II. IN SITU GROWTH OF TMD (MoTe₂/MoSe₂) SL

This is the first demonstration of in situ growth of 2D material SLs. As shown in Fig. 2, we demonstrate an MBE MoSe₂/MoTe₂ SL on GaAs (111)B substrates. GaAs substrate was chosen as high quality epi ready GaAs substrates are commercially available and the surface can be made quasi-van der Waals employing a surface termination with chalcogen atoms. Replacement of surface arsenic with sulfur retaining 1 × 1 reconstruction has been observed in GaAs (111) immersed in NH₄S solution ^{Reference Sugiyama, Maeyama and Oshima44} and such sulfur terminated GaAs has been previously used for MBE growth of layered materials like MoSe₂. ^{Reference Ueno, Shimada, Saiki and Koma45} In this work, we deoxidize GaAs in situ (at a thermocouple temperature of 730 °C) and then anneal under a Te flux for 5 min, followed by cool down to the growth temperature of 380 °C (thermocouple temperature). The effect of this tellurium treatment is shown in Fig. 3. When the surface is just deoxidized without Te treatment [Fig. 3(a)], the surface is heavily pitted with ∼10 nm deep triangular craters, which results in a broken RHEED pattern from the surface. When the surface treatment is performed [Fig. 3(b)], the surface becomes much smoother showing wider and shallower triangles, which are most likely GaAs/Te islands with atomic step heights as shown in Fig. 2(f). Correspondingly, the RHEED patterns become sharp and streaky as in Fig. 2(d). This shows that there are atomic motions on the surface assisted by Te atoms. Furthermore, the high angle annular dark field-scanning transmission electron microscope (HAADF-STEM) image in Fig. 2(f) shows the very top row of the As–Ga atomic pair of the GaAs substrate appears brighter than the As–Ga pairs below, strengthening the claim of Te replacing As on the GaAs surface.

FIG. 2. (a) Growth diagram of the MoTe₂/MoSe₂ SL on GaAs (111)B. (b) Raman from the SL (on GaAs as well as exfoliated onto SiO₂/Si) compared to Raman signal from CVT grown bulk MoSe₂ and bulk MoTe₂. (c) RHEED oscillation from a sample grown under the same condition but for a shorter duration for each period. Blue lines show where the chalcogen is switched and the dashed red line to the solid red line mark one period of the oscillation. (d) Evolution of RHEED signal from the sample shown in Fig. 2(c), RHEED streaks from GaAs are superimposed by the red line, MoTe₂ in yellow and MoSe₂ in green. (e) HAADF-STEM image of the SL and EELS map of Te M_4,5 edge, which clearly shows abruptness of different layers. (f) Higher magnification image of the red box in Fig. 2(e) showing how MoTe₂ climbs over an atomic step on GaAs.

FIG. 3. AFM images of (a) deoxidized GaAs surface without tellurium treatment, and (b) deoxidized GaAs surface followed by Te treatment.

As shown in the growth diagram in Fig. 2(a), once the substrate is cooled down to the growth temperature and stabilized, Mo and Te molecular beams are simultaneously incident on the substrate. The Mo molecular beam is from an e-beam source and Te and Se are from Kundsen cells. All the data presented in Fig. 2, except for the RHEED oscillations, are for a growth with 480 s for each period. The Mo flux is kept constant and the Te and Se fluxes are alternated. The key point drawn from Fig. 2(b) is that Raman from the SL is superimposition of Raman from both MoSe₂ and MoTe₂, which is expected but it is interesting to note that the Raman efficiency from MoTe₂ is much lower than that from MoSe₂. Also, as some of the Raman peaks were masked by peaks from the longitudinal optical (LO) and transverse optical (TO) phonons in GaAs, ^{Reference Abstreiter, Bauser, Fischer and Ploog46} the MoTe₂/MoSe₂ SL was exfoliated onto a SiO₂/Si substrate and Raman was carried out to confirm the peaks. Raman spectroscopy, measured using a 488 nm laser, an incident power of 633 µW and averaged over several scans, shows a MoSe₂ A_1g peak at 242.02 cm⁻¹ and E_2g peak at 286.3 cm⁻¹, which is consistent with the observations on MBE grown MoSe₂ on HOPG. ^{Reference Vishwanath, Liu, Rouvimov, Mende, Azcatl, McDonnell, Wallace, Feenstra, Furdyna, Jena and Grace Xing24} The E_2g peak seems to comprise of 2 very close peaks, one from MoSe₂ and one from MoTe₂. Figure 2(d) shows the RHEED evolution along the growth of the SL. It can be clearly seen that the RHEED streak spacing is inverse of the lattice spacing shown in Fig. 1(a). The ratio of the inverse of the RHEED streak spacing closely matches the ratio of reported lattice spacing. ^{Reference Böker, Severin, Müller, Janowitz, Manzke, Voß, Krüger, Mazur and Pollmann47} The ratio of the lattice spacing of (111) GaAs:MoTe₂:MoSe₂ from the literature is 3.997 (a_GaAs/√2):3.522:3.299 (where “a” is lattice constant of GaAs), which is equivalently 1:0.88:0.82. The ratio of the inverse of RHEED streak spacing of the Red, Yellow, and Green lines is 1:0.88:0.81, which is within the error margin given the RHEED patterns from the TMD layers are diffused.

