Introduction

ABO3-type perovskite-structured transition metal oxides and their structural variances (e.g., Ruddlesden-Popper, Brownmillerite, infinite-layer phases) have been extensively studied because of their remarkable physicochemical properties, including metal-to-insulator transition (MIT)1,2,3,4, superconductivity5,6, ferroelectricity7,8, notably high ionic conduction9,10, and surface catalytic activity11,12,13,14. Oxygen-based defects, which may be present as isolated oxygen vacancies (VO), vacancy clusters, or ordered oxygen vacancy channels (OVCs), can affect and even dominate these properties4,15,16,17,18,19. The topotactic phase transitions (TPT) as a result of oxygen content change can lead to intriguing concurrent changes in electronic, optical, and magnetic properties16,18,19. Memristors, synaptic transistors, and high-density memories based on such TPTs have been designed and tested, offering significant promise in the future oxide electronics and neuromorphic computing4,20,21. Reliability of the existing types of devices and our ability to create advanced and robust signal processing systems are predicated on the understanding of the coupling between composition, structure, electronic properties, and external stimuli.

Strontium ferrite (SrFeOx) is a convenient platform to study reversible redox activity and associated property changes. Oxygen-deficient brownmillerite-structured SrFeO2.5 (BM-SFO) (Fig. 1a, c), with ordered oxygen vacancy channels, can exhibit facile, highly anisotropic oxygen ion transport and low energy diffusion barriers within the OVCs, making these materials attractive for use as electrocatalysts and oxygen membranes16,17. BM-SFO is a G-type antiferromagnetic insulator, whereas the stoichiometric perovskite-structured SrFeO3 (P-SFO) (Fig. 1b) is an antiferromagnetic metal1,4,15, which suggests that a zoo of transitory electronic behaviors may be realized by accessing metastable phases along the BM-SFO → P-SFO transition pathways. For example, memristive switching behavior is attributed to the dynamic formation/breaking of a conductive P-SFO filament in the parent matrix of insulating BM-SFO22,23. However, an atomistic understanding that bridges the evolution of the microstructure and properties in such materials is still lacking, preventing their predictive and practical use.

Fig. 1: Characterization of BM-SFO thin films and the concept of oxidation of BM-SFO by in-situ TEM.
figure 1

a–c Structure models (projected along [010]) of (a) BM-SFO with OVCs parallel to the substrate, (b) P-SFO, and (c) BM-SFO with OVCs perpendicular to substrate. d Structure model highlighting an OVC plane shown in (c) but viewed along [100]. e–f HAADF-STEM images of BM-SFO grown on STO(001) and LSAT(001), respectively, demonstrating the controlled orientation of OVCs. Scale bar is 2 nm. g Schematic illustrations showing the electron-beam illumination of the cross-sectional STO/SFO sample. The yellow spheres represent the positively charged surface of the specimen as a result of electron beam illumination. h Schematics of electron-beam induced electrical field (dark blue arrows) generated during TEM imaging and its effect in promoting O2- diffusion. The orange spheres represent O2- ions moving in the opposite direction of electron-beam-induced electrical field. The light brown color represents the area under electron beam illumination.

Full size image

In this work, we show that controlling the orientation of the OVCs by epitaxial growth allows us to exploit the structural anisotropy to stabilize isomeric yet structurally and electronically distinct intermediates (SrFeO2.75). We use in-situ transmission electron microscopy (TEM) to directly activate and image the TPT from BM-SFO to P-SFO, where oxygen ions are supplied from the reducible substrates under a built-up electric field created by electron beam irradiation during TEM imaging. Oxygen diffusion pathways that allow us to selectively access metastable reaction intermediate phases are revealed using density functional theory (DFT) calculations. Our results provide atomic-scale insights into oxygen diffusion and redox-driven phase transition processes occurring in vacancy-ordered oxides, paving the way for a deliberate control of the metastable phases.

