Lithographically defined synthesis of transition metal dichalcogenides

Transition metal dichalcogenides (TMDs) promise to revolutionize optoelectronic applications. While monolayer exfoliation and vapor phase growth produce extremely high quality 2D materials, direct fabrication at wafer scale remains a significant challenge. Here, we present a method that we call ‘lateral conversion’, which enables the synthesis of patterned TMD structures, with control over the thickness down to a few layers, at lithographically predefined locations. In this method, chemical conversion of a metal-oxide film to TMD layers proceeds by diffusion of precursor propagating laterally between silica layers, resulting in structures where delicate chalcogenide films are protected from contamination or oxidation. Lithographically patterned WS2 structures were synthesized by lateral conversion and analyzed in detail by hyperspectral Raman imaging, scanning electron microscopy and transmission electron microscopy. The rate of conversion was investigated as a function of time, temperature, and thickness of the converted film. In addition, the process was extended to grow patterned MoS2, WSe2, MoSe2 structures, and to demonstrate unique WS2/SiO2 multilayer structures. We believe this method will be applicable to a variety of additional chalcogenide materials, and enable their incorporation into novel architectures and devices.


Introduction
Nanoscale materials with reduced dimensionality, such as zero-dimensional (0D) nanoparticles, onedimensional (1D) nanowires, and two-dimensional (2D) layered materials, have been the subject of intense research efforts in recent years due to the many emergent properties that arise at the nanoscale [1]. With the advent of a relatively easy-to-implement exfoliation process for creating one-off devices out of graphene, interest in 2D materials has surged. Transition metal dichalcogenides (TMDs), a family of layered 2D van der Waals materials, have gained interest from the research community thanks to their unique and highly tunable optical and electronic properties [2][3][4]. There have been considerable efforts to synthesize high-quality TMDs, both for fundamental studies and for integration in next generation devices. Early work [5][6][7] focused on the preparation of these materials by exfoliation from bulk crystals-similar to those for graphene. While this method is adequate for fundamental studies, it is limited when it comes to large-scale device integration since reproducible patterning and contamination-free fabrication is challenging. To overcome this limitation, a variety of direct-synthesis techniques have been developed-the most prominent among these being chemical vapor deposition (CVD) [8,9], metalorganic chemical vapor deposition (MOCVD) [10], and atomic layer deposition (ALD) [11]. Some hybrid approaches, such as post-growth annealing of ALD deposited nanocrystalline chalcogenide films [12], or chemical conversion of ALD oxides [13,14], have also been explored.
There are several key considerations for the successful fabrication and subsequent device integration of TMDs. (1) Controllable synthesis. This involves the growth of materials with high crystallinity, at specific Lithographically defined synthesis of transition metal dichalcogenides locations, and of desired thickness. (2) Material properties tuning. Tuning TMD properties can be accomplished by adjusting the structure itself, either in situ or post-growth (e.g. varying number of layers, intercalation, and doping) or by designing its environment [15][16][17][18][19]. (3) Integration of TMDs into devices while preserving their desired properties. Once TMDs with the desired properties have been produced, it is important to preserve their properties when exposing them to environmental conditions and to the fabrication processes required for integration into devices. The former is especially important for TMDs with poor air stability [20], while contamination and defects creation throughout the device fabrication process are concerns for all 2D materials. For example, lithographic processing can leave residues on the sensitive TMD surface [21], and passivation of TMDs with dielectrics or other materials, which is required for numerous applications, can lead to oxidation and/or creation of defects [22,23]. Passivation of TMDs is a challenging process due to the inertness of the basal plane of TMDs, hindering chemical and physical vapor deposition of uniform capping layers [24,25]. It follows that demonstration of a controllable, site-specific synthesis of TMDs, with pre-programmed structure and materials properties, in an isolated environment would be of major scientific and practical interest.
