Self-Limiting Layer Synthesis of Transition Metal Dichalcogenides

This work reports the self-limiting synthesis of an atomically thin, two dimensional transition metal dichalcogenides (2D TMDCs) in the form of MoS2. The layer controllability and large area uniformity essential for electronic and optical device applications is achieved through atomic layer deposition in what is named self-limiting layer synthesis (SLS); a process in which the number of layers is determined by temperature rather than process cycles due to the chemically inactive nature of 2D MoS2. Through spectroscopic and microscopic investigation it is demonstrated that SLS is capable of producing MoS2 with a wafer-scale (~10 cm) layer-number uniformity of more than 90%, which when used as the active layer in a top-gated field-effect transistor, produces an on/off ratio as high as 108. This process is also shown to be applicable to WSe2, with a PN diode fabricated from a MoS2/WSe2 heterostructure exhibiting gate-tunable rectifying characteristics.

Two-dimensional (2D) materials, and the heterostructures that can be created from them, have been widely studied due to their atomic-scale thickness, flexibility and unique electrical/optical properties [1][2][3][4][5] . However, there is still a need to develop a layer-controlled synthesis method capable of producing a uniform 2D material over large areas in order to ensure the reliable operation of optoelectronic devices whose properties are dependent on the number of 2D material layers. The well-established CVD process has allowed large-area graphene sheets to be used in various practical applications 6,7 , as this process is self-limited through a surface-catalyzed process based on the lower solubility of carbon in Cu than in Ni 8 . Since it is this self-limiting behavior that makes it possible to achieve monolayer (1L) graphene over 95% of the target growth area 8 , achieving a similar self-limiting behavior is clearly an important first step in the development of any new process for the large-area uniform growth of 2D materials.
Transition metal dichalcogenides (TMDCs) and their relevant 2D heterostructures (e.g., MoS 2 /WSe 2 and MoS 2 /graphene) have been the most heavily studied semiconducting 2D materials 3,4,[9][10][11][12][13] . Most recent research has been devoted to synthesizing uniform and layer-controlled TMDCs over large areas 9,14,15 , such as chemical vapor deposition and transformation of Mo and MoOx thin film 16,17,18 , but unlike graphene, the self-limiting growth of TMDCs with wafer-level layer controllability and uniformity has not yet been achieved. Atomic layer deposition (ALD) is known to be self-limiting, as the growth rate is dependent on the adsorption of precursor molecules rather than growth conditions such as exposure time 19,20 , but as growth occurs through the formation of multi-layer islands it is difficult to achieve the layer controllability needed when compared to other techniques such as CVD 21,22 . Maximizing the self-limiting behavior of the ALD process is therefore essential to achieving the layer controllability needed for a 2D structure, which requires not only careful optimization of the process conditions (e.g., temperature, pressure, exposure of precursor/reactant), but also the careful selection of the precursor and reactant 23,24 . Moreover, since the ALD process is entirely based on surface reaction, it is important to understand the surface characteristics of the material being deposited. For example, the ALD of metal oxides or metals on graphene is made difficult by the chemically inactive nature of the graphene surface [25][26][27] . As 2D TMDCs also have a chemically inactive surface, it is reasonable to expect they will exhibit a unique growth behavior during ALD when compared to conventional materials that are rich in dangling bonds 28  In this study, the self-limiting layer synthesis (SLS) of a 2D TMDC (MoS 2 ) is achieved through ALD by combining precursor exposure, purging, reactant exposure and a final purging into a single cycle. In this way, a point is reached at which the number of layers produced is determined purely by the growth temperature; a unique behavior that is directly attributable to the chemical inactivity of the 2D MoS 2 surface. The characteristics and layer uniformity achieved are subsequently assessed through spectroscopic and microscopic analysis, and the universality of the process itself is tested by applying it to the fabrication of a MoS 2 /WSe 2 heterostructure for use in a diode. Figure 1(a-c) contains AFM images and height profiles of MoS 2 synthesized through 120 ALD cycles at growth temperatures of 500, 700 or 900 °C. By transferring this MoS 2 to new SiO 2 substrates it was found that the thickness produced was 2 nm at 500 °C, 1.4 nm at 700 °C and 0.8 nm at 900 °C, which corresponds to the thickness of tri-, bi-, and mono-layer (3L, 2L, and 1L) MoS 2 29 . The Raman spectra (λ exc = 532 nm) in Fig. 1(d) shows that the 1L MoS 2 exhibits E 1 2g and A 1g modes from in-plane and out-of-plane vibrations at 384.6 cm −1 and 404.8 cm −1 , but these shift to 384.1 and 405.2 cm −1 with 2L MoS 2 , and to 383.5 and 406.8 cm −1 with 3L MoS 2 . The peak distance between

Self-limiting layer synthesis of MoS 2
2g and A 1g is often used to determine the number of MoS 2 layers, as an increase in layers is accompanied by a softening of the E 1 2g mode frequency and a stiffening of the A 1g mode frequency 30,31 . In this case, the calculated peak distances of 20.2 cm −1 for 1L, 21.1 cm −1 for 2L, and 23.3 cm −1 for 3L all agree well with previously reported values for MoS 2 29,32,33 . Thus, both the AFM and Raman results show that it is the growth temperature that determines the number of MoS 2 layers by SLS.