Results

Synthesis and characterization of BM-SFO thin films

Figure 1a, c show the structure model (viewed along [010]) of BM-SFO with two orientations of OVCs. BM-SFO is composed of alternating oxygen-deficient FeO4 tetrahedral layers (i.e., OVCs) and fully coordinated FeO6 octahedral layers. If OVCs are aligned parallel to the interface (Fig. 1a), oxygen ion (O2-) diffusion along the out-of-plane direction (i.e., from octahedral FeO6 sublayers to tetrahedral FeO4 sublayers) would proceed with a higher energy barrier21,24. In contrast, if OVCs are perpendicular to the interface (Fig. 1c, d), O2- migrating from the substrate can directly intercalate into OVCs without crossing the FeO6 layers and, therefore, with a lower energy barrier. We note that manipulating the orientation of OVCs has been realized in a number of systems (such as cobaltites25,26,27 and ferrites10,19,28,29,30) over the years. In addition to conventionally used controls of the thin film deposition parameters and substrates, Han et al31 recently reported on controlling the OVC ordering in SrCoO3-δ thin layers via ionic liquid gating (ILG), which may open a path toward the creation of oxitronic devices.

Here, we choose BM-SFO thin films with differently orientated OVCs as our studied materials to probe the effect of OVCs on anisotropic O2- migration. Epitaxial BM-SFO thin films with the thicknesses of 15–35 nm were deposited on (001)-oriented SrTiO3 (STO) and (LaAlO3)0.3(Sr2AlTaO6)0.7 (LSAT) single crystal substrates by using pulsed laser deposition (PLD). STO (a = 3.905 Å) and LSAT (a = 3.868 Å) substrates were used because of their small in-plane lattice mismatches with BM-SFO and their reducibility, which can serve as an oxygen source/vacancy sink32,33,34. XRD θ − 2θ scans along the out-of-plane direction (Supplementary Fig. 1) demonstrate that OVCs of BM-SFO can be stabilized to be either parallel to the interface with STO or perpendicular to the interface with LSAT. Moreover, high-angle annular dark-field scanning TEM (HAADF-STEM) images (Fig. 1e, f) further verify the characteristic of vacancy ordering in BM-SFO, where the OVCs appear as dark stripes repeating in alternate layer15,19,28,35. High-quality BM-SFO thin films with different orientations of OVCs enable us to further investigate the details of anisotropic oxygen transport in these materials by in-situ TEM.

Oxidation of BM-SFO thin films by in-situ TEM

In-situ TEM is used to drive the oxidation of BM-SFO and monitor subsequent phase transitions. Previous studies have shown that electron beam illumination on an insulating sample during TEM imaging can generate secondary and Auger electrons that subsequently leave the sample surface36,37,38,39. The outer shell of the sample can become positively charged due to the loss of these electrons and an additional electric field is generated (Fig. 1g–h and Supplementary Fig. 2). It has been reported that a few volts can be induced at an electron beam dose of 103−104 e Å−2 s−1 40, which could be sufficient to drive the O species migration from the reducible substrate STO or LSAT to BM-SFO15,21,40,41. Furthermore, the samples are heated to elevated temperatures (200 to 300 °C) to lower the O migration energy barrier in our setup during in-situ TEM experiments, allowing us to further tune the reaction speed so that the structural evolution from BM-SFO to P-SFO can be monitored at an appropriate time scale. It is worth noting that, in our experiment, neither electron beam illumination nor the elevated temperature heating could work alone to induce adequate oxygen ions flux to oxidize the BM-SFO under vacuum, which is consistent with former results1,42. While combining the electron beam-induced electric field and low-temperature thermal field, the adequate O ions migration from STO to SFO could be triggered and thus realized the oxidation process of SFO.

Oxygen ion diffusion across the OVC layers

In-situ TEM studies on BM-SFO/STO, in which OVCs are parallel to the interface (Fig. 1a, e), allowed us to capture the dynamical phase transition processes (Supplementary Movie 1, recorded under 300 °C sample heating). We note that the rate of O accumulation in SFO is determined by two processes: O ions migration from STO to SFO during the in-situ TEM experiments and O loss from SFO thin film that occurs at 300 °C under vacuum. The loss of O from SFO reduces the effective oxygen flux in SFO and, therefore, decreases the P-SFO nucleation rate. It took almost 4 minutes for the nucleation of the P-SFO phase to take place. Figure 2a–c show snapshot TEM images taken from Supplementary Movie 1, demonstrating the formation and evolution of the P-SFO phase in the BM-SFO matrix. The high-resolution TEM images corresponding to the starting BM-SFO and emerging P-SFO are shown in the panels of Fig. 2d, f, respectively. As shown in the bottom panels of Fig. 2g, i, the Fast Fourier Transform (FFT) patterns obtained from the TEM images match well with the simulated diffraction patterns that were based on the crystal structures of the corresponding phases. Furthermore, HAADF and annular bright-field (ABF)-STEM imaging for BM-SFO and P-SFO phases are also conducted, shown in Supplementary Fig. 3, to demonstrate the details of structure change at atomic scale.

