WorksPhysics and Astronomy
The Unreconstructed α-AlO(0001) Surface is Inhomogeneous and Rough
Abstract
Alumina (AlO) is a key material for thin-film growth and heterogeneous catalysis, where the atomic surface structure critically impacts performance. Using noncontact atomic force microscopy (nc-AFM) combined with density functional theory (DFT) calculations, we challenge the common assumption that the unreconstructed -AlO(0001) surface is atomically flat and uniformly Al-terminated. This widely accepted bulk termination satisfies polarity compensation requirements but results in highly undercoordinated surface Al cations at the surface. Despite substantial inward relaxation of these Al cations, we find that the (1 1) surface remains inherently metastable, relative to the thermodynamically stable surface reconstruction that forms at high temperatures above 1000 °C. Nc-AFM imaging of the unreconstructed surface reveals a rough and disordered morphology, with only nanometer-scale regions exhibiting the ordered Al-terminated (1 1) structure. Our results show that the unreconstructed AlO(0001) surface is intrinsically inhomogeneous, reconciling conflicting experimental observations and challenging the validity of commonly used atomistic models.
Introduction
The thermodynamically stable form of the -(0001) surface is the stoichiometric () reconstruction[1], which forms at high temperatures above 1000 °C. Under conditions relevant to technological applications, the surface typically remains unreconstructed. In the corundum lattice of , a non-polar (Tasker-type II[2]) (0001) surface can only be created by cutting between aluminum planes (Figure 1a), exposing threefold-coordinated Al cations at the surface (Figure 1b, Supplementary Data 1). This highly undercoordinated termination results in a high surface energy of the bulk-truncated surface. To increase their coordination, the surface Al atoms relax strongly inward into the underlying oxygen plane (Figure 1c, Supplementary Data 2). Our DFT calculations show an inward displacement of the surface Al atoms by Å, in agreement with earlier theoretical works[3, 4, 5] and consistent with studies of the unreconstructed surface by diffraction and scattering experiments[6, 7, 8, 9, 10]. Although the surface relaxation lowers the energy by more than 50% (blue dotted line in Figure 1d), it does not fully stabilize the surface. Thermodynamic stability is achieved only upon formation of the () reconstruction, which eliminates the undercoordinated surface Al sites via bonding to surface oxygen[1], leading to a further reduction in surface energy (red solid lines in Figure 1d). This reconstruction, however, only forms at high temperatures, when the increased atomic mobility enables an extensive reorganization across multiple atomic layers. The reconstruction preserves the overall stoichiometry, i.e., the composition of both the unreconstructed and the reconstructed surfaces remains . The bulk-terminated (1 × 1) surface remains energetically unfavorable across the entire range of oxygen chemical potentials (see Computational Methods for conversion to pressures and temperatures). The Al-terminated (1 × 1) surface (Figure 1c,Figure 1e) is therefore an intrinsically metastable, transient structure of the (0001) surface. Yet, it has long served as the standard model for epitaxial growth, adsorption, and surface chemistry studies[7, 11, 12, 13] [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26].
Figure 1. The Al-terminated – surface is metastable. (a) Corundum - bulk structure with the non-polar (0001) cutting plane (dashed line). (b) Bulk-truncated surface exposing undercoordinated Al cations (Supplementary Data 1). (c,e) DFT-relaxed Al-terminated surface: surface Al atoms relax inward into the O plane, adopting nearly threefold-planar coordination (Supplementary Data 2). The unit cell is marked by a black rhombus. (d) Ab initio phase diagram: although the relaxation lowers surface energy, the
Experimental results, however, suggest a more complex reality. A central enigma is the reactivity of the unreconstructed surface toward water. Theory consistently predicts facile dissociation with low activation barriers16,17,27–31, in line with the expectation that undercoordinated Al cations should be highly reactive. Thermal desorption and vibrational spectroscopy experiments, however, have yielded contradictory results. While some studies report dissociative adsorption19,32–34, others find that the unreconstructed surface is not easily hydroxylated, or even unreactive under ambient conditions20,21,28. Under ultrahigh vacuum (UHV) conditions, hydroxylation is typically limited to low coverages34–36, which was interpreted as dissociation at defects, steps and other surface heterogeneities22,32,35,36. Full coverage has only been achieved at elevated pressures34, unusually high exposures35,36, or with supersonic molecular beams28. Water desorption temperatures vary by several hundred Kelvin and the interpretation of desorption spectra remains controversial. Petrik et al.20 found only molecular adsorption by infrared spectroscopy despite observing thermal desorption peak shapes resembling those previously attributed to dissociatively adsorbed water35,36. Thissen et al. proposed that molecular and dissociative species coexist22. Here, we show that the disagreement between theory and experiment arises from the fact that the majority of the unreconstructed surface differs from the commonly assumed idealized structure model.
Under wet conditions, the surface transforms into a gibbsite-like termination that is O-terminated, fully hydroxylated, and ordered[7] [15, 37]. Differences in preparation conditions may lead to different surface terminations28, as reflected by the wide range of reported isoelectric points (pH 3.1–8) for nominally identical single crystals38. These ambiguities highlight the need for a well-defined model surface with known surface structure as a prerequisite for understanding and controlling alumina surface chemistry at the molecular level.
Spatially resolved imaging offers a route to visualize the surface structure and resolve the discrepancies between the predicted and observed reactivity of alumina surfaces. Yet atomic resolution on the clean, unreconstructed surface has remained elusive39,40. Recent advances in noncontact atomic force microscopy (nc-AFM), particularly the development of the qPlus sensor41, have enabled major progress in atomic-scale characterization of insulating materials, and judicious tip functionalization can now provide chemical sensitivity on the single-atom level42. Applied to the reconstruction, this methodology has already provided a detailed atomic model of this complex, lowest-energy structure[1].
