As the silicon complementary metal–oxide–semiconductor (CMOS) technology scaling, driven by the Moore’s Law, is running out of steam, atomically thin semiconducting TMDs, together with their polymorphism, provide promising alternatives for next-generation electronic devices 1-4. Despite their huge potential and a decade of development, 2D TMD materials still remain an early-stage technology 4. One important reason is the lack of a large-scale synthesis method that can reliably produce high-quality 2D TMDs with controlled phase and layer number 5. At present, layered TMDs are typically synthesized by chemical vapor deposition (CVD) methods with a vapor-solid-solid (VSS) mode, in which solid precursors become vaporized, transported by gas flow, attach onto the solid substrates, then diffuse to bond with solid products 6,7. However, such a VSS mode can only supply a limited amount of vapor precursor species nonuniformly because of a significant vapor concentration drop along the direction of gas flow 8. This thermodynamically favors decomposition of the as-synthesized TMD compounds, and therefore drives defect creation during the CVD synthesis. As a result, high concentration of defects, and defect-stabilized metastable crystalline phases instead of pure defect-free phase are often obtained via VSS growth (Fig. 1a, right). This makes the phase control of monolayer even more challenging. Even worse, certain members of TMDs, especially for those with natural thermal instability, simply cannot be synthesized by the existing CVD methods. For instance, the CVD synthesis of monolayer 2H MoTe2 is still missing owing to the defect-triggered phase transition from 2H to 1T′ 2.
In order to address these challenges, here we present a new liquid-vapor (LV) synthesis mechanism and take full advantage of liquid phase of precursors to synthesize TMDs. Unlike the conventional VSS or other solid precursor-based CVD methods, liquid precursors are introduced as a reservoir, providing sufficient amounts of active species that significantly facilitate the synthesis reaction of TMDs and suppress the decomposition of the target TMD compounds. Even for TMDs with as-formed defects, the sufficient reactants provided in such a LV mode maximize the chance of defects repairing at high temperature. Therefore, the LV method is expected to produce high-quality 2D TMDs, and enable synthesis of TMDs with controlled phases that are unattainable via conventional VSS method.
We experimentally tested this LV method by using metal tellurides as a case study. 2D metal tellurides are highly attractive for topological physics and phase engineering due to their strong spin-orbit coupling and polymorphism 9-12. The phase versatility of metal tellurides can greatly facilitate phase engineering such as phase heterostructure 2,13 and phase transition 12, which are promising for device applications, including high-performance 2D transistors with ohmic contact, topological transistors, reconfigurable circuits and memory devices, etc. Despite their great potential, the low electronegativity of tellurium and the weak metal-telluride bonds make this family of 2D TMDs particularly vulnerable to the undesirable chemistry in the VSS mode of CVD processes. Moreover, the tendency of decomposition and oxidation often leads to very defective synthesis, and for many cases, even a complete failure. In addition, 2D metal tellurides exhibit diverse polymorphs with similar formation energies, which makes the precise phase control even more challenging. To our knowledge, CVD grown monolayer 2H MoTe2 (with phase control) has not been reported before.
In this work, we experimentally demonstrated that the proposed LV method was successful in synthesizing diverse high-quality metal telluride, particularly MoTe2, with precise phase and layer-number control. Furthermore, main element tellurides flakes (SnTe and GaTe) were also synthesized through the LV approach. Our experimental findings have unveiled a new LV-facilitated process with on-site nucleation and on-demand reactant supply in the liquid phase, which is fundamentally distinctive from the conventional VSS process. As a general phase-guided method, this LV method can be applied to a wide range of precursors and material systems.
