2 substrate and the VO2 film, and between the film and LAO spacer. Furthermore, the STEM image of pristine VO2 on TiO2 (001) substrate shows (supplementary Fig. 1) that the epitaxial atomic arrangement of VO2 is identical for VO2 heterostructures and thin films. This data is consistent with in-situ RHEED patterns of the deposited VO2 films and suggests that the films are both atomically smooth and single-crystalline (Supplementary Fig. 2). The atomic force microscopy (AFM) images of the complete heterostructures further confirm the high quality of the growth by showing atomically smooth film surfaces (Supplementary Fig. 3).

To study the correlation between modulation-doping-induced carrier density changes and the changes in the MIT characteristics, we deposited several modulation-doped heterostructures with varying thicknesses of the VO2 layer ranging from 1.5 nm to 9.5 nm, while keeping the thickness of the TiO2-x layer unchanged at 3 nm. Thomas-Fermi screening lengths in VO2 are expected to be on the order of 1 nm (Supplementary Note 1) and, therefore, the highest n-type carrier densities are expected for the lowest VO2 film thickness used in this study (~1.5 nm). Heterostructures with VO2 films thinner than ~1.5 nm were not attempted due to the expected titanium interdiffusion at the VO2 film and TiO2 substrate interface49,50. We note that interfacial titanium interdiffusion will be present in thicker films as well. However, at VO2 thicknesses greater than 1.5 nm, there is still an observable MIT.

A summary of the θ–2θ X-ray diffraction (XRD) measurements, performed at room temperature, for a 9.5 nm VO2 film and the VO2/LAO/TiO2-x/LAO heterostructures on TiO2(001) substrates are shown in Fig. 2a. Clearly, the angular positions of the VO2(002) Bragg reflection peaks are identical for both the 9.5 nm VO2 film (purple spectrum) and the 9.5 nm VO2 heterostructure (blue spectrum). Furthermore, it is evident that the angular position of the Bragg reflection is essentially independent of the thickness of the VO2 film in the heterostructure. Additionally, there were no significant changes in the θ-2θ X-ray diffractograms between VO2 films and heterostructures with the same VO2 thickness (Supplementary Fig. 4). Reciprocal space maps also confirm that all samples are coherently strained in the plane of the TiO2(001) substrate. The out-of-plane rutile c-axis lattice parameter is identical for the thin film and the heterostructures for all thicknesses of VO2 (Supplementary Fig. 5). Based on θ-2θ XRD measurements, reciprocal space maps, and cross-sectional STEM imaging we conclude that there are no changes in the lattice parameters between VO2 films and heterostructures. And therefore, changes in strain cannot account for the reduction in TMIT observed in modulation-doped VO2 heterostructures (see Supplementary Figs. 1 and 4 and Supplementary Table 1). We also note that the reflections for the LAO spacer and capping layers and the TiO2-x dopant layer are absent, suggesting that these layers are not crystalline.

Fig. 2: XRD and temperature-dependent electrical transport in VO2 heterostructures.
figure 2

a High-resolution θ-2θ XRD spectra for 9.5 nm VO2 thin film and for VO2 heterostructures with varying thicknesses. For nomenclature simplicity, we distinguish VO2 thin films and heterostructures with a VO2 thickness of ‘t’ as tVO2 and tVO2-het, respectively. b Resistance versus temperature plots for the same set of samples as shown in a. Resistance values presented here are normalized to the resistance at 330 K (also see Supplementary Fig. 7). c A comparison of the changes in TMIT versus the changes in the rutile C-axis lattice parameter (ΔTMIT vs ΔCR) for this work and other previously published work using W- and Mo-doping9,10, oxygen vacancy doping22 and strain11. The relative changes are compared to the undoped and unstrained states in the case of bulk doping and for strained VO2, respectively. [also see Supplementary Table 1].

