WorksPhysics and Astronomy
Site-Specific Orbital Symmetry Breaking in Epitaxial Graphene on Silicon Carbide
Abstract
The epitaxial graphene grown on SiC interface has been the subject of tremendous research efforts to control its electronic transport properties in the quest for graphene electronic applications. The electronic structure of the epitaxial graphene buffer layer on SiC(0001) is investigated at the atomic scale using monochromated electron energy loss spectroscopy (EELS) in an aberration-corrected scanning transmission electron microscope (STEM). The C -edge reveals spectral fine structures distinctly different from free-standing graphene layer, providing direct experimental evidence of a modified bonding environment. Ab initio calculations providing a site-resolved probe of orbital character to elucidate the origin of these spectral features: covalent bonding between one-quarter of the buffer-layer carbon atoms and the substrate silicon atoms induces partial hybridization, quenching the out-of-plane states and disrupting -conjugation across the interfacial layer. Substrate-imposed symmetry breaking further lifts the in-plane orbital degeneracy. This orbital-engineering framework, rooted in -state quenching and site-specific symmetry breaking, establishes a quantitative basis for tailoring graphene-substrate interfaces toward next-generation low-energy transistor applications.
Introduction
As silicon-based device technology approaches its fundamental physical scaling limits, epitaxial graphene on SiC has emerged as a high-performance candidate for next-generation nanoelectronics. [1] Although pristine graphene is a zero-bandgap semimetal, the epitaxial layer in direct contact with the (0001)-SiC substrate — commonly designated the buffer layer (BL)[2] — exhibits semiconducting character with a reported bandgap of approximately 0.6 eV and hole mobilities surpassing those of germanium, properties that render it particularly suited to logic transistor applications. [1] The synthesis of high-quality graphene by thermal decomposition of hexagonal SiC has further consolidated this system as a viable nanoelectronic platform, [3, 4, 5] motivated by the prospects of nanoribbon-defined bandgap engineering[6, 7] and high electron mobility. [8] The BL is further characterized by a structural reconstruction,[9] intrinsic -type doping from Si dangling bonds, [10] and partial covalent bonding to the underlying Si-terminated SiC surface. [2, 11] Decoupling of the BL via intercalation of H, O, F, or Li,[12] [13, 14, 15, 16] or rapid thermal quenching, 17 yields quasi-free-standing graphene with tunable carrier type and concentration, enabling the engineering of novel two-dimensional heterostructures. 18 Nevertheless, the relationship between the electronic structure of the BL and its atomic configuration on the (0001) SiC surface remains incompletely understood, representing a fundamental obstacle to the industrial application of this material.
Atomic-scale core-level electron energy-loss spectroscopy (EELS) in the aberration-corrected scanning transmission electron microscope (STEM) is uniquely suited to addressing this problem.19 It offers electronic structure sensitivity at the spatial resolution of a single atomic plane or column,20 which provides relevant insight on, e.g., the oxidation state of metals and rare earths,21,22 the charge-carrier distribution in superconductors,23–25 the atomic-site distortion at surfaces and interfaces,26,27 etc., and enables mapping orbital directionality. 28–31
Previous investigations by Nicotra et al. 32 and {citet}`Palacio et al. {citep}`33 established atomic-scale differentiation of C-K EELS between the BL and adjacent graphene layers, reporting enhanced -bonding character and additional spectral weight between the and peaks in the BL. However, the instrumental energy resolution employed in these studies ( eV) likely precluded the resolution of finer spectral features intrinsic to the BL. Recent interpretation of the C-K electron energy-loss near-edge structure (ELNES) at the graphene–SiC interface have relied on linear combinations of bulk SiC and free-standing graphene, 34 an approach that overlooks the distinct electronic identity of the BL. Lacking theoretical simulations, this methodology cannot resolve how the interface geometry collapses the Dirac cone to form a bandgap.
In this Letter, epitaxially grown graphene multilayers on (0001) 6H-SiC are investigated using a monochromated electron beam at a meV energy resolution. Core-loss EELS measurements of the carbon K-edge, acquired from both the BL and the quasi-free-standing graphene, are combined with full-potential finite-difference method near-edge structure (FDMNES) simulations to elucidate the role of in-plane orbital symmetry in the observed bandgap. The C -edge spectra reveal a pronounced spectroscopic contrast between the BL and the overlying quasi-free-standing graphene. The distinct orbital character resolved in the projected density of states demonstrates the broken symmetry at carbon sites non-bonded to SiC substrate, providing a rigorous electronic-structure basis for the deviation of the BL from pristine graphene and, consequently, for the emergence of the substrate-mediated bandgap.