The merits of LV method can be understood from both kinetic and thermodynamic perspectives. The CVD synthesis involves two reaction processes: (1) formation: M (g) +xTe (l) MTex (s); and (2) decomposition: MTex (s) MTe(x-y) (s) + yTe (g) (M is metal), as shown in Fig. 1a. We highlight the vital role of decomposition as it is inevitable in conventional CVD growth, and is evidenced by defects and lattice disturbance in CVD-grown 2D metal tellurides (Supplementary Fig. 1). In order to synthesize high-quality materials, it is crucial to promote the formation process and impede the decomposition process. We use the LV strategy to exploit the synergism of equilibrium and kinetics to achieve this goal. From the kinetics point of view, the reaction rate of formation process sensitively depends on the precursors’ concentration. It can be expressed as , where is the rate constant, c(M) and c(Te) denote the concentration of metal-based and tellurium precursors, respectively. Thus, a higher concentration of Te will boost the kinetic growth rate ( ) of 2D metal tellurides. From a thermodynamics point of view, high concentration of Te will impede the decomposition of as-grown tellurides, thus minimize the probabilities of defect creation, and shift the equilibrium to the side of repairing defects according to the Le Châtelier's principle. For the LV growth, liquid reservoirs of Te supply a significantly higher concentration of on-site Te precursors than Te vapor in the conventional VSS method, since the Te concentration severely drops along the direction of the gas flow. Since the rate of defects repairing follows a power law against the concentration of Te, high concentration for liquid Te ensures fast repairing rate, which forms stoichiometric 2D tellurides without vacancies. In contrast, for the VSS mode, because of insufficient supply of Te, the growth rate of MTex should be much slower, and the as-grown MTex is prone to decompose, thus generating defects (Fig. 1a, right). In short, owing to sufficient reactants in liquid, the LV strategy can greatly facilitate growth, and minimize defects through the kinetic and thermodynamic control.
The process of LV growth is illustrated in Fig. 1b. Pre-deposited Te precursor begins to melt and forms liquid droplets when the furnace temperature is ramped up above the Te melting temperature (~450 ℃) (Fig. 1b, ii). Subsequently, as the metal precursor species vaporize, the Te liquid reservoir continuously collects surrounding gaseous precursor species (Fig. 1b, iii). The liquid-vapor-solid triple junction (solid is the substrate) is the major growth front. MTex continuously precipitates out at the liquid-solid interface (Fig. 1b, iv and Fig. 1c). As the growth proceeds, droplet vaporizes and the triple phase line moves inward (Fig. 1b, v and Fig. 1d). Consequently, the MTex flake has inward growth at the triple phase junction, and the growth terminates when the droplet is all consumed (Fig. 1b, vi and Fig. 1e). Simultaneously, the flake grows outward and increase lateral size through attachment of Te atoms from the in-situ saturated Te vapor and metal precursor atoms at the growing edge. Liquid precursor provides nucleation site and boosts growth. Moreover, liquid reservoir prevents defects generation and decomposition by continuously providing sufficient Te precursor in-situ during the growth.
Several experimental phenomena in the synthesis of metal tellurides are in good agreement with the proposed LV mechanism. Optical Microscope (OM) images illustrate several key stages during the LV growth of MoTe2 (Fig. 1, c to e). The droplet-shaped particles were found on MoTe2 flakes when the growth is terminated at an early stage (Fig. 1c). Black regions (labeled as light blue circle in the inset of Fig. 1f) on MoTe2 flakes were sometimes observed after the LV growth. Raman spectrum of the black region indicates that it is a mixture of elemental Te crystal and 2H MoTe2, while other regions of the flake only show Raman features of 2H MoTe2 (Fig. 1f). These black regions could be the residues of droplets. Furthermore, circular morphology of as-grown MoTe2 as well as coffee rings suggest the presence of the original Te droplet and many MoTe2 flakes forms from that (Fig. 1g and Supplementary Fig. 2). The MoTe2 grown out of the liquid region indicates outward growth also exists, which could result from the reaction between in-situ vaporized Te from liquid reacts and Mo precursors (Fig. 1g, left panel). Moreover, we found that droplet sometimes crawls on the surface and MoTe2 precipitates out, leaving a trace behind. The MoTe2 shows the migration trace of the droplet, and the thicker region (whiter region) at the tip would be the location of the droplet residue (Supplementary Fig. 2). For other metal tellurides such as GaTe system, flower-shaped GaTe was sometimes obtained with a droplet-like center and surrounding “petal” flakes (Fig. 1h). Some traces can be seen, which could be original Te droplet, now shrinking much in size (Fig. 1h). Raman spectra verify that the petals are the precipitated GaTe crystal (Supplementary Fig. 3). The GaTe flakes precipitate at the liquid-solid interface. The droplet does not completely deplete during the growth, then the liquid in the center condensed during the cooling process.