This work demonstrates a novel synthetic method that enables the fabrication of 2D TMDs with lithographically-predefined structures, and thickness control with few-layer precision. The structures are uniform across greater than centimeter scale and can grow to lateral depths exceeding 100 µm using techniques amenable to wafer-scale processing. The 2D material is grown as a buried layer-encapsulated by SiO 2 without further processing steps-providing a platform for contamination-free integration into devices. The synthesis is achieved by first encapsulating an ALD grown metal oxide (MO x ) film between SiO 2 layers, then patterning and etching to expose the edges of the metal oxide. The buried MO x film is subsequently chemically converted into a metal chalcogenide, laterally from the exposed edges. The approach promotes lateral (inplane) growth of TMDs, in a process we term 'lateral conversion'. The method is distinct from widely used MOCVD, CVD, and ALD methods in which films are directly deposited-or grown on a substrate, and is also distinct from conversion of exposed ALD oxide films to chalcogenides [13,14,[26][27][28]. The presented lateral conversion methodology and the properties of the resulting materials were studied in detail for the conversion of WO x to WS 2 . The films were characterized by a variety of analytical techniques, including x-ray photoelectron spectroscopy (XPS), hyperspectral Raman imaging, scanning electron microscopy (SEM), and transmission electron microscopy (TEM). To show the effectiveness and generalizability of this technique, the process was extended to demonstrate the growth of WSe 2 , MoS 2 , and MoSe 2 , and we believe it will be applicable to other relevant chalcogenides. To demonstrate one of the unique capabilities of this method, we fabricated stacked multilayer structures composed of repeating SiO 2 /WS 2 units. We believe this unique lateral conversion methodology will further enable a variety of other unique advanced architectures, such as complex (3D) nanophotonic structures [29,30].

Results and discussion
The fabrication of lithographically patterned TMD structures via lateral conversion is achieved by a several-step process, which is illustrated in the crosssectional schematic presented in figure 1(a). First, a multilayer film is deposited: a metal-oxide (MO x ) thin-film is deposited on a Si/SiO 2 thermal oxide substrate by plasma enhanced ALD. ALD allows angstrom-level precision of the layer thickness and can be varied to achieve the desired number of TMD layers, as will be shown later. To create a buried layer for lateral conversion and to ensure the preservation of a clean, high-quality interface between the SiO 2 and the metal-oxide, the MO x film is subsequently covered with a few nanometers of SiO 2 by ALD. Additional SiO 2 is deposited by plasma enhanced chemical vapor deposition (PECVD) to increase the barrier layer thickness while reducing total deposition time. The substrate is then patterned using standard photolithography techniques, and dry etched with CHF 3 + Ar, resulting in the structure depicted in figure 1(a)(left). The photoresist is removed and the samples are transferred to a tube furnace reactor. The process of chemical conversion of the MO x layer into MX 2 is carried out in two steps. First, the sample is annealed in an Ar/H 2 gas mixture, reducing the MO x layer to metal, as depicted in figure 1(a)(center). Next, it is chalcogenated by exposure H 2 X to form the TMD (MX 2 ). The resulting structure is depicted in figure 1(a)(right).
The process outlined above was used to convert WO 3 to WS 2 . True-color white light reflectance microscope (WLRM) images, showing substrates at key steps in the process, are presented in figure 1(b). The difference in color in the images is due to changes in thickness and index of refraction of the films as their composition changes. Figure 1(b)(left), shows an asmade patterned substrate with 4.8 nm of WO 3 deposited on 250 nm SiO 2 thermal-oxide on silicon, and coated with ~55 nm of SiO 2 . The WO 3 /SiO 2 layers appear light blue, with the etched regions (~100 nm deep) appearing white. After reduction in an Ar/H 2 gas mixture at 750 °C, the tungsten layer is composed primarily of metallic tungsten, and appears purple, as shown in figure 1(b)(center). The layer was then sulfidized by introducing H 2 S into the Ar/H 2 mixture at 750 °C, resulting in formation of WS 2 encapsulated by SiO 2 . To visually observe the difference between converted and unconverted regions, the sulfidation step was intentionally stopped before converting the entire film. In figure 1(b)(right), converted regions are observed only around etched holes, with WS 2 appearing dark blue. This basic demonstration of the process shows that the conversion proceeds laterally from the lithographically patterned edges and demonstrates the effectiveness of the SiO 2 capping layer in preventing penetration of sulfide into the reduced metallic layer from the top. Figure 1(c) shows three structures that were lithographically patterned and converted to WS 2 with increasing time or temperature (from top to bottom). These are arbitrarily patterned 'test structures', used to demonstrate the type of patterning that can be achieved with this method. It shows the conformal nature of the growth front, demonstrating that the extent of lateral conversion (ELC) can be controlled to create ribbon-like structures around the edges of a pattern when shorter conversion times are used, or to fully convert arbitrary shapes, filling areas when longer conversion times are used.
