On-stack two-dimensional conversion of MoS₂ into MoO₃

Figures 1(a) and (b) show optical micrographs of 1L and 2L MoS₂ (1L_MS and 2L_MS, respectively) obtained before and after a series of exposures to low-frequency O₂ plasma (see methods for the details). The accumulated exposure time (t_ox) represents the total reaction time. The optical contrast of both samples significantly decreased at t_ox ≥ 26 s, and the flake of 1L_MS could be barely identified at t_ox = 242 s. The change in the optical contrast was attributed to the conversion to 1L MoO₃ (1L_MO) with negligible absorption in the visible range, as will be shown below. A similar contrast change was observed for nL_MS (n = 3–4) (see figure S1). Figure 1(c) shows the Raman spectra of 1L_MS obtained in the ambient conditions at various t_ox. Among the four Raman-active modes of bulk 2H-MoS₂ (space group ${D}_{6h}^{4})$ [42], the in-plane ${{\rm{E}}}_{2{\rm{g}}}^{1}$ and out-of-plane ${{\rm{A}}}_{1{\rm{g}}}$ were shown to be useful in determining the number (n) of layers of thin MoS₂ samples [6]. Although their symmetry representations vary depending on n (E' and A₁' for odd-numbered nL_MS belonging to the ${D}_{3h}^{1}$ space group; E_g and A_1g for even-numbered nL_MS belonging to the ${D}_{3d}^{3}$ space group) [43], the bulk-notations will be used for simplicity according to the original report for 1L [6]. The frequency difference (Δω) of the two Raman peaks was found to be 18.5 cm⁻¹, and their widths were comparable to those of the previous report [6], which confirms that the sample is single-layered. The lack of the defect-activated LA(M) peak also verifies its high structural quality (figure S2) [44]. As further direct evidence, atomically resolved transmission electron microscope (TEM) images and diffraction patterns were obtained for freestanding 1L samples (figure S3).

Zoom In Zoom Out Reset image size

Oxidation induces a few distinctive changes in the Raman spectra. The in-plane Raman mode softened by ∼1 cm⁻¹ on the first exposure for 2 s and downshifted further with the other mode moving in the opposite direction when subjected to additional exposure (t_ox = 8 s). Since the newly evolving Raman peaks have both lower and higher peak frequencies than their pristine counterparts, their frequency difference, Δω_(NC), is ∼5 cm⁻¹ larger than that of pristine 1L_MS. Both peaks also broadened asymmetrically and displayed attenuated intensity. For t_ox = 26 s, when significant loss was observed in the optical contrast in figure 1(a), for example, the Raman intensities I(${{\rm{E}}}_{2{\rm{g}}}^{1})$ and I(A_1g) decreased to ∼5% of those for the pristine sample, with both peaks moving further away from each other and further broadening. The additional exposure (t_ox = 80 s) led to no detectible signal for both peaks, indicating complete loss of MoS₂. Within the phonon confinement model, the oxidation-induced downshift (upshift) of ${{\rm{E}}}_{2{\rm{g}}}^{1}$ (A_1g) can be explained by the relaxation of the fundamental Raman selection rule [45]. In essence, the defects confining phonons allow scattering of phonons with energies lower (higher) than ${{\rm{E}}}_{2{\rm{g}}}^{1}$ (A_1g) near the zone center. The low (high)-frequency shoulders of ${{\rm{E}}}_{2{\rm{g}}}^{1}$ (A_1g) that caused the asymmetric broadening are also due to the defects, which selectively allow scattering of the phonons near the M points in the double-resonance Raman scattering mechanism [46]. Indeed, the spectra of partially oxidized 1L and 2L (figures 1(c) and (d)) are well described by the five Voigt functions, including the defect-activated TO(M), LO(M) and ZO(M) [44, 45]. As discussed in detail below, plasma-generated reactive oxygen species create sub-oxide defects (MoS_xO_1−x) on the basal plane of MoS₂, essentially forming a nanocrystalline (NC) phase MoS₂ sheet.