Fig. 2: Phase transformations of BM-SFO → SrFeO2.75 → P-SFO observed on BM-SFO/STO with starting OVCs parallel to STO(001).
figure 2

a–c Selected in-situ TEM images (see Supplementary Movie 1) of the same region of a ≈35 nm SFO film illustrating stages of the phase transition process over ≈300 s: the initial state of BM-SFO (a), appearance of the P-SFO inclusions (b), and propagation of the P-SFO phase (from the bottom of the image) (c). d–f High resolution TEM images and g–i, Fast Fourier Transform patterns (bottom left) and simulated diffraction patterns (bottom right) for (d) BM-SFO, (e) SrFeO2.75, and (f) P-SFO. The scale bars are 2 nm−1. j Structure model of bulk-like SrFeO2.75. k Schematics showing the creation of oxygen vacancies (VOs) in SFO/STO by annealing in vacuum, and subsequent replenishment of oxygen by annealing in 18O2. l ToF-SIMS depth profiles of 18O display the enrichment level for a 15 nm SFO film grown on STO(001) and a STO reference sample. The SFO/STO interface location, marked by the black dashed line, was confirmed by the secondary ion signals of FeO- and TiO-.

Full size image

It is worth to note that the formed P-SFO phase can quickly transform back to BM-SFO if the electron beam is blanked as we conduct the in-situ TEM experiment at elevated temperature (300 °C), suggesting rapid O emission from P-SFO due to vacuum annealing. We used a lower heating temperature of 200 °C during another in-situ TEM experiment to slow down the O loss process, which also allowed us to capture an intermediate phase displaying a checkerboard-like structure in the transition regions (see Supplementary Movie 2 and Fig. 2e). In this case, it took about 1 min and a half for the starting of phase transition, which is much shorter than the waiting time for P-SFO nucleation under the heating temperature of 300 °C, suggesting the much slower O emission from SFO thin film under 200 °C. The intermediate phase’s FFT pattern well matches with the simulated diffraction pattern of a previously observed SrFeO2.75 phase (Cmmm space group, Fig. 2h)43,44. The structure of the SrFeO2.75 phase (Fig. 2e, j) shows a checkerboard pattern composed of alternating columns of octahedral FeO6 and pyramidal FeO5. This ordering corresponds to the formation of oxygen-deficient FeO5 connectivity (chain of VOs) along the beam projection direction.

As the experiments were conducted in ultra-high vacuum, there is no other oxygen source but the STO substrates. In addition, according to the built-up electrical field shown in Fig. 1g, we expect the oxygen ions needed to heal the VOs to form SrFeO2.75 or P-SFO should come from the reducible STO substrate, resulting in local oxygen deficiency at the vicinity of film/STO interface. To further confirm this, we collected cross-sectional STEM-electron energy loss spectroscopy (EELS) maps at the SFO/STO heterostructure at room temperature after in-situ TEM experiments. As shown in Supplementary Fig. 4, the O intensity for STO exhibits a visible decrease in the region close to the SFO/STO interface (≈30 nm in width), supporting our hypothesis. The emergence of clusters-like contrast (indicated by circles in Supplementary Fig. 4) close to the SFO/STO interface also indicates the agglomeration of VOs45,46.

To further visualize the oxygen exchange between SFO and STO, we performed 18O labeled, time-of-flight secondary-ion mass spectrometry (ToF-SIMS) analysis. One BM-SFO/STO sample (≈15 nm thickness) and a STO(001) substrate were first annealed in vacuum at 700 °C for 1 h. For BM-SFO/STO sample, vacuum annealing (VA) could promote oxygen loss in SFO, which would subsequently draw oxygen from the STO substrate. This is similar to the well-established phenomenon of oxygen transfer from the substrate to the film deposited by PLD32,33. After VA, the BM-SFO/STO sample and the STO substrate (STO reference marked in Fig. 2l) were annealed at 650 °C for 0.5 h in a tube furnace backfilled with 50 Torr of 18O2.

