Controllable synthesis of molybdenum tungsten disulfide alloy for vertically composition-controlled multilayer

Abstract

The effective synthesis of two-dimensional transition metal dichalcogenides alloy is essential for successful application in electronic and optical devices based on a tunable band gap. Here we show a synthesis process for Mo_1−xW_xS₂ alloy using sulfurization of super-cycle atomic layer deposition Mo_1−xW_xO_y. Various spectroscopic and microscopic results indicate that the synthesized Mo_1−xW_xS₂ alloys have complete mixing of Mo and W atoms and tunable band gap by systematically controlled composition and layer number. Based on this, we synthesize a vertically composition-controlled (VCC) Mo_1−xW_xS₂ multilayer using five continuous super-cycles with different cycle ratios for each super-cycle. Angle-resolved X-ray photoemission spectroscopy, Raman and ultraviolet–visible spectrophotometer results reveal that a VCC Mo_1−xW_xS₂ multilayer has different vertical composition and broadband light absorption with strong interlayer coupling within a VCC Mo_1−xW_xS₂ multilayer. Further, we demonstrate that a VCC Mo_1−xW_xS₂ multilayer photodetector generates three to four times greater photocurrent than MoS₂- and WS₂-based devices, owing to the broadband light absorption.

Introduction

The band gap modulation of two-dimensional (2D) transition metal dichalcogenides (TMDCs) has been intensively studied, because of their various applications in optoelectronic devices such as photodiodes, phototransistors and solar cells^1,2,3. It is well known that the band gap of 2D TMDCs is dependent on the number of layers^4,5,6. In addition, alloying 2D TMDCs through the synthesis of Mo_1−xW_xS₂, Mo_1−xW_xSe₂ or MoS_2xSe_2(1−x), for example, is another way of practically modulating the band gap. This is an effective approach because of the good thermodynamic stability at room temperature of the alloys, as predicted by theoretical calculations^{7,8,9,10,11,12,13,14}. Furthermore, recent studies have shown that a vertically composition-controlled (VCC) 2D TMDCs multilayer is feasible for the high performance optoelectronic devices due to functionality of interlayer such as interlayer transition^{15,16,17,18,19,20,21,22}. However, the reported synthesis processes for 2D TMDCs alloy and VCC 2D TMDCs multilayer, such as exfoliation, chemical vapor deposition and transfer, are limited in respect of systematic control of the composition and the number of layers, and clean interface for strong interlayer coupling^9,11,21,23. Hence, an improved synthesis process for 2D TMDCs alloy and VCC 2D TMDCs multilayer is highly required.

Atomic layer deposition (ALD), which is based on surface reactions between precursors and reactants, has benefits such as high purity, thickness control on the atomic scale and large area uniformity^24,25. In particular, it is suitable for the synthesis of alloy thin films with precisely controlled composition using the super-cycle method^26,27,28. In addition, a continuous super-cycle process with different cycle ratios can produce a VCC multilayer with a clean interface²⁹. In a previous report, we have shown that atomically thin, layer-controlled and wafer-level uniform 2D WS₂ can be synthesized by sulfurization of ALD WO₃ thin films³⁰.

Here we report a synthesis method of Mo_1−xW_xS₂ alloys by sulfurization of super-cycle ALD Mo_1−xW_xO_y alloy thin films. Using this method, we systematically control the composition and layer number (from mono- to tri-layers) of Mo_1−xW_xS₂ alloys by controlling the cycle ratio between the ALD MoO_x and WO₃. The bandgaps of the Mo_1−xW_xS₂ alloys are precisely controlled as functions of the composition and layer numbers of each respective alloy, as measured based on the photoluminescence (PL) spectra. Scanning transmission electron microscopy (STEM) shows the mixing of Mo and W atoms with shared metal atom sites in monolayer Mo_1−xW_xS₂ alloy. Furthermore, we develop a process to synthesize a VCC Mo_1−xW_xS₂ multilayer using a sequential super-cycle ALD process—specifically, 5 continuous super-cycles of ALD with different cycle ratios for each super-cycle. Ultraviolet–visible spectrophotometer analysis shows that the synthesized VCC Mo_1−xW_xS₂ multilayer has stronger interlayer coupling than that of a stacked VCC Mo_1−xW_xS₂ multilayer fabricated by the individual transfer of each monolayer Mo_1−xW_xS₂ alloy. This can be attributed to the clean interface between each layer in the synthesized sample^15,21.

Results

MoS₂ synthesis

Previously, we reported the synthesis of WS₂ using sulfurization of ALD WO₃ thin film with a one-step sulfurization process at 1,000 °C (ref. 30). These synthesized WS₂ exhibit smooth and continuous surfaces with layer controllability from mono- to tetra-layer. Based on this result, we sulfurized ALD MoO_x thin film (nine cycles, optimization of ALD MoO_x is represented in Supplementary Fig. 1) using a one-step sulfurization process at 1,000 °C (see experimental section) to synthesize MoS₂. Figure 1a,b comprises scanning electron microscope (SEM) and atomic force microscopy (AFM) images of sulfurized MoO_x thin film using the one-step sulfurization process at 1,000 °C. In contrast to WS₂, however, the MoS₂ shows a rough and non-continuous surface, and the measured root mean square (r.m.s.) is much larger (1.4 nm) than that of the SiO₂ substrate (0.37 nm). We surmise that this discrepancy between the MoO_x and WO₃ thin films sulfurized at the same temperature (1,000 °C) is caused by the relatively lower vaporization temperature of MoO_x (∼700 °C) in comparison with WO₃ (over 1,100 °C)³¹. In other words, the MoO_x is vaporized before the conversion to MoS₂ is complete, resulting in a rough surface.