In this work, we apply qPlus-based nc-AFM to the unreconstructed surface and show that, contrary to the prevailing view, it is intrinsically inhomogeneous and rough, with the ordered Al-terminated structure occurring only as a minority termination.
termination remains unfavorable relative to the stable reconstruction. Red lines indicate the coexisting O-rich and O-poor variants of the reconstructed surface[1]. Panels (a,c) modified from ref. 1.
Results
samples annealed in a tube furnace in air exhibited a well-defined terrace-step morphology in ambient AFM, albeit with noticeable roughness within individual terraces (Supplementary Figure S1). After transfer to UHV, the samples were further annealed up to 900 °C in low partial pressure up to mbar, below temperatures where the reconstruction forms, as confirmed by the absence of reconstruction spots in low-energy electron diffraction (LEED) (Supplementary Figure S2). Similar preparation procedures have been employed in the literature to obtain the unreconstructed surface, although the reported annealing temperatures vary [6, 9, 10] [43, 44, 45]. High-temperature annealing removed surface contamination, as verified by x-ray photoelectron spectroscopy (XPS) (Supplementary Fig. 3). No contaminants were detected, aside from a minute trace of fluorine on the sample shown in Figure 2. This fluorine signal was not reproduced on other samples exhibiting the same morphology and does not affect the results presented.
Figure 2. Morphology and atomic structure of AlO(0001). (a) Nc-AFM topography of the unreconstructed surface obtained by annealing at °C, recorded with a Cu-terminated tip in constant-frequency-shift mode ( Hz). Bright areas correspond to topographic protrusions and dark areas to depressions; the overall height range spans approximately three atomic steps (interlayer distance 216 pm). A representative height profile (along the orange line) is shown in the inset. (b) Constant-height image acquired with a Cu-terminated tip showing nanometer-sized islands surrounded by rough and disordered regions. Dark corresponds to attractive tip-surface interaction and indicates topographically higher islands, brighter areas represent lower-lying regions. Faint protrusions in the lower-lying areas (blue dotted lines) indicate the absence of periodic order. The frame at the right of the topography image (a) indicates the position of the constant-height image in (b). (c,d) High-resolution nc-AFM images of the islands acquired with (c) a Cu tip and (d) a CuOx tip, showing contrast inversion at surface Al sites. The measured size of the unit cell (orange) is consistent with the lattice constant of 475 pm. (e,f) Simulated AFM images reproduce the contrast of the Al-terminated surface for (f) a positively charged Cu tip and (g) a negatively charged CuOx tip. (g) nc-AFM topography of the reconstructed surface (annealed at 1300 °C) with a height variation below a single atomic step; the inset shows atomic-scale contrast within the reconstruction unit cell (outlined in red). Imaging and simulation parameters and details of sample preparation are provided in the Methods section and Supplementary Note 1.
Figure S1. Ambient AFM images of the surface before and after annealing in air. AFM images were acquired under ambient conditions using a cantilever-based AFM operated in tapping mode. (a) Before annealing, the surface exhibits a rough morphology with irregular, corrugated step edges. (b) After annealing in a tube furnace at in air for 10 hours, smooth step edges and flat terraces are observed. (c) The height profile along the blue line in (b), averaged over the line width of five pixels, shows a monoatomic step height of and an average terrace width of . (d) Detailed AFM image of the annealed sample, with contrast adjusted to highlight residual roughness within a single terrace. (e) Height profile taken along the red line in (d), shows considerable roughness within individual atomic terraces. Due to the finite sharpness of the tip and the short lateral scale of the roughness (cf. Figure 2a in the main text), the true peak-valley roughness is expected to strongly exceed the height differences in the line scan.
Figure S2. Direct transformation of the unreconstructed AlO(0001) surface into the reconstruction. LEED patterns (120 eV) acquired after annealing the unreconstructed surface at progressively higher temperatures from 900 °C (left) to 1300 °C (right), each for approximately 45 minutes in either mbar or UHV (900–1050 °C in mbar ; 1300 °C in UHV). Further experiments showed that the oxygen partial pressure during annealing does not affect the resulting LEED patterns. The onset of the reconstruction is indicated by the emergence of additional diffraction spots (red arrows). The (1 × 1) bulk unit cell is marked in blue; the two reconstruction domains, rotated by , are highlighted in orange and red.
In UHV, the morphology was examined with nc-AFM in constant-frequency-shift mode, where the tip follows the surface at a distance dominated by long-range forces. The unreconstructed surface appeared rough, with nanoscale height variations spanning several monoatomic steps (Figure 2a) and no periodic order. In contrast, the reconstructed surface was markedly smoother, with height variations well below a single atomic step (Figure 2g), arising mainly from corrugation within the reconstruction unit cell (Figure 2g, inset) and clearly apparent periodic order.
In regions with low corrugation, atomic resolution was achieved in constant-height mode (Figure 2b–Figure 2d), at a closer imaging distance. The images confirmed the rough and laterally inhomogeneous morphology of the unreconstructed surface, with nanometer-sized islands of ordered hexagonal symmetry consistent with a unit cell (Figure 2b). The chemical identity of these locally Al-terminated regions was established using functionalized tips: with a Cu tip, Al cations appeared as bright (repulsive) features (Figure 2c), whereas with a negatively charged -functionalized tip42 the same sites appeared dark (attractive) (Figure 2d). The termination of the tip was verified by a “fingerprint” image shown in Supplementary Fig. 4. The contrast inversion between the differently terminated tips, closely reproduced by AFM simulations (Figure 2e,Figure 2f), arises from electrostatic interaction between the tip and the surface Al atoms. The same qualitative contrast was consistently observed over a range of tip–sample distances (Supplementary Fig. 5) and is well described by the calculated electrostatic potential near the surface (Supplementary Fig. 6). Approaching the tip closer to the surface led to increased interaction and imaging instabilities before entering the Pauli repulsion regime. The images were thus acquired at tip–sample distances dominated by the electrostatic interaction to ensure stable imaging conditions. In both experiment and simulation, the contrast is dominated by the undercoordinated surface Al cations, whereas the close-packed oxygen sublattice is not resolved. Outside the ordered islands, the surface appeared disordered; faint protrusions (highlighted in Figure 2b) indicate the absence of a periodic structure just below the Al-terminated islands.
