Introduction

The field of complex oxides has been largely dominated by perovskite oxides, which are among the most extensively studied ABO3 phases. Their widespread interest stems from their diverse functional properties, including ferroelectricity, colossal magnetoresistance, multiferroicity, and metal-insulator transitions (MITs). This versatility arises from their inherent instability to energy-lowering structural distortions, ultimately governed by the imbalance of ionic radii at the A and B sites, which is quantified by the Goldschmidt tolerance factor1,2. In nearly half of all perovskites, this instability manifests as an antiferrodistortive octahedral rotation pattern, a−a−c+, resulting in an orthorhombic phase with space group Pbnm3,4. Interestingly, hexagonal oxides exhibit similar tolerance factors and often compete with orthorhombic polymorphs, as for example in the nickelates5. However, when the ionic radii of the A and B cations become more comparable, i.e., rionic(A) ≈ rionic(B), the system transitions towards the regime of ordered corundum compounds6. This parent crystal structure is illustrated in Fig. 1a, d with the \({{{\rm{R}}}}\bar{3}{{{\rm{c}}}}\) space group, of which Al2O3, Cr2O3 and V2O3 are most prominent examples. By defining A and B to be non-equivalent and ordered in the unit cell, the \({{{\rm{R}}}}\bar{3}\) ilmenite and polar R3c lithium niobate (LNO) symmetries can be established, as shown in Fig. 1b, c, respectively. These layered corundum compounds differ both structurally and chemically from the class of perovskites. Their small tolerance factor carries profound implications. First, they exhibit a high crystal density, where both A and B cations reside in equivalent octahedral environments. Second, the presence of face-sharing octahedra results in extremely short A–B distances, facilitating direct metal–metal bonding-distinct from the hybridization of oxygen atoms in A–O–B bonds found in perovskite materials. Lastly, the A and B sites can be occupied by similar transition metals, giving rise to strong electron correlations and associated phenomena, such as superconductivity and MITs7.

Fig. 1: Crystal structures of the ordered corundums.
Fig. 1: Crystal structures of the ordered corundums.
Full size image

Vanadium, chromium and oxygen atoms are represented in red, blue and gray, respectively. a, d Conventional and primitive corundum unit cell (\({{{\rm{R}}}}\bar{3}{{{\rm{c}}}}\)). The 6 atomic layers are highlighted as well as the ABC stacking. b, e conventional and primitive ilmenite unit cell (\({{{\rm{R}}}}\bar{3}\)). c, f Conventional and primitive LNO unit cell (R3c).

In light of these considerations, the stabilization of new ordered corundum thin films offers intriguing opportunities in the field of functional oxides. On the one hand, the combination of two magnetic sublattices with strong electron correlations can lead to MITs; on the other hand, new multiferroic compounds can be stabilized if ordered in the polar space group R3c8,9. All together, polar metals with magnetoelectric coupling can be targeted10,11. To date, ordered corundum materials have been primarily prepared as powders by solid-state12,13,14,15,16,17,18,19 or wet-chemical synthesis20,21,22 methods, and as thin films starting from mixed precursors23,24,25,26. As an alternative approach, we propose to leverage the atomic-layered growth via oxygen-assisted molecular beam epitaxy (MBE) to enable sub-unit cell engineering of corundum compounds and the stabilization of ordered corundum phases. This method provides several key advantages: access to a wider variety of A–B cation combinations, the opportunity to design tailor-made superlattice configurations, and the possibility of device integration.

In this work, we experimentally demonstrate the feasibility of constructing Cr2O3/V2O3 superlattices with sub-unit-cell stacking. As the ultimate realization of this approach, a 1-monolayer (1 ML) superlattice is obtained, which, to the best of our knowledge, represents the first stabilization of the ilmenite phase of CrVO3. The ordering of the alternating layers in this sub-unit cell regime has been directly evidenced by scanning transmission electron microscopy (STEM) combined with energy dispersive X-ray spectroscopy (EDX) and electron energy loss spectroscopy (EELS). Additionally, Raman spectroscopy has been employed to confirm symmetry lowering in these superlattices, as well as to verify the local ordering of the 1 ML CrVO3 superlattice into the \({{{\rm{R}}}}\bar{3}\) symmetry of the ilmenite structure. Finally, first-principles calculations predict that CrVO3 is a ferromagnetic insulator, consistent with previous studies9,27.

