The physics of solids is largely determined by their energy band structures. Therefore, the investigation and control of distinct electronic band dispersions assume a great length to understand and discover new states of the matter. One of the exotic electronic states is a type of flat band, which is predicted to host high-temperature superconductivity1,2,3,4, fractional quantum Hall effect5,6, and ferromagnetism7,8,9. In superconductors, a flat band can boost the coupling constant and the transition temperature (Tc) as a result of an enhanced density of states at the Fermi level (EF)10,11. This mechanism was utilized to explain the unexpected superconductivity in rhombohedral graphite and twisted graphene12,13,14. Other examples of flat band materials are Kagome lattices in which the flat band stems from destructive quantum interference due to the frustrated lattice geometry7,8. The Kagome-type flat bands have been observed in FeSn and Fe3Sn2 by angle-resolved photoemission spectroscopy (ARPES) and in Co3Sn2S2 by scanning tunneling spectroscopy (STS)8,9,10. However, the complexity of the electronic structure in the photoemission data and the lack of momentum resolution in STS make the observations elusive. Computational efforts have also been made focusing on designing flat band transition metal dichalcogenides (TMDs) through the formation of the Moiré superlattices, which could support strongly correlated physics at higher temperatures15,16,17. All these previous works conclude that the flat band media could be fertile to many novel states of the matter. However, the limited number of the materials with such non-trivial bands hinders future studies.

Motivated by earlier studies, we investigate the surface electronic structure of VSe2 TMD grown on the surface of Bi2Se3 topological insulator (TI) and show the emergence of a flat band in the electronic states. This flat band covers the entire kxky plane of the Brillouin zone (BZ) and displays dispersionless behavior along the kz direction as well. Furthermore, circular dichroism ARPES (CD-ARPES) measurements reveal that the CD signal of the flat band reverses the sign at several points within the BZ. Another notable observation is that the VSe2 overlayer and the emergence of the flat band do not reshape the Dirac cone of Bi2Se3 in the vicinity of the Dirac point (DP) unlike the case of transition metal doping which opens a large gap at the DP18. We also observe Moiré patterns in VSe2 domains of monolayer (ML) thickness and stripe-type patterns in bare Bi2Se3 through scanning tunneling microscopy (STM). Further elucidations on the crystalline and chemical properties of the system are provided by scanning transmission electron microscopy (STEM), and micro-spot low-energy electron diffraction (µLEED). Our results demonstrate a rich physics in this system and suggest a large family of materials as possible emergent flat bands and thus will motivate future studies based on heterostructure formed by other quantum materials such as superconductors.


Structural properties

Bi2Se3 and VSe2 are layered materials with their atomic stacking geometry shown in Fig. 1a. The layers in each compound are separated by van der Waals (vdW) gaps with weak covalent out-of-plane bonds connecting the layers. These properties allow for the formation of well-ordered VSe2/Bi2Se3 heterostructure despite the large in-plane lattice mismatch of around 20% between the two materials19,20. Figure 1b, c depict the relevant core-levels of such structures 0 ML, 0.3 ML, 2 ML, and 3 ML VSe2 on 12 quintuple layer (QL) Bi2Se3. Upon deposition of the VSe2, the Bi 5d peaks of Bi2Se3 located at 25.1 and 28 eV remain at the same binding energies, indicating the absence of V metals at the interface and/or in the bulk (Fig. 1b)21. This also shows that VSe2 surface deposition does not modify the chemical potential of Bi2Se3 seen as the absence of an electron or hole doping effect. Compared with pristine Bi2Se3, the Se 3d peak, however, appears at 0.1 eV higher binding energy for VSe2 grown sample. The difference in binding energy is possibly related to the charge density wave (CDW) phase of VSe219. Furthermore, V 2p1/2 and 2p3/2 peaks of VSe2 shown in Fig. 1c are located at 513 and 520.6 eV binding energies corresponding to +4 oxidation states being in agreement with the recent report20.