XPS spectra acquired at different emission angles (Supplementary Fig. 7) enable a qualitative comparison of surface and bulk composition. The Al 2p intensity measured at 70° emission (enhanced surface sensitivity) shows a markedly reduced aluminum-to-oxygen ration compared to the spectrum collected in normal emission, indicating a lower surface concentration of aluminum on the unreconstructed surface. This observation is incompatible with a uniformly Al-terminated surface and supports our nc-AFM finding that the Al-terminated structure occurs only as a minority phase, contributing minimally to the overall XPS signal. Comparison with the reconstructed surface (Supplementary Fig. 8) shows a nearly unchanged Al:O ratio, consistent with the previous finding that wide band-gap insulators are difficult to reduce[1]. The O 1s peak of the reconstructed surface is considerably broader than that of the unreconstructed surface, reflecting the diverse coordination environments present within the large unit cell.
Even after stepwise annealing in small temperature increments, the unreconstructed surface did not evolve into a flat, laterally uniform morphology. The ordered islands did not grow laterally with increasing temperature. Instead, their size increased only at the onset of surface reconstruction when the islands adopted structural motifs characteristic of the reconstruction (Supplementary Fig. 9). Above °C, the surface transformed into an atomically flat and well-ordered reconstruction (Figure 2b). No intermediate reconstructions were observed on clean samples, whose surface cleanliness was verified by XPS. LEED patterns recorded after successive annealing steps (Supplementary Figure S2) confirm this direct transition, in contrast to earlier reports of intermediate reconstructions46.
Once formed, the reconstructed surface remained stable and did not revert to the structure, even after annealing in an background up to mbar (Supplementary Figure S10). This observation contrasts with earlier reports suggesting reversibility under oxidizing conditions47, but agrees with recent results by Smink et al. 48. The ab initio phase diagram (Figure 1d) further supports this irreversibility, showing that the reconstructed phase is thermodynamically favored across the entire range of accessible oxygen chemical potentials. The reconstruction also exhibited remarkable environmental stability, remaining unchanged after controlled exposure to ultrapure liquid water (Supplementary Figure S10).
Figure S10. Stability of the reconstructed surface under oxidizing conditions and water exposure. (a) LEED patterns (120 eV) acquired before and after annealing in mbar , and (b) before and after exposure to ultrapure liquid water[5, 6], show no change in the reconstruction pattern, demonstrating the irreversibility and environmental stability of the surface. Improved focusing of the electron optics yielded enhanced contrast in (b).
Discussion
These results demonstrate that the Al-terminated structure is not a realistic description of the unreconstructed surface. Instead, the surface is rough and inhomogeneous, with only nanometer-scale islands of ordered termination. Reconciling this result with available structural measurements requires a closer look at the original data. Quantitative LEED-IV measurements on insulating samples are intrinsically challenging, because charging restricts the electron-energy range, and inaccurate electron energies due to charging compromise the -factor, the accepted measure of agreement between theory and experiment. Refs. [8, 9] reported satisfactory -factors only when introducing large anharmonicity of the top Al atoms, an effect that, as noted in ref. [9], could equally indicate static disorder. Grazing-incidence x-ray diffraction measurements often cited as evidence for Al termination[10] tested several structure models that yielded similar values. The same method also resulted in an early model of the surface49, which has since been substantially revised based on recent nc-AFM measurements[1]. LEED patterns obtained in the present work (Supplementary Figure S2) are comparable in quality to those previously reported35,36,46,47, suggesting that the observed periodicity in electron diffraction primarily reflects the underlying bulk lattice, with a thin, disordered surface layer covering most regions of the sample. In comparison with the reconstruction, the LEED pattern of the unreconstructed surface exhibits weaker spot intensities and enhanced diffuse background (Supplementary Fig. 11). Consistent with this observation, RHEED data reported in the literature show broader diffraction spots and stronger diffuse background intensity in the patterns of the unreconstructed surface compared to those of the reconstruction48,50,51.
The rough and irregular surface observed with nc-AFM helps reconcile prior conflicting reports of water adsorption on . The images reveal only nanometer-scale islands with the Al-terminated structure, while the surrounding regions appear disordered and remain inaccessible to atomic-resolution imaging. Considering the inertness of amorphous alumina layers formed upon oxidation of metallic aluminum, these disordered regions likely contain configurations in which surface atoms adopt higher coordination and are therefore unreactive52. Such regions would only allow molecular water adsorption, whereas the small degree of dissociative adsorption, previously attributed to surface defects, can be rationalized as occurring at the Al-terminated patches, which cover only a small fraction of the surface.
The lateral inhomogeneity of the unreconstructed surface has important implications for epitaxial growth on sapphire. The nanometer-scale domains separated by disordered regions provide a structurally non-uniform template and may influence film quality, particularly during the early stages of growth. For two-dimensional materials and ultrathin films, where the interface structure plays a critical role, such irregularities may strongly affect nucleation and growth. In contrast, the thermodynamically stable reconstruction yields a flat and well-ordered morphology and has been reported to promote adhesion during metal growth43 and recently shown to substantially improve the quality of grown on sapphire53.
