Colossal oxygen vacancy formation at a fluorite-bixbyite interface

Oxygen vacancies in complex oxides are indispensable for information and energy technologies. There are several means to create oxygen vacancies in bulk materials. However, the use of ionic interfaces to create oxygen vacancies has not been fully explored. Herein, we report an oxide nanobrush architecture designed to create high-density interfacial oxygen vacancies. An atomically well-defined (111) heterointerface between the fluorite CeO2 and the bixbyite Y2O3 is found to induce a charge modulation between Y3+ and Ce4+ ions enabled by the chemical valence mismatch between the two elements. Local structure and chemical analyses, along with theoretical calculations, suggest that more than 10% of oxygen atoms are spontaneously removed without deteriorating the lattice structure. Our fluorite–bixbyite nanobrush provides an excellent platform for the rational design of interfacial oxide architectures to precisely create, control, and transport oxygen vacancies critical for developing ionotronic and memristive devices for advanced energy and neuromorphic computing technologies.

O wing to the critical role of tunable oxygen defects, mixed ionic and electronic conducting oxides have become key components in many energy technologies, including energy conversion and generation 1 , solar hydrogen production 2 , humidity/gas detection 3 , and catalytic conversion of toxic species in automobile exhausts 4 . In particular, ceria (CeO 2 ) has attracted extensive attention for its good ionic conductivity and remarkable redox capability 5 and hence is widely used as a heterogeneous catalyst 6 and oxygen storage component 4 . Because the key element is the oxygen vacancy 7 , the creation of oxygen vacancies is a prerequisite for such CeO 2 -based ionic materials and devices. CeO 2 has a fluorite-type structure with a coordination number of 8 for Ce to O atoms 8 , so aliovalent chemical doping (e.g., yttrium, samarium, and gadolinium) into CeO 2 is one of the conventional approaches 9 to create oxygen vacancies. Nanoscaling, interfacing, and straining thin films are also known mechanisms to create oxygen vacancies [10][11][12] . However, the implementation of these approaches for practical applications requires further advancements because the fundamental understanding of oxygen vacancy formation at the atomic level is still lacking.
It was proposed that a space charge region (SCR) [12][13][14] could be formed by interfacing the fluorite CeO 2 (space group Fm 3m, a = 5.412 Å) with a bixbyite oxide, such as Y 2 O 3 (space group Ia 3, a = 10.607 Å), along the [100] direction 15,16 . The bixbyite structure has a "pseudofluorite" structure with an ordered array of vacant oxygen sites occurring on every fourth site (Supplementary Fig. 1), providing an ideal atomic arrangement for forming an interfacial charge layer that originates from the valence mismatch between Ce(4+) and Y(3+). However, the formation of an atomically well-defined (001) interface between the two binary oxides is energetically unfavorable, as the (001) surface has the highest surface energy among the low-index surfaces for both materials. In contrast, the (111) surfaces of fluorite and bixbyite structures are energetically more favorable because of their low surface energy 17,18 . Thus, determining the viability of the artificial creation of oxygen vacancies at an atomically well-defined (111) interface and understanding the (111) fluorite-bixbyite interface could open a new avenue toward the development of highperformance oxide-based ionotronic devices. Moreover, the Ce 4+ O 2 /Y 3+ 2 O 3 interface offers an interfacial condition similar to that of semiconductor and perovskite-oxide heterointerfaces having polar discontinuity [19][20][21] . Therefore, understanding the details of the interfacial oxygen vacancy formation reported herein will provide useful insights into developing a novel concept of two-dimensional (2D) ionic channels using atomically engineered oxide interfaces.
In this work, we investigate the viability of colossal formation of oxygen vacancies in CeO 2 /Y 2 O 3 superlattices with atomically well-defined (111) interfaces formed within a nanobrush architecture (Fig. 1). The nanobrush superlattices offer many advantages over 2D thin films, including much larger surface areas and a larger number of interfaces than exist in planar films, which are useful for many applications in various technologically relevant areas 22 .