To test our ability to control the thickness of the TMD layers, samples with 2.4, 3.6, 4.8, 7.2 and 9.6 nm thick WO x films were fabricated, corresponding to desired WS 2 films with 3, 5, 8, 11 and 15 layers, respectively. The samples were converted using the process parameters described above. Figure 1(d)(left) shows representative high-resolution TEM cross-sectional images of the WS 2 layers, labeled with the corresponding deposited WO x film thickness. For simplicity, going forward we will refer to samples using their initial ALD oxide thicknesses. The cross-sections show highly-oriented WS 2 films with programmable thickness control, and a measured interlayer spacing of ~6.2 Å, which closely matches the lattice spacing in the c-direction of pristine WS 2 . The average number of WS 2 layers in the film were counted at 30 locations along the 10 µm cross-sections, and summarized in figure 1(d)(right). The error bars describe the span of the thickness values (maximum and minimum number of layers counted), which shows excellent agreement to the expected number of layers (demarcated by the orange line) that were calculated based on the nominal thickness of the WO x layer. Additional TEM results and analysis are presented later.
XPS was used to follow the chemical changes at each stage of the conversion process (figure 1(e). Since the target layer is buried beneath SiO 2 , depth-resolved spectra were acquired by controllably sputtering the sample. The spectral shift of the W4f 7/2 peak, often referred to as chemical shift, is a clear signature of W oxidation state. The peak appears at binding energies of 31.3 eV for W metal, at 32.4 eV for WS 2 , and at 33.6 eV and 36.1 eV for WO 2 and WO 3 , respectively. Spectra were analyzed to reveal the chemical state by fitting with a minimal number (2-3) of components, each composed of a W4f doublet and a broad W5s feature at higher binding energy. The spectrum of the ALD tungsten oxide precursor is indicative of a complex oxide, requiring at least three oxide-like comp onents including WO 2 , WO 3 , and one with an intermediate oxidation state. Complex oxide-character is often seen in ALD oxide films and has been reported for ALD tungsten oxide [31]. The W4f feature changes drastically as a result of the reduction step and becomes almost exclusively metallic in character, with less than 8% residual oxide components. Following sulfidation, the W4f signature is almost exclusively of WS 2 in character, with less than 2% residual WO x -like species.
While the changes in color observed by WLRM images appear to provide a good estimate of the ELC, additional spectroscopic methods are needed to study the properties of the converted layer in more detail. Raman spectroscopy of WS 2 can allow us to extract the film's quality, orientation, and thickness [32]. WLRM images and Raman spectra, presented in figure 2, were collected for five samples with increasing WO x thickness, which were converted to WS 2 using identical conditions. WLRM images are shown in Figure 2(a)(left column). The corresponding confocal micro-Raman images shown in figure 2(a)(center column), map the integrated intensity of the WS 2 spectral component, which was extracted from the hyperspectral image data collected for each sample. The average Raman spectra are shown figure 1(a)(right column). In the three thickest samples, 3 regions are demarcated on the center Raman map by a white dashed line. The intensity of the Raman signal is highest in region 1 near the exposed edge. This is followed by a significant decrease in uniformity and intensity of the Raman signal from region 1 to region 2, indicating that the quality and uniformity of the WS 2 film is decreased in this region. The intensity of the Raman signal then drops off abruptly at the interface between region 2 and the unconverted metal layer in region 3. These observations are consistent with a diffusion driven mechanism for the conversion process: complete conversion to sulfide near the exposed film edge, a two-phase region-where residual unreacted tungsten and WS 2 are present together, and an abrupt conversion endpoint that is defined by the diffusivity of the sulfur in the buried layer.
