HfO₂ on UV–O₃ exposed transition metal dichalcogenides: interfacial reactions study

Two-dimensional transition metal dichalcogenides (TMDs), as part of the mono- and few-layer materials library, have potentially exceptional properties such as dangling bond-free surfaces and thickness-tunable band gap energies [1, 2], making them outstanding candidates as channel materials for beyond Si-CMOS device structures [3]. The electronic and optoelectronic properties of MoS₂, a readily available material in nature, have been extensively studied in recent years [4]. However, the presence of a Schottky barrier at the metal/MoS₂ interface has limited the MoS₂-based field-effect transistors (FETs) performance [5]. Moving forward into the exploration of other members in the TMD library, several efforts have focused on studying 2D WSe₂, where field-effect mobility (μ_FE) values up to ∼500 cm² V⁻¹ s⁻¹ have been reported [6]. Furthermore, band–band tunneling properties of WSe₂ for tunnel FET applications have been demonstrated [7]. Following these efforts, MoSe₂ has been implemented in FETs, exhibiting a promising performance (i.e. μ_FE ∼ 150–200 eV) according to recent reports [8, 9]. It has been demonstrated theoretically [10] and experimentally [11–13] that the limiting effect on mobility associated with Coulomb scattering can be reduced by a having a dielectric material on the 2D channel, motivating several studies related to dielectrics on TMDs. However, it has also been shown that, given the relative inertness of the TMDs surfaces, the direct atomic layer deposition (ALD) of high-k dielectrics on clean, un-treated surfaces results in the formation of clusters or islands instead of continuous films[14, 15]. Previously, it was shown that uniform, pin-hole free and thin (∼4 nm) Al₂O₃ films can be obtained by performing a ultraviolet (UV)–O₃ surface pre-treatment on MoS₂, where covalent oxygen functionalization of the MoS₂ top-most sulfur layer occurred, while avoiding molybdenum oxidation [16]. Alternative processes have been reported in order to improve the ALD high-k nucleation on MoS₂ such as oxygen plasma pre-treatment [15], the use of ozone as an oxidant precursor during ALD of Al₂O₃ [17], and metal-oxide seed layers [18].

Yet, whether the proposed processes can be directly extended to other TMDs remains an important question, since different compositions could result in variation of surface properties. Here, in situ monochromatic x-ray photoelectron spectroscopy (XPS) studies were used to characterize the effect of a UV–ozone treatment of MoSe₂ and WSe₂ single crystal surfaces. Also, the surface reactivity towards oxidation of these TMDs was compared and modeled by density functional theory (DFT) calculations. It was found that surface oxides formed upon UV–O₃ exposure of the diselenides surface. Additionally, XPS monitored the interfacial chemistry between ALD-HfO₂ and the UV–O₃ treated TMD surfaces. Finally, the HfO₂ film uniformity on the UV–O₃ treated TMDs was investigated and correlated with the interfacial chemistry.

The UV–O₃ treatments and ALD studies described here were performed in situ using an ultrahigh vacuum (UHV) cluster tool described elsewhere [19]. XPS was carried out using a monochromated Al K$\alpha$ x-ray source (hν = 1486.7 eV) and equipped with a seven channel analyzer, using a 15 eV pass energy. The XPS peak deconvolution was achieved with the software Aanalyzer^® [20], using Voigt functions with independent control of the Lorentzian and Gaussian components, and applying a dynamic Shirley background subtraction. The top-most layers from MoS₂ (SPI Supplies), MoSe₂ (2D Materials) and WSe₂ (Nanoscience Instruments) bulk crystals (all (0001) orientation) were mechanically exfoliated using Scotch^® Magic™ tape, and the freshly as-exfoliated bulk materials were loaded immediately into UHV. All the UV–O₃ exposures were performed at room temperature as described in [16]. In an interconnected chamber, the sample surfaces are placed within a few mm from a UV lamp (low pressure mercury lamp) in the presence of O₂ (O₂ partial pressure ${{P}_{{{{\rm O}}_{{\rm 2}}}}}~=~900 mbar)$ to generate ozone [21]. The ALD experiments were performed using a Picosun ALD reactor, with an operating pressure of ∼10 mbar. The hot wall reactor is integrated to the cluster tool to enable surface analysis without spurious contamination from atmospheric exposure [19]. For the HfO₂ deposition, tetrakis (dimethylamino) hafnium (TDMA-Hf) and H₂O were used as the precursors, having a substrate deposition temperature of 200 °C. The precursor pulse and purge times are 0.1 and 4 s, respectively. High purity (99.999%) N₂ was used as the precursor carrier and purging gas. After one full-cycle of TDMA-Hf and H₂O, the samples were taken out from the ALD reactor and transported to the XPS analysis chamber through a transfer tube under UHV (∼1 × 10⁻¹⁰ mbar). After analysis, the samples were transported back to the ALD reactor for further deposition. This process was repeated for subsequent depositions for a total of 30 ALD cycles. Tapping-mode atomic force microscopy (AFM) images were obtained ex situ using an atomic probe microscope Veeco, Model 3100 Dimension V. High angle annular dark field scanning transmission electron microscopy (STEM) was performed using a JEOL ARM200F with probe spherical aberration (Cs) corrected operated at 200 kV. TEM sample preparation was accomplished using a FIB-SEM Nova 200.

