Introduction

Complex oxides display an unprecedented diversity of fascinating functionalities arising from the strong interactions between their charge, spin, orbital and lattice degrees of freedom¹. Besides the non-isovalent chemical doping, the d-band-filling control can be achieved at oxide interfaces by the electric field effect. Various intriguing phenomena have been observed, e.g. insulator-superconductor transition², metal-insulator transition³ and magnetoelectric effects^4,5. Owing to the potential discontinuity at complex oxide heterointerfaces, marked charge modification might occur, leading to electronic reconstructions and further the emergent electronic and magnetic states^6–16. However, modification via either the electric field effect or interfacial charge transfer only works within a confined quasi-two-dimensional region near the interface due to the short electrostatic screening length of complex oxides¹⁷. Recent advance in achieving the three-dimensional (3D) d-band-filling control is through tuning the oxygen coordination number of the transition metal ions^18–20, limited to the oxides that adopt the ordered oxygen-deficient structures as the thermodynamically stable phase.

Among the 3d transition metal oxides, the most remarkable property distinguishing cobalt oxides from the others is that Co ions can accommodate various spin states dictated by the relative strength of the crystal field splitting Δ_cf and intra-atomic (Hund’s) exchange energy Δ_ex. For example, octahedrally coordinated Co³⁺ ions in LaCoO₃ (LCO) could have three spin states (Fig. 1a–c): low-spin (LS, S = 0) t_2g⁶ (where Δ_cf is slightly larger than Δ_ex), intermediate-spin (IS, S = 1) t_2g⁵e_g¹, and high-spin (HS, S = 2) t_2g⁴e_g² (where Δ_ex is barely larger than Δ_cf)^21–23, while the selection decided by Hund’s rule states that Co²⁺ always adopts the HS t_2g⁵e_g² state (Fig. 1d). The combination of the valence and spin states determines the electrical and magnetic properties of LCO. Bulk LCO is a nonmagnetic insulator owing to the LS state Co³⁺ ions^23,24. As Δ_cf is very sensitive to the Co–O bond length and angle²⁵, the IS or HS state of Co³⁺ can be stabilised by the in-plane tensile strain in LCO thin films, resulting in a ferromagnetic (FM) insulating ground state^25–30. It was reported that two metastable, anion-ordered phases were prepared by the special ‘kinetically controlled’ reduction of LCO so that the Co d-band-filling could be modified between 3d⁶ and 3d⁷ as the lattice coordination varied^31,32. However, it is still challenging to tuning the band filling by design, since the perovskite LCO is the most thermodynamically stable crystalline phase³³.

Fig. 1

Spin states of Co ions and electronic band structure of LCO/LTO.

a–d Spin states of Co ions. Δ_cf, crystal field splitting, Δ_ex, intra-atomic exchange energy. e Schematic diagram of 30u.c.-LaCoO₃ (LCO)/n-LaTiO₃ (LTO) heterostructure grown on SrTiO₃ (STO) substrate. We refer to this heterostructure as C30/Tn. The black arrow indicates the shared oxygen sites at the interface. f Electronic band structures of LCO and LTO. E_F, Fermi level; Δ_eff, charge-transfer gap; U_eff, Mott–Hubbard gap. g The final equilibrium state wherein the ΔE_F can be removed by charge transfer from Ti to Co. ΔΦ, electrostatic potential drop.

In this work, we report the 3D modulation of the d-band-filling and topotactic phase transition of LCO by designing LCO/LaTiO₃ (LTO) heterostructures (Fig. 1e). Due to the significant Fermi level (E_F) mismatch at the heterointerface, electrons transfer from LTO to LCO, which drives the perovskite LCO instability. Oxygen ions are then released from LCO to form local lattice configurations with a reduced Co valence state. The redistribution of these local configurations results in the 3D topotactic phase transition throughout the LCO film. The reduction of LCO can be tuned by the thickness of LTO, leading to a magnetic transition from FM to antiferromagnetic (AFM) ordering. Our findings provide a simple way to control the electronic configuration for the exploration of 3D emergent properties of complex oxides by designing heterointerfaces. Such 3D modification breaks the limitation of the electric field effect, which only works with an ultra-thin oxide film, leading to potential applications in neuromorphic computing and iontronics.