Analysis of Raman spectral features can be used to extract information about the WS 2 thickness. It has been shown that as the thickness of exfoliated WS 2 sample decreases, the intensity ratio of the 2LA to A 1g mode increases, while the separation between the two modes decreases [33]. Due to significant overlap between spectral components of WS 2 Raman signature, the spectrum must be fit with a set of Lorentzian functions for the intensity of the 2LA mode to be accurately extracted ( figure 2(b)). Each spectrum in the Raman hyperspectral image has been individually fitted. The relevant values have been extracted and plotted against the nominal WO x thickness (fig-ure 2(c)). The intensity ratio of the 2LA/A 1g modes increases with decreasing sample thickness, mostly as a result of the decline of the A 1g mode intensity; also, the 2LA mode position is red shifting, while the A 1g mode is blue shifting at the same time. These trends are almost identical to the ones referenced above [33], and can be interpreted as microscopic confirmation of the thickness and structure of the films. The Raman results confirm that the thickness of the WS 2 layer is defined by that of the original WO x layer, and also show that we can interpret the color contrast in WLRM images to measure the ELC. These are valuable feedback parameters for the study of the kinetics and thermodynamic mechanisms of this process.
While Raman spectroscopy is a good indicator of general material quality, atomically-resolved imaging by TEM can help us better understand the morphology, microstructure, and conversion mechanisms. Figure 3(a) shows a WLRM image (left), confocal Raman map (center), and an SEM image (right). The WLRM image and Raman map show variations in color and intensity, consistent with those discussed above, and were used to select the specific sites from which TEM cross sections were extracted. The exposed edge is on the left-hand side, with conversion proceeding from left to right. The boxes drawn on the SEM image indicate the regions from which two, ~14 µm long, 10-50 nm thick TEM cross-sections were extracted by focused ion beam (FIB) milling [34] (labeled 'FIB 1' and 'FIB 2'). Panoramic TEM images of the cross-sections are shown in figures 3(b) and (c) (top). A series of six high-resolution TEM (HRTEM) images, allowing detailed structural characterization, were taken at regular intervals-roughly corresponding to their positions below the panoramic image. In figure 3(b), the HRTEM images reveal a highly-oriented continuous WS 2 film near the exposed edge on the left-hand side of region 1. This region is correlated with the highest intensity area seen in Raman map in figure 3(a). Proceeding to the right, into region 2, the WS 2 layers show more disorder, and occasionally small grains of unconverted W are observed. Modulations in layer numbers and evidence of gaps between grains are also seen. These modulations explain the intensity variations observed in the Raman images. In figure 3(c), as we approach the end of the converted region, midway through region 3, the film is almost exclusively composed of pure W metal, with small amounts of WS 2 visible at the grain boundaries between the W domains. These observations are consistent with a diffusion mechanism paired with chemical reactions that immobilize sulfur as it is incorporated into WS 2 . The presence of small amounts of WS 2 past zone 3, at the grain boundaries of the W film, indicates the importance of grain boundary diffusion in the conversion process. Figure 3(d) shows a cross-sectional HRTEM image of the thinnest sample studied, synthesized by lateral conversion of a 2.4 nm thick WO 3 film. The resulting WS 2 is highly crystalline and consists of 4 ± 1 layers showing a high degree of in-plane alignment. No mixed WS 2 /W or unconverted regions were observed along the entire cross-section, and the WS 2 film was observed to terminate abruptly at the transition to W. Such precise control over the number of layers and the ability to grow WS 2 on micron scale, together with lithographic compatibility and pre-existing protection layer makes this material a promising candidate for few-layer electronic and optoelectronic devices. Further scale down of the precursor oxide and conversion process optimization may lead to (near-) monolayer material.