Results and Discussion
We investigate a specimen comprising five free-standing graphene layers and an epitaxial buffer layer (BL) grown on a single-crystalline (0001) 6H-SiC substrate, capped with the topological insulator (see Supporting Information), as illustrated by the STEM-high-angle annular dark-field (HAADF) image in Figure 1(a). The projected C -edge and corresponding HAADF image acquired at the graphene/SiC interface are displayed in Figure 1(a,Figure 1b). In the free-standing graphene layers, the feature is clearly resolved at eV, while the band spans the range 293.5 eV to 296 eV. The sharp excitonic resonance at eV, characteristic of -hybridised carbon,35 is well localised on each individual graphene layer.29 Since the layers above the BL exhibit nearly identical near-edge fine structures, the following discussion focuses on the electronic structure of the first graphene layer (see Figure 1(c)), hereafter referred to as the graphene layer (GL), for clarity. and conciseness. A comparison between the C- ELNES of the GL and the BL is shown in Figure 1(d) and Supporting Information, Figure S1; the BL spectrum differs strongly from that of the GL at energies below 297 eV, while the intensity of the continuum of states beyond this energy is comparable. The and peaks are less intense in the BL than in the GL. Furthermore, Figure 1(d) reveals additional spectral features, indicated by arrows, most notably an intensity contribution in the energy range 288.5 eV to 293.5 eV, situated between the and resonances, which is likewise apparent in the 2D spectral map presented in Figure 1(b). The peak intensity decreases, as does the excitonic peak at eV, which is attributed to a core-excitonic resonance. 33,36 The reduced intensity of the BL feature relative to the GL reflects changes in core-excited-state screening arising from covalent bonding at the BL–SiC interface, together with local structural and chemical disorder. Quantifying these experimentally observed excitonic effects (unlike disorder) requires going beyond the single-particle framework employed in our current FDMNES simulations.
Figure 1. Graphene/SiC specimen and layer-by-layer core-loss electron energy loss spectroscopy. (a) High-resolution HAADF-STEM cross-sectional view of the graphene/SiC specimen. (b) HAADF-STEM image of the graphene/SiC interface acquired as part of the spectrum image. C K-edge ELNES projected along the layer-stacking direction, in which the ELNES of each pixel is summed over the corresponding horizontal line of the datacube. (c) Representative unit cell of the surface reconstruction illustrating the graphene layer (GL) and buffer layer (BL) configurations. Superposition of the relaxed structural model onto the cross-sectional HAADF-STEM image to illustrate the SiC–buffer layer interfacial distance. (d) C K-edge energy-loss near-edge structures of the individual layers: the buffer layer and the first graphene layer.
Further inspection of the experimental spectra reveals a subtle intensity increase just above the Fermi level, at the onset of the peak. Several of these spectral features are consistent with those present in previously reported spectra, 32–34 though a comprehensive theoretical interpretation has thus far been lacking. To elucidate the origin of each spectral signature specific to the BL, FDMNES simulations of the C excitation, governed by the dipole-selection rule (), were performed using the finite difference method (FDM), which employs a free-shape potential without the spherical approximation (see Supporting Information, Methods), thereby maximizing accuracy in the near-edge region. 37,38 The simulations adopted the large realistic surface reconstruction, a structural model that has recently demonstrated improved accuracy in describing the electronic, 39 vibrational, and thermal properties of the BL. 40 As a representative case, side views of the GL and BL atomic structures, together with the superposition of the relaxed structural model onto the cross-sectional HAADF-STEM image, are presented in Figure 1(c), with full details on the relaxed structural models provided in the Supporting Information, Methods.
Prior to modeling the structurally complex BL, simulations were first performed on GL as a benchmark to validate the observed spectral features. In the experimental geometry employed here (see Supporting Information, Methods, Figure S2, and Ref. 29), the scattering vector (in units where ) is a superposition of in-C-plane () and out-of-C-plane () components, with normal to the graphene basal plane, such that the measured C K-edge spectrum (Figure 2(a) and Figure 2(e)) reflects a weighted integration over both orientational contributions. The simulations explicitly decompose this anisotropic response: the channel, involving transitions into unoccupied states, and the channel, dominated by transitions into the manifold with additional contributions, are computed independently and shown in Figure 2(b). A linear combination of the two geometric components , following the formalism of Hébert et al. 41 (see Supporting Information, Methods), reproduces the experimental spectrum. This assignment is further corroborated by the partial density of states (DOS) projected onto the carbon orbitals shown alongside the spectra. The sharp onset resonance is unambiguously attributed to pure states, corresponding to the resonance characteristic of graphene, while the broader weight at higher energy losses primarily originates from , , and derived states. Notably, as shown in Figure 2(c), the degeneracy of the and orbitals is largely preserved. These simulations are fully consistent with the canonical hybridization of the graphene lattice, in agreement with previous results, 35,36,42,43 providing a clear spectroscopic fingerprint for experimentally observed GL spectra.