Figure 1: Sulfurization of MoO_x thin films.

(a) SEM and (b) AFM images of sulfurized MoO_x thin film using one-step sulfurization process at 1,000 °C. SEM and AFM images of sulfurized MoO_x thin film using two-step sulfurization process with first-sulfurization temperatures of (c,d) 600 and (e,f) 800 °C, respectively. Scale bars, (a,c,e) 200 nm and (b,d,f) 0.5 μm.

Therefore, we examined the effect of sulfurization temperature on the roughness of the sulfurized ALD MoO_x thin film. To achieve this, we conducted a two-step sulfurization process, which consists of a low-temperature first step for the sulfurization of the MoO_x and a high-temperature second step to enhance the MoS₂ crystallinity. The first-sulfurization temperatures were set to lower (600 °C) and higher (800 °C) temperatures than the vaporization temperature of MoO_x (700 °C), while the second-sulfurization temperature and process time were kept at 1,000 °C and 150 min, respectively (see Methods section). The roughness of the sulfurized MoO_x thin films in accordance with first-sulfurization temperature was then compared using SEM and AFM (Fig. 1c–f). Figure 1c,e shows SEM images of the sulfurized MoO_x thin films for first-sulfurization temperatures of 600 °C and 800 °C, respectively. The sulfurized MoO_x thin film at a first-sulfurization temperature of 600 °C has a smooth and continuous surface, while the MoO_x thin film sulfurized at a first-sulfurization temperature of 800 °C has a rough and non-continuous surface. AFM analyses (Fig. 1d,f) illustrate the variations in the roughness of the sulfurized MoO_x thin films more clearly, which is due to the differing first-sulfurization temperatures. The r.m.s. value for the MoS₂ sulfurized at 600 °C is very low (∼0.4 nm) and is close to the r.m.s. value of the SiO₂ substrate (0.37 nm). In contrast, the r.m.s. value of the MoS₂ in the 800 °C case is relatively high (0.8 nm). As a result, a first-sulfurization temperature of 600 °C results in MoS₂ with uniform and continuous surfaces, due to the fact that the first-sulfurization temperature is lower than the vaporization temperature of MoO_x, as we assumed. Based on this result, we used a two-step sulfurization process with a 600 °C first-sulfurization temperature to synthesize continuous MoS₂ and Mo_1−xW_xS₂ alloys.

Next, layer-number-controlled MoS₂ was synthesized utilizing the two-step sulfurization process described above. Figure 2a–d shows the AFM images and height profiles of the transferred MoS₂, which were synthesized by sulfurizing MoO_x thin films deposited by 6, 9 and 12 ALD cycles. The measured thicknesses of the synthesized MoS₂ were ∼1, 1.6 and 2.3 nm for 6, 9 and 12 MoO_x ALD cycles, respectively. These thicknesses correspond to mono-, bi- and tri-layer (1, 2 and 3l) MoS₂, considering that the height of 1l MoS₂ on SiO₂ is ∼1 nm and the spacing between the first and second MoS₂ layers is ∼0.6 nm (refs 3, 4). As reported previously, the larger AFM-measured spacing between the first MoS₂ layer and the substrate, compared with that between the MoS₂ layers, is caused by the effect of distinct tip–sample and tip–substrate interactions^3,30,32. Also, the apparent colour gains of the transferred 1, 2 and 3l MoS₂ are observed in optical microscopy (OM) images (Supplementary Fig. 2). It should be noted that the MoS₂ is not formed in the case of an ALD MoO_x thin film with an ALD cycle number of <3 (Supplementary Fig. 3). This is attributed to a nucleation delay during the initial growth of the MoO_x, and similar behaviour was observed during the synthesis of WS₂ by sulfurization of ALD WO₃ (ref. 30). After the nucleation delay, 1l of MoS₂ is formed by the sulfurization of each three-cycle ALD MoO_x thin film sample (∼0.8−0.9 nm in thickness). This observation agrees with a previous report, where ∼1 nm of MoO_x film transformed into a 1l MoS₂ via sulfurization³³. The stoichiometry calculated from X-ray photoemission spectroscopy (XPS) result is 2 (S/Mo) as shown in Supplementary Fig. 4. As a result, we can systematically control the layer number of MoS₂ by controlling the ALD MoO_x cycle number.

Figure 2: Characterization of MoS₂.

(a–c) AFM images and (d) height profiles (along with white dashed line in AFM images) of transferred MoS₂ on SiO₂ substrate for 1l, 2l and 3l thickness, respectively. Scale bars, 0.5 μm. (e) Raman spectra and (f) PL spectra for 1l (red), 2l (blue) and 3l (black) MoS₂ on SiO₂ substrate. (g) HRTEM image of 1l MoS₂ at a selected region and (inset) FFT pattern. Scale bars, 2 nm.