The Raman spectra of multi-layered samples showed similar spectral changes with a layer-by-layer oxidation behavior. The in-plane mode of 2L_MS in figure 1(d), for example, showed a noticeable broadening due to the M-point phonons for t_ox = 8 s and became sharper with 40% attenuated intensity for t_ox = 26 s. Additional exposure (t_ox = 80 s) led to a further decrease in intensity and significant broadening. The NC-phase observed for t_ox = 8 s is formed on the top MoS₂ layer, with the bottom layer remaining intact. The disappearance of the NC-phase features for t_ox = 26 s can be attributed to complete oxidation of the top layer, which is only weakly coupled with the still pristine bottom layer responsible for the remaining half of the Raman signals. The second appearance of the broadening for t_ox = 80 s indicates that the oxidative attack reached the bottom layer, forming NC-phase domains. Additional exposure (t_ox = 242 s) led to a complete loss of Raman signal. These observations indicate that the oxidation proceeds in a layer-by-layer manner and is much slower for inner layers that are protected by the outer oxides.

Since optical SHG requires a lack of inversion symmetry [47], it can serve as a sensitive symmetry probe for chemical modification occurring in thin MoS₂. As shown in figure 2(a), a strong SHG signal occurred at 400 nm when the fundamental 800 nm Ti:sapphire laser beam with a nominal pulse width of 140 fs was irradiated on non-centrosymmetric pristine 1L_MS [48]. An input power dependence of the SHG peak intensity (I_SHG) confirmed the two-photon process (figure S4). The intensity of the SHG peak (I_SHG) indeed decreased with the plasma oxidation and finally reached zero for 1L_MO (t_ox = 20 s), which correlates nicely with the Raman intensity variation of MoS₂ (figure 2(b)). Thus, the attenuation of I_SHG can be attributed to the destruction of the 1L MoS₂ crystal. As shown in figures 2(a), 2L MoS₂ presents a striking contrast to the case of 1L. I_SHG(2L_MS) is ∼60 times smaller than I_SHG(1L_MS) because even-number-layered MoS₂ crystals are centrosymmetric and thus do not support SHG. The residual signal of 2L_MS is attributed to a minor interlayer asymmetry induced by the presence of the SiO₂/Si substrate [49]. A series of oxidation treatments, however, activated and then deactivated SHG by the 2L flake (figure 2(a)). Figure 2(b) reveals that I_SHG intensity increased at the expense of the Raman signal represented by I(A_1g). Since I_SHG reached its maximum at t_ox = 30 s, for which the Raman intensity is close to that of 1L_MS, the treated sample essentially corresponds to 1L_MO/1L_MS. This assignment is also supported by the fact that additional treatments led to a decrease in I_SHG, indicating oxidative degradation of the bottom sulfide layer. The lack of SHG signal from the completely oxidized layers (1L_MO and 2L_MO) might be explained if the 2D oxides are a centrosymmetric crystal such as bulk α-MoO₃ and belong to the space group of ${D}_{2h}^{16}$ [50]. As will be shown below, however, the 2D oxides are far from α-MoO₃ and more likely to be amorphous and isotropic, thereby not generating SHG signals.

Zoom In Zoom Out Reset image size

Despite the degradation of the crystalline lattice, the crystallographic orientation could still be determined for partially oxidized MoS₂ since I_SHG is strongly dependent on the polarization angle $(\theta -{\theta }_{0})$ between the fundamental polarization $({\mathop{E}\limits^{ \rightharpoonup }}_{\omega })$ and armchair direction $(\mathop{{\rm{A}}{\rm{C}}}\limits^{ \rightharpoonup }),$ as defined in figure 2(c). For polarization-resolved SHG measurements, samples were rotated to vary θ, and the SHG signal parallel to the input polarization, ${{\boldsymbol{I}}}_{{\rm{SHG}}}^{\parallel },$ was collected using a polarizer located in front of the detector. The polar plot for 1L_MS in figure 2(d) revealed the 6-fold symmetry of ${{\boldsymbol{I}}}_{{\rm{SHG}}}^{\parallel }\propto {\cos }^{2}3(\theta -{\theta }_{0})$ [48], which also determined that θ₀ = 15.0° ± 0.2°, the angle of $\mathop{{\rm{A}}{\rm{C}}}\limits^{ \rightharpoonup }$ with respect to a preset laboratory axis ($\mathop{{\rm{L}}{\rm{A}}{\rm{B}}}\limits^{ \rightharpoonup }).$ When oxidized to give ∼0.4 L_MS, as judged from its Raman spectra (t_ox = 15 s), ${{\boldsymbol{I}}}_{{\rm{SHG}}}^{\parallel }$ decreased by ∼75% but still showed the same angular dependence as the pristine 1L_MS. The oxidation-induced modulation in I_SHG (figures 2(a) and (b)) and the robust θ-dependence demonstrate that the SHG process can be utilized to characterize chemical changes in 2D materials.