To gain mechanistic understanding of the factors governing the kinetic and thermodynamic aspects of the lateral conversion process, we measured the ELC as a function of sample thickness, temperature, and reaction time. Other parameters, such as gas flow rates, pressure, and temperature ramp rates were kept constant. Two methods were used to evaluate the ELC: (1) Raman line profiles were acquired perpendicular to the exposed edge using a Horiba confocal microscope, and (2) color changes were observed by WLRM and measured. From the Raman line profile, the integrated intensity of the A 1g and the 2LA + E 1 2g peaks were extracted. Figure 4(a) shows a plot of intensity as a function of distance (top) and the corresponding WLRM image (bottom) for a laterally converted sample. Raman line scans and WLRM color changes were observed to be correlated in a large number of samples, thus we determined that WLRM can be used as a fast and accurate proxy measurement to determine the ELC. The ELC shown for the following studies were measured by programmatically extracting the intensity profiles from WLRM images based on color differences between the converted and unconverted regions.
To study the kinetics of the conversion process, we reduced the samples in Ar/H 2 atmosphere at 750 °C for 120-180 min to ensure full reduction of the WO x layer. Subsequently, samples of varying thicknesses were sulfidized for times varying from 15 to 120 min. The measured ELC values are shown in figure 4(b). For the thinnest samples the ELC is smaller than 1.4 µm and too close to the optical microscopy resolution to establish accurate time dependence. For all other samples, the ELC increases with increasing sulfidization time. The ELC for the 3.6 nm samples is somewhat flat, but shows an increase for long conversion times. The ELC for the 4.8 nm samples increase linearly with time, while the two thickest samples show a steep increase after 60 min. Figure 4(c) shows the temperature-dependent conversion experiments. Samples were annealed in H 2 at 750 °C, then sulfidized at temperatures ranging between 650 and 800 °C for 30 min. Similar to the time-dependent experiments, the ELC for the thinnest samples (<2 µm) are on the order of the optical microscopy resolution limit, making the temperature dependence of the 2.4 nm thick sample hard to measure. For all other thicker samples, the trend generally shows that increasing reaction temperature results in a larger ELC. However, they do not follow the expected Arrhenius behavior, suggesting that the kinetics are more complex than a single path random walk diffusion. The trend breaks down beyond 800 °C, with samples showing erratic ELC. We suspect this is due to the formation of silicon disulfide from around the edges of the substrate where the silica passivation layer is thinnest. A related observation is that samples containing small cracks or scratches (exposing Si to H 2 S) are prone to formation of silicon disulfide. The observation of SiS 2 formation seemed to correlate with inconsistent WS 2 growth. We performed experiments on quartz substrates, which showed similar time and temperature dependence, but were able to withstand higher temperatures, supporting premise that Si plays a role.
Though the time/temperature dependent conversion rates do not seem to follow a simple diffusion-rate mechanism, we speculate that there are three mechanisms for the diffusion of H 2 S through the film that occur simultaneously, each with different rates. (1) Diffusion through the relatively large space between van der Waals bonded layers, (2) grain boundary diffusion through W and WS 2 , and (3) diffusion through voids inside W or WS 2 , which were formed as a result of volume shrinkage during the reduction of WO 3 to W metal. Each of these likely contribute to the complex non-Arrhenius diffusion behavior observed.
The reduction step plays an important role in the lateral conversion process. Interestingly, when H 2 S or H 2 S/H 2 gas mixture was used without a separate reduction step, very little conversion was observed. The small amount of conversion that was observed, generally did not penetrate very deeply into the film or was very non-uniform. To assess the impact of the reduction step, samples were reduced in H 2 -varying the reduction time from 15-60 min while holding the subsequent sulfidation time constant at 120 min. The results presented in figure 4(d) show that the ELC increases with increasing reduction time. This suggests that: (a) the reduction of WO x to W metal also proceeds laterally, (b) that the reduction step is crucial for uniform lateral conversion to WS 2 , and (c) that the ELC is limited by the extent of reduction. It is worth noting that direct deposition and lateral conversion of metallic W, without the requirement for a reduction step, may also be interesting to study, however, the method we demonstrate has several significant benefits. ALD provides uniform, angstrom-level thickness-control over large areas and allows the ability to coat high aspect ratio structures-potentially enabling more diverse 3D architectures and applications of TMDs created using this method. Additionally, metal films below ~5 nm are difficult to achieve without significant island formation, which would likely result in discontinuous and inconsistent TMD films.
