Introduction

Two-dimensional transition metal dichalcogenides (2D TMDC’s), often depicted as MX2 (M = W, Mo, … & X = S, Se, …), have attracted substantial interest due to a wide variety of applications including catalysis, energy storage and microelectronic devices1. For microelectronic applications, 2D TMDC’s are promising alternatives or supplements for today’s silicon-based technology, owing to their inherent ultra-thin structure enabling superior electrostatic gate control and carrier confinement compared to 3D bulk semiconductors2,3. In addition, 2D TMDC’s have a high carrier mobility4, sizeable band-gaps5 and the potential for defect-free interfaces resulting from their smooth, self-passivated surfaces free of dangling bonds6.

However, the lack of dangling bonds does not allow for uniform material growth when done directly on the MX2 surface, which is critical for many microelectronic- and other applications. Nucleates only grow at defect sites or less stable areas such as grain boundaries, which have a low density in highly crystalline films. This results in island-like growth and poor quality of the deposited material7.

Enhancing material growth requires increasing the number of nucleation sites i.e., functionalizing the MX2 surface either via a covalent- or non-covalent approach. In the covalent approach, molecules are grafted on the MX2 surface leading to chemical functionalization for which the density of grafted molecules is often difficult to control and not stable at elevated temperatures8,9. For example, Azcatl et al. greatly enhanced dielectric deposition on multilayer MoS2, MoSe2, and WSe2 surfaces via UV/O3 exposure, generating an oxygen-terminated surface that was ‘self-cleaned’ after atomic layer deposition (ALD) at high temperatures10,11. Follow-up studies by Zhao et al. revealed various electrically active traps at the dielectric/MX2 interface, which could be ameliorated by high temperature anneals12,13. The electrically active traps originated from unintentional doping of the upper layer during UV/O3 functionalization, protecting the underlying layers. Multilayer MX2 films are inherently less susceptible to surface effects compared to monolayer systems, whose properties are drastically changed upon chemical bond formation8. In the non-covalent approach, physisorbed molecules interact via van der Waals (vdW) forces or electrostatic interactions. Since the physisorbed molecules are only weakly bound to the surface, the temperature for subsequent material growth is restricted to avoid desorption. For example, dielectrics grown on physisorbed seeding molecules suffer from a poor density and impurities, resulting from unpurged precursor gases and growth byproducts7,14. Besides single physisorbed molecules, seeding layers have been widely explored owing to their higher thermal stability7,15. Different types of seeding layers e.g., polymer-16, organic-17 and metal oxide-based18, have been demonstrated to result in uniform material growth. For example, a sub nm SiO2 seeding layer enables good dielectric growth, but contamination and poor interface quality give rise to charge trapping at the MX2/seeding layer interface, degrading performances2,15. Therefore, the ideal seeding layer has a self-passivated surface to establish a vdW interface, not degrading the desired MX2 properties.

Another popular member of the 2D family is graphene (Gr), but its utilization as seeding layer simply defers the deposition problem to the next self-passivated surface19. However, great effort has been put into covalently functionalizing graphene as reported by De Feyter and colleagues8,20,21. Using functionalized graphene as a sacrificial seeding layer has been demonstrated in the past22. Nourbakhsh et al. selectively functionalized the upper layer of a bilayer graphene stack using a pulsed oxygen plasma, after which an Al2O3 dielectric was grown via ALD resulting in a high performance, top-gated graphene field-effect transistors (FETs). However, the graphene oxide (GrO) contribution to the total gate capacitance can become significant as it scales with the O/C ratio23. Therefore, a GrO seeding layer can only be utilized for application tolerating its quantum capacitance, for which a large array of logic gates and integrated circuits is still envisioned24, as well as sensor focused applications25.

A seeding layer provides another advantage in forming a physical barrier that limits ambient exposure of the MX2 layer, which is prone to aging effects i.e., spontaneous oxidation upon ambient exposure26,27. An encapsulation-based method is, however, typically considered suboptimal as it complicates further processing or leads to damage upon removal28. However, this can be circumvented by integrating the passivation layer e.g., as a seeding layer for subsequent material growth.

In this study, GrO is used as a sacrificial seeding layer to passivate and enable uniform, high-\(\kappa\) dielectric growth on monolayer WS2. Our approach involves 3 main steps (1) graphene transfer; (2) graphene cleaning i.e., removal of transfer polymeric residues by combining solvent- and hydrogen downstream plasma source (DPS) treatments; (3) graphene functionalization via a dry UV/O3 oxidation process. Subsequently, a HfO2 dielectric is deposited via ALD on the GrO seeding layer. The paper consists of two sections. The first part focuses on the characterization of the different process steps. This is demonstrated for graphene transferred onto SiO2 to simplify sample preparation. The second part applies the learnings of the first to a Gr/WS2 stack, demonstrating proof of concept. A graphene-based nucleation- / passivation layer offers the flexibility to adapt the surface functionalization to the precursor gas chemistry, enabling alternative material growth (other than dielectrics) while maintaining a high-quality interface.