Fig. 1: Structural characterization of the VSe2/Bi2Se3 heterostructures.
figure 1

a Schematic representations of the top and side views of Bi2Se3 and VSe2 crystal structures. Hexagonal BZs with the high-symmetry points are given in the lower part of a. b Core-level photoemission spectra of Bi 5d and Se 3d of 0, 0.3, 1, 2, and 3 ML VSe2/12 QL Bi2Se3 heterostructures. c V 2p core levels for 0.3, 1, 2, and 3 ML VSe2/12 QL Bi2Se3 heterostructures. The atomic stoichiometry of Se to V is computed to be 2 by using the peak areas and photoionization cross-sections. Bi 5d and Se 3d peaks were recorded with 110 eV synchrotron radiation, whereas V 2p peaks were conducted by using 700 eV synchrotron radiation. d HAADF-STEM cross-section image of 3 ML VSe2/12 QL Bi2Se3 heterostructure. The color contrast in d is correlated with the atomic number (Z-contrast). e, f Room-temperature STM images of Bi2Se3 (at sample bias 100 mV, set point 1 nA) and VSe2 (at sample bias 80 mV, set point 1 nA) surfaces, respectively, obtained from 0.3 ML VSe2/12 QL Bi2Se3. Yellow parallelogram in f represents the unit cell of the Moiré pattern.

To further explore the system, we show a high-angle annular dark-field (HAADF)-STEM cross-section image of a 3 ML VSe2/12 QL Bi2Se3 heterostructure in Fig. 1d. Bi2Se3 and VSe2 exhibit regular atomic layers with smooth interfaces and vdW gaps marked with red arrows in Fig. 1d. On the other hand, the interface spacing between the Bi2Se3 and VSe2 layers comparably smaller than the vdW gaps which could strongly modify the local electronic structure. Furthermore, the STEM energy dispersive X-ray spectroscopy elemental maps presented in Supplementary Fig. 1 show the atomic distribution of Bi in the Bi2Se3 layers, V in VSe2 layers, and Se across the heterostructure as expected.

To study the local crystal structure, Fig. 1e, f show the STM image of Bi2Se3 and VSe2 regions of a 0.3 ML VSe2/12 QL Bi2Se3 sample, respectively. The Bi2Se3 surface has a stripe-like pattern similar to Cs and Fe doped Bi2Se322. The STM image of the VSe2 domains presented in Fig. 1f exhibits a Moiré pattern with ~2 nm × 2 nm superstructure. This differs from the previous studies conducted on VSe2/graphene19. Moiré pattern can be formed by a small misfit between the in-plane lattice parameters of the film and the underlying material or the relative rotation of two layers to each other, or both. By contrast, the lattice mismatch between the VSe2 and Bi2Se3 is quite large (about 20%). Unfortunately, we cannot make a quantitative analysis for precise determination of the in-plane lattice parameters or the atomic displacement due to the limitation in our STM data taken at room-temperature experiment. However, similar Moiré pattern formation is also observed on ML MoSe2 grown on a graphene substrate whose origin is attributed to the lattice mismatch between the multiple unit cells of the two materials23. Thereby, the Moiré pattern in VSe2 could be formed due to the small mismatch between four–five-unit cells of Bi2Se3 (4aBS = 16.56 Å or 5aBS = 20.7 Å) and five-unit cells of VSe2 (5aVS = 16.8 Å or 6aVS = 20.16 Å) for the rotationally aligned lattice geometry. Alternatively, the Moiré pattern could form be formed by the rotational misalignments of Bi2Se3 and VSe2 atomic lattices. Moreover, the details of the STM data reveal that the layer height is 6.8 Å for VSe2 (Supplementary Fig. 2) being in the line with the recent findings24.