At this point, it remains speculative why the unreconstructed surface is rough and whether this is an intrinsic property. Even for a simplified model system (see Computational Methods, Supplementary Fig. 12 and Supplementary Data 3), the energetic cost of forming steps on unreconstructed is remarkably low, which may facilitate roughening and contribute to the observed morphology. The stepped configurations allow locally higher coordination of surface Al atoms compared to a flat, Al-terminated structure, thereby reducing the energetic penalty associated with undercoordinated Al sites. The observed roughness of the unreconstructed surface may also originate from the surface termination present before UHV preparation. Samples exposed to ambient conditions can develop a gibbsite-like termination (a fully hydroxylated O-termination)54. An Al-terminated surface and a hydroxylated, O-terminated surface should be interconvertible by de/hydration, as shown in the molecular dynamics study of Hass et al.15. However, this transformation is hindered by substantial kinetic barriers and does not occur readily, as demonstrated by Yue28 and others[55] [20, 21, 29, 34, 35, 36]. Converting a hydroxylated surface into an Al-terminated structure requires substantial mass transport, inherently leading to roughness.
In summary, these results show that the simple, Al-terminated surface, widely used in theoretical models of unreconstructed , cannot be experimentally obtained. Even under pristine UHV preparation conditions, the unreconstructed surface is rough and inhomogeneous, and the expected Al-terminated surface occurs only as a minority phase on a surface that nonetheless exhibits a periodicity in electron diffraction. Recognizing this intrinsic complexity calls for a re-evaluation of how alumina is used as a substrate for thin-film growth and how its surface chemistry is understood.
Methods
Sample preparation
Polished - single crystals from Crystec GmbH were cleaned by several sonication cycles in a heated ( °C) pH-neutral detergent solution (3% Extran MA 02, Merck) followed by thorough rinsing with ultrapure (Milli-Q, Millipore, 18.2 MΩ·cm, < 3 ppb total organic carbon) until no polishing residues were observed by ambient AFM (Agilent 5500). The cleaned crystals were then annealed in air at 1100 °C for 10 hours in a tube furnace, resulting in wide atomic terraces (of approximately 500 nm) separated by straight, monoatomic step edges ≈ 220 pm high (Supplementary Figure S1). Samples were mounted on tantalum Omicron-type sample plates using spot-welded Ta wires. To minimize contamination, the sample plates and wires, as well as the crystals were each separately boiled in ultrapure water, followed by an additional boiling after assembly. The samples were then introduced into an ultrahigh vacuum (UHV) system and annealed at 820 °C to 900 °C in mbar for 1.5 hours in the preparation chamber (base pressure < mbar) using a tungsten filament heater, calibrated by measuring the temperature of a blank sample plate with a thermocouple vs. filament power. This annealing step leads to desorption of impurities. Sample cleanliness was checked by XPS, see Supplementary Fig. 3. To prevent adsorption of residual gases, the annealed samples were transferred to the analysis chamber (base pressure < mbar) at elevated temperature, prior to cooling to room temperature.
Noncontact atomic force microscopy
The nc-AFM measurements were performed at 4.7 K using an Omicron qPlus LT STM/AFM equipped with a cryogenic differential amplifier[56]. A qPlus sensor41 ( kHz, N/m, –16,000) with an electrochemically etched tungsten tip was used for imaging. The AFM tips were conditioned in UHV by field emission, Ar self-sputtering[57], and voltage pulsing on clean Au(100) and Cu(110) surfaces. Oxygen-terminated tips were formed by controlled indentation and voltage pulsing on a Cu(110) surface partially covered with an oxygen-induced added-row reconstruction, prepared by brief (15 s) exposure to ( mbar) at 250 °C. A stable CuO tip termination was confirmed by evaluating contrast variations over the Cu–O rows, as described in ref. 42; see Supplementary Fig. 4. Unless otherwise noted, imaging was conducted in constant-height mode using sample bias voltages between 0.0 V and −0.2 V to minimize the local contact potential difference (LCPD). The images were filtered in the Fourier domain to remove electrical and mechanical noise.
X-ray photoelectron spectroscopy
X-ray photoelectron spectroscopy (XPS) measurements were conducted in the preparation chamber using a SPECS XR 50 non-monochromatized Al K source and a SPECS Phoibos 100 hemispherical analyzer. Spectra were acquired at pass energies of 20 eV (individual spectra) and 60 eV (survey scans) in normal emission or at an emission angle of 70° relative to the surface normal to enhance surface sensitivity. To compensate for surface charging, a uniform energy offset of 7.8 eV was applied to align the O 1s binding energy to a nominal reference value of 531.0 eV. A linear Shirley background was subtracted from the O 1s and Al 2p spectra before analysis.
For visual comparison of peak shapes and relative intensities, each spectrum was normalized to the integrated area of the corresponding 1s peak. The Al 2p spectra were divided by the same normalization factor, enabling comparison of relative Al contributions between different emission angles and surface preparations. For quantitative determination of elemental ratios, atomic sensitivity factors[58] of 0.711 for 1s and 0.234 for Al 2p were applied to the background-subtracted peak areas.
Low-energy electron diffraction
Low-energy electron diffraction (LEED) patterns were recorded in the same chamber using a SPECS ErLEED 150. Screen inhomogeneities were corrected using flat-field (LEED image of a polycrystalline holder) and dark-frame (screen voltage off) images[59].
Ab initio calculations
Density functional theory (DFT) calculations were performed with the SCAN meta-GGA (generalized gradient approximation) exchange-correlation functional[60], as implemented in the Vienna Ab-initio Simulation Package (VASP)[61, 62]. The lattice parameters and of the - bulk cell deviated from experimental values[63] only by +0.1% and −0.2%, respectively, indicating that the bulk structure is reproduced well by this functional. The bulk unit cell was optimized using a cutoff energy of 800 eV, and a Monkhorst-Pack grid to integrate the Brillouin zone of the hexagonal cell. Slab models were created from the DFT-optimized bulk structure by cleaving along the non-polar (0001) planes (Figure 1a). Periodically repeated stoichiometric slabs with two equivalent faces were decoupled by a vacuum region of 15 Å. The surface calculations were performed using a lower cutoff energy of 500 eV and the same k-point mesh sampling used for the bulk cell. Structure optimization used the conjugate-gradient algorithm until the norms of all the forces on the atoms were smaller than 0.01 eV/Å. To compare surfaces with different stoichiometries, the surface free energies were evaluated using the formalism of ab initio atomistic thermodynamics[64]. Zero-point energy (ZPE) and vibrational entropy contributions were not explicitly calculated, as these terms largely cancel out when comparing related oxide terminations and therefore do not alter the qualitative stability ordering. The oxygen chemical potential in the ab initio phase diagram in Figure 1d is defined as:
where describes the temperature dependence at a reference pressure [64]. The energy of an isolated molecule in the gas phase was obtained in a spin-polarized calculation. This calculation converged to a magnetic moment of 2.0 , consistent with the triplet ground state of molecular oxygen. The SCAN functional was employed because it significantly improves predicted heats of formation for oxides[65] and provides a more accurate description of binding energies and magnetic states compared to previous GGA functionals[60], ensuring reliable reference energies.