Results and Discussion

Graphene cleaning via solvent- and hydrogen downstream plasma source treatments

Chemical vapor deposition (CVD) is widely used to produce high quality graphene at scale29, but requires graphene to be transferred from the growth- to target substrate. Detailed descriptions of both the growth and transfer processes can be found in the method section. A sacrificial polymer film i.e., poly(methyl methacrylate) (PMMA) is spin-coated directly on graphene to preserve its structural integrity during transfer. Afterwards, a solvent treatment is carried out to remove the bulk polymer material (see methods). Figure 1a–d shows atomic force microscopic (AFM) images of graphene’s surface before and after AFM tip cleaning. Here, the AFM tip acts like a brush that shoves physisorbed molecules towards the scan edges, after which the change in topography can be used to assess layer thickness30. The contamination build-up noted in Fig. 1d corresponds to PMMA transfer residues, which are difficult to remove due to the \(\pi\)-\(\pi\) stacking between the polymer chains and graphene’s conjugated network31,32. To weaken this interaction, polymer degradation via radiative- or thermal treatments has been explored since the higher solubility of the smaller fragments makes them easier to remove33. Conversely, radicals produced in these processes may also chemically react with graphene defects which complicates the cleaning process34.

Fig. 1: PMMA transfer residue cleaning via H2 Downstream Plasma Source (DPS).
figure 1

a, d AFM images after tip cleaning of solvent treated graphene without plasma exposure, where PMMA residues pile up on the scan edge (4 µm2) when brushed away by the AFM tip. b, e No PMMA pile-up is noted for graphene treated with a solvent and 3- or c, f 6 s H2 DPS exposure.

Hydrogen-based downstream plasma treatments have been explored as alternative cleaning methods on account of their scalability31,35,36,37. In this configuration, a baffle separates the sample from the plasma discharge which prevents ion bombardment and only allows diffusion of thermal radicals to the sample surface. The hydrogen radicals induce a scission of the polymer chain, leading to smaller and volatile etch products. Details on the etching mechanism of PMMA by hydrogen radicals are discussed elsewhere38. Figure 1b–e shows tip-cleaned AFM images after solvent treatment and 3 s H2 DPS exposure, revealing no significant contamination build-up at the scan edges. Extending the H2 DPS exposure to 6 s shows a similar contamination build-up (Fig. 1c–f), suggesting the marginal amount of residues still present may be inorganic in nature or originate from post-cleaning ambient exposure. Note some spots are present in all AFM tip-cleaned areas, which correspond to graphene bi- or multilayer islands as discussed in the Supplementary information (Supplementary Fig. 1).

The hydrogen radicals may also interact with the graphene layer itself. Despiau-Pujo et al. predicted that atomic hydrogen species can penetrate, adsorb or be reflected from the graphene surface depending on their energy39. Raman characterization assesses graphene surface modifications resulting from the H2 DPS treatment. Figure 2b, c shows Raman mappings of solvent treated graphene before and after 3 s H2 DPS exposure, respectively. The white mark on the left is part of a reference scratch used for consecutive measurements. The color scale corresponds to an ID/IG ratio, typically used to assess graphene quality. An increase in D (~1350 cm−1) to G peak (~1590 cm−1) ratio suggest a larger defectivity, since the D phonon mode requires a structural defect for its activation40. Some growth-related non-uniformity can be observed in Fig. 2b, which is to be expected given the large mapping scale (20 µm2). A small increase in ID/IG ratio is noted after H2 DPS exposure and additional quantitative analysis is provided in the Supplementary information (Supplementary Fig. 2). The slight ID/IG increase is attributed to limited hydrogenation, since no macroscopic defects are present in the AFM images (Fig. 1) after plasma exposure.

Fig. 2: Dry UV/O3 functionalization of graphene.
figure 2

a Effect of UV/O3 oxidation on the wettability of cleaned graphene, showing an increase in hydrophilicity. b Raman mapping before, c after 3 s H2 DPS treatment and d after 4.5 min UV/O3 exposure, showing a significant increase in ID/IG ratio i.e., graphene functionalization. e Normalized XPS C1s spectra measured before and after 6 min UV/O3 exposure showing an increase in shoulder peak ~286.5 eV i.e., C-O single bond, basal plane functionalities.

Plasma hydrogenation of graphene has been reported, leading to C-H bond formation that changes carbon’s sp2- to a sp3 hybridization (D peak increase) and breaking of the C=C translational symmetry (G peak decrease)41. Even though plasma hydrogenation is known to be reversible e.g., via thermal annealing41, its minor impact is tolerated since graphene’s conjugated network will be sacrificed anyway during subsequent UV/O3 oxidation. Hereafter, a solvent- and 3 s H2 DPS treated layer will be referred to as cleaned graphene.

Graphene functionalization via dry UV/O3 oxidation

The cleaned graphene is subsequently functionalized via dry UV/O3 oxidation, forming the sacrificial seeding layer. This method is known to be reversible upon UV irradiation42, demonstrating the gentle nature of the oxidation process. A soft functionalization is essential since damaging the graphene lattice would expose the underlying WS2 monolayer during further processing. On the one hand, the graphene layer requires sufficient oxidation to generate adequate carbon-oxygen functionalities. On the other hand, the carbon lattice should not be physically damaged due to excessive UV/O3 exposure e.g., UV light induced heating42 to also retain its passivation properties. Establishing this delicate balance requires a thorough characterization of the UV/O3 treatment.