The first-principles modeling, based on DFT [22, 23] with plane wave basis sets and Projector Augmented Wave pseudopotentials [23, 24] as implemented in the Vienna ab initio Simulation Package [22, 25, 26] were conducted to find the energetics of oxidation. The electronic wave functions were represented by plane wave basis with a cutoff energy of 500 eV. The exchange and correlation interactions were incorporated as a functional of the generalized gradient approximation [24, 27, 28]. A 5 × 5 supercell of host MX₂ monolayers (M = W or Mo and X = S, Se) was used for the DFT calculations. Each model has ∼16 Å vacuum to avoid interaction between replica images as a result of the periodic boundary conditions. The k point grid of 12 × 12 × 1 was adopted for Brillouin zone sampling. The energy and Hellmann-Feynman force convergence criteria chosen were 10⁻⁴ eV and 0.01 eV Å⁻¹, respectively.

In order to determine whether oxygen functionalization by UV–O₃ was extendable to TMDs beyond MoS₂, a room temperature UV–O₃ treatment was performed on WSe₂ and MoSe₂, while MoS₂ was also used as a control sample for comparison with previous work [16]. XPS spectra of MoS₂, MoSe₂, and WSe₂, after 3, 6 and 15 min of UV–O₃ exposure, is shown in figures 1(a)–(c) respectively, while table 1 summarizes the bonds formed after UV–O₃, showing their binding energy (BE) and ratios of the integrated intensity of the corresponding peaks with respect to the bulk signal. The sequential UV–O₃ exposure on MoS₂ shows that S–O bond formation is detected after 3 min, and by increasing the exposure time, i.e. 6 and 15 min, the oxygen adsorption saturates to form 1 ML [16]. Under these conditions, no Mo–O bond formation is detected, as previously reported. It has also been reported that UV–O₃ can cause molybdenum oxidation in MoS₂, when the exposure is much longer (60 min) [29]. This implies that the oxidation process is dose and time dependent.

Zoom In Zoom Out Reset image size

In the case of MoSe₂, the 3 min UV–O₃ exposure initially results in the formation of Se–O bonding, without causing Mo oxidation. The nature of the Se–O results analogous to the S–O observed in MoS₂ [16], which may be described as chemisorbed oxygen on the surface, i.e. covalent bond formation between oxygen and the top most selenium layer, while preserving the covalent bonding within the MoSe₂ layer. However, in contrast to MoS₂, the 6 min UV–ozone exposure results in sub-stoichiometric molybdenum oxide (MoO_x) formation. The MoO_x peak shows a symmetric lineshape, suggesting that Mo is present in only one of the seven different phases that has been identified for MoO_x, x = 2.75–3 [30, 31]. In the Se 3d region, the previously observed Se–O bond is below the detection limit, however, a new peak at ∼59 eV indicates that selenium is further oxidized forming a selenium sub-oxide SeO_x. In addition, an extra feature at ∼0.6 eV higher in binding energy than the bulk MoSe₂ peak is observed in both Mo 3d and Se 3d core levels (blue peak in figure 1(b)).