Observation of the flat band in a VSe2/Bi2Se3 heterostructure

To examine the band structure, the binding energy vs. kx plots are given in Fig. 2a for 12 QL Bi2Se3 and in Fig. 2b–d for various thickness of VSe2 on 12 QL Bi2Se3. Bi2Se3 exhibits the typical band structure with the linear Dirac surface states (DSSs) forming the Dirac cone with the Dirac point (DP) at 0.36 eV below EF25. Upon deposition of 0.3 ML VSe2 on the surface of Bi2Se3, a flat band at 0.47 eV binding energy and with a ~0.18 eV bandwidth emerges in the surface electronic structure (Fig. 2b). The flatness of the band is well distinguished in the ARPES maps where bulk bands and the DSS of Bi2Se3 strongly disperse as a function of kx, while the flat band retains dispersionless across the \({\bar{\mathrm{{\Gamma}}}} - {\bar{M}}\) high-symmetry lines. VSe2 growth also induces the well-known M-state quantization26 of the bulk valance band of Bi2Se3 shown in Fig. 2b. The flat band can be resolved in 1 ML and, less intense, in 2 ML VSe2 (Fig. 2c, d) and further increasing the thickness of VSe2 to 3 ML leads to disappearance of the flat band from the ARPES map (Fig. 2e). On the other side, the thicker VSe2 grown on a Bi2Se3 exhibits the same electronic feature with the one grown on a highly ordered pyrolytic graphite (HOPG) substrate (Fig. 2f), indicating that the Bi2Se3 sublayer plays a crucial role in the formation of the new electronic states. It is also noticeable that the nature of the flat band stays unchanged with increasing VSe2 thickness as its binding energy and dispersionless character remain the same. The only observable evolution is in the spectral intensity which gradually decreases with increasing VSe2 thickness. This emphasized in Fig. 2g displaying the energy distribution curves (EDCs) taken at kx = ‒0.25 Å−1 from each ARPES maps. The peak stemming from the flat band can be seen up to 2 ML VSe2, while it is not visible for thicker films. This is likely to be correlated with the electron mean free path and suggests that this state is localized at the interface between the two materials. This can be further confirmed by taking advantage of energy dependency of the electron mean free path. In Fig. 2h, the EDCs at kx = ‒0.3 Å−1 for a 3 ML VSe2/12 QL Bi2Se3 heterostructure are shown for three photon energies. As expected, due to the enhenced electron mean free path27, a shoulder on the high binding energy side of V 3d develops with increasing photon energy from 50 to 220 eV, supporting interface-induced flat band argument.

Fig. 2: Observation of the flat band in the VSe2/Bi2Se3 heterostructures.
figure 2

a Experimental electronic structures of a 12 QL Bi2Se3 sample. be Electronic structure of 0.3, 1, 2, and ML VSe2/12 QL Bi2Se3 heterostructures, respectively. f Electronic structure of 1 ML VSe2 grown on a HOPG substrate. Spectra in af were collected with 110 eV linear horizontal polarized lights along the \(\bar {\Gamma} - \bar M\) direction in the BZ. BVB represents the bulk valance band of Bi2Se3. g Corresponding EDCs obtained along the kx = −0.25 Å. h Photon energy-dependent EDCs of 3 ML VSe2/12 QL Bi2Se3 obtained with 50, 110, and 220 eV photons. A Tougaard background is subtracted from each spectrum. The corresponding ARPES maps are presented in Supplementary Fig. 3. i, j Constant energy counters at the EF and the EFB for 1 ML VSe2/12 QL Bi2Se3 sample, respectively. The momentum maps were recorded with 169 eV photons and the corresponding ARPES map can be found in Supplementary Fig. 4. k Wide momentum window ARPES map of the 1 ML VSe2/12 QL Bi2Se3 sample obtained with 220 eV photon energy along the \(\bar {\Gamma} - \bar M\) direction of the BZ. The EDC on the top of k is obtained at the EF. Blue and red hexagons in i, j correspond to the BZ of VSe2 and Bi2Se3, respectively. In-plane lattice parameters of 4.14 Å for Bi2Se3 and 3.356 Å for VSe2 were employed to compute the BZs27,31.