Figure 2a shows the effect of UV/O3 oxidation on graphene’s wettability. The contact angle gradually decreases over 6 min, after which a constant angle of ~20° is reached. The hydrophobicity decrease is explained by the polar carbon-oxygen functionalities formed, for which the reaction mechanism is reported elsewhere43. However, a contact angle of 20° is comparable to the wettability measured for SiO2 substrate after 6 min UV/O3 (Supplementary information, Supplementary Fig. 3). Consequently, the drop in hydrophobicity may be caused by graphene oxidation or SiO2 substrate exposure, suggesting graphene etching.

Raman mappings are recorded before and after 4.5 min UV/O3 to verify the presence of graphene (oxide), shown in Fig. 2c, d, as well as standard (1D) Raman spectroscopy after 7.5 min—see Fig. 5b. Overall, the ID/IG increases significantly after UV/O3 exposure, confirming graphene oxidization. The dismissal of graphene etching is based on the large remaining intensity of both the D and G peaks even after 7.5 min UV/O3 indicating that, although oxidized, graphene remains present on the SiO2 substrate.

Additionally, X-ray photoelectron spectroscopic (XPS) analysis monitors the change in graphene’s surface state i.e., the type of chemical group formation. Figure 2e shows normalized XPS spectra of the C1s core level of graphene before and after 6 min UV/O3, with Table 1 summarizing the peaks deconvolution. No significant spectral changes are noted ~287.5 and ~288.9 eV, associated with double bond carbon-oxygen functionalities (i.e., carbonyl and carboxyl groups, respectively)44,45,46. These double bond functionalities require σ-electrons of the carbon-carbon bonds i.e., the graphene lattice itself to be sacrificed, whereupon it would lose its passivation properties. However, there is an increase in peak intensity ~ 286.5 eV, associated with single bond carbon-oxygen functionalities (i.e., epoxy- and hydroxyl groups), created by sacrificing the \(\pi\)-electrons from graphene’s conjugated network46,47. Exposing graphene to a dry UV/O3 exposure thus predominantly forms carbon-oxygen single-bond i.e., basal plane functionalities.

Table 1 Deconvolution peaks of XPS C1s spectra

Evaluation H2 DPS cleaning and UV/O3 oxidation via back-gated GFET

A graphene field effect transistor (GFET) is exposed to H2 DPS and dry UV/O3 treatments to evaluate their effects on graphene’s transport properties. Details on GFET fabrication are summarized in the method section. Figure 3a illustrates the back-gated, 4-probe GFET, whose dimensions are summarized in Table 2. Additional optical images and Raman spectra are provided in the Supplementary information (Supplementary Fig. 4). Figure 3b, c shows I-V and resistance curves before and after H2 DPS cleaning and 7.5 min UV/O3 oxidation. After transfer, characteristic transport is noted but the Dirac point (Vk) is heavily shifted. The Vk shift towards positive gate voltage results from p-doping caused by the PMMA residues shown in Fig. 1a–d48,49,50,51.

Fig. 3: Evaluation of graphene’s transport properties after H2 DPS cleaning and UV/O3 oxidation.
figure 3

a Illustration 4-probe GFET. b ID-VGS characteristics of backgated GFET before- and after H2 DPS exposure, showing a shift in Vk which results from PMMA residue removal. After 7.5 min UV/O3, graphene’s conductance is significantly suppressed. c Graphene’s sheet resistance as a function of VGS before- and after H2 DPS exposure.

Table 2 Summary of GFET device dimensions and figures of merit

After H2 DPS cleaning, Vk shifts towards 0-point voltage, again demonstrating the effectiveness of the cleaning process. Table 2 summarizes the GFET figures of merit, extracted based on the work of de la ref. 52 and is explained in more detail in the Supplementary information (Supplementary Fig. 5). There is a small loss in current after H2 DPS cleaning (Ion/Ioff : 5.9 \(\to\)5.4), while both the electron- (µelectron) and hole mobilities (µhole) increase albeit asymmetrical (µelectron : 602.7 \(\to\) 1107.1 cm2V−1s−1, µhole : 1890.0 \(\to\) 2020.5 cm2V−1s−1). These 3 observations are explained by considering the following 3 effects: (1) The small loss in current is attributed to limited graphene hydrogenation, as demonstrated in Fig. 2b, c and in line with the results of ref. 53. (2) The µelectron and µhole enhancement is explained by the removal of the PMMA residues, which otherwise act as scattering sites in accordance with ref. 54. (3) Finally, D.B. Farmer et al. observed that graphene p-doping via diazonium functionalization does not drastically change the hole conductance, while the electron conductance was heavily suppressed due to a neutrality point misalignment caused by the channel doping55. Conversely, in this study, H2 DPS cleaning removes the neutral point misalignment caused by the p-dopant i.e., PMMA residues, which results in the asymmetric, larger increase in µelectron compared to µhole.