Figure 2. C K-edge ELNES and Simulations of GL and BL on SiC. (a) Experimental C K-edge ELNES spectrum of the graphene layer (GL). (b) Calculated in-plane () and out-of-plane () spectral components (shaded), together with their orientationally weighted sum. (c) Calculated partial density of states (DOS) of GL, with the degenerate and contributions shaded for clarity and shown alongside the calculated spectrum. Dashed line denotes the Fermi level (). An energy threshold () was applied to select transitions into unoccupied states above (see Computational Details). (d) Schematic of the GL/SiC interface distinguishing non-bonded () and bonded () carbon atoms (black spheres, carbon; blue spheres, silicon). (e) Experimental C K-edge ELNES spectrum of the buffer layer graphene (BL). Calculated C K-edge ELNES and the decomposed and components for the (f) and (g) sites of BL.
Figure S2. Schematic of the scattering geometry. The incident probe (wavevector ) converges onto the sample with convergence semi-angle ; the inelastically scattered electrons are collected within a semi-angle about the optic axis. For a representative scattered electron at scattering angle , the scattering vector is decomposed into a component along the beam direction, (set by the characteristic angle of the C -edge energy loss), and a component transverse to it, .
Having established a spectroscopic baseline for the GL, the simulation was extended to the more structurally complex BL. In the BL, approximately one quarter of the carbon atoms form covalent bonds with the underlying Si atoms of the SiC substrate (Figure 2(d)). This partial covalent bonding breaks the lateral homogeneity of the BL, giving rise to two electronically inequivalent carbon sites: those covalently bonded to Si (hereafter C) and those remaining unbonded (hereafter C), where C sites retain a local chemical environment analogous to that of freestanding graphene. To capture this structural heterogeneity, ELNES calculations were performed independently for the carbon sites within the BL. Although the reconstruction possesses no exact symmetry, considering only the immediate local environment of each atom reveals an approximate threefold symmetry, which was exploited to reduce the 338 inequivalent carbon sites to 70 representative sites (See Supporting Information, Methods, and Figure S3) requiring explicit calculation. Each of the 70 calculated site spectra was itself constructed as a linear combination of the two geometric components, and , following the same procedure applied to the GL reference. The resulting site-resolved spectra were then weighted and summed to obtain the total simulated response, explicitly accounting for the core-level shifts and the anisotropic momentum-transfer contributions from the and channels at each of the corresponding and sites (Supporting Information, Figure S4). The site-resolved simulation of the sites (Figure 2(f)) reveals that the resonance is predominantly governed by the out-of-plane channel, consistent with the purely -derived character of the states in an -hybridized carbon network. The resonance, by contrast, shows contributions from both the and channels, reflecting the mixed in-plane and out-of-plane character of the unoccupied -derived states. This channel decomposition is in good agreement with the anisotropic response established for the GL and confirms that the site retains, to a first approximation, the spectroscopic character of a GL-like carbon environment.
Figure S3. Atomic subset used for ELNES calculations within the supercell. (a) The shaded red region (solid triangle) denotes the irreducible wedge of BL atoms, as defined in Ref. 6. The total spectrum is evaluated as the per-atom symmetry-weighted average over these 70 sites (), explicitly accounting for the unit-cell site multiplicity: for general positions in the interior, for sites along mirror boundary lines, and for the site located on the 3-fold rotation axis.
Figure S4. ELNES calculations for the subset of inequivalent (a) and (b) atoms as shown in Figure S3.
Nevertheless, a comparison between the spectrum and the GL reference (see Figure 2(b) and Figure 2(f)) reveals a systematic enhancement of the spectral intensity of BL in the intermediate energy range of 288.5–293.5 eV, situated between the and resonances. This pre- intensity enhancement is less dominant in the GL simulation and therefore constitutes a spectroscopic fingerprint of the modified electronic structure induced by SiC substrate hybridization. Relative to the GL (Figure 2b), the calculated resonance of the sites in the BL (Figure 2f) is considerably broader and weaker. This arises mainly from the structural disorder of the BL, which modifies both the lineshape and the chemical shift of the resonance from site to site. Although the calculation overestimates this effect, the experimental spectra also show a lower intensity in the BL than in the GL, together with a weak additional intensity on its low-energy side (Figure 1d and Supporting Information, Figure S1). Inspection of the simulated spectrum further corroborates this interpretation: the spectral weight of the site is significantly dominated by resonances, with a marked suppression of the feature, providing direct spectroscopic evidence for the conversion from to hybridization driven by covalent bond formation with the underlying Si atoms. Importantly, the channel of the and simulations also exhibits a non-negligible contribution within the 288.5–293.5 eV energy window (Figure 2(g)), demonstrating that the intermediate-energy intensity receives contributions from both the and momentum channels. This observation necessitates a rigorous comparison with the projected onto the relevant orbital symmetries, in order to unambiguously identify the electronic states responsible for the spectral intensity enhancement between the and resonances and to disentangle the competing contributions of substrate-induced hybridization from intrinsic GL features.
