Direct atomic layer deposition of ultrathin aluminium oxide on monolayer $MoS_2$ exfoliated on gold: the role of the substrate

In this paper we demonstrated the thermal Atomic Layer Deposition (ALD) growth at 250 {\deg}C of highly homogeneous and ultra-thin ($\approx$ 3.6 nm) $Al_2O_3$ films with excellent insulating properties directly onto a monolayer (1L) $MoS_2$ membrane exfoliated on gold. Differently than in the case of 1L $MoS_2$ supported by a common insulating substrate ($Al_2O_3/Si$), a better nucleation process of the high-k film was observed on the 1L $MoS_2/Au$ system since the ALD early stages. Atomic force microscopy analyses showed a $\approx 50\%$ $Al_2O_3$ surface coverage just after 10 ALD cycles, its increasing up to $>90\%$ (after 40 cycles), and an uniform $\approx$ 3.6 nm film, after 80 cycles. The coverage percentage was found to be significantly reduced in the case of 2L $MoS_2/Au$, indicating a crucial role of the interfacial interaction between the aluminum precursor and $MoS_2/Au$ surface. Finally, Raman spectroscopy and PL analyses provided an insight about the role played by the tensile strain and p-type doping of 1L $MoS_2$ induced by the gold substrate on the enhanced high-k nucleation of $Al_2O_3$ thin films. The presently shown high quality ALD growth of high-k $Al_2O_3$ dielectrics on large area 1L $MoS_2$ induced by the Au underlayer can be considered of wide interest for potential device applications based on this material system.

plasma pre-treatments of the MoS2 surface have been used to create hydroxyl groups for conventional thermal ALD at 200 -250 °C, resulting in the deposition of uniform Al2O3 films with thickness down to 1.5 nm [21]. In spite of these beneficial effects, the damage and chemical modifications introduced by these plasma pre-treatments can affect the electronic transport in MoS2 devices. Besides thermal ALD, plasma-enhanced ALD (PEALD) processes have been also recently investigated to grow very thin films (<5 nm) of Al2O3 and HfO2 on MoS2 samples with different layer numbers [22,23]. In particular, electrical characterization of 1L, 2L, and 3L MoS2 back-gated transistors before and after HfO2 PEALD revealed the occurrence of plasma damage, resulting in significant degradation of the electronic properties especially for 1L MoS2 [23].
In addition to these processes involving a chemical modification of MoS2 surface, non-covalent functionalization with thin organic (e.g., perylene derivatives) [24] or inorganic (e.g., SiO2 nanoparticles) [25] seeding layers has been also explored to promote the thermal ALD growth of thin Al2O3 films on MoS2. However, the use of these interlayers ultimately limits the minimum thickness of the dielectric and may affect the electrical quality of the interface.
This short overview about ALD of high-k dielectrics on TMDs indicates that the seeding layers and pre-functionalization approaches explored so far presents some disadvantages, while direct thermal ALD of ultra-thin films would be highly desirable. In this respect, the interaction of atomically thin MoS2 layers with the underlying substrate is expected to play an important role in the ALD nucleation stage, similarly to what observed for monolayer graphene residing on some specific substrates [26,27]. As an example, Dlubak et al. [26] reported an enhanced Al2O3 nucleation on CVD-grown 1L graphene residing on the native metal substrates (Cu, Ni), that was ascribed to an improved ALDprecursor adsorption due the electrostatic effect of polar traps located at graphene/metal interface [28,29]. More recently, the uniform growth of ultra-thin (2.4 nm) Al2O3 films by direct thermal ALD (at 250 °C) on monolayer epitaxial graphene on 4H-SiC(0001) has been ascribed to the beneficial effect of the carbon buffer layer at the interface with the substrate [30]. To the best of our knowledge, analogous substrate effects on the ALD nucleation onto 1L TMDs have not been reported so far.
In this paper, we investigated the ALD growth of ultra-thin (<4 nm) Al2O3 films on 1L MoS2 produced by gold assisted mechanical exfoliation from bulk crystals [31,32,33,34]. This method exploits the strong Au-S interaction to exfoliate large area (cm 2 ) MoS2 membranes, predominantly formed by monolayers, on a gold substrate. These high crystalline quality membranes can be subsequently transferred on insulating substrates [33,34].
Using identical ALD conditions on 1L MoS2 membranes supported by gold (MoS2/Au) or by Al2O3 (100 nm)/Si substrate (MoS2/Al2O3/Si), the typical inhomogeneous coverage by Al2O3 islands was observed in the case of the 1L MoS2/Al2O3/Si system, whereas the formation of a highly uniform Al2O3 insulating film (3.6 nm thick) was observed on the 1L MoS2/Au sample. This excellent uniformity is the result of an enhanced ALD nucleation on MoS2 surface due to the interaction with Au substrate, giving rise to 50% Al2O3 surface coverage after only 10 ALD cycles, and >90% coverage after 40 cycles. In this respect, micro-Raman and micro-photoluminescence spectroscopy analyses of 1L MoS2/Au and 1L MoS2/Al2O3/Si samples before and after ALD processes provided an insight on the role of the substrate-related doping and strain in the Al2O3 nucleation and growth.