Results and discussion

Sub-unitcell engineered CrVO3

CrVO3 thin films were deposited by oxygen-assisted MBE on 1 × 1 cm Al2O3-(0001) substrates at 700 °C in ultra-high vacuum (chamber base pressure 10−10 mbar). The molecular oxygen partial pressure was kept constant at 2 × 10−6 mbar during the deposition, as measured with a residual gas analyzer connected to the growth chamber. The negative in-plane lattice mismatch (compressive strain) of bulk V2O3 and Cr2O3 with the substrate is −4.2% and −4.3%, respectively. Therefore, a Cr2O3 buffer layer was always deposited first on the Al2O3 surface and followed by fully relaxed CrVO3 thin films28. The fluxes of evaporated Cr and V were calibrated with a quartz crystal microbalance at 0.03 Å/s, while reflection high-energy electron diffraction (RHEED) was used to monitor in situ the layer-by-layer epitaxial growth of the crystalline oxide films. Figure 2a shows the RHEED pattern at the end of the growth of a Cr2O3 buffer layer, confirming the epitaxial growth of the film. The RHEED intensity can be recorded as a function of time as represented in Fig. 2b: the measured periodic oscillations indicate the layer-by-layer growth regime of the corundum thin film, and the period of about 40 s corresponds to the time required to form a complete single atomic layer at the calibrated growth rate. The RHEED oscillations, monitored during the growth of the Cr2O3 buffer layer, were used to calibrate and fix the shutter-switching times for the entire growth of the superlattices. The superlattice structures were obtained by alternating the opening and closing of the Cr and V cell shutters according to the desired periodicity. The targeted structures consist of a sequence of the following repetition units: 3V + 3Cr atomic layers, 2V + 2Cr atomic layers, and 1V + 1Cr atomic layers. Finally, a CrVO3 thin film with a 1:1 Cr:V stoichiometry was grown by co-deposition of Cr and V as a benchmark for the superlattice structures. Simple Cr2O3 and V2O3 epitaxial thin films on sapphire are used as a ref. 29.

Fig. 2: Structural and EELS characterization of the superlattice thin films.
Fig. 2: Structural and EELS characterization of the superlattice thin films.
Full size image

a RHEED pattern at the end of the growth of the Cr2O3 buffer layer along the \(\langle 1\bar{2}10\rangle\) direction. The blue rectangle indicates the region where the intensity has been recorded as a function of time. b RHEED intensity as a function of the deposition time. Each periodic oscillation corresponds to the completion of a single atomic layer. c HRXRD (ω−2θ scan) of the (006) peak of the corundum substrate, buffer layer and superlattice thin films. d Cross-sectional HAADF-STEM image of the 3 ML superlattice thin film along the \(\langle 1\bar{2}10\rangle\) ZA. The interface with the Cr2O3 buffer layer is visible in the bottom part of the image. The inset plot represents the HAADF intensity along the z-direction and averaged across the width of the image. e STEM-EELS spectra near the V L-edge, O K-edge and Cr L-edge measured over an area of several nm2 across the thickness of the thin films. The V L3-edge is used to align and normalize the intensity of the spectra. The Cr2O3 and V2O3 references were obtained from epitaxial thin films grown in the same MBE system. f XRR of the series of superlattice thin films.