Full size image

Next, we discuss the variations in the MIT characteristics for the same set of samples as used in the XRD studies. As shown in Fig. 2b, TMIT systematically decreases with decreasing VO2 thickness in the heterostructure. Note that the decrease in TMIT for thin films of VO2 is thickness-independent (Supplementary Fig. 6), suggesting that the decrease in TMIT for VO2 heterostructures is intrinsic to heterostructure formation. Furthermore, the sheet resistance of VO2 heterostructures in the insulating state also decreases (Supplementary Fig. 7). Except for the heterostructure with the thinnest VO2 layer (1.5 nm), all films continue to show a positive temperature coefficient of resistance, suggesting metallicity above TMIT.

We summarize our observations and compare the changes in TMIT with other previously published studies in Fig. 2c (also see Supplementary Fig. 8 for details regarding TMIT calculation). Significantly, there is a nearly 60 K change in TMIT for the thinnest heterostructures without any measurable changes to the rutile C-axis. In contrast, any comparable change in TMIT in the literature is associated with ΔCR greater than 0.5 pm. This demonstrates control over the MIT in VO2 without any measurable structural changes in the VO2 heterostructures proposed and synthesized in this work. We note that a decrease in TMIT was observed for elemental doping of VO2 with n-type dopants such as W and Mo, while an increase in TMIT was observed for hole-doping with elements such as Cr and Al. There is no W or Mo in any of the heterostructures in this work, and both La and/or Al doping can be ruled out because they would (if anything) lead to hole-doping resulting in an increase in the TMIT. This is contrary to the decrease in TMIT observed in these VO2 heterostructures.

To measure the extent and type of doping, we performed temperature-dependent Hall measurements. Hall measurements show an enhancement in the carrier densities in the insulating state with the Hall coefficient indicative of electron-doping (Fig. 3a). On the other hand, carrier densities in the insulating phase increased from ~6 × 1017 cm−3 (for 9.5 nm VO2 thin film) to ~2.8 × 1019 cm−3 (for 9.5 nm VO2 heterostructure) to a highest of ~5 × 1021  cm−3 for the 2.5 nm VO2 heterostructure. Contrastingly, the metallic state carrier densities remain identical (~6 × 1022 cm−3) across all the VO2 heterostructures and are consistent with previous reports12,25.

Fig. 3: Carrier concentration-dependent MIT in VO2 heterostructures.
figure 3

Plots of temperature-dependent a carrier densities and b carrier mobilities for tVO2 and tVO2-het samples as mentioned in the legends. Carrier density in the insulating state increases with decreasing VO2 thickness while carrier mobility decreases. c A phase diagram from the results in a. The dotted lines connected across the data points are a guide to the eye.

Full size image

However, carrier mobility decreases in both the insulating and metallic states as the thickness of the VO2 layer decreases (Fig. 3b). In the metallic state, this is potentially due to contributions from interfacial scattering, which increases with decreasing film thickness. In the insulating state, the decrease in carrier mobility could result in part from increased electron-electron scattering and interfacial scattering. A summary of the changes in carrier concentration is plotted against TMIT in Fig. 3c. There is a clear correlation between the TMIT and the carrier density with the highest carrier density of ~5 × 1021 cm−3 stabilizing the metallic state of VO2 to a TMIT of ~237 K (Supplementary Fig. 8). Importantly, the continuous increase in carrier density with decreasing VO2 thickness without any lattice parameter changes is suggestive of modulation-doping in VO2 heterostructures.

To further establish that most of the carriers are induced by modulation doping, we prepared two additional control samples. The first is a 7.5 nm VO2 thin film with a 2 nm LAO cap layer (7.5VO2-LAO) and the second is a 7.5 nm VO2 heterostructure with a 3 nm stoichiometric TiO2 layer (7.5VO2-LAO-TiO2). We compared the MIT characteristics of these two samples with the MIT characteristics of the 7.5 nm VO2 thin film (7.5VO2) and a 7.5 nm VO2 heterostructure with a TiO2-x dopant layer (7.5VO2-het). A summary of sheet resistance versus temperature data is shown in Supplementary Fig. 9. The 7.5 nm VO2 heterostructure with a TiO2-x dopant layer has the lowest sheet resistance and the lowest TMIT with a TMIT change of ~20 K. By contrast, the decrease in TMIT was restricted to ~8 K after the deposition of the 2 nm LAO layer. Remarkably, there is no further decrease in TMIT in the heterostructure with a stoichiometric TiO2 layer, suggesting that the TiO2-x dopant layer is required for the observed large change in TMIT.