Previous studies in MoS₂ correlated the appearance of peaks at higher BE from the main bulk peak, for both Mo 3d and S 2p, with the existence of molybdenum oxysulphide (MoS_xO_y) [32–34], where the BE position of such peaks was dependent on the x and y values. A peak at a slightly higher BE (∼0.5–0.8 eV) than the characteristic MoSe₂ peak in Se 3d could be also correlated with metallic Se [35], however, the fact that both core levels showed this high BE feature makes this chemical state assignment less probable. Another possible assignment could be a sub-stoichiometric MoSe_x phase, however, the existence of MoSe_x would cause the transition metal peak to shift lower in BE and the chalcogen peak to broaden [36], and this situation is not observed. Thus, the feature in blue in figure 1(b) is associated with MoSe_xO_y formation, caused by oxidation during UV–O₃ exposure. Finally, after UV–O₃ for 15 min, the MoO_x peak shifted to 0.4 eV higher BE which is consistent with a higher oxygen content in MoO_x [30, 37]. Also, the MoO_x oxide to bulk peak intensity ratio increases in comparison to the 6 min ratio, showing that the oxygen is inserted deeper within the material. In the case of Se 3d, the sub-stoichiometric SeO_x shows a shift to 0.4 eV higher BE, suggesting complete oxidation to SeO₂.

In comparison, figure 1(c) shows the XPS spectra of WSe₂ for the same UV–O₃ exposure times. After UV–O₃ exposure for 3 and 6 min, Se–O bonding is detected however the Se–O to bulk WSe₂ peak intensity ratio is lower than that for MoSe₂. W oxidation in the form of WO_x is detected after 6 min of exposure. Similar to MoSe₂, the UV–O₃ exposure of WSe₂ presents tungsten oxyselenide bonding (WSe_xO_y) after 15 min of exposure, in addition to WO_x and SeO_x formation.

Oxidation studies of these TMDs in the bulk form were performed previously by Jaegermann, et al [38], where reactivity towards oxidation was found to go from lower to higher in the sequence: MoS₂ < MoSe₂ < WSe₂. This tendency was related to the degree of mixed metal d-states from the transition metal with the p-states in the chalcogenide, based on ionization energies arguments. However, the focus of that study was the photoelectrochemical oxidation in solution, using electrolytes (i.e. K₂SO₄) as oxidizers, which might introduce additional variables in the oxidation processes. Thus, DFT calculations were performed in order to have a better understanding of the energetics involved in the oxidation of TMDs.

Figures 1(d)–(f) shows optimized structures for MoS₂, MoSe₂ and WSe₂, respectively, when an oxygen atom is adsorbed on top of the chalcogen (O–Se, O–S) and when oxygen substituted the chalcogen to form a transition metal–oxygen bond (O–Mo, O–W). The calculated formation energies of the oxygen adsorption and chalcogen substitution with oxygen are listed in table 2, which were obtained using the following equation:

$$\begin{eqnarray}&&\begin{array}{lllllllllllllll} {{E}^{{\rm form}}}=E\left( {\rm O}:M{{X}_{{\rm 2}}} \right)-E\left( {\rm pristine} \right) \\ \quad \quad \;\;-n{{\mu }_{{\rm O}}}+n{{\mu }_{{\rm S}}}, \\ \end{array}\end{eqnarray} \tag{ 1 }$$

where E^form is the formation energy of the adsorbed or substitutional oxygen on MX₂ surfaces with total DFT energy E (O:MX₂), E (pristine) is the DFT energy of the pristine MX₂ surfaces ((M = W or Mo and X = S, Se), μ_O and μ_S are the reference chemical potentials of O (oxygen molecule) and S (bulk sulfur) and n is the number of adsorption or substitutional O atoms. Under this definition, negative values of the formation energy refer to thermodynamically stable oxidation process.

Table 2.  Formation energies of oxygen adsorption ($E_{f}^{{\rm ad}})$ and oxygen replacement ($E_{f}^{{\rm rep}})$ on MoSe₂ and WSe₂ calculated by DFT.

