Thin film preparation. Epitaxial SrFeO2.5 thin films with the thickness of 15–30 nm were grown on (001)-oriented SrTiO3 and LSAT single crystal substrates by using PLD.19 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). The cross-sectional lamella lifted out using FIB was thinned down to ~ 200 nm at 30 kV, followed by 5kV and 2 kV milling down to ~ 50 nm, and the 1 kV setting was used for the final milling. 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/(Å2 s) 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. It should be noted that heating to temperatures in excess of 300 °C led to faster oxygen loss, promoting the partial reduction of the newly formed P-SFO phase. 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 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 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 Fe and Ti.
Ab initio simulations. Supported SrFeOx films were represented using two models. In the periodic slab model, the BO2-terminated substrate was represented by one unit cell of SrTiO3 with the in-plane coordinates of all atoms fixed in their centro-symmetric positions. BM-SFO film supported on this substrate has the thickness equivalent to four crystallographic perovskite cells. The vacuum gap was set to ~ 15 Å. In the bulk model, the BM-SFO was represented by the supercell equivalent to the 4×4×4 cubic perovskite supercell. In both cases, we considered two orientations of the oxygen vacancy channels (OVCs) in BM-SFO – parallel and perpendicular to the substrate plane. In addition, we used 8×4×2 supercell for the OVCs perpendicular to the substrate to investigate the relative stability of configurations formed by partial occupancy of the in-plane and out-of-plane oxygen vacant sites. To investigate the strain effects, we considered four sets of in-plane lattice parameters (a = b) corresponding to LaAlO3 (3.790 Å), LSAT (3.868 Å), SrTiO3 (3.905 Å), and DyTiO3 (3.950 Å) substrates. In the bulk model, the off-plane lattice parameter was optimized.
The calculations were performed using the Vienna Ab initio Software Package (VASP) 38,39. The Perdew-Burke-Ernzerhof exchange-correlation functional modified for solids (PBEsol) 40 and the projector augmented wave pseudopotentials 41, as implemented in VASP, were used. The energy cut-off was 500 eV. Gamma point only was used for energy minimization with respect to the internal coordinates and the off-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 approach42. 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 approach 43. 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.