The MoS₂ were further characterized using Raman, PL and high-resolution TEM (HRTEM). The Raman spectra (λ_exc=532 nm) for 1, 2l and 3l MoS₂ are shown in Fig. 2e. The MoS₂ exhibit in-plane and out-of-plane vibrations modes at 386.6 and 406.5 cm⁻¹ (E′ and A′₁) for the 1l, 385.6 and 407.6 cm⁻¹ (E_g¹ and A_1g) for 2l, and 384.7 and 408.5 cm⁻¹ (E′¹ and A′₁) for 3l (ref. 34). From the Raman spectra, we calculated the relative peak distance between the in-plane and out-of-plane modes, which is closely related to the layer number of the MoS₂ due to the softening in the in-plane and stiffening in the out-of-plane mode frequencies, with increasing layer numbers^35,36. The calculated relative peak distances are 19.9, 22 and 23.8 cm⁻¹ for the 1, 2 and 3l samples, respectively, which are in good agreement with previously reported values for synthesized MoS₂ (refs 37, 38, 39).

The PL spectra dependence on the layer number of the MoS₂ is shown in Fig. 2f. The spectrum of the 1l MoS₂ shows PL peaks at 1.89 eV and 2.01 eV, which are correlated to the A₁ and B₁ direct excitonic transitions of the MoS₂, respectively. With increasing layer number, weak PL peaks are observed at 1.87 eV and 2.00 eV for the 2l, and 1.86 eV and 1.99 eV for the 3l. The red shift and low intensity of the PL peaks with increasing layer number is due to the band gap transition from direct to indirect, which is consistent with the dependency of the PL peak on the layer number^4,5,6,40. These Raman and PL results again confirm the layer controllability of MoS₂ using the ALD process. Figure 2g is an HRTEM image for the synthesized 1l MoS₂. The MoS₂ shows a honeycomb-like structure with lattice spacing of 0.27 nm and 0.16 nm for the (100) and (110) planes, respectively. In addition, sixfold coordination symmetry is observed in the fast Fourier transformation (FFT) image (inset of Fig. 2g). The approximate domain size is 10−20 nm, similar to that of previously reported synthesized MoS₂ and WS₂ using the sulfurization of MoO_x and WO₃ thin films^30,41.

Mo_1−xW_xS₂ alloy synthesis

A super-cycle ALD-based Mo_1−xW_xS₂ alloy synthesis process was developed based on the synthesis processes for 2D MoS₂ (this study) and WS₂ (previous study)³⁰. The overall synthesis scheme for the Mo_1−xW_xS₂ alloy is illustrated in Fig. 3a. First, we conducted 10 cycles of WO₃ ALD to address the nucleation delay of the ALD WO₃ (ref. 30) (not shown in Fig. 3a). Subsequently, one super-cycle ALD process consisting of n cycles of ALD MoO_x and m cycles of ALD WO₃ was conducted and the deposited Mo_1−xW_xO_y alloy thin films were sulfurized. We used varying cycles for MoO_x (n) and WO₃ (m) in one super-cycle to deposit 0.8−0.9-nm-thick composition-controlled Mo_1−xW_xO_y alloy thin films to create a 1l Mo_1−xW_xS₂ alloy. This was based on the growth rate of ALD MoO_x (2.7 Å per cycle) and WO₃ (0.9 Å per cycle), as shown in Supplementary Table 1. Figure 3b–d shows the XPS spectra of the 1l MoS₂, 1l WS₂ and sulfurized Mo_1−xW_xO_y alloy thin films with different n and m numbers in one super-cycle. All measured XPS results were normalized by S2p_3/2 peak intensity and calibrated to the C1s peak at 285 eV. With increasing n/m ratio, the intensity of the Mo3d peaks increased, while the W5p_3/2 and W4f peaks decreased. Furthermore, the peak positions for Mo3d and W4f gradually shifted to higher binding energies, from 232.2 eV and 229.1 eV to 232.5 eV and 229.4 eV for Mo3d_3/2 and Mo3d_5/2, respectively, and from 34.8 eV and 32.6 eV to 35.0 eV and 32.8 eV for W4f_5/2 and W4f_7/2, respectively. In addition, the S2p peaks shifted to lower binding energies, from 163.5 eV and 162.4 eV to 163.3 eV and 162.2 eV for S2p_1/2 and S2p_3/2, respectively. This small shift in peak position is attributed to the enhanced electron attraction strength of S and the reduced electron attraction strength of W, following increased Mo content due to smaller electronegativity of Mo (2.16) than that of W (2.36) as previously reported⁷. It is noteworthy that the Mo⁶⁺ 3d_3/2 peak, which is attributed to the Mo–O bonding, is not observed in the Mo3d spectra; this indicates the absence of O species.

Figure 3: Synthesis and XPS of Mo_1−xW_xS₂ alloy.

(a) Synthesis procedure of super-cycle ALD for Mo_1−xW_xS₂ alloy. XPS measurements for (b) Mo3d, (c) W4f and (d) S2p core levels in the 1l Mo_1−xW_xS₂ alloy with different n and m numbers in one super-cycle. All measured XPS results are normalized by S2p_3/2 peak intensity.

We calculated the Mo, W and S concentrations from the XPS results for the Mo3d, W4f and S2p peaks, respectively, to examine the Mo_1−xW_xS₂ alloy composition. Table 1 presents the calculated concentration and W composition, x. The calculated x value is dependent on the n and m numbers in a single super-cycle, and yields x=0.8 for n=1 and m=6, x=0.6 for n=2 and m=4, and x=0.3 for n=3 and m=1. Also, the calculated stoichiometry is 2 (S/(Mo+W)). This shows that the W composition (x) in the Mo_1−xW_xS₂ alloys can be systematically modulated by changing the values of n and m in one super-cycle.