The epitaxial growth of the buffer layer and of the superlattice thin film on the sapphire substrate is also confirmed by high-resolution X-ray diffraction (XRD) ω−2θ scans and high-angle annular dark-field STEM (HAADF-STEM) images in Fig. 2c, d. Additionally, the STEM-EELS spectra, measured across the entire thickness of the films, confirm the oxidation states of Cr3+ and V3+. The V3+ oxidation state is confirmed by comparison of the V L-edge fine structure with both the V3+ reference spectrum plotted in Fig. 2e and the previously reported data by Lin et al.30 and Hébert et al.31. Similarly, the Cr3+ oxidation state is verified using the Cr3+ reference spectrum shown in Fig. 2e and by applying the method proposed by Arévalo-López et al.32, in which the intensity ratio between the Cr L3 and L2-edges is evaluated. The measured L3/L2 ratio, ranging from 1.68 to 1.75 across the series of superlattice thin films, lies well within the expected interval for Cr3+, thereby confirming its trivalent state. It is also noteworthy that the buffer layer exhibits the Cr3+ oxidation state. The fact that both Cr and V are stabilized in the 3+ oxidation state is particularly significant for two reasons. First, it rules out other common vanadium oxidation states, such as V4+ and V5+33,34. Second, it marks a difference with existing M2+TiO3 and M2+MnO3 ilmenites, where two metals are arranged in a layered structure with 2+ and 4+ oxidation states, respectively12,15,18,35. Possibly, this can be attributed to the high stability of the Cr3+ cation36 in comparison to Cr2+ and Cr4+. As a result, V3+ seems to be stabilized when confined between Cr2O3 layers. A logical extension would be to examine vanadium in combination with elements that are readily oxidized to M2+ and/or M4+ states.

Compared to the Cr2O3 buffer layer, the superlattice thin film exhibits a distinct oscillation of the HAADF signal along the out-of-plane direction, attributed to the different atomic number (Z) contrast between Cr and V. However, such contrast remains weak to effectively highlight and measure the periodicity of Cr–V superstructures with shorter periods (see Supplementary Fig. 1) because of the small Z difference.

In addition to the (0006) and (00012) reflections that emerge from the ABC stacking in the corundum structure with space group \({{{\rm{R}}}}\bar{3}{{{\rm{c}}}}\), additional diffraction occurs in the 3 ML and 2 ML superlattices due to the artificially introduced periodicity. This results into higher-order reflections in the X-ray reflectivity (XRR) (see Fig. 2f) with a d-spacing of d3ML = 16.5 ± 0.4 Å and d2ML = 11.0 ± 0.4 Å, respectively. On the other hand, the removal of the glide reflection in the 1 ML superlattice with expected \({{{\rm{R}}}}\bar{3}\) space group symmetry would lead to a (0003) reflection in the diffractogram at 2θ ≈ 19° corresponding to d1ML = 4.7 Å. However, such a reflection proved to be hard to detect, possibly due to the background and noise levels.

To confirm the measured periodicity and evaluate the interface quality of the periodical superstructures, STEM-EDX was performed. Figure 3 includes the elemental maps of all three superlattice thin films, as well as the corresponding line profile calculated along the out-of-plane direction. The Cr and V signals are periodic, with the Cr maxima matching with the V minima. The period of the superlattice was calculated as the average distance between the Cr and V maxima, including the standard deviation to quantify its variability. In particular, the periods d3ML = 16.3 ± 0.9 Å and d2ML = 11.0 ± 1.2 Å are found to be in agreement with XRR. On the other hand, the 1 ML film presents a Cr–V periodicity only locally, d1ML = 7.9 ± 0.7 Å, explaining why the (0003) reflection was harder to detect in the X-ray diffractograms. Elemental mapping was also performed by STEM-EELS on the 2 ML and 1 ML samples. An absolute quantitative analysis is challenging due to the overlap of the V, Cr, and O edges (see Fig. 2e), but a relative estimate of the spatial distribution of Cr and V can be obtained by model-based quantification with the pyEELSmodel software37. The resulting spatial distributions of Cr and V along the out-of-plane direction in the 2 ML and 1 ML superlattices are plotted in Fig. 4a, b together with the HAADF signal. STEM-EELS confirms the ordering of the 2 ML superlattice (period d2ML = 10.9 ± 1.3 Å) and the local cation ordering in the 1 ML thin film. However, the alignment of the Cr–V spatial distribution with the HAADF signal also highlights the errors in the 2Cr + 2V layers stacking; e.g., the Cr (or V) peak occasionally spans over 3 HAADF peaks corresponding to 3 atomic layers. The deviations from the designed Cr–V stacking periodicity can be explained by extrinsic factors, such as instabilities of the Cr and V evaporation fluxes during the deposition (Supplementary Figs. 2 and 3). However, intrinsic factors like interdiffusion cannot be ruled out, and they can impact the quality of the superlattice interfaces. Notably, antisite defects are expected to be structurally more favorable in ABO3 ordered corundum structures than perovskites because of the similar oxygen environments for A and B cations. Both sites have a six-fold oxygen coordination in a distorted octahedra, while in the perovskite lattice, A and B are surrounded by 12 and 6 O atoms, respectively. In the specific case of CrVO3, we have shown that Cr and V have also the same valence state (Fig. 2e). In addition, Zhao et al.9 calculated that the properties of ordered ABO3 corundums are not affected by the swap of A and B cations, suggesting a limited impact of antisite defects.

