Since controlled positioning of 35 single Xe adatoms on a single crystal Ni(110) surface in the seminal work by Eigler and collaborators1, scanning tunneling microscopy (STM) has been considered not only as an imaging and spectroscopic technique but also as the unique instrument capable to fabricate designer structures at the atomic-scale2,3. Over the years, STM manipulation of surface adsorbates4–10 and surface atom vacancies11–13 has progressed from both technical and conceptual points of view (Fig. 1a). Particularly, design of specific systems facilitating reliable STM operation has shown fabrication of unprecedently complex structures through thousands of controlled manipulation events. For example, CO molecules adsorbed on Cu(111) at cryogenic temperatures were arranged in hierarchies performing basic logic operations5 or in artificial lattices expressing highly tunable electronic bands8,14,15. Even higher stability of surface vacancies in Cl layer on Cu(110) enabled atom-by-atom realization of a one kilobyte memory12, establishing state-of-the-art status for STM manipulations.
However, transfer of such accomplishments from metal substrates to more application relevant systems is challenging. It requires careful combination of STM manipulation protocols with material design, synthesis and processing. Hydrogenated Si(100)11,13 is one of the most successful examples, where Simmons and collaborators used STM-stimulated desorption of single H atoms to fabricate devices realizing single atom transistors16 or coupled qubits17. For these applications, the synthesis and processing steps are indeed critical, as STM lithographic patterns form only chemically active sites for phosphine molecule adsorption, from which phosphorous dopants are subsequently incorporated into the crystalline lattice, thermally activated and finally protected by a few layers of post-deposited Si.
Larger band-gap semiconductor, silicon carbide (SiC), constitutes another technologically important material with applications varying from high-power microelectronics18 to quantum information processing19,20. Controlled thermal decomposition of the top SiC layers creates also a possibility of growing epitaxial graphene-SiC heterostructures21–26. Due to their unique electronic properties and compatibility of the SiC(0001) substrate with general semiconductor processing strategies, these epitaxial graphene-based systems attract significant attention27–35. Particularly, improved synthesis methods of a high-quality, long-range ordered interface between the first epitaxial graphene layer (buffer layer; Cbuffer) and SiC have realized a semiconducting graphene, a long-time challenge36–38. In this case, Cbuffer consists of sp2 hybridized C atoms arranged in quasi-hexagonal (6×6)SiC moiré pattern. This graphene-like network is separated by sp3 hybridized C atoms covalently bonded to Si from the substrate. It was also shown that Cbuffer and the underlaying top SiC layer form incommensurate lattices37, slightly deviating from the (6\(\sqrt{3}\)×6\(\sqrt{3}\))R30°SiC reconstruction model39. The resulting local stress additionally modulates the interface structure by producing larger patches of decoupled graphene. Importantly, the breaking of covalent bonds between Cbuffer and SiC is associated with charge redistribution across the interface, as shown globally by rapid cooling synthesis25 or hydrogenation40,41.
Here, we use these properties of the epitaxial graphene to design a platform for STM manipulation of a bilayer graphene (BLG) heterostructure on the SiC(0001) substrate. We show that the inherent local stress from the incommensurate lattices forming the Cbuffer-SiC interface under BLG creates a fine balance for local covalent coupling of Cbuffer to the SiC substrate; which consequently could be reversibly controlled by application of an electric field with the use of STM setup. Similar electric-field-induced manipulation procedures for graphene heterostructures on hexagonal boron nitride (hBN) have provided highly tunable control over graphene local doping levels at the nanometer scale42–45. In contrast to this approach, which is based on redistribution of charged defects within the underlaying hBN, our work provides a far more precise STM manipulation landscape via synthesis of a ferroelectric two-dimensional interface46 directly underneath graphene layers.
