Molecular beam epitaxy of the van der Waals heterostructure MoTe₂ on MoS₂: phase, thermal, and chemical stability

Mechanical exfoliation and stacking of different mono- to few layers of van der Waals materials on top of each other enables the creation of novel devices and materials [1–9]. The functionality of these materials also led to a renewed interest in directly growing van der Waals heterostructures by molecular beam epitaxy (MBE) [10]. Direct growth of these materials has the advantage of being a scalable process and enabling materials fabrication on a wafer scale. MBE of van der Waals materials has been initially explored in the 1990s [11–15]. During this time the benefits of relaxed lattice matching conditions due to the weak interlayer interactions appeared promising for growing heterostructures with a variety of properties. These studies also resulted in a number of fundamental materials characterization of such grown films. For instance, a transition from an indirect to direct gap for monolayer materials has been already observed in angle resolved photoemission (ARPES) in 2001 on such grown monolayers [16]. In order to make devices, or to characterize novel properties induced by interfaces, a better understanding or even just achieving the growth of various van der Waals materials is required. While growth by chemical vapor deposition (CVD) has accomplished the fabrication of heterostructures of different van der Waals materials [17, 18]. CVD growth is currently difficult to control and typically the reported results only represent a few well-defined islands over the entire wafer. Nevertheless these local heterostructures are very useful for fundamental materials characterization and prototype device characterization. Physical vapor growth or (MBE) is an alternative approach for van der Waals epitaxy [19, 20]. MBE potentially allows the fabrication of heterostructures on wafer-scales, which also enables surface and interface characterization by non-local techniques like ARPES. Generally, MBE has the advantage over other growth methods by allowing control over elemental deposition rates and easy switching from one material to another, thus in principle enabling precise control for heterostructure growth. A challenge for many van der Waals materials, such as transition metal dichalcogenides (TMDs), is that the deposited materials are volatile either in elemental form or as small molecules. This causes a low sticking coefficient on the growth-substrate, especially on inert van der Waals substrates. As a consequence, MBE generally requires low growth temperatures and chalcogen rich conditions.

In this communication we examine the growth and properties of (sub)monolayer MoTe₂ on bulk MoS₂ substrates. α-MoTe₂ has a sub-1 eV band gap and thus is particularly interesting for infrared optical applications. MoTe₂ has two well-known structural phases [21]. Both phases are van der Waals materials, i.e. structures with strong in-plane covalent bonding and weak interlayer interactions. The α-phase of MoTe₂ is isostructural to MoS₂ or MoSe₂, i.e. it forms the hexagonal 2H-MoTe₂ structure, with an in-plane lattice constant of a = 3.519 Å. This α-phase is semiconducting with an indirect band gap of 0.88 eV for the bulk material [22]. The β-phase, on the other hand, is a semimetal and has a monoclinic structure. The rectangular unit-cell of the monolayer has lattice constants of a = 6.330 Å and b = 3.469 Å. In the β-phase the Mo-atoms are displaced from the center-plane of the monolayer, giving rise to a buckled monolayer structure. Furthermore, the Mo-atoms form quasi-one dimensional zigzag chains along the b-direction. The reduced Mo–Mo separation along this chain gives rise to increased metal-metal interactions, which also causes strongly anisotropic conductivity. Figure 1, shows structural models for the two MoTe₂ polymorphs. Historically, the phase transition from the α- to the β-phase was reported to occur between 820 °C and 880 °C [23], with the higher transition temperature for slightly tellurium depleted material (MoTe_1.90). In a very recent paper the phase transition was revisited and much lower phase transition temperatures were found [24]. In this recent study the phase transition temperature was found to be significantly lower and dependent on small variations in the composition. For slightly Te-deficient materials the transition already started at ∼500 °C and for Te-enriched phase the α-phase was found to be stable up to 750 °C. Once transformed into the β-phase, this structure remains metastable to low-temperatures without transforming back into the α-phase, allowing its characterization at or below room temperature [25].