We start with MoTe2 as an example to demonstrate the advantages of the LV method. Monolayer 2H MoTe2 is a promising key material as the on-chip light source for next-generation silicon-based photo-electronics 14. Moreover, monolayer 2H MoTe2 provides a great platform for developing phase-change devices because of the reversible phase transition of monolayer 2H to 1T′ MoTe2 12. The 1T′ MoTe2 is the Weyl superconductor with an edge supercurrent 15, and monolayer 1T′ is a quantum spin Hall insulator 16. The synthesis of atomically thin MoTe2 with controllable polymorphism is a huge challenge due to the fact that several crystal phases (2H 2,17, 1T′ 15,16, 1T 18, Td 19, d1T 20, and 2Hd 21) with similar energies (energy difference between 2H and 1T′ MoTe2 <0.04 eV per unit formula) and tendency of decomposition during growth 2. Since monolayer 1T′ MoTe2 is more stable than 2H MoTe2 under Te deficiency 2, and Te vacancy is always generated during conventional growth 22, an efficient CVD method to synthesize monolayer 2H MoTe2 is still lacking.
The synthesis of 2D MoTe2 with specific phase can be realized via LV method by optimizing the thermodynamic and kinetic factors. Our phase-selective synthesis strategy is based on controlling Te and temperature which tunes the relative stability of the 2H and 1T′ phases. Maintaining a Te enriched environment during growth to avoid Te deficiency should be a key point to obtain monolayer 2H MoTe2. High-quality monolayer 2H MoTe2 can be obtained using MoCl5, H2 and Te (details in Supplementary Information, Supplementary Figs. 4 to 8). To confine the as-grown MoTe2 in monolayer form, the amounts of both Te and MoCl5 were finely controlled. Low mass flux of MoCl5 was realized by the ampoule tube design and the low amount of MoCl5 precursor. The as-grown monolayer MoTe2 flake shows hexagonal shape, a typical shape of 2H single crystals (with size ranging from a few to tens of micrometers, Fig. 2a). The step height of the flake is ~1 nm, consistent with the height of exfoliated monolayer 2H MoTe2 (Fig. 2b).
Raman spectroscopy and polarized second-harmonic generation (SHG) spectroscopy were used to investigate the crystal phase, uniformity and optical anisotropy properties of the 2H MoTe2 flake. Raman spectrum displays two peaks at 171 cm-1 (A1g mode) and 236 cm-1 (E2g1mode) (Fig. 2c). The strong intensity of A1g mode and absence of B2g1 mode (290 cm-1) confirming the monolayer 2H nature of the flake 23. The uniformly distributed color in both Raman intensity map of A1g (Fig. 2d) and E2g1 (Supplementary Fig. 10) modes, together with no B2g1 signal over the whole flake region (Supplementary Fig. 10), suggest the uniformity of the monolayer flake. The SHG spectrum of the flake (Fig. 2e) displays a strong SHG signal and six-fold SHG pattern, consistent with the SHG feature of monolayer 2H MoTe2. Furthermore, the SHG mapping (Fig. 2e, inset) shows uniform intensity with hexagonal shape without boundary lines, indicating single crystal nature of the monolayer flake.
Monolayer 2H MoTe2 film has also been successfully obtained via LV synthesis. The monolayer 2H film is uniform and continuous, with only the edge region of the sample showing discontinuity (Fig. 2f). Atomically resolved annular dark field scanning transmission electron microscopy (ADF-STEM) reveals the hexagonal atomic lattice of monolayer 2H MoTe2 (Fig. 2g). X-ray photoelectron spectroscopy (XPS) measurement was conducted to examine the elemental composition, bonding types and phase purity of LV grown monolayer film. The high-resolution Mo 3d spectrum shows two prominent peaks at 228.3 eV (3d5/2) and 231.5 eV (3d5/2) which are attributed to the Mo-Te bonds (Fig. 2h). The two peaks at 573 eV and 583.4 eV in the Te 3d spectrum can also be assigned to the Mo-Te bonds (Fig. 2h). These features are consistent with that of bulk 2H MoTe2 crystal and of CVD grown few-layer 2H MoTe2 17. The XPS peaks for 1T′ MoTe2 were not detected in the sample, indicating that our CVD film is 2H MoTe2 with high phase purity. Moreover, the atomic ratio between Mo to Te is ~1:2, suggesting the as-grown MoTe2 is stoichiometric.