Consistent with this, the carrier density in the 7.5 nm VO2 heterostructure with the TiO2-x dopant layer is ~7 × 1019 cm−3 compared to a carrier density of 2 × 1019 cm−3 for the 7.5 nm VO2 capped with 2 nm LAO (Supplementary Fig. 10). It is possible that amorphous (disordered) LAO could host ionized donors and enable modulation-doping33. However, we found that the amorphous LAO deposited for these experiments is insulating, suggesting that any ionized donors should be below the measurement threshold of electrical resistivity measurements. We estimate that such ionized donors in LAO, if any, should have a carrier density of ~1019 cm−3 (assuming a carrier mobility of 0.01 cm2/V-s) or lower, putting an upper bound on the number of carriers contributed by the spacer layer.

As further proof of modulation-doping, we performed resistance-temperature measurements on heterostructures with varying LAO spacer layer thicknesses of 2 nm, 4 nm, and 10 nm. The thickness of VO2 is fixed at 7.5 nm and that of TiO2-x dopant layer is fixed at 3 nm for all three heterostructures. As the thickness of the LAO layer (tLAO) increases, the probability of charge transfer from the TiO2-x dopant layer to the VO2 layer decreases (Fig. 4a, b). Consistent with this, at the highest tLAO of 10 nm, where the lowest amount of charge transfer is expected from the dopant layer, TMIT is the closest to that of a 7.5 nm VO2 film with a 2 nm LAO cap layer, but without the dopant layer. Clearly, the dopant layer does not significantly affect the transition temperature when a 10 nm thick LAO spacer layer is used (Fig. 4c). By contrast for the heterostructure with tLAO = 2 nm, the TMIT is shifted by ~20 K as discussed earlier. Since the thicknesses of the other layers are fixed, the larger shift in TMIT for the tLAO = 2 nm heterostructure implies an increased charge transfer for the thinner spacer layers. This further reinforces the central conclusion that the shift in the transition temperature is enabled by modulation doping.

Fig. 4: Transport characteristics of heterostructures as a function of spacer layer thickness.
figure 4

Schematic representation of quantum mechanical tunneling of charge carrier across the spacer layer for a 2 nm and b 4 nm thicknesses of LAO (tLAO). ‘Ψ’ is the electronic wave function. The schematics represent a decrease in the transfer of charge carriers from the dopant layer (TiO2-x layer) to the VO2 layer as a function of increasing tLAO. c Resistance-Temperature plots comparing the MIT characteristics of 7.5 nm VO2 modulation-doped heterostructures employing tLAO of 2 nm, 4 nm, and 10 nm with the MIT characteristics of a 7.5 nm VO2 film with a 2 nm LAO cap layer, but without any dopant layer.

Full size image

To probe the electronic band bending that enables electron accumulation in the VO2 channel layer, we performed bulk-sensitive hard X-ray photoelectron spectroscopy (HAXPES)51 measurements at the P22 beamline in the PETRA III synchrotron at DESY. We note that standard ultra-violet photoemission (UPS) or soft X-ray photoemission measurements cannot facilitate a probing depth sufficient to reach the VO2/LAO interface that is buried beneath multiple layers of the heterostructure. To capture interfacial band bending in VO2, HAXPES measurements were performed in both the insulating phase (at 200 K) and the metallic phase (at 310 K) for VO2 heterostructures with VO2 thicknesses of 1.5 nm, 3.5 nm, 4.5 nm, and 7.5 nm, and for a VO2 thin film with a thickness of 7.5 nm as a reference.

