Atomic layer deposition of a uniform thin film on two-dimensional transition metal dichalcogenides

Two-dimensional (2D) materials have been highlighted as promising materials that can exceed the current limitation of the three-dimensional (3D) bulk materials. Unlike bulk materials, 2D materials have been known to possess superior electrical, optical, and mechanical properties that cannot be obtained from conventional 3D materials.^1–4 For instance, graphene, the first reported 2D material, had been highlighted as a promising material that can reach a breakthrough in current electronics.⁵ Owing to its hexagonal-honeycomb lattice structure, carbon atoms contain sp²-hybridized bonded layer; thus, it exhibits extremely high conductivity and carrier mobility with only a nanometer-thick layer.⁶ Despite these unprecedented characteristics, however, it has not been widely used as a material for electronic devices since it has no intrinsic bandgap, which is essential for a switching characteristic. Many researchers have tried to open the bandgap of graphene but had some problems obtaining a certain level of bandgap using graphene, retaining its pristine characteristics.^7–9 For this reason, 2D transition metal dichalcogenide (TMD) has been considered a promising material for the channel material of electronics instead of graphene. Characteristics of TMDs are not only superior compared to those of conventional materials but also can be varied from a metal to a semiconductor and even to an insulator by a variety of combinations of metals and chalcogen elements.¹⁰ Molybdenum diselenide (MoS₂), one of the representative TMD materials, has been highlighted as a promising channel material of the transistor, showing remarkable electrical characteristics that include high mobility and an excellent on/off ratio.¹¹ In addition, direct to indirect bandgap transition of ultrathin 2D TMDs depending on the number of layers^12,13 and excellent mechanical durability^14–16 is used for the purpose of high-performance optical or mechanical device fabrication. Likewise, because of their superior properties, most 2D TMD research studies have focused their attention on the investigation of their own properties and expansion to the various applications using these peculiar characteristics over a decade.

To utilize 2D TMDs, deposition on the TMDs is frequently required in order to fabricate various devices^17,18 or synthesize a hybrid material for energy applications with inorganic/organic materials.^19,20 To date, there are numerous deposition techniques, including vapor deposition (physical or chemical), solution process, etc. Among them, atomic layer deposition (ALD), one of the specialized chemical vapor depositions that consists of sequential exposure of chemical species, is the most appropriate deposition method for nanoscale device fabrication.^21,22 Based on self-limiting growth characteristics, precursor molecules chemisorb onto the surface reaction sites and subsequent reactant molecules react with the adsorbed precursor species forming a subnanometer thick film within a cycle. Owing to its unique deposition manner, various advantages, including atomic-scale thickness controllability and large area uniformity, can be obtained. Moreover, high conformal coating even on a 3D-complex structure like nanoflakes is also available.^23–25 However, since the 2D material has quite a different surface chemistry compared to conventional 3D materials, the growth characteristics of ALD on the 2D material shows a significantly different aspect. Unlike conventional 3D materials, there is a lack of dangling bond on TMDs and this leads to a chemically inert surface on the basal plane of these materials.¹⁰ This indicates that no chemisorption of the chemical species is available on the surface. Even the exfoliated or synthesized 2D TMDs has inherent defects that are generated during the process, uneven distribution of reaction sites on TMDs induces nonuniform deposition on the surface after the ALD process,¹⁷ leading to a critical limitation on applying ALD on 2D materials. Therefore, many research studies have been focused on the uniform deposition of ALD-grown material without destroying or damaging a pristine TMD structure.

In this paper, ALD on 2D TMDs will be highlighted from the perspective of growth characteristics. There are numerous reports of ALD on 2D TMDs, but we will only focus on the growth mechanism and the methods for excellent uniformity. In terms of surface chemistry, dissimilar deposition mechanism of ALD on 2D TMDs will be discussed by referring to previous reports. Moreover, diverse approaches for uniform deposition of the ALD-grown material on 2D TMDs and its applications will be described in detail by suggesting the outlook and prospect of ALD on 2D materials.

The notable difference between conventional materials and 2D TMDs is the presence of the reaction sites on the surface. As mentioned in the Introduction briefly, a conventional material has a dangling bond on top of the surface that enables reactions with different chemical species. In the case of ideally infinite 2D materials, on the other hand, there was no dangling bond theoretically because of its peculiar crystal configuration²⁶ that indicates 2D materials are chemically inert. The absence of the dangling bond on the basal plane is one of the major advantages of 2D TMDs, which can be avoided by electrical deterioration from interface traps.²⁷ From the perspective of the ALD process, however, free of dangling bond suppresses a material growth on the surface. For instance, when the precursor or the reactant is exposed to the reactor on a dangling bond-rich substrate, a chemical reaction occurs on top of the surface by reacting with the dangling bonds (i.e., surface reaction sites), and, therefore, the film starts to grow. However, since there is lack of a reaction site on pristine 2D TMDs, film growth inhibition on the basal plane of 2D TMD is observed when ALD is conducted on the 2D materials.^17,28

In the case of surface properties of nonideal finite 2D TMDs, it has slightly different surface chemistry. Initially, most of the 2D TMD research was conducted with a TMD flake that was mechanically²⁹ or chemically exfoliated^30,31 from the single-crystal TMD. Among them, a mechanical exfoliation process that uses an adhesive tape for peeling off the ultrathin 2D TMD flake from the bulk TMD crystal ensures TMD flakes with single crystalline, high purity, and immaculateness. In particular, since a mechanically exfoliated 2D TMD flake has a very low level of defect concentration, using this flake helps us to understand the intrinsic characteristics of 2D TMD.^32,33 However, some defects or edge sites that exist on the exfoliated flake, generated during synthesis or transfer process (e.g., point defect or dislocations), also have a significant effect on its surface characteristics as well.^34,35 Since these have dangling bonds that are chemically reactive,^36–39 defects or edge sites of exfoliated TMD act as reaction sites for the gas molecules or chemicals.^40,41 While the exfoliated 2D TMDs showed superior characteristics that are suitable for laboratory-scale device fabrication, it is not appropriate for scaling up of the large-scale fabrication and controlling thickness and size of the flakes that are essential for characteristic modulation. Accordingly, the development of fabrication methods for fabricating 2D TMD flakes with large scale and high uniformity have also been widely researched. In contrast to exfoliation, however, CVD-grown TMDs showed a relatively high level of defect densities due to imperfection of the synthesis process, which induces various structural defects in the material.⁴² From the perspective of surface chemistry, a high level of defect densities on 2D TMD imply that a reaction with the following chemical species would be more favorable on the CVD-grown TMD than on the exfoliated one.

