Operando Study of Thermal Oxidation of Monolayer MoS2

Abstract Monolayer MoS2 is a promising semiconductor to overcome the physical dimension limits of microelectronic devices. Understanding the thermochemical stability of MoS2 is essential since these devices generate heat and are susceptible to oxidative environments. Herein, the promoting effect of molybdenum oxides (MoOx) particles on the thermal oxidation of MoS2 monolayers is shown by employing operando X‐ray absorption spectroscopy, ex situ scanning electron microscopy and X‐ray photoelectron spectroscopy. The study demonstrates that chemical vapor deposition‐grown MoS2 monolayers contain intrinsic MoOx and are quickly oxidized at 100 °C (3 vol% O2/He), in contrast to previously reported oxidation thresholds (e.g., 250 °C, t ≤ 1 h in the air). Otherwise, removing MoOx increases the thermal oxidation onset temperature of monolayer MoS2 to 300 °C. These results indicate that MoOx promote oxidation. An oxide‐free lattice is critical to the long‐term stability of monolayer MoS2 in state‐of‐the‐art 2D electronic, optical, and catalytic applications.

and/or increased FWHM. [9,10] The confidence level for the ratio quantification accuracy of XPS is set to 10 % to reduce the misinterpretation attributed to the signal to noise ratio in the Mo 3d region.
The vibrational modes of as-grown and etched MoS2 monolayers were studied by Raman spectroscopy. The Raman spectroscopy was performed by WiTec 500 AFM/micro-Raman Scanning microscope and HORIBA Scientific LabRAM HR Evolution Spectrometer with a 532 nm excitation laser. The transmission electron microscope (TEM) image was taken by FEI Titan The temperature of the sample was accurately controlled in the reactor using a custom-made heating unit and K-type thermocouple to monitor the local temperature of the monolayers. The operando XAS data were collected online under dynamic environmental conditions of gas and temperature, and the XANES spectra were further processed using the ATHENA software which is part of the Demeter package, a graphical interface of the IFEFFIT code. [11] Simulations of the XAS spectra were performed via the OCEAN code [12,13] that uses QUANTUM ESPRESSO [14] for ground-state density functional theory (DFT).
Statistical analysis. The operando and ex-situ XAS experiments were performed during three different beam time runs at SSRL-SLAC that consisted of three days of experiments, each. The beam times were separated by months. For each XAS experimental run, fresh CVD-grown MoS2 monolayers were synthesized. From each CVD batch, the wafer was divided into 10 mm x 5 mm pieces uniformly covered with MoS2 monolayers. Half of the batch was used as as-grown samples, and the other half was subject to the etching and transfer protocol to generate MoOx-free MoS2 monolayer samples. Ultimately, the oxidation onset temperature and consistent spectra of asgrown and etched monolayer MoS2 was measured for three different batches of samples, at different times of the year. The XAS analysis was done with ATHENA software (part of the DEMETER package v0.9.26). Linear combination fitting analysis (LCFA) reconstructs the samples' spectra using the model standards (i.e., etched monolayer MoS2, MoO2, and MoO3). The goodness of fit is reported using the R-factor and the reduced chi-squared. As defined for LCFA of the XANES data ATHENA's author describes the R-factor as the mean square sum of the misfit at each data point. For the reduced chi-squared, due to the solitary nature of time sequence XAS measurements/scans, ATHENA assumes 1 as the value of measurement uncertainty, resulting in very small values for chi-squared. Hence, although single values cannot assert the goodness of the fit, it is possible to relatively compare successive fits as performed in this work.

