2.1 In Situ Observations of Fe Exsolution from SrTi_1−xFe_xO₃ Films

To meticulously observe the formation and crystallization of Fe particles as well as the evolution of the underlying thin oxide layers, in situ transmission electron microscopy (TEM) was employed during the annealing process of SrTi_1−xFe_xO₃ films (where x ranges from 0.1 to 0.7), synthesized via sol–gel spin-coating under a H₂ gas flow (P_oxy = 10⁻¹⁷ Torr). Figure 1g–l presents a sequence of TEM bright-field (BF) images capturing distinct regions of the sample with a 30% Fe concentration (SrTi_0.7Fe_0.3O₃). These images were extracted from a video recording the sample undergoing staged heating at temperatures of 500, 600, 700, 750, 800, and 850 °C, sequentially. Accompanying each image, the respective selected area electron diffraction (SAED) pattern is displayed to understand phase evolution and transformation (see the expanded views in Figure S1, Supporting Information). At 600 °C, we observe a number of Fe particles, with diameters ranging from 3.5 to 6.0 nm (Figure S2b, Supporting Information), exsolved from the SrTiO₃ film. The SAED pattern of the image reveals that the particles are crystalline Fe, as identified by the presence of weakly diffracted spots (evidenced by the green arcs in the SAED pattern of Figure 1h and JCPDS 87-0722) while the film is amorphous, as demonstrated by the diffused rings of SrTiO₃ film. The size of Fe particles increases with temperature, approaching saturation at a range of 6.5 to 10.5 nm when the temperature exceeds 750 °C (Figure S2b and S3b, Supporting Information). The film remains amorphous up to 750 °C and crystallizes at 800 °C (see the relevant SAED patterns, JCPDS 79-0176). Schematics in Figure 1aa describe these stages of Fe exsolution from an amorphous SrTiO₃ film, followed by the subsequent crystallization of the oxide film. Significantly, this trend continues across various Fe compositions, with the notable exception that the temperature required for film crystallization increases as the Fe composition rises (Figure 1). Specifically, the film with 70% Fe is not crystallized below 850 °C. We observe that the diffracted spots from the Fe particles are not detected below 700 °C in the 10% Fe sample, 600 °C in the 30% Fe sample, and 600 °C in the 50% Fe sample, attributable to the minimal quantity of exsolved Fe atoms.

To confirm the presence of Fe crystals on amorphous films, high-resolution transmission electron microscopy (HRTEM) was employed during annealing at 750 °C for the sample with 30% Fe doping. Figure 2f presents the HRTEM image of a single Fe crystal, alongside its corresponding fast Fourier transform (FFT) pattern, as shown in Figure 2g. The crystal structure of the particle is identified as BCC, which is the preferred structure below 911 °C,^[37] and the lattice spacings of (110) and (020) are 0.201 and 0.174 nm, respectively, in agreement with those of pure Fe crystal (JCPDS 87-0722). In contrast, the supported oxide film exhibits an amorphous nature without lattices.

2.2 Quantitative Analyses of Exsolved Fe Particles during Reductive Annealing

Real-time observations enable us to quantify the size evolution and density distribution of Fe particles at different temperatures through direct measurements of individual particles. We observe that the sizes of Fe particles rapidly reach saturation within tens of seconds and remain constant over an extended period (≈10 min), without further nucleation or coarsening after reaching to designated temperatures. Figure 2a,b shows how temperature affects the average diameter and density of the particles, respectively, across all Fe concentrations. It is evident that as temperature increases, the sizes and densities of the particles either increase or remain nearly constant until reaching saturation, beyond which the SrTiO₃ films crystallize because Fe exsolution occurs under suppressed particle coarsening. The only exception is the sample with 50% Fe doping, where the particle density decreases with temperature, opposite to the trend of average diameter, due to severe particle coarsening during exsolution caused by the high doping concentration and high atomic diffusivity^[38] in the amorphous substrate. However, particle coarsening ceases after the crystallization of the film because diffusivity becomes lower in the crystalline state.^[38] In the case of 70% doping, despite the higher doping level, both particle size and density increase simultaneously. This is because a substantial amount of Fe has already exsolved at around 500 °C, resulting in significant coarsening at lower temperatures (see Figure 1s–z) and minimal increases in area and volume with temperature compared to the 30% Fe and 50% Fe samples (see Figure 2d,e). As a result, the spacing between particles becomes considerably larger, and with further temperature increases, almost no additional coarsening occurs. Thus, exsolved atoms either attach to existing particles, further increasing their size, or form new particles on the broad substrate areas between the existing particles.