Since 2D TMDs consist of two or more different elements, transition metal and chalcogen, it is much more possible to have various types of defects than graphene which only consists a single carbon element. A previous study showed various point defects of 2D MoS₂ such as a monosulfur vacancy, a disulfur vacancy, a vacancy complex with Mo and three neighboring sulfur atoms, a vacancy complex with Mo and three disulfur pairs, antisite defects with a Mo atom, etc.⁴² In the case of graphene, the adsorption tendencies of chemical species on graphene depends on the type of defects.⁴³ Likewise, the adsorption of chemical species on 2D TMDs differs from what the chemical species or a kind of defect sites are. A previous report investigated adsorption of various molecules on 2D MoS₂.⁴⁴ The authors calculated adsorption energies between various molecules (CO₂, CO, O₂, H₂O, NO, N₂, and H₂) and basal plane or defects (S vacancy, S divacancy, Mo vacancy, one or two Mo atoms on S mono- and divacancy, and one or two S atoms on Mo mono- and divacancy) of 2D MoS₂ by using density functional theory (DFT) calculation. They revealed that all the molecules have larger adsorption energy on defect sites than that of on the basal plane, and the adsorption energy varied depending on the surface species and the type of chemical species. Consequently, defects on 2D TMDs can play a role as highly reactive sites and trap sites for different molecules by exposure to chemical functional groups.⁴⁵ 

Based on the classical ALD theory, a film is formed on reaction sites of the surface by the chemical reaction between chemical species (i.e., precursor and reactant).²¹ This chemical reaction, named as dissociative chemisorption, occurs with the help of external energy, forming a chemical bond between adsorbent (solid) and adsorbate (gas). During the ALD process, since chemical species could only react with surface reaction sites of the substrate and residual molecules that are not reacted would be purged out, this self-limiting growth based on the chemisorption is irreversible in general.⁴⁶ 

In contrast to this chemisorption-based reaction, another adsorption mechanism, physisorption, becomes more and more important for the modern ALD process, especially on 2D materials. Unlike the chemisorption-based process, when chemical species approaches on top of the 2D surface, these are physically attached on the surface with weak intermolecular force (e.g., van der Waals force) without making bond modification between an adsorbent and adsorbate. Although physisorption is one of the fundamental bonding theories, this had been seldom considered in the ALD process until carbon-based materials (i.e., carbon nanotube⁴⁷ or graphene^40,48,49) came out to the world since most of the ALD reaction were conducted on the substrate with full of dangling bonds that enable a favorable chemical reaction on the surface. However, since graphene itself lacks dangling bonds on the top of the surface, it showed a chemically inert surface, resulting in growth inhibition or abnormal nucleation growth on top of the graphene surface.^40,50,51 To overcome this limitation, many researchers have tried to physically attach molecules for the purpose of artificial adsorption sites for the reaction with subsequent chemical species.

Figure 1(a) shows a plot of Lennard–Jones potential model, which is generally used for molecular adsorption model. In this figure, the depth of the potential well indicates the attraction force between solid (substrate) and gas (precursor or reactant), and the x axis indicates a distance between them. As shown in the figure, the attraction force of physisorption is relatively low (∼40 meV) with a relatively large distance (>3 Å) compared to that of chemisorption (∼1 eV) with a small distance (<3 Å).^52,53 Because of its weak bond strength and relatively long distance, physically attached molecules can be easily detached by external energy source, for instance, thermal heating⁵⁴ or long purging time.⁵⁵ That is, since physisorption does not induce bond modification between the adsorbate and the adsorbent, this could be reversible. Figure 1(b) shows a simple schematic of ALD on the 2D material with and without physisorption. As we mentioned in the previous paragraph briefly, ALD on the 2D substrate is favorable on the defect or edge sites, not on the basal plane. However, if some chemical species are physically attached to the basal plane of the 2D material, it may act as a reaction site for the following precursor, resulting in nucleation on the basal plane. In other words, if molecules can be physically attached on the entire basal plane, uniform thin film deposition on dangling bond-free 2D TMDs would be also available by ALD.

In this paragraph, we will discuss the recent progress of ALD on 2D TMDs to obtain high uniform films even with a sub-10 nm thickness level that is essential for nanoscale device integration with high performance. Over a decade, numerous reports of ALD on 2D TMDs have been reported. In this report, however, we will mainly focus on the growth mechanism of ALD on 2D TMDs from a viewpoint of surface chemistry. Also, various methods for uniform thin film deposition will be described in detail. All reports of ALD on 2D TMDs hitherto are summarized in Table I as a function of the synthesis method of TMD.

The initial research of ALD on 2D TMDs began on the exfoliated TMD flakes from single-crystal TMD. In 2011, Radisavljevic et al. investigated device properties of MoS₂ top-gate transistors with or without high-k dielectrics.¹¹ For a fabrication monolayer MoS₂ field-effect transistor (FET) without high-k dielectrics, the MoS₂ monolayer (∼6.5 Å) was mechanically exfoliated onto a 270-nm-thick SiO₂/degenerately doped silicon substrate. Since the mobility of 2D materials combined with high-k dielectrics could be improved owing to dielectric screening,^56–59 the authors deposited ALD HfO₂ on monolayer MoS₂ without any treatment. HfO₂ was chosen high-k dielectrics due to its high dielectric constant (k ∼ 25) and large bandgap (∼5.7 eV).⁶⁰ For the deposition of HfO₂, tetrakis(dimethylamino)hafnium (TDMAHf) was used as an Hf precursor, and H₂O was used as a reactant, and 30 nm of HfO₂ was deposited on the MoS₂ flake for top-gate dielectric. After deposition, the electrical properties of FET were greatly enhanced owing to the screening effect, as mentioned before. However, even though direct ALD on an untreated MoS₂ surface might induce numerous nucleation, their result showed insignificant change on surface morphology before and after ALD HfO₂. In general, nucleation and growth of ALD films favorably occur on surfaces that have a high concentration of hydroxyl (–OH)-terminated groups, which is considered as adsorption sites for the precursors.⁶¹ However, although the mechanically exfoliated TMD flake has no inherent –OH species, relatively flat thin film was observed. This result was verified later that, when the thickness of HfO₂ was quite thick, the nucleated HfO₂ started to be connected eventually, forming a relatively flat thin film.⁶² 

