Atomic-layer materials have recently attracted attention, because they exhibit a variety of anomalous properties distinct from those of their three-dimensional bulk counterparts, as exemplified by the Dirac-cone states in graphene1 and the high-temperature superconductivity in ultrathin iron-based superconductor films.2 Whereas atomic-layer materials are inherently unstable, owing to their tendency to progress toward long-range order,3 such characteristics may facilitate the exploration of new crystal phases that cannot usually be realized in bulk. For instance, the local crystal structure at the surfaces of transition metal dichalcogenides (TMDs) is known to be sensitive to crystal defects, chemical reactions and thermal effects,4, 5, 6, 7, 8, 9 thus suggesting the feasibility of crystal-phase control in two dimension. The structural transition between 2H- (trigonal prismatic structure, D3h) and 1T- (octahedral structure, Oh) polymorphs (see Figure 1a) in monolayer MoS2 is a typical example that cannot be achieved in the bulk phase.5, 6, 7 Bulk NbSe2 has several polymorphs: 3R, 4H and 2H, although the essential structural unit is identical, trigonal prismatic. Here we term a monolayer consisting of this structural unit 2H-NbSe2.

Figure 1
figure 1

Characterization of monolayer 2H- and 1 T-NbSe2. (a) Schematic crystal structure of monolayer 2H- and 1T-MX2 (M=transition metal; X=chalcogen), shown in top and side views. (b, c) Reflection high-energy electron diffraction (RHEED) patterns of bilayer (BL) graphene and monolayer NbSe2 (mixed phase), respectively, obtained along the [100] direction of the silicon carbide (SiC) substrate. The top panels illustrate the sample and the substrate. (d) Constant-current STM image of a surface area of 100 × 50 nm (sample bias voltage Vs=−1.0 V and set-point tunneling current It=100 pA). (e) Height profile along a cut indicated by yellow lines in d. (f, g) High-resolution STM images measured in the surface region enclosed by the rectangle in d for type-A (Vs=−0.5 V) and type-B (Vs=+0.6 V) islands, respectively. (h) Representative tunneling spectra for type-A (red) and type-B (blue) islands.

Among TMDs, bulk 2H-NbSe2 has been the target of intensive studies, because it exhibits a unique ground state in which an incommensurate charge density wave (CDW) and superconductivity coexist below 7 K.10 Recent success in fabricating monolayer and multilayer 2H-NbSe2 films has revealed drastic changes in the superconducting and CDW transition temperatures with changes in the number of layers, thus providing a valuable opportunity for studying key questions regarding the interplay among dimensionality, the CDW and superconductivity.11, 12, 13, 14 In contrast to the intensively studied 2H-NbSe2,15, 16 the 1T counterpart has not yet been synthesized either in bulk or as an ultrathin film. Although a 1T-like phase has been observed locally around defects17, 18, 19 in thermally treated bulk samples,20 AgxNb2Se3 (Koslowski et al.21) and Nb1−xTixSe2 (Di Salvo et al.22), it has been difficult to selectively fabricate a pristine sample of a single phase for precise physical measurements as well as for device applications. It is thus of great importance to fabricate an ultrathin film of pristine 1T-NbSe2 and to investigate its electronic states to reveal as-yet-unknown electronic properties. Here, combining molecular-beam epitaxy and state-of-the-art electron spectroscopy methods, we succeeded in selectively fabricating 1T and 2H single-phase monolayer NbSe2 on bilayer graphene and investigated the anomalous electronic states in detail.

Materials and methods

A monolayer NbSe2 film was grown on bilayer graphene by using molecular-beam epitaxy in ultrahigh vacuum (3 × 10−10 Torr). Bilayer graphene was prepared by annealing an n-type Si-rich 6H-SiC(0001) single-crystal wafer,23, 24 by resistive heating at 1100 °C in high vacuum—better than 1.0 × 10−9Torr—for 20 min. A monolayer NbSe2 film was grown by evaporating Nb on the bilayer graphene substrate in a Se atmosphere. The substrate was kept at 500–590 °C during epitaxy. The as-grown film was annealed at 400 °C for 30 min and then transferred to the angle-resolved photoemission spectroscopy (ARPES)-measurement chamber without breaking vacuum. The growth process was monitored by reflection high-energy electron diffraction. The custom-made scanning tunneling microscopy (STM) system was operated at T=4–6 K under ultrahigh vacuum (below 2 × 10−10 Torr). The ARPES measurements were carried out using an MBS-A1 (MB Scientific AB, Uppsala, Sweden) electron-energy analyzer with a high-flux helium discharge lamp and a toroidal grating monochromator at Tohoku University. The energy and angular resolutions were set at 16 meV and 0.2º, respectively. The Fermi level (EF) of samples was calibrated with a gold film deposited onto the substrate. First-principles band-structure calculations for free-standing 2H- and 1T-NbSe2 monolayers were carried out by using the Quantum Espresso code.25 Spin–orbit interactions were included in the calculations. The plane-wave cutoff energy and the k-point mesh were set to be 30 Ry and 12 × 12 × 1, respectively. The thickness of the inserted vacuum layer in a model crystal was set to ~10 Å.