Results
Precision synthesis of a (111) fluorite-bixbyite nanobrush. Figure 1a shows a cross-sectional high-angle annular dark-field (HAADF) image obtained using scanning transmission electron microscopy (STEM) from a 1.4-μm-thick CeO 2 /Y 2 O 3 nanobrush superlattice grown on a Y 2 O 3 -stabilized ZrO 2 (YSZ)-buffered (001) silicon substrate. The Z-contrast (Z denotes the atomic number) HAADF image clearly reveals alternating stacks of CeO 2 (Z Ce = 58) and 200 nanobrush superlattice was grown by pulsed laser deposition under an extremely nonequilibrium supersaturated condition, which is far away from the conventional condition optimized for the layer-bylayer growth of 2D thin films. Despite the porous architecture with single-crystalline bristles and the large thickness, the created nanostructure was highly epitaxial and (001)-oriented, as confirmed by X-ray diffraction (XRD) 2θ−θ scans and pole figure measurements ( Supplementary Fig. 2). Based upon the XRD data, the epitaxial orientation relationship of the CeO 2 /Y 2 O 3 nanobrush superlattice on the YSZ-buffered (001) silicon substrate was Y 2 O 3 (001)∥CeO 2 (001)∥YSZ(001)∥Si(001), a finding also confirmed by selected area electron diffraction ( Supplementary Fig. 3). While a variety of oxide nanostructures-such as nanorods 23,24 , nanowires 25 , nanobelts 26 , and nanonails 27 -have been successfully demonstrated, our nanobrush architecture is a revolutionary new form of oxide nanostructure in which the growth is governed by both diffusion-limited aggregation and the shadow effect (Supplementary Fig. 4). The details of the growth mechanism for oxide nanobrush architectures were reported elsewhere 28 . Interestingly, a characteristic chevron pattern can be seen within each bristle. This chevron pattern originates from the {111} facet formation for both CeO 2 and Y 2 O 3 , owing to the lowest surface energy 17,18 . The interface connection of the {111} facets of the CeO 2 and Y 2 O 3 sublayers was the key to the spontaneous formation of oxygen vacancies on the CeO 2 side, as discussed below.
We performed atom probe tomography (APT) on the CeO 2 -Y 2 O 3 nanobrush to reveal the three-dimensional (3D) geometry of the nanobristles. Owing to the small nanobristle size (only~30 nm in the lateral dimension), the growth of a nanobristle on a micron-wide silicon-tip for APT was highly challenging. Nevertheless, we successfully grew a nanobrush sample on a special silicon microtip array and prepared an APT specimen via focused ion beam (FIB) milling, as shown in Supplementary Fig. 5. Figure 1b shows an APT image (see also Supplementary Movie 1 for 3D view) of a (CeO 2 ) 6 u.c. /(Y 2 O 3 ) 2 u.c. nanobristle. The APT investigation confirmed the chevron pattern on the nanobristles. The overall "Christmas tree-like" shape of the bristles and the chevron pattern on each bristle are only visible along the <110> direction, whereas the <100> directional view reveals a horizontal layer-by-layer stacking of superlattices, clearly visualizing the 3D geometry of the nanobrush samples. Figure 1c schematically shows top and cross-sectional views of the oxygen coordination mismatch at the (111) interface between the fluorite and the bixbyite, depicting the underlying mechanism for the formation of oxygen vacancies. Only oxygen atoms are shown to reveal the difference in the oxygen network between (111)CeO 2 and (111)Y 2 O 3 . While the structure of a (001) fluorite/ bixbyite interface is rather simple ( Supplementary Fig. 1), the oxygen network along the (111) plane is rather complex, forming a triangular network of oxygen vacant sites in Y 2 O 3 . The crosssectional views cut along the C1, C2, and C3 directions clearly visualize the oxygen mismatch between the two materials. This interfacial oxygen coordination mismatch can explain the underlying mechanism behind the formation of the interfacial oxygen vacancies. To cope with the valence mismatch in Y 3+ vs. Ce 4+ , the following two scenarios were considered: (1) formation of an oxygen interstitial in Y 2 O 3 and (2) formation of an oxygen vacancy in CeO 2 .
Artificial formation of oxygen vacancies at the (111) interface.