Results

The interfaces of complex oxides have been proposed to follow different band-alignment rules due to the invalidity of the single work–function approximation³⁴. As schematically illustrated in Fig. 1e, f, since the corner oxygen sites are shared at the interface, the oxygen 2p states should be aligned^9–13. The difference in the charge-transfer gap Δ_eff, which is between the filled O 2p and unoccupied upper-Hubbard 3d states³⁵, determines the E_F mismatch (ΔE_F) across the interface. For LCO (3d⁶), it is a narrow gap charge-transfer insulator with Δ_eff ~ 0.1 eV³⁶, while in a prototypical Mott–Hubbard insulator, LTO (3d¹) has small Hubbard splitting U_eff ~ 0.2 eV and large Δ_eff ~ 4.5 eV (its 3d band is located far above its oxygen 2p band)³⁵. Therefore, there is a huge ΔE_F (~4.4 eV) at the LCO/LTO interface, resulting in electron transfer from the lower-Hubbard Ti 3d band of LTO to the empty upper-Hubbard Co 3d band of LCO. Such electronic reconstruction causes a rigid 3d band shift, i.e. the elevation of the Co 3d band and the lowering of the Ti 3d band, as illustrated in Fig. 1g. Note that the above scenario is confined within the 2D region at the interface, and the modification of the band-filling can be described as Co³⁺ (3d⁶) + Ti³⁺ (3d¹) → Co²⁺ (3d⁷) + Ti⁴⁺ (3d⁰).

The LCO (30 u.c.)/LTO (n) (referred to as C30/Tn in the following, u.c. = unit cell) heterostructures were grown on TiO₂-terminated SrTiO₃ (STO) substrates by pulsed laser deposition (see ‘Methods’ section). We first explore the charge transfer across the LCO/LTO interface via electron energy loss spectroscopy (EELS). Figure 2a shows the line-by-line Co L-edge spectra from the LCO/LTO interface (uppermost curve) to the LCO/STO substrate interface (bottom curve) of a C30/T15 heterostructure. Surprisingly, we do not observe any clear peak shifts at different positions, against the interfacial charge-transfer picture. Together with the roughly unchanged Co L_3,2 ratio (Fig. 2b), we conclude that the valence state of Co in the C30/T15 is spatially homogeneous. However, upon comparing it with the LCO film without LTO (C30/T0), the Co L₃ peak of the C30/T15 shifts towards lower energy (Fig. 2c), while the peak position of the La M-edge is identical, indicating the decrease of the valence state of Co ions in the C30/T15. Figure 2e shows the Co L-edge X-ray absorption spectroscopy (XAS) results. The curve of C30/T15 has split L₃ peaks, which shift to lower energy compared with that from the C30/T0 sample. The peak shift and L₃ peak splitting are the signatures of the Co²⁺ valence state^15,16. The difference of the averaged O K-edge spectra (Fig. 2d) of the LCO in the C30/T15 and C30/T0 also indicates charge redistribution. There is a pre-peak feature in C30/T0 originating from the excitation of O 1s–O 2p on the O 2p holes created by the hybridisation of O 2p with Co 3d^37,38. However, such a pre-peak is absent in the C30/T15, since there are no O 2p holes at the Co²⁺ sites, as previously revealed in CoO³⁹ and proton-inserted HSrCoO_2.5¹⁸. The charge redistribution could also be verified via XAS measurements of the Ti L-edge (Fig. 2f). The features exhibit excellent agreement with those of the Ti⁴⁺ charge state and are remarkably different from the spectra of Ti³⁺, consistent with calculated Ti oxidation states in the LTO layer from the spatially resolved STEM-EELS (Supplementary Fig. 2), demonstrating electronic reconstruction on the Ti site, e.g. Ti 3d¹ to 3d⁰. The expected electron transfer from Ti to Co indeed occurred in the C30/T15; however, the length scale breaks the 2D limit, reaching the 3D scale.