It is also interesting to note that starting from the second MoTe₂ layer polycrystalline ring develops along with the streaky RHEED. This is correlated with the waviness in the MoTe₂ in the 2nd period of the SL and MoSe₂ in the 2nd period of the SL as seen from the cross-sectional HAADF-STEM image in Fig. 2(e). This polycrystallinity deteriorates RHEED oscillations, therefore, the RHEED oscillation presented in Fig. 2(c) is from a different growth under the same growth condition but with thinner layers, which was used to calibrate the growth rate. Since Mo is a refractory metal and the substrate temperature is low, the sticking coefficient of Mo can be safely assumed to be 1 at the surface thus the growth rate is limited by the Mo deposition rate. The RHEED oscillation period is taken to be the duration of one monolayer. From the 2nd and 3rd oscillation in Fig. 2(c), one oscillation period (duration between the dashed red line and the solid red line) can be estimated to be ∼160 s, i.e., a growth rate of 1 monolayer per 160 s. In the 2nd oscillation, it is taken as double of the duration between the minimum and maximum. In the 3rd oscillation, as the duration between the 2 minima. It is an approximation since, Mo flux from e-beam changes slightly between growths and during growth itself. Based on this observation, each half-period of the reported SL growth in Fig. 2(e) is carried out for 480 s, which corresponds to 3 monolayers. However, it is worth noting that in the region in Fig. 2(e) marked by an orange box we can see that there are 3 monolayers of MoTe₂ and 4 monolayers of MoSe₂ in the 1st period, and there are 4 layers of MoTe₂ and MoSe₂ in the 2nd period. Together with the substrate surface features of atomic heights, this suggests that observation of RHEED oscillation doesn't necessarily mean a full layer-by-layer growth. In fact, delay in the RHEED intensity and oscillation magnitude is a result of cumulative interference from a roughening film. In spite of that during growth the Se and Te flux is about 3 orders magnitude higher than the Mo flux and there are no breaks between the Se pulse and Te pulse to remove excess Se or Te from the surface, very abrupt MoTe₂ and MoSe₂ interfaces have been achieved. This is evidenced by the atomic resolution electron energy loss spectroscopy (EELS) map of the Te M_4,5 edge of the SL. The sharpness of the map is demonstrated by the fact that the Te signal takes a bend over the GaAs step where the layers bend. Figure 2(f) shows how the MoTe₂ layers climb over an atomic step on GaAs and that there is no chemical bond between the 2D material and the 3D substrate treated with Te.

III. IN SITU GROWTH OF TMD/QUASI-2D CRYSTAL (MoX₂/Bi₂X₃) SL

Diced sapphire substrates were first cleaned sequentially in chloroform (30 min), acetone (15 min), methanol (5 min), and hydrofluoric acid (HF) (10%, 2 min). The substrate was then loaded into the MBE chamber, heated to 800 °C and held for 30 min, then cooled to 200 °C. Once the temperature was stabilized the first 2–3 monolayers of Bi₂X₃ (X denotes Se or Te) was deposited. RHEED from sapphire is observed to disappear; when the substrate is heated gradually to the growth condition of 340 °C (avoiding any overshoot), the Bi₂X₃ RHEED pattern appears. This is similar to the growth reported for Bi₂X₃ on GaAs (001). ^{Reference Liu, Smith, Cao, Chen, Fan, Zhang, Pimpinella, Dobrowolska and Furdyna48} Subsequently, the Bi₂X₃ and MoX₂ SL was grown. The reason for this increase in Te flux before MoTe₂ growth on Bi₂Te₃ is presented elsewhere. ^{Reference Vishwanath, Liu, Rouvimov, Azcatl, Wallace, Furdyna, Jena and Grace Xing49}