Another interesting observation is that the flat band overlaps with the lower branch of the Dirac cone in the vicinity of kx = 0 Å−1 without inducing any prominent change in its spectral shape. This can be even better seen in the films with the thicker VSe2 coverage confirming that the flat band, DSSs, and the dispersive V 3d state of VSe2 coexist in the surface electronic structure (Fig. 2b–d and Supplementary Fig. 5). It is also found that the DP of Bi2Se3 does not experience an energy shift upon surface deposition of VSe2. This indicates the absence of any band bending effect which is consistent with the behavior of Bi 5d core levels presented in Fig. 1b.

To further investigate the flat band, a kxky intensity plots at EF and at the binding energy of the flat band (EFB) for a 1 ML VSe2/12 QL Bi2Se3 sample are shown in Fig. 2i, j, respectively. The Fermi surface is dominated by a flower-like electron pocket formed by the V 3d orbitals of VSe2 centered at \({\bar{\mathrm{{\Gamma}}}}\) point as similar to the earlier observation for VSe2 grown on different substrates28. The constant energy cut at EFB is instead quite featureless. Besides, a weak residual of the start-like features of VSe2, no new, distinct dispersion is seen along the any direction of the BZ. This indicates that the flat band fills the entire BZs of Bi2Se3 and VSe2 as depicted by blue and red hexagons in Fig. 2i, j, respectively. This can be also seen in Supplementary Fig. 4 where the spectra taken along the different directions in the BZ for fixed kx and ky momentum all show the existence of the flat band. Such electronic state spread over a large momentum area can significantly enhance the electronic correlation yielding quantum effects at very high-temperatures. It is also worth noting that the LEED pattern of the sample shows stretched diffraction spots along the rotational direction indicating the presence of the rotationally misaligned VSe2 domains (Supplementary Fig. 2a) with respect to each other and to the Bi2Se3 substrate. The rotational misfit of ±3° estimated from µLEED pattern, however, is too small for a band to span whole BZ and to induce a fully occupied constant energy counter in the momentum space.

In Fig. 2k, we also present an ARPES map covering two BZ centers. The flat band connects two \({\bar{\mathrm{{\Gamma}}}}\) points to each other. Another intriguing realization in this spectra is that the Bi2Se3 and VSe2 shares the same in-plane lattice constants. This is consistent with our LEED measurements, which does not show distinct diffraction patterns arising from VSe2 and Bi2Se3 (Supplementary Fig. 2a). This is supported with the EDC taken at the Fermi level and given on the top of the spectra in Fig. 2k. Two zone centers are separated with kx = 1.9 Å−1 yielding a 4.03 Å in-plane lattice parameter. This is smaller than in-lane lattice parameter of Bi2Se3 (a = 4.545 Å)19, while larger than the one of VSe2 (a = 3.356 Å)23. This strong lattice relaxation can significantly modify the electronic structure of both material in the vicinity of the interface and induce new electronic states.

Photon energy-dependent electronic structure

In ARPES experiments, by recording the electronic structure with a wide photon energy range, a k vs. kz or binding energy vs. kz dispersions can be also extracted. This method allows studying the dispersion of the energy bands along the kz (out-of-plane) direction to distinguish the two-dimensional (2D) bands from the dispersive bulk bands. Such spectra acquired at varying photon energies (from 45 to 120 eV with 5 eV steps) for a 1 ML VSe2/12 QL Bi2Se3 heterostructure are given in Fig. 3. In the plot of kz vs. ky dispersion at EF, DSSs marked with dashed black lines exhibit no kz dependence (Fig. 3a). Similar spectrum at EFB given in Fig. 3b shows that a high spectral intensity along the ky = 0 Å−1 originates from the bottom of the Dirac cone of Bi2Se3. Away from the ky = 0 Å−1, the plot has non-vanishing spectral intensity contributed by the flat band. This can be better seen in the momentum distribution curves obtained at various kz points (Fig. 3c) in which each spectrum exhibits always finite density of states along the ky momentum direction. This implies the dispersionless nature of the flat band along the kz momentum direction. To further validate this observation, we present the binding energy–kz plots along the ky = ±0.25 Å−1 in Fig. 3d, e, respectively. The plots clearly show that the flat band at 0.47 eV binding energy is kz independent confirming its non-bulk derived nature. We should also note that the M-shape bulk band located in the vicinity of 1 eV binding energy exhibits a nearly nondispersive feature along the kz as shown in Fig. 3d, e. To further reveal the details of the flat band, Fig. 3f depicts the EDCs taken at different kz points. One can see that the EDC of the flat band does not exhibit a kz-dependent evolution in the binding energy and bandwidth, providing a signature that it has 2D nature and originates from single type of atomic orbital.

