Two dimensional semiconducting materials based on transition metal dichalcogenides (TMDС) are promising candidates to replace the conventional semiconductors in nanoelectronics and sensorics^[1−4]. The ultrathin layers of 2D materials with tunable bandgap such as MoS₂, MoSe₂, WS₂, and WSe₂ with high carrier mobility are highly demanded for low power device applications. The fabrication of single-crystalline high-quality layers of these materials has recently been the focus of many research groups^[5−9]. So far, the most of studies have been conducted on the flakes of the 2D materials transferred to various substrates by exfoliation technique. Attempts to grow these materials epitaxially have also been realized in a number of works by CVD^[5−7], MOCVD^[8], and MBE^[9]. The growth of large-scale films of 2D materials have been discussed in Ref. [10, 11]. The exfoliated and epitaxially grown WS₂ films have been studied with a large variety of experimental techniques including Raman spectroscopy^{[12, 13]}, high-resolution transmission electron microscopy^{[14, 15]}, atomic force microscopy^[16], photoluminescence^[15], etc. The studies utilizing X-rays were mainly related to X-ray photoemission spectroscopy (XPS)^{[17, 18]} which is a surface sensitive technique and cannot be effectively applied to thick films or thin films capped with a protective layer. Only a few studies were done with X-ray absorption spectroscopy^[19] and (for thick WS₂ films) with conventional X-ray reflectometry^[20]. To our knowledge, no results on resonant X-ray reflectometry of few monolayer thick TMDC films have been reported so far. Due to the monolayer scale thickness of the 2D films, implementation of the standard methods to probe depth dependence of the structural and chemical properties is not straightforward. E.g., it is hard to apply etching followed by X-ray photoemission spectroscopy or secondary ion mass spectrometry to analyze films that are so thin. Among other characterization techniques, high-resolution transmission electron microscopy enables a detailed cross-sectional image with close to atomic resolution, however this approach requires considerable efforts to prepare an appropriate sample cut, whereas the preparation procedure itself can influence the sample structure. In the present work we demonstrate application of non-destructive methods of conventional and resonant X-ray reflectometry to ultrathin WS₂ films grown by pulsed laser deposition onto sapphire substrates. An important advantage of a synchrotron experiment is that photon energy can be varied in a wide range and in particular can be tuned to the absorption edges which gives the X-ray scattering methods the power of being selective to particular chemical elements as well as to their electronic^{[21, 22]} and magnetic states^[23−25].

As it was recently shown, the method of resonant X-ray reflectometry can be made more effective by analyzing two-dimensional maps of reflectance measured as a function of photon energy and incident angle^[26−30]. By fitting the experimental maps with a dedicated modelling software it is possible to restore depth profiles of density, chemical composition, oxidation state, crystallographic environment and magnetization state^{[26, 27]}. The possibility to distinguish optically similar sublayers was discussed in Ref. [31] in relation to soft X-rays and in Ref. [32] in relation to hard X-rays. As it was further demonstrated in Refs. [33−36] in particular cases the fitting algorithm can be effectively used to restore also the spectral shapes of the atomic scattering factors within the studied materials. In this case there is no need to independently obtain reference spectra of the scattering length density.

In the present paper we apply hard X-ray resonant reflectometry 2D mapping technique to the study of ultrathin WS₂ layers epitaxially grown on Al₂O₃ substrate. The challenge addressed in this work is related to the ultra-low thickness of the films potentially complicating the non-destructive extraction of depth profiles of chemical composition and oxidation state. The element selectivity was achieved in this experiment by tuning the photon energy to the L absorption edge of W at 10 210 eV. Fitting of the resonant reflectance maps was carried out with freely varied scattering length density spectral shapes which made it possible to discover the presence of a subnanometer layer of WO_x residing on top of WS₂, the conclusion supported by the differences in optical densities and oxidation state dependent energy positions of the W L₃ absorption edge in tungsten disulfide and tungsten oxide.