Interface of graphene buffer layer and SiC
Figure 1b displays a cross-section of the epitaxial BLG interface in the form of charge a density map modeled by our density functional theory (DFT) calculations (see Methods for details). It consists of three layers of C atoms atop a SiC bilayer. The top two layers of purely sp2 hybridized C form the AB-stacked BLG. In contrast, the hybrid third layer (Cbuffer) possesses some sp3 hybridized atoms, which are covalently bonded to the undelaying Si as manifested by the shared charge densities (red ellipses). The Cbuffer and SiC lattices are rotated by 30° creating a complex moiré pattern of these covalent bonds, and thus the effective corrugation of the whole BLG heterostructure exhibits a faint, quasi-hexagonal (6×6)SiC periodicity (Fig. 1c). The proposed (6\(\sqrt{3}\)×6\(\sqrt{3}\))R30°SiC and (13×13)Cbuffer supercell, which is used in our calculations, still includes a small tensile strain (~ 0.4%)48 in comparison to the unit cell of freestanding graphene. As a result, part of the quasi-hexagonal moiré pattern in the relaxed DFT structure is more protruded (green color in Fig. 1c) due to breaking of additional covalent bonds between Cbuffer and SiC (Extended Data Fig. 1).
The above model corresponds well to our low-bias STM topography of the ultrahigh vacuum synthesized epitaxial BLG (for synthesis details see Methods and ref.49) presented in Fig. 1d (all STM experiments were performed at ~ 78 K). Positions of the protruded parts of the BLG interface, exemplified by the white arrows, are spanned by the (6×6)SiC lattice unit vectors. They may occupy any of the three alternating corners of the quasi-hexagonal pattern (Fig. 1e). Cross-section of the high-resolution STM in Fig. 1f shows that these protrusions extend only about 25 pm from the regular moiré corrugation. This small value supports the findings of our DFT and points to the local changes of covalent coupling within the buried interface between Cbuffer and SiC as a natural origin of the observed protrusions50. Moreover, due to difference in thermal expansion coefficients of graphene and SiC25, the covalent coupling between Cbuffer and SiC should depend on kinetic conditions during synthesis. In this work, extended annealing times at the SiC graphitization step (1h at ~ 1570 K) followed by progressive two-step cooling under ultrahigh vacuum resulted in protrusion surface densities of about 0.4 per one (6×6)SiC unit cell [~ 1.2 per (6\(\sqrt{3}\)×6\(\sqrt{3}\))R30°SiC], matching the DFT model predictions.
Evidence of electric-field-induced interface manipulation
Interestingly, stable configurations of the moiré pattern protrusions change upon STM imaging performed with larger bias voltages, as seen in Fig. 2a under the atomically resolved top BLG lattice. The observed dynamic behavior depends on polarity of the applied voltage. Particularly, large negative sample biases decrease the average number of protrusions, while large positive biases increase it (Fig. 2b). This manipulation process is reversible. Additionally, imaging at low sample biases and tunneling currents exhibits hysteresis effects depending on the previously used conditions (Extended Data Fig. 2). These results particularly exclude origination of these features from heteroatoms, i.e. Si, consistent with former studies showing a chemically homogenous interface51.
Figure 2c presents a quantitative analysis of the protrusion density as a function of STM sample bias voltage for a given area and tunneling current. The experiment consisted of two data sets of STM images obtained for sequentially increasing negative and positive sample bias voltages (Extended Data Fig. 3). Before collecting each set, the area was imaged with a large negative voltage, which resulted in low starting surface densities of protrusions [~ 0.2/(6×6)SiC]. For the negative sample biases, the number of protrusions initially increases up to the level close to native surface densities [~ 0.4/(6×6)SiC], and then progressively decreases to the initial, low value. In contrast, for the positive voltages we observe a monotonic, close to linear, increase in surface densities up to the level reaching about one protrusion per (6×6)SiC unit cell. For the positive voltages, the density also increases linearly with the logarithm of tunneling current (Fig. 2d). Such a weak tunneling current dependence, together with the above bias voltage polarity dependences, point to the STM-induced, local electric field42,52−54 as a source of the interface manipulation.