Fig. 3: Photon energy-dependent electronic structure of a 1 ML VSe2/12 QL Bi2Se3 heterostructure.
figure 3

a, b kykz dispersions at the EF and the EFB, respectively. Dashed red lines in a mark the DSSs. c MDCs at different kz points. d, e Binding energy vs. kz maps at ky = ±0.25 Å−1, respectively. Dashed cyan colored lines in d and e represent the dispersion of the flat band along the kz direction. f EDCs at various kz points to study the spectral shape of the flat band. ARPES maps for the plots were conducted along the \({\bar{\mathrm{{\Gamma}}}} - {\bar{\mathrm{K}}}\) direction in the BZ. Data were conducted from the 1 ML VSe2/12 QL Bi2Se3 heterostructure at 10 K.

Circular dichroism ARPES

CD-ARPES has gained great attention due to its feasibility to investigate the helical spin–orbit texture in topological surface states29. The principle of the method is the spectral weight differences in ARPES arising from the opposite helicity of the circularly polarized lights. CD-ARPES is then obtained from \(\left[ {\left( {I_{{\mathrm{RCP}}} - I_{{\mathrm{LCP}}}} \right)} \right]/\left[ {\left( {I_{{\mathrm{RCP}}} + I_{{\mathrm{LCP}}}} \right)} \right]\) where IRHP and ILHP are photoemission intensities for right hand circular polarized (RCP) and left hand circular polarized (LCP) lights, respectively. Thus, we have recorded the band structure of 0.3 ML VSe2/12 QL Bi2Se3 sample with RCP and LCP, shown in Fig. 4a, b, respectively. The corresponding CD-ARPES is presented in a binary color map in Fig. 4c (red: negative-CD and blue: positive-CD). The bulk bands of Bi2Se3 dispersing below 0.8 eV binding energy show a strong CD signal as seen in Fig. 4a–c. CD signal of the DSSs exhibits a spectral weight switching from the −ky to +ky regions when changing the excitation energy from RCP to LCP. For clarity, the DC signal vs. ky is plotted in Fig. 4d at 0.1 eV binding energy where the CD is positive for left and negative for the right side of the Dirac cone, marked with vertical arrows. Further away from the ky = 0 Å−1, the plot in Fig. 4d still shows non-zero CD. This is likely originating from the V 3d orbitals, which dominate the density of states at the EF for VSe219.

Fig. 4: CD-ARPES.
figure 4

a, b ARPES maps of 0.3 ML VSe2/12 QL Bi2Se3 sample recorded with RCP and LCP lights, respectively. c Computed CD-ARPES. d, e CDs as a function of ky at 0.1 eV binding energy and the EFB, respectively.

