van der Waals epitaxy of 2D h-AlN on TMDs by atomic layer deposition at 250 °C Open Access

Semiconducting two-dimensional (2D) material is one of the promising candidates as a channel material in optoelectronic devices^1–3 and field-effect transistors (FETs) in future monolithic 3D ICs^4–6 owing to their attractive intrinsic bandgap values, self-passivated surface, and atomic flatness. Even though the 2D transition metal dichalcogenide (TMD) materials have been widely studied, finding an attractive gate dielectric stack for TMD transistors has been challenging.^7–9 To construct ultra-scaled TMD FETs, integrating nm-thin uniform high-k dielectric films is necessary for obtaining thin equivalent oxide thickness (EOT) and excellent channel electrostatic control. A few-nm thin atomic layer deposition (ALD) high-k HfO₂ film has proven to be an excellent manufacturable gate dielectric stack for the most advanced Si FET.^10–12 The ALD method based on self-limiting chemical reactions between the precursors and the reactive sites of the substrate surface is well suited for producing smooth and uniform films. However, previous research of dielectric oxide deposition by ALD on an inactive TMD surface often led to unsatisfactory coverage with pinholes and even uncovered areas.^13,14 The surface coverages of HfO₂ on TMD could be improved by forming a less-well-controlled seed layer^15,16 or mild plasma treatments,^41,42 which may damage the TMD surface. On the other hand, hexagonal boron nitride (h-BN) was widely used to demonstrate that a 2D gate dielectric material can enhance the mobility of carriers in the channel and the quality of the semiconductor–insulator interface of the 2D FET, as evidenced by better subthreshold swing.^17,18 However, the h-BN layer is formed through chemical vapor deposition (CVD) at elevated temperatures above 1000 °C and then transferred onto TMDs using the exfoliation process.^19–21 Therefore, a higher-k (than h-BN) 2D dielectric that can be ALD deposited on TMDs at the CMOS-technology compatible temperature as the gate dielectric or as the interfacial layer between TMD and ALD HfO₂ would be ideal. h-AlN has an excellent lattice match with Mo(W)S₂, and its dielectric constant of 8.73 is much higher than that of h-BN (∼4.7).

This study developed an ALD method to grow 2D h-AlN layers on top of the TMD's surface. Generally, the temperature for growing AlN films using ALD is between 150–300 °C.^22,23 High-resolution transmission electron microscopy (HR-TEM) images of h-AlN grown on several TMDs show that h-AlN has layered structures, and the crystal quality is better when TMD and AlN lattice-constant mismatch is smaller. Based on the x-ray scattering results, the h-AlN layered structure has its hexagonal lattice aligned with the TMD lattice. Based on these observations, we believe that the AlN growth mechanism is van der Waals epitaxy (VDWE).^24,25 Compared to ALD HfO₂ directly on the TMD surface, we demonstrate that adding h-AlN as an IL has the advantage of notable improvement in the HfO₂ roughness and uniformity. However, the electrical properties of the h-AlN/TMD interface still need to be studied.

This study reports h-AlN growth on various TMD materials (WS₂, MoS₂, and WSe₂) using the plasma-enhanced ALD (PEALD) method. Before the AlN deposition, remote H₂/N₂ mixture plasma surface pretreatment on TMD was performed. The best grazing-incidence wide-angle x-ray scattering (GIWAXS) intensity and FWHM were achieved with equal N₂ and H₂ flow rates of 50  sccm, a plasma power of 30 W, and a time of 400 s as shown in Table I. Then, trimethylaluminum (TMA) and remote N₂ plasma at 300 W were used as the precursors of Al and N. After AlN was formed, a TiN capping layer was in situ deposited using tetrakis(dimethylamido)titanium (TDMAT) and N₂ plasma at 300 W to prevent oxidation. All ALD processes were performed at 250 °C. The AlN growth rate on TMD layers was 0.58–0.68 Å per cycle. We sought to elucidate the growth mechanism of the AlN growth on the TMDs surface by also preparing a reference sample of AlN grown on sapphire. The crystallographic properties of the layered h-AlN/TMDs were examined using synchrotron-based GIWAXS with the detector at 9.5 cm from the sample to access q∼5 Å⁻¹ at Taiwan Photon Source (TPS) BL25A. The GIWAXS data were collected with an area detector. High-resolution transmission electron microscopy (HR-TEM) was used to probe the layered h-AlN/TMD microstructure and obtain atomic-scale images. The surface morphologies and roughness of the samples were examined with atomic force microscopy (AFM).

