Introduction

Transition metal oxides (TMOs) offer a considerable number of emergent physical phenomena, such as superconductivity, magnetism and (multi)ferroicity. These phenomena originate from the complex interplay among charge, spin, orbital and lattice degrees of freedom associated with the d states of transition metals1,2,3. Due to their inherently coupled nature, even a small change in the orbital degree of freedom can bring about drastic evolution of the underlying electronic structure2,3,4,5. Therefore, finding a tuning knob for controlling the relative occupancy of d orbitals has been a central issue to tailor the intriguing physical properties of TMOs. The outcome of such a study, thus, not only deepens our fundamental understandings but also allows us to envision novel TMO-based electronic devices3,4,5.

Recent advances in atomic-scale synthesis of TMO heterostructures and artificial superlattices (SL) offer unprecedented opportunities for the control of orbital states that cannot be realized within bulk counterparts. Examples include (1) controlling the physical interaction and relevant orbital energy levels at the interface between a cuprate and a manganite, resulting in a systematic tuning of the orbital occupation5,6; (2) inducing high-Tc cuprate-like orbital states in a LaNiO3/LaAlO3 perovskite heterostructure by controlling the electron occupation of particular d states with the eg symmetry (x2-y2 and 3z2-r2)7,8,9; and (3) modifying the eg orbitals systematically to optimize the catalytic activities at the surface of TMOs10,11,12. The charge transfer needed for a catalytic reaction was found to be facilitated by the d-level near the chemical potential.

Up to date, however, the orbital control has been rather limited to a subset of d electronic states, i.e. either within t2g or eg states. Despite a huge number of reports on the control of the electron number within the respective t2g or eg states (e.g. Refs. 7,9), the artificial control of relative electron populations across the two subsets has not been realized because the t2g-eg separation, called 10Dq, is usually a few eV (in typical perovskite TMOs), much larger than the energy scale we can engineer e.g., with heterostructuring. In this Report, we demonstrate that a dramatic control over the occupation between the t2g and eg orbital states can be achieved in quasi-2-dimensional (2D) artificial SLs of a TMO heterostructure. The SLs are composed of LaCoO3 (LCO) as an active layer and LaAlO3 (LAO) as a spacer layer. LCO is a unique perovskite which has a “twin” of quantum many-body states. In the case of bulk LCO, a low spin (LS) state (t2g6; S = 0) is the ground state and a high spin (HS) state (t2g4eg2; S = 2) is the first excited state, with an energy difference less than 50 meV13,14,15 (see Supplementary Information for details of the discussions on the intermediate spin state16). At room temperature, bulk LCO has a mixed spin state of 40% HS + 60% LS due to thermal excitation13,14,15; namely, ∼1 electron (hole) occupies the eg (t2g) orbital in bulk LCO. Therefore, the room temperature electronic state of bulk LCO has been considered as a Mott insulator14,17. The energies of the twin states are in delicate balance and thus the electron populations of those states could be tuned with a minimal energy cost by applying strain or controlling dimensionality through heterostructuring.

Results

Figure 1 shows a schematic diagram of our approach to control the inter-multiplet orbital occupation in LCO. At the ultrathin limit, one of the eg orbital levels, namely, 3z2-r2 (z is the direction of the surface normal, i.e. along the crystallographic [001] direction) can be modulated to change the insulating nature of LCO. In particular, the quantum confinement effect can selectively increase the energy of orbitals concentrated along the z direction (3z2-r2 and xz/yz). Moreover, in heterostructures with a spacer layer such as LAO, certain chemical interactions at the interface between the LCO layer and the spacer layer can influence the strength of the Co 3d(3z2-r2)-O 2p hybridization3,8,18. Such interactions can affect the total energy of the HS state while keeping the total energy of the LS state almost unchanged. Thus, heterostructuring as in a quasi-2D oxide system offers a means to modify the delicate population balance between two quantum many-body states (HS and LS).

Figure 1
figure 1

Reducing dimensionality to control the d-orbital population in (LaCoO3)n(LaAlO3)n.

Two competing spin-orbital configurations, i.e. low-spin (S = 0; t2g6) and high-spin (S = 2; t2g4eg2), are shown for the Co3+ ions. The two spin states in 3D bulk LaCoO3 are nearly degenerate, resulting in a mixed-spin state at room temperature, whereas the low-spin state can be induced in quasi-2D superlattices by changing the energies of the 3z2-r2 and xz/yz states.

