Fast growth of large-grain and continuous MoS2 films through a self-capping vapor-liquid-solid method

Most chemical vapor deposition methods for transition metal dichalcogenides use an extremely small amount of precursor to render large single-crystal flakes, which usually causes low coverage of the materials on the substrate. In this study, a self-capping vapor-liquid-solid reaction is proposed to fabricate large-grain, continuous MoS2 films. An intermediate liquid phase-Na2Mo2O7 is formed through a eutectic reaction of MoO3 and NaF, followed by being sulfurized into MoS2. The as-formed MoS2 seeds function as a capping layer that reduces the nucleation density and promotes lateral growth. By tuning the driving force of the reaction, large mono/bilayer (1.1 mm/200 μm) flakes or full-coverage films (with a record-high average grain size of 450 μm) can be grown on centimeter-scale substrates. The field-effect transistors fabricated from the full-coverage films show high mobility (33 and 49 cm2 V−1 s−1 for the mono and bilayer regions) and on/off ratio (1 ~ 5 × 108) across a 1.5 cm × 1.5 cm region.

A part from graphene, transition metal dichalcogenides (TMDs) with atomic thickness are the most renowned two-dimensional (2D) materials because of their excellent electrical and optical properties [1][2][3][4][5][6] . Their robust physical properties in atmosphere enable their practical applications in novel optoelectronic devices 7,8 . For electronics, TMDs with atomic thickness, which inherently have no surface dangling bonds, are immune to mobility degradation and short channel effects in contrast to conventional three-dimensional materials, such as Si and GaAs [9][10][11] . Such materials have layer-dependent bandgaps from near infrared to visible regions 12,13 , and thus, TMDs are favorable for energy or optical applications 7,8,14,15 . Furthermore, the difference between Berry curvature at the K and K′ valleys of monolayer TMDs generates new opportunities for valleytronics 16,17 . Despite these remarkable properties, TMDs still have limitations arising from spatial nonuniformity. Therefore, the fabrication of high-quality and large-grain films is thus crucial for TMDs. Currently, chemical vapor deposition (CVD) is the most recognized method for producing high-quality monolayer TMDs because of its low cost and scalability [18][19][20][21][22] . Conventional CVD methods of TMD fabrication are based on the reaction of gas-phase chalcogens (e.g., S and Se) and metal oxides (e.g., MoO 3 and WO 3 ) [18][19][20][21][22] . Generally, in gas-phase reactions, the grain size of TMDs is limited by the high nucleation density and typically is <500 µm. Recently, researchers have used various methods, which include fabricating at a high temperature 23 , inserting diffusion barriers 24 , and using an extremely small amount of precursors 25 , to reduce nucleation density and to increase the surface diffusion length for growing large TMD crystals. These methods can produce comparatively large crystals but considerably reduce the coverage of TMD crystals 25 , which hinders their practical applications.
For bulk materials, Czochralski 26,27 method enables the production of a single-crystal ingot with a diameter of up to 300 mm by vertically pulling a solid seed crystal from a liquid source 28 . This method provides the unprecedentedly high uniformity of conventional bulk materials at an ultra-large scale. Moreover, a liquid source can more easily dissolve other solid dopants than a gas-gas reaction 29 . Therefore, it is desirable to grow solid crystals from liquid sources. For the growth of TMDs, Li et al. recently proposes the vapor-liquid-solid (VLS) reaction for fabricating high-quality MoS 2 nanoribbons from a liquid precursor on a sodium chloride (NaCl) single crystal 30 . First, NaCl reacts with MoO 3 to form a eutectic compound (Na 2 Mo 2 O 7 ), which has a relatively lower melting point and exists in a liquid phase under growth conditions (generally 700-800°C). Second, the sulfur vapor is rapidly dissolved into the liquid and reacts to form a solid-state monolayer MoS 2 on NaCl. However, because of the low wettability between the NaCl and liquid Na 2 Mo 2 O 7 droplets, this method can only generate MoS 2 nanoribbons, which considerably limits its application. This problem can be solved by growing on other substrates with a better wettability 31 .