Consequently, the total surface area (R²) and volume (R³) of the particles, plotted against temperature at various Fe concentrations, exhibit a consistent general trend as observed in Figure 2a,b (Figure 2d,e), suggesting Fe atoms continue to exsolve with increasing temperature and no further Fe atoms exsolve after reaching saturation. This behavior is corroborated by the particle size distributions, which show collapsed Gaussian plots across the temperature range (Figure 2c and S3, Supporting Information). The thermal stability observed after the crystallization of the film, characterized by negligible aggregation or coarsening of particles, is attributed to the Fe particles embedded in the SrTiO₃ films, as demonstrated in Figure 2h.

Specifically, after exposure to air, the Fe particle is oxidized to the Fe₃O₄ particle while remaining embedded in the SrTiO₃ film. The particle has a FCC crystal structure, with lattice spacings of 2.96 nm for the (220) plane and 2.53 nm for the (311) plane (JCPDS 88-0866) (see the HRTEM image and FFT pattern in Figure S4, Supporting Information). The Fe composition remains nearly constant throughout the film, staying below 0.5% (less than 2.5% of the B site's composition), as exhibited in Figure 2h,i. This indicates nearly all Fe atoms are exsolved from the amorphous film during reductive annealing because the amount of extracted Fe atoms is saturated before the crystallization of the SrTiO₃ film (Figure 2). This trend continues beyond 50% Fe, which is referred to as “amorphous exsolution”. This amorphous exsolution contrasts sharply with conventional exsolution from crystalline oxides, termed “crystal exsolution”. In the 30% Fe-doped crystalline SrTiO₃ film, the Fe composition remains constant at 5.5% starting from ≈5 nm below the film surface, with only 0.5% of Fe atoms in total composition (2.5% of B site's composition) being exsolved, even under a lower oxygen partial pressure (P_oxy = 10⁻¹⁹ Torr) than that used for amorphous exsolution (see the cross-sectional TEM image and energy dispersive X-ray spectrometer (EDX) line scan in Figure S5, Supporting Information and the plan view TEM image in Figure 6p). Significant exsolution occurs only near the film surface. Such insufficient exsolution typically occurs in various oxide systems during crystal exsolution.^[34-36] Additionally, we note that the saturated R² of the particles in the samples with 30% and 50% Fe doping are similar due to the increased averaged particle sizes with higher Fe concentration.

2.3 X-ray Diffraction (XRD) and X-ray photoelectron Spectroscopy (XPS) Analyses of SrTi_1−xFe_xO₃ Films after Exsolution

To corroborate the crystal structures of the phases observed during the in situ TEM work, XRD spectroscopy was performed. The XRD data acquired from the samples with various Fe doping concentrations after annealing at 850 °C under P_oxy = 10⁻¹⁷ Torr are shown in Figure 3a. The spectra from the samples with 0%–50% Fe doping exhibit strong SrTiO₃ peaks (JCPDS 79-0176). Above 70% Fe, the SrTiO₃ peaks are completely absent because the SrTiO₃ is amorphous. The (220) and (311) peaks of Fe₃O₄ appear at 30%–70% Fe doping (JCPDS 88-0866), indicating the oxidation of Fe particles, except at 10%, where the amount of exsolved Fe is too small to be detected. The XRD spectra obtained from the 30% Fe samples annealed at various temperatures and under P_oxy = 10⁻¹⁷ Torr are exhibited in Figure 3b, confirming the crystallization process of the Fe particles and the SrTiO₃ film, as observed using in situ TEM in Figure 1.