The samples were grown by means of pulsed laser deposition (PLD) on Al₂O₃(0001) substrates. The samples were designed to have 3 and 5 WS₂ monolayers (1 ML = 6.3 Å). The film thicknesses were chosen sufficiently low taking into account the potential usage of these films in top gated field effect transistors. Some samples were capped with a few nm protective layer of Ag to prevent contamination during the contact with air. We used an advanced PLD setup (SURFACE, Germany) equipped with a KrF excimer laser to ablate stoichiometric WS₂ target. The growth rates of WS₂ and Ag were of the order of 0.1 Å per pulse as measured with a quartz microbalance. The deposition was performed in argon atmosphere at a pressure of 0.05 mbar. The growth was performed at 500 °C and was raised up to 700 °C for 30 min after finalization of the growth process in order to improve crystallinity. The surface morphology was measured using atomic force microscopy (Ntegra NT-MDT), the estimated roughness was about 0.2 nm (Fig. 1).

The resonant X-ray reflectometry studies were carried out at BL3A beamline of the Photon Factory synchrotron (Tsukuba, Japan). The reflectivity maps were measured as a series of θ–2θ scans (incident angles in the range of 0°–3.5°) with the photon energy scanned with a step of 3 eV across the L₃ absorption edge of tungsten at 10 210 eV. The intensity of the reflected light was measured with a silicon drift detector placed behind a slit and tuned to cut off higher monochromator orders, inelastic scattering and fluorescence. In addition to reflectance, we have also studied X-ray absorption spectra of the WS₂ films (Fig. 1(b)). This was performed by measuring X-ray fluorescence with the same SDD detector tuned to pass only the fluorescence line of W. Conventional non-resonant reflectometry was carried out using a 4-circle Bruker diffractometer at a fixed photon energy corresponding to Cu Kα emission.

Visualization, fitting and analysis of resonant reflectance maps was performed with the aid of a dedicated software^{[26, 27]} that calculates reflectance of a multilayer using recursive Parratt formalism^[37]. The knowledge of scattering length densities (SLD) of the layer’s materials is required to perform these calculations. To fit the non-resonant reflectivity, the values of atomic scattering factors are usually taken from the Henke tables^[38] while the optimal layer thicknesses, densities and roughnesses are picked up by the fitting software (e.g. GenX^[39], ReMagX^[40], and OPAL^[22], etc.) in order to minimize the misfit. The situation is drastically different across the absorption edges where both real and imaginary parts of the optical density exhibit a sharp energy dependence. This makes it more difficult to fit experimental data, but at the same time makes it possible to distinguish individual chemical elements. The common approach accepted in multiple works^{[23, 26, 27, 36]} including those of our group is to derive Im(SLD) part from X-ray absorption spectroscopy measurements of a reference material and then to obtain Re(SLD) part through Kramers–Kronig (KK) relations. While this approach looks quite straightforward, it must be taken into account that the absorption measurements are prone to instrumental errors^{[41, 42]}, and moreover the absorption spectra of the reference materials may be different from those of the materials inside the studied heterostructure. In the present work we did not use reference data to model reflectivity. The SLD spectral shapes were constructed by the fitting routine using a semi blind approach assuming that the imaginary part of SLD across the tungsten absorption edge exhibits a Lorentzian shape^[43] summed with a step function:

1$\begin{array}{ll}{f}_2\left(E\right)=\hfill & f_2^0\cdot 4E{E}_0{w}^2/\left(\left(E_0^2-E^2\right)+{\left(2Ew\right)}^2\right)+\hfill \\ \hfill & S\cdot \left(\frac{1}{2}+\frac{1}{\pi }\text{arctg}\left(E-E_0\right)\right)+f_2^{\text{preedge}}\mathrm{.}\hfill \end{array}$

Here $f_2^0$ and w define the peak height and width respectively, E₀ defines the energy position of the peak maximum, S defines the height of a step present in the L absorption edge spectrum and $f_2^{\text{preedge}}$ defines the preedge adsorption level. The parameters describing the Lorentzian shape of the absorption peak were allowed to vary freely during the fitting routine. The real part of the SLD was then calculated using the KK transformation. The benefit of the described approach is that the energy position of the peak is defined automatically taking into account the monochromator misalignment and the energy shifts related to the oxidation state and crystallographic environment of the atom. The fitting is performed using the high-performance OpenCL code implementing a differential evolution algorithm^[44].