After 7.5 min UV/O3, graphene’s conductance is suppressed to the nA range. The drop in current results from the sacrifice of graphene’s π-conjugated network to form the basal plane functionalities as evidenced by Fig. 2e. The evolution of graphene’s transport properties substantiates a clean graphene surface post H2 DPS cleaning. Subsequent UV/O3 oxidation functionalizes the basal plane, converting it into a seeding layer.

HfO2 ALD on GrO/SiO2

A HfO2 ALD process is applied to the UV/O3 functionalized graphene to test the seeding layer growth characteristics. Figure 4 shows the Hf areal density measured by Rutherford backscattering spectrometry (RBS) after ALD on graphene layers functionalized with various UV/O3 exposure times. 1 min UV/O3 leads to a non-linear growth, indicating a growth delay. Extending the UV/O3 exposure to 6 min results in linear growth and a significant increase in overall Hf deposition, demonstrating the enhanced surface functionalization. The growth enhancement is also confirmed by scanning electron microscopic images (Supplementary information, Supplementary Fig. 6), showing graphene exposed to 1 min UV/O3 is only decorated by HfO2 island after 124 ALD cycles, whereas 6 min UV/O3 results in more uniform growth and layer closure.

Fig. 4: HfO2 ALD on graphene oxide seeding layers.
figure 4

Hf areal density measured by RBS showing a similar growth on 6 min UV/O3 functionalized graphene when benchmarked against a SiO2 i.e., well-functionalized substrate.

In addition, the HfO2 growth is benchmarked against SiO2, representing a model substrate with a well-functionalized surface. The Hf density measured on SiO2 is comparable to 6 min UV/O3 exposed graphene, demonstrating that good dielectric growth can be achieved after adequate UV/O3 functionalization.

Fabrication and characterization of 2D Gr/WS2 heterostack

Detailed descriptions of the monolayer WS2/graphene growth and transfer processes can be found in the method section. Extensive characterization of the Gr/WS2 heterostack is presented in the Supplementary information which includes AFM tip cleaning and Raman spectroscopy, demonstrating a clean, high-quality graphene layer (Supplementary Fig. 7). In addition, the high-quality Gr/WS2 vdW interface is evidenced by Raman phonon and Photoluminescence (PL) quenching resulting from the interlayer coupling (Supplementary Fig. 7). Lastly, the graphene and WS2 monolayer thicknesses are confirmed via Raman- and PL characterization (Supplementary Fig. 8).

Substrate dependency of selective graphene oxidation towards underlying WS2 monolayer

The Gr/WS2 heterostack is exposed to a dry UV/O3 treatment to functionalize graphene’s basal plane as demonstrated in the first part of the paper. Figure 5 shows Raman spectra of UV/O3 exposed graphene transferred on (a) monolayer WS2 and (b) 90 nm SiO2. For Gr/SiO2, the D peak significantly increases after 7.5 min UV/O3 and longer exposure leads to Raman quenching. For Gr/WS2, a similar D peak increase requires up to 30 min of UV/O3 exposure. The rate of graphene oxidation appears to be substrate dependent and is a peculiar phenomenon that impacts the process flow. Authors theorize it results from UV light-induced charge transfer between the graphene and WS2 monolayer. This hypothesis is based on the interplay of the following two phenomena.

Fig. 5: Graphene dry UV/O3 oxidation substrate dependency.
figure 5

Stacked Raman spectra of UV/O3 exposed graphene a on WS2 showing significant modifications after 30 min, whereas b when transferred on SiO2 only 7.5 min suffice. c Authors explain this substrate dependency on account of UV-light induced, ultrafast charge transfer between the graphene and WS2 monolayer, lowering graphene’s electron density and resulting in a slower UV/O3 oxidation rate compared to a SiO2 substrate.

First, the reaction mechanism during UV/O3 oxidation is discussed. Atomic oxygen is generated by exposing oxygen gas to UV light (\(\lambda\) = 184.9 nm), resulting in ozone formation. The ozone molecule dissociates (\(\lambda\) = 253.7 nm) into an oxygen molecule and atomic oxygen. The atomic oxygen diffuses to the graphene layer where it adsorbs in a ‘bridge’ position i.e., in between two adjacent carbon atoms due to the favorable hybridization of the 2p orbitals of the carbon and oxygen atoms56. In other words, the oxygen atom chemisorbs onto graphene by consuming π-electrons of two carbon atoms to satisfy its octet configuration, forming an epoxide group42. Importantly, this reaction does not consume any electrons making up the carbon σ-bonds and thus leaves the graphene lattice intact. Consequently, a dry UV/O3 exposure results predominantly in epoxy functionalization of graphene’s basal plane, both theoretically56,57,58 and experimentally verified by the XPS results shown in Fig. 2e.

Second, the Gr/WS2 band structure is considered and how it changes upon UV light exposure. In the ground state, the Gr/WS2 band structure is simply a superposition of the individual bands due to their weak vdW interactions, without substantial charge redistribution59,60. However, ultrafast charge transfer can occur upon excitation e.g., via electron pump or UV light exposure. This charge transfer is demonstrated by the PL quenching of the WS2 monolayer before and after graphene capping, enabled by the high quality vdW interface (Supplementary information, Supplementary Fig. 7). During charge transfer from the WS2 monolayer to graphene, holes in the WS2 valence band maximum (VBM) tunnel faster than electrons in the conduction band minimum (CBM) due to a stronger coupling of the VBM with graphene’s acceptor states59. When excited continuously, the difference in electron and hole diffusion rates leads to a positive charge build-up in the graphene layer59.