Full size image

This scenario in ultrathin LCO is clearly supported by the results of the dynamical mean field theory (DMFT) calculations19. Figure 2 comparatively shows the electronic structures of an ideal bulk (3-dimension; 3D) and a single layer (2D) LCO system (for details of the calculations, see Supplementary Information16). The total/orbital-resolved density of states (DOS) for the 3D and 2D systems are presented in Figures 2a and 2b, respectively. The corresponding momentum-resolved spectral functions are shown in Figures 2c and 2d. The energy distribution of the orbital states shows broad Co eg (narrow Co t2g) bands located between +1 and +4 eV (−1 and +1 eV). O 2p bands are mostly located below −2 eV. In the case of 3D LCO, a portion of the eg orbital is occupied (the red-shaded area in the DOS near E = −0.8 eV), while a portion of the t2g orbital is unoccupied (the blue-shaded area in DOS just above the Fermi level). This configuration manifests a mixed spin state in 3D LCO at room temperature14,15. In contrast, in the 2D case, it is clearly shown that the eg bands are empty and the t2g bands are almost fully occupied, suggesting an electron transfer from the eg orbital to the t2g orbital.

Figure 2
figure 2

Dynamical mean field theory results.

Total/orbital-resolved density of states (DOS) and spectral functions as a function of energy and momentum for (a, c) 3D and (b, d) 2D configurations, respectively. Each orbital-resolved DOS is normalized so that the total area becomes unity. A yellow arrow in (c) denotes the nearly dispersionless t2g band. Possible optical transitions α, β and γ are indicated by arrows.

Full size image

It should be noted that the total DOS near E = −0.5 eV is significantly enhanced as the dimensionality is reduced (3D → 2D; from Figure 2a to Figure 2b). This is primarily due to the increase in the number of oxygen ions that participate in the O p – Co d hybridization in 2D LCO; in 3D LCO, the total DOS comprises 3(O pσ) + 6(O pπ) + 2(Co eg) + 3(Co t2g), while in 2D LCO, 4(O pσ) + 8(O pπ) + (x2-y2) + (3z2-r2) + (xy) + 2(yz/zx). Therefore, the charge transfer nature becomes much enhanced in 2D LCO even with a small O pπ DOS at E = −0.5 eV.

The redistribution of the d-orbital occupation in dimension-controlled LCO yields a dramatic change in the nature of the electronic ground state from a Mott insulator to a band insulator. In the spectral function of the 3D LCO sample (Figure 2c), nearly non-dispersive and incoherent states (denoted by a yellow arrow) are observed just above the Fermi energy. These states have the Co t2g orbital character, reflecting that the 3D LCO is a Mott insulator, as we denoted the optical transition across the Mott gap by the arrow α. In contrast, the 2D LCO sample is found to be a band insulator, in which the gap at the Fermi level is defined by the energy separation between the fully occupied t2g and empty eg bands (instead of t2g); the corresponding optical transition is represented by the arrow β. It is also shown that the lower eg band edge along the Γ-X direction shifts to higher energy in the 2D LCO case. The blueshift of the transition β is presumably caused by the fact that the coordination of the Co3+ ion with the upper and lower layers is absent and the hopping in the out-of-plane direction is suppressed. The dimensional control can stimulate the electronic quantum phase transition as well as the sizable redistribution of the orbital occupation in the 2D LCO sample.

In order to experimentally validate the theoretical prediction on the control of orbital occupation, we designed TMO SLs composed of ultrathin LCO layers embedded within the large band-gap insulator LAO. Systematic dimensional control was conducted by reducing the periodicity of SLs below a few atomic unit-cells; namely, (LCO)n/(LAO)n SLs (n = 2, 6 and 10). The LCO SLs were fabricated on NdGaO3 (NGO) (110) substrates using pulsed laser epitaxy (see Supplementary Information16). Figure 3 shows x-ray diffraction reciprocal space maps of the three samples (n = 2, 6 and 10) near the (103) reflection of the substrate. The indices on the satellite peaks reflect the periodicity of the SLs. For more detailed structural analyses, see Supplementary Information16.

Figure 3
figure 3

X-ray diffraction reciprocal space maps of the three (LaCoO3)n(LaAlO3)n superlattices with n = 2, 6 and 10, near the (103) reflection of the substrate.

The indices on the satellite peaks reflect the superlattice periodicity. All of the H values in the HKL-coordinate of the superlattice peaks (red spots) were unity, suggesting negligible influence of strains on the atomic structure of the LaCoO3 superlattices.