Table 1 Calculated composition of Mo_1−xW_xS₂.

The synthesized composition-controlled Mo_1−xW_xS₂ alloy from super-cycle ALD Mo_1−xW_xO_y alloy thin films were characterized using AFM, Raman and PL, as shown in Fig. 4. The AFM images and height profiles of the transferred Mo_1−xW_xS₂ alloys are represented in Fig. 4a−d and they show good uniformity and continuity (also see OM images in Supplementary Fig. 5). The measured thicknesses of the Mo_0.2W_0.8S₂, Mo_0.4W_0.6S₂ and Mo_0.7W_0.3S₂ alloys were all ∼1 nm, corresponding to the 1l thickness of Mo_1−xW_xS₂ alloy. Furthermore, 2 and 3l Mo_1−xW_xS₂ alloys can be synthesized using two- and three-super-cycle ALD Mo_1−xW_xO_y alloy thin films (Supplementary Fig. 6). As a result, our super-cycle ALD-based Mo_1−xW_xS₂ alloy synthesis process can systematically control the layer number, as well as the composition of the resultant alloys through manipulation of the super-cycle ALD process.

Figure 4: Characterization of Mo_1−xW_xS₂ alloy.

(a–c) AFM images and (d) height profiles (along with white dashed line in AFM images) of transferred 1l Mo_1−xW_xS₂ alloy on SiO₂ substrate for x=0.8, 0.6 and 0.3, respectively. Scale bars, 0.5 μm. (e) Raman spectra and (f) PL spectra for 1l Mo_1−xW_xS₂ alloy on SiO₂ substrate for x=1, 0.8, 0.6, 0.3 and 0. (g) PL peak position versus W composition (x) graph. Error bars represent s.d. of PL peak position in five-times repeatedly synthesized Mo_1−xW_xS₂ alloy.

Figure 4e shows the Raman spectra of composition-controlled 1l Mo_1−xW_xS₂ alloys. The 1l WS₂ (x=1) exhibits first-order modes: out-of-plane (A′₁) and in-plane (E′) modes at 417 cm⁻¹ and 357 cm⁻¹, respectively, and a second-order mode: 2LA(M) at 353 cm⁻¹ (ref. 30). The A′₁ mode shifts to a lower frequency with decreasing W composition, while the E′ mode related to WS₂ does not noticeably shift with the reduction of intensity. In addition, an E′ mode related to MoS₂ appear at x=0.8 and shifted to a higher frequency with a reduction in W composition. The specific peak position dependency on W composition is represented in Supplementary Fig. 7, and the W composition dependence of the Raman spectra of the 1l Mo_1−xW_xS₂ alloy is consistent with previous reports^7,8,9.

The normalized PL spectra of the composition-controlled 1l Mo_1−xW_xS₂ alloys are shown in Fig. 4f, also the x values versus the average PL peak positions and s.d. of five-times repeatedly synthesized 1l Mo_1−xW_xS₂ alloys are plotted in Fig. 4g. As the value of x increased from 0 to 1, the averaged PL peak position initially decreases from 1.885 to 1.863 eV, and then gradually increases to 2.021 eV. This non-linear PL peak position behaviour with changing x is the so-called ‘bowing effect’, and has also been reported for other semiconducting alloys and exfoliated 1l Mo_1−xW_xS₂ alloys^9,42,43. The bowing effect in 1l Mo_1−xW_xS₂ alloy can be described by the Equation (1),

where b is a bowing parameter. After fitting the experimental results as shown in Fig. 4g (red solid curve), a b value of 0.25±0.03 eV was extracted. The extracted b value is comparable to that of the previous experiment (0.25±0.04 eV) and simulation (0.28±0.04 eV) results⁹. Furthermore, the s.d. of the five-times repeatedly synthesized 1l Mo_1−xW_xS₂ alloys are small within the range of 0.008 to 0.01, which indicates that the process has good reliability in terms of composition control. Thus, the PL result confirms that we modulate the band gap of the Mo_1−xW_xS₂ alloy by reliably controlling the composition. Moreover, the band gap can also be modulated by controlling the layer number (see Supplementary Fig. 8).

Figure 5a is the HRTEM image of the 1l Mo_0.4W_0.6S₂ alloy (x=0.6). The Mo_0.4W_0.6S₂ alloy shows a periodic atomic arrangement with a honeycomb-like structure and sixfold coordination symmetry, similar to the 1l MoS₂ shown in Fig. 2f. To distinguish between the W and Mo atoms in the 1l Mo_0.4W_0.6S₂ alloy, we analysed the Mo_0.4W_0.6S₂ alloy using STEM annular dark-field and energy dispersive X-ray spectrometry (EDX). Figure 5b is the STEM-ADF image of the 1l Mo_0.4W_0.6S₂ alloy. Brighter and less bright spots, which correspond to W and Mo atoms, respectively, are clearly resolved in the ADF image, as previously reported⁴⁴. The calculated Mo/W ratio from the atom count in Fig. 5b is 0.42:0.58, which differs by <5% from the XPS-measured stoichiometry. In addition, the EDX result in Fig. 5c supports the presence of W, Mo and S species in the 1l Mo_0.4W_0.6S₂ alloy. To extract a clear intensity difference between the W and Mo atoms, we performed an inverse FFT by applying a mask to the yellow dashed square region in Fig. 5b. Figure 5d–e shows the inversed FFT image (Fig. 5d) and intensity profile (Fig. 5e) along with the yellow solid line in Fig. 5d. Although S atoms are not distinguishable in our result as a result of the displacement of S atoms at 200 kV operation voltage by the knock-on mechanism⁴⁵, the W and Mo atoms are clearly observable, confirming that these elements share the metal atom sites⁴⁴. The preference for Mo or W atoms at the neighbouring sites of W atoms is evaluated by degree of alloying that can be calculated by Equation (2)^23,44,