Finally, the graphene oxidation substrate dependency is summarized (Fig. 5). UV light generates atomic oxygen, but also excites carriers in the WS2 monolayer. Constant UV exposure continuously generates electron-hole pairs where WS2, holes tunnel faster than WS2, electrons, leading to a positive charge build-up in graphene that lowers the effective electron density. Consequently, its π-electrons are less prone to bind atomic oxygen i.e., slows down the oxidation process. As a result, graphene oxidation requires longer UV/O3 exposure when transferred on WS2 compared to SiO2, where no interlayer charge transfer takes place. XPS characterization of the C1s core level of UV/O3 exposed Gr/WS2 is carried out to verify the substrate dependency (Supplementary information, Supplementary Fig. 10). However, no significant changes in peaks contributions are noted, possibly due to interference from the extra signal originating from WS2 monolayer, lowering the overall measurement sensitivity.

The dry UV/O3 oxidation functionalizes graphene but should not affect the underlying WS2 monolayer. PL spectroscopy is widely used to monitor changes in MX2 monolayers on account of its extreme sensitivity61, but cannot be utilized in this case due to interlayer quenching as explained in the Supplementary information (Supplementary Fig. 7). Raman spectroscopy is another optical technique used to detect WS2 modifications62, but which is not prone to interlayer quenching. Figure 6a shows the Raman response of WS2 monolayer after different UV/O3 exposures with- (a) and without graphene capping (b). Graphene capped WS2 shows no loss in Raman signal after graphene cleaning nor over 30 min UV/O3 exposure, while uncapped WS2 shows significant Raman quenching after 7.5 min UV/O3. The loss in Raman response suggests UV/O3 induced layer damage63, which is prevented after graphene capping. XPS is carried out to monitor changes in Wmetal to Woxide chemical states resulting from the UV/O3 oxidative treatment. Figure 6c shows the W4f core level of graphene capped- and uncapped WS2 treated for various UV/O3 exposure times. The deconvoluted spectra are included in the Supplementary information (Supplementary Figs. 11 and 12) and a detailed description of the deconvolution method is provided in the method section. The stacked plot shown in Fig. 6c summarizes the Wmetal and Woxide peak contributions, which show no significant changes after graphene capping, H2 DPS cleaning and 7.5 min UV/O3 oxidation compared to the as grown WS2 monolayer. Over 15–30 min UV/O3, a gradual, small increase in Woxide contribution is observed, in contrast to uncapped WS2 for which a significant increase in Woxide contribution is noted after 7.5 min. In comparison, Woxide/Wtotal after 30 min UV/O3 is 25.2% / 80.7% for graphene capped/ -uncapped WS2. These results confirm the loss in Raman signal in Fig. 6b is caused by WS2 oxidation, which is prevented after graphene capping that maintains its passivation properties throughout the oxidation process. The limited Woxide increase is attributed to limited imperfections in the graphene layer, anticipated from the large-scale (2.5 × 2.5 cm2) growth and transfer processes (methods). These imperfections e.g., point-defects and grain boundaries readily oxidize compared to the basal plane and do not passivate during extended UV/O3 exposure.

Fig. 6: Passivation properties of the graphene oxide seeding layer.
figure 6

Raman spectra of UV/O3 exposed WS2 (normalized w.r.t Si 520 cm−1) a with graphene capping showing no loss in signal over 30 min UV/O3, whereas b uncapped WS2 shows signal quenching after 7.5 min. c Normalized XPS W4f spectra, showing graphene passivated WS2 exhibits limited increase in Woxide peak contributions i.e., layer damage compared to uncapped WS2.

In addition, uncapped WS2 exposed to ambient i.e., 24 h per day light exposure in a controlled environment with 40.2% relative humidity & 19.1% O2 gas concentration, shows a similar increase in Woxide/Wtotal of 81.8% compared to uncapped WS2 treated with 7.5 min UV/O3 (Supplementary information, Supplementary Figs. 13 and 14). Therefore, it is reasonable to assume that graphene capping not only protects the underlying WS2 monolayer in a strong oxidative environment e.g., dry UV/O3 exposure, but also mitigates spontaneous ambient oxidation i.e., aging effects26,27.

HfO2 ALD study on GrO/WS2

A HfO2 ALD study is carried out on GrO/WS2 to evaluate the GrO seeding layer’s growth characteristics. Figure 7a shows the Hf areal density measured by RBS after ALD on uncapped- and graphene capped WS2 exposed to 30 min UV/O3. A substantially enhanced Hf deposition is noted on GrO/WS2 compared to direct deposition on WS2, for which there is a strong nucleation delay. Benchmarking the HfO2 growth on GrO/WS2 against SiO2 reveals similar trends, again demonstrating the GrO surface is well-functionalized and results in good material deposition.