The on-stack transformation was found to proceed in a layer-by-layer manner and could be controlled with a high thickness resolution, even allowing formation of single-layer MoO₃. To monitor morphological changes from the transformation, an amplitude-modulated, non-contact AFM was exploited. The height profile across the pristine 2L_MS-1L_MS area in figure 3(a) revealed a step with a height of 0.70 ± 0.1 nm, which agrees well with the interlayer spacing (0.62 nm) of 2H-MoS₂ [51]. Note that the height of the 1L_MS-SiO₂ step varied over 1.2–1.8 nm from sample to sample due to the chemical and electrostatic contrast between dissimilar materials [52, 53] and interlayer molecular species trapped during the exfoliation. When 2L_MS was completely oxidized (t_ox = 242 s), the 2L_MO-1L_MO step height increased to 1.8 ± 0.1 nm (figures 3(b) and (e)). As depicted in figure 3(f), the step height corresponds to the thickness of single-layer MoO₃ and is ∼2.5 times the interlayer spacing (0.69 nm) of α-MoO₃ [50]. A similar change was observed for the 2L_MS-1L_MS step of another sample when treated for t_ox = 242 s (figures 3(c) and (d)). After the same treatment, however, the 3L_MS-2L_MS step height changed only by ∼0.2 nm. In contrast to the 2L region, the Raman spectrum of the 3L region showed that ∼15% of the lowermost layer remained intact (see figure S5 for Raman spectra obtained after each step of the sequential oxidations). Thus, the step height measured across the 3L-2L region corresponds to the thickness of the partially oxidized bottom layer.

Zoom In Zoom Out Reset image size

Despite the increase in thickness, MoO₃ layers were still highly flat, as seen in the AFM height images in figure 3. The change in the surface morphology of the 2D oxides was further quantified by the root-mean-square roughness (R_q), or standard deviation of height distribution, shown in figure 3(e). The R_q of 1L_MS was 170 ± 10 pm, which is slightly higher than that of graphene [54] and equivalent to that of completely oxidized 1L_MO, 180 ± 10 pm for t_ox ≥ 80 s, as shown in figure S6. Notably, however, the surface roughness showed a spike of 230 ± 10 pm for t_ox = 2 s and a sharp decrease for t_ox = 8 s. Similar spikes were also observed for 2L_MS and 3L_MS when treated for 2 s. The R_q values of 2L_MO and 3L_MO (170 ± 10 pm) are also equivalent to those of their pristine counterparts. We attribute the roughness surge to clusters of partially oxidized MoS₂, possibly molybdenum oxysulfides (MoS_xO_y) [55] that serve as nucleation sites during the plasma oxidation. This picture is also corroborated by the drastic PL quenching for t_ox = 2 s, as will be shown below. However, the Raman spectra did not show a noticeable change (figures 1(c) and (d)) because the majority of the sample remained intact for the short reaction time.