Measurements of non-resonant X-ray reflectivity were carried out on a laboratory diffractometer at the photon energy of E = 8051 eV (λ = 1.54 Å). The samples with WS₂ thickness of 3 and 5 ML have been studied. The experimentally measured reflectivity profiles are shown in Figs. 2(a) and 2(c) together with the profiles obtained by fitting. The low amplitude high frequency jitter present in the experimental data is related to detector noise rather than to structural features of the studied samples and is not taken into account during modelling. With the non-resonant SLD values taken from the Henke tables, the only parameters optimized by the fitting routine were thickness, densities and roughnesses of the layers. Interestingly, it was not possible to reproduce the experimentally measured reflectivity curves following this approach and assuming a single layer model. The simplest model providing a reasonably good agreement with the experimental data was a trilayer consisting of a transition layer residing at the Al₂O₃/WS₂ interface, a WS₂ main layer with close to nominal optical density and a surface layer with an increased optical density. The thicknesses of the transition layer (5 Å) and the surface layer (8 Å) were similar for both studied samples while the thickness of the main WS₂ layer was expectedly larger for the sample designed to have 5 monolayers of WS₂. Interestingly the model suggests that the scattering length density of the top layer is higher than that of WS₂ which indicates higher concentration of tungsten at the surface. Though, in principle some tungsten segregation may occur during the growth from a sulfur deficient plasma plume, the formation of purely metallic W at the surface of WS₂ is excluded as in this case the optical density of the capping layer would be considerably higher (up to 4 r_e Å^–3). A more reasonable agreement with the tabular optical densities is obtained for a model in which the WS₂ layer is capped with 8 Å of tungsten oxide WO_x (Fig. 2).

The formation of the WO_x layer on top of WS₂ is related to the photo-oxidation reaction which is known to convert WS₂ in the monolayer form to tungsten oxide. The ambient oxidation of WS₂ is known to be intensified in the presence of water which acts as a catalyst for oxidation. Initiated by photon-mediated electronic band transitions, this reaction is dependent on the photon energy and photon flux^[45]. The monolayer thick WS₂ films were reported to degrade over time in ambient air under illumination and the source–drain currents flowing in the WS₂ based 2D channel transistors were shown to drop by orders of magnitude^[46]. Most of the studies related to photo-assisted WS₂ degradation were conducted with visible and UV light, showing that the WO_x oxide layer is formed after days or weeks of irradiation. While oxidation during sample transportation and storage cannot be excluded in our case, it is also possible that the oxidation was activated by X-ray photons during reflectivity measurements carried out in non-controlled ambient conditions. As the films studied in the present work were thicker than 1 monolayer, it is reasonable to assume that the oxidation did not affect the full depth of the film. As mentioned in Ref. [20], when a bilayer or monolayer WS₂ sheets are exposed to oxygen plasma, the entire volume gets oxidized. However, for thicker layers, only outer sheets are oxidized, leaving an intact WS₂ under thin film of WO_x.

To get a deeper insight into the oxidation process occurring in ultrathin PLD grown WS₂ films, a 2D resonant reflectance mapping experiment was carried out using variable energy synchrotron radiation. The 5 ML WS₂ layer studied in this experiment was capped with 30 Å Ag layer in an attempt (though not successful as will be shown below) to slow down ambient oxidation of the WS₂ layer. The experimentally measured resonant reflectance map is presented in Fig. 3(a). In contrast to our previous works where resonant reflectivity of much thicker 30–60 nm films was studied, the map presented here exhibits very smooth intensity variation. Apart from thickness oscillations resulting in horizontal stripes, a wide vertical ridge is observed on the map at approximately 10 210 eV where the L₃ absorption edge of tungsten is located. The ridge shape resembles that of the real part of the atomic scattering factor as can be seen in Fig. 1(b) showing the reflectance spectrum measured at 2° of incidence.