Likewise, there is no specific obstacle to deposit relatively thick dielectric (>15 nm) on dangling bond-free TMD substrates using ALD. However, the thickness of the dielectric should be decreased for the integration of complementary metal oxide semiconductor (CMOS).⁴¹ In addition, the electrical properties of TMD-based FET would be remarkably improved when the thickness of the gate dielectric is small enough, especially at the sub-10 nm region, resulting in better control of the channel and larger drive current.⁵⁹ Therefore, deposition of ultrathin high-k dielectrics with excellent uniformity on 2D TMD layers is a prerequisite for high-performance device fabrication, and after that time, many studies have focused on the deposition that satisfies these requirements.

The initial approach to obtain a uniform thin film (<10 nm) on 2D TMDs was controlling of process temperature. In 2012, Liu et al. reported Al₂O₃ ALD on exfoliated MoS₂ and boron nitride (BN) with an experimental and theoretical method.²⁸ The deposition of ALD Al₂O₃ was conducted on mechanically exfoliated 2D MoS₂ and BN films using trimethylaluminum (TMA) and H₂O by differing process temperatures to investigate the effect of temperature on the uniform deposition. Figure 2(a) shows atomic force microscopy (AFM) images of MoS₂ and BN substrates after 111 cycles of ALD Al₂O₃ depending on the process temperature (200, 300, and 400 °C, respectively). The nominal thickness of Al₂O₃ that was calculated based on the growth rates of Al₂O₃ on the SiO₂ substrate was about 10 nm since there was no temperature dependence on growth rates of ALD Al₂O₃ on SiO₂ from 200 to 400 °C, indicating chemical adsorption of TMA on –OH groups was not affected by temperature from 200 to 400 °C. However, the morphology of ALD Al₂O₃ on BN and MoS₂ was strongly affected as a function of growth temperature. At 200 °C, ALD Al₂O₃ that was deposited on BN and MoS₂ showed excellent uniformity without voids or pinholes. As the process temperature of ALD Al₂O₃ increased, however, many voids and pinholes started to be formed. When the process temperature was increased up to 400 °C, numerous islands of Al₂O₃ were observed on the surface rather than on a thin film, regardless of the substrate. Surface coverage of ALD Al₂O₃ on MoS₂ that is calculated from these AFM images was slightly higher than that on BN, regardless of the process temperature. The aspect of the film growth depending on the process temperature is closely related to the physisorption of precursor on MoS₂ and BN. As discussed in Sec. III, precursor (or reactant) molecule would be physically adsorbed on the dangling bond-free substrate, forming a weak van der Waals interaction. As the temperature increases, however, some physisorbed molecules could be desorbed by breaking a weak bond and nonuniform reaction sites on TMDs remain. While the physisorbed molecules on the basal plane gradually detached as the process temperature increased, edge sites of TMDs still remained and acted as reaction sites because of the presence of dangling bonds.^40,50 Accordingly, the ALD process could not deposit uniform Al₂O₃ films on BN and MoS₂ surfaces as the process temperature increased.

In order to investigate the interaction between precursors and the surface of 2D materials and understand the reaction mechanism of initial ALD cycles on 2D materials, adsorption energy between TMA and MoS₂ or BN were calculated using DFT calculation. The physical adsorption energy of TMA on the BN surface was 8.7 kcal/mol, which was larger than that of H₂O on BN. The physisorption energy of TMA on MoS₂ was 21.9 kcal/mol, which is larger than that of H₂O on MoS₂. This indicates that TMA is more strongly physisorbed on 2D materials than H₂O. In addition, TMA has the largest interaction energy with MoS₂ because of the following reasons; positively charged Al can more favorably interact with negatively charged N and S than negatively charged O, and polarizabilities of TMA and MoS₂ are larger than those of H₂O and BN, respectively. Based on these results, Lennard–Jones potential is depicted in Fig. 2(b). These calculation results confirmed the higher coverage of Al₂O₃ on MoS₂ than on BN in the high temperature range (300–400 °C), which corresponds to the coverage difference. The authors revealed that temperature dependent growth indicates that ALD Al₂O₃ on the 2D substrate is mainly controlled by physisorption of the precursors, instead of chemisorption.

From both experimental and calculation results, the authors deduced that the growth of ALD Al₂O₃ on 2D materials is critically affected by physisorption, not chemisorption. Since physical adsorption on 2D TMDs is highly affected by not only adsorption energies but also polarizabilities of chemical species and TMD, a proper selection of the precursor and the reactant depending on the 2D material is highly demanded to enhance physisorption on a dangling bond-free substrate.

Even though lowering process temperature is one of the possible methods to obtain a uniform thin film on 2D TMDs, however, it also brings undesirable effects as well. For instance, low temperature ALD leads to poor film density⁶³ and much incorporation of impurities from precursor (e.g., C, N, O, etc.)⁶⁴ due to insufficient activation energy for the reaction between precursor and surface reaction sites. For these reasons, it is highly required to obtain uniform film using ALD without degrading the film quality.

In this section, various surface treatment methods will be discussed to acquire a highly uniform thin film on the 2D TMD substrate. Diverse methods including not only attaching molecules but also generating –OH species or defects on the TMD surface will be described in detail.