Fig. 7: HfO2 ALD on GrO/WS2.
figure 7

a Hf areal density measured by RBS showing a similar growth for 30 min UV/O3 functionalized GrO/WS2 compared to SiO2 i.e., well-functionalized substrate. b HfO2 layer roughness extracted from AFM images revealing a growth delay for uncapped WS2, whereas the lower and constant HfO2 layer roughness on GrO/WS2 demonstrates uniform growth and fast layer closure.

The HfO2 layer roughness is measured via AFM to examine surface coverage and layer closure throughout the ALD process (Supplementary information, Supplementary Fig. 15). Figure 7b plots the HfO2 layer roughness after ALD on GrO capped- and uncapped WS2. Direct growth on WS2 shows a gradual increase in layer roughness which stabilizes only after 248 cycles ~5–6 nm, while a constant roughness of ~1–1.5 nm is measured on GrO/WS2. The large increase in roughness together with nucleation delay demonstrate the poor, island-like growth when done directly on the WS2 monolayer, similar to the trends observed in Fig. 4. The low, constant roughness of the HfO2 layer grown on GrO/WS2 demonstrates uniform material deposition and fast layer closure even in the early stages of the ALD process. Additionally, scanning transmission electron microscopic images are provided in the Supplementary information (Supplementary Fig. 16) that show a smoother HfO2 layer on GrO/WS2 compared to direct growth on WS2, confirming the GrO seeding layer leads to enhanced material deposition.

In summary, we present a flexible approach to facilitate the nucleation and HfO2 growth on monolayer WS2. Our method utilizes a graphene oxide seeding layer, serving a dual purpose (1) as passivation layer, allowing for a perfect van der Waals interface and (2) as seeding layer, for which the surface chemistry can be tailored to enable other types of material growth.

After transfer, a two-step cleaning process (solvent + H2 downstream plasma treatments) removes polymeric transfer residues from the graphene surface. The large-scale hydrogen plasma cleaning is highly selective towards graphene, verified by GFET analysis. The cleaned graphene is then functionalized selectively towards the underlying WS2 via dry UV/O3 oxidation, during which basal plane functionalities are formed that leave the carbon lattice and its passivation properties intact. Interestingly, the oxidation rate is lower when graphene is transferred on WS2 compared to SiO2. This substrate dependency possibly results from ultrafast charge transfer via the interlayer coupling between graphene and the WS2 monolayer. Graphene capping passivates the underlying WS2 during oxidative UV/O3 exposure and, therefore, is expected to also protect against aging effects.

The carbon-oxygen functionalities serve as nucleation sites in a subsequent ALD process, significantly enhancing HfO2 growth compared to direct deposition on WS2 and a similar deposition when benchmarked against a well-functionalized surface in SiO2. Moreover, the constant HfO2 roughness measured throughout the ALD process indicates uniform deposition and fast layer closure. Our methodology is expected to be generally applicable to different 2D TMDC flavors since the different treatments are applied to the sacrificial graphene (oxide) layer and not directly to the MX2 film. Ultimately, a graphene-based seeding layer offers adaptability i.e., tailored functionalization to enable alternative material growth, while preserving a van der Waals interface to prevent degradation of the 2D TMDC properties.

Methods

WS2 growth

The WS2 monolayer was grown via metal organic chemical vapor deposition on 300 mm silicon wafers covered with a 90 nm thermal oxide. At first, the wafer was heated under high-purity N2/H2 until a chamber temperature of 750 °C was reached. The growth started with a nucleation step, where W(CO)6 (partial pressure ~9 × 10−3 Torr) and H2S gases were introduced (precursor partial pressures ratio H2S/W(CO)6 = 5 × 103) for 10 min at 750 °C at a chamber pressure of 150 Torr. Next, the chamber temperature was increased to 900 °C and the chamber pressure lowered to 20 Torr to enhance lateral growth and promote layer closure.

Graphene growth and transfer

Two types of graphene were used in this study: commercially available graphene produced by Graphenea and graphene produced at imec, which were used for the first- and second part of the paper respectively.

Graphene growth at Graphenea was performed in a Black Magic Pro 2 × 8” CVD furnace from Aixtron Ltd on a copper foil. A H2 pre-annealing was done at 900 °C, after which the temperature was raised to 1000 °C and methane gas introduced in the chamber until the chamber pressure reached 25 mbar. After 30 min, the methane was evacuated, the growth chamber purged and cooled down to room temperature in Ar. Next, the malleable Gr/Cu foil catalyst was covered with a PMMA coating and then laminated with an adhesive polymer carrier to provide a more rigid support during transfer. The catalyst was etched via a wet etching process performed in borosilicate glass tanks using FeCl3 as an etchant. Several consecutive ultra-pure deionized H2O and acidic rinses of diluted solutions of hydrochloric acid (37%) were used to wash the FeCl3 etchant and minimize the metal traces concentration, after which the graphene was dried via a N2 flow. The graphene layer was then laminated via a dry method onto the SiO2 target wafer. Namely, the adhesive carrier/PMMA/graphene stack was roll-laminated at a pressure above 1 bar at a temperature of 150 °C, after which the carrier polymer was removed leaving PMMA/graphene layers behind on the SiO2 wafer. At imec, the 300 mm wafers were manually cleaved into coupons for further processing.