To account for the possibility of matrix effects during ToF-SIMS analysis, one non-18O2 annealed BM-SFO/STO sample was measured (Supplementary Fig. 5). The oxygen signal intensity shows a notable increase upon sputtering through the BM-SFO film into the STO substrate (Supplementary Fig. 5b). Using the ratio of the total oxygen signal intensity on the film and substrate sides (i.e., (16O + 18O)film/(16O + 18O)substrate) yields a ratio of 0.80, which is close to the expected stoichiometric ratio of oxygen in BM-SrFeO2.5/SrTiO3 of 0.83. Therefore, this result suggests that the matrix effect, which has been well described in previous reports47,48,49,50, does not play a dominant role in the oxygen signal intensity change upon sputtering through the BM-SFO film into the STO substrate.

Based on prior reports47,48,49,50, matrix effects can significantly influence signal intensity but play a minimal role in isotopic ratio counting. Indeed, as demonstrated in Supplementary Fig. 5, the ratio of 18O/(16O + 18O) for a non-18O2 annealed BM-SFO/STO remains constant at the natural abundance of 18O (Supplementary Fig. 5c), even though the signal intensity changes from BM-SFO to STO. Hence, we utilized the 18O enrichment level, defined as 18O/(18O + 16O), to assess and visualize the oxygen exchange occurring at the interface. Figure 2l presents the depth profiles obtained from ToF-SIMS, illustrating the 18O enrichment level for both 18O2 annealed samples.

The 18O enrichment level is ≈74% at the near-surface region of STO reference, which undergoes an exponential decay to natural abundance level (≈0.2%) within a ≈100 nm range (see Supplementary Fig. 6). In comparison, the 18O enrichment level is higher (≈80% in SFO film) in the BM-SFO/STO sample due to the strong surface oxygen exchange between 18O2 and SFO. An even higher 18O enrichment level (up to ≈87%) is also observed on the STO side close to the STO/SFO interface, indicating that the oxygen vacancies in the STO substrate created during the VA process have been filled by 18O2 annealing (see schematics shown in Fig. 2k). We also performed 18O annealing experiment on an as-received STO substrate and observed similar 18O concentration distribution with that of VA STO substrate (see Supplementary Fig. 6). These results clearly demonstrated that oxygen exchange of the STO substrate can occur at 650 °C. Moreover, the oxygen diffusion coefficients of STO at 650 °C was determined by fitting the 18O depth profiles for both VA and non-VA STO substrates34,51,52,53,54. As shown in Supplementary Fig. 6, the bulk diffusion coefficient (D*) for VA and non-VA STO substrates were found to be 1.2×10−10 and 3.1 × 10−11 cm2 s−1, respectively, in agreement with previously reported values51,52.

Oxygen ion diffusion within the OVCs

We now consider the case of BM-SFO/LSAT, in which OVCs are perpendicular to the interface (Fig. 1c, f). As shown in Supplementary Movies 3 and 4, compared to BM-SFO/STO, the phase transition occurred more readily (almost no waiting time) in BM-SFO/LSAT during in-situ electron beam shower and 200 °C sample heating. We attribute this to the low diffusion barrier and high migration rate of O ions along OVCs in BM-SFO. Interestingly, two reaction fronts can be observed during the phase transition process, indicating a two-step reaction during the oxidation from BM-SFO to P-SFO. In Fig. 3a–d, the sequences of TEM images taken from Supplementary Movie 3 clearly show the phase transformation from BM-SFO (Fig. 3a, e) to P-SFO (Fig. 3d, g, l). Figure 3b–c capture an intermediate phase separated by two reaction fronts (marked by white and yellow dashed lines). A magnified view of the intermediate phase circled in Fig. 3b is displayed in Fig. 3f. High-resolution HAADF-STEM image (Fig. 3h) together with its lattice spacing mapping (Fig. 3i) clearly reveal the phase boundaries, and the three phases are assigned as BM-SFO, SrFeO2.75, and P-SFO. The experimentally measured Sr-Sr in-plane spacings for BM-SFO are ≈4.3 Å and 3.4 Å, due to the modulations from FeO4 and FeO6 connectivity, respectively (Fig. 3m). In contrast, the Sr-Sr distance of P-SFO is converged to one value of around 3.9 Å (Fig. 3m), as expected from a cubic structure coherently strained to an LSAT substrate. Compared to BM-SFO (SrFeO2.5), the long and short Sr-Sr atomic distances of the intermediate phase are measured to be ≈4.0 Å and 3.8 Å, respectively (Fig. 3m). According to our DFT modeling, the intermediate SrFeO2.5 phase has a structure that is significantly different from the bulk-SFO2.75 phase shown in Fig. 2j. In this case, oxygen ion migration from the reducible LSAT substrate could directly intercalate into the OVCs without involving mass transfer from/to the FeO6 sublayers. The SrFeO2.75 structure contains OVCs characteristic of FeO5 pyramidal connectivity (model in Fig. 3k) as a result of selectively healing VOs. The decrease in the IP Sr-Sr atomic distance between the sub-stoichiometric layers compared to BM-SFO is expected due to partial healing of VOs19,55,56.