Experimental
The gold substrate used for MoS2 mechanical exfoliation was prepared by sequentially depositing a 10 nm Ni adhesion layer and a 15 nm Au film with DC magnetron sputtering on top of a SiO2/Si sample. This process resulted in a very smooth Au surface with RMS roughness (<0.2 nm), suitable for the MoS2 exfoliation procedure [34]. This latter was performed by pressing a bulk molybdenite stamp on the surface of a freshly prepared Au/Ni/SiO2 sample, in order to avoid the adsorption of contaminants (e.g. adventitious carbon) on Au surface, which would reduce the 1L MoS2 exfoliation yield [35]. This exfoliaton results in a large area mostly composed 1L MoS2 with some 2L regions identified by optical contrast, AFM and Raman spectroscopy.
The Al2O3/Si substrate used for transferring the Au-exfoliated MoS2 was prepared by DC-pulsed RF reactive sputtering of 100 nm Al2O3 on a Si wafer. The transfer procedure of the large areas MoS2 membranes from gold to this insulating substrate is discussed in details in Ref. [34].
Thermal ALD of Al2O3 thin films on MoS2 was carried out in a PE-ALD LL SENTECH Instruments GmbH reactor, using TMA and H2O as the aluminum precursor and co-reactant, respectively. All depositions were carried out at a temperature of 250 °C and with a pressure of 10 Pa. Initially, a process consisting of 80 ALD cycles was simultaneously carried out on both 1L MoS2/Au and 1L MoS2/Al2O3/Si systems, to compare the Al2O3 coverage uniformity. After observing the beneficial effect of the Au substrate on the uniformity of the Al2O3 growth on 1L MoS2, the nucleation and growth mechanisms on the 1L MoS2/Au were investigated in a more detail, by performing shorter ALD processes (10 and 40 deposition cycles).
The surface roughness, coverage fraction and thickness of the deposited Al2O3 on MoS2 were evaluated by tapping mode Atomic Force Microscopy (AFM), morphology and phase, using a DI3100 equipment by Bruker with Nanoscope V electronics. Sharp silicon tips with a curvature radius of 5 nm were used for these measurements. Furthermore, the electrical insulating properties of the very thin Al2O3 films deposited on 1L MoS2/Au were evaluated by conductive AFM (C-AFM) analyses [36] using the TUNA module and Pt-coated silicon tips.
Micro-Raman spectroscopy and micro-photoluminescence (PL) measurements of MoS2 on the different substrates and before/after the ALD growth of Al2O3 were carried out using an Horiba HR-Evolution micro-Raman system with a confocal microscope (100× objective) and a laser excitation wavelength of 532 nm. Figure 1 shows the comparison between the AFM surface morphologies of Al2O3 simultaneously deposited at 250 °C by 80 ALD cycles on the surface of the 1L MoS2/Al2O3/Si sample (a) and of the 1L MoS2/Au sample (b), respectively. A very inhomogeneous Al2O3 coverage, resulting in a root mean square roughness RMS=2.5 nm, is observed on the surface of 1L MoS2 transferred on the Al2O3/Si substrate. Furthermore, the height distribution in Fig.1 (c) shows two components, related to bare and Al2O3 covered MoS2 areas, due to 70% coverage and an average Al2O3 islands height of 4 nm. This scenario, schematically depicted in the inset of Fig.1(c), is consistent with the commonly reported island growth during direct thermal ALD on MoS2 surface. Differently, for 1L

Results and discussion
MoS2 supported by the Au substrate ( Fig.1(b)), a pinhole-free Al2O3 layer with a very flat morphology is observed after 80 ALD cycles. The deposited film exhibits a very narrow height distribution   After demonstrating the formation of a homogeneous 3.6 nm Al2O3 insulating film on top of 1L MoS2/Au by 80 ALD cycles at 250 °C, we investigated the film nucleation and growth stages by AFM analyses performed after a reduced number of ALD cycles at the same temperature. Fig.3(a) and (d) show two tapping mode morphological images acquired on 1L MoS2/Au samples after 10 and 40 ALD cycles, respectively. In particular, after 10 cycles, a very irregular and ultrathin coating can be deduced from the morphological image, resulting in a RMS0.4 nm, slightly higher than the 0.2 nm value measured on the bare 1L MoS2/Au sample. On the other hand, a grain-shaped morphology of the deposited Al2O3 film can be clearly observed after 40 ALD cycles (see Fig.3(d)), suggesting the occurrence of 3D growth of Al2O3 islands on top of the inhomogeneous nucleation layer formed at lower number of cycles. A quantification of the coverage percentage is very difficult from these morphological images. On the other hand, the Al2O3 coated and uncoated 1L MoS2 areas can be clearly distinguished in the corresponding AFM phase maps ( Fig.3(b) and (e)), as the phase signal is known to be very sensitive to the surface properties of materials. In particular, the red and black contrast in these two images correspond to the Al2O3-covered and uncovered 1L MoS2, respectively. Furthermore, the histograms