The chemical transformation was revealed by XPS. Because the probe size of the XPS instrument was larger than the typical size of the exfoliated samples (<20 μm), large-area MoS₂ films grown by the CVD were used (see methods for the details of preparation). While the effective thickness was found to be 2L based on the peak difference (Δω), their Raman intensity was lower than that of mechanically exfoliated samples, suggesting lower crystallinity (figure S7). Figure 4(a) presents the Mo 3d spectra obtained as a function of t_ox. The pristine sample showed the strong doublet of Mo⁴⁺ arising from MoS₂, with a binding energy (E_B) of 229.5 eV (232.6 eV) for 3d_5/2 (3d_3/2), which is consistent with the average literature value of E_B(3d_5/2) = 229.25 eV for bulk MoS₂ crystals [56]. The Mo⁶⁺ species of the pristine sample, responsible for the minor doublet with E_B = 232.7 eV (235.7 eV) for 3d_5/2 (3d_3/2), were attributed to residual MoO₃ (<20% of the total Mo species) formed before or after the CVD growth. Upon plasma treatment, the Mo⁶⁺ doublet grew significantly at the expense of the Mo⁴⁺ doublet, indicating the conversion to MoO₃. The oxidation was also confirmed by the decrease in the intensities of the S 2 s peak (E_B = 226.7 eV) in figure 4(a) and the S 2p doublet (E_B = 161.9 eV for 2p_3/2) in figure 4(b). The current data cannot exclude the possibility that a small fraction of Mo⁵⁺ species (E_B = 231.1 eV for 3d_5/2) [57], representing oxygen vacancy defects generated by incomplete oxidation, may be present.

Zoom In Zoom Out Reset image size

To shed more light on the chemical transformation, we obtained optical absorption spectra using reflectance spectroscopy (figure 5, see methods for the details). For a very thin film (thickness $\ll$ λ) supported on a thick transparent substrate with a refractive index of n₀, the fractional change in reflectance $({{\rm{\delta }}}_{{\rm{R}}})$ is given as follows: ${\delta }_{{\rm{R}}}=\tfrac{R-{R}_{0}}{{R}_{0}}=\tfrac{4}{{n}_{0}^{2}-1}{\rm{A}},$ where R, R₀ and A are the reflectance of the film, reflectance of the bare substrate, and absorbance [58]. Pristine 1L_MS–3L_MS exfoliated on quartz substrates showed characteristic absorption peaks at 1.90, 2.05 and 2.85 eV, which were denoted, respectively, by A, B and C as per an earlier work [59]. The two former excitonic peaks originate from the direct-gap transitions between the valence and conduction bands at the K points in the Brillouin zone [5]. The high energy peak, C, arises from nearly degenerate multiple excitonic states [60] or transitions across nested valence and conduction bands [61]. All three peaks downshifted in energy with increasing thickness [5]. When the samples were treated for t_ox = 80 s, which is expected to oxidize ∼1.7L, the overall absorbance decreased significantly for all thicknesses. The A and B peaks of 1L, in particular, almost disappeared, with the major absorption edges relocated to >2.7 eV, which agrees well with the fact that MoO₃ is a wide-bandgap semiconductor [62]. The partially oxidized 2L exhibited downshifts of the A and B peaks but an upshift of the C peak, and 3L exhibited similar shifts but to a lesser degree. The spectral changes are attributed to the oxidation-induced defects and can be explained by the drastic changes in the electronic band structures of O-substituted MoS₂ [63]. To determine the optical bandgap of the 2D MoO₃, Tauc plot analysis was taken for the UV/Vis absorption spectra of CVD-grown large-area samples (figure S8). Despite some inconsistencies in the literature [64], it is generally accepted that an optical gap energy (E_g) of an amorphous semiconductor can be extrapolated using ${(\alpha {\rm{h}}\nu )}^{1/m}\propto (h\nu -{E}_{{\rm{g}}})$ [65, 66], where α, ${\rm{h}}\nu$ and m are, respectively, the absorption coefficient, photon energy and an exponent specifying the nature of the optical transition (m = 2 for an indirect allowed transition; m = 1/2 for a direct allowed transition). By choosing m = 2 for α-MoO₃ as an indirect bandgap semiconductor [67], E_g was determined to be ∼3.0 eV for 1L_MO and 2L_MO. The resulting bandgap energies are in a good agreement with those of α-MoO₃ crystals [62] and thermally deposited MoO₃ films [68, 69]. Despite the similar bandgap energies and apparent layered structure of plasma-generated MoO₃, however, no Raman signal could be detected from the 2D oxides, possibly due to their amorphous nature. A careful Raman analysis in comparison with mechanically exfoliated thin α-MoO₃ crystals set an upper bound for α-MoO₃ content of ∼1% (figure S9).