Fig. 3: Phase transformations of BM-SFO → SrFeO2.75 → P-SFO observed on BM-SFO/LSAT with starting OVCs perpendicular to LSAT(001).
figure 3

a–d Time-stamped in-situ TEM images taken from Supplementary Movie 3 illustrating a two-step phase transition process. The yellow and white dashed lines indicate the two reaction fronts. e–g High-resolution TEM images matching the highlighted regions shown in (b) and (d) and j–l, their corresponding structure models. h HAADF-STEM image of an SFO/LSAT sample showing the coexistence of three different phases. i In-plane lattice spacing (between Sr-Sr ions) map of the blue box marked region shown in (h). m Comparison of experimentally measured values (symbols) with DFT calculated results (dashed lines) for in-plane Sr-Sr interatomic distances among the three phases. The error bars of ≈0.1 Å is determined considering the finite pixel size of the experimental HAADF image.

Full size image

Discussion

To gain atomic-scale insights into the electronic and structural changes upon BM-SFO oxidation and the details of different oxygen diffusion pathways related to the orientation of OVCs, we turn to DFT simulations (details are summarized in Methods). Figure 4 summarizes the relationships between oxygen content and SFO stability, oxygen diffusion pathways, and electronic properties. We first consider the case of BM-SFO/STO, in which OVCs are parallel to the interface and a bulk-like intermediate phase SrFeO2.75 was observed upon BM-SFO oxidation. We mimic the experimentally observed O transfer from STO to SrFeOx by continuously incorporating O atoms into one OVC plane of the BM-SFO supercell. Our calculations show that such O accumulation is thermodynamically preferred up to x≈2.7 (Fig. 4a). These additional O species pull electron charge from the neighboring ions and, as formally O2– ions (2p6), diffuse along the OVC parallel to the interface with STO with the calculated barrier of ≈0.65 eV (Supplementary Fig. 7). Further incorporation of O into the same OVC until all vacant sites are occupied (x = 2.75 for the supercell used here) is cost neutral. This configuration corresponds to the formation of a full P-SFO layer; accordingly, the P-SFO/BM-SFO interface advances to the next OVC layer. We define the SrFeO2.75 intermediate formed through vacancy layer accumulation as LA-SFO as shown in Fig. 4a. As the concentration of the additional O species in the OVC increases, the number of vacant sites available for their diffusion decreases which slows the in-plane diffusion down. At the same time, O species in the vicinity of the P-SFO/BM-SFO interface become progressively less negative due to a competition between the preferred O2– and Fe3+ electronic configurations, which leads to the formation of O(2–δ)– (2p6–δ) and Fe3+γ (3d5–γ) ions (see Supplementary Fig. 8). This depletion of the O 2p band destabilizes the oxygen sublattice, thus promoting local restructuring. We found that for the fixed concentration of additional oxygens (x = 2.75), distributing them over all OVCs, rather than confining them to one OVC is energetically preferred (Fig. 4a). The most stable configuration corresponds to the bulk SrFeO2.75 phase (purple dot in Fig. 4a), as observed experimentally (Fig. 2e) and illustrated in Fig. 2j.