Figure S1. Additional representative STEM-EELS data supplementing Figure 1 of the main text. (a) High-resolution HAADF-STEM cross-sectional image of the graphene/SiC specimen. (b) Carbon K-edge ELNES spectra extracted from individual layers, namely the buffer layer (BL) and the first graphene layer (GL).
To identify the origin of the spectral features observed at the C -edge, the simulated BL spectrum was obtained by averaging the calculated site-resolved spectra over all carbon atoms in the BL, grouped into their inequivalent and sites, constituting approximately 25% and 75% of the carbon atoms (see Supporting Information, Methods), consistent with the known structural periodicity of the reconstruction. As shown in Figure 3a,Figure 3b, the simulated spectrum reproduces the experimental C -edge, capturing both the pronounced suppression of the resonance and the redistribution of spectral weight into the region relative to the GL reference. The simulation additionally exhibits a minor high-energy shoulder on the peak of the GL, arising purely from the component, that is not clearly resolved in the experimental spectrum. In the BL, the – separation is reduced by approximately 0.5 eV with respect to the GL (Figure 3b), so that the resonance is not only broadened but its maximum is also shifted to lower energy. Since the strain is largely relieved in the model, we attribute this reduction to the corrugation of the BL and to the coexistence of and sites, whose contributions are offset from each other by slightly less than 1 eV. The overall agreement with experiment is consistent with approximately 25% of the carbon atoms in the BL being covalently bonded to the Si-terminated SiC substrate, with the remaining 75% retaining a non-bonded, -like coordination, in line with the 25:75 : interfacial bonding ratio obtained independently from DFT structural relaxation. 40 Nonetheless, some discrepancies between the experimental and calculated spectra remain and can be ascribed to the implementation of the core hole, which underestimates the intensity, as well as to electron-beam channeling effects, which induce deviations from the geometric weights of and .
Figure 3. Orbital Reconstruction and Broken Symmetry. (a) Experimental C -edge ELNES spectra acquired from the graphene layer (GL) and buffer layer (BL). (b) Simulated C -edge ELNES spectra for the GL and BL. The BL spectrum is decomposed into contributions from the C and C sites, with the total spectral response reproduced by a linear combination of 25% C and 75% C character. (c) In-plane bond angles and bond lengths at the C and C sites extracted from the structural model. Partial densities of states projected onto the (d) C and (e) C sites reveal the lifting of orbital degeneracy arising from the broken structural symmetry at the C sites. Dashed line denotes the Fermi level ().
Having quantitatively established the site distribution within the BL, we proceeded to examine the pDOS for the GL, and the BL (i.e., and ) sites individually. In the GL, the three-fold rotational symmetry of the honeycomb lattice enforces strict degeneracy of the in-plane and orbitals, as reflected in their identical pDOS profiles. A direct comparison of the and contributions at the and sites of the BL reveals that this degeneracy is selectively and inequivalently lifted across the two populations upon covalent pinning to the SiC surface, fundamentally reorganizing the electronic structure relative to that of epitaxial GL. At sites, which represent the minority (25%) population, the 25:75 : ratio implies that each atom is coordinated by an average of three nearest neighbors. The out-of-plane covalent C–Si bond to the underlying SiC substrate induces partial rehybridization and locally suppresses aromatic delocalization, converting the three in-plane bonds into nearly equivalent single C–C bonds ( Å), as shown in Figure 3c,Figure 3d. Because all three neighbors are chemically equivalent, the local threefold symmetry is preserved and the degeneracy remains essentially intact. In contrast, each site is, on average, coordinated by two and one nearest neighbor (with some local deviations from this idealized count). The two bonds to neighbors retain a graphene-like aromatic character (1.40–1.42 Å), whereas the bond to the neighbor, whose aromaticity is suppressed by its bonding to Si, is elongated and inequivalent ( Å), breaking the local threefold rotational symmetry, as evident from the structural models in Figure 3c. This asymmetric bonding environment, together with an out-of-plane displacement markedly smaller than that at C sites, results in a pronounced splitting of the - and -derived states, as shown in Figure 3e.
This lifting of orbital degeneracy at the C sites reflects the anisotropic strain and broken local symmetry that arise from the proximity of covalently bonded C neighbors within the superstructure, even though the C atoms themselves do not form direct bonds with the substrate. This result is particularly significant, as it demonstrates that the electronic perturbation associated with substrate bonding extends beyond the immediately bonded atomic sites and propagates non-locally across the carbon network.