Figures 4(b) and 5(b) shows that Raman from the SL is a superimposition from individual components of respective SLs, Bi₂Te₃/MoTe₂ and Bi₂Se₃/MoSe₂. Again, even though in both SLs the thickness of MoX₂ is almost equal, the Raman signal from MoTe₂ is much lower and broader than from MoSe₂, which is consistent with the observation in the MoTe₂/MoSe₂ SL (Fig. 2). The MoTe₂ A_1g peak at ∼171 cm⁻¹ and E_2g at ∼230 cm⁻¹ (obtained by Lorentz fitting to the broad peaks) can be better resolved due to the absence of MoSe₂ Raman peaks in the neighborhood, which are close to the values from bulk MoTe₂ at 174.4 and 235.8 cm⁻¹. 1T-MoTe₂ has been reported to have very different peak positions from 2H-MoTe₂, ^{Reference Park, Yun, Kim, Park, Chae, An, Kim, Kim, Kim and Lee36} therefore, this observed shift and broadening in MoTe₂ peaks are more likely due to the small grain size in the MBE grown MoTe₂ as compared to the bulk. We assume there is no systematic zero error since the E_2g peak of Bi₂Te₃ is at 103.8 cm⁻¹ and the ${\rm{A}}_{1{\rm{g}}}^2$ peak is at 134.7 cm⁻¹, consistent with the reported values for bulk Bi₂Te₃. ^{Reference He, Wang, Qiu, Delaney, Beck, Kidd, Chancey and Gao50,Reference Richter, Kohler and Becker51} The bulk value measured by us is consistent with the reported value for bulk 2H-MoTe₂. ^{Reference Yamamoto, Wang, Ni, Lin, Li, Aikawa, Bin Jian, Ueno, Wakabayashi and Tsukagoshi52} RHEED in Fig. 4(c) is in line with the expectation that Bi₂Te₃ has a wider lattice spacing than MoTe₂ hence a narrower spacing in the reciprocal space. It is worth noting that RHEED from MoTe₂ is much more diffused than that from Bi₂Te₃, pointing toward a greater disorder and smaller grains in MoTe₂. This is similar to the small grains reported in MBE MoSe₂ on graphene, ^{Reference Vishwanath, Liu, Rouvimov, Mende, Azcatl, McDonnell, Wallace, Feenstra, Furdyna, Jena and Grace Xing24,Reference Ugeda, Bradley, Shi, da Jornada, Zhang, Qiu, Ruan, Mo, Hussain, Shen, Wang, Louie and Crommie25} owing to the low adatom surface diffusivity of refractory metals. Bi₂Te₃ RHEED sharpness recovers during the subsequent growth on MoTe₂, adding strength to the above claim.

FIG. 4. (a) Growth diagram of the Bi₂Te₃/MoTe₂ SL, (b) Raman spectrum from the SL compared to that from bulk MoTe₂ and the *reported spectrum ^{Reference He, Wang, Qiu, Delaney, Beck, Kidd, Chancey and Gao50} for bulk Bi₂Te₃, (c) RHEED showing that MoTe₂ has a smaller lattice constant than Bi₂Te₃ hence the reverse in the reciprocal space, (d) HAADF-STEM image of the SL showing the abruptness of different materials, (e) higher magnification image of the red box in Fig. 4(d) showing that the lattice constant of the 2 materials is consistent with that of the bulk and how a MoTe₂ layer stitches to a partial layer of Bi₂Te₃.

FIG. 5. (a) Growth diagram of the Bi₂Se₃/MoSe₂ SL, (b) Raman spectra from the SL compared to that from bulk MoSe₂ and MBE grown Bi₂Se₃ on sapphire, (c) HAADF-STEM image of the SL, (d) EDX line scan of the SL showing abruptness of the Mo and Bi species, the widening of the edge is due to sample motion during imaging, (e) high-resolution transmission electron microscopy (HRTEM) of MoSe₂ grown at 340 °C on Bi₂Se₃, the same growth condition as the SL, (f) HRTEM of MoSe₂ grown at 400 °C on Bi₂Se₃, (g) AFM of Bi₂Se₃ grown on sapphire at 340 °C, the same growth condition as the first layer in the SL, (h) AFM of MoSe₂ grown at 340 °C [sample in Fig. 5(e)], and (i) AFM of MoSe₂ grown at 400 °C [sample in Fig. 5(f)].