Fig. 4: DFT calculations of SFO stability, oxygen diffusion pathways, and electronic density of states.
figure 4

a Energy gain due to incorporation of oxygen into OVCs parallel (||) to the substrate. Incorporation proceeds continuously into one OVC layer up to ≈75% occupancy (x ≈ 2.7 in our supercell) with the formation of LA-SFO2.75 phase; reorganization of the oxygen sublattice at larger x through disruptive diffusion leads to the formation of DD-SFO2.75. b Potential energy surface (PES) for an O2- (large spheres) diffusion across Fe-centered polyhedra (brown) FeO6 layers into a neighboring OVC calculated using nudged elastic band climbing image (NEB-CI) method; insets show the A (initial) and C (final) configurations for each step; the transient configuration (B) in between renders the overall diffusion barrier of ≈1.1 eV. c Energy gain due to oxygen incorporation into OVCs perpendicular (⊥) to the substrate (Ueff = 0 eV) for sequentially occupied VO-ip and VO-oop sites (red) and sequentially occupied VO-oop and VO-ip sites (blue), leading to the formation of the intercalation ID-SFO2.75 phase. d PES for O2- diffusion along perpendicular to OVC. Insets show the local atomic configurations for the first four minima. One-electron density of states (DOS) for e, BM-SFO (x = 2.5) and f, ID- and DD-SFO2.75, and P-SFO (x = 3) phases. Shaded areas show DOS projected on Fe 3d states. Fermi energy is at 0 eV.

Full size image

To shed light on the kinetics of the lattice reorganization, we investigated the mechanisms of oxygen ion diffusion across the FeO6 layer into the next OVC in the out-of-plane direction, as schematically indicated in Fig. 1a. Our simulations suggest that this diffusion proceeds via a two-step mechanism (A–B–C path in Fig. 4b), whereby each step has the barrier of ≈0.6 eV. Notably, the transient configuration (B) formed after the first diffusion step can reverse to the original configuration (A) with the barrier of only 0.1 eV, which renders the overall diffusion barrier across the FeO6 plane of ≈1.1 eV, i.e., nearly twice as large as that for diffusion within an OVC plane (≈0.65 eV). Since the formation of bulk-SFO2.75 requires a diffusion process disrupting continuous FeO6 planes by forming checkerboard arrangements of FeO6 and FeO5 polyhedra, we refer to it as disruptive diffusion (DD) thereafter and the corresponding bulk phase as DD-SFO2.75 (Fig. 4a).

For the case of BM-SFO/LSAT, in which OVCs are vertically aligned to the interface and a new intermediate phase was observed upon BM-SFO oxidation. To analyze the effect of oxygen incorporation and to establish the atomic structure of this observed intermediate phase, we examined the stability of SFO depending on the arrangements of the oxygen species using DFT simulations. Figure 1d shows the side view (along [100] direction) of a vertically aligned OVC in which only connectivity of FeO4 tetrahedra is visible. Healing of the in-plane VO indicated by solid red circles (defined as VO-ip) will lead to the formation of Fe-O-Fe bonds parallel to the interface. In comparison, the VOs formed between two Fe atoms along the out-of-plane direction (indicated by dotted circles) are defined as VO-oop, and healing those will lead to Fe-O-Fe bonds perpendicular to the interface. Along the one-dimensional (1D) FeO4 tetrahedral chain, the energies needed to heal the alternating VO-ip and VO-oop may vary depending on the sequence of oxygen incorporation and the substrate-induced strain. We considered several configurations of the O species occupying VO-ip only or VO-oop only sites and varied the concentration of these additional species (see Fig. 4c). According to our simulations for the 8 × 2 × 4 supercell (see Methods) at low O concentration, it is energetically favorable for the O to occupy VO-oop sites (blue symbols in Fig. 4c). The preference to occupy VO-oop over VO-ip sites persists through the entire 2.5 < x ≤ 2.75 range. An additional oxygen in the BM-SFO lattice is more stable at the VO-oop site than at the VO-ip site by ≈0.2 eV. However, for the SrFeO2.75 phase, this energy difference reaches 0.35 eV (for Hubbard U = 0 eV). Preferential occupation of the VO-oop sites is attributed to unconstrained lattice relaxation in the off-plane direction, while in-plane relaxation is suppressed due to epitaxial constraints. Accordingly, the out-of-plane lattice parameter decreases with increasing occupancy of the VO-oop sites and remains essentially unchanged with occupancy of the VO-ip sites. At x = 2.75, our DFT modeling predicts that the stable structure should have all VO-oop sites occupied and all VO-ip sites vacant (see Supplementary Fig. 9), which is selected as the structural model for the experimentally observed reaction intermediate shown in Fig. 3f. Moreover, the long and short Sr-Sr distances (indicated by the dashed lines in Fig. 3m) in the modeled SrFeO2.75 are calculated to be ≈4.0 Å and 3.7 Å, respectively, which are in good agreement with the experimentally measured values. Since this SrFeO2.75 intermediate phase is derived by intercalation diffusion (ID) along the chains of adjacent VO-ip or VO-oop sites (see Fig. 1d), we define it as the ID-SFO2.75 phase.