Zoom In Zoom Out Reset image size

The change in the PL spectra in figure 6 revealed further details of the chemical modifications occurring in 2D MoS₂. The PL spectrum of the pristine 1L_MS mainly consists of the two exciton peaks A and B, respectively located at 1.84 and 2.00 eV, which is in good agreement with the absorption spectra. Both originate from the direct-gap transitions between the conduction and valence bands at the K points in the Brillouin zone [5, 7]. First, we note that the PL is drastically quenched by the plasma treatments. After a 2 s exposure, for example, the intensity of A (I_A) decreased by ∼65%, although the Raman spectra in figure 1 show no significant change in I$({{\rm{E}}}_{2{\rm{g}}}^{1})$ and I$({{\rm{A}}}_{1{\rm{g}}}).$ The quenching can be attributed to the plasma-generated defects that were confirmed by the defect-induced Raman peaks, as will be further discussed below. When further defects were introduced to form the NC phase (t_ox = 8 s), the PL intensity was only ∼10% of the pristine value. The completely oxidized 1L_MO gave no PL signal, which is consistent with the fact that its E_g is larger than the excitation photon energy of 2.41 eV. 2L_MS in figure 6(b) shows a similar defect-mediated PL quenching—∼60% decrease for t_ox = 2 s and almost zero PL signal for t_ox = 8 s, which introduced a significant number of defects judging from the Raman spectrum (figure 1(c)). Upon further oxidation (t_ox = 26 s), however, the PL intensity increased back to ∼30% of that for the pristine 2L_MS. We note that the Raman spectra showed that the top layer was almost completely oxidized with the bottom layer mostly intact (figure 1(d)). This suggests that the system essentially became a 1L_MO/1L_MS/SiO₂ sandwich, which further corroborates the layer-by-layer oxidation. Thicker samples showed a similar trend, but the PL intensity oscillation of nL_MS could be seen only when t_ox was selected in such a way that (n − 1)L_MO/1L_MS was nicely formed.

Zoom In Zoom Out Reset image size

Structural defects in crystalline semiconductors have various modes of interaction with excited charge carriers, and their interactions become more significant in lower dimensional materials due to confined wave functions within a tighter space. For carbon nanotubes with low lying dark exciton states, sp³-type defects form new sets of radiative energy levels granting an enhanced PL quantum yield [70]. Defects may trap excited free charge carriers and mediate ultrafast Auger decay of photoexcitation in quantum dots [71]. Excitons can be localized at defects and radiatively decay at lower energies, as shown for 1L MoS₂ irradiated with α particles [72] and 1L WS₂ treated with Ar plasma [73]. Defects in some 2D semiconductors have also been shown to serve as single photon emitters [74, 75]. Localization followed by non-radiative decay is one among various quenching routes of excitons, which also include exciton-exciton annihilation and electron–phonon interactions [76]. The PL quenching observed at the very early stage of the oxidative conversion can be attributed to partially oxidized defects, including the nanometer scale protrusions that were responsible for the roughness spikes (figure S6). The higher defect-sensitivity of PL can be more clearly seen in the comparison between Raman and PL intensities given as a function of t_ox (figure S10).

The overall change can be summarized by the scheme for 3L_MS with terraces of 2L_MS and 1L_MS presented in figure 3(f). At the early stage of reactions (t_ox = 2 s), the top surface of 3L_MS exposed to the gaseous oxidants undergoes partial oxidation forming nanometer-scale oxysulfides (MoS_xO_y) clusters that are responsible for the increased R_q. At this stage, the average thickness of the top layer is not much different from that of pristine 1L_MS or the interlayer spacing of 2H-MoS₂ crystals. Although the Raman spectra reveal no significant change, the PL intensity of 3L_MS drops significantly due to the defects serving as PL quenchers. Additional exposure to oxidants induces the NC-phase Raman peaks and almost complete quenching of the PL. When the first top layer is completely converted to MoO₃, its thickness increases to 1.8 nm, which is 2.5 times the interlayer spacing of α-MoO₃ crystals. At this stage where 1L_MO/2L_MS is formed, the PL intensity can be recovered back to that of 2L_MS because of the lack of quenching sites within the remaining 2L_MS. Further exposure will repeat the above reactions for the second top layer but with a greatly decreased reaction rate.