The most interesting observation in transmission electron microscopy (TEM) imaging of this SL sample is that a lateral junction between a quintuple layer Bi₂Te₃ and a trilayer MoTe₂ is seen, shown in Fig. 4(e). This is the first ever observation of a lateral junction between a quasi-2D layer and a 2D layer. We see that the top 3 layers of the Quintuple (Te–Bi–Te) transition into (Te–Mo–Te). Such a junction might open paths for novel devices not envisioned before involving structurally different layered materials.

A substrate temperature of 340 °C was chosen to grow the Bi₂X₃/MoX₂ SL samples presented in Figs. 4 and 5(a)–5(d) since the growth of Bi₂X₃ was optimized near 340 °C. In our previous work MBE growth of MoSe₂ was carried out at a substrate temperature of 400 °C. To explore the effects of the substrate temperature, growths of MoSe₂ at 340 and 400 °C were carried out on Bi₂Se₃, which was grown on sapphire at 340 °C as described previously. In case of MoSe₂ growth at 400 °C, the first layer of MoSe₂ was grown at 340 °C and then substrate was heated to 400 °C for subsequent layers to avoid any dissociation of Bi₂Se₃. Atomic force microscopy (AFM) imaging of the as grown Bi₂Se₃ at 340 °C [Fig. 5(g)] shows triangular domains consistent with other reports. ^{Reference Liu, Smith, Cao, Chen, Fan, Zhang, Pimpinella, Dobrowolska and Furdyna48} When AFM images from MoSe₂ growth at 340 °C [Fig. 5(h)] are contrasted to those at 400 °C [Fig. 5(i)], significant differences in surface morphology are observed. The MoSe₂ at 340 °C follows the contours of the underlying Bi₂Se₃ and hence any pits in Bi₂Se₃ are also seen after growth of MoSe₂. But when the growth is done at 400 °C we observe features similar to that observed on thick MBE MoSe₂ on HOPG, ^{Reference Vishwanath, Liu, Rouvimov, Mende, Azcatl, McDonnell, Wallace, Feenstra, Furdyna, Jena and Grace Xing24} i.e., tall protrusions enclosing large smooth regions. TEM imaging of the 2 samples [Figs. 5(e) and 5(f)] reveals that MoSe₂ grown at 400 °C has a greater waviness but the interface between Bi₂Se₃ and MoSe₂ is sharp in both cases. Further investigations are necessary to understand the exact mechanisms behind these observations.

Finally, energy dispersive x-ray spectrum (EDX) line scan shown in Fig. 5(d) is consistent with the expected 40% Bi in Bi₂Se₃ and 33.3% Mo expected in MoSe₂. Local variations in the height of Bi₂Se₃ triangular grain along the width of the TEM sample and slight sample drift during the scan likely contribute to the overlap in the Bi and Mo signals in contrast to the sharp interface in HAADF-STEM [Fig. 5(c)].