To further test this interpretation, and to expose the orbital symmetry lifting in a more transparent structural setting, simulations were also performed on a deliberately simplified model.44 Although this structure departs substantially from the realistic superstructure — most notably in its absence of significant corrugation at both the GL and BL (Figure 2d and see Supporting Information, Figure S5) — it nonetheless provides a valuable platform for isolating and verifying the key physical mechanisms identified in the full supercell calculations. A comparison of the simulated GL (Figure S6), C (Figure S7), C (Figure S7), and total BL (Figure S8) spectra against the corresponding experimental data, obtained using the simplified model structure, is provided in the Supporting Information. As anticipated, the simplified structure reproduces the principal experimental spectral features, although the splitting is less well resolved than in the calculations. This is likely a consequence of the strain (around 8%) imposed on the carbon layer, which is forced to match the SiC lattice parameter in the simplified structure, whereas in the large realistic cell it is almost completely relieved.
Figure S5. (a) A simplified () supercell of the ()R30 model structure is employed as a computationally efficient approximation of the experimentally observed ()R30 superstructure. (b) Top view of the buffer layer illustrating the distinct covalently bonded (C) and non-bonding carbon (C) sites with respect to the underlying SiC substrate.
Figure S6. (a) Experimental -edge ELNES spectrum acquired from the graphene layer (GL). (b) Calculated in-plane () and out-of-plane () spectral components (shaded) of the () supercell of the () model structure, along with their orientationally averaged superposition. (c) Schematic structural representation of the GL (carbon atoms, black spheres; silicon atoms, blue spheres).
Figure S7. (a) Experimental C -edge ELNES spectrum acquired from the buffer layer (BL). Calculated C -edge ELNES spectra of the supercell of model structure along with the decomposed and spectral components for the (b) covalently bonded () and (c) non-bonding () carbon sites of the BL. (d) Schematic structural representation of the BL/SiC interface, distinguishing covalently bonded () and non-bonding () carbon atoms (carbon atoms, black spheres; silicon atoms, blue spheres).
Figure S8. (a) Experimental C -edge ELNES spectra acquired from the graphene layer (GL) and buffer layer (BL). (b) Simulated C -edge ELNES spectra for the GL and BL of the supercell of the model structure. The BL spectrum is decomposed into contributions from the covalently bonded (C) and non-bonding (C) carbon sites, with the total spectral response reproduced by a linear combination of 25% C and 75% C character. (c) Schematic cross-sectional representation of the GL/BL/SiC heterostructure alongside top views of the BL and GL.
Crucially, however, the pDOS results obtained from the simplified model are in excellent agreement with those of the supercell: the degeneracy remains essentially intact at the C sites (Figure 3(d) and Supporting Information, Figure S9), while a pronounced splitting of the - and -derived states is recovered at the C sites (Figure 3(e) and Supporting Information, Figure S10), consistent with the symmetry reduction imposed by the surrounding covalently bonded network. The convergence of results across two structurally distinct models demonstrates that the observed orbital degeneracy lifting is a robust, physically transferable feature of the BL electronic structure. This degeneracy lifting has a direct spectroscopic consequence: splitting of the - and -derived states at the C sites shifts part of the in-plane spectral weight to lower energy, contributing to the low-energy shoulder observed experimentally in the BL. It is not, however, the sole origin of this shoulder: the C and C contributions are offset by slightly less than 1 eV (Figure 3b), and the reduced – separation arising from BL disorder likewise moves the spectral weight to lower energy. The shoulder thus reflects the combined effect of these three contributions.
Figure S9. (a) The simulated ELNES spectrum at the C site of the supercell of model structure, with the corresponding pDOS shown in panels (c) and (d). Panel (d) displays exclusively the in-plane and contributions at the C site. Dashed line denotes the Fermi level () (b) The bond lengths and angles are uniform across all C sites, reflecting the partial rehybridization induced by the out-of-plane covalent C–Si bond formed at the graphene–SiC interface. Despite the rehybridization the in-plane hexagonal symmetry of the lattice remains intact, indicating that the structural perturbation introduced by the substrate is spatially localized within the buffer layer. The DOS shown here were calculated in the presence of a core hole.
Figure S10. (a) The simulated ELNES spectrum at the C site of the supercell of the model structure, with the corresponding DOS shown in panels (c) and (d). Panel (d) displays exclusively the in-plane and contributions at the C site. Dashed line denotes the Fermi level (). Unlike the C sites, the C configuration retains a planar coordination, yet exhibits inequivalent C–C bond lengths of 1.62, 1.47, and 1.49 Å, indicative of a broken local threefold rotational symmetry. This structural asymmetry is reflected in the DOS, which reveals a lifting of the / orbital degeneracy. This behavior is attributed to the anisotropic bonding environment imposed by the commensurate reconstruction, whose full periodicity and inequivalent site distribution are more faithfully represented in the larger supercell geometry.