To investigate the relationship between the oxygen chemical environment and different Fe doping levels, XPS analysis was conducted on the same samples, as shown in Figure 3a. The O 1s spectra for samples with varying Fe concentrations are presented in Figure 3c. The spectra primarily exhibit three peaks corresponding to different oxygen species in SrTiO₃: lattice oxygen (O²⁻) at 529.2 eV, highly oxidative oxygen (O⁻, ${\mathrm{O}}_2^{-2}$) at 530.3 eV, and surface oxygen (OH) at 531.8 eV.^{[36, 39, 40]} The lattice O peak retains the same binding energy up to 50% Fe but it shifts to a higher binding energy of 529.6 eV at 70% Fe, likely due to the formation of amorphous SrTiO₃ (a similar trend is observed in the Sr 3d spectra shown in Figure S6a, Supporting Information).^[41] Furthermore, up to 50% Fe, the surface oxygen and the highly oxidative oxygen peaks retain the similar binding energies, while their intensities increase relative to the lattice oxygen peaks as temperature increases (see the ratios in Figure 3e). This suggests the generation of more oxygen vacancies during Fe exsolution at higher Fe concentrations, which promotes the formation of other oxygen-containing species, such as the OH group, on the film surface.^[36] The Fe 2p spectrum can be deconvoluted into six peaks with two main peaks of Fe 2p_3/2 and Fe 2p_1/2 at 710.2 and 723.7 eV, as shown in Figure 3d. The two peaks (in green) located at 710 and 722.5 eV can be attributed to 2p_3/2 and 2p_1/2 of Fe²⁺ species.^{[42, 43]} And the other two peaks (in orange) at binding energies of 712.5 and 725 eV correspond to 2p_3/2 and 2p_1/2 of Fe³⁺ species.^{[41, 44]} The remaining two peaks (in purple) at 717.7 and 729.7 eV are satellite peaks, indicating the formation of Fe₃O₄ particles on the film surface.^[45]

2.4 PEC Water-Splitting Performances of SrTi_1−xFe_xO₃ Samples after Exsolution

We further investigated the effect of Fe doping on the PEC water-splitting performance of the SrTi_1−xFe_xO₃ samples used as photoanodes, following thermal reduction at 850 °C under P_oxy = 10⁻¹⁷ Torr. Linear sweep voltammetry (LSV) curves reveal a significant increase in photocurrent densities for all photoanodes under illumination compared to dark conditions. This increase ranges from 18.75% to 251.72% at 1.23 V versus reversible hydrogen electrode (RHE), depending on the Fe concentration (Figure 4a and Table S2, Supporting Information). The photocurrent density at 1.23 V versus RHE initially rises with increasing Fe concentration, peaking at 5.10 mA cm⁻² for 30% Fe, which is an 11.86-fold improvement over bare SrTiO₃. This improvement is attributed to the electric field in the space-charge region at the interface between the SrTiO₃ photoanode and Fe₃O₄ cocatalysts, as described in Figure 4e,f, which facilitates the separation of photogenerated charges.^[46] The current density achieved in our work is superior to those reported in previous studies (see Table S1, Supporting Information). We will discuss the detailed estimation of the band diagrams of the materials before and after they join each other, and the effects of the junction in the next section.

However, beyond 30% Fe, the photocurrent density dramatically decreases by 52% to 2.46 mA cm⁻² at 50% Fe due to increased Fe₃O₄ particle sizes and the similar surface coverage of the particles as at 30% Fe (Figure 2a,d). This results in longer specific diffusion lengths for the photo-holes, and reduced light absorption.^{[47, 48]} The photocurrent density continues to decrease by 96% to 0.19 mA cm⁻² at 70% Fe due to the formation of an amorphous SrTiO₃ film (see Figure 1z), which can form numerous defects that act as trap recombination centers within its forbidden band.^{[49, 50]} Electrochemical impedance spectroscopy (EIS) supports the LSV results, showing that the diameter of the semicircle decreases up to 30% Fe, and then increases (Figure 4b).^{[51, 52]}

