Introduction

Transition-metal dichalcogenides (TMDs) have attracted increasing interest because of their unique electrical and optical properties1,2. TMDs have a layered structure with the common chemical formula of MX2 (M=Mo and W; X=S, Se, and Te), and have polymorphs, such as hexagonal 2H, octahedral 1T, distorted octahedral 1T′, and orthorhombic Td. While widely-studied S- and Se-based TMDs are typically stabilized in the hexagonal 2H structures, MoTe2 is crystallized into either the 2H or the 1T’ phase, owing to the small energy difference (<0.05 eV per formula unit) between the two phases3. Thus the synthesis of high-quality and large-area MoTe2 films with a precisely controlled phase is particularly important for practical applications in novel devices.

Chemical vapor deposition (CVD) is one of the most promising methods to synthesize TMD films, and the successful CVD growth of phase-controlled 2H- and 1T’-MoTe2 has been reported4,5. While the CVD growth of well-aligned TMD films has been recently demonstrated6, large-area epitaxial growth of continuous and seamless TMD films with low defect density has been still a challenging issue. Molecular-beam epitaxy (MBE) has recently received increased attention as an alternative method. Because of its superiority in controlling the thickness and composition of epitaxial films, MBE has been expected to synthesize atomically thin and high-quality TMD films over the entire wafer7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23.

In this work, we investigated the structure and morphology of the MBE-grown MoTe2 monolayer on GaAs(111)B using scanning tunneling microscopy (STM), electron diffraction, and X-ray photoelectron spectroscopy (XPS). We show that both 2H and 1T′ phases coexist in the as-grown MoTe2 film and that the MoTe2 film after the annealing under the Te-rich conditions exclusively consists of the 2H phase. In the epitaxial 2H-MoTe2 film, in addition to the azimuthally aligned domains, the presence of in-plane twisted domains rotated by ± 0.77 is identified: the rotation of domains provides a slight better lattice mismatch than the non-rotated domains. Both orientations are found to be stable, as confirmed by first-principles calculations. While earlier studies have shown that MoTe2 can be reversibly transformed from the 2H phase to the 1T′ phase4,24, and that the 1T′ phase is favored by the condition of high temperatures and low Te concentrations4,24, the 2H to 1T′ transition has not been observed in the present experiments. Instead, we found that the annealing of the MoTe2 film under Te-deficient conditions causes the desorption of Te, leading to the formation of nanowire-like structures.

Results and discussion

Epitaxial growth of MoTe2

Figure 1a–c shows a series of reflection high-energy diffraction (RHEED) patterns taken during the formation of MoTe2 films on GaAs(111)B. After the growth for 80 min at 300 C, broad and weak streaks are observed in the RHEED patterns [Fig. 1b]. The in-plane lattice constant extract from the RHEED patterns is 0.36 nm, indicating that the 2H-MoTe2 were epitaxially grown with their [11$$\overline{2}$$0] direction being parallel to the GaAs[1$$\overline{1}$$0] direction. As indicated by arrowheads, there exist streaks at the half-order positions in both [1$$\overline{1}$$0] and [11$$\overline{2}$$0] directions. As suggested in earlier papers18,19,20, the coexistence with the 1T′-MoTe2 could account for the present result: the superposition of three equivalent rotational alignments of the 1T′-MoTe2 lattice on GaAs well explain the appearance of half-order reflections in Fig. 1b (see Supplementary Fig. 1)

The existence of the 1T’ phase in the as-grown MoTe2 sample is further confirmed by XPS measurements. Shown in Fig. 2a are high-resolution XPS spectra of Mo 3d and Te 3d measured from the as-grown MoTe2 sample. Both Mo 3d and Te 3d spectra consist of two components. For the components with higher binding energies, Mo 3d5/2 and Mo 3d3/2 peaks are located at 228.6 and 231.8 eV, respectively, and the peak positions of Te 3d5/2 and Te 3d3/2 are 573.3 and 583.7 eV, respectively. These values are in good agreement with previously reported values for 2H-MoTe225,26. The components located at lower binding energies by ~0.4 eV relative to the 2H phase could be assigned to the 1T′ phase5,25. The Te/Mo ratio is 2.1–2.3 for both 2H and 1T′ phases, indicating the formation of nearly stoichiometric MoTe2 films. The existence of the 1T′ phase was also confirmed by the Raman spectroscopy (see Supplementary Fig. 2). We also carried out the growth experiments at substrate temperatures of 200 and 400 C and obtained the almost same RHEED and XPS results from the as-grown films.