The LV grown monolayer 2H MoTe2 exhibits excellent photoluminescence property. The photoluminescence (PL) spectrum of monolayer film shows a prominent PL peak centered at ~ 1135 nm (Fig. 2i), consistent with the direct band gap of monolayer 2H MoTe2 at ~1.09 eV. Repeated measurements on various locations of the same film yield very similar position and intensity of the peak (Supplementary Fig. 11). The PL peak of the LV grown sample has a comparable intensity with the exfoliated monolayer sample, and different locations yield PL peaks with similar position and intensity, demonstrating high-quality and homogeneity for the LV grown monolayer 2H MoTe2. Owing to the comparable direct bandgap with Si, superior optical property, as well as its 2D nature (without dangling bonds on the surface and can be easily integrate with Si), LV grown monolayer 2H MoTe2 could be an excellent candidate for integration with Si photonics.
The LV technique can synthesize high-quality 2H MoTe2 with controlled layer numbers via adjusting the amount of MoCl5 precursor. The thickness of 2H MoTe2 is found positively correlated with the amount of MoCl5. Thus few-layer MoTe2 is produced through increasing the amount of MoCl5 in the ampoule. The step height of the hexagonal shaped flake is about ~2 nm, suggesting it is tri-layer MoTe2 (Fig. 3a). Angle dependent polarized SHG displays six-folder pattern (Supplementary Fig. 12a), indicating the noncentrosymmetric nature of the flake. Given the flake is ~2 nm height, the flake should be tri-layer 2H MoTe2. SHG mapping of the flake indicates there is no boundaries in the flake, suggesting the flake is single crystal MoTe2 (Supplementary Fig. 12b). The Raman spectrum shows three peaks at ~171 cm-1 (A1g mode), ∼234 cm-1 (E2g1 mode), and ∼290 cm-1 (B2g1 mode) (Fig. 3b). Since the out-of-plane B2g1 mode is Raman inactive for monolayer and bulk 2H MoTe2, the flake should be few-layer 2H MoTe2. The intensity maps of the A1g peak and E2g1 peak display homogenous color over the flake region, suggesting high homogeneity of the few-layer 2H MoTe2 flake (Supplementary Fig. S13). Since the ratio of B2g1/ E2g1 increases as the layer number of 2H MoTe2 decreases, the uniformly distributed color over the flake in B2g1/ E2g1 map further indicates the flake has uniform thickness (Fig. 3c). High-resolution transmission electron microscopy (HRTEM) image shows a hexagonal atomic lattice without obvious defects. Moreover, selected-area electron diffraction (SAED) of the flake displays hexagonal symmetry pattern (Fig. 3d). These features are consistent with crystal structure of 2H MoTe2, suggesting the high crystalline quality of LV-grown 2H MoTe2. Transfer characteristics of a few-layer 2H MoTe2 transistor displays ambipolar behavior with a current ON/OFF ratio of ~103 at room temperature (Supplementary Fig. 14). No obvious hysteresis is observed during the gate voltage scanning forward and backward. The near zero charge neutrality point indicates the as-grown MoTe2 flake is intrinsic without obvious doping, in strong contrast with the heavy p-type doping 2H MoTe2 grown by previous methods 13,17. Therefore, LV grown MoTe2 provides a good materials platform to investigate the intrinsic properties of 2H MoTe2.
Phase-selective (2H or 1T′) growth of MoTe2 can be realized via LV method through tuning temperature or Te concentration. With less amount of Te pre-deposited on substrate and a lower growth temperature, high-quality 1T′ MoTe2 flakes will be obtained. Willow-leave like flakes were scattered on mica substrate with certain preferential directions (Fig. 3e). The elongated shape comes from the in-plane anisotropic feature of 1T′ MoTe2, with the long direction (preferential growth direction) along the zigzag Mo atomic chain. SAED from 1T′ MoTe2 shows rectangular pattern, consistent with the lattice structure of 1T′ MoTe2 (Supplementary Fig. 15). Raman spectrum of a flake displays several Raman peaks: Ag modes at 85 cm-1, 115 cm-1, 130 cm-1, 164 cm-1, 253 cm-1 and 272 cm-1, as well as Bg modes at 93 cm-1 and 103 cm-1 (Fig. 3f). These features are consistent with the Raman spectrum of monolayer 1T′ MoTe2 24. Interestingly, the intensity of these modes is strongly dependent on the angle of the flake (Supplementary Fig. 16). Since Raman intensity is closely related to crystalline orientation because of the anisotropic nature of 1T′ 25, the angle-dependent Raman intensity suggests that the flake is single crystal. The uniformly distributed color in the Raman intensity map of Ag mode at 164 cm-1 (Fig. 3g) further indicates the homogenous and single crystalline feature of the 1T′ MoTe2. In contrast, only sparse and small 1T′ MoTe2 flakes are obtained (Supplementary Fig. 17) via VSS growth.