To confirm the viability of highly confined formation of oxygen vacancies at a (111) interface in a CeO 2 -Y 2 O 3 nanobrush, we performed a systematic STEM study. Figure 2 shows HAADF and low-angle annular dark-field (LAADF) images obtained from both the tip and middle parts of a [( nanobristle. In the HAADF images ( Fig. 2a, b), the chevron pattern of a stack of alternating CeO 2 and Y 2 O 3 layers is clearly visible. As indicated in Fig. 2e, the angle between the two {111} facets is~70.5°, which can readily be seen when viewed along the [110] direction, confirming the formation of (111) interfaces. The STEM images also reveal that the outer portion of each layer is thinner than its center part. In addition, a fast Fourier transform analysis clearly shows a fully coherent interface (i.e., free from any dislocations) formed between CeO 2 and Y 2 O 3 layers (Supplementary Fig. 6). Moreover, at the edge region of the bristles in the HAADF images, the CeO 2 layers encase the Y 2 O 3 layers. The capping yields vertically connected CeO 2 ( Supplementary Fig. 7). This may be attributed to the lower Ehrlich-Schwoebel barrier 29 for CeO 2 on Y 2 O 3 than for Y 2 O 3 on CeO 2 .
The two interfacial reconstruction scenarios mentioned previously, i.e., (1) formation of an oxygen interstitial on the Y 2 O 3 side and (2) formation of an oxygen vacancy and subsequent Ce 3+ ions on the CeO 2 side, can result in a local lattice distortion. Such structural disorder can be detected by LAADF imaging, as it leads to electron dechanneling of the incident e-beam 30,31 . As shown in Fig. 2c, d, the LAADF images reveal an enhanced intensity at the interface between CeO 2 and Y 2 O 3 . Through a careful examination of the images, we found that the bright contrast originated from the CeO 2 side of the interfaces (see the yellow lines in Fig. 2a-d). Note that we have not observed any bixbyite Ce 2 O 3 structures at the CeO 2 /Y 2 O 3 interface, as confirmed by HAADF imaging (Supplementary Fig. 8). Considering the fact that the lattice mismatch between Ce 2 O 3 and Y 2 O 3 is~10%, the formation of the bixbyite Ce 2 O 3 phase may not be energetically favorable at the interfaces. LAADF imaging also confirmed that the structurally distorted layer owing to the chemical expansion spanned over~1 nm in thickness (Supplementary Fig. 9). In addition, a previous TEM study reported a large lattice expansion occurred in CeO 2 nanoparticles, owing to the formation of oxygen vacancies with Ce 3+ ions, without changing the original fluorite structure 32 . Thus, our results suggest that the highly localized formation of a large number of oxygen vacancies on the CeO 2 side of the interfaces was responsible for the observed electron dechanneling. The formation of interfacial oxygen vacancies was further confirmed by a geometric phase analysis (GPA) of STEM images. As shown in Fig. 2f, the GPA result reveals a significant lattice expansion by up to 3% for the CeO 2 interfacial regions, consistent with the LAADF results. Furthermore, we can also exclude the possibility of the formation of oxygen interstitials on the Y 2 O 3 side of the interface-one of the two interfacial reconstruction scenarios mentioned earlier based upon the STEM data-as we have not observed any atomic structural changes in Y 2 O 3 interfacial layers.
Interfacial cerium valence modulation. The interfacial oxygen vacancy formation was further investigated by electron energyloss spectroscopy (EELS), which highlighted the elemental distributions, atomic configurations, and bonding states of both Ce and O. Figure 3a shows an elemental map produced using the Ce-M 4,5 edge in the vicinity of the (111)CeO 2 /(111)Y 2 O 3 interface, revealing interfaces that are not only structurally well-defined but also chemically sharp. Background-subtracted Ce-M 4,5 edge spectra obtained from each atomic plane across the (111)CeO 2 / (111)Y 2 O 3 interface are shown in Fig. 3b. The peak positions of Ce-M 4,5 edges are shifted to a lower energy-loss region for the reduced CeO 2 interfacial layers, as compared with that of the CeO 2 interior layer. The Ce-M 4,5 edges originate from transitions between the 3d and 4f states (M 4 corresponds to the 3d 3/2 to 4f 5/2 transition and M 5 corresponds to the 3d 5/2 to 4f 7/2 transition) combined with ligand hole effects and valence-sensitive fine structures. Thus, spatially resolved chemical information 33  obtained from EELS measurements. Using linear combinations of Ce-M 4,5 reference spectra from CeO 2 and Ce 2 O 3 34 , the valence states of the reduced cerium atoms near the interface were roughly estimated to be Ce +3.3 (the nearest to the interface) and Ce +3.6 (the second nearest), which correspond to CeO 1.65 and CeO 1.8 , respectively, as illustrated in Fig. 3c. These oxidation states led to 17.5% and 10% of missing oxygen atoms, respectively, which was indeed a large amount, especially considering the robust interface lattice structures observed by STEM. Background-subtracted O-K spectra averaged along the (111)O planes show a considerable number of oxygen vacancies are formed at interfaces to cope with the interfacial charge mismatch (Supplementary Fig. 10). It is worth stressing that such an interfacial charge modulation is a powerful method of creating a high concentration of oxygen vacancies. Interestingly, our STEM and EELS results suggest that the formation of oxygen vacancies at the (111) interface between the fluorite and the bixbyite is not attributable to the creation of SCRs in the CeO 2 layer, in which the calculated Debye length (λ) of~2.4 nm at 300 K 16,35 can result in a~10 nm thick regime of overlapping space charges (4λ) 12 .