To investigate the dichroism effect in the flat band, the CD at EFB is also plotted as a function of ky in Fig. 4e and it exhibits sign inversions at ky = 0 Å−1 and ky = ±0.5 Å−1, and the maxima at ky = ±0.25 Å−1. This shows that similar to the DSSs, the CD in the flat band also exhibits helical texture where opposite ky momentums have opposite signs of the CD. Notably, zero CD signal is also observed as white color in the CD-ARPES along the ky = 0 Å−1. This depicts the nodal line, which was proposed to be the characteristic feature of the 2D electronic structure30. In particular, the CD signal of the DSSs depends on the incident photon energy assigning it to the final state effect in the photoemission process29. This was discussed in ref. 30 with details where they propose the non-trivial connection between the spin–orbit texture and the CD signal. Thereby, the helical CD texture and the nodal line band suggest that the flat band could be topologically non-trivial.

Temperature-dependent electronic structure

To further study the nature of the flat band in detail, we present the temperature-dependent ARPES maps of 1 ML VSe2/12 QL Bi2Se3 in Fig. 5a–d and 0.3 ML VSe2/12 QL Bi2Se3 in Fig. 5f–i for 10, 100, 200, and 300 K, respectively. For both samples, the temperature does not induce a prominent change on the nondispersive nature of the flat. This can be seen in the EDCs obtained along the ky = −0.25 Å−1 from the ARPES maps (Fig. 5e–j). It is clear from the data that temperature does not affect the line shape or the binding energy of the flat band. At any given temperature, the flat band is located at 0.47 eV. Only noticeable change is the broadening of all spectral features with increasing temperature as can be expected from phonon contributions to the total lifetime. On the other side, these data constitute a strong indication that the flat band is not likely to originate from disorder or impurities. The disorder usually shows a considerable dependence on the temperature as a modulation in the density of defects. This refers that the binding energy and the density of the states of the flat band should change if it is formed by the impurities or disorders. This can be found in a recent study where the disorder induced kink like states in Bi2Se3 can be diminished by lightly annealing the sample31. The temperature-dependent ARPES data also show that the flat band is not due to surface impurities. It is expected that the impurities will segregate into the bulk of Bi2Se3 and locate in sublattices with increasing temperature32. This will significantly modify the chemical potential of the system. A number of ARPES studies have been devoted to studying the effects of the adatoms on the band structure of Bi2Se3 and the only change found has been electron or hole doping18,26,32. No experimental work has shown indication of any new states or flat like bands induced by surface impurities on the surface electronic structure of a TI. Furthermore, resonant ARPES experiments also revealed that the transition metal impurity states evolve onto the surface states rather than forming a flat band33,34. Finally, it is worth mentioning that the ARPES maps taken at room temperature can be compared with the room-temperature STM data presented in Fig. 1f, leading to the conclusion that the Moiré pattern and the flat band electronic states coexist in our system.

Fig. 5: Temperature-dependent electronic structure of the VSe2/Bi2Se3 heterostructures.
figure 5

ad ARPES maps of 2 ML VSe2/12 QL Bi2Se3 at 10, 100, 200, and 300 K, respectively. e EDCs obtained along the ky = −0.25 Å−1 from each spectrum. fj Same as ae but for 0.3 ML VSe2/12 QL Bi2Se3 sample. Spectra were obtained with 110 eV photons.


Here we present a detailed ARPES study on a VSe2/Bi2Se3 system and showed unexpected formation of a flat band on the surface electronic structure on this heterostructure. One of the first scenario to consider as the origin of the flat band is the surface V impurities. Even though the theory predicts nearly nondispersive features induced by impurities or disorders in TIs31,35, experimental studies shows that these states evolve onto the surface states rather than being flat in the momentum space33,34. Also, in contract to our results, these states are expected to strongly modify the Dirac cone and open an energy gap at the DP31,35. Another scenario to consider would be the existence of superlattices as seen in \(\sqrt 3\) × \(\sqrt 3\) silicene superstructure by STS where the local density of states forms the electronic Kagome lattice36. Interface dislocation or strain can also flatten the original bands by introducing pseudo-magnetic field term to the Hamiltonian in Moiré superstructures4. Furthermore, a pronounced band flattening in this scenario requires superstructure patterns with at least a few tens of nanometers periodicity which is much larger than one observed in the present case. However, in contrast to our observations, the flat band discussed within the superlattice frameworks is dispersionless only in the BZ of the superstructure17,36.