In addition to WS 2 , we were able to extend our lateral conversion technique to fabricate MoS 2 , WSe 2 , MoSe 2 , and a stacked multilayer structure. Figure 5 shows optical images and corresponding Raman spectra for these TMDs. Raman modes were assigned based on those observed in previous work [35][36][37][38], and confirm the composition of the films. While the conversion extent of the WSe 2 and MoSe 2 materials do not demonstrate the near-perfect contour that we observe for WS 2 and MoS 2 , we believe it can be improved with some process optimization, and it suggests that the lateral conversion method should be generalizable to the growth of other TMDs. In figure 5(d), a stacked multilayer structure is shown, consisting of four WO x layers with different thickness-separated by SiO 2 -laterally converted to WS 2 . A WLRM image of the multilayer structure with three distinct colors around etched holes is shown (left), and a TEM cross-section of the same structure is shown (right). Due to the thickness differences between the WO x layers, the ELC are different for each layer, resulting in the multi-colored concentric rings that were observed by WLRM. TEM measurements show well-aligned, alternating layers of WS 2 and amorphous SiO 2 . The topmost layer, however, consists of metallic W and does not show conversion. This is also evidenced by the display of only three colors in the WLRM image as opposed to the expected four colors. We suspect that the layer failed to convert due to redeposition of SiO 2 during RIE, obstructing the opening through which chalcogen diffusion should occur. Optimization of this process is required, nonetheless, we think this is a significant result, showing a unique capability and flexibility of this approach.

Conclusion
In conclusion, we demonstrate the growth of TMD materials by lateral conversion of metal-oxide layers encapsulated between SiO 2 layers. The approach enables the formation of precisely patterned TMD structures using standard lithographic techniques and tools widely used in industry and academia, such as ALD and dry etching. The converted materials were analyzed using WLRM, Raman spectroscopy, XPS, and TEM. Raman measurements were able to clearly identify the TMDs, and provided evidence for the reduction in number of WS 2 layers with reduced initial WO x thickness. Cross-sectional TEM images of the laterally converted films revealed the morphology, microstructure, conversion mechanisms, and fewlayer thickness control of the WS 2 layer. Kinetic and thermodynamic aspects of the lateral conversion mechanism were studied, exploring changes in conversion rate as a function of film thickness.
Looking forward, we believe that layer control and crystallinity of the materials can be further optimized. Synthesis of TMDs encapsulated between SiO 2 or other protective layers should act to preserve the asgrown material quality without degrading its properties or contaminating it during subsequent processing steps, thus providing a way to grow and improve the long-term stability of oxidation-prone materials. This method can likely be extended to be used with other technologically useful encapsulating dielectric materials or epitaxial substrates, such as h-BN, graphene, other high-k dielectrics, sapphire, etc. Because ALD allows conformal coating of complicated and highaspect ratio structures, this method could also enable integration of TMDs into 3D architectures. Finally, different metals and chalcogens could be used to synthesize a large variety of different TMD materials and TMD vertical or lateral heterostructures.