Figure 4c displays the applied-bias photon-to-current conversion efficiency (ABPE) (see details in Supporting Information). Initially, ABPE increases with Fe concentration: 0.47% at 0.45 V for 10% Fe and 1.54% at 0.62 V for 30% Fe, representing over a 2.61-fold and 8.56-fold improvement compared to bare SrTiO₃, respectively. However, at higher Fe concentrations, photoconversion efficiency significantly decreases to 1.00% at 0.42 V for 50% Fe and falls below that of bare SrTiO₃ at 70% Fe. Long-term stability tests of the photoanodes were conducted at 1.23 V versus RHE under illumination for 24 h, as shown in Figure 4d. The photoanode with 30% Fe demonstrates the highest consistency with only a decline rate of ≈3% per day and the highest current density. This is attributed to the largest surface coverage of Fe₃O₄ particles on the SrTiO₃ photoanode and smaller particle sizes compared to those at 50% Fe (Figure 2a,d). These particles enhance charge separation/transfer and passivate the surface states of the photoanode, thereby preventing photocorrosion by consuming the remaining photoinduced holes.^{[39, 40]} For other Fe concentrations, photocurrent density and/or stability decrease considerably due to the reduced surface coverage of Fe₃O₄ particles and/or larger particle sizes (Figure 2a,d). We confirm that the SrTiO₃ film and Fe₃O₄ particles remain largely unchanged after the oxygen evolution reaction (OER) reaction. This is supported by the XRD and XPS spectra of the 30% Fe-doped SrTiO₃ sample obtained after the stability test (Figure S7 and S8, Supporting Information). The data show results that are nearly identical to those measured prior to the stability test. Additionally, the dissolved Fe from the 30% Fe-doped sample in the electrolyte after a 24-h stability test (see Figure 4d) was analyzed using inductively coupled plasma mass spectrometry. The concentration of dissolved Fe was estimated to be ≈0.48 nmol, which is about 2.8% of the total Fe content in the Fe₃O₄ particles embedded in the SrTiO₃ film, consistent with the stability test results. This small amount of dissolved Fe confirms the high stability of the optimized sample.

2.5 Estimating Electronic Band Structures and Understanding PEC Efficiencies

To estimate the band structures of the optimized photoanodes, the 30% Fe-doped SrTiO₃, and to understand the impact of the SrTiO₃/Fe₃O₄ heterostructures on PEC performance, we utilized a range of analytical techniques. These include UV–visible absorption spectroscopy (UV–vis), ultraviolet photoelectron spectroscopy (UPS), Mott-Schottky (M-S), photoluminescence (PL), time-resolved photoluminescence (TRPL), incident photon-to-electron conversion efficiency, charge separation/transfer efficiencies, electrochemically active surface area (ECSA), and Faradaic efficiency (FE) measurements (see details of the techniques in Supporting Information).

The UV–vis absorption spectra of SrTiO₃, Fe₃O₄, and 30% Fe-doped SrTiO₃ photoanodes are displayed in Figure 5a, to assess their absorption capabilities in the 300 to 800 nm range. The SrTiO₃ and Fe₃O₄ films show light absorption edges at ≈374 and 748 nm, respectively. This information facilitates the estimation of the photoanodes’ bandgaps, calculated to be 3.24 eV for SrTiO₃^{[53, 54]} and 1.75 eV for Fe₃O₄,^{[55, 56]} by drawing a tangent to the curve in the Tauc plot ((αhν)^1/2 versus photon energy, Figure 5b) (see Equation (S2), Supporting Information). Notably, anchoring Fe₃O₄ particles into the SrTiO₃ films enhances light absorption compared to pristine SrTiO₃ and extends the light absorption region to ≈550 nm (Figure 5a). These behaviors contribute to the improved PEC performance as demonstrated above.

The electronic structures of SrTiO₃ and Fe₃O₄ were investigated using UPS (Figure 5c). This analysis determines the work function and the valence band maximum (VBM) relative to the Fermi level (E_F).^[57] The work functions are calculated to be −4.45 eV for SrTiO₃ and −5.90 eV for Fe₃O₄. Using the bandgaps obtained from UV–vis measurements, the valence band positions are determined to be −7.15 eV for SrTiO₃ and −6.46 eV for Fe₃O₄. Correspondingly, the conduction band positions are calculated to be −3.90 eV for SrTiO₃ and −4.76 eV for Fe₃O₄. The data obtained from UV–vis and UPS measurements are depicted in Figure 4g, providing an estimation of the band structures of the SrTiO₃ and Fe₃O₄ photoanodes. The resulting band alignment confirms the formation of a type-II heterojunction between SrTiO₃ and Fe₃O₄ (Figure 4h), indicating a structure that facilitates efficient charge separation, beneficial for the oxygen evolution reaction.