Computational modeling of surface steps
Surface steps on -(0001) were modeled using a surface unit cell (in-plane dimensions 9.51 Å 28.53 Å), derived from the relaxed slab model (Figure 1c). To investigate steps of varying depth and width, between one and twelve units were systematically removed from the topmost layers, while maintaining the overall stoichiometry. These asymmetric slabs were terminated by a bulk-truncated Al layer at the bottom to avoid a polar surface. During geometry optimization, the bottom-most layers of the slab were fixed, while the upper layers were allowed to relax. Periodically repeated slabs were separated by a 15 Å vacuum gap. To efficiently explore the large configuration space, machine-learned force fields (MLFFs) based on the Gaussian Approximation Potential (GAP) approach (VASP version 6.5.1) were employed. The MLFFs were trained sequentially on several stoichiometric systems: bulk -, and slabs, slabs containing various step structures, a three-dimensional cluster derived from bulk -, and stepped slabs. The training data were generated from molecular dynamics (MD) simulations performed between 1000 K to 2300 K (near the melting point), resulting in a robust MLFF comprising over 2300 local atomic configurations. The trained MLFF was then used to identify low-energy step configurations via parallel tempering simulations[66], followed by geometry relaxations using simulated annealing. These simulations were carried out in prediction-only mode, without additional ab initio calculations. The parallel tempering simulations employed 24 replicas with logarithmically spaced temperatures from 150 K to 2400 K, ensuring high exchange probabilities. Swaps between adjacent replicas were attempted every 200 steps, yielding acceptance ratios above 0.3. Each run consisted of 1,000,000 ionic steps with a time step of 0.7 fs. The lowest-energy configurations obtained from these searches were subsequently fully relaxed using standard DFT calculations.
AFM simulations
Simulated AFM images based on the DFT-relaxed structure (Figure 1c) were generated using the Probe-Particle Model[67, 68, 69]. This approach accounts for the electrostatic potential above the surface (obtained from DFT calculations), Lennard-Jones potentials, and the elastic response of the tip. The lateral and vertical spring constants of the Cu tips ( N/m, N/m) and the CuOx tips ( N/m, N/m) were adopted from ref. 42. The effective tip charges were set to +0.05 e (Cu tip) and −0.05 e (CuOx tip). The simulated contrast was qualitatively unchanged with variations of the tip charge magnitude. The oscillation amplitude in each simulation matched the amplitude used in the corresponding experimental image. Because the exact tip–surface distance is not known experimentally, simulations were performed for a range of tip heights. The simulated images shown correspond to the tip–sample separations that yielded the best agreement with the experimental contrast.
Data availability
All data generated or analyzed during this study are included in this published article and its supplementary information files.
Acknowledgements
This work was supported by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant Agreement No. 883395) and funded in part by the Austrian Science Fund (FWF) [10.55776/F8100]. The computational results have been achieved using the Austrian Scientific Computing (ASC) infrastructure. For open access purposes, the authors have applied a CC BY public copyright license to any author accepted manuscript version arising from this submission. The authors acknowledge TU Wien Bibliothek for financial support through its Open Access Funding Program.
Author Contributions
J.B. conceived the research project. J.I.H-R. acquired and analyzed the experimental data. A.C. performed the computational modeling. J.I.H-R. and D.K. developed the sample preparation protocol. M.S. and F.M. supported computational modeling and data analysis. J.B. and M.S. supported experimental data acquisition and analysis. U.D. acquired funding. J.B. and U.D. supervised the project. J.I.H-R., A.C., U.D., and J.B. wrote the manuscript with contributions from all authors.
Competing interests
The authors declare no competing interests.
Additional Information
Correspondence and requests for materials should be addressed to Jan Balajka
The Unreconstructed -AlO(0001) Surface is Inhomogeneous and Rough
Johanna I. Hütner-Reisch; Andrea Conti; David Kugler; Florian Mittendorfer; Michael Schmid; Ulrike Diebold; Jan Balajka
These authors contributed equally to this work.
Supplementary Figures
Figure 3. Survey XPS spectrum of the unreconstructed Al₂O₃(0001) surface annealed in vacuum. The spectrum was recorded on a sample annealed at ≈900 °C (corresponding to the AFM image in Figure 2d) with a pass energy of 60 eV and an emission angle of 70° from the surface normal. Transitions originating from the Al₂O₃ sample are labeled in black. A small F 1s peak is present in this spectrum but was not observed on other samples with the same surface structure and is therefore not considered to affect the morphology. The O 1s (Mg Kα) ghost peak stems from x-rays emitted by the Mg anode adjacent to the Al anode in the dual-anode x-ray source used in this experiment. A spurious C 1s signal (indicated in gray) is caused by a measurement artifact and does not originate from the sample, as it remained unchanged with varying sample potential. The spectrum was corrected for surface charging by shifting the energy axis such that the O 1s binding energy is 531.0 eV.
Figure 4. Nc-AFM “fingerprint” image of the (2 × 1) oxygen superstructure on Cu(110) used to verify the tip termination. The image was recorded with the same tip used to image the unreconstructed surface shown in Figure 2d in the main text. The fingerprint confirms the tip termination[1] and supports the assignment of the attractive species on the unreconstructed surface as Al cations. The image was acquired in constant-height mode at 5 K with an oscillation amplitude of 500 pm and a sample bias of 0 V. The increase in contrast toward the bottom of the image is due to a decreased tip–sample distance.