Zoom In Zoom Out Reset image size

Semiconducting to metallic phase transitions are also known to exist for other TMDs. However in these cases the structural transitions are induced by electron-donation. For instance in MoS₂ a transition from the stable 2H or 3R structure to so-called 2T or 2T' structure is induced by Li-doping (Li is used in chemical exfoliation of these materials and thus metallic phases are often obtained in this process) [26–28]. This structural transition is similar to that of MoTe₂ with respect that it increases Mo–Mo interactions in one crystallographic direction. Recently, it has been suggested to use this phase transition in MoS₂ to pattern a MoS₂ monolayer with metallic and semiconducting regions by inducing the phase transition locally [29]. Similar proposals may be made for MoTe₂, where the metallic phase could be formed, for example, by local heating, rather than chemical charge donation. However, for achieving this on a wafer scale, uniform growth of α-MoTe₂ is a prerequisite. Here we study the growth of MoTe₂ on MoS₂ substrates. We find that the obtained phase is strongly dependent on the growth conditions, especially the growth rate, with the desired α-MoTe₂ phase only obtainable at extremely slow growth (monolayer in hours) and under tellurium rich conditions. Also, tellurium is easily lost from the film by vacuum annealing, resulting in a tellurium-deficient phase that remains planar but favors quasi-one dimensional morphologies. Importantly, this tellurium poor phase exhibits only a very small band gap and thus despite its fragmented morphology could still be used as an electric contact material. The structure of this phase may be related to the β-phase of MoTe_2, which also exhibits quasi-one dimensional structures due to the formation of Mo–Mo chain structures.

For the α-MoTe₂/MoS₂ interface we observe the formation of a moiré-structure due to the different lattice constants between the monolayer and the substrate. This moiré structure indicates that the MoTe₂ monolayer grows rotated by ∼56° relative to the MoS₂ substrate. Finally, we point out that the films we have grown are unstable in air and easily oxidize. However, the formed MoO₃TeO₂ maintains the 2D-shape of the grown islands and thus this may be a useful approach for the formation of an ultrathin dielectric oxide layer on MoS₂.

The experiments are conducted in a UHV system consisting of a preparation chamber for MBE of chalcogenides and a surface analysis chamber. The preparation chamber is equipped with a K-cell for tellurium and a mini e-beam evaporator for Mo evaporation. The sample manipulator allowed for radiative, or e-beam heating of the substrate, which was mounted on a Ta-sample plate. The MoS₂ substrate was a commercially available synthetically grown MoS₂-single crystal from 2D Semiconductors Inc. The crystal was cleaved in air and subsequently annealed in vacuum at 300 °C for 4 h prior to the film growth. The cleanliness of the sample was checked by x-ray photoemission spectroscopy (XPS) and no contamination was detected. Impurity levels below the detection limit of the XPS (∼1%–2% of a monolayer) can, however, not be excluded. For film growth the K-cell for Te-evaporation was stabilized at 240 °C for 10 min. This K-cell temperature gave approximately a deposition rate of 0.05 nm min⁻¹ determined by a quartz crystal monitor. The Mo deposition was measured with a built-in flux monitor and was kept constant during deposition. After stabilization of the sources the sample temperature was increased to growth temperature and subsequently the shutters of the deposition sources were opened and the sample rotated to face the deposition sources. The main interest in this study is samples grown on MoS₂, however, for calibration and determination of oxidation-states we also grew a few samples on graphite (HOPG) under the same growth conditions. This was mainly done to avoid overlap of the Mo peaks from the substrate and the grown MoTe₂ films. After completion of the growth the sample was transferred in situ to the analysis chamber with help of a magnetically coupled transfer arm. In the analysis chamber the sample manipulator was also equipped with a calibrated e-beam heater, which enabled heating of the sample to specific target temperatures by radiative or e-beam heating. The accuracy of the annealing temperature is estimated to ±40 K. The mu-metal analysis chamber was equipped with an Omicron Sphera-II electrostatic analyzer for photoemission spectroscopy. For XPS a Mg/Al dual anode x-ray source and for ultraviolet photoemission spectroscopy (UPS) a VUV He-discharge lamp was used. For the XPS measurements reported in this manuscript the Mg anode was used. The emitted photoelectrons were detected at an emission angle of 53° (measured from the surface normal) to increase surface sensitivity. For quantitative measurement of the MoTe₂ composition we used MoTe₂ grown on HOPG. The composition of the film was then determined from Mo/Te peak intensities taken the atomic sensitivity factors of the elements as well as the transmission function of our analyzer into account. For optimized preparation conditions we found 67 ± 3 at% Te and 33 ± 2 at% Mo in the grown films in agreement with MoTe₂. Also, post-air exposure analysis of the oxidation state of Mo and Te was done on HOPG, as well as for Te on MoS₂.