Fig. 3: Cross-sectional STEM-EDX chemical maps.
Fig. 3: Cross-sectional STEM-EDX chemical maps.
Full size image

All data were acquired along the \(\langle 1\bar{2}10\rangle\) zone axis, with the 3 ML, 2 ML, and 1 ML samples arranged from left to right. The top row contains the HAADF-STEM images with the corresponding STEM-EDX maps, and the bottom row shows the intensity profiles extracted from the maps. V and Cr distributions are mapped using the intensity of their respective Kα lines. The 3 ML sample map was processed with a Wiener filter, while the corresponding intensity profile presents the unfiltered data. At the bottom, a portion of the Cr2O3 buffer layer is also visible. To minimize the impact of surface steps on the chemical maps, the 2 ML and 1 ML samples were imaged from FIB lamellae thinner than 20 nm. Given the short interlayer distances of the 2 ML and 1 ML films, multi-frame imaging and spectroscopy data were acquired and aligned to correct for sample drift and scanning distortions. All scalebars represent 5 nm.

Fig. 4: STEM-EELS and Raman measurements confirm the periodicity of the superlattice thin films.
Fig. 4: STEM-EELS and Raman measurements confirm the periodicity of the superlattice thin films.
Full size image

a, b HAADF-STEM and EELS profiles of the 2 ML and 1 ML superlattices. Vertical dashed lines are drawn matching the peaks in the HAADF signal, each corresponding to a single layer of cations. c Raman spectra of the superlattice thin films, compared to pure V2O3, Cr2O3, and a co-deposited CrVO3 film. All the V2O3 and CrVO3 films are grown on a Cr2O3 buffer layer. As discussed in the text, an additional Ag mode appears in the CrVO3 samples around 685 cm−1. The intensity of this mode is normalized to the intensity of the A1g mode at 550 cm−1 and is shown in d. The error bars represent the standard deviation of the signal noise. e Illustration of the Raman-activated Ag mode. The expansion (contraction) of the red (blue) octahedra is represented by the DFT-calculated eigenvectors (black arrows).