Figure 5. Height-dependent nc-AFM images of the Al₂O₃(0001)-(1 × 1) surface. (a–e) Experimental constant-height nc-AFM images acquired with a CuOₓ tip at progressively smaller tip–sample separations; indicates the height difference relative to the first image (a). The contrast remains qualitatively unchanged over the accessible distance range. At closest approach (e), strong tip–surface interactions lead to imaging instabilities, visible as horizontal streaks. Therefore, the Pauli repulsion regime was not reached experimentally, and the images shown in the manuscript were acquired at tip–sample distances dominated by electrostatic forces. (a′–e′) Simulated AFM images calculated for corresponding increments of tip–sample separation (between 700 and 500 pm) reproduce the qualitative contrast.
Figure 6. DFT-calculated electrostatic potential and vertical electric field near the Al₂O₃(0001)-(1 × 1) surface. (a,b) Two-dimensional maps (12 Å × 12 Å) of (a) the local electrostatic potential and (b) the vertical electric field component , calculated above the surface of the relaxed structure model (Supplementary Data 2) using DFT. Each row corresponds to a vertical distance relative to the plane defined by the topmost surface Al atoms, where . The electrostatic potential is was referenced to the vacuum level () by subtracting the planar-averaged potential evaluated at the center of the vacuum region (≈ 6.87 V above the calculated Fermi level). The vertical electric field was computed as the negative gradient of the electrostatic potential along the surface normal (), with the positive -direction pointing from the slab into the vacuum. In the bottom two rows ( Å), the field in the vicinity of the Al cores cannot be accurately calculated by the PAW approach[2, 3]; these regions are indicated by gray disks. The bottom panels ( Å) include an overlay of the relaxed surface atomic structure (Al: blue; O: red), with the (1 × 1) unit cell outlined in black. At tip–surface distances relevant for nc-AFM imaging ( Å), the contrast in both and is dominated by the surface Al atoms, consistent with the experimentally observed contrast (Figure 2d,Figure 2e).
Figure 7. XPS spectra of the unreconstructed (0001) surface recorded at different emission angles. The spectra were acquired at a pass energy of 20 eV at 0° (normal emission, light blue line) and 70° relative to the surface normal (dark blue) to increase surface sensitivity. Sample charging was corrected by applying a uniform energy shift to all spectra to align the O 1s peak to a binding energy (BE) of 531.0 eV. After subtraction of a linear Shirley background, each spectrum was normalized to the integrated area of the corresponding O 1s peak. The Al 2p spectra were divided by the same normalization factor as the O 1s to allow comparison of relative Al contributions at different emission angles. (a) O 1s and (b) Al 2p spectra of the unreconstructed (0001) surface after annealing at °C in mbar . The absence of an OH-related shoulder at the high-BE side of the O 1s peaks shows that the surface is not hydroxylated. Quantitative analysis of the peak areas using atomic sensitivity factors yields [O]/[Al] ratios of 1.52 (0°) and 2.41 (70°). The reduced Al 2p intensity at 70° emission indicates a lower relative Al concentration near the surface, consistent with a predominantly oxygen-terminated outermost layer. Photoelectron diffraction effects may additionally contribute to the reduced Al contribution.
Figure 8. Comparison of grazing-emission XPS spectra of the unreconstructed and reconstructed surfaces. The spectra were acquired at a pass energy of 20 eV at an emission angle of 70° from the surface normal to enhance surface sensitivity. Sample charging was corrected by applying a uniform energy shift to all spectra to align the O 1s peak to 531.0 eV. After subtraction of a linear Shirley background, each spectrum was normalized to the integrated area of the corresponding O 1s peak. The Al 2p spectra were divided by the same normalization factor to allow comparison of relative Al contributions between the two surfaces. (a) O 1s and (b) Al 2p spectra of the unreconstructed surface (blue, annealed at °C in mbar ) and the reconstructed surface (red, annealed at °C in mbar ). Quantitative analysis using atomic sensitivity factors yields ratios of 2.41 for the unreconstructed surface and 2.19 for the reconstructed surface (both at 70° emission). The similar relative Al 2p intensities indicate that both surfaces are predominantly oxygen-terminated and exhibit comparable stoichiometry.
Figure 9. Island size increases only when the reconstruction starts to form. (a) Nc-AFM image of a representative island ( nm wide) on the unreconstructed AlO(0001) surface (see Figure 2 in the main text). The lateral size of the islands did not increase with annealing temperature. The image was acquired in constant height with a CuO-terminated tip; Al atoms appear dark (attractive). (b) After prolonged annealing at 820 °C in mbar O for 2.7 h, the islands grew (here nm wide), but no longer exhibited the structure. Instead, they displayed structural motifs characteristic of the reconstruction[4]: octahedrally coordinated surface Al atoms surrounded by four O atoms in a square geometry (white squares), and tetrahedrally coordinated Al atoms surrounded by three O atoms (white triangles). Island coarsening thus coincides with the onset of the reconstruction. The nc-AFM image was recorded with an oscillation amplitude of 500 pm and 1.5 V sample bias. The contrast is adjusted to highlight features on the island; the lower terrace (white) is not resolved. (c) Fully reconstructed surface obtained by annealing at 1300 °C in mbar O for 2.5 h, shows an atomically flat morphology and well-ordered structure. At this stage, terrace widths exceeded 100 nm. Images (b) and (c) were acquired in constant height with a Cu-terminated tip; surface O atoms appear dark (attractive).