In addition, the analysis chamber also housed a room temperature scanning tunneling microscope (STM) (Omicron STM 1). Electrochemically etched W-tips were used. The STM was used for imaging and local I–V spectroscopy. dI/dV derivatives were obtained numerically. Scanning tunneling spectroscopy (STS) maps were acquired by taking I–V spectra at every data point and subsequently the measured tunneling current at these data points for specific bias voltages were plotted. STS maps enabled the identification of MoTe₂ phases.

3.1 . Direct MoTe₂ growth

Submonolayer of MoTe₂ is deposited by co-vapor deposition of Mo and tellurium on a heated MoS₂ substrate. The film morphology is very sensitive to the growth rate and the Mo/Te ratio. For the case of growing MoTe₂, where any excess Te re-evaporates from the surface, the growth rate is primarily determined by the Mo-flux. In this study we mainly modify the Mo-flux and thus a fast growth rate also corresponds to a higher Mo/Te flux ratio, while a slow growth also corresponds to Te-rich conditions. Growth temperature below 200 °C causes formation of 2D-islands with dendritic shape reminiscent of a fractal growth [30]. This is shown in figures 2(a) and (b). A dendritic or fractal growth is expected at low temperatures for which molecular species that attach to the island have not enough mobility to diffuse along the edges of the growing island to form an equilibrium island shape. Commonly one would expect that the dendritic growth of such islands could be overcome by increasing the growth temperature in order to increase diffusion and thus enable the formation of equilibrium island shapes. Unexpectedly, however, we observe a different growth behavior. With growth temperatures above ∼350 °C the islands become very elongated and form meandering islands, as shown in figure 2(c). Such a drastic change in growth is explained if the grown islands are of a different structural phase and/or composition. If such a different phase has strongly anisotropic edge energies, elongated islands may result as an equilibrium shape rather than compact island-shapes. Increasing the growth temperature does not only modify the surface diffusion but also decreases atomic/molecular residence times on the substrate. For example a preferential desorption of Te-species could result in a different composition of the grown film. Unfortunately, an exact measurement of the composition of the sub-monolayer films by XPS is made impossible because our substrate contains the same transition metal, i.e. Mo, as the film. Below, we show that annealing of α-MoTe₂ islands also show a transition into quasi one-dimensional strands with a concurrent loss of Te and thus we assign both morphologies to the same fundamental MoTe_2−x phase.