Furthermore, irrespective of the model, the spectral region at both C and C site types remains predominantly of , , and character. At the C sites, however, the orbitals contribute significantly to the region (Supporting Information, Figure S9 and Figure S11), as expected from the formation of the C–Si bond. The contribution is smaller than the and contributions at the maximum (294.5 eV), but it dominates between 290 and 293 eV, at the energy of the low-energy shoulder of the resonance. This shift of the spectral weight towards low-energy is a fingerprint of the covalent interaction with the underlying Si atoms. Its distribution differs markedly from that of a perfect hybridization, for which the , and contributions would be equivalent. This is consistent with a partial, rather than complete, rehybridization at this site. By contrast, the contribution is much weaker at the C sites (Figure S11), as expected for carbon atoms that are not bonded to the substrate and therefore retain a predominantly -like configuration. The site-resolved analysis thus reveals that the electronic structure reorganization within the BL cannot be attributed solely to local rehybridization at the covalently bonded C sites; rather, it emerges from a collective symmetry breaking that propagates across the extended carbon network via non-local modulation of in-plane orbital degeneracy — a fundamentally cooperative electronic response to substrate interaction.

Figure S11. (a) Simulated ELNES spectra at and sites of the superstructure, illustrating the spectral distinction arising from differential hybridization. (b) pDOS at sites, where covalent bond formation with the subsurface Si atom induces pronounced -like rehybridization, manifested as a suppression and concomitant downward energy shift of the contribution. (c) pDOS at sites, where the retention of robust intensity at higher energies confirms continued participation in the network. Dashed line denotes the Fermi level (). An energy threshold () was applied to select transitions into unoccupied states above (see Computational Details). The DOS shown here were calculated in the presence of a core hole.
The inclusion of orbital contributions reveals a further consequential departure from graphenic behavior that directly extends this symmetry-breaking picture. At sites, the formation of covalent bonds with the underlying Si atom drives significant -like hybridization, as evidenced by the suppression and concomitant downward energy shift of the intensity projected in the (Supporting Information, Figure S11), localizing the electron in a low-energy bonding state and withdrawing it from the itinerant conduction network. At sites, however, the robust intensity at higher energies (Supporting Information, Figures S9, S10, and S11) confirms the bonding, despite a symmetry-broken local environment induced by inequivalent C–C bond lengths. The consequences are twofold: the conducting channel is depleted at bonded sites, while the induced symmetry breaking propagates laterally to neighboring sites, lifting the degeneracy of their frontier orbitals — consistent with the splitting identified in the analysis above. The occurrence of these two mechanisms — -electron localization via rehybridization at sites and site-specific orbital degeneracy lifting at sites — constitutes a peculiar electronic structure characteristic of the graphene BL–SiC interface. Both mechanisms lift the degeneracy of the graphene sublattices, a necessary condition for opening a bandgap at the Dirac point.[1] [45] This sublattice symmetry breaking, driven by the selective covalent bonding of sites to the underlying SiC substrate, has been experimentally evidenced by the absence of dispersive -bands in angle-resolved photoemission spectroscopy measurements of the BL and the emergence of localized non-dispersive states below the Fermi energy,45,46 and distinguishes the electronic character of the BL fundamentally from the quasi-free-standing graphene character of the epitaxial overlayer.
Conclusion
In summary, monochromated core-loss EELS combined with ab initio calculations have been employed to resolve the electronic structure of the epitaxial graphene BL on SiC at atomic plane resolution. The C K-edge fine structure of the BL is markedly distinct from that of quasi-free-standing graphene, with the weakening of the resonance and redistribution of spectral weight into the region quantitatively reproduced by a 25% bonded and 75% non-bonded C site decomposition consistent with the surface reconstruction. Beyond structural confirmation, the site-resolved partial density of states analysis provides direct electronic insight into the BL’s orbital symmetry. Covalent bonding at C sites drives -like rehybridization that localizes electrons into low-energy bonding states, depleting the conduction network, while the induced symmetry breaking propagates to C sites through a selective lifting of orbital degeneracy. These findings, accessible only to high-spatial-resolution core-level spectroscopic techniques, underscore the unique capability of sub-Å core-loss EELS to disentangle orbital-level electronic reconstruction at buried graphene interfaces, providing atomic-scale insight into the origin of the buffer-layer bandgap and suggesting that substrate hybridization represents a viable route to tailoring the electronic structure of epitaxial graphene.
Acknowledgement
VKP and MB acknowledge funding from Agence Nationale de la Recherche (Grant ANR-22-CE29-0019). The EELS experiments were performed at SuperSTEM, the EPSRC national facility for advanced electron microscopy (through grants EP/V036432/1 and EP/W021080/1). The authors thank Orsay Physics for lamella preparation by focused ion beam milling. M.B. acknowledges the School of Chemical and Process Engineering at the University of Leeds for a Visiting Associate Professorship.