In 2013, McDonnell et al. reported ALD HfO₂ on a mechanically exfoliated MoS₂ substrate. TDMAHf and H₂O were used for HfO₂ deposition at 200 °C.¹⁷ In this report, 30 nm of HfO₂ deposited on MoS₂ covered the entire MoS₂ layer but not uniformly, unlike previous reports.^11,28,65 The different thing from those and this report was presumed by the residue removal process. In order to fabricate TMD-based FET, subsequent coating of organic materials followed by removal with solvents is typically carried out after the transfer or lithography process.^65,66 However, residues that are generated during these processes are hardly removed even after UHV annealing,⁶⁷ thus, these remained surface contaminants may wact as reaction sites. To investigate whether intentionally attached molecules (i.e., organic or solvent contaminants) would promote nucleation of ALD HfO₂ on MoS₂ flakes, they were soaked into acetone or N-methyl-2-pyrrolidone (NMP) or spin-coated with poly(methyl methacrylate) (PMMA) prior to ALD HfO₂. As shown in Fig. 3(a), even though there was no significant change after ALD when MoS₂ was submerged in acetone, the clear difference in phase images of AFM was observed after PMMA coating or the NMP treatment. From the AFM images, the author deduced that the organic residue or solvent, which remained on the MoS₂ substrate acted as a nucleation promotion site that improved uniformity of ALD HfO₂. They also confirmed the effect of purge time of the ALD process on the physisorption of chemical species. With extremely long purge time between precursor and reactant exposure (over 6 h), no HfO₂-related peaks were observed from x-ray photoelectron spectroscopy (XPS) [Fig. 3(b)], while there were clear Hf4f and O1s peaks when the precursor and reactant purge time were 5 and 10 s, respectively. This could be a piece of evidence that physisorbed precursor molecules can be detached by long purge time. Therefore, physisorption or molecule attachment (functionalization) would enhance the nucleation of ALD on the low number of reaction sites on the TMD surface.

In 2016, Park et al. successfully deposited Al₂O₃ on exfoliated WSe₂ by organic functionalization using the titanyl phthalocyanine (TiOPc) monolayer.⁶⁸ They were motivated by a previous report of uniform ALD on graphene using TiOPc.⁶⁹ From the previous first-principle calculation, they found a strong binding energy between dimethyl aluminum, surface species that were generated by dissociation of TMA with surface hydroxyl group, and elements of TiOPc (C, N, or O), over 1.5 eV, which indicates that TMA would attach on the TiOPc layer favorably, resulting in reaction sites for subsequent reactants. In this report, WSe₂ was deposited on highly ordered pyrolytic graphite (HOPG) using molecular beam epitaxy (MBE), then a monolayer of TiOPc was also deposited using MBE. From the scanning tunneling microscopy (STM) images after TiOPc deposition, shown in Fig. 3(c), highly ordered morphology without defect was observed, indicating that uniform and thin TiOPc monolayer was well deposited on MBE-grown WSe₂. The energy level that is represented with color profiles, and this shows that every single TiOPc molecule has a single bright peak that is negatively charged, providing potential binding sites for polar ALD precursors [Fig. 3(d)]. To confirm that this TiOPc is effective for nucleation of following ALD, ALD Al₂O₃ was deposited on bare and TiOPc-deposited WSe₂. As shown in Figs. 3(e) and 3(f), after TiOPc deposition, smooth and flat Al₂O₃ was obtained (RMS roughness ∼0.15 nm), which is comparable to the roughness of conventional ALD Al₂O₃ roughness,⁶³ while ALD on bare WSe₂ showed large pinholes (RMS roughness ∼3.6 nm).

Plasma treatment is one of the most frequently used techniques in the modern fabrication process that readily primes any surface for better acceptance of the following process.^70,71 Owing to highly energetic species that are made during plasma ignition, functional groups would be generated on the substrate surface by the reaction between reactive ion/electron species and surface group.⁷² In particular, this kind of surface treatment has been widely used for carbon nanotubes (CNTs)⁷³ or graphene,⁷⁴ which have lack of dangling bond on them, to promote a nucleation of ALD by generating hydroxyl species on top of the surface. However, it has critical obstacle that pristine substrate may get a damage after plasma ignition because of ion bombardment from ion/electron species that have high kinetic energy.⁷⁵ Since 2D TMDs are considerably thin, compared to bulk materials, the effect of plasma damage on 2D TMDs is much critical than that on conventional bulk materials. Therefore, it is highly desirable to functionalize the surface of the ultrathin 2D TMD surface with less damage.

In 2013, Kim et al. reported the effect of oxygen-plasma pretreatment on the growth of ALD Al₂O₃ and HfO₂ on 2D MoS₂.⁷⁶ To minimize damage from plasma, they used remote inductively coupled plasma (ICP) which plasma glow and surface treatment region was in a separate reactor. Using this remote plasma surface treatment, the TMD surface was not subjected to intense ion and electron bombardment during the treatment process.⁷⁷ Firstly, the authors investigated the growth characteristics of ALD Al₂O₃ and HfO₂ on pristine MoS₂ at 250 °C as a function of thickness. Multilayered 2D MoS₂ was prepared by mechanical exfoliation, and TMA, tetrakis(ethylmethylamino)hafnium (TEMAHf), and H₂O were used for ALD Al₂O₃ and HfO₂. Before the ALD process, oxygen plasma was treated on the MoS₂ flake to improve the following precursors’ adsorption. When nominally 1-nm-thick Al₂O₃ was deposited on pristine MoS₂, the poor surface coverage of Al₂O₃ was observed (∼23%), while when the thickness of Al₂O₃ was increased to 10 and 30 nm, the surface coverage was also increased to 77 and 97%, indicating the connection of individual Al₂O₃ islands. However, there were still uncovered grain boundaries even after 30 nm Al₂O₃ deposition. In the case of HfO₂ deposition, the more uniform film was deposited on MoS₂. However, numerous pinholes were still observed because of different reactivity of basal plain and grain boundary with TEMAHf precursor. These results are consistent with previous reports that are mentioned in Sec. IV A.