Figure 5: Atomic arrangement and mixture of Mo_0.4W_0.6S₂ alloy.

(a) HRTEM image of 1l Mo_0.4W_0.6S₂ alloy at a selected region, and (inset) FFT pattern. Scale bars, 2 nm. (b) STEM-ADF image of 1l Mo_0.4W_0.6S₂ alloy at a selected region and (c) corresponding EDX spectrum. Scale bars, 1 nm. (d) Inverse FFT image with masking applied to yellow dashed square region in b. Scale bars, 1 nm. (e) Intensity profile of yellow solid line in d. (f) Coloured W atoms with light brown, blue, red, dark red, yellow, green and violet for six, five, four, three, two, one and zero number of neighbouring Mo atoms.

where P_observed is the averaged ratio of number of neighbouring Mo atoms to total neighbouring sites of W atoms, and P_random is the total ratio of Mo atoms in the examined layer. Figure 5f represented differently coloured W atoms depending on number of neighbouring Mo atoms: light brown, blue, red, dark red, yellow, green and violet for six, five, four, three, two, one and zero number of neighbouring Mo atoms. The calculated degree of alloying is 99%, which indicate that there is no preference for Mo or W atoms at the neighbouring sites of W atoms and a random mixture of our 1l Mo_1−xW_xS₂ alloy.

A VCC Mo_1−xW_xS₂ synthesis

The composition controllability of our ALD-based Mo_1−xW_xS₂ alloy synthesis process enables synthesis of a VCC Mo_1−xW_xS₂ multilayer with a clean interface, strong interlayer coupling and broadband light absorption. We sulfurized a VCC Mo_1−xW_xO_y thin film that was deposited by a sequential super-cycle ALD process, so as to synthesize a VCC Mo_1−xW_xS₂ multilayer, as shown in Fig. 6a. First, we conducted 20 cycles of WO₃ ALD on a SiO₂ substrate, corresponding to 1l WS₂. We immediately performed three super-cycles of Mo_1−xW_xO_y ALD with different super-cycle n and m numbers, in the following order: n=1 and m=6, n=2 and m=4, and n=3 and m=1. Last, we conducted three cycles of MoO_x ALD (n=3) corresponding to 1l MoS₂. The deposited VCC Mo_1−xW_xO_y thin film was sulfurized to convert it into a VCC Mo_1−xW_xS₂ multilayer. Figure 6b,c shows an AFM image and height profile of the transferred VCC Mo_1−xW_xS₂ multilayer, with a measured thickness of ∼3.5 nm. This thickness, synthesized by five sequential ALD super-cycles, corresponds to a 5l Mo_1−xW_xS₂ alloy, which is consistent with each super-cycle result for the 1l Mo_1−xW_xS₂ alloy.

Figure 6: Characterization of VCC Mo_1−xW_xS₂ multilayer.

(a) Sequential super-cycle ALD procedure and schematic structure of a VCC Mo_1−xW_xS₂ multilayer. (b) AFM image and (c) height profiles (along with white dashed line in AFM image) for a VCC Mo_1−xW_xS₂ multilayer. Scale bars, 0.5 μm. (d) Calculated atomic concentration and relative concentration ratio of Mo and W from ARXPS measurement. (e) Raman spectra for a VCC Mo_1−xW_xS₂ multilayer. (f) Raman peak position of A_1g and MoS₂-like E¹_2g modes from fitted Raman spectra (red and blue solid line) and from measured Raman spectra of 1l Mo_1−xW_xS₂ alloy (black dashed line). (g) Calculated Raman peak distances between A_1g and MoS₂-like E¹_2g modes from fitted Raman spectra (red solid line) and from measured Raman spectra of 1l Mo_1−xW_xS₂ alloy (black dashed line).

The different composition concentrations of the bottom and top layers in the VCC Mo_1−xW_xS₂ multilayer were analysed using angle-resolved XPS (ARXPS). Figure 6d shows the calculated atomic and relative concentration ratios of the Mo and W from the ARXPS measurement (ARXPS spectra are shown in Supplementary Fig. 9). The Mo concentration increased from 18.6 to 20.9%, while the W concentration decreased from 15.7 to 13.5%, with increasing emission angle from 0 to 70° (red line). The Mo/W concentration ratio increased from 1.17 to 1.55 with increasing emission angle (blue line). Although the exact atomic concentration according to position in the VCC Mo_1−xW_xS₂ multilayer cannot be calculated because of the larger depth resolution of the XPS measurement in comparison with the VCC Mo_1−xW_xS₂ multilayer thickness, the emission angle dependency of the Mo and W concentration indicates Mo-rich and W-rich concentration in the upper and lower layers of the VCC Mo_1−xW_xS₂ multilayer, respectively. As a result, ARXPS shows that the VCC Mo_1−xW_xS₂ multilayer has VCC characteristics. Notably, the calculated stoichiometry ratio was 2 (S/(Mo+W)) in all ARXPS results.