Results and discussion

We used the van der Waals epitaxy technique26 to fabricate monolayer NbSe2 (Ugeda et al.13) by using bilayer graphene grown on silicon carbide as a substrate. Figure 1b shows the reflection high-energy electron diffraction pattern of bilayer graphene on 6H-SiC(0001), which signified a 1 × 1 streak pattern and spots originating from the bilayer graphene and the buffer layer beneath it, respectively.23, 24 After co-evaporation of Nb and Se atoms onto a substrate kept at 530 °C in ultrahigh vacuum, the reflection high-energy electron diffraction intensity of spots was reduced and a new 1 × 1 pattern appeared (Figure 1c). This behavior is characteristic of a TMD ultrathin film, as has been observed in monolayer WSe2 and TiSe2 on bilayer graphene.23, 24 STM revealed the formation of monolayer NbSe2 islands (Figure 1d), whose height (~ 0.7 nm; see line profile in Figure 1e) was nearly equal to the distance between adjacent NbSe2 layers in bulk 2H-NbSe2, 0.63 Å.10 We found that this film was composed of two different types of islands. One, called type-A here, exhibited a 1 × 1 atomic structure (Figure 1f), whereas the other, type-B, showed no clear atomic order. The former islands showed characteristic V-shaped density of states with a finite density of states at zero-bias voltage in the dI/dV curve at T=6 K, a result reminiscent of the spectral features of bulk 2H-NbSe2.18 This island type is probably made of 2H domains. However, the dI/dV curve of type-B islands showed markedly different characteristics; the density of states vanished in a wide energy region of ±0.2 eV with respect to EF, indicating an insulating nature with an energy gap of ~0.4 eV. These results demonstrated that both metallic and insulating crystal phases coexisted in this monolayer NbSe2 film.

To further clarify the nature of the two different islands, we performed ARPES on this film. It was expected that the characteristic spectral features from both domains would be simultaneously observed in an ARPES spectrum, because this technique probes a relatively large surface area (~2 × 2 mm2). As shown in Figure 2a, the ARPES-intensity plot along the ΓM cut resolved several dispersive bands in the valence band region, reflecting the high single-crystallinity of the film. Figure 2b shows the valence-band ARPES-intensity plot of bulk 2H-NbSe2, which is useful for distinguishing the contribution of the 2H phase in a monolayer film, because the overall experimental valence band dispersion in this wide energy range is known to be similar between the bulk and the monolayer.13, 15, 16 In previous studies, the band calculations have revealed several differences in the band dispersions near EF between the bulk and the monolayer, such as the presence or absence of the kz dispersion in the Nb 4d band and the EF crossing or non-crossing of the anti-bonding Se pz band.13, 27 A side-by-side comparison of the ARPES-intensity plots in Figure 2a and b revealed that the observed band structures shared several common features, for example, the band near EF around the M point, which crosses EF midway between Γ and M, and the hole-like bands with the top of the dispersion at the binding energies (EB) of 0.9 and 1.2 eV at Γ, as highlighted by the blue dashed curves. In Figure 2c, some corresponding bands are visible in the calculated band structure for monolayer 2H-NbSe2, whereas several quantitative differences are also apparent in the energy of the bands (see, for example, the hole-like bands at EB~1 eV at Γ). In addition to the bands ascribed to the 2H phase, Figure 2a shows additional bands that are absent in bulk and monolayer 2H-NbSe2,13, 15, 16 such as a relatively flat feature at EB—0.4 eV around Γ (here denoted the α-band), a faint hole-like band at Γ (β-band) and a larger hole-like band (γ-band), as highlighted by the red dashed curves. These bands are likely to be of 1T phase in origin, because a corresponding feature in the calculated bands for monolayer 1T-NbSe2 (Figure 2c) is apparent, particularly in the β- and γ-bands. Intriguingly, although the calculations predict the EF crossings of hole-like bands around Γ, the ARPES intensity was strongly suppressed around EF, thus suggesting an opening of an energy gap. Apparently, this gap was not an ordinary band gap, because the first-principles calculation failed to reproduce it, as shown in Figure 2j, which highlights the essential difference between the experimental and the calculated band dispersions around EF. The opening of such an energy gap could not be explained in terms of carrier doping, because the band calculation for monolayer 1T-NbSe2 showed no band gap around the Γ point, irrespective of the location of the chemical potential (Figure 2c). Instead, the observed gap may have arisen from a Mott–Hubbard gap, because the similar material bulk 1T-TaSe2 shows almost identical band dispersions in the Mott phase.28, 29, 30 In this context, the α-band was assigned to the lower Hubbard band. Moreover, the insulating type-B islands in the STM image probably arose from the 1T phase, taking into account the similar energy scales for the gap below EF found by ARPES and STM (~0.25 eV), as seen from a comparison of Figures 1h and 2j.