Phonon spectroscopy to probe symmetry breaking

The introduced local symmetry breaking in the superlattices was evaluated by polarized Raman spectroscopy. As shown in Fig. 4, an additional band is detected around 685 cm−1 in the CrVO3 samples in comparison to the sesquioxides Cr2O3 and V2O329,38. This is the highest observed frequency mode in the range 100–1150 cm−1, and its intensity increases as the superlattice period is reduced (Fig. 4d). A similar observation was made by Baraton et al.21 for FeCrO3 powders obtained by co-precipitation, where they identified a new peak, the highest-frequency phonon mode around 700 cm−1, which was not observed in the parent oxides. As noted by the authors, the presence of the extra Raman band is a proof of cation ordering. The additional band is assigned to the inactive A2g mode of A2O3 corundum (\({{{\rm{R}}}}\bar{3}{{{\rm{c}}}}\) space group), which becomes the Raman active Ag mode in the ABO3 ilmenite structure (\({{{\rm{R}}}}\bar{3}\) space group). Both modes correspond to the contraction and expansion of AO6 and BO6 octahedra in successive (0001) planes (Fig. 4e), resulting in the inactive A2g if A=B in the parent A2O3 corundum oxide, but Raman active (Ag) if A≠B in the ABO3 ilmenite structure. The same Ag assignment was made by Wang et al.39 for MgTiO3: 10 Raman modes are observed in the Raman spectrum as predicted by group theory and match the density functional theory (DFT) calculated frequencies for an ilmenite structure. The Ag mode is detected at 715 cm−1. Similarly, we calculated the frequencies of the Raman active modes in ilmenite CrVO3 (Supplementary Fig. 4): the calculated highest frequency mode is the Ag vibration described above, and we assign it to the highest observed frequency mode at 685 cm−1. To further confirm the Ag assignment, the Raman spectra were also collected as a function of the polarization angle (Supplementary Fig. 5). Consistent with the selection rules for the Ag mode, the band intensity exhibits a pronounced and gradual decrease when transitioning from parallel (VV) to cross (VH) polarization, providing a more robust evidence for the mode assignment.

Two additional A2g modes are expected to be Raman active in the ilmenite \({{{\rm{R}}}}\bar{3}\) symmetry with their expected frequencies in the ranges ~186 and ~395 cm−1 as predicted by DFT (Supplementary Table 1). The first overlaps with the A1g-derived Ag mode, giving shape to an asymmetric peak (Supplementary Fig. 6). The second was not detected, either due to a smaller Raman polarizability or an overlap with the peaks of the buffer layer. Taking all the factors into account, we have demonstrated the presence of \({{{\rm{R}}}}\bar{3}\) symmetry features in the Raman spectra of all the superlattice thin films and the co-deposited CrVO3. The last question to address is why such features are not observed solely in the 1 ML sample, which is designed to adopt the ilmenite structure with \({{{\rm{R}}}}\bar{3}\) symmetry. It is plausible to argue that the appearance of these features is caused by the symmetry breaking at the interfaces between the Cr2O3 and V2O3 layers. The same activation of a Raman inactive mode has already been observed in perovskite-based superlattices by Seong et al.40. Such an interpretation would also be consistent with the increase in intensity of the Ag mode as the density of interfaces in the superstructures rises (i.e., with the decreasing period of the superlattice), as shown in Fig. 4d. In the broader context, it indicates that the cations in the 1 ML sample exhibit at least local ordering, as previously demonstrated by STEM-EDX and STEM-EELS. Furthermore, it suggests the presence of a self-ordering mechanism in the co-deposited sample, though less effective than the periodic opening and closing of the shutters, as evidenced by the weaker Ag peak. This result should not be surprising for two reasons. First, the structural stability of ilmenite CrVO3, as calculated in this work, aligns with previous literature findings9,27. Second, ilmenite structures have already been successfully synthesized using methods, such as solid-state reactions and co-precipitation, which do not impose superstructure periodicity and must therefore rely on a form of self-ordering12,15,18,19,20,21,41,42.

Ferromagnetic insulating ilmenite

By DFT calculations, the relative stability of both the non-centrosymmetric ilmenite and the polar LNO structure were evaluated as a function of Hubbard correction (U), as shown in Fig. 5a. It can be anticipated that a Hubbard value of 3 eV is considered an appropriate approximation for CrVO3, based on prior works of V2O3 and Cr2O343,44. Henceforth, it is inferred that the ilmenite structure is energetically favored. Subsequently, the electronic properties were evaluated, as depicted in Fig. 5a. An insulating behavior is observed with an estimated energy gap of 1.28 eV. Finally, the magnetic ordering and main magnetic interactions were evaluated by considering five different short-period spin arrangements (see Supplementary Fig. 7). By an energy comparison, it is shown that FM ordering is energetically favored above 2 eV, with magnetic moments MV = 2.02 μB and MCr = 3.07 μB, in excellent agreement with prior work9. Although a strong exchange interaction of 52.2 meV exists along the short Cr–V dimer, the relative energy stability of FM is only 14 meV/f.u. compared to the lowest AF state (at U = 3 eV). This might suggest a relatively weak FM ordering.