Fabrication and conversion
Metal oxide (MO x ) thin-film, between 2-10 nm was deposited on a Si wafer with 250 nm thermal oxide in an Oxford FlexAl ALD at 300 °C. An oxygen plasma was used as the oxygen source. The tungsten precursor was bis(tert-butylimino) bis(dimethylamino)tungsten(VI). The molybdenum precursor was molybdenum hexacarbonyl. Our process demonstrated linear cycle-to-thickness growth dependence with WO 3 and MoO 3 deposition rates of 0.5 Å/cycle and 0.7 Å/cycle respectively. The film thickness was confirmed using an in situ Woolam M-2000 ellipsometer. The MO x film was subsequently covered with 4.3 nm of SiO 2 by ALD without breaking the vacuum, using tris(dimethylamino)silane and oxygen plasma. Substrates were transferred to an Oxford 80 + PECVD to grow an additional 50 nm of SiO 2 , using 1% SiH 4 in Ar and N 2 O (1182 sccm and 710 sccm, 350 °C, 10 W). Various test-patterns were created on the substrate using ma-N 1420 negative tone photoresist, and standard photolithography techniques. The samples were dry etched with Ar + CHF 3 (25 sccm + 35 sccm, 100 W, 30 mTorr) in an Oxford 80 + Reactive Ion Etching system to a depth of about 100 nm. After etching, the photoresist mask was removed in hot acetone and samples were annealed for 10 min in O 2 plasma (50 sccm, 150 W, 100 mTorr) to remove photoresist residues and the polymer layer known to form on trench sidewalls during dry etching. Larger wafers were then cleaved into smaller samples for conversion. To prevent the chalcogen precursor from attacking the exposed silicon on the edges of the substrates during the conversion step, the edges of the substrates were placed face-down and coated with 50 nm of SiO 2 in the PECVD system. Next, the substrates were transferred to a MTI Crystal 1200x-RTP-4 tube furnace reactor for conversion. To reduce the MO x film, the samples were heated to between 650 °C-750 °C in 40 s, in 1000 sccm Ar and 50 sccm H 2 at atmospheric pressure, and held for up to 2 h. The metal was then chalcogenated by introducing 25 sccm of H 2 S into the Ar/H 2 mixture at between 650 °C-850 °C at a pressure of 100 torr. To ensure accurate conversion times, rapid gas exchange was achieved by quickly pumping to base pressure and refilling with the desired Ar/H 2 /H 2 S mixture. At the end of the reaction time, the system was again pumped to base pressure and refilled with Ar and allowed to cool. The selenide samples were prepared using a similar procedure, in a Lindberg Blue HTF55322C tube furnace, a 1″ quartz tube, 500 sccm Ar, 25 sccm H 2 , and 12.5 sccm H 2 Se at 760 torr. Annealed temperature of 750 °C, and conversion temperature of 650 °C.
Optical characterization WLRM images were obtained using a Leica DM2500-M microscope.
Raman spectroscopy maps were performed on a custom Raman microscope built on a Nikon Eclipse Ti inverted microscope. The sample was illuminated with a 532 nm laser which was focused to a diffraction limited spot with a Nikon 100 × 0.95 NA objective. The sample was rastered with a MadCityLabs Piezo scan stage. Spectra were recorded with an Acton 2300i spectrometer with an 1800 grooves mm −1 grating and a Princeton Instruments PIXIS CCD. The sample is scanned over an area up to 49 µm 2 . Data acquisition is performed using Scope Foundry Software [39] and processed using Python's matplotlib [40], scipy [41] and hyperspy [42] libraries. Raman maps represent integrated intensities of the WS 2 signal component extracted by means of principal component analysis using non-negative matrix factorization algorithm [43].
Raman line scans and point spectra were also obtained on Horiba Jobin Yvon LabRAM ARAMIS confocal microscope, with 532 nm Nd:YAG laser, focused by a 100× objective of 0.9 numerical aperture and the spectra were processed using the LabSpec 6 software.
X-ray photoelectron spectroscopy XPS measurements were performed using Thermo Scientific K-Alpha x-ray photoelectron spectrometer system, using a monochromatic Al Kα x-ray (hv = 1486.7 eV), equipped with an ion gun for depth profiling. The latter was operated in a monatomic mode at 1 keV, with typical sputtering rates of SiO 2 of 5 nm min −1 .
Electron microscopy SEM measurements were performed using Zeiss Ultra 55 SEM at an accelerating voltage of 5 kV. Crosssectional samples were milled using a FEI Helios G4 UX dual beam focused ion beam (FIB). The sample surface was first protected using electron beam induced deposition of carbon and platinum followed by additional platinum layer deposition using and ion beam. Next a sample was millled-out using a standard procedure common for thin film sample preparation. The sample was mounted on a Copper grid and thinned to 10-50 nm thickness. TEM imaging was performed using a JEOL 2100F TEM, at an accelerating voltage of 200 kV. The image contrast and brightness were adjusted in Gatan Digital Micrograph 3.0.