For the 7.5 nm VO2 film measured at 200 K (insulating state), the binding energies of the V 2p3/2 and V 2p1/2 core-level peaks were observed to be ~515.8 eV and ~523.1 eV, respectively (see Fig. 5a). These measured binding energies (see Supplementary Fig. 11 for binding energy calibration procedure) are consistent with previous reports52,53,54. Importantly, a systematic shift of the main component of the V 2p3/2 peak to higher binding energies is observed for the VO2 heterostructures, with the highest increase in binding energy (~250 meV) observed for the heterostructure with the thinnest VO2 layer (1.5 nm), as seen in the inset. This is also in agreement with the highest carrier densities and the lowest TMIT being observed for the heterostructures with the thinnest VO2 layers. For measurements performed on VO2 in the metallic state, no such binding energy shift was observed (Fig. 5b and Supplementary Fig. 12). This is consistent with the complete screening of interfacial electric fields at the metallic VO2/LAO interface (Fig. 5c, d). The presence of binding energy shifts observed in the insulating state of VO2 and their absence in the metallic state of VO2 further support carrier doping by chemical potential shifts (modulation-doping) in the insulating state for VO2 heterostructures as proposed in this work.

Fig. 5: Band bending and core level spectral changes in VO2 heterostructures.
figure 5

A comparison of V 2p core-level spectra of modulation-doped VO2 heterostructures for a the insulating (200 K) and b the metallic states (at 310 K). A clear shift in the V 2p levels is seen in the insulating state spectra but not in the metallic state spectra. Schematics in c and d show the expected band-bending in the modulation-doped heterostructures for the insulating and metallic states respectively. Band-bending is expected in the insulating state and not in the metallic state. Two additional spectral features not seen in VO2 thin films are labeled P1 and P2.

Full size image

Photoemission data also showed two remarkable features in the V 2p spectra. The first, labeled P1 in Fig. 5a, b, is a lower binding energy shoulder around ~514.5 eV in the insulating and metallic states. The presence of this spectral feature at lower binding energies was proposed to signify non-local screening from coherent 3d1 states near EF in the metallic phase of VO254,55. Interestingly, the intensity of P1 in the insulating state increases with increasing carrier density and decreasing VO2 thickness in the heterostructures. The emergence of this peak in the insulating state spectra for VO2 heterostructures suggests that the additional charge transferred to the VO2 channel layer enables non-local screening that was previously observed only in the metallic phase of VO2.

To further quantify the evolution of the P1 peak across the metallic and insulating phases, we compared the metallic and insulating state spectra of 1.5 nm and 7.5 nm VO2 heterostructures in Supplementary Fig. 13. The intensity of the P1 peak in the insulating state of the 7.5 nm VO2 heterostructure is observable but small in comparison to the P1 peak in the metallic state (Supplementary Fig. 13a). The intensity difference spectrum shows a large difference at ~514.5 eV (Supplementary Fig. 13c) further confirming that the non-locally screened shoulder is negligibly small in the insulating state when compared to the metallic state. Remarkably, the non-locally screened shoulder is quite predominant in the insulating state spectra of 1.5 nm VO2 heterostructure (Supplementary Fig. 13b). The intensity difference spectra between the metallic and insulating states show a very small difference between the two phases (Supplementary Fig. 13d). These trends are consistent with the differences in the carrier densities between the metallic and insulating phases. The carrier density ratio between the metallic and insulating states is about 10 for the 1.5 nm VO2 heterostructure compared to about 1000 for the 7.5 nm VO2 heterostructure.

The second remarkable feature in the photoemission data is a high binding energy spectral feature at ~517.5 eV. This feature is associated with the V 2p3/2 peak and labeled P2 in Fig. 5a, b. A corresponding feature is also observed for the V 2p1/2 peak but is more smeared out and appears as broadening on the higher-binding-energy side at ~525 eV. In general, higher binding energy spectral features are associated with higher oxidation states in photoemission. The presence of V5+ is a possibility. However, an increase in the oxidation state from V4+ to V5+ cannot explain the observed electron-doping in VO2 heterostructures. Generally, electron doping should decrease the V4+ oxidation state in VO2, and therefore, an increase in the oxidation state of vanadium cannot explain the increase in electron density in the VO2 heterostructures. Therefore, V5+, even if present, has no bearing on the MIT observed in heterostructures.