Supporting Text
The state-of-the-art Operando XAS reactor. The apparatus herein described in Figure S7 enables operando XAS studies of atomically-thin and ultra-dilute samples. The objective is to obtain XAS data under varying and dynamic environmental conditions (gas mixture balanced with ultra-high purity He, ambient pressure, and from room temperature to 400 °C). For the operando XAS studies, the samples are subject to a heat-treatment under different gases at atmospheric pressure using our in-house developed reactor. A custom-made heating unit locally increases the temperature of the sample. An integrated K-type thermocouple is used to monitor the local temperature of the sample in real-time, while a temperature controller and a power supply are used to control the heating rate and final temperature of the substrate. Mass flow controllers are utilized to control the gas mixture and the total flow into the reactor. The gas mixture flows in and out of the gas-tight reactor exposing the sample to a very stable and constant gas volume and partial pressure per time unit. X-rays are illuminated from the front X-ray transparent window of the reactor and the XAS signal is recorded using a metallic collector which is kept on a constant potential difference to collect the electron yield from the samples (Table S1). [15] Theoretical analysis of XANES spectra transition by the monolayer MoS2 oxidation. Periodic DFT calculations were carried out for a MoS2 monolayer without S-vacancies as shown in Figure   S13. We also considered models where S-vacancies at different levels of concentration were generated, and the Mo atoms thus exposed are coordinated by O atoms, i.e., substitution of surface sulfur atoms by oxygen. To prepare the models, a 2H-MoS2 monolayer with a supercell of lateral size (4 ✕ 4) was constructed, and some of the S atoms of the model were replaced by O. For all models thus prepared, a vacuum region of 18 Å was used to decouple the periodic images. We used the Vanderbilt method (GBRV pseudopotential library) [16] and RPBE functional. [17] The kinetic energy cutoff was chosen to be 500 eV and integration was carried out in the reciprocal space with (2 ✕ 2 ✕ 1) Monkhorst-Pack k-points. All calculations were spin-polarized and performed with the QUANTUM ESPRESSO package. [14] The obtained DFT-optimized structures (see Figure S13) were used to simulate their Mo L3-edge spectra using the OCEAN package. [12,13] This first-principles code generates X-ray absorption spectra based on both ground-state DFT and the numerical solution of the Bethe-Salpeter equation (BSE) within a basis of electron and hole states (and associated core-hole dielectric screening) provided by the DFT Kohn-Sham orbitals. [18,19] The DFT electronic structure was calculated within the generalized gradient approximation using the QUANTUM ESPRESSO code. [14] The efficient numerical sampling of the Brillouin zone was enabled through the use of the Shirley interpolation scheme. [20] The DFT plane-wave basis cut-off energy was set to 100 Ry using the PBE functional within the generalized gradient approximation (GGA). [21] The k-points used in the OCEAN calculations were 3 × 3 × 1 for all supercell models.
The real-space mesh for the BSE calculation was 12 × 12 × 24. The radius of the sphere in which the local basis is calculated was set to 3.0 Bohr to construct the PAW-style optimal projector functions (OPF). The screening of the core-hole interaction was done in real space using the random phase approximation up to a radius around the core of 6.0 Bohr. [22] The calculated Mo L3edge XANES spectra were numerically broadened via convolution with a Lorentzian with a halfwidth at half maximum (HWHM) of 0.5 eV.
Reasons for selecting the CVD method and alkaline-bath transfer treatment. We synthesized monolayer MoS2 by CVD method and removed MoOx using alkaline-bath transfer treatment due to their advantages. First, the mechanical exfoliation of monolayer MoS2 contains MoOx impurities [7,23] that are likely intrinsic impurities of the bulk MoS2 crystals [24] or introduced during processing, such as thermal annealing or O2 plasma to remove the tape residue during the mechanical exfoliation process. [23] Second, exfoliated MoS2 samples are typically composed of both monolayers and multi-layers, which are good for applications where you can pinpoint the monolayers. On the contrary, our chemical vapor deposition (CVD) method grows large flakes and areas of MoS2 monolayers, [1,2,25] which are suitable for XAS measurements as the X-ray beam illuminates broad parts of the sample. Third, the transfer and etching method is a common protocol to transfer CVD grown MoS2 monolayers to other substrates. [26,27] We did not observe significant structural