Remarkably, despite the absence of sharp features in reflectance, the use of the fitting algorithm enabled distinguishing multiple sublayers present within the studied film. Similar to the non-resonant case described above, the simplest "as-designed" model consisting of the Ag-capped WS₂ film on Al₂O₃ substrate did not result in close resemblance to the experiment. A low-density transition layer had to be placed between the WS₂ and the substrate in order to reproduce the sharp reflectance minimum observed at θ = 0.5°–0.6°. In addition to this, a tungsten oxide layer had to be added between the WS₂ layer and the Ag cap. Finally, a low-density capping layer was put on top of Ag to account for possible presence of a thin film of absorbed water. The modeled reflectance map is presented in Fig. 3(b), a close resemblance to the experimental map (Fig. 3(a)) is confirmed by constant energy reflectance profiles shown in Fig. 3(c) for discrete photon energies. The structure optimized by the fitting routine and giving the minimum achievable misfit is presented in Fig. 4(a) in the form of two maps showing real and imaginary parts of the scattering length density as a function of photon energy and depth. The SLD intensity shown in the form of such a map is an effective way to visualize the in-depth variation of the spectral features of absorption and refraction within the studied multilayer and is very effective in case when reflectance maps are measured at multiple absorption edges^[32] or when the shape or position of the absorption edge is affected by the change in the oxidation state or crystallographic environment within the multilayer^[27]. The more conventional constant-energy and constant-material cuts through the SLD maps are shown in Figs. 4(b) and 4(c). The most surprising result is the presence of a 32 Å low density transition layer at the interface between the main WS₂ layer and the substrate. While a similar layer was also observed in the laboratory X-ray reflectivity experiment, the synchrotron experiment indicates the presence of a less dense interface layer. This can be the result of partial film exfoliation taking place because of the heating caused by high intensity synchrotron beam during the reflectivity measurements. This is not surprising as the WS₂ film is not strongly attached to the substrate by weak Van-der-Waals force. A thin 8 Å layer of what looks to be WO_x residing on top of the WS₂ resembles the similar layer observed in the non-capped sample measured by non-resonant reflectometry and discussed above. The 15 Å capping layer of Ag was found to have a density substantially lower than expected, likely due to the fact that silver was deposited at 100 °C rather than at room temperature and formed nanoislands with much reduced effective density and thickness. The non-uniformity of the Ag capping layer also explains the fast oxidation of the WS₂ layer. Interestingly, the fitting routine was able to provide information about the oxidation state of W in the studied multilayer. The L₃ absorption peak was placed at 10 206 eV for the WS₂ layer and at 10 210 eV for the tungsten oxide layer. The shift in the absorption peak positions is readily visible in the density maps in Fig. 4(a) and the SLD spectra in Fig. 4(c). This is in agreement with the fact that tungsten exhibits 4+ oxidation state in WS₂ and 6+ oxidation state in WO₃ tungsten oxide. The similar shifts of the L₃ absorption edges were observed in the works^{[47, 48]} related to X-ray absorption spectroscopy studies of WS₂ and WO₃. While it is difficult to determine the exact oxidation state from the performed measurements, it can be concluded that a mixture of WO₂ and WO₃ is present in the surface layer above WS₂. It must be noted that conclusions on oxidation state of elements within thin uncapped layers could be also drawn from surface sensitive XPS measurements. However, when the material of interest is buried or capped with a protective layer, the benefits of resonant XRR over XPS become evident.

2D mapping in resonant X-ray reflectivity has been successfully applied to ultrathin TMDC layers. The blind fit approach enabled identification of tungsten oxide and tungsten disulfide in the sublayers helped obtaining a remarkably good agreement between the experimental and modeled maps. It has been shown that resonant X-ray reflectivity can be quite effectively applied to the very thin TMDC films consisting of a few monolayers distinguishing subnanometer sublayers of oxidized tungsten on top of the film. Importantly, taking into account the strong photo-oxidation of the TMDC films under illumination and in the presence of water vapor, it would be advisable to perform experiments in controlled environment such as dry nitrogen or vacuum and to store films in darkness prior to experiment. At the same time, photodegradation itself is an important, though undesirable feature, presenting interest to researchers developing devices based on ultrathin 2D layers of TMDC materials. As it had been demonstrated in the present work, the photodegradation can be nondestructively studied by resonant X-ray reflectivity with the photon energy tuned to the L absorption edge of the transition metal. It must be also noted that care must be taken during synchrotron measurement of reflectivity from ultrathin TMDC layers as the high intensity synchrotron beam can cause partial exfoliation of the studied film due to local heating. Importantly the presented method can also be applied to various multilayer structures where a chemical reaction occurs at the interface changing the oxidation state or the crystallographic environment of the atoms. Where possible the validity of the conclusions derived from resonant reflectometry should be confirmed by independent methods such as X-ray photoemission spectroscopy. However, it must be taken into account that unlike reflectometry which can be quite sensitive to the buried layers, XPS is mostly surface sensitive and cannot be applied to buried layers or layers with a protective capping. Finally, it must be noted that the presented map modelling approach can be very useful at the planning stage of a synchrotron experiment related to resonant reflectivity as it provides the knowledge of the angle/energy combinations at which reflectance becomes sensitive to particular physical properties of the studied samples.