Supporting Information for: Site-Specific Orbital Symmetry Breaking in Epitaxial Graphene on Silicon Carbide
Vinod K. Paidi; Yves Joly; Alex Taekyung Lee; Guillaume Radtke; Vlado K. Lazarov; Demie Kepaptsoglou; Quentin M. Ramasse; Matthieu Bugnet
Methods
Experimental Details
The epitaxial graphene/SiC specimen was synthesized by thermal decomposition of SiC, at 1300 °C in ultrahigh vacuum (UHV); BiSe was subsequently deposited by molecular beam epitaxy at 275 °C to 325 °C.[1, 2] The cross-section STEM lamella was prepared by focused ion beam milling at ORSAY PHYSICS (Fuveau, France). The thickness of the specimen in the regions of investigation was evaluated to nm by Fourier-Log deconvolution of low-loss EELS spectra. The STEM-EELS experiments were carried out using a Nion HERMES MC100 microscope, equipped with a monochromator, a aberration corrector up to the 5th order, a Gatan Enfinium spectrometer, and operated at 60 kV. The convergence and collection semi-angles were mrad and mrad, respectively. Comprehensive details regarding the STEM-EELS experimental methodology are provided in Ref. [2] The STEM-EELS experimental datasets presented in this work were processed and analyzed using GATAN Digital Micrograph and the HyperSpy open-source Python library.³ Background subtraction was performed by fitting the pre-edge with a power-law function, prior to normalisation of the ELNES spectra by fitting the post-edge continuum with a linear function. The resulting spectra were subsequently normalised to the calculated edge-jump intensity, scaling the post-edge continuum to unity.
Computational Details
ELNES simulations were performed using the FDMNES code[4, 5] via the NRIXS_mono module within a density functional theory (DFT) framework. The magnitude of the scattering vector relevant to our simulations was estimated by decomposing into components parallel and perpendicular to the incident beam direction, and , with the characteristic angle at the energy loss , given by the relativistically corrected expression , and the scattering angle (see Figure S2). For the C -edge ( eV) at kV, mrad. Although our on-axis collection aperture accepts scattering angles up to a collection semi-angle mrad, considerably larger than , the inelastic signal is not uniformly distributed over this range. For parallel illumination (taken here to simplify the argument, though the reasoning remains the same for our finite convergence angle), the scattering intensity follows a Lorentzian angular distribution, , which is sharply peaked at small angles and decays as beyond ; consequently the signal collected within the aperture is dominated by scattering at angles of order rather than by the outer angular limit . Taking as representative of this weighting, the corresponding momentum transfer is , i.e. , which gives Å with Å at 60 kV (Figure S2). The electronic structure was solved using the finite difference method (FDM), which employs a full-potential approach by solving the Schrödinger equation on a real-space grid. The FDM was selected over the alternative multiple-scattering theory (MST) approach owing to its superior accuracy, despite the associated increase in computational cost. All FDMNES calculations reported here were performed within the dipole approximation; all other contributions (e.g., monopole, electric-quadrupole) were explicitly excluded.
DFT-optimized structures from Ref. [6] were used as starting geometries and were slightly modified to construct the present graphene/SiC slab models. We use DFT with the projector augmented wave (PAW) method[7] and the revised generalized gradient approximation (PBEsol),[8] as implemented in the VASP software package.[9] A plane wave basis set with a kinetic energy cutoff of 400 eV is used. For graphene on SiC, we modeled graphene on the SiC(0001) surface. The lateral lattice vectors were fixed using the optimized bulk SiC lattice constant, corresponding to a commensurate graphene/ SiC(0001) cell. The bottom surface of the SiC slab was C terminated and passivated with H atoms, whereas the top surface was Si terminated and in direct contact with graphene. SiC slabs containing two bulk layers were considered, resulting in simulation cells comprising 1216 atoms, including the graphene layer, SiC slab, and H passivation atoms. A vacuum region of 18.5 Å was used to avoid interactions between periodic slab images. We used a -centered k-point mesh of size . During structural relaxation, all H atoms were kept fixed, while the remaining atoms were relaxed until residual forces on all unconstrained atoms were below 0.01 eV/Å. To account for van der Waals interactions, the Grimme D3 correction was used.[10]
For the ELNES simulations, a 9 Å cluster radius was sufficient for the DFT-optimized superstructures, whereas a smaller 7 Å radius was used for the simplified supercell of the model.[11] The ELNES calculations were performed independently for the carbon sites within the buffer layer. It is worth noting that the reconstruction possesses no exact symmetry; however, considering only the immediate local environment of each atom reveals an approximate threefold symmetry, which was exploited to reduce the 338 inequivalent carbon sites to 70 representative sites (Figure S3). For the simplified supercell of the model, by contrast, the carbon -edge ELNES was obtained by summing the contributions from the two inequivalent absorbing carbon atoms within the unit cell. Although minor variations in the local coordination environments yield marginally distinct spectral responses across individual sites, these discrepancies were found to be negligible, thereby validating the site-averaged spectrum as an accurate representation of the system’s intrinsic electronic structure.