We performed calculations using the density functional theory (DFT) via Quantum ATK developed by Synopsys. The DFT calculations used the Perdew–Burke–Ernzerhof (PBE) exchange–correlation function and the PseuDojo-medium basis set. The nonlocal dispersive forces (the van der Waals force) were added into DFT calculation with the Grimme DFT-2D method. The HfO₂/h-AlN/WS₂ structure was modeled as stable four unit cells of m-HfO₂²⁶ with orientation (110), h-AlN (1 to 4 layers), and monolayer WS₂. The mesh cutoff was 120 Hartree, and Monkhorst–Pack k-points grids of 3 × 7 × 1 were used. The structure was relaxed along the z-direction. The force minimization was 5 meV/Å for all structure relaxations. For the study of the band structure and density of states, the Brillouin zone was sampled by 11 × 11 × 1 k-points.

Figure 1 presents cross-sectional HR-TEM images of TiN/AlN on WS₂/sapphire, on MoS₂/sapphire, on WSe₂/sapphire, and on sapphire directly. The microstructure of the AlN film exhibited a 2D layered structure on the TMDs surface compared with that of amorphous growth on the sapphire surface. These films were grown by ALD in the same equipment under the same condition and at the same time. The growth rates of the layered h-AlN film on TMDs (0.58–0.68 Å per cycle) were slower than the growth rate of AlN on the sapphire substrate (0.88 Å per cycle), probably because the chemically inactive surface of TMDs allowed only vdW physisorption during layered h-AlN deposition on TMDs.^27,28 The growth mode of the layered h-AlN film on TMD crystals may be similar to the Frank–van der Merwe (FM) mode^29,30 in forming the layer-by-layer stacked structure shown in Figs. 1(a)–1(c). Figure 1(d) clarified that the structural features of layered h-AlN were due to TMD rather than the sapphire substrate. The layered h-AlN was also grown on the WS₂/SiO₂(quartz) substrate. Figure 2(a) presents the high-angle annular dark-field (HAADF)-TEM of the TiN/AlN/WS₂/SiO₂ structure. Each layer of the TiN/AlN/WS₂/SiO₂ structure is sharply identified. Figure 2(b) presents the HR-TEM image from a selected region highlighted in Fig. 2(a). Even though the SiO₂ substrate is non-crystalline, AlN on WS₂ still exhibits the layered feature. Figures 2(c)–2(i) present a spatial elemental mapping of Ti, N, Al, W, S, Si, and O in the TiN/AlN/WS₂/SiO₂ structure. The microscopy data clearly show that the layered AlN ordering results from crystal WS₂, not the amorphous SiO₂ substrate. Thus, the layered h-AlN growth on the van der Waals surface of TMD materials is VDWE growth. Epitaxial growth also requires a close lattice match between the deposited material and the substrate. In this work, TMD surface free energies were not much different;^31–33 therefore, their lattice mismatch mainly influenced the layered h-AlN film growth on 2D crystals. The in-plane lattice constants of AlN, WS₂, MoS₂, and WSe₂ are 3.13, 3.19, 3.19, and 3.33 Å, respectively, and the in-plane lattice mismatch between AlN and WS₂, MoS₂, and WSe₂ is 1.88%, 1.88%, and 6.01%, respectively. Due to the relatively larger lattice mismatch between AlN and WSe₂, the first layer of h-AlN on WSe₂ was delaminated as a buffer layer in Fig. 1(c). In contrast to the layered h-AlN on TMDs, the growth mode of AlN directly on a sapphire substrate was amorphous, as shown in Fig. 1(d).

To verify these local microstructure observations, we utilized GIWAXS measurement to investigate the low-dimension materials' surface structure and provided structural morphology information over the entire sample.^34–36 Synchrotron GIWAXS was employed to analyze the structured features of the TiN/layered h-AlN/TMD stacks. The measurement geometry of GIWAXS is shown in Fig. 3(a). The sample is illuminated under a grazing-incidence angle, and our coordinate system is such that the x–y plane is the sample plane, and the x-axis lies in the scattering plane. Figures 3(b)–3(d) present the 2D reciprocal space map (on the left) and scattered intensity profile (on the right) of TiN/AlN on WS₂/sapphire, on MoS₂/sapphire, and on WSe₂/sapphire, respectively. As seen from the reciprocal space map, the vertical stripe along the crystal truncation rod (CTR) of these samples shows that the crystalline morphology of layered h-AlN is similar to the layered structure of TMDs underneath. The overlapped stripe pattern of layered h-AlN and Mo(W)S₂ shows that the layered h-AlN crystal growth followed the lateral facet of (10$1¯$0) Mo(W)S₂ in Figs. 3(b) and 3(c). The structural features of MoS₂, WS₂, and WSe₂ by GIWAXS measurement are presented in supplementary material note 1. In contrast, the AlN/WSe₂ sample exhibits two individually streak lines at positions 2.18 and 2.29 of q_r, indicating that the (10$1¯$0) layered h-AlN spacing is smaller than that of WSe₂, although the layered h-AlN crystals are also growing along the [10$1¯$0] direction in Fig. 2(d).