In order to enhance surface coverage of the ALD film on MoS₂, O₂ plasma was treated prior to film deposition by varying treatment time (10, 20, and 30 s). The surface coverage of ALD Al₂O₃ films was significantly increased only after 10 s of plasma pretreatment (93%), compared to that without plasma treatment (77%). Nearly completely covered Al₂O₃ films would be formed with 30 s of plasma pretreatment, showing that the surface of reaction sites would be increased as the plasma treatment time increased. A similar tendency of coverage improvement with plasma pretreatment was also confirmed for the ALD HfO₂ case. They confirmed optimal plasma treatment time for uniform ALD on 2D MoS₂ as 30 s. They also deposited both Al₂O₃ and HfO₂ with various thickness after 30 s plasma treatment on MoS₂, and SEM images of them are shown in Fig. 4(a). From these figures, nearly complete deposition of 10-nm-thick Al₂O₃ and HfO₂ were available on MoS₂ after 30 s of O₂ plasma treatment. Figure 4(b) shows root mean square (RMS) surface roughness of ALD Al₂O₃ and HfO₂ that were extracted from AFM images as a function of various plasma pretreatment times. The RMS surface roughness gradually decreased as the plasma pretreatment time increased on both Al₂O₃ and HfO₂, indicating that more surface reaction sites were generated by increasing treatment time. Thus, dielectrics were more uniformly deposited as the plasma-treated time increased.

After the optimization of the plasma treatment process, XPS and Raman spectroscopy were measured to investigate what makes the improved growth nature of ALD dielectric on MoS₂. From the XPS spectra of MoS₂ [Fig. 4(c)], Mo⁶⁺ peaks which correspond to MoO₃ appeared after 30 s of the O₂ plasma treatment, while intensities of Mo⁴⁺ and S²⁻ was slightly decreased, indicating that some MoS₂ layers were transformed to MoO₃ after plasma treatment. MoO₃ is known as hydrophilic material that contains numerous hydroxyl (–OH) species on the surface,⁷⁸ thus the transformation to MoO₃ could lead to an improvement of surface coverage of ALD films. Also, from Raman spectroscopy analysis, a decrease of peak intensities of two MoS₂ Raman peaks (E¹_2g and A_1g) were observed, which indicates the oxidation of MoS₂. However, there was no identifiable peak shift by the additional physical damage. From this paper, they concluded that the improvement of surface roughness after the plasma is mainly due to the oxidation of the surface, however, undesirable MoO₃ formation followed as well.

In 2018, Hong et al. reported direct plasma treatment on the CVD-grown MoSe₂ layer to obtain uniform and high quality of dielectric to passivate bulky MoSe₂ from unintentional molecular adsorption which may induce deterioration of FET performance.⁷⁹ MoSe₂ is one of the 2D TMDs that has smaller bandgap (1.57 eV for monolayer, 1.09 for bulk) than MoS₂ (1.88 eV for monolayer, 1.23 for bulk), therefore, better electrical properties could be anticipated in MoSe₂ FET.^80,81 Unlike previous reports that used remote plasma, they used direct plasma to oxidize the topmost MoSe₂ layer intentionally. The oxygen-plasma treatment (100 W, 30 SCCM of O₂, 10⁻²Pa) was conducted on 50 nm of MoSe₂ layer for 1 min prior to passivation. Then, 40 nm of Al₂O₃ was deposited on the MoSe₂ using TMA and H₂O at 200 °C. Cross-sectional and high-resolution transmission electron microscopy (TEM) images of multilayer MoSe₂ is shown in Fig. 4(d). From these figures, around 3 nm of MoO₃ was observed after direct plasma exposure, which corresponds to three to five uppermost MoSe₂ layers were transformed. Notwithstanding the relatively strong plasma exposure, the rest of the MoSe₂ layers seem to retain its crystallinity. Since the thickness of the MoSe₂ layer was around 50 nm, the oxidation of the topmost MoSe₂ layer might not be critical to the entire MoSe₂, and the oxide layer (MoO_x) may act as a nucleation promotion layer between MoSe₂ and Al₂O₃. From XPS analysis, as shown in Fig. 4(e), clear peak shifts were observed from overall spectra, that indicates oxygen incorporation into MoSe₂.⁸² Notable thing from these spectra is that Se–O, similar species to S–O that we mentioned in previous section, was also observed after direct plasma treatment. And these Se–O species would also assist chemical species to be adsorbed on top of the surface.

Recently, Huang et al reported H₂O plasma treatment for the uniform dielectric deposition on the MoS₂ substrate.⁸³ They used H₂O plasma to generate –OH species on top of TMD surface since H₂O has hydrogen atom itself, unlike O₃ or O₂ plasma, hydroxyl radical (–OH*) can be generated by plasma ignition.^84,85 As the generated hydroxyl groups adsorbed on the MoS₂ substrate acting as reaction sites, uniform and high quality Al₂O₃ was obtained on the MoS₂ flake. Using H₂O plasma treatment, they successfully obtained uniform Al₂O₃ film of on exfoliated MoS₂ down to 1.5 nm, which is a state-of-art result of thickness of uniform deposition on TMD material. Moreover, from the Raman and PL analysis, the peak intensities were not decreased after H₂O plasma treatment, rather increased because the carbonaceous contaminants on the mechanically exfoliated MoS₂ were removed during the plasma process (self-cleaning effect),⁸⁶ suggesting a promising method for the TMD surface treatment.

As described, although plasma treatment can promote nucleation of ALD on TMD surface by forming dangling bonds on top of the surface, it also damages ultrathin TMD itself because of highly reactive species that are generated during plasma ignition, resulting in a deterioration of its intrinsic and unique characteristics. Thus, forming a dangling bond on 2D TMDs without detriment should be carried out. For these reasons, researchers made efforts to develop a facile way that can generate reaction sites on the 2D TMD surface without damaging itself.