Fig. 2

EELS and XAS of C30/Tn heterostructures.

a The Co L-edge EELS spectra of C30/T15 at different positions from STO/LCO to the LCO/LTO interface. The bottom curve is from the STO substrate/LCO interface, and the uppermost curve is from the LCO/LTO interface. b The corresponding Co L_3,2 intensity ratio. The LCO/LTO interface sets for zero. The averaged c Co L-edge and d O K-edge EELS spectra of C30/T15 and C30/T0. The shadow area in d indicates the distinct difference in the pre-peak feature. The spectra were averaged from the line scan in the 30u.c. LCO block. e Co L-edge XAS spectra of the C30/T15 and C30/T0 samples. f The Ti L-edge XAS spectra of the STO substrate, C30/T15, and 100-nm YTiO₃ film from ref. ¹⁰.

This 3D charge redistribution affects the lattice structure, which could be observed macroscopically. The bulk LTO has a GdFeO₃ orthorhombic structure with pseudocubic lattice parameter a_pc = 3.969 Å⁴⁰; while the structure of LCO is rhombohedral with pseudocubic lattice parameter a_pc = 3.826 Å (Fig. 3a)⁴¹. Figure 3b shows the X-ray diffraction (XRD) θ–2θ scan patterns of the LCO/LTO grown on STO (a = 3.905 Å) substrates. Without the LTO (C30/T0), only diffraction peaks from LCO (besides those of the STO substrates) were observed (black line). The presence of clear Kiessig fringes indicates the excellent crystalline quality of the film. For the C30/T10 (magenta line) and C30/T15 (light blue line) heterostructures, topotactic phase transitions of LCO occur by interfacing with LTO. The phase in the C30/T10 is consistent with the La₃Co₃O₈ phase, which has a monoclinic u.c. with a_m = 5.559 Å, b_m = 5.415 Å and c_m = 11.773 Å (space group P2₁) (Fig. 3a)³². Each u.c. consists of double CoO₆ octahedral layers and single CoO₄ tetrahedral layers along its c_m axis, which could be represented as pseudotetragonal with a_t = 3.879 Å and c_t = c_m/3 = 3.925 Å. The emergent phase in the C30/T15 is identified as brownmillerite (BM) La₂Co₂O₅, which adopts a distorted orthorhombic u.c. (space group Pnma), with a_o = 5.445 Å, b_o = 15.869 Å and c_o = 5.692 Å (Fig. 3a)³¹. It could also be considered pseudotetragonal, with a_t = 3.937 Å and c_t = b_o/4 = 3.967 Å. In La₃Co₃O₈ or La₂Co₂O₅, the CoO₆ octahedral and CoO₄ tetrahedral sub-layers are stacked alternatively with long-range order along the c_t axis, giving rise to the (0 0 k/3) or (0 0 k/2) fractional diffraction peaks (indices with respect to the average perovskite u.c.) as shown in Fig. 3b.

Fig. 3

Topotactic phase transition in C30/Tn.

a Schematic crystal structure of perovskite LaCoO₃, La₃Co₃O₈, and brownmillerite La₂Co₂O₅, respectively. Enlarged details are presented to show the stacking sequence of CoO₆ octahedrons and CoO₄ tetrahedrons along the out-of-plane direction. b X-ray diffraction data of LCO/LTO heterostructures. Blackline, C30/T0; orange line, C28/T7; magenta line, C30/T10; light blue line, C30/T15. All indices correspond to the averaged perovskite u.c. Peaks from STO substrates are denoted by asterisks. c High-resolution high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of C30/T15 taken along the STO [110] direction. The scale bar corresponds to 0.5 nm. The inset image shows a simulated STEM image viewed along the [001]_o direction based on the structure of brownmillerite La₂Co₂O₅. The yellow dashed line indicates the LCO/LTO interface, while the red dashed line shows the region used to calculate La–La interatomic distances. d Out-of-plane La–La distance along the red dashed line shown in c. The error bar shows the standard deviations of the averaged measurements. The cyan dotted line and green dashed line indicates the out-of-plane lattice parameter of LaCoO₃ (C30/T0) and La₂Co₂O₅ (C30/T15) obtained from b, respectively.