To obtain atomic-level insights into the kinetics of the oxidation process for the case of BM-SFO/LSAT, we determined the oxygen diffusion pathways and activation energies using DFT, in which the BM-SFO film was explicitly strained to the substrate. Figure 4d shows the potential energy surface (PES) calculated by displacing an O2– ion along the OVC and optimizing the positions of all other atoms at every step. The deep and shallow energy minima correspond to the O2– occupying VO-oop and VO-ip sites, respectively. The insets show the corresponding configurations of the selected O2– ion (shown in red) migrating from the in-plane to out-of-plane to in-plane configuration. The steep rise of the left side of the plot reflects the interaction of the extra oxygen in the OVC and the stoichiometric substrate. The calculated diffusion barriers vary between ≈0.6 and ≈0.8 eV, suggesting that O2– diffusion proceeds through the entire film with a preference of occupying the VO-oop sites, which explains the mechanism of the formation of the transient ID-SFO2.75 phase observed experimentally.

The compositional and structural differences between the SFO phases are reflected in the differences of their electronic properties. Figure 4e, f shows the one-electron densities of states (DOS) calculated for the BM-SFO, DD-SFO2.75 with in-plane (||) and out-of-plane (⊥) FeO5 pyramidal connectivity, ID-SFO2.75 with occupied out-of-plane O sites, and P-SFO. In the case of BM-SFO (x = 2.5), both || and ⊥ OVC configurations show a band gap of over 1 eV and nearly identical band edge DOS profiles. The band gap is underestimated with respect to the experimental values of ≈2.0 eV21, as expected for PBEsol. For x = 2.75, intercalation type diffusion (ID-SFO2.75) results in the band gap closure, indicating a metallic behavior. In contrast, bulk-phase DD-SFO2.75 formed by oxidizing in-plane OVC remain nonmetallic regardless of the orientation of the remaining VO channels. As the oxygen content increases to x = 3.0, the DOS magnitude near the Fermi level increases continuously indicating enhanced metallic conductivity. The band gap closure upon increasing oxygen content from x = 2.5 to x = 3.0 is driven by the appearance of unoccupied Fe 3d states (shown with shaded areas in Fig. 4e, f), which are depleted by the incorporation of O species into the OVCs. Thus, the electronic properties of the SrFeOx films, particularly the onset for insulator-to-metal transition, can be controlled not only by changing the oxygen content but also by controlling the oxidation pathways to access the appropriate precursor phases.

In summary, we demonstrated the ability to promote and guide the transfer of oxygen species from reducible substrates to oxidize differently oriented BM-SFO thin films and monitor the structural changes and phase transitions using in-situ TEM. By combining precisely controlled synthesis of precursor SFO with in-situ control of the electron-beam-induced electric field and sample heating, we were able to activate oxygen diffusion along selected pathways in BM-SFO that allows us to access two different reaction intermediates – Disruptive Diffusion-SFO2.75 and Intercalation Diffusion-SFO2.75 phases before the films were fully oxidized to metallic P-SFO. For the case of OVCs parallel to the interface, O diffusion is found to involve both FeO6 octahedral and FeO4 tetrahedral sublayers, whereas the phase transition from BM-SFO to P-SFO proceeds through a DD-SFO2.75 intermediate state via a disruptive diffusion process. In contrast, intercalation-only O diffusion is realized in samples displaying OVCs perpendicular to the interface, which results in a meta-stable ID-SFO2.75 phase containing alternating FeO6 octahedral and FeO5 pyramidal sublayers. Our DFT calculations predict that differences in crystal field associated with these two SrFeO2.75 intermediates result in qualitatively different electronic structures. It can be envisioned that selectively activating intercalation type diffusion may lead to faster ion transport, lower power assumption, and metastable intermediate with unique properties that can be harnessed for energy and information storage applications.