The formation of Mo_1−xW_xS₂ alloy with different compositions in a VCC Mo_1−xW_xS₂ multilayer was analysed using Raman spectroscopy. Figure 6e shows the Raman spectrum of a VCC Mo_1−xW_xS₂ multilayer, which exhibits strong peaks for A_1g, MoS₂-like E¹_2g and WS₂-like E¹_2g+2LA(M) modes. Each Raman peak can be fitted using a Lorentzian function to the Raman spectrum of the Mo_1−xW_xS₂ alloy with x=0, 0.3, 0.6, 0.8 and 1. The fitted Raman spectrum was compared with the measured Raman spectrum for the 1l Mo_1−xW_xS₂ alloy, with respect to variations in the peak position and peak distances of the A_1g and MoS₂-like E¹_2g modes, depending on W concentration. The A_1g and MoS₂-like E¹_2g peak positions from the fitted Raman spectrum are represented in Fig. 6f with the measured Raman peak positions for the 1l Mo_1−xW_xS₂ alloy (black dashed line, the same as Supplementary Fig. 7). The variation in the fitted Raman peak position with increasing W concentration in the Mo_1−xW_xS₂ alloy is the same as the variation in the measured Raman peak position for the 1l Mo_1−xW_xS₂ alloy: A_1g shifts to a higher frequency with an increase in W concentration, while the MoS₂-like E¹_2g modes downshift. Figure 6g shows peak distances between the A_1g and MoS₂-like E¹_2g modes from the fitted Raman spectrum (red solid line) and measured Raman spectrum of the 1l Mo_1−xW_xS₂ alloy (black dashed line), which are 3−4 cm⁻¹ larger than that of the Raman spectrum of the 1l Mo_1−xW_xS₂ alloy. This is due to the softening in the MoS₂-like E¹_2g mode frequency and stiffening in the A_1g mode frequency. Similar behaviour, that is, increasing peak distances with increasing layer number, is also observed in MoS₂ (refs 35, 36) and WS₂ (ref. 30) because of the reduced long-range Coulomb interaction between the effective charges, which is induced by an increase in the dielectric screening. These results for the fitted Raman spectra are in good agreement with the dependency of the peak positions on the W composition given by the measured Raman results, and the dependency of the peak distances on layer number in 2D TMDCs. Thus, we can conclude that the fitted Raman spectra show the formation of a Mo_1−xW_xS₂ alloy with different compositions in a VCC Mo_1−xW_xS₂ multilayer.

As a result, it can be stated that the ARXPS and Raman results show the VCC characteristics of a VCC Mo_1−xW_xS₂ multilayer. Also, these findings indicate that the vertical interdiffusion of the Mo and W atoms during the sulfurization process have no critically effect on the VCC characteristics. A similar result was observed in a previous report, in that MoO_x/WO₃ thin film was converted to MoS₂/WS₂ without the formation of a Mo_1−xW_xS₂ alloy, indicating the limited interdiffusion of Mo and W atoms⁴⁶. Further, it is noteworthy that we verified the validity of ARXPS and Raman measurements as a means of characterizing the VCC Mo_1−xW_xS₂ multilayer via characterization of a VCC Mo_1−xW_xS₂ multilayer synthesized with a reversed vertical composition profile (see Supplementary Fig. 10).

Since the interlayer coupling affects interlayer transition^15,21,47,48, strong interlayer coupling in a synthesized VCC Mo_1−xW_xS₂ multilayer was evaluated using comparison of interlayer transition in three difference sample types as shown in Fig. 7a. Sample 1 is a stacked VCC Mo_1−xW_xS₂ multilayer fabricated by the transfer of each differently composed Mo_1−xW_xS₂ alloy onto glass substrate, while sample 2 is the same as sample 1 but annealed at 200 °C for 15 min in an Ar ambient atmosphere to enhance the interlayer coupling by the removal of residual molecules^21,48. Sample 3 is a transferred VCC Mo_1−xW_xS₂ multilayer on glass substrate, which was annealed at 200 °C for 15 min in an Ar ambient atmosphere. Ultraviolet–visible spectrophotometer measurements for samples 1, 2 and 3 (Fig. 7b) illustrate that these have broadband light absorption properties due to the sum of the light absorption from the differently composed Mo_1−xW_xS₂ alloys. In previous reports, the absorption spectrum of the interlayer transition could be obtained by comparing the intensity difference between the absorption spectra of the weakly interlayer-coupled sample and that of the strongly interlayer-coupled sample^15,48. Based on these reports, we extracted the interlayer transition absorption spectrum by subtracting the absorption spectrum of sample 1 from that of sample 2 and of sample 3, since sample 1 has the weakest interlayer coupling of the three samples as a result of the contamination at the interface caused by the layer transfer process^15,21,48. The extracted absorption spectra of the interlayer transition are shown in Fig. 7c. The sample 2–sample 1 spectrum (black solid line) shows a small absorbance peak at 1.87 eV, while the sample 3–sample 1 spectrum (red solid line) shows an absorbance peak that is over five times stronger than the sample 2–sample 1 absorbance peak at the same position. Specific observations on the origin of the absorbance peak position from the interlayer transition (1.87 eV) are described in the Supplementary Information (Supplementary Fig. 11). The stronger absorbance peak of sample 3–sample 1 in comparison with that of sample 2–sample 1 indicates that sample 3 has stronger interlayer coupling compared with sample 2. In other words, a VCC Mo_1−xW_xS₂ multilayer based on sequential super-cycle ALD has the strongest interlayer coupling among the three types of samples. We surmise that this strong interlayer coupling results from the absence of a transfer process, which eliminates the incorporation of residual molecules such as H₂O and organic contaminants^15,21,47,48.