The structural distortion induced by the inserted CoO₄ tetrahedral layer could be visualised directly in the real space via scanning transmission electron microscopy (STEM). Figure 3c displays a high-angle annular dark-field (HAADF)-STEM image of the C30/T15, taken along the STO [110] zone axis. The most straightforward observation is the lateral superstructure (parallel to the interface) with doubled periodicity, which could be seen as a signature of the BM phase with long-range alternative stacking of the octahedral and tetrahedral sub-layers¹⁹. We also performed a STEM simulation based on the BM La₂Co₂O₅. The consistency of the experimental and simulated STEM images seen along the c_o axis (inset of Fig. 3c) further confirms the existence of a periodically arranged CoO₄ layer. Such a superstructure is absent in the C30/T0 sample (Supplementary Fig. 3), consistent with the macroscopic structural characterisation of XRD.

The reciprocal space map reveals that throughout the topotactic phase transitions, La₂Co₂O₅ is still coherently strained on STO (Supplementary Fig. 4). The out-of-plane lattice parameter c_t and the volume of the pseudotetragonal u.c. show a monotonous increase as the LTO thickness increases, consistent with the monotonous rise of the proportion of Co²⁺ ions associated with the insertion of CoO₄ tetrahedral layers into the structure. Quantitatively, it should be noted that the ratio of CoO₆ to CoO₄ layers in the emerging superstructures equals 30/n in C30/T10 and C30/T15 cases, i.e. one u.c. LTO corresponds to the insertion of one tetrahedral layer into the octahedral matrix.

The in-plane lattice parameters of bulk La₂Co₂O₅ and LTO are larger than those of the STO substrate (a_STO ~ 3.905 Å, a_La2Co2O5 ~3.937 Å and a_LTO ~3.969 Å), indicating that the C30/T15 is subjected to compressive strain. From the STEM images, we measured the in-plane lattice parameters of the La₂Co₂O₅ and LTO blocks as 3.90 ± 0.05 Å (Supplementary Fig. 5), in accordance with the reciprocal space map. The out-of-plane La–La atomic distance varies between the CoO₆ (~3.67 ± 0.1 Å) and CoO₄ (~4.32 ± 0.1 Å) sub-layers, as shown in Fig. 3d. Such an interspacing modulation is attenuated towards the LCO/LTO interface and might be associated with the slight intermixing across the interface (Supplementary Fig. 6). Importantly, we observed that the LTO films in both the C30/T10 and C30/T15 maintained a well-defined perovskite structure without forming ordered LaTiO_3+δ phases⁴².

Next, we address the 3D modulation of the physical properties of the C30/Tn. As shown in Fig. 4a, the highly insulating behaviour and the thermally activated character of all heterostructures are clearly observed, with the sheet resistance in the order of ~10⁹ Ω/□ at room temperature. The sample with a thicker LTO has a higher content of Co²⁺ ions and larger resistance since electrons are easily localised on the Co²⁺ sites. By fitting the transport data (inset of Fig. 4a), we calculated the activation gap as ~0.32 ± 0.02 eV for the C30/T0, with a monotonous increase as the LTO thickness increases.