Herein, a self-capping vapor-liquid-solid (SCVLS) reaction, which can grow large single crystals and full-coverage TMD films, is proposed. A solid precursor comprising ultra-thin MoO 3 , SiO 2 , and NaF layers was used for the controllable eutectic reaction of MoO 3 and NaF at high temperature. The as-formed eutectic liquid (Na 2 Mo 2 O 7 ) rose to the surface and was sulfurized into MoS 2 seeds. These seeds, acted as a self-capping layer, redirected the rising liquid into a horizontal direction. The residual liquid was continuously pushed along the growth direction and eventually sulfurized to form new MoS 2 at the edge of the MoS 2 seeds. This growth mechanism enables fabrication of ultra-large (~1.1 mm) single crystals. Moreover, continuous large-area MoS 2 film with large-grain size (~450 µm) can also be fabricated using thicker precursor. By controlling the kinetic factors of this reaction, the layer number can be controlled and large bilayer MoS 2 (~200 µm) can be achieved. In this study, the quality and uniformity of MoS 2 grown using this method are evaluated through electron microscopy, optical spectroscopy, and electrical measurements. For electrical measurements, both mono-and bilayer MoS 2 field-effect transistors (FETs) show high mobility (33 and 49 cm 2 V −1 s −1 ), large on/off ratio (5 × 10 8 ), and high current density (up to 230 and 390 µA µm −1 ). The large-grain, continuous film exhibits high performance across a 1.5 ×1.5 cm area, making the SCVLS method promising for practical applications.

Result
Material synthesis and growth mechanism. Figures 1a and S1 show that a smooth MoO 3 layer was grown on c-plane sapphire through plasma-enhanced atomic layer deposition (PEALD). SiO 2 and NaF layers were stacked on top of the MoO 3 layer through sputtering and thermal evaporation, respectively. The SiO 2 layer acted as a diffusion membrane to control the amount of MoO 3 vapor that broke the SiO 2 layer ( Fig. 1b and Supplementary  Fig. 2), diffused upward and reacted with the NaF layer at a temperature higher than 500°C to form liquid-phase Na 2 Mo 2 O 7 and gas-phase MoO 2 F 2 ( Fig. 1b and Supplementary Fig. 3). Simultaneously, the consumption of NaF generated holes and pathways in the NaF layer, which allowed Na 2 Mo 2 O 7 and MoO 2 F 2 to gradually rise to the top surface of the NaF through the pressure gradient and capillary phenomenon (Fig. 1c). Meanwhile, sulfur vapor was introduced into the system and rapidly dissolved in the eutectic liquid (Na 2 Mo 2 O 7 ) that rose to the surface. As discussed in the first VLS paper, the products of this VLS reaction were MoS 2(s) and sulfur oxides (SO 2(g) and SO 3(g) ) 30 . Moreover, the molten liquid surface provided a temporarily atomic-flat and defection-free surface with a low nucleation density 23,32 . The oversaturated MoS 2 precipitated as seed layers on the liquid surface (Fig. 1e). The as-formed MoS 2 seed layers blocked the route for sulfur to dissolve into the liquid and redirected the underlying liquid to move horizontally. The unsaturated liquid then emerged to the surface at the MoS 2 edge, and this was where the SCVLS reaction primarily occurred. Therefore, MoS 2 laterally grew into large crystals (Fig. 1f, g). Millimeter-sized MoS 2 single crystals were obtained using this SCVLS method. The large triangular MoS 2 flakes are single crystals in nature, as validated by diffraction analysis at multiple spots in a large flake ( Supplementary Fig. 4). The zoom-in image of a MoS 2 edge exhibits many bilayer fringes (Fig. 1h), which were a result of the precipitation of the residual liquid at the edges of the MoS 2 flakes during the rapid cooling process. These fringes validate the existence of the liquid phase during growth and the aforementioned mechanism. In contrast to the conventional CVD method, wherein the nonuniform gas flow in the furnace often gives resultant film of poorer uniformity 33 , extending the SCVLS method to a wafer scale is facile because the precursor rises uniformly from the bottom surface of the growth substrate. Supplementary Fig. 5 shows a full-coverage MoS 2 film grown on a sapphire of 3 × 3 cm 2 (size was only limited by the CVD tube size). This process can also be extended to grow MoSe 2 by replacing sulfur with selenium, result of which is shown in Supplementary Fig. 6.