After the sample was annealed at 600 C under the Te flux of 3 × 10−7 Torr, as shown in Fig. 1c, the integer-order streaks became sharp and intense, while fractional-order streaks disappeared. This means that the majority of the 1T′-MoTe2 phase transforms into the 2H phase. Figure 2b shows Mo 3d and Te 3d spectra measured from the annealed MoTe2 sample. Both Mo 3d and Te 3d spectra show single components of Mo(2H) and Te(2H). Their positions are the same as those of the 2H phase in the as-grown sample [Fig. 2a], and the Te(2H)/Mo(2H) ratio remains unchanged (2.1–2.2). As we will show later, the relatively high Te flux (2–3 × 10−7 Torr) is an important factor in the formation of the 2H phase.

Figure 1d shows a typical low-energy electron diffraction (LEED) pattern taken from the annealed sample. In addition to the diffraction spots from MoTe2(0001) and GaAs(111)B, satellite spots surrounding the integer-order spot of GaAs are observed. As can be seen in the schematic drawing of the LEED pattern [Fig. 1e], the satellite spots could be ascribed to the double diffraction of electrons at the MoTe2 film and the GaAs(111)B substrate below (see Supplementary Fig. 3). We note that satellite reflections are also observed in RHEED patterns [Fig. 1c].

A typical STM image of the annealed MoTe2 film is shown in Fig. 1f. While the MoTe2 monolayer is nearly completed, the nucleation of the second layer islands (~0.7 nm in height) is locally observed, which is due to the kinetics of the MBE process. The MoTe2 film has a root-mean-square roughness value of 0.16 nm and is more uniform and continuous as compared to MBE-grown MoSe212,16,17,21,23 and WSe223.

Twisted and non-twisted MoTe2 domains

Figure 3a shows a magnified image of the MoTe2 film: bright features showing hexagonal close-packed structure are clearly seen, which corresponds to a Moiré pattern due to the lattice mismatch between MoTe2 and GaAs. The peak-to-peak corrugation was found to be ~0.08 nm [see inset in Fig. 3a]. The spacing of bright features is ~2.9 nm, in good agreement with the value estimated from the in-plane lattice constants of MoTe2 (0.3519 nm) and GaAs (0.3998 nm). Carefully observing the image, we found that the bright features are not exactly aligned in the same directions: there exist domains with slightly different azimuthal orientations. In the upper left region, the hexagonal protrusions are aligned along the [11$$\overline{2}$$0] direction, while domains in the right and center regions are rotated by +6.6 and −6.6, respectively.

When the GaAs(111)B and MoTe2(0001) lattices are perfectly aligned, as shown in Fig. 3b, the (8 × 8) MoTe2 supercell commensurately located on the (7 × 7) GaAs(111)B, which reduces the effective lattice mismatch from −12 to 0.59%. The observed Moiré periodicity of 2.9 nm is close to seven times the lattice constant of GaAs (2.8 nm) and eight times the lattice constant of MoTe2 (2.8 nm) [Fig. 3d]. The lattice mismatch is further reduced to −0.39% by rotating the MoTe2 lattice by ±0.77. As shown in Figs. 3c, e, this epitaxial geometry results in the rotation of the Moiré pattern by ±6.6 from the [11$$\overline{2}$$0] direction, in good agreement with the observed STM features [Fig. 3a]. The resultant Moiré spacing for the rotated geometry [Fig. 3e] is 3.0 nm. We note that the rotation angle of 0.77 is so small that the rotation of MoTe2 domains could not be confirmed by the present LEED observations [Fig. 1d].