Fig. 4

Transport and magnetic properties of C30/Tn.

a Sheet resistance of C30/Tn heterostructures as a function of temperature. The inset shows the activation charge gap of C30/Tn. b Temperature dependence of magnetisation M(T) curves for C30/Tn under field cooling measured by applying an in-plane magnetic field H = 0.1 Tesla. c Magnetic hysteresis loops M(H) measured at 10 K for C30/Tn heterostructures. The saturation magnetisation reaches ~0.9 μ_B/Co, evidencing the stoichiometric LCO of the C30/T0³⁰. The disappearance of FM ordering in the La₃Co₃O₈ and La₂Co₂O₅ indicates an AFM ground state as in their bulk form^{31, 32}.

There is a substantial change of the magnetic ground states associated with the topotactic phase transitions, as shown in Fig. 4b, c. The tensile-strained LCO film (C30/T0) has the FM ground state. However, for the C30/T10 (La₃Co₃O₈) and C30/T15 (La₂Co₂O₅), the FM ordering vanishes without any indication of magnetic transition. Considering the spin state and charge state of Co ions (Fig. 1a–d), in the C30/T0, Co³⁺ ions could adopt either the LS or HS state: an LS/HS mixture state²⁸. The FM insulating ground state is established via the superexchange interaction within the HS–LS-HS Co³⁺ configuration according to the so-called Goodenough–Kanamori–Anderson rules⁴³. However, this interaction is weak and easily vanishes when the HS Co²⁺ ions reach a threshold ratio of ~12.5%³⁰, since the exchange coupling between two Co²⁺ with a half-filled e_g orbital turns out to be strong and AFM. This explains the disappearance of FM ordering in the La₃Co₃O₈ and La₂Co₂O₅: they are actually in an AFM ground state as in their bulk form^31,32.

Discussion

The LCO/LTO interfacial electron transfer has been expected in 2D^9–15. Surprisingly, however, 3D modulation of the band-filling mediated by the topotactic phase transitions is observed. To determine the mechanism, we performed first-principle calculations on the structural stability of LCO (Supplementary Note 2). The results reveal that, with the extra transferred electrons, the perovskite structure of LCO becomes unstable, while the reduced CoO₄ tetrahedral lattice configuration with released O²⁻ ions shows a significant energy lowering. When 1u.c. of LTO is grown on the LCO, interfacial electron transfer occurs and induces the formation of a CoO₄ tetrahedral sub-layer. As shown in Fig. 1g, after the interfacial charge transfer the final equilibrium state is a balanced charge state near the interface with the perovskite LaCo²⁺O₃ and LaTi⁴⁺O₃. However, interfacial charge transfer would trigger the phase transition of LCO, then the electrostatic charge at the LCO side could be eliminated. Increasing the LTO thickness, charge transfer might occur between LaTi⁴⁺O₃ at the interface and following LaTi³⁺O₃. Due to there still has electron at Ti d-band and the uncharged LCO at the interface, consecutive charge transfer across the LCO/LTO interface and the kinetic process in LCO where the initial CoO₄ sub-layer diffuses into the LCO block occur, allowing the formation of a new CoO₄ sub-layer at the interface. This consecutive process results in the ordered distribution of the CoO₄ sub-layers inserted into the octahedral matrix, quantitatively evidenced by the experimental observation that each u.c. LTO corresponds to the insertion of one CoO₄ layer, i.e. La₃Co₃O₈ and La₂Co₂O₅ forms in the C30/T10 and C30/T15, respectively. Such a quantitative dependence maintains in the C50/T25 with the LCO thickness increased up to 25u.c., indicating that the mechanism of the band-filling modification is 3D indeed. In brief, the 3D band-filling modulation realised in the current work is triggered by the 2D interface charge transfer, and mediated by the topotactic phase transition all through the LCO film. We notice a recent report on LCO/LTO superlattices¹⁵, in which the CoO₄ tetrahedral layer was not detected. This might be due to the distinct lattice relaxation in the ultra-thin LCO layers (2u.c.) that suppresses the formation of CoO₄ tetrahedrons, or the symmetric interfaces at both sides of LCO that result in the random arrangement of CoO₄ tetrahedrons.