Figure 11. Supplementary Fig. 11 | LEED images and diffuse background of unreconstructed vs. reconstructed . LEED images (acquired at room temperature, eV, inverted grayscale) of (a) the unreconstructed surface annealed at and (b) the reconstructed surface annealed at (same data as Supplementary Figure S2, but without flat-field correction). Some diffraction spots of the unreconstructed surface are broader than others, as expected for a surface with a high density of steps (broad spots occur under out-of-phase conditions7). Corresponding intensity profiles along the indicated lines are shown below; distances along the lines are given in reciprocal lattice units (r.l.u.) For the profile plots, the intensities were normalized such that the integrated intensity of all diffraction peaks (after background subtraction) is equal for both images. The unreconstructed surface exhibits broader and less intense diffraction peaks and an enhanced diffuse background, in contrast to the sharp and intense peaks with lower background observed for the reconstructed surface. These results are consistent with the rough and irregular morphology observed by nc-AFM (Figure 2 in the main text) and supports the interpretation that the observed pattern primarily reflects the underlying bulk lattice.
Figure 12. The energetic cost of forming steps on the surface is low. The lowest-energy structure with steps (Supplementary Data 3) identified in the parallel tempering simulations is higher in energy by 0.93 eV per simulation cell compared with the relaxed surface (Figure 1c in the main text). The simulation cell contains two steps (one up and one down, each 9.5 Å long). The calculated energy difference therefore corresponds to a step energy of ≈49 meV/Å. For comparison, forming a hypothetical, vertical facet of a monoatomic step height (≈2.2 Å), using the surface energy of the relaxed configuration (120 meV/Å, Figure 1d), would yield a step energy of ≈264 meV/Å. Thus, the calculated step energy of ≈49 meV/Å is remarkably low. As only a limited number of configurations were examined, structures with even lower step energies may exist. The atomic configuration helps rationalize the low energy: near the step and on the lower terrace, only oxygen atoms are exposed at the surface, while aluminum atoms are at least fourfold coordinated. The same tendency to avoid threefold-coordinated Al drives the formation of the thermodynamically stable reconstruction[4]. The low step energy indicates that the driving force for flattening a rough surface is weak. At the same time, it implies that atoms at the step occupy low-energy (strongly bound) configurations, which reduces step mobility. These considerations may explain why, in our experiments, the steps do not disappear at temperatures below those required for the formation of the reconstruction (cf. Supplementary Fig. 9).
Supplementary Notes
Supplementary Note 1 | Additional information for Figure 2.
The image in panel (a) was acquired with a oscillation amplitude and panels (b) and (c) using an oscillation amplitude of . Images in panels (a), (b) and (c) were acquired on a sample annealed at in for 90 minutes, and 0 V bias. The image in (g) was recorded with an oscillation amplitude of and 1 V bias. The inset of (g) was acquired on a sample annealed at 1300 °C in for 60 minutes using a -terminated tip at −0.1 V bias with oscillation amplitude. Image (d) was acquired on a sample annealed at in for 90 minutes, using a amplitude and −0.2 V sample bias.
Supplementary References
Schulze Lammers, B., Yesilpinar, D., Timmer, A., Hu, Z., Ji, W., Amirjalayer, S., Fuchs, H. & Mönig, H. Benchmarking atomically defined AFM tips for chemical-selective imaging. Nanoscale 13, 13617–13623 (2021).
Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 17953–17979 (1994).
Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758–1775 (1999).
Hütner, J. I., Conti, A., Kugler, D., Mittendorfer, F., Kresse, G., Schmid, M., Diebold, U. & Balajka, J. Stoichiometric reconstruction of the surface. Science 385, 1241–1244 (2024).
Balajka, J., Pavelec, J., Komora, M., Schmid, M. & Diebold, U. Apparatus for dosing liquid water in ultrahigh vacuum. Rev. Sci. Instrum. 89, 083906 (2018).
Balajka, J., Hines, M. A., Debenedetti, W. J. I., Komora, M., Pavelec, J., Schmid, M. & Diebold, U. High-affinity adsorption leads to molecularly ordered interfaces on in air and solution. Science 361, 786–789 (2018).
Henzler, M. Quantitative evaluation of random distributed steps at interfaces and surfaces. Surf. Sci. 73, 240–251 (1978).
References
- [1]Hütner, J. I., Conti, A., Kugler, D., Mittendorfer, F., Kresse, G., Schmid, M., Diebold, U. & Balajka, J. Stoichiometric reconstruction of the Al₂O₃(0001) surface. Science 385, 1241–1244 (2024).DOI
- [2]Tasker, P. W. The stability of ionic crystal surfaces. J. Phys. C: Solid State Phys. 12, 4977 (1979).DOI
- [3]Manassidis, I., De Vita, A. & Gillan, M. J. Structure of the (0001) surface of α-Al₂O₃ from first principles calculations. Surf. Sci. 285, L517–L521 (1993).
- [4]Godin, T. J. & LaFemina, J. P. Atomic and electronic structure of the corundum (α-alumina) (0001) surface. Phys. Rev. B 49, 7691–7696 (1994).pubmed.ncbi.nlm.nih.gov/10009515
- [5]Jarvis, E. A. A. & Carter, E. A. Metallic character of the Al₂O₃(0001)-(√31 × √31)R ± 9° surface reconstruction. J. Phys. Chem. B 105, 4045–4052 (2001).DOI
- [6]Ahn, J. & Rabalais, J. W. Composition and structure of the Al₂O₃{0001}-(1 × 1) surface. Surf. Sci. 388, 121–131 (1997).DOI
- [7]Eng, P. J., Trainor, T. P., Brown Jr., G. E., Waychunas, G. A., Newville, M., Sutton, S. R. & Rivers, M. L. Structure of the hydrated α-Al₂O₃(0001) surface. Science 288, 1029–1033 (2000).DOI
- [8]Walters, C. F., McCarty, K. F., Soares, E. A. & Van Hove, M. A. The surface structure of α-Al₂O₃ determined by low-energy electron diffraction: aluminium termination and evidence for anomalously large thermal vibrations. Surf. Sci. 464, L732–L738 (2000).