Full size image

It is clearly shown that all of the H values in the HKL-coordinate of the SL peaks (red spots) were unity, indicating that the in-plane lattice constants were coherently maintained with respect to the substrate. This result suggests negligible influence of strain among the LCO SLs. Note that the lattice mismatch of bulk LCO (pseudocubic) and LAO on a NGO substrate is 1.6% and 1.9%, respectively. However, it has also been shown that the electronic structure in LCO thin films is rather insensitive to the lattice strain, while the magnetic properties showed strong dependence20. The insensitiveness against the strain might be due to the peculiar electronic structure of LCO. As shown in Figure 2, there exists O 2p density of states near the Fermi level. According to our DMFT study on the strain dependence (not shown), the strain mostly influences the O 2p energies but not much the Co 3d energies. This suggests that valence and consequently, ionic size of Co3+ can hardly change under external stresses. Therefore, we can tell unambiguously that the evolution in the electronic structure found in this work, is originated mainly from the dimensionality rather than the strain effects.

The evolution in electronic structure predicted by the DMFT calculations is verified experimentally using optical spectroscopy. Figure 4 shows optical conductivity (σ(ω)) of the three SLs. For comparison, σ(ω) of a 13 nm-thick LCO film is attached. We observed strong absorption features at ħω ∼3 eV (thin gray lines) and 1.5 eV (thin yellow triangles), together with a weak absorption feature at ħω ∼ 0.5 eV (filled red triangles). Based on the results of the DMFT calculations (Figure 2), we can attribute these features to a O p-Co d charge transfer transition (γ ∼ 3 eV), a Co d(t2g)-d(eg) transition (β ∼ 1.5 eV) and a Co d(t2g)-d(t2g) transition (α ∼ 0.5 eV), respectively. It is clearly observed that as n decreases, the weak absorption at ħω ∼0.5 eV (α) is systematically suppressed and disappears at (LCO)2/(LAO)2. Based on the DMFT results, the disappearance of the excitation α can be explained in terms of the crossover of the spin-orbital ground state from a mixed HS + LS configuration to a LS configuration with decreasing the dimension, which is generically accompanied by a crossover from a Mott insulator to a band insulator; the optical transition between the Hubbard bands of Co t2g states disappears with decreasing n, and the optical gap is defined by the transition from fully occupied Co t2g bands to empty eg bands. Also, the blueshift of transition β, predicted by the DMFT calculation results, is indeed observed in the experimental σ(ω), as denoted by the yellow triangles in Figure 4. The excellent agreement between theoretical and experimental results confirms that our SLs indeed undergo the dimensional crossover of the electronic structure, as illustrated in Figure 1.

Figure 4
figure 4

Optical conductivity spectra of (LaCoO3)n(LaAlO3)n superlattices (n = 2, 6 and 10) and a 13 nm-thick LaCoO3 film.

With decreasing n, the optical transition within Co t2g (i.e., t2g → t2g) was suppressed, opening a charge gap. Thin lines are fitting curves highlighting each peak.

Full size image

Discussion

Polarization-dependent O K-edge X-ray absorption spectroscopy (XAS) data further reveal the details of orbital occupation change in our (LCO)n/(LAO)n SLs with variation in the dimensionality (see the Supplementary Information for details16). Since the O K-edge XAS probes the transition to the unoccupied O 2p–Co 3d hybridized states, we can monitor the change in the 3d orbital occupancy of Co3+ ions directly. Figure 5a shows the isotropic XAS spectra ([(E // c) + 2 × (E // ab)]/3) measured at room temperature. The spectrum from a 13 nm-thick LCO film (gray dots) is included in the figure for comparison. The main features near the photon energy ħω = 529 eV and the shoulder near ħω = 528 eV are attributed to the unoccupied Co eg and Co t2g states, respectively15,21,22. Features near ħω = 532 eV reflect the Al 3sp states in the LAO substrate. In contrast to the case of thick LCO, the unoccupied Co t2g state nearly disappears with decreasing n, while the main structure of the unoccupied Co eg state is clearly enhanced as shown by the arrow in Figure 5a. This spectral evolution provides concrete evidence of the electron transfer from the eg state to the t2g state.

Figure 5
figure 5

(a) O K-edge XAS isotropic spectra of (LaCoO3)n(LaAlO3)n superlattices (n = 2, 6 and 10) and the thick LaCoO3 film taken in a fluorescence yield (FY) mode. The intensity of the t2g (eg) feature near 528 eV (529 eV indicated by a vertical dotted line) decreases (increases) with decreasing n, suggesting a dramatic transition to a low-spin state. The measurement geometry is appended in the inset. (b) Polarization-dependent spectra and (c) the X-ray linear dichroism (XLD) spectra highlighting the orbital character. Features near 532 eV reflect the Al 3sp states in the LaAlO3 substrates.