Figure 7: Absorbance and photoinduced current of VCC Mo_1−xW_xS₂ multilayer.

(a) Schematics of three sample types for ultraviolet–visible spectrophotometer measurement. (b) Absorption spectra of sample 1 (black solid line), sample 2 (blue solid line), and sample 3 (red solid line) and (c) extracted absorption spectra of interlayer transition using subtraction of sample 1 from sample 2 (black solid line) and from sample 3 (red solid line). (d) Spectral and (e) time-resolved photocurrent of a VCC Mo_1−xW_xS₂ multilayer, 5l WS₂ and 5l MoS₂ photodetectors.

A VCC Mo_1−xW_xS₂ multilayer exhibits a broadband light absorption property, as well as strong interlayer coupling. Thus, the VCC Mo_1−xW_xS₂ multilayer has promising potential use as an active layer in an efficient photodetector. To evaluate the photoinduced response of the VCC Mo_1−xW_xS₂ multilayer, we observed the spectral and time-resolved photocurrent of a VCC Mo_1−xW_xS₂ multilayer photodetector and compared it with 5l WS₂ and 5l MoS₂ photodetectors (see the Methods section for details of device fabrication, and see Supplementary Fig. 12 for AFM images of the 5l MoS₂ and WS₂ and their I–V characteristics). Figure 7d shows the dependence of the photocurrent on the illumination energy for the VCC Mo_1−xW_xS₂ multilayer, 5l WS₂, and 5l MoS₂ photodetectors for a voltage drain to source (V_ds) of 5 V. The continuum power spectral density is represented in Supplementary Fig. 13. The VCC Mo_1−xW_xS₂ multilayer photodetector generates a broadband photoinduced current from 1.2 to 2.5 eV, because of its broadband light absorption property. In contrast, the 5l WS₂ and 5l MoS₂ photodetectors generate narrower photocurrents than the VCC Mo_1−xW_xS₂ multilayer, at 1.3 and 2.1 eV for the 5l WS₂ photodetector and 1.2 and 1.8 eV for the MoS₂ device; these values correspond to the 5l WS₂ and MoS₂ bandgaps. We then examined the time-resolved photocurrent measurement using white-light illumination, as shown in Fig. 7e (result using specific laser wavelength is shown in Supplementary Fig. 14). The white light was first turned off for a period of 5 s, and then turned on for 5 s with the biasing V_ds=5 V. The drain current (I_ds) increased on activation of the light and decayed following removal of the incident light. The induced photocurrents were 39 pA, 13 pA and 11 pA for the VCC Mo_1−xW_xS₂ multilayer, 5l WS₂ and 5l MoS₂ devices, respectively. Hence, the VCC Mo_1−xW_xS₂ multilayer generates three to four times greater photocurrent than 5l WS₂ or 5l MoS₂, which is attributed to broadband light absorption. Thus, we concluded that the VCC Mo_1−xW_xS₂ multilayer is promising as regards use as an efficient photodetector with broadband light absorption. Furthermore, the broadband light absorption property is feasible for various optoelectronic applications such as solar cells^49,50.

Discussion

In summary, we developed an ALD-based Mo_1−xW_xS₂ synthesis process using sulfurization of super-cycle ALD Mo_1−xW_xO_y thin film. We studied the sulfurization process of ALD MoO_x thin films to produce uniform and continuous MoS₂. The synthesized ALD-based Mo_1−xW_xS₂ alloy show good stoichiometry, uniform and continuous surfaces, controlled composition and layer numbers, and mixing of Mo and W atoms. Moreover, we developed a simple method to synthesize a VCC Mo_1−xW_xS₂ multilayer with a clean interface, which shows stronger interlayer coupling than that of a stacked VCC Mo_1−xW_xS₂ multilayer fabricated using the transfer process. Further, we have shown that the VCC Mo_1−xW_xS₂ multilayer has promising potential applications as an efficient photodetector, because of its broadband light absorption capability. It should also be noted that the ALD-based TMDCs alloy synthesis process is not only limited to Mo_1−xW_xS₂, and we expect that similar process strategies can be developed for other TMDCs materials and their vertical stacks.

Methods

MoO_x film growth and characteristics

A 6-inch ALD chamber containing a loadlock chamber was used for the deposition of the MoO_x films. The films were deposited on SiO₂(300 nm)/Si substrates by plasma-enhanced ALD using Mo(CO)₆ and O₂ plasma at a 200 °C growth temperature. The temperature of the bubbler containing Mo(CO)₆ was maintained at 35 °C to produce adequate vapour pressure, and vapourized Mo(CO)₆ molecules were transported into the chamber by pure argon (99.999%) carrier gas. The O₂ flow and plasma power were fixed at 300 s.c.c.m. and 200 W, respectively. An ALD cycle consists of four steps: Mo(CO)₆ precursor exposure (t_s), Ar purging (t_p), O₂ plasma reactant exposure (t_r) and another Ar purging (t_p). In the ALD MoO_x process, the t_s, t_p and t_r were fixed at 5 s, 12 s and 5 s, respectively. Optimization of the ALD MoO_x process is described in the Supplementary Fig. 1.