Although the Raman analysis suggests that the 2D oxides are amorphous, their high flatness and modulation in SHG and PL intensities indicate that the oxidation reaction proceeds in the layer-by-layer manner. Thus, it is likely that the oxides are also layered like α-MoO₃. Since the number density of Mo in a single α-MoO₃ layer [67] is 18% higher than that for 1L_MS [51], a 2D conversion into the crystalline oxide would lead to a large tensile strain (∼9%) when the areal density of Mo is maintained or generate voids requiring significant mass transport to compensate for the difference in the Mo densities. Moreover, the average thickness of 1L_MO (figure 3) is 2.5 times the interlayer spacing of α-MoO₃. These facts suggest that the 2D oxides are much less dense than α-MoO₃. Despite the apparent flatness and lack of pits or cracks in the 2D oxides, these considerations lead us to conclude that there must be numerous structural irregularities, including fine voids, that cannot be detected by the AFM probes. Our results also showed that the plasma oxidation is highly effective for the topmost sulfide layer and increasingly slower for the next buried layers. Such voids may serve as a route for the oxidants required for the reactions underneath the top layer.

In summary, we demonstrated a model 2D chemical reaction that converts 1L sulfides into 1L oxides, thus allowing MoO₃/MoS₂ heterostructures. Single and few-layer MoS₂ prepared via micromechanical exfoliation of 2H-MoS₂ crystals were treated with strong oxidants generated from O₂ plasma. The early stage of the reaction was detected by defect-induced Raman peaks, drastic quenching of PL signals and sub-nm protrusions in AFM images. As the reaction proceeded from the uppermost layer to buried layers, PL and SHG signals showed characteristic modulations revealing a layer-by-layer conversion. The plasma-generated 2D oxides, confirmed as MoO₃ by XPS, were found to be amorphous but highly flat with a surface roughness of 0.18 nm, comparable to that of 1L MoS₂. The rate of oxidation quantified with Raman spectroscopy decreased very rapidly for buried sulfide layers due to protection by the surface 2D oxides. As exemplified in this work, the on-stack chemical transformation can be applied to other 2D materials in forming otherwise unobtainable materials and complex heterostructures, thereby expanding the palette of 2D material building blocks.

Single and few-layer MoS₂ samples were prepared by mechanical exfoliation [3] of molybdenite, a natural mineral of 2H-MoS₂ (SPI). Silicon wafers with 285 nm thick SiO₂ layers were used as substrates. Ultrathin layers of MoS₂ were identified using an optical microscope, and their numbers of layers and quality were determined using Raman spectroscopy [6]. To prepare thin α-MoO₃ crystalline flakes as a Raman standard (figure S9), a similar mechanical exfoliation was applied to a powder of MoO₃ (Materion).

Two different CVD growth methods were used in synthesizing large area MoS₂ films. In one approach for the samples used for XPS measurements, Mo metal of 0.5 nm for 2L_MS was deposited on SiO₂/Si substrates by e-beam evaporation. The samples were positioned in a quartz tube furnace, which was then evacuated to a low vacuum. The samples were heated to 750 °C under a flow of Ar at 50 ml min⁻¹. Since the surface of deposited Mo films usually oxidizes when exposed to ambient air during transfer to the furnace, H₂ gas was briefly introduced at 750 °C to reduce the Mo oxides. After the pre-annealing and reduction processes, a H₂S/H₂/Ar gas mixture (1:5:50) was introduced for 15 min to sulfurize Mo films into MoS₂. The pressure in the furnace was maintained at 300 mTorr during the sulfurization step. To enhance the crystallinity of grown films, the samples were further annealed briefly at 1000 °C under a flow of the H₂S/Ar gas mixture (1:50). In the approach for the samples used in UV/Vis measurements, 5–10 mg of MoO₃ powder (Sigma-Aldrich) in a quartz boat was placed at the center of the furnace, and quartz substrates were placed near and downstream of the boat. The furnace was heated to 600 °C at a rate of 20 °C min⁻¹ under Ar gas flow at a rate of 200 ml min⁻¹. At 600 °C, H₂S gas was introduced for 30 min at a rate of 1 ml min⁻¹ to sulfurize MoO₃ into MoS₂. Then, the samples were rapidly cooled to room temperature. The average thickness and quality of the grown films were characterized by their Raman and PL spectra.