At imec, graphene was grown via CVD in a Black Magic BM6 from Aixtron Ltd on a 2.5 × 2.5 cm2 Platinum foil64, which was cleaned by acetone-, short oxygen plasma- and piranha solvent treatments. The growth chamber was gradually heated up to 1170 °C in a H2-rich environment at a chamber pressure of 750 mbar, after which a waiting time of 6 min was set to ensure the Pt foil was properly annealed. Next, a H2/CH4 gas mixture (800/4 sccm) was introduced into the chamber for 40 min. Subsequently, the chamber was cooled down and when the chamber temperature dropped below 700 °C, the CH4 precursor was switched off. The H2 supply was turned off only when the temperature dropped below 150 °C. The rich-H2 environment during the cooling stage aims to etch multilayer nucleation sites resulting from carbon supersaturation of the Pt foil. Next, a PMMA layer was spin coated on the graphene to protect and maintain structural integrity during transfer. The PMMA/Gr/Pt stack was then put in a deionized water bath at 80 °C for 18 h to promote water intercalation in between the graphene and Pt foil. Subsequently, a frame was glued onto the PMMA/Gr/Pt stack to assist in handling the graphene after delamination and facilitating alignment with the target substrate. Afterwards, a wet electrochemical method was used to transfer graphene to the monolayer WS2 or 90 nm SiO2 substrate, which is described in detail elsewhere65. Essentially, graphene is delaminated from the Pt growth foil by the intercalation of cations which do not reduce within the electrochemical window of TEAH solvent. The ion intercalation was identified as the critical component for delamination rather than the hydrogen bubble formation resulting from the redox reaction of water at the Pt interface. Then, the graphene layer was rinsed and put to float in a deionized water bath and laminated onto the target substrate by lowering the water level, with help of a specially designed substrate holder and the frame glued on the PMMA/Gr stack. After lamination, the sample holder was placed on a hotplate at 80 °C under an angle of 45° to gently remove the water trapped in between the graphene and target substrate. After several hours, the stack was cut out of the frame and placed into a vacuum chamber on a hotplate at 50 °C overnight to further remove any residual water. In addition, XPS measurements show no Pt residues after graphene transfer that would otherwise degrade the interface quality (see Supplementary information, Supplementary Fig. 9).

Solvent treatment

The solvent treatment consisted of an overnight (±12 h) acetone dip, followed by a short (5 min) isopropanol dip to remove acetone residues. Both solvent treatments were carried out at room temperature.

H2 Downstream Plasma Source treatment

A Lam Research downstream plasma etch system (2300 Microwave stripper) was used to carry out the H2 dry cleaning treatments. Graphene coupons were placed in 300 mm silicon pocket wafers for processing. The etching process consisted of a 60 s heat-up step at 300 °C, followed by a low-power H2 DPS exposure for which a highly diluted plasma chemistry was used i.e., He/H2 5000/50 sccm ratio at 300 °C, 1 Torr and 500 W plasma power37. A pre-conditioning was performed prior to the graphene dry cleaning, to clean out the chamber and avoid carry-over-effects from preceding plasma processes. This pre-conditioning used He/H2 chemistry in a 3000/900 sccm ratio at 300 °C,1 Torr and 1000 W plasma power. The cleanliness of the graphene surface was then verified by tip cleaning AFM, explained further on.

Graphene UV/O3 oxidation

A UVO-Cleaner Model 144AX (Jelight) was used to carry out the UV/O3 treatment with an oxygen gas flow of 300 L h−1, without additional sample heating, the reaction chamber at room temperature and atmospheric pressure, and a mercury lamp as UV source (Power settings: 100–120 V, 60 Hz), located above the sample surface.

HfO2 atomic layer deposition

HfO2 ALD was carried out in a 300 mm ASM Eagle XP4 tool (cross flow ALD reactor). The ALD precursors used were hafnium (IV) chloride and water (pulsed first) at a temperature of 300 °C.

Raman and photoluminescence spectroscopy

A Horiba Jobin-Yvon HR800 Raman tool was used for the Raman measurement on graphene and WS2 with a 532 nm laser at 3 mW laser power (or 50% ND filter), 1000 µm confocal hole, 1800 grooves mm−1 grating, 100 × 0.9 NA Olympus objective, and a double exposure of 2 (accumulations) x 30 s (integration time) with spike filter to eliminate cosmic spikes. The Raman mappings consisted of a 20 × 20 \(\mu\) m scan for which a 1 \(\mu\) m step-size was used, shown in Fig. 2.

Note that direct peak comparison between the two sets of Raman spectra of the Gr/WS2 and Gr/SiO2 stacks (Fig. 5) is not possible since a significantly lower laser intensity was used for the Gr/WS2 stack. This is because the high laser intensity, typically used for graphene, can induce WS2 oxidation in the presence of intercalated water originating from ambient or the wet transfer process used in this work66. Nevertheless, the claims made regarding the graphene layer remain valid since no similar quenching is observed for graphene on WS2 compared to graphene on SiO2 after 30 min of UV/O3 exposure, as shown in Fig. 5.