In 2014, Azcatl et al. investigated the effect of the room temperature ultraviolet-ozone (UV-O₃) treatment on the growth of ALD Al₂O₃ on 2D MoS₂.⁸⁷ UV-O₃ treatment is one of the traditional treatment methods which can remove surface contaminants by the reaction with highly reactive radicals that are generated from UV and O₃ reaction.⁸⁸ But nowadays, it becomes a more captivating technique to make a hydrophilic surface by forming a hydroxyl species on the surface.⁸⁹ Likewise, using this surface treatment, it is anticipated that ALD on 2D TMD could obtain a smooth film retaining its original structure, and this was already confirmed on graphene.^90,91 They proposed a nondestructive MoS₂ functionalization method without breaking sulfur-molybdenum bonds by using UV-O₃ exposure. To remove the top layers from bulk MoS₂, the authors performed mechanical exfoliation by using adhesive tape. After the exfoliation, the authors loaded the sample into the ultrahigh vacuum (UHV) within 5 min and transferred to a chamber for the UV-O₃ treatment.

in situ XPS was used to investigate chemical state changes of the MoS₂ surface after UV-O₃ exposure without affecting external contamination from ambient. Figures 5(a) and 5(b) show XPS spectra of S2p, Mo3d, O1s, and C1s before and after UV-O₃ treatment. As shown in Fig. 5(a), a small peak was newly observed on Mo3d after the UV-O₃ treatment, which corresponds to S–O, however, the oxygen-related peak (i.e., Mo-O) was not found. In S2p spectra, the additional doublet peak near 164.8 eV was observed after UV-O₃ exposure. It is coincident with the appearance of an O1s peak [Fig. 5(b)], which corresponds to the S–O feature that was found in Mo3d. The newly generated S2p peaks were matched with S⁴⁺ which corresponds to the previous reports on MoS₂ oxidation,⁹² instead of S²⁻. From these spectra changes, it could be deduced that S–O was formed on top of the MoS₂ surface without Mo–S bond cleavage. They also calculated that the functionalized oxygen was formed with a monolayer coverage by calculating peak intensities of S2p (I_S-O/I_s), which confirms that functionalization using UV-O₃ would make plenty of reaction sites for subsequent precursors on MoS₂. Carbon residue on the surface was also eliminated as well after UV-O₃ treatment [Fig. 5(b)] since UV-O₃ is quite effective to remove carbonaceous contaminants on the surface, as mentioned earlier in this section. All the changes described above occurred when UV and O₃ were exposed simultaneously since UV can not only greatly increase the reactivity of O₃ (Ref. 88) but also highly reactive O radicals could be generated by O₂ dissociation by UV absorption.⁹² From the DFT calculation, they showed that the formation of S–O bonds is energetically favorable without cleaving Mo–S bonds. Also, they claimed that O is hard to substitute S atom since S vacancy formation energy was quite large.

To investigate growth characteristics and uniformity of film growth on treated MoS₂, ALD Al₂O₃ on UV-O₃ functionalized MoS₂ was conducted, and surface morphology was measured by AFM and high-resolution TEM (HR-TEM). Figure 5(c) showed AFM images of Al₂O₃-deposited MoS₂ at 200, 250, and 300 °C before and after the UV-O₃ treatment and cross-sectional TEM images after the treatment. Without the UV-O₃ treatment, nucleation occurred selectively on edges sites of MoS₂, regardless of temperature. At 200 °C, the density of Al₂O₃ clusters on the MoS₂ is higher than those of higher temperature conditions because the thermal desorption of chemisorbed oxygen species was suppressed at relatively low temperature. In contrast, the coverage is significantly increased on UV-O₃ functionalized MoS₂. Al₂O₃ films showed very small RMS roughness values of 0.14, 0.17, and 0.30 nm, with ALD process temperatures of 200, 250, and 300 °C, respectively. From the HR-TEM cross-section images of Al₂O₃ on UV-O₃ functionalized MoS₂, Al₂O₃ films with excellent uniformity were observed at 200° and 250 °C. However, Al₂O₃ deposited at 300 °C exhibited island-type growth, suggesting that S–O species that were generated after UV-O₃ treatment would stand up to 250 °C, and started to be decomposed or desorbed at 300 °C. From the XPS analysis after ALD Al₂O₃ on functionalized MoS₂, the S–O species was disappeared after ALD Al₂O₃ from the XPS analysis, which indicates surface S–O species acted as a sacrificial layer for ALD nucleation. Moreover, there was no Mo–Al or S–Al peak, indicating a noncovalent bond of Al₂O₃ on MoS₂, similar to the previous result of HfO₂ on MoS₂.¹⁷ 

The authors also showed the surface chemistry of WSe₂ and MoSe₂ with the UV-O₃ treatment.⁹³ The study showed that reactivities of TMDs diverse with their molecular species, and the different oxidation with the UV-O₃ treatment. Unlike MoS₂, both MoSe₂ and WSe₂ showed oxidation of transition metal (i.e., Mo and W) after 6 min of UV-O₃ exposure, and this might due to the reactivity difference toward oxidation. From the previous report, WSe₂ is the highest oxidation tendency, and MoS₂ is the lowest among the three TMDs.⁹⁴ This infers that MoSe₂ or WSe₂ are more sensitive to form artificial nucleation sites without damaging the original materials, compared with MoS₂. However, it was found that the surface oxides generated on MoSe₂ and WSe₂ by UV-O₃ exposure were reduced or eliminated by the ALD process. As shown in Figs. 7(d) and 7(f), only after one cycle of ALD HfO₂ on UV-O₃ treated MoSe₂, the oxidized MoSe₂ species (MoSe_xO_y) started to be diminished and a new Mo⁵⁺ peak that might be induced by reduction of Mo⁶⁺ was observed. This phenomenon, referred to as “self-cleaning effect,”⁹⁵ was originated from the ligand exchange between surface oxide species and Hf precursor since the formation of HfO₂ is thermodynamically favorable than retaining MoO_x.^96,97 On the other hand, when HfO₂ was deposited on WSe₂, WO_x species were not completed removed after 30 cycles of ALD HfO₂ even though HfO₂ is more favorable to be formed than retaining WO_x. This might because of the following reasons: further reaction such as dissociation of W–O might be preceded before ligand exchange with Hf precursor,⁹⁸ or high stability of WO_x hindered the reduction of WO_x.⁹⁹ This implies that even UV-O₃ treatment has been shown as promising method for surface functionalization, the effect might be dissimilar to the different 2D TMDs.