M-S measurements were performed to gain qualitative insights into the formation of p–n junctions in the photoelectrodes (Figure 5d). The positive and negative slopes confirm that SrTiO₃ and Fe₃O₄ are n-type and p-type semiconductors, respectively.^{[46, 58]} The flat band potentials (E_fb) of SrTiO₃, Fe₃O₄, and 30% Fe-doped SrTiO₃ were determined by extrapolating the x-intercepts from the M-S equation^[59] (see Equation (S3), Supporting Information). The flat band potentials obtained are −0.05 V for SrTiO₃, 0.36 V (E_fn) and 1.23 V (E_fp) for 30% Fe-doped SrTiO₃, and 1.52 V for Fe₃O₄ (V vs RHE), respectively. These values demonstrate that the E_fb of SrTiO₃ shifts positively and the E_fb of Fe₃O₄ shifts negatively after forming the type-II heterostructure between SrTiO₃ and Fe₃O₄. These shifts indicate the Fermi levels of n-type SrTiO₃ shift downward, and those of p-type Fe₃O₄ shift upward, along with energy bands in the heterostructure. Such behavior increases band bending, enhancing the separation of photogenerated charges and facilitating more photogenerated holes reaching the electrode surface (Figure 4g,h). Additionally, the carrier density (N_D) in the photoanodes can be estimated from the slopes of the linear portions for the M-S plots^[60] (see Equation (S3), Supporting Information). The carrier densities are 2.64, 27.3, and 9.54(E_fn) and 68.1(E_fp) × 10¹⁸ cm⁻³ for SrTiO₃, Fe₃O₄, and 30% Fe-doped SrTiO₃, respectively, implying that the Fe₃O₄ particles embedded in SrTiO₃ enhance charge transport properties, thus improving PEC performance.

PL emission spectra serve as an effective tool to understand of the recombination process of photogenerated electrons and holes. A higher intensity in the steady-state PL emission spectrum indicates faster recombination of photogenerated charges, which reduces photocatalytic activity.^{[61, 62]} Figure 5e shows the steady-state PL spectra of SrTiO₃% and 30% Fe-doped SrTiO₃ photoanodes, with prominent peaks around 525 nm, mainly attributed to the oxygen vacancies near the SrTiO₃ surface.^[63] The PL intensity of 30% Fe-doped SrTiO₃ photoanodes is significantly lower than that of SrTiO₃, indicating that the SrTiO₃/Fe₃O₄ heterostructure effectively suppresses the recombination of photogenerated charge carriers. The TRPL curves of SrTiO₃% and 30% Fe-doped SrTiO₃ photoanodes were fitted by Equation (S4), Supporting Information (Figure 5f), and the average lifetimes (τ_avg) were calculated according to Equation (S5), Supporting Information. As listed in Table S3, Supporting Information, the average lifetime of the 30% Fe-doped SrTiO₃ photoanode (11.42 ns) is longer than that of the SrTiO₃ (3.43 ns), confirming that Fe₃O₄ particles embedded in the SrTiO₃ surface effectively reduce charge recombination within the film.

To assess the photoresponse of the photoanodes, IPCE measurements were performed in the 300–650 nm wavelength range at 1.23 V versus RHE. The 30% Fe-doped SrTiO₃ exhibits an IPCE value of 79.7% at 339 nm, significantly higher than the ≈10.9% observed for pristine SrTiO₃ at the same wavelength (Figure 5g). The 30% Fe-doped SrTiO₃ photoanode shows excellent solar-to-current conversion efficiency across the 300–590 nm spectrum, consistent with the photocurrent density observed in the LSV curve. The onset wavelength of the photoanode is red-shifted compared to that of SrTiO₃, as verified by their Tauc plots (Figure S9, Supporting Information), indicating the establishment of a type-II band alignment.

To uncover the mechanisms behind the enhanced PEC performance, studies on bulk charge separation and surface charge transfer dynamics of SrTiO₃ and 30% Fe-doped SrTiO₃ photoanodes were conducted by comparing their LSV curves obtained under the water and sulfite oxidation conditions (Figure S10, Supporting Information). The efficiencies of bulk charge separation (η_bulk) and surface charge transfer (η_surf) were calculated to determine the effects of Fe₃O₄ cocatalysts, using Equation (S8) and (S9), Supporting Information, respectively. The 30% Fe-doped SrTiO₃ exhibits a η_bulk value of 80%, which is 3.2 times higher than the 25% observed for SrTiO₃ at 1.23 V versus RHE (Figure 5h). This increase is mainly attributed to the type II band alignment at the interface between the SrTiO₃ film and Fe₃O₄ particles, promoting effective charge separation. Additionally, the η_surf value for 30% Fe-doped SrTiO₃ reaches 90%, which is 3.6 times greater than the 25% for SrTiO₃ at 1.23 V versus RHE (Figure 5i). This superior η_surf for the 30% Fe-doped SrTiO₃ is due to Fe₃O₄ particles enhancing charge transfer kinetics and minimizing surface recombination through surface passivation.