A Horiba Jobin–Yvon HR800 Raman tool was used for the PL measurements on WS2 with a 532 nm laser at 0.05 mW laser power (or 1% ND filter), 1000 µm confocal hole, 600 grooves mm−1 grating, 100 × 0.9 NA Olympus objective, and a double exposure of 1 (accumulations) x 15 s (integration time) with spike filter to eliminate cosmic spikes.

Water contact angle

An OCA contact angle system (Dataphysics) was used with a standard CCD camera (768×576 25 fps intercalated), Hamilton 500 µL syringe and needle thickness of 0.52 mm. A water droplet was dispensed onto the sample surface by manually lowering and lifting the syringe set-up. Sufficient time (~30 s) was taken for the droplet to reach equilibrium wetting, after which a snapshot was made and used to calculate the contact angle of the water droplet and sample surface (Static WCA method). The CA measurements were performed within 5 min of UV/O3 exposure to minimize the effect of ambient contamination.

X-ray photoelectron spectroscopy

The measurements were carried out in Angle Integrated mode using a QUANTES instrument from Physical electronics. The measurements were performed using a monochromatized photon beam of 1486.6 eV, a 100 microns spot size with charge neutralization.

CasaXPS software was used to analyse and deconvolute the C1s and W4f spectra. In case of the W4f core level, the spectra were deconvoluted by fitting doublets corresponding to W(0) i.e., WS2 at ±32 & 34 eV, W( + IV) i.e., WO2 at ±34 & 36 eV and W( + VI) i.e., WO3 at ±35.3 & 37.846, while respecting an area ratio of ¾ resulting from spin-orbit coupling67. An extra peak was added at high binding energy corresponding to the W5p3/2 subshell46. A Lorentzian asymmetric line shape was used to extract the different peak contributions and normalized to account for any acquisition-related intensity variations. The deconvoluted spectra together with the peak numerical values are provided in the Supplementary g information (Supplementary Figs. 11, 12, 14).

Rutherford backscattering spectroscopy

A 6SDH tandem accelerator (National Electrostatics Corporation) was used where the accelerated ions were magnetically examined to eliminate contaminants and determine the ion energy. Afterwards, the ion beam was collimated to 1 mm in size and introduced to the scattering chamber, where the sample was mounted on a manipulator. The experimental settings included a particle energy of 1.523 MeV, He+ beam, scattering angle of 170°, sample tilt angle of 11°, a beam current of 22 nA and detector calibration with 68 KeV offset, 1.4925 KeV gain and a FWHM 0.1626E-01 MeV.

The error bars correspond to the absolute error, calculated via Eq. (1):

$$\sqrt{\left(\frac{{Sum}}{4}\right)^{2}+({Sum}* 0.02)^{2}}$$
(1)

Where Sum corresponds to the fitting value of the element in question i.e., the amount of the element in atoms cm-2.

Since the RBS signals of Hf and W overlap, a reference WS2 sample was measured to estimate the W signal contribution and was later subtracted in case of all samples that contained both Hf and W in their stack. This subtraction was also included in the error calculation.

(Tip cleaning) Atomic force microscopy

Tip cleaning AFM was carried out using a Nanoscope V AFM tool (Bruker) in lateral force microscopy (LFM) mode under an angle of 90°. A HQ-NSC19/AIBs cantilever was used with a spring constant of ~0.5 N m−1. Due to the low spring constant, scans were carried out with very small loading forces (<1 nN) that ensured the underlying graphene surface did not get damaged during the mechanical scratching process. Subsequent scans were performed in peak force QNM mode to determine the topography and contaminant thickness. Since this an oscillating type of mode, no lateral forces were exerted during the scans that could displace the residues over the graphene surface.

Scanning electron microscopy

Top-down scanning electron microscopic images were recorded with a FEIS VeriosTM apparatus using a 3 kV beam energy, 0.10 nA beam current, 5.0 µs dwell time, a through-the-lens detection system of secondary electrons with a horizontal field of view and a working distance of 2.6 mm.

Transmission electron microscopy

Cross transmission electron microscopic images were recorded via the Titan3 G2 60-300 (FEI) with a FEG electron source, operated at 200 kV. A spin-on-carbon (SOC) cap layer was deposited to protect the surface during the focused ion beam sample preparation.

Graphene Field Effect Transistor fabrication and measurement

For GFET fabrication, a graphene layer was grown, transferred on 90 nm SiO2/Si, and cleaned as described earlier. Next, graphene patterning was done via photolithography i.e., a positive resist (IX845) was spin coated and baked for 1 min at 120 °C before optical exposure. After, the resist/graphene stack was developed in OPD 5262 and etched using an oxygen plasma. After graphene patterning, the bulk resist was removed by the solvent treatment described earlier. Source and drain were deposited by lift-off using the same photolithography process and resist mentioned earlier. The metal contacts consist of 2 nm Ti / 50 nm Pd. After lift-off, the devices were again solvent treated and additionally cleaned by the H2 DPS plasma protocol described earlier.

For device measurements, a Keithley 4200 parameter analyser was used with a Cascade probe station in ambient without light exposure during measurement. The gate voltage was swept by changing the potential applied to the chuck/sample holder i.e., a back gated set-up.