IV. IN SITU GROWTH OF HETEROSTRUCTURES INVOLVING MORE THAN 3 MATERIALS

We have grown layered crystals (MoSe₂, MoTe₂, SnSe₂, WSe₂) on a 3D substrate (GaAs, CaF₂) and quasi-2D crystals (Bi₂Se₃, Bi₂Se₃) with very small band gaps (∼200 meV), and growth of quasi-2D materials like SnSe (∼0.86 eV bulk indirect band gap) ^{Reference He, Wang, Qiu, Delaney, Beck, Kidd, Chancey and Gao50} and GaSe on a 2D substrate is equally interesting. Of particular interest is the Sn–Se system, because SnSe has been shown to have exceptional thermoelectric properties ^{Reference Zhao, Lo, Zhang, Sun, Tan, Uher, Wolverton, Dravid and Kanatzidis53} and SnSe₂ is a layered material with high work function ideal for tunnel junctions. Monolayer and few layer SnSe has only been demonstrated to date using solution synthesis. ^{Reference Li, Chen, Hu, Wang, Zhang, Chen and Wang54} Taking guidelines from previously reported growth conditions for MBE growth of SnSe and SnSe₂ on layered mateials ^{Reference Shimada, Ohuchi and Koma55,Reference Schlaf, Louder, Lang, Pettenkofer, Jaegermann, Nebesny, Lee, Parkinson and Armstrong56} we grew a heterostructure involving 3 materials: Bi₂Se₃, MoSe₂, and SnSe₂. The corresponding growth diagram is shown in Fig. 6(a). Raman measurements show superimposition of expected layered materials. The shift of SnSe₂ peaks is likely due to difference in stacking order of bulk SnSe₂ and MBE SnSe₂. ^{Reference Mead and Irwin57–Reference Vishwanath, Liu, Rouvimov, Furdyna, Jena and Grace Xing59} The cross-sectional TEM [Fig. 6(c)] image reveals that majority of the deposited Sn–Se was in form of SnSe₂, but there were some regions of SnSe. What is very interesting is that a band of SnSe forms at the interface between MoSe₂ and SnSe₂. Our hypothesis is that even though the growth temperature was lowered from 340 to 240 °C for the growth of SnSe₂, the growth temperature of 240 °C wasn't stabilized for long enough, essentially resulting in a higher than desired growth temperature for the first few layers. Indeed we have consistently observed that SnSe grows at a higher temperature than SnSe₂. While a separate manuscript ^{Reference Vishwanath, Liu, Rouvimov, Furdyna, Jena and Grace Xing59} is under preparation to report the MBE growth of SnSe and SnSe₂, we present here two sets of growth. Samples in Fig. 6(d) (SnSe _x grown at 240 °C on MBE MoSe₂/Bi₂Se₃/Sapphire) and Fig. 6(e) [SnSe grown at 240 °C on exfoliated MoS₂/SiO₂ (300 nm)/Si] differ primarily by substrate alone. Samples in Figs. 6(e) and 6(f) [SnSe₂ grown at 200 °C on exfoliated WSe₂/SiO₂ (300 nm)/Si] differ primarily by growth temperature. Although using STEM imaging only SnSe is seen on exfoliated MoS2 in Fig. 6(e), inorder to see if there is SnSe₂ in other regions of the film, X-ray photoelectron spectroscopy (XPS) was performed on the sample. In agreement with the STEM imagining, the XPS spectra for Fig. 6(e) shows that majority of the as grown film is indeed comprised of the SnSe phase, at a binding energy of 54.3 eV in Se 3d and 485.5 eV in Sn 3d, which are consistent with the values reported in literature. ^{Reference Boscher, Carmalt, Palgrave and Parkin60} Yet, in addition to this phase, a small but significant amount of SnSe₂ was also detected, where the corresponding peaks appear at higher binding energy (0.6 eV for Se 3d and 0.8 eV for Sn 3d) than the SnSe peaks. In addition to the SnSe _x film, a signal from metallic Se was detected at a binding energy of 54.7 eV, suggesting that some selenium remained unreactive in a metallic form (Se⁰). Finally, the presence of surface oxides, SnO _x and SeO _x , were also detected in this sample, which could be generated due to air exposure prior to XPS measurements (Fig. 7).

FIG. 6. (a) Growth diagram of the Bi₂Se₃/MoSe₂/SnSe₂ heterostructure, (b) Raman spectrum from the heterostructure compared to those from MBE grown Bi₂Se₃ on sapphire, bulk MoSe₂ and bulk SnSe₂, (c) HAADF-STEM image of the heterostructure, (d) higher magnification image of the interface of SnSe₂ on MoSe₂, (e) HAADF-STEM image of another MBE sample showing that single phase SnSe can be grown on MoS₂, (f) HAADF-STEM image of yet another MBE sample showing that single phase SnSe₂ can be grown on WSe₂.

FIG. 7. XPS on sample in Fig. 6(e) (SnSe on exfoliated MoS₂ on SiO₂/Si) (a) Se 3d core level showing the presence of SnSe, SnSe₂, metallic Se, and SeO _x , (b) Sn 3d core level showing the presence of SnSe, SnSe₂, and SnO _x , (c and e) low intensity signals from Mo–S bonds are detected, however both Mo 3d and S 2p overlap with selenium related regions from SnSe _x complicating the deconvolution for MoS₂, and (d) the stoichiometry for SnSe and SnSe₂ are close to the expected values.

This shows that by optimizing and modeling parameters like growth temperature, interfacial thermal resistance, and the thermal conductivity of the various layers, it is possible to grow single phase SnSe _x (x = 1 or 2) or a mixed phase with varying concentrations of SnSe₂ and SnSe.

VI. INSTRUMENTATION DETAILS