Even though a uniform thin film could be deposited using the plasma or UV and O₃ treatment, slight damage during the process is inevitable, resulting in a deterioration of device properties because of its high reactivity with TMDs. To avoid this, in 2014, Zou et al. reported a new way to deposit a uniform thin film on 2D TMDs by insertion of ultrathin buffer layer, which can promote nucleation and growth of ALD-grown film between TMD and dielectric [simple schematic is shown in Fig. 6(a)].¹⁰⁰ For the ALD film growth promotion, three different metals (Mg, Al, Y) were thermally evaporated on the MoS₂ substrate with a nominal thickness of 1 nm and then oxidized for several hours in the drying oven. Nucleation with a buffer layer, shown in Fig. 6(b), showed quite different growth nature depending on what the buffer layer material was. When HfO₂ was deposited on the MgO buffer layer, numerous granularlike nucleation of HfO₂ was observed with voids and pinholes and this might be due to the gap between large MgO nanoparticles, while smooth film growth on Al₂O₃ or Y₂O₃ buffer layers. In contrast to MoS₂, buffer metal oxide has numerous reaction sites on top of the surface, reaction with subsequent chemical species is favorable, making a uniform film on top of the buffer layer. In terms of melting point, Y has the highest melting point among three different metals, which means less probability of coalescence during the ALD process. Even the melting point of a thin film is lower than that of the bulk film, the thin Y₂O₃ film could resist any nucleation or coalescence during the ALD process. In addition, since Y is thought to have a good wettability to 2D materials,¹⁰¹ uniform and compact formation of buffer and MoS₂ could be possible. From Raman analysis [Fig. 6(c)], no notable change was observed after buffer layer and dielectric deposition, indicating a buffer layer deposition without damaging the original MoS₂ structure. Utilizing the MoS₂/Y₂O₃/HfO₂ stack, they could obtain a uniform HfO₂ thin film by further scaling down HfO₂ to 9 nm, and the fabricated device exhibits excellent electrical properties, showing the excellent interface quality and scalability.

Son et al. also reported a uniform thin film deposition on MoS₂ flake using buffer layer insertion prior to ALD deposition.⁶² For the buffer layer, which promoted nucleation of Al₂O₃, 1-nm-thick Al was evaporated on the MoS₂ flake and naturally oxidized. Since the surface energy of Al is much higher (1.27 × 10³mJ/m²)¹⁰² than that of MoS₂ (46.5 mJ/m²),¹⁰³ granular Al was formed on top of the MoS₂, not a film. However, through the oxidation process, the thickness of Al was increased around 4 nm due to the metal-to-oxide transformation,¹⁰⁴ resulting in a strain on MoS₂ that could modulate the electrical properties.¹⁰⁵ The AFM images of 3-nm-thick Al₂O₃ deposition on MoS₂ with and without the buffer layer are shown in Fig. 6(d). Without this buffer layer, surface roughness and coverage of ALD Al₂O₃ on MoS₂ without buffer layer were 1.969 nm and 61.89%, respectively, while those with buffer layer were 0.649 nm and 99.29%, respectively. These results confirm that the insertion of an ultrathin buffer layer between 2D TMD and dielectric not only could promote the film growth on TMD but leave the pristine nature of TMD intact. However, by inserting an additional layer, the increase of total thickness which might be detrimental for nanoscale device fabrication is also inevitable.

As discussed in Sec. IV B 3, UV-O₃ treatment showed optimistic results for the generation of reaction sites on MoS₂ without damaging the host material. Nevertheless, some papers have reported that long exposure of UV on TMDs may induce photodegradation, which could deteriorate the electrical properties of TMDs.^106,107 Thus, researchers tried to find another way to make full of reaction sites on 2D TMDs without using UV light.

In 2014, Cheng et al. reported an improvement of ALD nucleation on MoS₂ using O₃ as a reactant. O₃ has been known as a strong oxidant that is frequently used for the ALD process.^108–111 Moreover, because of the high reactivity, various advantages including low impurities and high density would be anticipated when it was used as the reactant of ALD. However, since it was revealed that O₃ itself could not form surface S–O functional groups, reaction sites for subsequent precursors, on MoS₂,¹⁷ they used a two-step deposition process for ALD Al₂O₃. At 200 °C, ALD Al₂O₃ using TMA and H₂O with 3 nm of nominal thickness showed poor uniformity on the MoS₂ flake, similar to the previous reports that mentioned before. However, the uniformity of ALD Al₂O₃ significantly improved when O₃ was adopted for reactant instead of H₂O, but still a lot of pinholes existed in the film. To improve the surface morphology, five cycles of ALD Al₂O₃ was conducted at 30 °C to form a seeding layer prior to ALD Al₂O₃ at 200 °C, and it showed fairly smooth RMS roughness (0.23 nm) compared to that of pristine MoS₂ (0.18 nm). Since reactive O* that were generated from O₃ dissociation could provide the following precursors to reaction sites by forming weak S–O bond, artificially generated surface reaction species could be also desorbed or diffused on the surface as the temperature increases, leading a high roughness. Therefore, by employing a low temperature process in advance, undesirable effects such as thermal desorption of reaction sites would be avoided. From Raman spectra of MoS₂ after O₃ exposure as a function of process temperature, there was nearly no changes before and after O₃ treatment since MoS₂ is thermally stable against the oxygen-rich atmosphere up to 370 °C¹¹² due to its high dissociative binding energy.⁹⁸ Therefore, using O₃ as a reactant could supply surface reaction sites on the 2D TMD surface without damaging or braking its original structures, which has great potential for future TMD-based electronics.

As mentioned in Sec. IV B 4, plasma treatment using O₂ or H₂O can make surface hydroxyl species, which enables reaction with the following chemical species for ALD. However, it may also damage the pristine material, resulting in a device performance deterioration.⁸⁷ In particular, since the total thickness of 2D flakes that are used for FET fabrication is quite thin (<10 nm), partially oxidized or damaged 2D flake result in quite different characteristics with those of undamaged materials.