For oxidation, samples were treated with either of two low-frequency plasma instruments operated at 50 kHz, a commercial unit (Femto Science Inc., Cute-1MP) and a quartz-tube-type unit. The partial pressure of O₂ in the two plasma chambers was 540 and 300 mTorr, respectively. For oxidation reactions, oxygen plasma was maintained for a pre-specified period of t_ox at a power of 10 W. Because of the differences in the instruments and their detailed procedures, the apparent oxidation was ∼3 times faster for the former than the latter. Thus, t_ox given in this work has been corrected for the difference.

The Raman and PL spectra of the samples were obtained with a microscope-based (Nikon, Ti) Raman setup that is equivalent to that detailed elsewhere [77, 78]. Briefly, the 514 nm excitation beam from a solid state laser (Cobolt, Fangdango) was focused onto an ∼1 μm diameter spot using an objective lens (40X, numerical aperture = 0.60). The back-scattered signal was collected by the same objective and guided to a spectrograph (Princeton Instruments, SP2300) combined with a liquid nitrogen-cooled charge-coupled detector (CCD) (Princeton Instruments, PyLon). The spectral resolutions judged from the FWHM of the Rayleigh line were 3.0 and 12 cm⁻¹ for the Raman and PL spectra, respectively.

A similar microscope-based spectroscopy setup was employed for the SHG detection. A train of 140 fs pulses from a Ti:Sapphire laser (Coherent, Chameleon) operated at 800 nm was focused onto a 1.6 μm spot in FWHM with an objective lens (40X, numerical aperture = 0.60). The backscattered SHG signal centered at 400 nm was collected and fed to a spectrograph (Andor, Shamrock 303i) equipped with a thermoelectrically cooled CCD (Andor, Newton). To vary the polarization angle of the fundamental laser with respect to the MoS₂ lattice vectors, samples were rotated in a rotational mount with an angular accuracy better than 0.2°. For polarized detection, the SHG signal was filtered with a polarizer located in front of the spectrograph.

The topographic details of the samples were investigated using an atomic force microscope (Park Systems, XE-70). The height images were obtained in non-contact mode using Si tips with a nominal tip radius of 8 nm (MicroMasch, NSC-15). The AFM tip was driven to a free oscillation with an amplitude of 20 nm and engaged in amplitude-modulated scanning with an amplitude set-point of ∼14 nm.

The elemental information of the CVD-grown samples was obtained using an x-ray photoelectron spectrometer (Thermo Scientific, K-Alpha^TM+). XPS measurements were carried out with Al K_α line (1.4866 keV). The binding energy of the photoelectrons was calibrated with respect to Mo(IV) 3d_5/2 (E_B = 229.5 eV), which was more reliable than the C 1 s peak (E_B = 284.6 eV) originating from carbon residues on the samples.

For the samples exfoliated onto quartz substrates, fractional changes in reflectance were used to obtain absorption spectra. As a broadband Vis/NIR light source, the output of a tungsten-halogen lamp (iiSM Inc., Mighty Light Beam) was collected with a multimode optical fiber (core diameter of 50 μm) and guided to the micro-spectroscopy setup used for the SHG measurements. The FWHM of the focus spot was 0.8 and 2.0 μm in the visible and NIR ranges, respectively. For the samples CVD-grown on quartz substrates, the absorption spectra were obtained with a UV/visible spectrometer (Jasco, V-650).

For TEM measurements, MoS₂ flakes exfoliated onto SiO₂/Si substrates were transferred to carbon-film grids with 2 μm holes (Quantifoil) using an isopropyl alcohol and KOH solution [79]. The samples were analyzed using an aberration-corrected FEI Titan Cubed TEM (FEI, Titan³ G2 60-300), which was operated at 80 kV acceleration voltage with a monochromator. The microscope provides sub-angstrom resolution at 80 kV and −11 ± 0.5 μm of spherical aberration (C_s).