Downscaling of STM manipulation to single moiré sites
Small effective electric fields during soft STM scanning conditions, e.g. \(\left|U\right|<0.4\)V and \(I <100\) pA, do not modify metastable configurations of the interface protrusions (but high-density phases are stabilized by large positive voltages). This fact enables nanometer-scale STM patterning of BLG/SiC heterostructure from its bottom interface as exemplified in Extended Data Fig. 4. More importantly, it also allows scaling down of our manipulation protocols to reach lateral precision of a single moiré (6×6)SiC lattice site (Fig. 3a). In this case, we used the soft scanning conditions to stabilize the position of the STM tip above a certain corner of the moiré pattern. Then, to create a protrusion under constant-height STM operation conditions, we initialized a positive sample bias voltage ramp loop (Fig. 3b). An abrupt rise of the tunneling current signal, followed by a clear hysteresis, manifests the formation of a protrusion underneath (due to decrease of distance between the tip and the top graphene layer). This is proven by a subsequent STM scan (Fig. 3a, middle panel). This manipulation process is not only reversible, as shown by a similar procedure performed for negative voltage (Fig. 3c), but also highly reproducible (Extended Data Fig. 5 and Supplementary information STM movies).
To determine the influence of the interface protrusions on the electronic properties of BLG, we also performed a scanning tunneling spectroscopy experiment (Fig. 3d). Differential conductance \(\text{d}I/\text{d}V\) spectra obtained above the corners of the moiré interface pattern match closely to the former STM reports of epitaxial BLG on SiC(0001)47,49,55−57, particularly featuring a n-doped BLG system with a Dirac point located about 0.33 eV below the Fermi level. Comparison between spectra above protrusions and a regular interface shows consistent shift of about 10 mV towards positive voltages for the \(\text{d}I/\text{d}V\) spectra of protrusions. Due to efficient screening by the large density of electronic states from the top BLG layers, such small changes are expected42,57. Local p-doping of BLG above a protrusion supports the mechanism of electric-field-induced charge redistribution at the interface, which accompanies detachment of Cbuffer from the SiC substrate (Fig. 3e). This interpretation is also consistent with the photoelectron spectroscopy experiments performed after global decoupling of epitaxial graphene25,40, presenting the same trend of p-doping towards a neutral, freestanding system.
Local STM tip position dependence
Finally, we investigated a detailed position dependence of our local manipulation protocol for formation of protrusions (Fig. 4). To probe different lateral locations of the moiré corrugation pattern, we set a series of positive sample bias voltage ramps over a grid consisting of 86×86 points spanning an area of 10×10 nm2. First, from the fragmented STM image collected during the experiment (Fig. 4b), we obtain a distribution of the metastable protrusion locations. It is important to note that since soft imaging conditions were used in between positive voltage ramps, the local structure is not stabilized by the STM-induced electric field. The protrusions closely follow the (6×6)SiC lattice, but the corresponding sites are not equivalent. Moreover, interplay between local stress within the heterostructure preclude the formation of a complete lattice, i.e. the formation of a dense protrusion pattern in one part, annihilates them in the other. This effect is clearly seen via analysis of the single voltage ramp spectra performed away from the (6×6)SiC lattice sites. In this case, as the tip is in between two or three switchable interface locations, we often observe multiple switching events caused by inherent instabilities (Fig. 4c). In contrast, for locations directly above protrusions, we generally do not register any switching events, because once a protrusion is formed, it is stable under positive voltage conditions.
Figures 4d and 4e present examples of single-point spectra from the experiment for single and multiple switching events, respectively. The single events dominate (apart from zero events), they generally exhibit larger changes in tunneling current signals and larger threshold voltages (Fig. 4f). This behavior is caused mainly by the lateral distribution of single switching events close to the locations of expected protrusions. These results clarify the reproducibility of the previously described manipulation protocol by presenting its limitations, which are particularly strongly related to the position of the STM tip with respect to the (6×6)SiC moiré lattice.