Full size image

Moreover, the polarization dependence in the XAS data reveals that the eg-to-t2g electron transfer can be facilitated by an increase in the 3z2-r2 orbital energy. Figure 5b highlights the anisotropy in the electronic structure. The polarization-dependent XAS spectra for n = 2 and n = 10 SLs are shown in Figure 5b. The (E // c) data are shown as dots while the (E // ab) data are shown as solid lines. The eg peak in the (E // c) spectra are located at slightly higher energies compared with their respective peak in (E // ab) counterparts, suggesting that the 3z2-r2 state is at a higher level than the x2-y2 state.

The corresponding X-ray linear dichroism [XLD = (E // c)-(E // ab)] is displayed in Figure 5c with a dotted zero-line for guidance to eyes. The spectral difference is more clearly observed in the XLD spectra. The positive (negative) signs in the XLD indicate the dominance of the 3z2-r2 (x2-y2) character. The XLD signal is enhanced with decreasing n from 10 to 2. Such anisotropy in the eg orbitals manifests ‘lifting’ of the 3z2-r2 level under low dimensionality5,8,12,18,23. The energy difference between the 3z2-r2 state and the x2-y2 state is estimated to be ∼1 eV, which is very close to the energy distance between the leading peak positions in the unoccupied 3z2-r2 and x2-y2 DOS in the DMFT result (see Figure 2b). For more clear assignments, we also performed Co L2,3-edge XAS and the corresponding XLD measurements. The analysis of the Co L2,3-edge data using cluster model calculations confirmed the XLD results of the changes in the eg orbital states (see Supplementary Information for more details16).

Our combined theoretical and experimental results clearly revealed how the d-orbital occupation with different symmetries (t2g and eg) can be manipulated by two-dimensional layering of atomically thin Mott insulator-band insulator SLs. By controlling the delicate competition between the crystal field splitting and electron correlation effect, we found that a subtle change in the 3z2-r2 orbital could be induced as a consequence of a d-orbital reconstruction from a t2g6-δegδ state to a complete t2g6 state. The lattice relaxation for all of the LCO SLs was not observed, implying that the evolution in the electronic structure is indeed driven by the reduced dimensionality rather than the lattice strain effect. The reduced dimensionality altered the insulating nature of LCO with a spin state crossover from LS+HS to LS. Therefore, our demonstrated controllability of electron occupation between the t2g and eg orbital states with atomic-scale SLs manifests that an accurate growth control of oxide heterostructures can open up many opportunities for discovering new physical properties or functionalities such as catalysis and photovoltaics, where the energy of d level is an essential parameter.

Methods

Sample fabrication

(LCO)n/(LAO)n SLs (n = 2, 6 and 10) were grown on NdGaO3 (110) substrates by pulsed laser epitaxy at 700°C under an oxygen atmosphere (100 mTorr). A KrF excimer laser (λ = 248 nm) with a laser fluence of ∼1 J cm−2 was used to ablate sintered LCO and single-crystalline LAO targets. The crystallinity of the fabricated SL samples were confirmed by lab x-ray diffraction (XRD) and hard XRD at the BL14B1 beamline of the Shanghai Synchrotron Radiation Facility in China.

Optical spectroscopy

Optical conductivity spectra (σ(ω)) were obtained by reflectance measurements in the energy region of 0.1–1.0 eV and variable angle spectroscopic ellipsometry (JA Woollam V-VASE) in the energy region of 0.74–4.5 eV. The σ(ω) of the LCO layers in the SLs were obtained by optical simulations implemented in the W-VASE software. The optical constants of bulk LAO were measured independently for the SL optical model calculations.

X-ray absorption spectroscopy

Soft X-ray absorption spectroscopy (XAS) measurements were performed at the 2A beamline at the Pohang Light Source (PLS) in Korea, which is equipped with an elliptically polarized undulator (EPU). The energy resolution was approximately 0.1 eV. We measured the absorption coefficients in both the fluorescence yield (FY) and total electron yield (TEY) modes. Note that we have not observed any discernable difference between the TEY and FY data in the energy region of interest in the O K-edge XAS spectra. Here, we changed the polarization of light by tuning the phase and gap in the EPU while fixing the angle of incidence with respect to the surface normal to the SLs to 70°, rather than by rotating the samples. This process circumvents any issues associated with the finite probing depth of the soft X-rays. The spectra of E // σ reflect only in-plane orbital hybridization, whereas the spectra of E // π reflect mostly (88%) perpendicular hybridization and partially (12%) in-plane orbital hybridization. The isotropic term ([(E // c) + 2× (E // ab)]/3) and the XLD ((E // c) - (E // ab)) at the Co L2,3-edge were deduced from the raw data of (E //σ) and (E //π) using the relationships of (E //σ) = (E // ab) and (E // π) = sin270° (E // c)+ cos270° (E // ab), respectively. The spectra were normalized maintaining the total spectral weight after subtracting the contributions of the NdGaO3 substrates.