Methods

Thin film preparation

Epitaxial SrFeO2.5 thin films with the thicknesses of 15–35 nm were grown on (001)-oriented SrTiO3 and LSAT single crystal substrates using PLD19. The laser pulse (248 nm) energy density was ≈2 J cm−2, the repetition rate was 1 Hz. The substrates were heated to 700 °C during deposition and the growth oxygen pressure was kept at 0.1 mTorr. After growth, the samples were cooled down to room temperature under the same oxygen pressure.

In-situ TEM experiments

The TEM samples used for in-situ TEM were prepared using a focus ion beam scanning electron microscopy (Helios). In-situ TEM experiments were conducted using an FEI Titan 80-300 TEM equipped with an aberration corrector for the objective lens and a Gatan furnace-based heating holder. The accelerate voltage of 300 kV and electron beam dose rate of ≈103 e Å−2 s−1 were used in the in-situ TEM experiments. The TEM samples were heated to elevated temperatures (200 to 300 °C) during the experiments to promote the reaction, making it suitable for in-situ TEM observation. The high angle annular dark-field (HAADF) scanning transmission electron microscopy (STEM) image, annular bright-field (ABF) STEM image and electron energy-loss spectrum (EELS) mapping were conducted using JEM ARM200F. The collection angle for HAADF and ABF imaging were 90-370 mrad and 10–23 mrad, respectively. The probe current of ≈20 pA was used for STEM imaging and EELS mapping to minimize the electron beam induced phase transition. The Dual-EELS was used for the energy calibration of Fe-L edge with the simultaneously acquired zero loss spectrum.

SIMS measurements

ToF-SIMS measurements were performed using a ToF-SIMS V (ION-TOF GmbH, Münster, Germany) mass spectrometer equipped with a time-of-flight analyzer of a reflectron type. A dual-beam non-interlaced depth profiling strategy was used, in which a 1.0 keV Cs+ beam (∼40 nA, 200 µm × 200 µm scanning area) was used for sputtering and a 50 keV Bi32+ beam (∼0.05 pA, 50 µm × 50 µm scanning area at the Cs+ crater center) was used for negative spectra data collection. One BM-SFO sample was first annealed in vacuum at 700 °C for 0.5 h to promote further oxygen loss in SFO. After vacuum annealing, the SFO/STO sample, together with an untreated STO(001) substrate (Reference STO), were annealed at 650 °C for 0.5 h in a tube furnace backfilled with 50 Torr of 18O2 (97% purity, Cambridge Isotopes). The SFO/STO interface location was confirmed by the secondary ion signals of FeO- and TiO-.

Ab-initio simulations

The calculations were performed using the Vienna Ab-initio Software Package (VASP)57,58. The Perdew-Burke-Ernzerhof exchange-correlation functional modified for solids (PBEsol)59 and the projector augmented wave pseudopotentials60, as implemented in VASP, were used. The energy cut-off was 500 eV. SFO was represented using the periodic model approach and a supercell equivalent to the 4 × 4 × 4 extension of the pseudo-cubic perovskite crystallographic cell. Gamma point only was used for energy minimization with respect to the internal coordinates and the out-of-plane lattice parameter, the electronic structure for the energy minimum configuration was recalculated using 2 × 2 × 2 Monkhorst-Pack k-points mesh, 4 × 4 × 4 mesh for used for DOS calculations. The total energy was converged to 10–5 eV. The Hubbard U correction for Fe 3d states (Ueff = U – J = 3 eV) was applied using Dudarev’s approach61. The 1 × 2 × 1 k-mesh and Ueff = 0 eV were used for the 8 × 2 × 4 supercell. Atomic charges were calculated using the Bader’s approach62. The diffusion pathways and activation energies were calculated for the SrTiO3 substrate (a = b = 3.905 Å) using the nudged elastic band (NEB) method and eight NEB images unless stated otherwise. Energy gain due to oxygen incorporation was calculated with respect to the gas-phase O2 molecule.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.