P2 was also observed in VO2 samples capped with 2 nm LAO (Supplementary Fig. 14). Therefore, we have also inspected the La 3d5/2 and Al 1s spectra to look for any chemical shifts associated with a redox or chemical reaction at the VO2/LAO interface. As shown in Supplementary Fig. 15, there are no observable changes to the spectra across heterostructures. Finally, interfacial oxygen vacancy creation remains a possibility. However, any oxygen vacancy creation should lead to V3+ and an associated low-binding-energy feature in both the metallic and insulating state spectra. However, the spectra do not show any signatures of oxygen vacancies in VO2 heterostructures. Therefore, we rule out any oxygen-vacancy induced carrier doping in these heterostructures.

Furthermore, the intensity of P2 is carrier density dependent. For all V 2p spectra in the metallic state, where the carrier density is independent of the VO2 thickness, the intensity of this additional peak relative to the main V 2p3/2 peak is also independent of the VO2 thickness, with the intensity ratio of P2 to V 2p3/2 being close to 1. Contrastingly, the intensity of P2 increases with the decreasing film thickness in the insulating state of VO2. The intensity ratio of P2 to V 2p3/2 approaches the intensity ratio observed for the metallic state spectra at the highest carrier density of ~5 × 1021 cm−3 in the insulating state. These carrier-density-dependent changes suggest that this new spectral feature is intrinsic to the heterostructure. However, this additional spectral feature might benefit from further spectroscopic investigation with complementary techniques such as X-ray absorption spectroscopy to confirm its origins.

Furthermore, since there is a strong correlation between electron density and P2 peak intensity, we performed an LDA + DMFT Anderson impurity model calculation for the undoped and electron-doped VO2 to examine the V2p XPS spectral changes with electron doping. The calculation of the V2p XPS spectrum shows that electron doping creates a satellite peak at approximately 517.5 eV (see Supplementary Note 2 and Supplementary Fig. 16 for details of the calculations). Since the V-O covalency for the V3+ (d2) state is weaker than V4+ (d1) one, the binding energy of V3+ in electron-doped VO2 increases due to a weaker bonding and anti-bonding splitting in the XPS final states. Therefore, we assign the P2 peak to a satellite induced by electron doping.

The combination of electron transport and HAXPES data show that VO2 heterostructures facilitated effective modulation-doping and carrier densities as high as 5 × 1021 cm−3 could be achieved using this approach. The highest carrier densities correspond to electron doping of ~0.2 e-/vanadium. This is an extremely high dopant density at which conventional rigid band models predict metallization in doped correlated insulators56.

Bulk-sensitive valence-band HAXPES spectra recorded for the same set of heterostructures as discussed in Fig. 5 are shown in Fig. 6 and Supplementary Fig. 17. The insulating-state spectra for all VO2 heterostructures (blue) exhibit nearly zero spectral intensity at the Fermi level while an appreciable non-zero spectral intensity is observed for the higher-temperature metallic-state spectra (orange). These spectra further confirm that VO2 continues to undergo an MIT even in the presence of electron densities as high as ~0.2 e−/vanadium. The presence of MIT at such high doping levels, without any observable changes in the lattice parameters (Fig. 2a and Supplementary Fig. 4), points to a possible renormalization of the electronic structure with doping.

Fig. 6: Robust MIT in VO2 at high carrier densities.
figure 6

A comparison of the V 3d valence band (VB) spectra of modulation-doped VO2 heterostructures for the insulating (200 K, blue) and metallic states (at 310 K, orange) for different VO2 film thicknesses of a 7.5 nm VO2 film and b 7.5 nm, c 4.5 nm, d 3.5 nm, and e 1.5 nm VO2 heterostructures. There is a clear spectral weight shift across the MIT for all the samples with the insulating state being robust even for the heterostructure with a VO2 thickness of 1.5 nm, corresponding to carrier doping of ~0.2 e−/Vanadium.