	$E_{f}^{{\rm ad}}$ (eV)	$E_{f}^{{\rm rep}}$ (eV)
	O–S	O–Mo
MoS2	−1.12	−1.912
	O–Se	O–Mo
MoSe2	−0.284	−2.504
	O–Se	O–W
WSe2	−0.132	−2.814

According to the formation energies shown in table 2, O–S bond formation in MoS₂ is more energetically favorable and therefore more stable in comparison to O–Se in MoSe₂ and WSe₂. The DFT calculations also indicate that oxygen desorption will occur more easily in MoSe₂ and WSe₂, in comparison to MoS₂. The formation energy required for the replacement of the chalcogen to form a direct bond between oxygen and the transition metal is found to be more energetically favorable in the case of MoSe₂ and WSe₂ than for MoS₂, which is consistent with the XPS results after UV–O₃ exposure. The relative lower stability of the O–Se bond in comparison to O–S implies that desorption of oxygen, and possibly the 'kick-out' [39] of O–Se species, can leave unsaturated bonds for the transition metal, which then are readily available to form Mo–O and W–O bonds in presence of oxygen, such as during a UV–O₃ exposure. Experimentally, this oxidation process caused the formation surface oxides of 0.36 nm and 1 nm thickness in MoSe₂ for 6 min and 15 min of UV–O₃ exposure respectively, while the surface oxides in WSe₂ exhibited 0.33 nm and 0.66 nm thicknesses for the same mentioned exposure times, according with XPS thickness calculations.

These UV–O₃ exposure studies demonstrate that oxygen functionalization of MoSe₂ and WSe₂ surfaces is more difficult to control compared to MoS₂. Yet, the impact of having surface oxides on these TMDs prior to high-κ dielectric deposition by ALD is still of interest. Thus, HfO₂ was sequentially deposited by ALD on the UV–O₃ treated MoSe₂ and WSe₂, while the interface chemistry was monitored by XPS. For comparison purposes, all the samples discussed here received a UV–O₃ exposure for 6 min.

Figure 2(a) shows that after one ALD cycle on functionalized MoSe₂, the MoO_x peak intensity significantly decreases to ∼24% its initial intensity. In the Mo 3d region, a new peak is detected at 231.9 eV, which is designated as the Mo⁵⁺ oxidation state [40], indicating that the Mo–O bonds are reduced (i.e. O was transferred to Hf, resulting a lower oxidation state of Mo). Additionally, the SeO_x intensity decreases to below the detection limit, as shown in the Se 3d region. The decrease in oxide intensity during precursor pulsing in ALD, or 'self-cleaning', effect was previously studied for III–V materials, such as In_xGa_1−x As [41]. The self-cleaning reduction reactions that occurred after one cycle result in the gradual reduction of the oxidation state from Mo⁶⁺ to Mo⁵⁺. In addition, the fact that no Mo⁰ or other chemical state was detected suggests that MoO_x went through a ligand exchange reaction with the Hf precursor, forming a volatile molybdenum compound possibly of the form Mo(N(CH₃)₂)_x [42]. Based on the thermodynamic quantities, HfO₂, with a Gibbs free-energy of formation of −251.8 kcal mol⁻¹ [43], results in a more energetically favorable reaction product than MoO_x, assuming that the Gibbs free-energy of MoO_x is close to—159.65 kcal mol⁻¹ [44], which is the corresponding value for MoO₃.

Zoom In Zoom Out Reset image size

For the Se 3d feature, no lower oxidation states are detected (i.e. S⁰), indicating that SeO_x, a volatile compound, may be thermally desorbed from the surface in vapor phase [45]. Finally, the MoSe_xO_y peak is not observed after one cycle, suggesting that the MoSe_xO_y phase possibly recovers its initial state as MoSe₂ through the SeO_x desorption channel, thus contributing to the bulk Se peak signal intensity.

Further deposition of HfO₂ using five ALD cycles results in a decrease of the Mo⁶⁺ intensity to below the detection limit; only Mo⁵⁺ remains. As shown in figure 2(b), the Mo⁵⁺ to MoSe₂ integrated intensity ratio is ∼0.03, suggesting that most of the surface oxides undergo 'self-cleaning' reactions. The Se 3d spectra show no change after five cycles. After ten cycles, no interfacial oxides from MoSe₂ are detected, and further HfO₂ deposition only caused attenuation of the MoSe₂ features. Figure 3(c) shows that the chemical identity of HfO₂ remains constant through the sequential depositions.