Zoom In Zoom Out Reset image size

In order to obtain mainly α-MoTe₂ the Mo-flux was reduced and the substrate temperature was held at T = 200 °C. The low Mo-flux resulted in a slow growth of ∼0.16 ML h⁻¹ and Te-rich growth conditions. Under these conditions, growth of more compact islands was obtained. STM images of the morphology of these films are shown in figure 3(a) as a function of growth time. Under these growth conditions the island density does not increase with deposition time indicating that the nucleation density is determined by either defects in the substrate (heterogeneous nucleation) or the island density is given by the diffusion length of molecular species necessary for the formation of a stable nuclei. The latter implies a long diffusion length of the order of the island separation of 50–100 nm. A long diffusion length before the formation of a critical nuclei also explains that second layer islands are not observed until islands reach a size of the order of ∼35 nm in diameter. A detailed analysis of the growth mode will require more systematic analysis at different growth temperatures and deposition rates and is beyond the scope of the current communication. STS, shown in figure 3(b) indicate that the center of these islands are semiconducting with a band gap of approximately 1.03 ± 0.02 eV, however, some metallic regions are present at their perimeter. Closer inspection of the metallic region with atomic-resolution imaging indicates that these regions are β-MoTe₂. The STM images shown in figure 3(c), show a rectangular unit-cell with dimensions matching the expected monoclinic unit-cell of β-MoTe₂. Furthermore, the images compare very well with previously reported STM images of bulk β-MoTe₂ single crystals [24]. The semiconducting regions are α-MoTe₂ as the hexagonal atomic-structure of the atomically resolved regions, also shown in figure 3(c), indicates. The α-MoTe₂ domains also exhibit a triangular superstructure. A similar structure was observed for MoSe₂ grown on MoS₂ and was associated with a moiré-structure due to the lattice-constant differences between the substrate and the monolayer film [13]. The observed 2.6 nm periodicity of the superstructure can be reproduced with the lattice constants of MoTe₂ and MoS₂ assuming that the two lattices are rotated by ∼56° with respect to each other, as discussed below. Here we define the rotational angle such that 0° rotation corresponds to exactly the same structure of two subsequent layers, i.e. in the typical 2H-structure of TMDs the rotation of subsequent layers would be 60° in this notation. The band gap determined by STS matches closely the value reported previously by optical measurements for monolayer MoTe₂ [31]. This indicates that at least part of the MoTe₂ film is similar to freestanding exfoliated MoTe₂ monolayers, however, the question if the band structure is modulated within the moiré unit cell, as has been proposed for similar van der Waals heterostructures [32], require more detailed and higher resolution STS studies. The sample preparation discussed here should enable such studies in the future.

Zoom In Zoom Out Reset image size

UPS measurements of the grown films show a coexistence of a metallic and semiconducting phase. The metallic contribution can be reduced by annealing to ∼500 °C and reduced even further by covering the surface with Te prior to the annealing step. After annealing, all the excess Te is desorbed but the UPS spectra are better defined possibly indicating a reduction of defects and association of the metallic features with tellurium deficiency in the monolayer. The valence band spectrum for the MoTe₂ film annealed to 500 °C can be found in figure 4(b) and shows the VBM of the α-MoTe₂ phase to 0.78 eV, which is in reasonable agreement to the value of 0.8 eV determined by STS (see figure 3).

Zoom In Zoom Out Reset image size

3.2 . Morphology change in post-growth annealing

3.3. Oxidation of MoTe₂ films by air-exposure

Exposing submonolayer MoTe₂ films to air results in an oxidation of the film. Ambient-air AFM images are shown in figure 5(a), which may be compared to STM images taken of the same film in UHV prior to air exposure, shown in figure 5(b). The island structure is maintained. Measuring the island height (see supporting information figure S3), however, reveals that island after air exposure are significantly taller (1.02 ± 0.06 nm) than expected for MoTe₂ (0.6 nm). Subsequent characterization of the film by XPS shows that the Te peak shifts by 3.42 eV from 572.71 eV for MoTe₂ to 576.13 eV, as shown in figure 5(c). The latter peak position corresponds to TeO₂. To evaluate the Mo-3d peak positions after air exposure without having to de-convolute the contribution from the MoS₂ substrate we also grew sub-monolayer of MoTe₂ on HOPG-graphite. In these measurements, shown in figure S2, we observe a Mo-3d peak at 227.70 eV for MoTe₂ and a shift to 232.36 eV after air exposure. The latter peak position is consistent with MoO₃. Thus XPS indicates that the grown submonolayer MoTe₂ films are unstable in air and oxidize. Expected volume expansion upon transformation of MoTe₂ into MoO₃ and TeO₂ would also explain the larger island height, measured by AFM, compared to the height expected for monolayer MoTe₂. While these films are unstable in ambient air, attempts to oxidize the MoTe₂ by exposing up to 10⁻⁶ Torr pure O₂ at temperatures up to 500 °C, or in 10 Torr of pure O₂ at room temperature did not result in oxidation of the film. Thus it is possible that in addition to O₂ other oxidants like water are important to cause the observed oxidation. To understand the oxidation process better more studies under controlled atmospheres are required.