Mo_1−xW_xO_y Film Growth

MoO_x and WO₃ ALD processes²⁷ were used to deposit Mo_1−xW_xO_y film using super-cycle ALD (as shown in Fig. 3a) under the same chamber and deposition conditions described above for the ALD of MoO_x. After 10 cycles of WO₃ ALD to address nucleation delay²⁷, we conducted super-cycle ALD, which consists of n cycles of MoO_x ALD and m cycles of WO₃ ALD. The detailed process steps are shown in Supplementary Table 1.

Sulfurization processes

One-step process. To sulfurize the ALD MoO_x, the sample was placed in the centre of a tube furnace (1.2 inch in diameter). Initially, the sample was heated at 200 °C for 60 min under flowing H₂ (25 s.c.c.m.) and Ar (25 s.c.c.m.) gas, to remove any organic contaminants on the surface. Subsequently, the temperature was gradually increased from 200 to 1,000 °C at 13.3 °C min⁻¹, and this temperature was then maintained for 60 min with flowing Ar (50 s.c.c.m.) and H₂S (5 s.c.c.m.). Then, the sample was cooled to room temperature under a flowing Ar (50 s.c.c.m.) atmosphere.

Two-step process. Initially, samples were annealed at 200 °C as in the one-step process. Then, the temperature was gradually increased from 200 °C to first-sulfurization temperatures of 600 or 800 °C at 13.3 °C min⁻¹. The peak temperature (600 or 800 °C) was maintained for 60 min with flowing Ar (50 s.c.c.m.) and H₂S (5 s.c.c.m.). Subsequently, the temperature was gradually increased from the first-sulfurization temperatures to 1,000 °C at 13.3 °C min⁻¹ and was maintained at this temperature for 30 min with flowing Ar (50 s.c.c.m.) and H₂S (5 s.c.c.m.). The sample was cooled to room temperature under a flowing Ar (50 s.c.c.m.) atmosphere after the process was completed. Based on the results shown in Fig. 1, we used a two-step sulfurization process with a 600 °C first-sulfurization temperature to sulfurize MoO_x and Mo_1−xW_xO_y thin films in the MoS₂ and Mo_1−xW_xS₂ alloy synthesis process.

Transfer of MoS₂ and Mo_1−xW_xS₂

The as-synthesized MoS₂ and Mo_1−xW_xS₂ on the SiO₂ substrate were coated with polymethyl methacrylate (PMMA) by spin coating at 4,000 r.p.m. for 60 s. After curing of the PMMA at 100 °C for 15 min, the samples were immersed in 10% hydrogen fluoride solution to etch the SiO₂ layer. Subsequently, the samples were washed using deionized (DI) water and scooped onto a clean SiO₂/Si substrate. The PMMA was removed by acetone and washed away using isopropyl alcohol.

Characterization of MoS₂ and Mo_1−xW_xS₂

OM (Olympus DX51), Raman spectroscopy (HORIBA, Lab Ram ARAMIS; 532-nm laser excitation wavelength), AFM (VEECO, Multimode), PL (SPEX1403, SPEX; 532-nm laser excitation wavelength), absorbance with ultraviolet–visible spectrophotometer (JASCO Corporation, V-650), XPS (Thermo UK, K-alpha), SEM (JEOL Ltd, JSM-6701F), TEM (FEI Titan G2 Cube 60-300; accelerating voltage, 80 kV), STEM and EDX (JEM 2100F; accelerating voltage, 200 kV) analyses were employed to characterize the MoS₂ and Mo_1−xW_xS₂ alloy, and a VCC Mo_1−xW_xS₂ multilayer.

Fabrication and characterization of photodetectors

Photodetectors were fabricated from an as-synthesized VCC Mo_1−xW_xS₂ multilayer, 5l WS₂ and 5l MoS₂ on a SiO₂ (300 nm)/Si substrate by evaporating Au(40 nm)/Ti(1 nm) electrodes with 100-μm channel length. Electrical measurements were conducted using a Keithley 2400 (Keithley Instruments). The photocurrent was measured by modulating the laser beam with a mechanical chopper (1,000 Hz) and detecting the photocurrent with a current preamplifier and a lock-in amplifier. A monochromator was used for wavelength-dependent measurements of the photocurrent.

Parameters for analysis of XPS and Raman

We used Spectral Data Processor v4.1 for the XPS and Raman spectra fitting. In the fitting analysis of the XPS spectra, the full widths at half maximum (FWHM) were between 1.7 and 1.9 eV, the Lorentzian Gaussian Ratio was 2:8, the energy difference between the Mo3d spin-orbit doublet was set to 3.2 eV and the branching ratio was 2/3. In addition, we used Scofield Relative Sensitivity Factor for calculation of stoichiometry as represented in Supplementary Table 2. For the Raman spectrum fitting analysis, the FWHM was between 7 and 12 cm⁻¹.