Characterization of as-grown material. Compared with the conventional CVD technique, the final products of this SCVLS method were more complex (Fig. 2a), which comprised two parts: the top MoS 2 layer and the complex solid products generated by the quenched liquid within the NaF matrix. During the sulfurization process, sulfur vapor was primarily dissolved into the liquid through the exposed liquid-gas interface, and thus, the oversaturated liquid could continuously precipitate MoS 2 at the edge of the MoS 2 seeds. However, the reactions were not limited to the top surface. With a lower sulfur concentration, some incomplete sulfurization reactions and precipitations were observed within the NaF matrix (Supplementary Fig. 7). Upon cooling, the unsaturated liquid solidified and resided below the MoS 2 flakes or was buried in the NaF matrix (Fig. 2a). X-ray photoelectron spectroscopy (XPS)    was used to analyze the final products. Figure 2b, c shows Mo-3d and S-2p spectra obtained from regions covered by large MoS 2 flakes and regions with the exposed NaF matrix, respectively. For a region covered with MoS 2 , the Mo-3d peak of~230-228 eV could be deconvoluted into sharp Mo 4+ and broad Mo x+ . The Mo 4+ signal was obtained from the top MoS 2 , and the Mo x+ signal was obtained from the precipitates and solidified liquid phase in the NaF matrix, as shown using an XPS depth profile ( Supplementary Fig. 7). With a lower sulfur concentration in the matrix, the possible products for the precipitate and quenched liquid were amorphous MoS 2 , MoO 2 , and MoS x O y , comprising the Mo x+ signal. In addition, a small peak at~235 eV indicates Mo 6+ , which is the peak from Na 2 Mo 2 O 7 , indicating the presence of residual unreacted precursor below MoS 2 ; this supports the as-proposed horizontal transport of the liquid. For a region without MoS 2 , only a small amount of sulfur diffused into the NaF matrix and reacted with the liquid below the surface. Figure 2b, c shows a considerably weaker Mo x+ and sulfur signal, which indicates that no liquid rose to the top surface. The resultant products were further characterized using Raman spectroscopy (Fig. 2d). The region covered with MoS 2 exhibited sharp E 2g and A 1g peaks with a spacing of 19.5 cm −1 , thus validating the high quality and monolayer characteristics for the as-grown MoS 2 . The region that was not covered by MoS 2 exhibited no significant Raman signal, which validated the absence of any crystalline product in the NaF matrix. The as-grown monolayer MoS 2 could be readily transferred to various substrates using the conventional polymethyl-methacrylate (PMMA) method. Atomic force microscopy images and photoluminescence spectrum in Supplementary Fig. 8 and 9 also confirm the monolayer property. The insets of Fig. 2b Controlling the coverage and thickness of MoS 2 . With a suitable sulfur source, the growth rate of SCVLS process was controlled by horizontal transport rate of the liquid. Rapid horizontal mass transport of the liquid was crucial for growing large monolayer MoS 2 . The driving force stemmed from the diffusion and capillary phenomena of the high-pressure liquid and gas produced by the eutectic reaction (Fig. 3a). Because MoS 2 layers covered the top surface and confined the liquid flow, the vertical driving force was redirected horizontally, and both vertical and horizontal liquid transport was promoted with increasing amount of liquid source. The rapid SCVLS reaction could thus be performed under this condition, and the rapid growth rate abruptly increased the grain size of MoS 2 . However, when the driving force was weak, the low growth rate would result in more nucleation seeds on NaF, thus reducing the average size of the MoS 2 crystals. In some regions, the weak driving force was not sufficient to push the liquid to the surface. The sulfur vapor would slowly diffuse into the NaF matrix, react with the liquid, and eventually solidify. Therefore, no MoS 2 was grown on the surface under this condition, and this phenomenon reduced the coverage of MoS 2 .