Fig. 5: Functional properties of CrVO3 with \({{{\rm{R}}}}\bar{3}\) ilmenite structure.
Fig. 5: Functional properties of CrVO3 with $${{{\rm{R}}}}\bar{3}$$ R 3 ¯ ilmenite structure.
Full size image

a Energy comparison of the ilmenite (\({{{\rm{R}}}}\bar{3}\)) and the LNO (R3c) structures as a function of the Hubbard correction U for CrVO3. The background color and the crystal models indicate what structure is the most stable in the two regions of the phase diagram. b The Cr and V contributions to the partial DOS of ilmenite CrVO3. c Resistivity of the CrVO3 1 ML superlattice thin film. d Magnetization of CrVO3, V2O3 and Cr2O3 thin films as a function of the external magnetic field H at 10 K and with H parallel to the in-plane lattice direction.

Following up on the DFT calculations, a preliminary investigation into the functional properties of the 1 ML superlattice thin film was performed. The resistivity measurement as a function of temperature shows a clear insulating behavior as the sample is cooled down to 50 K from room temperature (Fig. 5c). The thin film is confirmed to be insulating by UV–VIS spectroscopy (see Supplementary Fig. 8), but a reliable estimate of the band gap proved to be challenging due to the overlap with the optical features of the Cr2O3 buffer layer, i.e. its 3.1 eV band gap and crystal field excitations of Cr3+ ions at 2.1 and 2.6 eV45. At last, the magnetic properties of the CrVO3 1 ML superlattice were assessed and compared to Cr2O3, V2O3 and co-deposited CrVO3 thin films. The measured magnetization M as a function of the magnetic field H is shown in the Fig. 5d. All the samples present a weak ferromagnetic response along both field directions (in and out-of-plane) and at both probed temperatures (300 and 10 K). The largest remnant magnetization is observed in the V2O3 thin film, in all the four combinations of H direction and sample temperature (see Fig. 5d and Supplementary Fig. 9); this is in contrast with the expected paramagnetic behavior of V2O346. This weak ferromagnetism could arise from extrinsic effects such as superparamagnetic clusters or glassy dynamics, which in V2O3 thin films may arise from magnetic impurities or defects associated with the epitaxial growth on sapphire. A similar ambiguity exists for ilmenite CrVO3; however, the CrVO3 thin films exhibit a larger coercive field (Hc ~ 85 Oe at 300 K and Hc ~ 130 Oe at 10 K for the 1 ML superlattice). Additional measurements are required to establish whether the increased coercive field reflects the presence of intrinsic long-range magnetic order.

Conclusions

The results presented in this work demonstrate an approach to the synthesis of ordered corundum oxides, leveraging the atomic layer-by-layer growth of epitaxial thin films. CrVO3 superlattice thin films have been grown by oxygen-assisted MBE, progressively reducing the thickness of the alternating Cr2O3 and V2O3 layers. Specifically, the M2O3 units consist of 3, 2 or 1 layer of cations, i.e., one half, one third or one sixth of the conventional corundum unit cell height. A detailed structural characterization showed how the superlattices can be controlled at the atomic scale, although the induced Cr–V alternation may be in competition with interdiffusion at the 1 ML limit at the growth temperature of 700 °C. Moreover, antisite defects are expected to be structurally more favorable if compared to perovskites due to the similar sixfold O-coordination of the Cr and V sites. Despite the challenges involved, this approach towards the synthesis of ordered corundum structures opens a promising path for the growth of artificial functional oxide thin films, whose properties depend on the layering configuration in the corundum lattice and the chosen pair of cations. First-principles calculations have shown the potential of 1 ML limit, i.e., ilmenite and LNO structures, while the properties of sub-unit cell superlattices need to be explored in detail in future studies.

Methods

The XRD ω−2θ scan and XRR experiments were conducted with a Rigaku TF-XRD-300 system, equipped with a dual rotating Cu anode and a HyPix-3000 detector. The system was operated with a Ge220 × 2 monochromator in the incident optics for the XRD measurements.