To evaluate the OER properties at the photoanode surface, the ECSA is typically measured. This is determined from the electrochemical double-layer capacitance (C_dl), which is directly proportional to the ECSA^[64] (see Equation (S11), Supporting Information). The C_dls of the SrTiO₃ and 30% Fe-doped SrTiO₃ photoanodes were calculated from the slopes of their plots of ΔJ versus scan rates (Figure 5j), derived from the cyclic voltammetry curves (Figure S11, Supporting Information). The 30% Fe-doped SrTiO₃ achieves a C_dl of 0.484 mF cm⁻², significantly higher than the 0.092 mF cm⁻² observed for SrTiO₃. This substantial increase in active sites at the surface is attributed to the Fe₃O₄ cocatalytic particles, which enhance the η_surf, as discussed in Figure 5i.

To ensure the photocurrent is effectively used for water oxidation, the Faradaic efficiency of the evolved O₂ and H₂ gases was measured. The gas evolution over time for SrTiO₃% and 30% Fe-doped SrTiO₃ photoanodes during a 3-h period is shown in Figure S12a,c, Supporting Information, respectively, with the corresponding Faradaic efficiencies in Figure S12b,d, Supporting Information. The results indicate stoichiometric production of H₂ and O₂, confirming successful water splitting. Figure 5k compares the actual O₂ evolution rates with theoretical predictions, showing that the 30% Fe-doped SrTiO₃ photoanode produces O₂ at a rate of 35.47 μmol h⁻¹, which is about 13 times higher than the 2.72 μmol h⁻¹ rate of SrTiO₃. Additionally, Figure 5l highlights that the Faradaic efficiency for the 30% Fe-doped SrTiO₃ reaches ≈93.1%, significantly surpassing the ≈79.0% efficiency of SrTiO₃, indicating more effective utilization of photogenerated holes for oxygen evolution. Figure 5m presents a comparison between the actual H₂ evolution rates and the theoretical expectations, revealing that the 30% Fe-doped SrTiO₃ photoanode achieves a production rate of 74.72 μmol h⁻¹, which is ≈12.3 times greater than the 6.06 μmol h⁻¹ rate observed for SrTiO₃. Furthermore, Figure 5n demonstrates that the Faradaic efficiency of the 30% Fe-doped SrTiO₃ reaches around 97.3%, significantly exceeding the 87.8% efficiency of SrTiO₃. This indicates a more efficient utilization of photogenerated electrons for hydrogen evolution.

2.6 Density Functional Theory (DFT) Calculations of Amorphous and Crystalline SrTi_1−xFe_xO₃

To explore the exsolution behavior in both amorphous and crystalline SrTi_1−xFe_xO₃ phases (x = 0–1.0), we employed DFT to assess electronic structures, bonding features, and vacancy formation energies (V_O and V_Fe). First, the density of states (DOS) (Figure S13c–f, Supporting Information) was analyzed to examine bonding and antibonding characteristics in TiO and FeO interactions, using crystal orbital Hamilton population (COHP)^[65] (Figure 6c,d). COHP values that are positive indicate bonding states, while negative values represent antibonding states. In SrTiO₃, the occupied states in the valence band and the unoccupied states in the conduction band are generated by TiO bonding and antibonding interactions, respectively. In SrTi_0.5Fe_0.5O₃, FeO interactions in both amorphous and crystalline phases show antibonding characteristics near the VBM (≈−1.0 to 0.0 eV), indicating weaker bond strengths compared to those in SrTiO₃. Notably, in the amorphous phase, FeO antibonding interactions are more pronounced than in the crystalline phase, indicating a reduction in FeO bond strength (see integrated COHP in Figure 6e). Although both SrFeO₃ phases are metallic, the crystalline phase significantly intensifies antibonding interactions.