To gain more insight into the layered h-AlN spacing on TMDs where |q| = 2π/d_hkl spacing and d_hkl spacing is the distance between the parallel planes of atoms, one may examine the (10$1¯$0) layered h-AlN spacing of TiN/AlN on WS₂/sapphire, on MoS₂/sapphire, and on WSe₂/sapphire in the scattering profile of Figs. 3(b)–3(d), and they are 2.76, 2.78, and 2.74 Å, respectively. The lateral facet of (10$1¯$0) layered h-AlN spacing on three TMDs is slightly larger than that of the theoretical value of h-AlN (2.71 Å).³⁷ This suggests that the layered h-AlN crystal grown on TMD is slightly tensile strained. The layered h-III-nitride materials have been extensively studied using theoretical calculations. When III-nitride materials become thinner, the internal electrostatic field problem was overcome, resulting in lower formation energy of the hexagonal structure.^38,39 From the thermodynamics point of view, the cohesive energy of a hexagonal structure is lower than that of a wurtzite structure when the thickness of III-nitride is thin and the bonding type changes from the sp³ to sp² hybridization, resulting in the preference of each ion for the trigonal planar configuration.^39,40 Akiyama et al.⁴¹ calculated the relative stability between the hexagonal and wurtzite structures of III-nitride materials as a function of the layer thickness and found h-AlN favored below 13 monolayers. It is consistent with our experimental results of the layered h-AlN structure on TMDs. In addition, the strain effect on h-AlN had been studied by Wu et al.⁴² Tensile strain can substantially stabilize the hexagonal symmetry of the h-AlN multilayer.

Figure 4 presents the microstructure and topography of HfO₂/WS₂/sapphire and HfO₂/layered h-AlN/WS₂/sapphire structures, in which hafnium oxide (HfO₂) was deposited by ALD using tetrakis(dimethylamido)hafnium (TDMAH) and H₂O as precursors at 250 °C with Ar carrier gas. In the HfO₂/WS₂/sapphire structure, the amorphous HfO₂ exhibited a non-uniform granular structure shown in Fig. 4(a) and a HfO₂ surface roughness of about 1.08 nm. In contrast, the amorphous HfO₂ layer was more uniformly grown on the layered h-AlN/WS₂ surface. The layered h-AlN/WS₂ structure was not degraded by HfO₂ growth in Fig. 4(b). The ordered structure of layered h-AlN/WS₂ remained intact, and the GIWAXS data were presented in supplementary material note 2. The shape of grains exhibited a hexagonal pattern in Fig. 4(d), which confirmed our GIWAXS analysis that the layered h-AlN grains followed the underlying WS₂ crystal. More importantly, HfO₂ on the WS₂ sample was granular, while HfO₂ on the layered h-AlN grain surface was uniform. The roughness of the HfO₂/AlN/WS₂ structure is ∼0.17 nm, which is less than that of HfO₂/WS₂ by about six times. These results suggest that the insertion of layered h-AlN as an interfacial layer between the high-k dielectric and pristine TMD channels can solve the coverage and roughness problem of high-k oxide deposition on TMDs. Finally, we theoretically examined the stability of the HfO₂/h-AlN/WS₂ structure with DFT calculations. Figure 5 shows the adsorption energy as a function of the number of h-AlN layers. The adsorption energy between h-AlN and WS₂ (green line) was consistent with the van der Waals interaction,^29–31 which was lower than the HfO₂/WS₂ structure (black dash line), indicating that the interface of h-AlN and WS₂ maintained a certain interaction distance. Compared to the h-AlN/WS₂ interface, the adsorption energy between HfO₂ and h-AlN (red line) is much larger when the number of h-AlN layers was more than a monolayer, i.e., the adhesion of HfO₂ on the h-AlN surface is strong. Importantly, the van der Waals interaction between WS₂ and h-AlN was maintained even with HfO₂ stacked on top of h-AlN (blue line) when the number of h-AlN layers is two or three.

In conclusion, this work clearly demonstrated layered hexagonal-AlN epitaxial deposition on monolayer TMDs by low-temperature PEALD. The growth mechanism of the layered h-AlN on TMD is van der Waals epitaxy and is facilitated by a close lattice match between layered h-AlN and TMDs. Synchrotron-based GIWAXS measurement revealed that layered h-AlN formed a 2D hexagonal lattice followed by the TMD lattice. We also demonstrated ALD HfO₂ deposition on layered h-AlN (as an IL)/TMD while leaving the van der Waals interface between h-AlN and TMD intact. This makes ALD h-AlN an interesting candidate for the gate dielectric or an interfacial layer of TMD transistors. The electrical property of the h-AlN/TMD interface still needs to be studied.

See the supplementary material for the structural features of MoS₂, WS₂, and WSe₂ monolayer, and the HfO₂/h-AlN/WS₂ structure.

This work was financially supported by the “Center for the Semiconductor Technology Research” from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan and also supported in part by the Ministry of Science and Technology, Taiwan, under Grant MOST 111-2634-F-A49-008. The authors thank Dr. Bang-Hao Huang, MSSCORPS CO., LTD, for the HR-TEM analyses.