Raman spectra were acquired at room temperature using a 100× objective with a numerical aperture of 0.95. An output laser power of 40 mW was used to avoid the overheating effects and possible damage to the sample. Raman spectra were collected in the spectral range 65–1160 cm−1 using an 1800 g mm−1 grating. Under optimal conditions, the spectral resolution is 0.02 cm−1. In turn, the vibrational properties were also evaluated as a function of polarization angle. Here, the polarization analyzer detector is adjusted from 0° (parallel polarization, VV) to 90° (cross polarization, VH) against the incident polarization. Raw Raman data were processed with Witec Project Plus software (version 2.08).

Cross-sectional lamellae were prepared by focused ion beam (FIB) for STEM imaging and elemental mapping using DualBeam Nova 600 NanoLab, then analyzed with a cold-FEG JEOL ARM200F microscope operated at 200 kV, equipped with a STEM-Cs corrector, a Centurio EDX spectrometer and a Gatan Tridiem electron energy loss spectrometer. EELS spectra were collected with a collection semi-angle of 16.7 mrad and a beam current of 120 pA. For the 1 ML and 2 ML samples, thinner lamellas were prepared on a Dual beam FIB/SEM Helios Nanolab 650. A more detailed analysis was performed on a Themis Z microscope, which is an aberration-corrected STEM, Thermo Fisher Scientific Titan3 60–300 operated at an acceleration voltage of 200 kV, equipped with a high-brightness field-emission electron source (X-FEG) and an electron monochromator excited to 1.4, providing an energy resolution better than 200 meV while keeping a spatial resolution of 1 Å. EDX and EELS were acquired simultaneously using a collection semi-angle of 56 mrad and a beam current of 50 pA. Multiple EELS data were acquired on a prototype Iliad EELS Spectrometer with Zebra EELS detector developed by Thermo Fisher with a dispersion of 0.05 eV/pixel47. Core-loss EELS data analysis was performed using model-based quantification with the pyEELSmodel software37. The core-loss data were aligned using the zero-loss peak and spatially binned to improve the signal-to-noise ratio of the spectra and facilitate the fitting procedure. Relative estimates of spatial variations of Cr and V can be extracted using model-based fitting reliably. The model consisted of three parts. First, a convexity constrained linear background48. Second, core-loss edges using the cross sections for O, V and Cr49. Third, a component considering multiple scattering based on the low loss signal. For EDX elemental mapping of the 2 ML and 1 ML superlattices, multi-frame EDX data was acquired, using the simultaneously acquired HAADF signal.

The resistivity of the 1 ML superlattice thin film was measured in the Van der Pauw configuration with a Keithley 4200-SCS parameter analyzer between 50 and 300 K. UV–vis absorbance spectra were acquired using a Tecan Infinite 200 PRO microplate reader. A magnetic property measurement system based on a superconducting quantum interference device magnetometer was used to characterize the magnetic properties of a selection of thin films. The relatively strong diamagnetic signal due to the substrate was fitted in the high fields regions (5 kOe < ∣H∣ < 30 kOe) and subsequently subtracted from the measured magnetization (see Supplementary Figs. 10 and 11).

All DFT calculations were carried out using the Vienna ab initio simulation package50. The interactions between electrons and ions were treated using the projector augmented-wave method51, and the electronic wave functions were expanded with a plane-wave cutoff energy of 600 eV. The PBEsol exchange-correlation functional52 was employed throughout, while the localized nature of the 3d electrons of the cations was treated by a Hubbard correction U within the rotationally invariant approach proposed by Dudarev et al.53. Structural relaxations were performed with a force convergence criterion of 0.005 eV/Å, and the Brillouin zone was sampled using a Γ-centered 12 × 12 × 12 k-point mesh. For the electronic and magnetic calculations, a dense 24 × 24 × 24 mesh was used with an energy convergence criterion of 10−6 eV. The phonon dispersions were evaluated by density functional perturbation theory and with the use of the PHONOPY package54.