of the phase distribution extracted from the two maps are reported in Fig.3(c) and (f), from which an Al2O3 coverage percentage of 50% and 93% were evaluated after 10 and 40 ALD cycles, respectively. This very high coverage after only 40 ALD cycles, corresponding to 1.4 nm Al2O3 thickness (measured by an AFM step-height analysis as in Fig.2(b)), demonstrates a very good nucleation on Au supported 1L MoS2. enhanced ALD nucleation on 1L-MoS2. On the other hand, in the case of 2L MoS2 (Fig.4(c), right image), the Au-S interaction is expected to be partially screened by the presence of the first MoS2 layer, resulting in a less-efficient ALD growth. A similar degradation of Al2O3 coverage homogeneity has been also observed in the case of bilayer or few-layers CVD graphene on the native copper substrate [26] and of bilayer epitaxial graphene on 4H-SiC(0001) [27]. Then, micro-Raman spectroscopy analyses have been performed to investigate the strain and doping status of 1L MoS2 residing on Au and Al2O3 substrates before and after the ALD growth. Fig.5(a) reports two representative Raman spectra for as-exfoliated 1L MoS2 on Au (reference) and after 80 TMA/H2O ALD cycles, resulting in the homogeneous 3.6 nm Al2O3 film deposition shown in Fig.   2(b). The corresponding Raman spectra for 1L MoS2 transferred onto the Al2O3/Si substrate (reference) and after the 80 ALD cycles are shown in Fig.5(b). From the comparison of the reference spectra on the two substrates, a significantly higher separation  between the in-plane (E') and outof-plane (A1') vibrational peaks is observed for 1L MoS2 on Au (21 cm -1 ) as compared to the case of 1L MoS2 on Al2O3/Si (18 cm -1 ). Such a difference is due to significant red-shift of the E' peak (mostly associated to the strain) and to a slight blue shift of the A1' peak (mostly related to the doping) for 1L MoS2 membrane on gold. Interestingly, the E' peak red-shift is further increased and the A1' peak blue-shift is slightly reduced after ALD of the uniform Al2O3 film on the gold supported membrane. On the other hand, only a slight red shift of the A1' peak was observed after 80 ALD cycles on the Al2O3 supported 1L MoS2, probably due to the inhomogeneous Al2O3 coverage (as shown in Fig.1(a)). In order to achieve a quantification of the strain  (%) and doping n (cm -2 ) for the 1L MoS2 membranes on the two different substrates before and after the ALD process, a correlative analysis of the A1' vs E' peak frequencies has been carried out in Fig.5(c), according to the procedure recently discussed in Ref. [34]. The red and black lines in  Fig.5(c), the reference 1L MoS2/Au sample is characterized by an average tensile strain of0.21% and p-type doping of n0.25×10 13 cm -2 , whereas an opposite compressive strain -0.25% and n-type doping n0.5×10 13 cm -2 are observed for 1L MoS2 transferred onto the Al2O3/Si substrate. Such n-type behaviour is consistent with the unintentional doping type commonly reported for exfoliated or CVDgrown MoS2, which has been associated to the presence of defects (e.g. sulphur vacancies) or to other impurities in the MoS2 lattice [37,38]. In the case of 1L MoS2 on Au, a strong electron transfer to the substrate is guessed, which overcompensates the native n-type doping, resulting in a net p-type behaviour. Furthermore, the tensile strain for 1L MoS2 on Au can be ascribed to the lattice mismatch between MoS2 and the Au surface, mostly exposing (111) orientation [39,40]. cycles indicates no significant changes in the tensile strain, consistently with the highly inhomogeneous Al2O3 coverage, and an increase of the n-type doping to n0.6×10 13 cm -2 . This latter can be ascribed to positively charged defects [41] at the interface between the poor quality Al2O3 film and 1L MoS2. In Fig.6, micro-PL spectra acquired on the two reference 1L MoS2 samples and after 80 cycles ALD growth are also reported, to further elucidate the impact of the substrate and of the deposition process on the optical emission properties of the direct bandgap 1L MoS2 membrane.
A prominent emission peak located at 1.84 eV can be observed for 1L MoS2 supported by Al2O3/Si, whereas a significant reduction of the PL intensity accompanied by the red shift of the main peak position at 1.79 eV is found for the reference sample on Au. A similar quenching of the PL intensity has been reported for 1L MoS2 exfoliated on Au [34,35] and for MoS2 functionalized with Au nanoparticles [42]. This behavior can be explained in terms of a preferential transfer of photoexcited charges from MoS2 to Au. In addition, the tensile strain of 1L MoS2 in contact with Au can also play a role in the reduction of the PL yield [43]. After 80 ALD cycles, only a small reduction of the PL intensity was observed for the 1L MoS2/Al2O3 sample, which can be explained by the highly inhomogeneous Al2O3 coverage and to the small interaction of MoS2 with the dielectric substrate. On