Subsequently, O and Fe vacancy formation energies were determined (Figure 6h,i) at the (111) surfaces (Figure 6f,g). The trends in vacancy formation energies for O and Fe are generally consistent: 1) vacancy formation energies in the amorphous phase are notably lower compared to the crystalline phase, a consequence of the diminished FeO bond strength discussed earlier; and 2) As Fe concentration increases, Fe vacancy formation energy decreases in both phases, attributed to the coordinate covalent bonding between Fe and O,^{[66, 67]} which weakens the bond strength in comparison to the covalent bonding between Ti and O.^[68-71] The weakening of the bonds promotes exsolution in both phases and causes instability in the SrTiO₃ structure during the crystal exsolution.^[36] Interestingly, the initial increase in O vacancy formation energy at around 20% Fe composition is linked to the delocalization of electronic structures in Fe and Ti.

2.7 Thermodynamic Mechanism of Amorphous Exsolution in SrTi_1−xFe_xO₃ in Comparison with Crystal Exsolution

Our concept of “amorphous exsolution” occurs when the energy required to form Fe vacancies is lower than the energy needed for crystallization from the amorphous state of the oxide support (illustrated by the red solid and short-dotted curves in Figure 6j). The entire mechanism is detailed in Figure 6k. In contrast, exsolution systems previously reported and fabricated through solution processes typically follow a “crystal exsolution” pathway, which occurs under conditions opposite to those required for amorphous exsolution (represented by the red long-dotted curve in Figure 6j). This mechanism adheres to the black landscape in Figure 6j.^[72-79] Notably, if the dopant cannot exsolve from the amorphous oxide, it must overcome a substantially higher energy barrier to form a vacancy within the crystal structure (indicated by the two-way arrow in Figure 6j).

Crystal exsolution systems are typically crystallized by calcining solutions in oxidizing conditions (air or oxygen environments) with low doping levels (less than 10% in B sites of perovskite), which inhibits oxygen vacancy formation and limits dopant exsolution.^[68] To investigate further, we annealed a 30% Fe-doped SrTiO₃ solution in different environments at 850 °C. In air or under O₂ gas flow, no exsolution was observed due to the high Fe vacancy formation energy barrier imposed by these oxidative conditions (Figure 6l and the long-dotted red curve in Figure 6j). In contrast, under vacuum conditions, a few Fe particles were exsolved on the SrTiO₃ films, attributed to a lower Fe vacancy formation energy barrier (Figure 6m and the short-dotted red curve in Figure 6j). Notably, under a H₂ gas flow (P_oxy = 10⁻¹⁷ Torr), the same condition in Figure 1, a significant amount of Fe particles (representing all Fe dopants) were exsolved on the film, due to the substantially reduced Fe vacancy formation energy barrier (Figure 3n and the solid red curve in Figure 3j).

To compare these conditions with crystal exsolution, we annealed the sample in Figure 6l under different degrees of reducing conditions. In vacuum, no exsolution occurs (Figure 6o), and under a oxygen partial pressure (P_oxy = 10⁻¹⁹ Torr) (lower than that in Figure 6n), only a small amount of Fe is exsolved (Figure 6p and S5, Supporting Information). These behaviors are expected due to the high free energy barrier to form Fe vacancies compared to the driving forces provided by the two reducing conditions. When a lower oxygen partial pressure (P_oxy = 10⁻²¹ Torr) is imposed, the film decomposes, exsolving both Sr and Fe, thereby forming SrO and Fe₃O₄ particles on the film after exposure to air (Figure S14a, Supporting Information), resulting in severely degraded PEC activity and stability (Figure S15, Supporting Information). Such poor PEC performance is ascribed to: 1) the large bandgap of SrO (5.7 eV) (Figure S16, Supporting Information), which causes reduced light absorption; 2) the large particles, which results in longer specific diffusion lengths for the photo-holes and reduced light absorption; and 3) a large amount of oxygen vacancies (Figure S14b, Supporting Information).^{[47, 48]} Detailed explanations on the PEC performance of the crystal exsolution and relevant analyses are provided in Section 5, Supporting Information. Importantly, the total volume (R³) of the SrO and Fe₃O₄ particles only reaches ≈71% of that in amorphous exsolution (Figure 2e and 6n), implying that significantly fewer Fe dopants can be exsolved by crystal exsolution without decomposing Sr, regardless of the reducing conditions applied, such as low oxygen partial pressures.