As characterised by STEM (Fig. 3c), although the LTO layer exhibits the Ti⁴⁺ oxidation state homogeneously (Supplementary Fig. 2b), it maintains the perovskite structure without any oxygen-excess phases detected. This evidences that the oxygen released from LCO could mainly diffuse through the LTO layer and desorb from the surface during growth. It is worth noting that the LTO layer would be charged within the above picture since the oxygen ions diffused from LCO should leave electrons in LTO when they desorb in the form of molecules. Such a charging effect might be compensated by the STO substrate and capping, which could be considered as electron reservoirs, or by randomly and locally distributed interstitial oxygen in the LTO lattice without being detected by XRD or STEM⁴⁴. Therefore, in the current case, in addition to provide the electrons that trigger the structural destabilisation of LCO, the LTO layer also acts as an oxygen getter layer²⁰. Associated with the oxygen out-diffusion, the final state of LCO is indeed the redox-reaction-driven, to the extent of 3D transformation. However, the observed effects cannot be determined by the redox reaction alone due to chemical potential mismatch⁴⁵. We fabricated LCO/oxygen-deficient LaAlO₃, and found that the capping layer does induce the redox reaction of LCO but not to the extent of topotactic transitions (Supplementary Fig. 7). We could also directly exclude the current observation is due to the redox reaction of LCO by depositing the amorphous LTO layer⁴⁶ (Supplementary Fig. 11). We also fabricated a heterostructure LCO (30u.c.)/STO (5u.c.)/LTO (15u.c.), in which the STO layer effectively blocks the interfacial electron transfer. This sample does not show the topotactic phase transition as observed in the C30/T10 and C30/T15 (Supplementary Fig. 8), indicating that the interfacial charge transfer has a key role in triggering the structural destabilisation and topotactic phase transition of LCO.

It is important to note that the CoO₄ tetrahedra induced by the interfacial electron transfer form the 2D network and remain in the layered arrangement during the diffusion away from the interface. This brings the opportunity to design different anion-ordered LCO phases by adjusting the LTO thickness, i.e. the metastable La_mCo_mO_3m−1 (m ≥ 2, an integer) can be obtained with arbitrary thickness with the macroscopic distinct electronic and magnetic ground states. Besides the La₃Co₃O₈ and La₂Co₂O₅ phases identified in XRD (Fig. 3b), we try to obtain the high-ordered phase, La₄Co₄O₁₁, by growing the C28/T7 heterostructure. As shown by the HAADF-STEM image in Fig. 5, there are 7 CoO₄ sub-layers (the dark stripes indicated by the red arrows) induced by the 7u.c. LTO, appearing as 2D insertions in parallel to the interface. However, their interspacing is not uniform. Different La_mCo_mO_3m−1 configurations are formed locally without long-range order, resulting in the disappearance of the fractional diffraction peaks in XRD (Fig. 3b). Further fine-tuning of the growth temperature or rate is necessary to control the kinetic process of the CoO₄ sub-layers diffusion, thus the monophased La_mCo_mO_3m−1 or even the superstructures of the phases with different m can be achieved by design.

Fig. 5

Nonuniform distribution of CoO₄ layers in C28/T7.

HAADF-STEM image of C28/T7 taken along the STO [110] direction. The scale bar corresponds to 1 nm. The red arrows indicate the CoO₄ layers.

In conclusion, we achieve the 3D modulation of the d-band-filling of LCO, mediated by the topotactic phase transitions, in the designed LCO/LTO heterostructures. The interfacial electron transfer from LTO to LCO triggers the destabilisation of perovskite LCO, inducing the formation of local lattice configurations with a lowered Co valence. The redistribution of these reduced sub-layers, whose number is determined by the thickness of LTO, mediates the 3D topotactic phase transitions. Our approach can also be extended to other complex oxides, e.g. to achieve the infinite-layer nickelate superconductor through the design of a Nd_0.8Sr_0.2NiO₃/LTO heterostructure⁴⁷. The mechanism proposed in the current work might inspire future works on neuromorphic computing and iontronics.

Methods