Zoom In Zoom Out Reset image size

In an analogous process, HfO₂ was sequentially deposited on UV–O₃ treated WSe₂ followed by XPS analysis. Figure 3(a) shows the evolution of WSe₂ chemical states through the experiment. It is shown that, in contrast to MoSe₂, the decrease in intensity of the transition metal oxide in WSe₂ is minimal after the first cycle. In fact, the WO_x to bulk intensity ratio remains constant, however the Se–O intensity decreases below the detection limit, and the SeO_x to WSe₂ ratio decreases by ∼44%. Also, the HfO₂ signal intensity after the first cycle of deposition on WSe₂ (figure 3(c)) is smaller in comparison to that obtained for MoSe₂ (figure 3(c)).

Even when the formation of HfO₂ is more energetically favorable than WO_x based on the Gibbs free energy (WO₃ ≈ −195.7 kcal mol⁻¹) [44], WO_x remains on the WSe₂ surface, exhibiting lower reactivity during the ALD 'self-cleaning' reduction reactions in comparison to MoO_x. Various hypotheses can be made to explain the limited HfO₂ growth. First, assuming a ligand exchange mechanism as in the case of MoSe₂, then dissociation of W–O bond would be needed. The dissociation energy of the W–O bond is 720 ± 71 kJ mol⁻¹ [46], which is much higher than the Mo–O dissociation energy 597.2 ± 33.5 kJ mol⁻¹ [46] thus the oxygen loss in WO_x and subsequent formation of a volatile W(N(CH₃)_x) compound requires a higher energy than MoO_x. Secondly, WO₃ and MoO₃ clusters are well known for their catalytic properties toward oxidation, where it was found that Mo⁶⁺ will be more easily reduced to Mo⁵⁺ than W⁶⁺ to W⁵⁺ due to the high stability of WO₃ [47]. This is consistent with our observation in MoO_x and WO_x reactivity after one ALD cycle. The capability of WO_x to remain in the W⁶⁺ oxidation state makes WO_x an effective oxidizing agent of organic molecules [34]. Thus, another possible scenario is the reaction between WO_x and the organic ligand in TDMA-Hf inhibiting ligand exchange reactions. In fact, from the C1s spectra, which overlap with a Se Auger feature, additional carbon species are detected in WSe₂ in comparison to MoSe₂ after the first cycle and subsequent cycles (see supplementary information (SI)).

After five cycles, the WO_x to WSe₂ peak ratio decreases by only 5% along with a WO_x peak shift to a lower BE of ∼0.2 eV, while the SO_x to WSe₂ peak ratio decreases by 31% from its previous value. Interestingly, extra features at identical lower BE positions from the bulk WSe₂ signals, for both W 4f and Se 3d are detected. The appearance of low BE peaks in WSe₂ was previously reported and attributed to a non-van der Waals surface, [48] characterized by edge planes and stepped surfaces. Such a non-homogeneous surface potential can give raise to different fermi level positioning, shifting the core levels in WSe₂ the same amount. Thus, these newly identified peaks can be identified as WSe₂ with a different fermi level in reference to the initial WSe₂. A non-van der Waals surface could be generated from desorption of surface oxides leaving regions in WSe₂ with unsaturated bonds and/or partially etched areas. After 30 ALD cycles, the SeO_x is below the detection limit, and the WO_x to WSe₂ peak intensity ratio decreases to a value of 0.08, with a final BE position ∼0.53 eV lower than the initial WO_x BE, due to lower oxygen content in WO_x [49]. This indicates that lower oxygen content in WO _x improves the rate of ligand exchange reactions between WO_x and TDMA-Hf.