- [9]Soares, E. A., Van Hove, M. A., Walters, C. F. & McCarty, K. F. Structure of the Al₂O₃(0001) surface from low-energy electron diffraction: Al termination and evidence for anomalously large thermal vibrations. Phys. Rev. B 65, 195405 (2002).
- [10]Guénard, P., Renault, G., Barbier, A. & Gautier-Soyer, M. Determination of the α-Al₂O₃(0001) surface relaxation and termination by measurements of crystal truncation rods. Surf. Rev. Lett. 5, 321–324 (1998).
- [11]Fu, J.-H., Min, J., Chang, C.-K., Tseng, C.-C., Wang, Q., Sugisaki, H., Li, C., Chang, Y.-M., Alnami, I., Syong, W.-R., Lin, C., Fang, F., Zhao, L., Lo, T.-H., Lai, C.-S., Chiu, W.-S., Jian, Z.-S., Chang, W.-H., Lu, Y.-J., Shih, K., Li, L.-J., Wan, Y., Shi, Y. & Tung, V. Oriented lateral growth of two-dimensional materials on c-plane sapphire. Nat. Nanotechnol. 18, 1289–1294 (2023).DOI
- [12]Liu, L., Li, T., Ma, L., Li, W., Gao, S., Sun, W., Dong, R., Zou, X., Fan, D., Shao, L., Gu, C., Dai, N., Yu, Z., Chen, X., Tu, X., Nie, Y., Wang, P., Wang, J., Shi, Y. & Wang, X. Uniform nucleation and epitaxy of bilayer molybdenum disulfide on sapphire. Nature 605, 69–75 (2022).DOI
- [13]Zhu, H., Nayir, N., Choudhury, T. H., Bansal, A., Huet, B., Zhang, K., Purot zky, A. A., Bachu, S., York, K., Mc Knight, T. V., Trainor, N., Oberoi, A., Wang, K., Das, S., Makin, R. A., Durbin, S. M., Huang, S., Alem, N., Crespi, V. H., Van Duin, A. C. T. & Redwing, J. M. Step engineering for nucleation and domain orientation control in WSe₂ epitaxy on c-plane sapphire. Nat. Nanotechnol. 18, 1295–1302 (2023).DOI
- [55]Zhang, Y., Nayir, N., Shin, Y. K., Mao, Q., Jeong, G.-U., Chen, C., Redwing, J. M. & Van Duin, A. C. T. ReaxFF study of surface chemical reactions between α-Al₂O₃ substrates and H₂O/H₂ gas-phase molecules. J. Phys. Chem. C 128, 18767–18781 (2024).DOI
- [56]Huber, F. & Giessibl, F. J. Low noise current preamplifier for qPlus sensor deflection signal detection in atomic force microscopy at room and low temperatures. Rev. Sci. Instrum. 88, 073702 (2017).DOI
- [57]Setvín, M., Javorský, J., Turčínková, D., Matolínová, I., Sobotík, P., Kocán, P. & Ošt’ádal, I. Ultrasharp tungsten tips—characterization and nondestructive cleaning. Ultramicroscopy 113, 152–157 (2012).DOI
- [58]Moulder, J. F., Stickle, W. F., Sobol, P. E., Bomben, K. D. Handbook of X-Ray Photoelectron Spectroscopy: A Reference Book of Standard Spectra for Identification and Interpretation of XPS Data. (Perkin-Elmer Corporation, Eden Prairie, Minnesota, U.S.A., 1992).
- [59]Schmid, M., Kraushafer, F., Imre, A. M., Kißlinger, T., Hammer, L., Diebold, U. & Riva, M. ViPERLEED package II: Spot tracking, extraction, and processing of I(V) curves. Phys. Rev. Research 7, 013006 (2025).DOI
- [60]Furness, J. W., Kaplan, A. D., Ning, J., Perdew, J. P. & Sun, J. Accurate and numerically efficient r²SCAN meta-generalized gradient approximation. J. Phys. Chem. Lett. 11, 8208–8215 (2020).DOI
- [61]Kresse, G. & Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558–561 (1993).
- [62]Kresse, G. & Furthmüller, J. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 15–50 (1996).DOI
- [63]Ishizawa, N., Miyata, T., Minato, I., Marumo, F. & Iwai, S. A structural investigation of α-Al₂O₃ at 2170 K. Acta Cryst. B 36, 228–230 (1980).
- [64]Reuter, K. & Scheffler, M. Composition, structure, and stability of RuO₂(110) as a function of oxygen pressure. Phys. Rev. B 65, 035406 (2001).DOI
- [65]Kothakonda, M., Kaplan, A. D., Isaacs, E. B., Bartel, C. J., Furness, J. W., Ning, J., Wolverton, C., Perdew, J. P. & Sun, J. Testing the r²SCAN density functional for the thermodynamic stability of solids with and without a van der Waals correction. ACS Mater. Au 3, 102–111 (2023).DOI
- [66]Sugita, Y. & Okamoto, Y. Replica-exchange molecular dynamics method for protein folding. Chem. Phys. Lett. 314, 141–151 (1999).DOI
- [67]Oinonen, N., Yakutovich, A. V., Gallardo, A., Ondráček, M., Hapala, P. & Krejčí, O. Advancing scanning probe microscopy simulations: A decade of development in probe-particle models. Comput. Phys. Commun. 305, 109341 (2024).
- [68]Hapala, P., Kichin, G., Wagner, C., Tautz, F. S., Temirov, R. & Jelínek, P. Mechanism of high-resolution STM/AFM imaging with functionalized tips. Phys. Rev. B 90, 085421 (2014).DOI
- [69]Hapala, P., Temirov, R., Tautz, F. S. & Jelínek, P. Origin of high-resolution IETS-STM images of organic molecules with functionalized tips. Phys. Rev. Lett. 113, 226101 (2014).pubmed.ncbi.nlm.nih.gov/25494078