Spatial distributions of oxygen defects at the (111) interface.
To clarify the role of the (111) interface in creating the oxygen vacancies, the defect formation energy of a neutral oxygen vacancy complex, ðCe 0 Ce À V ÁÁ O À Ce 0 Ce Þ , at the (111) interface was computed by density functional theory (DFT) calculations ( Supplementary Fig. 11). The DFT calculations indicated that the positive charge of oxygen vacancies was always compensated by two Ce 3+ ions ( Supplementary Fig. 12). In addition, the DFT results revealed that the formation energies of oxygen vacancies and Ce 3+ ions at the heterointerface were~1.8 and~0.5 eV, respectively. These values are lower than those in bulk CeO 2 as a result of the lower electronic state of the polaron (i.e., reduced cerium cations, Ce 0 Ce ) and the easier local lattice relaxation at the interface. These low formation energies increased the number of both the oxygen vacancies and the Ce 3+ ions at the interface, leading to the segregation of oxygen vacancies on the CeO 2 side of the interface, as shown in Fig. 4. The calculation result showing up to 10% oxygen loss at the interface is quantitatively in good agreement with experimental observations of the EELS data shown in Fig. 3b, c. It is also worth noting that the vacancy segregation was not driven solely by an electrostatic potential at the interface, as the (111) surface was nonpolar so that an SCR could not easily form, and the defect concentration in ceria was high enough to screen out any electrostatic potential (Supplementary Fig. 13). Therefore, the (111) fluorite-bixbyite interface could induce a larger number of oxygen vacancies without the creation of an SCR, which differs from the mechanism proposed for (001) interfaces of fluorite-bixbyite systems in previous studies 15,16 . As was confirmed by STEM imaging, classical force field (FF) simulation also confirmed that the CeO 2 side of the interfaces always remained as the fluorite structure, even though the interface had a high concentration of oxygen vacancies ( Supplementary Fig. 14).

Discussion
There have been several studies of creating oxygen defects, understanding the underlying mechanism for their formation, and probing their existence in mixed ionic and electronic conducting materials 13,14 . However, identifying a new means to artificially create and control highly confined interfacial oxygen vacancies has been a huge challenge. The present work elucidates that interfaces between the fluorite CeO 2 and the bixbyite Y 2 O 3 can provide an excellent platform for developing new ionic materials. Creating atomically sharp (111) fluorite/bixbyite interfaces in the nanobrush architecture provides a completely new approach to spontaneously creating oxygen vacancies through interfacial charge modulation. In contrast to the wellknown SCR effect, the (111) fluorite/bixbyite interface creates and confines oxygen vacancies within two to three atomic layers on the CeO 2 side of the (111) interface. It is remarkable that a large number of oxygen vacancies can be accommodated by taking advantage of the novel interface architecture while leaving the lattice structure intact. We further demonstrated a new "synthesis science" approach 36 to making a specific interfacial nanomaterial with a vertically aligned geometry and a largely increased area of the (111)oriented surface. As shown in Fig. 1a, our CeO 2 -Y 2 O 3 nanobrush has a highly porous topography. Therefore, the quantitative information on porosity was important for accurately determining the surface area, which was an extremely challenging task. The difficulty was overcome by using small-angle neutron scattering (SANS), as shown in Supplementary Fig. 15. The SANS results revealed that the upper limit of the porosity was~49%. More details on the SANS measurement and data analysis can be found elsewhere 28 . The high porosity is highly advantageous for many technical applications, in which a large contact area can improve (electro-)chemical activity, e.g., electrodes of solid oxide fuel/electrolysis cells 37 , oxygen permeation membranes 38 , chemical sensors 39 , and catalysts 40 . In our study, the SANS data confirmed that the~1.4-μm-thick nanobrushes had a surface area that is over 200 times greater than that of a 2D thin film. Therefore, these advances are major technical accomplishments that provide a new means to develop high-performance energy conversion devices, novel catalysts, nonvolatile memories, and memristive computing devices.