Zoom In Zoom Out Reset image size

Epitaxial growth of α-MoTe₂ on MoS₂ has been achieved at 200 °C in a tellurium rich environment. Lower residence time of tellurium compared to Mo is suggested to be responsible for the formation of sub-stoichiometric films with higher annealing temperature and/or reduced tellurium flux. The stoichiometric α-MoTe₂ monolayer on MoS₂ substrate exhibits a moiré structure with a periodicity of 2.6 ± 0.1 nm. Using the lattice constants for MoS₂ and MoTe₂ of 0.316 nm and 0.352 nm, respectively, one obtains this moiré periodicity by rotating the two lattices by 56°, as shown in figure 6(a). Note that for a 2H-stacking alternating layers are rotated by 60° thus a 56° corresponds to a 4° rotation with respect to a 2H-stacking sequence. Only one moiré structure is observed in STM studies and under the relatively high annealing temperatures it is reasonable to assume that this rotation represents the lowest energy configuration of this system. Like for single TDCs, the 2H-stacking is the energetically preferred stacking also for TDCs heterostructures [3]. Because of the lattice mismatch between the two layers it is, however, not possible to have this stacking configuration throughout the monolayer. For a related system of a bilayer of WS₂/MoS₂ the adsorption energies was calculated (by artificially matching the lattice constants) for the favored 2H-structure and compared to the structure with the chalcogen atoms in adjacent layers on-top of each other, rather than in the three-fold hollow sites of the 2H-structure (see figure 6(b) for configuration). For this system an energy difference of 0.0744 eV/cell was calculated. Thus the total energy of a realistic system may be minimized by reducing the number of MoTe₂ unit cells for which the Te-atoms are located directly above the sulfur atoms of the MoS₂ substrate. We evaluated the fraction of such energetically unfavorable configuration as a function of relative rotation angle between a MoS₂ and MoTe₂ layer to find the rotation angle that minimizes the configuration with chalcogen atoms of the two layers at the same lateral position. To do this, the two chalcogen-lattices are simulated with cosinusoidal waves, with maxima representing atom positions, using the following expression [13]:

$$\begin{eqnarray*}&&F\left({\bf{r}},{{\bf{g}}}_{1}{\boldsymbol{,}}{{\bf{g}}}_{2}\right)=\displaystyle \frac{2}{9}\left\{\mathrm{cos}({{\bf{g}}}_{1}{}^{\Circle }{\bf{r}})+\;\mathrm{cos}({{\bf{g}}}_{2}{}^{\Circle }{\bf{r}})\right.\\ &&\quad \quad \quad \quad \quad \quad +\;\left.\mathrm{cos}\left[({{\bf{g}}}_{1}+{{\bf{g}}}_{2}){}^{\Circle }{\bf{r}}\right]+\displaystyle \frac{3}{2}\right\};\end{eqnarray*}$$

where ${{\bf{g}}}_{1}$ and ${{\bf{g}}}_{2}$ are the primitive reciprocal lattice vector. Forming the product of the two wavefunctions gives a representation of the overlap of the chalcogen atoms in the adjacent layers as shown in figure 6(c). To ignore regions with negligible or no overlap of chalcogen atoms we define a cut-off below which we set all values to 0. Summation of the remaining wavefunction intensity and plotting the variation of this intensity as a function of rotation angle of the two sublattices allows us to determine at which rotation angle the summed intensity is lowest and thus the chalcogen overlap is minimal. Such a plot is shown in figure 6(d). A minimum is observed for a rotation angle of 56.1° or 63.8°, which is within our experimental value for the rotation between MoTe₂ and MoS₂. This minimum is quite stable for a reasonable range of cut-off values between 0.1 and 0.8. Thus the simple argument of minimizing the number of chalcogen atoms that are directly on top of each other in adjacent layers may give a simple estimation of the rotational angle between transition metal dichalcogen heterostructures in thermodynamic equilibrium. It has to be noted though that this is a very simplistic view and more detailed simulations of the adsorption energies of different moiré structures should be applied, despite the challenges involved in dealing with such large unit-cells.