Here, the vertical driving force was controlled using different amounts of MoO 3 sources. Figure 3b-d shows the optical images of the as-grown MoS 2 with different thicknesses of MoO 3 precursor layers. The grain size and coverage of MoS 2 abruptly increased with the increasing thickness of MoO 3 precursor (Fig. 3f). In order to estimate the grain size of the full-coverage film (Fig. 3d), the growth time was reduced from 10 ( Fig. 3d) to 1 min (Fig. 3e), to monitor the grain size before grains merged into continuous film. The average grain size of the full-coverage film was~450 µm, which is the largest recorded average grain size for completely covered MoS 2 film. Although there are some thick islands on the film (Fig. 3d and Supplementary Fig. 10), this large-grain continuous film can still demonstrate outstanding electrical performance shown in the later section. The trend of the coverage and average grain size versus the precursor thickness in SCVLS (Fig. 3f) is very different from that for the common gasphase reaction CVD. For gas-phase CVD, the average grain size of the continuous film is reduced by a factor of 10-100 compared with the largest isolated crystals because the larger amount of the precursor for growing continuous film abruptly led to the increased nucleation density and thus reduced grain size (Fig. 3g) 25,34 . However, for SCVLS method, the average grain sizes of the continuous film (450 µm) and largest isolated crystals (500 µm) are similar because of the self-capping effect and the fast transport of liquid. Furthermore, coverage of~82% was reached within 1 min of fabrication (Fig. 3e), which demonstrates the rapid growth rate (370 µm/min, see Supplementary Fig. 11). This may be a result from the fast transport of liquid assisted by the fluoride surface, which has been shown to enhance growth rate of 2D materials 35 . The driving force and nucleation density could be further controlled by tuning the thickness of SiO 2 membranes, growth temperature, and the thickness of NaF ( Supplementary  Figs. 12-14).
In addition to large-grain continuous film, layer-controlled growth of bilayer and multilayer are also attractive because of the better electrical performance of few-layer MoS 2 [36][37][38][39] . SCVLS can also control the layer number of MoS 2 with its special growth mechanism we proposed. In the growth condition of previous paragraphs, sulfur vapor was introduced before or during the liquid's rise to the surface. The small liquid droplets rapidly dissolved sulfur and formed monolayer MoS 2 capping seeds, which promoted the horizontal mass transport and formed large monolayer MoS 2 grains. It is noteworthy to mention that the SCVLS reaction can be dynamically controlled by changing the timing of sulfurization (Fig. 4a). When sulfur vapor was introduced later, the emerged liquid would form into a large droplet ( Supplementary Fig. 15). During sulfurization, the asformed small MoS 2 seeds were buried in the oversaturated liquid. Under this condition, fresh MoS 2 could be formed at the edge of the original seeds, and a second layer could be grown on the MoS 2 seeds (Fig. 4b). Large bilayer MoS 2 crystals could be fabricated by delaying the sulfurization timing for 2 min. Trilayer MoS 2 was occasionally observed when sulfurization is delayed. Figure 4c-e are the optical images of the transferred mono-, bi-, and trilayer MoS 2 on SiO 2 /Si substrates, respectively. The clear optical contrast shows the characteristics of the mono-, bi-, and trilayers of each MoS 2 . AFM images in Supplementary Fig. 16 also confirm the thickness of these samples. Figure 4d, e shows that the second and third layers are well-aligned with the bottom MoS 2 layer, thus indicating the epitaxial growth of an excess MoS 2 layer. The diffraction patterns manifest a 2H-type stacking order of the SCVLS reaction ( Supplementary Fig. 17) 30 . Raman spectra in Fig. 4f further validate the layer number and strong interaction between the mono-, bi-, and trilayer MoS 2 grown using the SCVLS method. The peak separations of E 2g and A 1g are 18  each layer changes the dielectric environment, thus softening the in-plane E 2g mode (red-shift). Moreover, the strong interaction between interlayer S increases the restoring force, thus stiffening the out-of-plane A 1g mode (blue shift) 40 . The photoluminescence spectra in Fig. 4g exhibit clear quenched signals for bilayer and trilayer MoS 2 because of the direct-indirect band gap transition for monolayer and bilayer MoS 2 4,6 . By employing dynamic control of sulfurization, a large bilayer single crystal (200 µm) is successfully synthesized, the grain size of which is comparable to the large bilayer crystals in the previous studies 37,38 . Moreover, the ability to change the layer number of MoS 2 by controlling the size of the droplet validated the SCVLS mechanism proposed in Fig. 1.