Methods
Synthesis of nanobrush samples. Micron-thick single-crystalline fluorite-bixbyite (CeO 2 -Y 2 O 3 ) nanobrush superlattices were fabricated using pulsed laser epitaxy (KrF, λ = 248 nm) with sintered CeO 2 and Y 2 O 3 targets. While all of the samples studied were grown on single-crystal (001) YSZ substrates, the samples for STEM and APT were grown on YSZ-buffered (001) silicon substrates without deteriorating the sample quality to avoid the charging effect resulting from the highly insulating YSZ substrate. Details on the growth of a YSZ buffer layer and the epitaxial relationship of YSZ and silicon can be found elsewhere [41][42][43][44] . All of the samples were deposited at 700°C at an oxygen partial pressure of 100 mTorr. The sample structure and crystallinity were characterized by high-resolution fourcircle XRD.
Scanning transmission electron microscopy. Cross-sectional TEM specimens were prepared using ion milling after mechanical thinning and precision polishing. HAADF and LAADF imaging and EELS analysis were carried out in a scanning transmission electron microscope (Nion UltraSTEM100) operated at 100 keV. The microscope is equipped with a cold field-emission gun and an aberration corrector for sub-Ångstrom resolution. Inner detector angles of 78 and 30 mrad were used for HAADF and LAADF imaging, respectively. The convergence semiangle for the electron probe was set to 30 mrad.
Atom probe tomography. APT specimens were prepared by first coating a silicon microtip array with the CeO 2 /Y 2 O 3 nanobristles. An FEI Nova 200 dual-beam scanning electron microscope/FIB was used to fabricate APT needles from the nanobrushes grown on the silicon microtip array. A 300 nm platinum layer was deposited by FIB, followed by several annular milling steps and a final 2 kV mill to fabricate needle-shaped specimens ready for APT analysis. After sample preparation, the specimen was loaded into the LEAP 4000XR and analyzed using laser mode with a 10 ps/100 pJ laser pulse, a 50 K base temperature, a 125 kHz pulse repetition rate, and a detection rate of 1 ion per 500 pulses. Reconstruction and analysis were performed using the CAMECA integrated visualization and analysis software package.
Density functional theory calculations. In this work, the DFT calculations were performed with VASP 45-47 using the projector augmented wave method. Generalized gradient approximation with the PBE exchange correlation functional 48 and an energy cutoff of 500 eV were used. To properly simulate Ce 3+ , a Hubbard U correction 49 was applied to the Ce 4f orbital with U = 5 eV. Hybrid Monte Carlo and molecular dynamics with classical force field. The vacancy and Ce 3+ concentration profiles were computed with a hybrid MC and MD algorithm using a classical FF. The simulation technique is well documented in the original paper 50 and our previous work 51 . The simulation was started with a uniform distribution of Ce 3+ and oxygen vacancies in the CeO 2 layer. In the MC simulation, only Ce 3+ and Ce 4+ ions were allowed to swap their positions, whereas in the MD simulation, the oxygen vacancies could diffuse to either Y 2 O 3 or CeO 2 layers. More than 10 5 MC trials and 6 ns MD simulations at 600°C were performed. The defect concentration profiles were obtained by averaging the last 5000 configurations of each single calculation.
Small-angle neutron scattering. Room-temperature SANS measurements were performed using EQ-SANS at the Spallation Neutron Source at Oak Ridge National Laboratory 52,53 . Data were collected in 60 Hz mode with a Q-range between~0.015 and 0.35 Å −1 . The 2D scattering data were reduced using the MantidPlot software according to standard procedures 54 . The SANS signals from CeO 2 -Y 2 O 3 nanobrush samples were subtracted the background scattering from a YSZ bare substrate.

Data availability
Data supporting the findings of this study are available from the corresponding author upon reasonable request.

Code availability
The VASP code, which is available for purchase, was used for the calculations in this paper. The configuration files used in these calculations are all available upon request from the authors.