Figure 2
figure 2

Electronic structure of monolayer 2H- and 1 T-NbSe2. (a) Plot of valence-band ARPES intensity for monolayer NbSe2 (mixed phase obtained at a substrate temperature TS of 530 °C) measured along the Γ-M cut. ARPES data were recorded at T=40 K at the He-Iα line (=21.218 eV). (b) As in a, but for a cleaved surface of bulk 2H-NbSe2. Blue dashed curves in a, b highlight the band dispersions that are commonly seen in a, b. Red dashed curves are a guide, indicating the bands that are absent in bulk 2H-NbSe2. (c) Calculated band structure obtained from the first-principles band calculations for monolayer 2H- (blue) and 1 T- (red) NbSe2 compared with the ARPES-intensity plot of monolayer NbSe2 (mixed phase; as in a, but plotted in grayscale). (d, e) ARPES intensity plotted as a function of wavevector and binding energy for monolayer 2H-NbSe2 (TS=515 °C) in the valence band and near-EF regions, respectively. (f) ARPES-intensity mapping at EF plotted as a function of the two-dimensional wavevector for monolayer 2H-NbSe2. The intensity at EF was obtained by integrating the ARPES intensity within ±10 meV of EF. The solid and dashed lines indicate the high-symmetry lines in the Brillouin zone for the (1 × 1) structure. (g) Constant-current STM image for monolayer 2H-NbSe2 (sample bias voltage Vs=500 mV) measured at 4 K. (h) STM image expanded in the area enclosed by the small rectangle in g (Vs=10 mV). The diamond corresponds to the 3 × 3 unit cell. (im) As in dh but for monolayer 1T-NbSe2 (TS=590 °C). The sample bias-voltage in l, m is Vs=−2 V. The diamond in m corresponds to the unit cell arising from the star-of-David clusters.

Having established the existence of a 1T/2H-mixed phase in monolayer NbSe2, we next addressed how to selectively fabricate these two phases. We found that the temperature of the bilayer graphene substrate (TS) during the epitaxy of the NbSe2 film was a major parameter controlling the crystal phase. Whereas the mixed phase presented in Figures 1 and 2a was obtained at TS=530 °C, we found that the 2H (1T) single phase was obtained at a slightly lower (higher) TS. As evident from Figure 2d and e, which show the ARPES intensities for the 2H phase (TS=515 °C), no obvious mixture of the 1T-phase bands was observed (compared with Figure 2a and d). Similarly, the ARPES intensity for the 1T phase (TS=590 °C) showed negligible contamination from the 2H phase, as evident from the absence of EF crossing bands, as shown in Figure 2i and j. This selective fabrication of the 2H and 1T phases is also highlighted in Figure 2f and k, which show that the Fermi surface (FS) of the 2H phase consisted of a large hexagonal pocket at Γ and another large pocket at K (Figure 2f), as in the bulk sample,15, 16 whereas the FS was completely absent in the 1T phase, as seen from the featureless intensity pattern at EF (Figure 2k). The control of crystal phases via temperature has been used in the chemical vapor transport method to obtain 1T and 2H phases in some bulk TMDs.31, 32, 33 Our results thus suggest that temperature control is a key factor for obtaining different crystal phases in both monolayer and bulk TMDs.