To investigate the stability of the epitaxial geometries observed in Fig. 3, we performed theoretical calculations. Seventeen structure models were studied, and the results for eight models, for which the lattice mismatch is less than 1% and the supercell size is 2.5–3.3 nm, are summarized in Table 1. We note that the majorities of the other nine models have smaller adsorption energy than the models shown in Table 1. The MoTe2 layer and GaAs(111)B surface in model C(H) has the same orientation, i.e., the twist angle (φ) between the two lattices is 0, while the others have a twist angle ranging from 0.769 to 21.787. As shown in Fig. 3a, most of the surface is covered with the two types of Moiré patterns, which correspond to models C and F. We found that these two models have a large adsorption energy of 20.49 and 20.44 meV Å−2, that explains the experiments well.

As shown in Table 1, the adsorption energies for models B and E are comparable to the experimentally observed two models (C and F). For those four models, both the twist angle (<2.7) and the lattice mismatch (<0.6%) are quite small. Thus, it is likely that both a small mismatch and a small twist angle are necessary for the energetical stability. As for a twist angle, the previous study has shown that the adsorption energy rapidly decreases, as the twist angle of TMDC heterobilayer increases27, being consistent with the present calculations (see also Supplementary Fig. 4). However, model E (φ = ~1.7) was observed only in a limited region, as can be seen in Supplementary Fig. 5, and model B (φ = ~2.7) was not observed. Since the dominant two orientations of C and F have smaller twist angles of φ = 0 and φ = ~0.8, respectively, it is suggested that the stacking geometry with a small twist angle is preferable, irrespective of the adsorption energy. While no definitive answer is available at this stage, we have some hypotheses: (i) a boundary between areas with a largely different twist angle is unstable and (ii) a quite small twist angle is preferable during the growth process.

In-plane rotational domains twisted by a specific angle have been found in the heteroepitaxy of layered materials28,29,30, which provide a better coincidence of the two lattices at the interface. On the other hand, we have reported that in the MoSe2/GaAs system, the directions of MoSe2 and GaAs with the highest linear atomic densities are aligned with each other and that the degree of the coincidence of lattice sites is less important in determining the epitaxial relationship21. The present results show that epitaxial MoTe2 films with and without the rotation are formed, suggesting that the two mechanisms compete with each other in the MoTe2/GaAs(111)B heteroepitaxy.

STM images similar to that in Fig. 3a have been observed on MoS214,30,31,32,33 and MoSe214,33, on Au(111) and have been interpreted as a Moiré pattern originating from the lattice mismatch between the film and substrate. On the other hand, another type of hexagonal pattern consisting of wagon-wheel-like structures has been commonly observed on MBE-grown MoSe28,10,15,34 and MoTe2 films11,18. Such wagon-wheel-like patterns are usually skewed and could not be explained by the Moiré effect, but reflect electronic states at the boundaries of inversion domains. As we will show below, STM images of MoTe2/GaAs show the two types of hexagonal patterns with different origins depending on the bias voltage.

Figure 4a, b show STM images taken from the same area with the bias voltages of −2.5 and −1.2 V, respectively. The STM image taken at −2.5 V [Fig. 4a] shows a periodic array of bright protrusions, similarly to the case for Fig. 3a, corresponding to the Moiré pattern. On the other hand, at −1.2 V, wagon-wheel-like patterns, that are distinct from the Moiré patterns in Fig. 4a, are clearly visible, indicating the existence of inversion domains in the monolayer MoTe2 film on GaAs(111)B. The existence of inversion domains is also evidenced by LEED observations (see Supplementary Fig. 6). Comparing the two images, we found there is no correlation between the two patterns. The circle A1 (A2) is located in the triangular center (the twin lines) in Fig. 4b, while both circles are at the center of bright features in Fig. 4a. Similarly, the position of the triangular center B2 in Fig. 4b corresponds to the center of bright features in Fig. 4a, and B1 is located between the bright features. Thus, it is likely that the observed Moiré patterns are predominantly topographic in origin, and are not related to the electronic states of inversion domain boundaries.