The PL spectra of the synthesized MoS 2 are shown in Fig. 1(e) as a function of the number of layers obtained. Note that the 1L MoS 2 spectrum exhibits a strong PL signal at 1.89 eV and a weak, wide PL signal at 2.05 eV, which correspond to the A 1 and B 1 direct excitonic transitions of MoS 2 14,34 . These signals weaken in the case of 2L MoS 2 , and become negligible with 3L MoS 2 , as the increasing number of layers induces a transition from a direct to an indirect band gap. This is concordant with previous results regarding the dependence of the PL signal on the number of layers 14,34,35 and further confirms the growth temperature dependent nature of the SLS of MoS 2 .
Given the good correlation between the Raman peak distances and the number of layers of MoS 2 obtained on a SiO 2 substrate, this was used a criterion to assess the effects of varying the number of process cycles from 40 to 250 at growth temperatures of 500, 600, 700, 800 and 900 °C (the Raman spectra for each point are presented in Supplementary Fig S2). From the results shown in Fig. 1(f), it is evident that the number of MoS 2 layers does not increase linearly with the number of process cycles, but rather saturates at a certain critical point determined by the synthesis temperature (500 °C for 3L, 600-700 °C for 2L, and 800-900 °C for 1L). This stands in stark contrast to conventional ALD, in which the thickness does in fact increase linearly with the number of process cycles. We can therefore only conclude that the growth mechanism of the current SLS process is totally different from that of conventional ALD.
This peculiar "self-limiting" behavior of ALD during the SLS process is believed to be caused by the inherently chemically inactive nature of the surface of TMDCs such as MoS 2 . Specifically, during the growth of the first layer, precursor molecules (MoCl 5 in this case) chemically adsorb to the abundant adsorption sites on the SiO 2 surface. However, once this initial layer is formed over the entire surface, any further chemical adsorption of precursor molecules is hindered by the absence of suitable adsorption sites on the newly created TMDC surface 28,36 . Synthesis is therefore forced to proceed through the physical adsorption of MoCl 5 molecules on MoS 2 ; with the adsorption/ desorption of precursor molecules under this physical adsorption-dominant regime being determined by the growth temperature. This can perhaps be better explained by the framework of the Lennard-Jones potential model: i.e., at lower temperatures molecules are trapped in a potential well because their thermal energy is less than the potential depth, whereas at higher temperatures they have sufficient thermal energy to escape 28,36 .
The surface potential of MoS 2 that is induced by the positive charge between it and the SiO 2 substrate also affects the potential depth of the precursor molecules; the surface potential of MoS 2 decreasing with an increasing number of MoS 2 layers due to their screening effect on the electric field 37 . This decrease in surface potential can reduce the potential depth of the MoCl 5 molecule on MoS 2 in the same way that the surface potential of physically adsorbed CH 4 on h-BN decreases with an increasing number of layers 38 . In other words, once a specific number of MoS 2 layers has been formed at any given growth temperature, any MoCl 5 molecules adsorbed onto the MoS 2 basal plane can be easily desorbed due to the reduced potential depth, thereby creating a self-limiting growth behavior.
It should be noted here that the SLS of 2D MoS 2 relies on using a sufficiently high process temperature to ensure the formation of layered 2D structure with chemically inactive surface. The lower crystallinity and non-layered 3D structure at lower growth temperatures causes deposition to proceed through chemical adsorption of precursor molecules, as is the case in the conventional ALD of MoS 2 without self-limiting growth behavior 21,22 . These temperature requirements make proper selection of the precursor essential, with MoCl 5 being used in this study due to its higher thermal stability relative to other metal organic precursors.