The selective fabrication of 1T and 2H phases was further corroborated by our STM measurements. We found that the 1T phase (Ts=590 °C) consisted of several monolayer 1T islands on the graphene substrate (Figure 2l), unlike the 2H counterpart, which appeared to grow layer by layer, as inferred from its large terrace size (Figure 2g). Intriguingly, closer examination of one of the 1T islands in Figure 2m revealed the modulation of density of states with periodicity (note that this modulation is not clearly resolved in the 1T/2H-mixed phase in Figure 1g, probably because of the low film quality). This result strongly suggests the formation of ‘star-of-David’ clusters, similar to the case of the Mott-insulating (commensurate CDW) phases of bulk 1T-TaS2 and TaSe2,28, 30, 34 thus supporting the Mott-insulating nature of the 1T phase. In contrast, the STM image of the 2H counterpart showed a 3 × 3 intensity modulation (Figure 2h), which followed the periodicity of the CDW in bulk and monolayer 2H-NbSe2.13, 26 These STM results, together with the ARPES results, firmly establish the intrinsic differences between the electronic states of the 1T and 2H phases.

To experimentally determine the temperature ranges suitable for the selective growth of pure 1T or 2H single phases, we performed systematic ARPES measurements for samples with various TS values. We evaluated the ratio of 1T and 2H components from the energy distribution curve along the ΓM cut on the hexagonal pocket, seen in the 2H phase (kx=0.5 Å−1). We numerically fit the energy distribution curve with two Lorentzian peaks originating from the 2H and 1T phases located at around 0 and 0.4 eV, respectively. In Figure 3b, the energy distribution curve for TS=515 °C showed a single peak at EF, thus supporting its purely 2H-phase nature. On increasing TS, the intensity of this peak was monotonically reduced, in accordance with the evolution of the 0.4 eV peak. This finding suggested that the ratio between the 1T and 2H phases systematically changes as a function of TS, as highlighted by the phase diagram in Figure 3a. We also found that TS values below 500 °C or above 600 °C were not suitable for the fabrication of monolayer NbSe2, owing to the presence of multiple 2H domains with unfixed axis orientations and/or the difficulty of maintaining a stable monolayer crystal phase. From this empirical evidence, together with the results of Figure 3a, we concluded that the TS values best suited for obtaining 2H or 1T single phases are 500–515 °C and 580–600 °C, respectively.

Figure 3
figure 3

How to selectively fabricate monolayer 2H- and 1 T-NbSe2. (a) Surface area ratio between the 2H and 1T domains, plotted as a function of TS, obtained by numerical fits to energy distribution curves (EDCs). The left and right panels illustrate the schematic density of states (DOS) for monolayer 2H- and 1T-NbSe2, respectively. (b) Near-EF EDC at kx=0.5 Å−1 for monolayer NbSe2 films fabricated at various substrate temperatures (TS=515, 530, 560 and 590 °C). The green curves are the numerical fittings with two Lorentzian curves (representing the 1T (red) and 2H (blue) domains) and a linearly decreasing background multiplied by the Fermi–Dirac distribution function. The ratio between the 2H and 1T phases was estimated according to the total spectral weight of the Lorentzians.

A previous study of monolayer 2H-NbSe2 (Ugeda et al.13) has suggested that two types of 2H domains, rotated by 30° with respect to each other, can exist when the film is fabricated at low TS (~330 °C). As described above, we found a way to selectively fabricate the single-domain 2H phase by keeping the TS above 500 °C. When two domains are mixed in a sample, the appearance of a complicated FS is expected. For example, besides the intrinsic FS at K, a replica FS should appear around M. In contrast, as shown in Figure 2d, our ARPES intensity mapping for the Ts=515 °C sample showed no signature of the FS at M, thereby strongly suggesting the single-domain nature of our 2H-phase sample. This conclusion is also supported by the experimental evidence that the FS at the Γ point exhibited a clear hexagonal shape consistent with the sixfold symmetry of a single domain.

The present success of the selective fabrication of insulating 1T and metallic 2H phases may contribute to realizing a field effect transistor based on semiconducting monolayer 1T-NbSe2 and metallic 2H-NbSe2. Although polymorph field effect transistors have recently been reported by using other TMDs such as MoS2 and MoTe2,8, 9 the unique Mott-insulating state of 1T-NbSe2, distinct from the states of the band-insulating 2H-MoS2 and 2H-MoTe2, would provide a new means of controlling a device, because the characteristics of the Mott-insulating state in the 1T phase are known to be altered by various physical parameters, such as pressure and disorder.34, 35 Furthermore, as monolayer 2H-NbSe2 exhibits superconductivity robust against magnetic fields owing to space inversion symmetry breaking and the resultant spin imbalance in the electronic states,12 the present result also opens a path toward atomic-layer electronics based on superconductor/semiconductor heterojunctions.