High-performance FET device. The electrical properties of the large monolayer MoS 2 crystal grown by the SCVLS method was examined by measuring the transport properties of MoS 2 FETs. Figure 5a shows an FET with a back-gate structure and 90-nmthick SiO 2 . Both mono-and bilayer MoS 2 FETs were fabricated as shown in Fig. 5b and Supplementary Fig. 18. Figure 5c displays the gate-dependent conductance of the MoS 2 FETs. Both devices have a very small hysteresis, indicating low defect and impurity induced trap density in the SCVLS growth and the device fabrication processes. The field-effect mobility was calculated using , where C G , L CH , W, V GS , and G stand for the back-gate capacitance, channel length, channel width, back-gate voltage, and sheet conductance of the channel, respectively. Because of its higher carrier density and stronger charge screening effect, bilayer MoS 2 has a smaller threshold voltage (V th ) and higher mobility. The mobilities of the mono-and bilayer MoS 2 FETs are 33 and 49 cm 2 V −1 S −1 , respectively.
These values are comparable to the exfoliated MoS 2 41 and the best reported values of CVD MoS 2 on SiO 2 38,42,43 , showing the high quality of MoS 2 grown through the SCVLS method. The temperature-dependent transport also confirms the quality of SCVLS MoS 2 as shown in Fig. 5d. For the monolayer device, a clear metal-insulator transition (MIT) was observed, which is generally detected when using a high-k dielectric layer to reduce Columbic scattering in the MoS 2 channel 2 . For back-gate devices without a high-k dielectric layer, a clear MIT occurs only when using high-quality MoS 2 with a low concentration of sulfur vacancies. In general, according to the Ioffe-Regel criterion 44 , MIT occurs when the critical channel conductance is approximately one quantum conductivity (e 2 /h) 2,44 . If the crossover point is in a lower carrier concentration region, this directly reflects the high-mobility property of MoS 2 45,46 . The carrier concentration of the transition point is calculated using where V T and V MIT are the threshold voltage and voltage at which the MIT occurs, respectively 47,48 . The n MIT of the SCVLS MoS 2 in this study is 4.3 × 10 12 cm −2 , which is even lower than the previously reported intrinsic exfoliated MoS 2 with low S-vacancy concentration 47 . This indicated the high quality and low sulfur vacancy concentration of the MoS 2 fabricated using the SCVLS method. Figure 5e is the output characteristics of a short channel (1.48 μm, see Supplementary Fig. 19) monolayer FET at various back-gate voltages. The linear behavior in the low source-drain voltage (V DS ) region shows a good Ohmic property of contacts. Figure 5f presents the semi-logarithmic of the gate-controlled current density. The device on/off ratio can reach 5 × 10 8  density in the monolayer MoS 2 is 230 µA µm −1 (390 μA μm −1 for bilayer), potentially comparable to the optimal reported in consideration of the difference in contact geometry 38,42 . Table 1 lists the recently reported monolayer MoS 2 back-gate FETs. MoS 2 grown by the SCVLS method exhibits the largest grain size and remarkable electrical performance compared with other CVD techniques. The good uniformity of the large monolayer crystal is confirmed by measuring 18 FETs in a 1-mm crystal (Supplementary Fig. 20). In addition, with the capability of growing large-grain and continuous film, we fabricated hundreds of devices across a 1.5 × 1.5-cm area, as shown in Fig. 6a, b. Figure 6c is the gate-dependent conductance of a hundred devices in the whole area. Ninety percent of devices have pure monolayer channel and show high mobilities (34 ± 7 cm 2 V −1 s −1 ) with small variation of V th (4.9 ± 2.3 V). Devices have larger variation in mobility and V th if their channels are bilayer, few-layer, monofew-layer junction, or monolayer with a small few-layer flake on top. However, the mobilities are still high for all of the devices because of the large-grain monolayer underneath ( Supplementary  Fig. 21). This demonstrates the advantage of using the SCVLS method for practical applications.