The uniformity of the MoS 2 obtained through SLS was evaluated at different scales through Raman mapping of the peak distances between the A 1g and E 1 2g modes. The Raman map of the 1L SLS MoS 2 in Fig. 2(a) shows a perfectly uniform distribution at a micrometer scale (20 μ m × 20 μ m), with statistical analysis (see Supplementary  Fig. S6) revealing the average peak distance and standard deviation to be 20.3 and 0.6, respectively. Uniformity at a wafer-level scale was measured by synthesizing 1L, 2L and 3L MoS 2 onto 1.5 × 9 cm 2 SiO 2 substrates; the substrate size being limited in this instance by the diameter (~3 cm) and length (15 cm hot zone) of the tube furnace used. It is clear from Fig. 2(b) that the color of the SLS MoS 2 is certainly dependent on the number of layers, but to assess the cm-scale uniformity, Raman spectra were measured at nine different positions along the length of the SLS MoS 2 . The peak distances and full-width at half-maximum (FWHM) of the E 1 2g and A 1g modes are plotted in Fig. 2(c) for each position, from which we see that with all samples the variation in peak distance and FWHM of the E 1 2g and A 1g modes with position is quite small (~2% for peak distance, ~5 and ~4% for the FWHM of E 1 2g and A 1g , respectively). In addition, the Raman peak distance varies from 20.2 to 23.4 cm −1 as the number of layers is increased, confirming that good uniformity and layer control is achieved at the wafer level through the SLS of MoS 2 .
The crystallinity and electrical performance of the 1L SLS MoS 2 was evaluated through high-resolution TEM (HRTEM) analysis and by using it in a top-gated field-effect transistor (FET). The low-magnification TEM image of 1L SLS MoS 2 in Fig. 2(d) reveals triangular, dark-contrast regions of 2L MoS 2 ; but as these represent only about 4% of the total area, the 1L SLS MoS 2 can be considered to have near-perfect ( > 95%) micrometer-scale layer uniformity. This growth of 2L MoS 2 on 1L MoS 2 could be due to the energetically favorable adsorption of MoCl 5 on defect sites such as sulfur vacancies or grain boundaries, as such adsorption of molecules is seen with graphene 26,27 . In the HRTEM image of the 1L SLS MoS 2 in Fig. 2(e), selected regions exhibit a honeycomb-like with a lattice spacing of 0.27 or 0.16 nm depending on whether it involves (100) or (110) planes. A six-fold coordination symmetry is also clearly evident in the fast Fourier transform (FFT) image in the inset of Fig. 2(e). The approximate grain size is 80-100 nm, though this could potentially be improved through further optimization of the process and substrate conditions. The electrical performance of the 1L SLS MoS 2 was evaluated by using it in the fabrication of a top-gated FET with Au(10 nm)/Ti(50 nm) electrodes and an ALD Al 2 O 3 (40nm) gate insulator. The room temperature performance of this FET at 10 −5 mTorr is shown in Fig. 2(f), which reveals an n-type behavior; the 0.2 cm 2 /V•s field effect electron mobility in the linear regime of the transfer curve agreeing with a previous report of a MoS 2 FET 39 . Interestingly, this 1L SLS MoS 2 FET also has a low subthreshold swing value of ~0.36 V/dec and an excellent on/off current ratio of ~10 8 that is higher than anything previously achieved with 1L MoS 2 18 , and is in fact comparable with a single crystal 39 .
Vertically stacked heterostructure. If the proposed self-limiting growth mechanism of SLS is valid, then it would be expected to apply to other 2D materials. This was therefore tested using mechanically exfoliated WSe 2 flakes on a SiO 2 (300 nm)/Si substrate, as demonstrated by the microscopy (OM) images in Fig. 3(a). This WSe 2 flake was confirmed through AFM and Raman analysis (See Supplementary Fig. S8(a-c)) to contain regions of both 2L WSe 2 (#2) and 12L WSe 2 (#3). The AFM image of the SLS MoS 2 produced on this WSe 2 flake at 800 °C ( Fig. 3(b)) shows that a thickness of 1.3 nm (or 2L WSe 2 ) was obtained, indicating that 1L MoS 2 is deposited on both 2L WSe 2 and the SiO 2 substrate under these conditions. Figure 3(c) shows the Raman spectra obtained at 3 different points of the SLS MoS 2 on WSe 2 /SiO 2 . In the SiO 2 region (#1), E 1 2g and A 1g Raman peaks for MoS 2 are observed at 385.2 and 405 cm −1 , respectively, with the peak distance of 20.2 indicating that 1L MoS 2 was obtained as expected. Raman peaks of WSe 2 (i.e., the sum of the E 1 2g and A 1g peaks at 249.8 cm −1 ) are observed in the 2L WSe 2 region (#2) along with peaks for MoS 2 (E 1 2g at 378.4 cm −1 and A 1g at 404.8 cm −1 ), indicating that MoS 2 was also synthesized on the WSe 2 flake. Furthermore, the absence of any Raman peaks related to MoSe 2 (E 1 2g at 286 cm −1 and A 1g at 244 cm −1 ) or WS 2 (E 1 2g at 356 cm −1 and A 1g at 420 cm −1 ) indicates that there is no significant mixing or alloying between the two 2D materials 15,40 . There is, however, a notable 7 cm −1 downshift in the E 1 2g peak of MoS 2 in the 1L MoS 2 /WSe 2 region relative to the MoS 2 /SiO 2 region. A similar downshift has been reported in the case of an interlayer-coupled 1L MoS 2 /1L WSe 2 heterostructure fabricated by transferring individual MoS 2 and WSe 2 flakes, with this being attributed to interaction between MoS 2 and WSe 2 41 . Meanwhile, the absence of any MoS 2 Raman peaks in the 12L WSe 2 (#3) region indicates that there is effectively no growth of MoS 2 on 12L WSe 2 , further supporting the idea that the self-limiting nature of the SLS process is layer dependent.