On the other hand, our results show the emergent character of our flat band, which is likely due to the formation of the interface in the heterostructure between Bi2Se3 sublayer and VSe2 top layer. This conclusion is supported by thickness and photon energy-dependent photoemission data showing that the flat band can be resolved with higher photon energies on thicker VSe2 films. As a first example of the dispersionless electronic excitation in a topologically non-trivial band structure, our results could open a new pathway in the critical field of experimental realization and control of novel quantum effects.



Molecular beam epitaxial growth (MBE) technique was employed to grow VSe2/Bi2Se3 and Bi2Se3 samples in a custom ultrahigh vacuum system located at the ESM beamline of NSLS-II. Se and Bi sources (5 N) were evaporated from the ceramic crucibles while the e-beam evaporation method was used for V (99.8% purity) source. All samples were grown on Al2O3(0001) substrates at 255 °C. Before the growth, the substrates were first degassed at 550 oC for three hours and flashed at 850 °C for 5 min. An ML VSe2 were grown on a freshly cleaved HOPG substrate at 255 °C after annealing the substrate at 350 °C for 2 h. Sample thicknesses were estimated within a 15% error bar by using a quartz thickness monitor and X-ray photoemission spectroscopy. Samples for ARPES and µLEED experiments were capped with 20 nm amorphous Se film before being removed from the MBE chamber.

Core-level spectroscopy

Core-levels were recorded at 21-ID-1 ESM beamline of National Synchrotron Light Source II (NSLS-II) by using a DA30 Scienta electron spectrometer at 10 K sample temperature.

Angle-resolved photoemission spectroscopy

ARPES experiments were performed at 21-ID-1 ESM beamline of NSLS-II by using a DA30 Scienta electron spectrometer. The pressure in the photoemission chamber was 1 × 10−10 Torr and samples were kept at 15 K during the experiment by a closed-cycle He cryostat. The energy resolution in the ARPES experiments was better than 15 meV with a spot size of ~20 µm. Before the ARPES experiments, samples were annealed at 220 °C for 30 min to remove the Se capping layer. The angle between the light and the surface normal of the sample is 55° at the normal emission during the ARPES experiments. The films were grounded with a tantalum clip. A part of the ARPES experiments was conducted at the linear undulator beamline at the Hiroshima Synchrotron Radiation Center BL-1 (Supplementary Fig. 5). Photon energy is converted to kz momentum space by using the free electron final state approximation \(\hbar k_z = \sqrt {2m_e\left( {E_{\mathrm{kin}}{\it{cos}}^2\theta + V_{\mathrm{o}}} \right)}\) where me is the free electron mass, Ekin is the kinetic energy of a photoelectron, and Vo is the inner potential taken as 11.8 eV for Bi2Se337.

TEM and STM microscopy

HAADF-STEM images were acquired with Hitachi HD2700C dedicated STEM with a probe Cs corrector operating at 200 kV at room temperature. Samples were prepared using the in-situ lift-out method on the FEI Helios 600 Nanolab dual-beam FIB. Final milling was completed at 2 keV. STM (Omicron VT- STM -XA 650) experiments were performed in an ultrahigh vacuum (UHV) system with a base pressure of 2 × 10−10 Torr at room temperature. All the STM images were observed in constant current mode using Pt/Ir tips. All bias values in the text refer to the bias applied to the sample. The STM images were analyzed using Gwyddion-2.55 software package. HAADF-STEM and STM experiments were conducted at the Center for Functional Nanomaterials, Brookhaven National Laboratory. Samples for STM were transferred with a vacuum suitcase.

Low-energy electron diffraction

µLEED experiment was performed at X-ray photoemission electron microscopy/low-energy electron microscopy (XPEEM/LEEM) endstation of the ESM beamline (21-ID-2).