Stability of MoTe2 under Te-deficient conditions

When the MoTe2 film grown with the Te flux of 3 × 10−7 Torr was annealed at 600 C under the lower Te flux of 1 × 10−8 Torr, as shown in Fig. 2c, additional components, Mo(X) and Te(X), are observed in the Mo 3d and Te 3d spectra at binding energies lowered by ~0.5 eV. MoTe2 has two well-known structure phases of semiconducting 2H and metallic 1T′, as mentioned earlier, and the peak positions of Mo 3d and Te 3d for the 1T′-MoTe2 are lower by ~0.4 eV than those of the 2H phase5,25. It has been reported that the thermally driven transition from the 2H to the 1T′ phase begins at temperatures higher than 500 C, and the transition temperature is lowered under the Te-deficient conditions24. Thus, at first sight, one may attribute the appearance of component X to the structure change from 2H to 1T′. This explanation, however, could be ruled out, because the Te/Mo intensity ratio for component X is 1.1–1.2, which is incompatible with the chemical composition of MoTe2.

Figure 5a, b show STM images of monolayer MoTe2 films. The images in Fig. 5a, b were observed from the samples after the annealing with the Te fluxes of 3 × 10−7 Torr and 1 × 10−8 Torr, respectively. For the sample annealed under the Te flux of 1 × 10−8 Torr b, nanowire (NW) -like structures are formed at the edges of holes, those are not identified in Fig. 5a. The formation of a similar NW-like structure has been found in previous STM observations9,35 and has been attributed to the transformation from 2H-MoTe2 driven by thermal annealing under vacuum9,26,35. Detailed analysis based on scanning transmission electron microscopy observations revealed that the NW-like structure has the formula Mo6Te626,35. The reported binding energies of Mo 3d and Te 3d peaks for the Mo6Te6 NWs are lower by ~0.4 eV than those of 2H-MoTe226, in good agreement with the present results [Fig. 2c]. In addition, the stoichiometry of Mo6Te6 is consistent with the observed Te/Mo ratio of 1.1–1.2. Thus, it is likely that the annealing of MoTe2 under the Te-deficient conditions causes the desorption of volatile Te, leading to the formation of Mo6Te6 NWs.

The formation of an NW-like structure strongly depends on the annealing temperature and the Te flux. We confirmed that the NW-like structure is formed after the annealing at 600 C under the Te flux lower than 1 × 10−7 Torr. On the other hand, at a lower annealing temperature, e.g., 530 C, the formation of an NW-like structure was suppressed even under the lower Te flux of 3 × 10−8 Torr. In addition, as we will show below, the Te flux during the growth at 300 C is also important in obtaining the pure 2H-MoTe2 phase. Shown in Fig. 5c is the STM image of the MoTe2 film grown with the lower Te flux of 1 × 10−8 Torr and annealed under the higher Te flux of 3 × 10−7 Torr: The formation of NW-like structures could not be completely suppressed. On the other hand, when the sample was grown and annealed under the Te flux of 1 × 10−8 Torr, the whole surface is covered with elongated and meandering strands [Fig. 5d], suggesting that the formation of NW-like structures is further facilitated.

In summary, we have studied the structure and morphology of MBE-grown MoTe2 on the Se-treated GaAs(111)B substrate. The as-grown MoTe2 film shows the coexistence of the 2H and 1T’ phases, and the MoTe2 films annealed under the Te-rich conditions are mainly composed of the 2H phase. In addition to highly-oriented MoTe2 domains with GaAs(111)B lattice, slightly (0.77) rotated domains are formed to accommodate the large lattice mismatch. Both types of domains are found to be stable by first-principles calculations. The 2H-MoTe2 film evolves into the NW-like structure under the Te-deficient conditions, but the transition from 2H to 1T’ was not observed in the present experiments. While, because of the limitation of our STM setup, the presented experimental data were obtained from small size samples, it is possible, in principle, to grow MoTe2 films over the entire 2-in. GaAs wafer. Thus, we believe that the present results will offer a strategy to realize the large-scale synthesis of monolayer TMDs with a precisely controlled phase and a new perspective for a better understanding of the MBE growth of TMDs.