Full size image

What is unclear however is the presence or absence of the structural phase transition. To probe this, we performed temperature-dependent XRD measurements at 320 K (metallic phase) and 200 K (insulating phase) for a 9.5 nm VO2 thin film and 9.5 nm and 4.5 nm VO2 heterostructures. Figure 7a shows that there is a clear shift in the angular position of the out-of-plane Bragg reflection. This was attributed to the rutile to monoclinic phase transition in previous studies11. Similar changes to the out-of-plane Bragg reflection were also observed for the 9.5VO2-het and 4.5VO2-het across the MIT (Fig. 7b, c). Additionally, the X-ray diffractogram measured at 280 K for the 4.5VO2-het shows rutile phase characteristics akin to that observed for the diffractogram measured at 320 K (Fig. 7d) suggesting that the electronic phase transition is concomitant with the structural phase transition.

Fig. 7: Temperature-dependent X-ray diffractograms of VO2 thin films and heterostructures.
figure 7

X-ray diffractograms of a 9.5VO2, b 9.5VO2-het, and c 4.5VO2-het measured in the insulating phase (at 200 K) and the metallic phase (at 320 K) of VO2. VO2 (\(\bar{4}02\))M and (002)R reflections can be clearly distinguished for all the heterostructures studied in this work. The \((\bar{4}02)\)M peak is the out-of-plane (of the substrate) Bragg reflection in the monoclinic phase of VO2 while (002)R is the out-of-plane reflection in the rutile phase. d A comparison of temperature-dependent XRD spectra of the 4.5VO2-het film measured at 200 K, 280 K, and 320 K. The XRD spectrum measured at 280 K shows rutile phase characteristics like the one measured at 320 K suggesting that VO2 remains in the rutile phase at 280 K in the 4.5 nm VO2 heterostructure.

Full size image

Moreover, in Supplementary Fig. 18, we have also compared the XRD spectra of 9.5VO2 and 9.5VO2-het both in the insulating and metallic phases. It is clear from the figure that there is an excellent overlap of the diffractograms for the film and heterostructure, including thickness fringes, for both the structural phases. This is evidence that modulation-doping decreases the phase transition temperature but does not suppress the structural transition. The presence of the structural phase transition even at carrier densities as high as 5 × 1021 cm−3 is suggestive of a strong electron-lattice coupling in VO2.

In summary, we demonstrated a purely electronic control of the MIT in modulation-doped VO2 heterostructures. Our work shows that the insulating state in VO2 is surprisingly robust even in the presence of electron doping as high as 0.2 e-/vanadium. Notably, all the films meet the Mott criterion (\({a}_{B}\cdot {n}_{C}^{\frac{1}{3}} > 0.25\), where aB is the effective Bohr radius and nC is the carrier density). Therefore, metallicity is expected at all temperatures based on the carrier densities achieved in these experiments. We note that a similar robust insulating state had also been found in modulation-doped nickelate thin films31. Perhaps, the development of theoretical models that go beyond the conventional carrier concentration independent rigid-band models will be required to understand electronic phase transitions in correlated electron oxides. An alternate explanation could be that the insulating state is favored at lower thicknesses due to interfacial disorder-induced Anderson-like localization of carriers, which will be more pronounced for the thinnest films. Further experiments will be needed to assess whether the insulating state is stabilized by the interfacial disorder.

A remarkable feature of this work is the possibility of bulk metallization in modulation-doped VO2. In general, in band semiconductors such as SrTiO3, conductivity modulation is achieved over a thickness of 1–2 nm in the vicinity of the channel/spacer interface33,57. In this work, a sharp MIT is observed for heterostructures with VO2 thicknesses as high as 9.5 nm, which are much higher than the Thomas Fermi screening length of ~1 nm. This is suggestive of the entirety of the film being metallized at the lowered TMIT after modulation-doping, even though the charge transfer densities are the highest at the interface (within 1–2 nm of the interface). While further studies are required to establish this beyond doubt, interfacial-doping induced bulk metallization of correlated electron insulators has implications for low-power electronics with high ON/OFF ratios32,58.

Finally, our work shows that modulation-doping is a powerful technique for achieving high carrier densities, close to those possible with elemental doping. Since our approach does not need any epitaxially matched spacer and dopant layers, it expands the library of materials that can be explored for the study of modulation-doping-induced electronic phase transitions of other related CEMs including complex oxides2 and pyrochlores59. This methodology, therefore, paves the way for exploring ‘pure’ electronic effects in correlated oxides and related systems. Such studies could also enable a fundamental understanding of band matching and relevant energy scales in complex oxides and, perhaps enable the discovery of new interfacial phases and devices that rely on phase transitions.