It should be noted that the Hf 4f integrated intensity is significantly different on MoSe₂ and WSe₂ for same number of cycles by comparing figures 2(c) and 3(c), where the final HfO₂ content on MoSe₂ is about three times that detected on WSe₂, suggesting differences in HfO₂ coverage. For such reasons, the surface topography of the HfO₂ films on the UV–O₃ exposed TMDs was also investigated by ex situ AFM. For comparison, HfO₂ was deposited (30 cycles of TDMA-Hf and H₂O) on as-exfoliated non-treated samples, where the dearth of reactive sites on the MoSe₂ and WSe₂ surfaces impeded uniform ALD nucleation, thus HfO₂ is deposited following an island-type growth mechanism, as observed on MoS₂ [14]. In contrast, on the UV–O₃ treated MoSe₂, the HfO₂ film is fully covered and pin-hole free (figure 4(a)). Therefore, the presence of surface oxides formed after UV–O₃ enhanced HfO₂ growth rate in comparison to non-treated MoSe₂, leaving as a result a completely covered HfO₂ film with a RMS roughness of only 0.15 nm. Interestingly, on the UV–O₃ treated WSe₂, HfO₂ islands with a triangular shape are formed with edge length and height in the range of 30–60 nm and 4–5 nm, respectively. It has been shown that triangular hole-like structures can be generated on WSe₂ by oxidative electrochemical etching promoted by AFM [50, 51] and STM [52, 53] tip-sample voltage. In this case, the presence of triangular hole-like structures on WSe₂ could have served as nucleation site for HfO₂, resulting in a quasi-ordered triangular cluster growth. This implies that oxidative etching of WSe₂ could have taken place during UV–O₃ exposure for 6 min and/or ALD self-cleaning reactions. In fact, the last hypothesis correlates with the newly detected WSe₂ XPS peaks (blue features in figure 3(a)) and in turn with the formation of a non-van der Waals surface, as discussed previously.

Zoom In Zoom Out Reset image size

In accordance with the AFM results, cross-section STEM images in figure 5(a) shows a uniform HfO₂ film with thickness of ∼4 nm on UV–O₃ treated MoSe₂, having a minimal variation in thickness across a length of 0.25 μm (see SI). Interestingly, the top-most MoSe₂ surface exhibits darker contrast regions that correspond to selenium-deficient MoSe₂ according to electron energy loss spectroscopy line scans across the HfO₂–MoSe₂ interface (shown in the SI). The ALD self-cleaning reactions occurred right at the MoSe₂ surface are likely to be involved in the generation of such selenium-deficient regions. It is also evident that the MoSe₂ layers below the darker contrast region are continuous and unaltered.

Zoom In Zoom Out Reset image size

Finally, the high magnification STEM image in figure 5(b) gives further insights in regard to HfO₂ on the UV–O₃ treated WSe₂. Partially etched WSe₂ layers are identified at the interface with the HfO₂ islands, which correlates with the XPS and AFM results described above. Here, the contrast from the partially etched WSe₂ layers is weaker at the HfO₂/WSe₂ interface than it is in the bulk WSe₂. This can be explained by considering that an STEM image is averaging cross-sectional information from a finite ∼100 nm thick material. Since the triangular hole-like structures observed in this work have an area smaller than the thickness analyzed in cross-section by STEM, their presence could only be observed in STEM as a weakening of the contrast, while some signal from the un-etched portions of that same layer will always be obtained. Clearly, various mechanisms are involved in the HfO₂ growth on UV–O₃ treated WSe₂. First, the presence of WO_x on the WSe₂ surface and its relative high stability during self-cleaning reactions slows down the HfO₂ growth rate in comparison to MoSe₂ for same number of cycles, generating a partially covered HfO₂ film. Secondly, the presence of triangular hole-like structures on WSe₂ generated by etching caused the preferential HfO₂ nucleation on such structures, forming triangular HfO₂ islands.

In summary, it is found that selective oxygen functionalization of the selenium-based TMDs studied here is limited by highly energetically favorable oxidation of the transition metal. These results highlight the importance that the TMD composition has on the reactivity towards oxidation. The investigation of the ALD of HfO₂ on the TMDs with initial surface oxides elucidates self-cleaning reduction reactions and the desorption of volatile species during the ALD process. According to the interface study, the surface oxides on MoSe₂ are completely removed by the self-cleaning reduction reaction upon HfO₂ deposition. In contrast, the self-cleaning effect is less effective in the oxide removal on WSe₂ mainly due to the relative stability that WO_x exhibits. Thus, the ability of assisting the HfO₂ nucleation is superior for MoO_x than WO_x. Finally, the coverage of HfO₂ on the TMD surfaces was improved by the UV–O₃ pre-treatment.

This work is supported in part by the Center for Low Energy Systems Technology (LEAST), one of six centers supported by the STARnet phase of the Focus Center Research Program (FCRP), a Semiconductor Research Corporation program sponsored by MARCO and DARPA. It is also supported by the SWAN Center, a SRC center sponsored by the Nanoelectronics Research Initiative and NIST. All the DFT calculations are performed using the computational resources of Texas Advanced Computer Center (TACC) at the University of Texas at Austin.