Each of the resulting site-specific spectra was itself constructed as a linear combination of the two geometric components, and , following the same procedure applied to the GL reference (as shown below). The relative weighting of the in-plane and out-of-plane FDMNES basis spectra was estimated using the formalism of H’ébert et al.[12] For the present cross-sectional geometry, the normal to the carbon plane is perpendicular to the electron beam, i.e., , for which the dependence on the azimuthal angle vanishes. Using , the integrated cross section then reduces to
where and are the out-of-plane () and in-plane () transition strengths, and , are geometric factors set by the convergence angle , collection semi-angle , and characteristic semi-angle . The coefficients and follow from the closed-form relativistic expressions of Ref.,[12]
with
The reduced angles and include the relativistic Lorentz factor of the incident electron,
where is the characteristic (relativistic) scattering angle associated with the energy loss . For mrad, mrad, and – eV (C -edge), this gives and , so that
where , are the approximate weights applied to the and FDMNES spectra.
Prior to obtaining the final spectrum, comprising approximately 75% and 25% character, the following procedure was used to account for core-level shifts: 1. The site-specific Kohn-Sham energy was estimated as , where and are the core-level binding energies. 2. The relative shift with respect to a reference site was computed as , where is the Kohn-Sham energy evaluated at an arbitrarily chosen reference site. 3. Each site-specific spectrum’s energy axis was shifted by its corresponding prior to weighting and summing the spectra to obtain the total simulated response. All spectra presented here include the core-hole in the final-state calculation, and an energy threshold filter of eV was applied. This filter suppresses simulated spectral intensity below , ensuring that the onset of the simulated spectrum coincides with the experimental threshold at which transitions into the lowest unoccupied states are physically accessible. [4, 5] The simulated spectra were subsequently broadened via Lorentzian convolution to account for finite core-hole and photoelectron lifetimes, with a full width at half maximum (FWHM) of 0.16 eV.
TOC Graphic
Figure 4 (PDF p. 42).
References
- [1]Liu, Y.; Li, Y. Y.; Rajput, S.; Gilks, D.; Lari, L.; Galindo, P. L.; Weinert, M.; Lazarov, V. K.; Li, L. Tuning Dirac States by Strain in the Topological Insulator Bi₂Se₃. Nat. Phys. 2014, 10, 294–299.
- [2]Bugnet, M.; Feder, M.; Lazarov, V. K.; Li, L.; Ramasse, Q. M.; Löffler, S.; Kepaptsoglou, D. M. Imaging the Spatial Distribution of Electronic States in Graphene Using Electron Energy-Loss Spectroscopy: Prospect of Orbital Mapping. Phys. Rev. Lett. 2022, 128, 116401.DOI
- [3]De La Peña, F. et al. hyperspy/hyperspy: HyperSpy 1.3. 2017.
- [4]Joly, Y. X-ray absorption near-edge structure calculations beyond the muffin-tin approximation. Phys. Rev. B 2001, 63, 125120.DOI
- [5]Bunău, O.; Joly, Y. Self-consistent aspects of x-ray absorption calculations. J. Phys. Condens. Matter 2009, 21, 345501.DOI
- [6]Radtke, G.; Lazzeri, M. Vibrational properties of epitaxial graphene buffer layer on silicon carbide. Phys. Rev. B 2025, 111, L220303.arxiv.org/abs/2506.16166
- [7]Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 1994, 50, 17953–17979.pubmed.ncbi.nlm.nih.gov/9976227
- [8]Perdew, J. P.; Ruzsinszky, A.; Csonka, G. I.; Vydrov, O. A.; Scuseria, G. E.; Constantin, L. A.; Zhou, X.; Burke, K. Restoring the Density-Gradient Expansion for Exchange in Solids and Surfaces. Phys. Rev. Lett. 2008, 100, 136406.pubmed.ncbi.nlm.nih.gov/18517979
- [9]Kresse, G.; Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 1999, 59, 1758–1775.DOI
- [10]Grimme, S.; Ehrlich, S.; Goerigk, L. Effect of the damping function in dispersion corrected density functional theory. Journal of Computational Chemistry 2011, 32, 1456–1465.DOI
- [11]Mattausch, A.; Pankratov, O. Ab initio study of graphene on SiC. Phys. Rev. Lett. 2007, 99, 076802.DOI
- [12]Hébert, C.; Schattschneider, P.; Franco, H.; Jouffrey, B. ELNES at magic angle condi-tions. Ultramicroscopy 2006, 106, 1139–1143, Proceedings of the International Work-shop on Enhanced Data Generated by Electrons.pubmed.ncbi.nlm.nih.gov/16934931