For further examination of SLS MoS 2 on WSe 2 , Raman mapping of the MoS 2 E 1 2g peak intensity and position was compared against OM images of 1L SLS MoS 2 grown on WSe 2 flakes on a SiO 2 substrate. In Fig. 3(d), regions confirmed by Raman analysis and AFM (See Supplementary Fig. S8(d)) to be 2L WSe 2 are indicated by white arrows, with the rest being bulk WSe 2 . The Raman map of MoS 2 E 1 2g intensity in Fig. 3(e) shows that a strong MoS 2 E 1 2g signal is observed only at 2L WSe 2 regions, indicating that MoS 2 was not synthesized on bulk WSe 2 . The Raman map of MoS 2 E 1 2g position (Fig. 3(f)) further supports the notion that MoS 2 grows only on 2L WSe 2 , which is accompanied by a downshift relative to the MoS 2 E 1 2g position on SiO 2 . This confirms the validity of using SLS to produce MoS 2 on other chemically inert surfaces such as WSe 2 , and indicates that the process has the potential for widespread application. As the SLS process clearly allows for much greater layer control than previously reported methods 10,42,43 , it represents a promising option for fabricating atomically thin functional devices such as PN diodes, light emitting diodes and inverters 10,12,44 . To test this, a PN diode was fabricated using a 1L SLS MoS 2 /2L WSe 2 heterostructure, with Fig. 4(a,b) showing the device structure and an OM image of the fabricated PN diode. Operation of this device is dependent on the back gate voltage, which as shown in Fig. 4(c), can be adjusted by varying the carrier concentration through electrical doping. In other words, the PN diode exhibits a gate-tunable characteristic, with an increase in gate voltage from -60 to 20 V changing the p-n rectifying configuration to n-n junction behavior. The calculated forward/reverse current ratio at V ds = |5 V| clearly shows this gate-tunable PN diode characteristic (inset of Fig. 4(c)). The forward/reverse current ratio of ~80 at V g = -60 V, is higher than previously reported for a PN diode based on 1L MoS 2 /1L WSe 2 (~50 at V ds = |8|V) 45 , but drops to 1.4 at V g = 20 V. This gate-tunable characteristic could be explained by a variation in carrier density with electrical doping, as consistent with a previous report 45 (also explain in Supplementary Fig. S10). Also of note is the fact that this PN diode exhibits a strong PL quenching property and photovoltaic effect, indicating a rapid carrier separation at the MoS 2 /WSe 2 junction 10,42 . Figure 4(d) shows the PL spectra for 1L MoS 2 , 2L WSe 2 and the heterostructure created from them. It is evident from this that the strong PL peak for the direct gap transition of 1L MoS 2 is greatly suppressed by the WSe 2 junction, which is attributed to the rapid separation of charge carriers 10 . Figure 4(e) shows the I-V characteristics of a MoS 2 / WSe 2 PN diode at Vg = -50 V with and without illumination by an incident optical power density of 14 W/m 2 . We see from this that the current increases with illumination due to the generation of optically excited carriers. The open circuit voltage of 0.2 V indicates a photovoltaic effect, with a calculated photoresponsivity of 33 mA/W at V ds = 1 V. As a result, we show the potential of SLS MoS 2 /WSe 2 structure in photovoltaic device as well as PN diode.

Conclusion
In summary, the synthesis of MoS 2 on a SiO 2 substrate has been successfully achieved through a new self-limiting process that allows the number of layers formed to be controlled by varying the growth temperature. Though the precise mechanism requires further study, this behavior is believed to be caused by the lack of dangling bonds on the surface of MoS 2 and the screening effect that MoS 2 layers have on the substrate's electric field. More importantly, this process can achieve excellent layer uniformity (up to 95%) over large areas at wafer-level scale. The resulting 2D MoS 2 can produce n-type behavior and a high on/off ratio when used in a top-gated FET, and can be grown on other chemically inert 2D materials such as WSe 2 . Indeed, a PN diode based on a MoS 2 /WSe 2 heterostructure is capable of a high forward/reverse current ratio, and exhibits a gate-tunable rectifying property attributable to