XPS was utilized to characterize elemental composition and bonding types of the 1T′ MoTe2 sample. The Mo 3d5/2 peak at 228.1 eV and Mo 3d3/2 peak at 231.2 eV can be attributed to Mo-Te bonds. The strong Te 3d5/2 and Te 3d3/2 peaks at 572.7 eV and 583.1 eV are also assigned to Mo-Te bonds, consistent with bulk 1T′ samples (Fig. 3h). There is no Mo oxide or Te oxide peaks in the high-resolution XPS spectra, indicating the CVD grown 1T′ MoTe2 is intrinsic without oxidation. The Te/Mo atomic ratio is estimated to be ~2.01, suggesting the CVD grown 1T′ MoTe2 is stoichiometric. The source-drain current versus voltage characteristic of the 1T′ MoTe2 grown from MoCl5 (Fig. 3i) is linear and symmetric, with a sheet resistance of ~3200 Ω/sq, indicating the metallic nature of the flake.
The generality of the LV approach was demonstrated for the synthesis of diverse 2D metal tellurides, including transition metal (MoTe2), IVA metal (SnTe) and IIIA metal tellurides (GaTe) with multiple crystalline phases (Fig. 4). β-SnTe is a topological crystalline insulator, and few-layer films show robust ferroelectricity at room temperature 26. GaTe has extremely high photoresponsivity and good thermoelectric properties, which has promising applications in photonics and thermoelectric devices 27. Liquid Te droplet works as a preferential reservoir for precursors from the gas phase, and subsequently as seed for the growth of 2D metal tellurides. We choose metal chloride instead of usual metal oxide to address the weak chemical reactivity of Te. The higher reactivity of chloride precursors leads to much faster reaction, synchronously avoid intermediate products and oxidation during growth. Moreover, the total oxygen-free growth condition can further ensure the high quality of as-grown metal telluride. For the synthesis of SnTe, we choose SnCl2, H2 and Te as precursors. After the LV growth, α-SnTe and β-SnTe were successfully obtained. The square shape of the flake (Fig. 4b) corresponds to the cubic crystal structure of β-SnTe. The Raman spectrum of the flake (Fig. 4c) shows two peaks at ~118 and ~139 cm-1, consistent with the Raman features of nano-sized β-SnTe 28. Interestingly, the low-temperature phase α-SnTe 29 is also stable at room temperature. The Raman peak at ~128 cm-1 (Fig. 4f) and EDS result (Supplementary Fig. 18) clearly prove that the trianglar flake is α-SnTe 29,30. GaTe flakes are obtained from GaCl3, Te and H2. Four Raman peaks at ~ 125, 143, 160, 171 cm-1 (Fig. 4i) are the same as that of the GaTe flakes exfoliated from bulk crystal 27, indicating the as-obtained flakes are GaTe.
In conclusion, we have developed a universal LV method to synthesize 2D TMDs. The LV method provides an ideal way to continuously supply precursors, facilitates growth, suppresses decomposition, and enables growth of 2D materials/phase that are inaccessible through the conventional VSS growth. Moreover, LV-grown 2D materials show high quality due to minimum defects. The advantages of LV are demonstrated by synthesis of 2D metal tellurides with high quality, better controlled phase and stoichiometry, which were hard to synthesize previously due to low bonding energy, tendency for decomposition, and complex crystalline phases. In particular, monolayer 2H MoTe2 film with superior properties was generated, thus enables diverse applications such as highly efficient near-infrared light-emitting diodes, near-infrared photodetectors, integration with silicon photonics, and phase-change devices. This method is of great significance for 2D metal tellurides with both fundamental and technological importance. We anticipate the LV strategy can be readily applied to synthesize many other high-quality quantum materials and heterostructures.