Zoom In Zoom Out Reset image size

Annealing of α-MoTe₂ islands above 500 °C causes a phase transformation that originates at the island edges. This temperature is close to the recently reported phase transition temperature for slightly Te-deficient bulk α-MoTe₂ to β-MoTe₂ [24]. Thus the observed transition at the edges of α-MoTe₂ to β-MoTe₂ is likely related to this phase transition. With further annealing, the film becomes strongly tellurium deficient and a MoTe_2−x phase is formed. This phase exhibits a strongly anisotropic structure consisting of short one-dimensional strands separated by holes in the layer. This structure is closely related to the structure that forms directly under low tellurium growth conditions and that results in long meandering strands. Since the high temperature β-MoTe₂ phase consists of one-dimensional Mo-zigzag chains we speculate that this structure may form the basic unit of these observed strands. However, unlike a contiguous β-MoTe₂ plane, we are observing only a few Mo-zigzag chains with tellurium-deficient edges. Regardless of the precise structure, the observation that long one-dimensional MoTe_2−x structures can be grown directly on a weakly interacting substrate like MoS₂ may make this an interesting structure for further investigation of quasi 1D materials. Although our studies have shown that the direct thermal transformation of monolayer semiconducting α-MoTe₂ into semimetallic β-MoTe₂ is possible, however, additional loss of Te at high annealing temperature in vacuum results in MoTe_2−x phase, which still exhibits an almost metallic (very small band gap semiconductor) electronic structure and also maintains a planar morphology on MoS₂. Thus even this tellurium deficient structure may still be used as a planar electrical contact to α -MoTe₂ monolayers. For stabilizing β-MoTe₂ our studies suggest that it is important to avoid loss of tellurium, which one may achieve by annealing in a sealed, tellurium rich environment.

Finally, the sub-monolayer films we have grown are not stable in air. Generally monolayer materials are more reactive than bulk materials. In our case a high density of step edges, metallic edges, and possibly other defects increases the reactivity [33]. Layer-by-layer oxidation of the surface of bulk TMDs has been reported previously using activated oxygen [34]. Simulations also showed for MoS₂ that oxidation is thermodynamically favored but kinetically limited which explains the air stability of MoS₂ and shows the important role defects play in the oxidation [35]. The surface oxidation of bulk TMDs or the oxidation of more reactive monolayers at the surface of a TMDs as shown here enables the formation of a conformal oxide layer. This oxide layer may be used subsequently as a seed for growth of an oxide-dielectric by, for example, atomic layer deposition [36]. Thus monolayer growth of a (defective) TMD on a less reactive TMD substrate or other layered materials like grapheme [37], is a potential approach for the growth of conformal oxide dielectrics, which are otherwise difficult to achieve.

Submonolayer growth of MoTe₂ on bulk MoS₂ substrates by MBE has been investigated by scanning tunneling microscopy and photoemission spectroscopy. For growth of α-MoTe₂ tellurium rich conditions and low growth temperatures are needed. However, the formation of minority β-MoTe₂ domains could not be avoided. The complete transformation of submonolayer α-MoTe₂ into β-MoTe₂ by vacuum annealing is prevented by tellurium desorption and formation of tellurium-deficient MoTe_2−x. The new tellurium-deficient MoTe_2−x forms elongated, quasi-1D structures, which may bear some resemblance to single strands of the anisotropic β-MoTe₂ phase. The grown α-MoTe₂ monolayer is rotated by ∼56° with respect to the MoS₂ substrate resulting in a moiré pattern with 2.6 nm periodicity. Finally the grown MoTe₂ layers are not air stable and readily oxidize to MoO₃ and TeO₂. The resulting oxide layer maintains, however, its two dimensional morphology and thus may be used as a seed layer for subsequent growth of oxide dielectric thin films.

Financial support from the National Science Foundation (DMR-1204924) is acknowledged.