changes of MoS2 from the transfer process ( Figure S4 Figure 4c shows a well-defined single S 2p3/2 and S 2p1/2 doublet ascribed to sulfur in the MoS2 environment. This is not the case when MoOxSy is present because the oxysulfides produce shoulders in the spectrum, change the peak ratio, and increase the 9 FWHM as reported in the literature. [28][29][30][31][32][33][34] Third, the chemical composition of MoOxSy was not captured by Raman spectroscopy (Figures S1 and S4), which would show a diverse mixture of peaks similar to MoS2, MoO2, Mo4O11, and MoO3 (e.g. E' and A1'). In practice, the MoOxSy compound is barely observed in literature except when the MoS2 synthesis conditions are not completely optimized, such as when the MoO3 precursor is too rich. From published studies, [34][35][36] MoOxSy is partially oxidized MoO3 precursors during the sulfurization process under S deficient conditions. The morphology of MoOxSy appears as thick orthorhombic particles rather than triangular crystalline MoS2 monolayer films. Furthermore, MoOxSy can be easily observed by an optical microscope due to its relatively large size of a few to tens of micrometers, which was not the case in this work. In conclusion, the existence and effect of MoOxSy are negligible.  The single crystalline domain (size of tens to around 100 µm) contains minimally exposed edges; hence, the edge sites to atomic bulk ratio is minimal. The effect of the edge sites in the oxidation process of monolayer MoS2 should be negligible in the samples. MoS2. [49] Based on the Raman spectroscopy data, it is concluded that the alkaline etching process does not structurally affect the 2H phase of monolayer MoS2. We could not detect MoO3 Raman characteristic peaks which were observed from MoO3 crystals in reference [50].    K-edge of bulk MoS2 [51][52][53][54] (a) and Mo L3-edge (b) XANES spectra of commercial bulk MoS2, [53,54] MoO2, [55] and MoO3 [55,56] as references (He atmosphere, ambient pressure, room temperature).

Reference MoS2 Layers Environment Temperature Time
Characterization Ref. [60] 1L Air 380 °C 10 min OM a Ref. [58] Few-layer Air 250 -310 °C 1 h XPS b , AFM c Ref. [61] Thick flake Air 360 °C 5 min AFM, SEM d Ref. [50] 1L, 2L O2/Ar 320 °C 1 h AFM, Raman e Ref. [62] 1 -5L H2O/Ar 500 °C 30 min AFM Ref. [63] 2L Air 330 °C 1h OM Ref. [64] 5L Air 320 °C 5 min AFM work. Our methodology employs the electron yield detection mode using tender X-rays (2-3 keV) for the characterization of ultra-diluted concentrations such as monolayer MoS2 as shown in Figure   S7. The electron-yield mode is practically based on the Auger electrons, photoelectrons, and secondary electrons overall, that escape from the sample's surface as a function of the energy of the incident X-ray radiation. After X-ray absorption, the electrons escaping the sample's surface ionize the present gas creating multiple electron-hole events which are ultimately events in the environmental gas-phase via impact ionization. Hence, this cascading effect leads to a significantly higher signal (electron yield) than those encountered by the conventional XAS modes. [15,[76][77][78][79] Table S4. Statistical parameters obtained from the linear combination fitting analysis (LCFA) of the operando Mo L3-edge X-ray absorption near edge structure (XANES) spectra when including/excluding different reference standards (MoS2, MoO2, MoO3). The analysis was done with ATHENA software (part of the DEMETER package v0.9.26) [11] within an energy range of −20 eV below to +30 eV above the edge (defined as the first derivative of the Mo L3-edge white line). LCFA reconstructs the sample spectra using the aforementioned model standards (i.e., etched monolayer MoS2, MoO2, and MoO3). The goodness of fit is reported using the R-factor and the reduced chi-squared. [11] As described by the author of the ATHENA software, here the R-factor is the mean square sum of the misfit at each data point. [11] For the reduced chi-squared, due to the solitary nature of time sequence XAS measurements/scans, ATHENA assumes 1 as the value of measurement uncertainty, resulting in very small values for chi-squared. Hence, although the single values cannot assert the goodness of the fit, it is possible to relatively compare successive fits as provided below. Using all three reference standards results in improvements of a factor of 2 and 5 for the average R-factor for all temperature points for both samples; and, an improvement on the fit by a factor of 2.5 and 5.5 for the average reduced chi-squared values when only considering MoO3 and MoO2 standards in the fit, respectively. The fits were performed assuming that all components weights' in the fitting add to unity and constraining the weights of the components as positive values.