Methods

Experiments

The growth experiments were carried out using a multi-chamber ultra-high vacuum (UHV) system consisting of MBE chambers for the growth of GaAs and MoTe221. The system is equipped with STM and XPS apparatuses for online characterization36. The Si-doped GaAs(111)B wafers (13 mm × 10 mm in size) were used as substrates. The clean GaAs(111)B-(2 × 2) surfaces were first prepared by MBE21,37. The samples were then transferred via UHV transfer modules to another MBE chamber for the Se treatment and the growth of MoTe2. The Se-treated GaAs(111)B surface is suitable for TMD growth because the surface is electrically passivated37. The MoTe2 films were grown on the Se-treated GaAs(111)B-(1 × 1) substrates using an electron-beam evaporator (Tectra e-flux2) for Mo, and a conventional Knudsen cell for Te. The beam-equivalent pressures (BEPs) of Mo and Te were measured using a beam flux monitor at the sample position. The BEP of Mo was controlled at 1 × 10−10 Torr and that for Te was varied in the range of 1 × 10−8 Torr ~3 × 10−7 Torr. Similar to the case for MoSe2/GaAs heteroepitaxy, the substrate temperature during the growth was kept at 200–400 C. The growth rates of MoTe2 on GaAs(111)B were 0.01–0.02 ML per minute, as deduced from RHEED observations. After the growth of MoTe2, the samples were annealed at 600 C under the Te molecular beam until the RHEED patterns no longer changed, the annealing duration being typically 20 min.

The growth processes were monitored by in-situ RHEED with electron-beam energy of 15 keV. The samples were also characterized by online STM (Omicron Micro STM), XPS (Surface Science Instrument M-Probe), LEED (OCI LEED 600), and Raman spectroscopy. All the STM images were collected at room temperature (RT) in the constant current mode with a tunneling current of 0.1 nA. XPS measurements were performed using monochromatic Al Kα radiation (1486.6 eV). Photoelectrons were detected at an angle of 35 from the surface. The LEED patterns presented in this paper were recorded at RT using primary electron-beam energy of 70 eV. The Raman spectra were recorded using laser excitation at 532 nm in a commercial Raman microscope (Renishaw InVia).

Calculations

Theoretical calculations were performed by using the PHASE/0 code38, which is based on density functional theory (DFT)39 and pseudo-potential schemes 40,41 with plane-wave basis sets42. For the exchange-correlation term, the PBE form was used43. For the van der Waals interactions, the DFT-D2 method was applied44. The cut-off energies for the wavefunctions and charge density were 25 Ry and 225 Ry, respectively. The number of k points sampled in the Brillouin zone was more than 5 × 5 per GaAs(111)-(1 × 1) surface unit cell. We confirmed by using the cut-off energy of 36 Ry and the doubled k-point mesh that the relative adsorption energy change is less than 0.2 meV Å−2. All the models were optimized to meet the force criterion of 0.01 eV Å−1. Each GaAs(111) slab consists of three bilayers and the thickness of a vacuum region is 1.6 nm. The top layer of a GaAs(111)B slab is covered with Se atoms based on the experimental conditions and the bottom layer is terminated by fictitious H atoms.

The calculated lattice constant for the hexagonal MoTe2 single layer is 0.3524 nm, while that for a hexagonal GaAs(111) is 0.3948 nm. In this study, the change in the lattice mismatch of MoTe2 grown on GaAs(111)B is quite essential to see the stability of MoTe2/GaAs(111)B heterostructures. Thus the lattice constant for GaAs is set to 0.4004 nm to follow the experimental ratio of 1.136 for MoTe2 and GaAs. This treatment is justified, because there is no chemical bond between the MoTe2 layer and GaAs slab, and the van der Waals interaction and the lattice mismatch of MoTe2 mainly determine the stability of the systems. Adsorption energy, Ead is defined as

$${E}_{{{{\rm{ad}}}}}=({E}^{{{{{\rm{MoTe}}}}}_{{{{\rm{2}}}}}}+{E}^{{{{\rm{GaAs}}}}}-{E}^{{{{{\rm{MoTe}}}}}{_{{{\rm{2}}}}/{{{{GaAs}}}}}})/S,$$
(1)

where $${E}^{{{{{\rm{MoTe}}}}}_{{{{\rm{2}}}}}}$$, EGaAs, and $${E}^{{{{{\rm{MoTe}}}}}{_{{{\rm{2}}}}/{{{{GaAs}}}}}}$$ are the calculated total energies for MoTe2 single layer, GaAs(111)B slab, and MoTe2/GaAs(111)B. S is the area of a superstructure. The sizes of the MoTe2 single layer and GaAs(111)B slab are indicated by (m × m) and (n × n) in Table 1, respectively. In this definition, the larger Ead is, the more stable the system is.