Recently, some researchers have reported a new way of uniform dielectric deposition on 2D TMDs using PE-ALD. Unlike thermal ALD, PE-ALD uses highly reactive chemical species that are generated during plasma reaction (e.g., O*, H*, etc.).^113–115 Because of high reactivity of the radicals, the use of PE-ALD not only enables lower process temperature but also can make a high density film, compared with thermal ALD.^116,117 In this research, they tried to minimize the damage of 2D TMDs from highly reactive plasma species, such as the use of remote plasma or relatively mild reactant source. In 2017, Price et al. reported uniform Al₂O₃ and HfO₂ deposition on exfoliated multilayer MoS₂ (6–8 nm) using remote PE-ALD.¹¹⁸ They used TMA and TEMAHf as precursors, and H₂O and O₂ plasma as reactants for ALD Al₂O₃ and HfO₂, respectively. Both Al₂O₃ and HfO₂ films grown by PE-ALD showed a smooth surface on several layers of MoS₂, regardless of growth temperature, while thermal ALD showed numerous nucleation on top of the MoS₂ flake (Fig. 7 shows AFM images of thermal and PE-ALD of Al₂O₃ on MoS₂). Using PE-ALD, they were able to obtain uniform thin film deposition down to 3.5 nm with subnanometer scale roughness. Although the uniform growth using thermal ALD was determined by the coverage of physisorbed molecules, since PE-ALD induce dangling bonds on the surface directly, physisorption is not the factor that affects the initial growth characteristics anymore. They could acquire ultrathin dielectric thin film (∼3.5 nm) on MoS₂ with subnanometer scale roughness. As they fabricated and characterized the MoS₂ FET with these dielectrics, the change of electrical properties was quite different. When PE-ALD Al₂O₃ was deposited on MoS₂ FET, PE-ALD HfO₂ process did not degrade the electrical properties of MoS₂ FET, rather improved the performance, contrary to ALD HfO₂ or PE-ALD Al₂O₃. From XPS analysis, they found that there was nearly no oxidation on MoS₂ flake after the PE-ALD process since MoO₃-related peak was not observed on Mo3d and S2s spectra, respectively. However, it is hard to say that there was no oxidation since the oxidation of the atop MoS₂ layer could not be detected because of the XPS detection limit. Instead, there was a slight negative peak shift of Mo3d and S2s after the PE-ALD process, indicating p-type doping.^119,120 PE-ALD Al₂O₃ showed a larger shift (∼1.2 eV), whereas, the PE-ALD HfO₂ showed a much smaller shift (∼0.4 eV), which indicates that PE-ALD HfO₂ affected less on MoS₂ than PE-ALD Al₂O₃.

Even this PE-ALD does not deteriorate the electrical characteristics of MoS₂-based FET, rather enhanced the performance, it could be possible since the number of MoS₂ was around 6–8 nm, which correspond 9–12 layers of MoS₂. In the case of thick flake, the effect of oxidation or damage could be negligible since only the uppermost TMD layer was affected by plasma species. Thus, they also investigated the effect of PE-ALD on relatively thin, mono-, bi-, and trilayer of MoS_2.¹²¹ They used CVD-grown MoS_2, which has numerous inherent defects on the surface compared to mechanically exfoliated. Similar to the previous reports with exfoliated MoS₂, electrical property degradation of monolayer MoS₂ FET, such as the V_th shift, I_on/I_off ratio decrease, was observed. However, in the case of FET using bi- and trilayer MoS₂, the deterioration of electrical properties was quite decreased, and this was clarified by TEM analysis that the uppermost layer of MoS₂ was damaged after the PE-ALD process. As shown in Fig. 7(c), the TEM image of trilayer MoS₂ after thermal ALD shows clear and distinct three layers between SiO₂ and HfO₂, which indicates no damage after the thermal ALD process. In contrast to thermal ALD, however, bleary three layers of MoS₂ were observed which possibly indicates plasma damage. The high-resolution TEM image [right inset image of Fig. 7(c)] clearly showed two MoS₂ layers on the bottom side and one blurry layer on the uppermost MoS₂ flake, verifying that the uppermost layer of MoS₂ was damaged during the PE-ALD process, and this result corresponds to the previous results of MoS₂ FET. From their previous reports, PE-ALD HfO₂ showed less deterioration of FET characteristics when it applied to MoS₂ FET than PE-ALD Al₂O₃. Thus, they also calculated adsorption energy of Hf precursor of MoS₂ depending on different surface configuration, pristine, sulfur vacancies, and H-passivated vacancies in MoS₂ using the first-principle calculation. From there calculation, the adsorption energy of the precursor was −1.06 eV for pristine, −1.33 for sulfur vacancies, and −2.52 eV for H-passivated vacancies in MoS₂. This result indicates that H-terminated vacancies in MoS₂ that were generated in the PE-ALD process are most stable among them, indicating PE-ALD could promote nucleation and growth film using ALD by forming these species on the topmost 2D TMD layer.

In this review, we discussed the extraordinary surface nature of 2D TMDs and the way to obtain uniform films on the TMD surface under 10 nm. Because of its fascinating and versatile characteristics of 2D TMDs, these can be applied not only to future electronics that need characteristics beyond the limits of conventional materials but also to diverse fields that cannot be acquired because of physical limits of conventional materials. For these applications, not only the development of 2D TMD synthesis for large-scale fabrication but also thin film deposition with excellent uniformity are both essential. Given the above, an ultrathin film is able to be obtained on 2D TMDs with negligible effect from the process, and the adequate process and chemical species can be selected depending on the host materials. These results provide a perspective of the ALD growth mechanism on 2D materials and solution for the deposition of ultrathin layers with high quality. Since the various methods have their pros and cons, we cannot suggest which method is the most promising for ALD on 2D TMDs. However, we predict that a new novel method, which can deposit a high quality and ultrathin film without the degradation of 2D TMDs properties, would be developed by comprehensive complementation of previous methods, which enables a breakthrough for the demonstration of various electrical or optical devices based on 2D materials.

This work was supported by the Materials and Components Technology Development Program of MOTIE/KEIT (No. 10080642, Development on precursors for carbon/halogen-free thin film and their delivery system for high-k/metal gate application and No. 10080527, Development of commercialization technology of high sensitive gas sensor based on chalcogenide 2D nanomaterial).