Discussion
In summary, a new concept of growing high-quality single-crystal 2D materials from the liquid precursor was proposed using the SCVLS method. The rapid horizontal mass transport promotes the lateral growth of 2D materials and allows the growth of More sophisticated sulfurization precursor such as H 2 S is expected to improve the layer number control or the uniformity of continuous films. Fabricating crystals by using the liquid-solid reaction, such as in doping and alloying, is one expected niche of this SCVLS method, which provides a new approach for synthesizing industrial-grade 2D materials for practical applications in 2D electronics.

Methods
Preparation of solid precursor. 3 × 3-cm c-plane (0001) sapphire substrate was cleaned first by deionized water then sonicated in acetone and isopropyl alcohol for 20 and 5 min, respectively. MoO 3 film with well-controlled thickness was grown on top of sapphire substrates with a homemade PEALD system using Mo(CO) 6 as precursor and oxygen plasma as the oxidation reactant. For each deposition cycle, a Mo(CO) 6 precursor pulse is provided into the chamber, then the excess precursor is purged away by argon, and finally oxygen plasma (up to 200 W) is used to oxidize the precursor and form uniform MoO 3 film. Thermal evaporation can be used to replace the PEALD process for depositing MoO 3 but will result in worse MoS 2 morphology ( Supplementary Fig. 22). SiO 2 film was deposited on top of MoO 3 layer by sputtering a 3-inch SiO 2 target with Ar plasma at a power density of 0.6 W/cm 2 in a radio-frequency magnetron sputtering system. NaF thin film was deposited onto the sample by heating NaF powder (Acros, 97%) loaded in a Mo boat in a high vacuum evaporator chamber (<5 × 10 −5 torr). For sputtering and thermal evaporation film, film thickness was monitored by a quartz crystal microbalance and the deposition rate was maintained at 0.1 Ås −1 . Samples were attached to a spinning sample holder to obtain high uniformity.
Growth of MoS 2 . High temperature growth was carried out in a 2-inch quartz tube and the temperature profile of the growth was controlled by a three-zone furnace. Sulfur powder (Aldrich, 99.98%), which was placed in an alumina crucible, and precursor sample held by a 3 × 3-cm 2 quartz plate were placed at the center of first and third hearing zone, respectively, as depicted in Supplementary Fig. 1c.   Supplementary Fig. 18. c Gate-dependent conductance of the devices across the large area.
A 5 sccm H 2 and 50 sccm Ar mixed gas flow was used as carrier gas and the pressure within the quartz tube was controlled to be 30 torr. The temperature at the sample ramped up at a rate of 40°C min −1 to 800°C and was held for 10 min. Sulfur vapor was introduced by ramping up the temperature at the first zone at a rate of 15°C min −1 and was held at the desired temperature during growth. After growth, the furnace was turned off and was fast-cooled using an industrial fan. The temperature ramping profile is shown in Supplementary Fig. 1d. For monolayer MoS 2 growth, A and B setpoints are reached at the same time. For bilayer MoS 2 growth, the B setpoint is reached 2 min later than the A setpoint.
Device fabrication and characterization. p-type heavily doped silicon wafers with 90-nm thermal oxide layers were used for back-gate FET devices. MoS 2 films/ isolated crystals were transferred to wafers through a conventional PMMA method. Optical lithography and oxygen plasma were used to define the MoS 2 strips. Then, the second lithography process defined the source-drain patterns. A 50-nm gold layer was thermally evaporated under high vacuum as the source-drain and back-gate electrodes. Before measuring electrical properties, FETs were annealed at 120°C under a 10 −3 torr vacuum for 10 h in a probe station (Lakeshore). Gate and source-drain voltage were applied by Kethley 6487 picometers. Raman and photoluminescence spectra were measured by a confocal system equipped with a 476-nm laser. XPS spectra were obtained using PHI VersaProbe system. Transmission electron analysis ( Supplementary Figs. 4, 17, and 23) was performed in JEOL AEM 2010F and JEOL AEM 2100F, which was equipped with a probe-type corrector for the spherical aberration of the objective lens. Both systems were operated at 200 kV for the analysis.

Data availability
The data that support the findings of this study are available from the corresponding author on reasonable request.