Methods

Prior to deposition, single-crystalline rutile TiO2 (001) substrates (Shinkosa, Japan) were treated using the procedure discussed previously60. All thin film samples were deposited using PLD (NEOCERA) with a 248 nm KrF laser. All VO2 thin films (both pristine films and heterostructures) were deposited on treated TiO2 substrates from a sintered V2O5 target with a laser fluence of ~1.5 J/cm2, 8 mTorr of oxygen pressure, and a growth rate of ~4.7 × 10−2 Å/pulse at a substrate temperature of 425 °C60. For all heterostructure samples, 2 nm thick LAO spacer layers were deposited at 10 mTorr of O2 pressure at a growth rate of ~5 × 10−2 Å/pulse using a single-crystalline LAO target (Shinkosa Japan). 3 nm thick TiO2-x dopant layers were then deposited using a TiO2-x single-crystalline target (Shinkosa Japan) at a growth rate of ~4.2 × 10−2 Å/pulse in 10−5 Torr of background vacuum. Finally, a 1 nm thick LAO layer was deposited under the same conditions used for the LAO spacer layer. Depositions of the spacer, dopant, and capping layers were all done at room temperature at a laser fluence of ~1.2 J/cm2. For all depositions, the substrate-to-target distance and the laser pulse frequency were maintained at 55 mm and 2 Hz respectively.

High-resolution Cu-Kα X-ray diffraction spectra for both the pristine and heterostructure VO2 films were recorded in standard θ–2θ geometry using a Rigaku Smart Lab X-ray diffractometer. LEPTOS 7.8 software by Bruker was used to determine film thicknesses and to calculate the differential strain between pristine and heterostructure films.

Cross-sectional scanning transmission electron microscopy (STEM) imaging and energy dispersive x-ray spectroscopy (EDS) mapping were performed using TITAN Themis microscope (60–300 kV) equipped with a probe corrector and super-X four quadrant EDS detector. The high angle annual dark field (HAADF)-STEM images were acquired at an operating potential of 300 kV with a convergence angle of 24.5 mrad, 160 mm camera length, and a dwell time of 12 μs per pixel. The images were further processed with Gatan digital micrograph software. The EDS maps were acquired using Velox software under similar microscopic conditions with a dwell time of 2 μs per pixel.

Sheet resistance vs temperature measurements were performed in Van der Pauw geometry using Keithley 2450 SMU and Eurotherm 2408 PID temperature controller. Continuous temperature scanning was carried out at a rate of 4 K/min for both heating and cooling cycles.

To extract carrier density and mobility, Hall measurements for all films and heterostructures were performed using Van der Pauw geometry in a PPMS-Dynacool equipment from Quantum Design and Keithley SMU 2450 from Tektronix. For these measurements, the magnetic field was swept from 0 T to 2 T to −2 T to 0 T at a scan rate of 100 Oe/s for different temperatures ranging from 200 K to 320 K.

High-resolution hard X-ray photoelectron spectroscopy (HAXPES) measurements61 were carried out with an incident photon energy of 6.2 keV at the sample temperatures of 200 K (insulating phase) and 310 K (metallic phase). Binding energy calibration was carried out using a high-resolution Fermi-edge measurement on a standard Au sample. Core-level and valence-band spectra were measured using a wide acceptance angle SPECS Phoibos 225HV hemispherical electrostatic analyzer in a near-normal emission geometry. The total energy resolution was estimated to be approximately 320 meV. Preliminary HAXPES measurements and sample screening were carried out using a lab-based HAXPES instrument at Temple University equipped with a wide acceptance angle ScientaOmicron EW4000 analyzer at a total experimental energy resolution of 450 meV.

To realize the origin of the P2 peak (at 517.5 eV) in the V 2p core level spectra (see Fig. 5), the simulations were performed on the electron-doped VO2 system using LDA + DMFT Anderson impurity model calculations. Details of the calculations are presented in Supplementary Note 2 and Supplementary Fig. 16.