Area-selective atomic layer deposition on 2D monolayer lateral superlattices

Area-selective atomic layer deposition (AS-ALD) enables bottom-up patterning by depositing materials only on predefined regions while blocking others. Conventional AS-ALD (CAS-ALD) relies on inert barrier layers to prevent chemisorption and reaction of ALD precursors, but suffers from resolution limits and imperfect selectivity due to physisorption and penetration of small reactive precursors (e.g., trimethyl aluminum) through barriers. Here, the authors observe and harness AS-ALD on a two-dimensional (2D) MoS2–MoSe2 lateral superlattice template. Leveraging lateral growth during CVD of van der Waals materials, the superlattice stripe widths (MoS2 and MoSe2) can be precisely controlled by precursor pulse durations, enabling ultra-high-resolution templates beyond optical diffraction limits. Unlike CAS-ALD, selectivity in this superlattice-based AS-ALD (SAS-ALD) arises from adsorption and diffusion of precursors on chemically inert basal planes (no dangling bonds) of MoS2 and MoSe2, allowing selective deposition in very narrow patterns and compatibility with highly reactive precursors.

The paper situates SAS-ALD within the broader AS-ALD field, citing prior work on area-selective deposition using self-assembled monolayers and inhibitor strategies, and highlighting limitations such as physisorption-induced defects and precursor penetration through barriers that degrade selectivity. It references lateral growth and stitching of 2D TMD superlattices via sequential CVD precursor supply as a means to achieve coherent, atomically thin superlattices with engineered strain and tunable widths. Prior studies on ALD nucleation on TMDs and graphene emphasize the chemically inert basal planes and challenges in initiating deposition, framing the need for a mechanism not reliant on surface reactions.

The approach comprises two main steps: (1) growth of monolayer lateral TMD superlattices by CVD/MOCVD and (2) ALD for selective deposition, supported by advanced characterization and multiscale simulations.

• CVD growth of MoS2–MoSe2 lateral superlattice: Conducted in a horizontal hot-wall quartz tube using MoO3 + KCl solid sources and alternating gas-phase chalcogen precursors: diethyl sulfide (DES) for S and dimethyl selenide (DMSe) for Se. Growth temperature 650–700 °C, pressure ~1 torr, time 30–60 min. A cycle comprises four steps: MoS2 growth (Ar/DES 110/10 sccm, duration ts = 10–300 s), MoS2 purge (2–8 s), MoSe2 growth (Ar/DMSe/H2 110/10/2 sccm, duration tse = 10–300 s), and MoSe2 purge (2–8 s). Widths of MoS2 and MoSe2 regions are tuned from tens to hundreds of nanometers by adjusting durations. Epitaxial growth and alignment achieved on c-plane sapphire; typical growth on Si/SiO2 yields random flake orientations. WS2–WSe2 superlattices were grown by MOCVD using tungsten hexacarbonyl with alternating DES/DMSe at 700–730 °C, <0.1 torr, 2 h.
• ALD processes: Conventional thermal ALD. • Al2O3 at 170 °C using TMA/H2O; pulse times 0.2 s for both TMA and H2O; 10 s Ar purge (100 sccm). • HfO2 at 170 °C using tetrakis(dimethylamido)hafnium(IV) and H2O; pulse times 1.1 s (Hf precursor) and 0.1 s (H2O); 17 s N2 purge (150 sccm). • Ru at 200 °C using Ru(C7H9)(C2H2O) and O2; Ar carrier/purge 500 sccm; O2 500 sccm; 200 cycles of: Ru injection 70 s, Ru purge 5 s, O2 injection 10 s, O2 purge 5 s. • Sb2Se3 at 70 °C using Sb(OC2H5)3 and [(CH3)3Si]2Se; Ar carrier 50 sccm, purge 200 sccm; 50 cycles of Sb inject 2 s, purge 5 s, Se inject 2 s, purge 10 s. • Te at 70 °C using BTMS-Te and Te(OEt)4 bubbled at 40 and 50 °C, delivered with 50 sccm Ar.
• Characterization: SEM for morphology; AFM (contact and non-contact) for surface topography and ripple structures; cross-sectional HAADF-STEM and EDS for compositional mapping and interfaces; EMPAD 4D-STEM for lattice constant and strain mapping; Raman for WS2–WSe2; UV-VIS for Al2O3 bandgap; device fabrication of MoSe2 nanoribbon FET using Al2O3 as hard mask.
• Simulations: MD (LAMMPS) with modified Stillinger–Weber potentials to model ripple formation and strain distribution due to lattice mismatch at low temperature (4.2 K) to minimize thermal deformation; DFT (VASP, PBE+D3) for adsorption energies, reaction energies, and diffusion barriers (CI-NEB, MLNEB/DyNEB); kMC (BKL algorithm) using site-dependent adsorption, desorption, and diffusion rates derived from Gibbs free energies (hindered translator/rotor models in ASE), with linear regression to map energies vs. local strain fields. Occupation maps and collision heatmaps computed at various temperatures to correlate with nucleation behavior.

• SAS-ALD on monolayer MoS2–MoSe2 lateral superlattices enables selective deposition on MoSe2 regions while blocking MoS2, demonstrated for Al2O3, HfO2, Ru, Sb2Se3, and Te.
• Patterning resolution: Achieved sub-10 nm half-pitch; minimum measured pitch 19.7 nm with 16 nm line width for Al2O3; controllable widths and pitches (e.g., width 15–48 nm at ~57 nm pitch; width ~45 nm at pitches 70–155 nm). Selectivity maintained up to 15 nm Al2O3 thickness. Long, uniform line arrays over micrometer scales demonstrated; orientation random on Si/SiO2 but aligned on c-plane sapphire due to epitaxial growth (selective deposition of ~1 nm Al2O3 verified by AFM).
• Initial nucleation: Early-stage (15 cycles) Al2O3 forms periodic islands within MoSe2 that nucleate with sub-20 nm periodicity and merge into continuous lines by ~60 cycles. EDS confirms Al localized on Se-rich (MoSe2) regions with clean interfaces.
• Strain-driven template: MD and 4D-STEM reveal periodic ripples in MoSe2 with alternating tensile crests and compressive valleys; MoS2 remains relatively flat with largely released strain. Periodic compressive sites (reduced lattice constant) in MoSe2 align with preferential nucleation.
• Reaction pathways ruled out: DFT shows direct dissociative reactions of TMA on basal planes are highly endothermic (e.g., ~1.59 eV on MoS2; ~2.03 eV on MoSe2), and H2O reactions even more unstable (~5.09–5.22 eV). At chalcogen vacancies, TMA/H2O adsorption is not strengthened (TMA ~−0.48 eV on vacancies vs. ~−0.66 to −0.67 eV on chalcogen top), and dissociative reactions (TMA 1.09 eV on Vs, 1.38 eV on Vse; H2O 1.22 eV on Vs, 1.29 eV on Vse) are still energetically unfavorable and would favor MoS2 over MoSe2, contradicting experiments.
• Adsorption and diffusion govern selectivity: TMA adsorption stabilizes under compressive strain; adsorption in MoSe2 valleys is >60 meV stronger than in MoS2 flat or MoSe2 crests and further stabilized by concave curvature. Diffusion barrier is lowest in compressed MoSe2 valleys (reported ~−0.09 eV), promoting rapid surface transport into valleys.
• Kinetics from kMC: At 443 K, TMA diffusion is fastest in MoSe2 valleys (~3× vs. MoSe2 crests), while desorption is slowest in valleys and ~10× faster in MoS2 than MoSe2, yielding longest residence times and highest collision frequencies in valleys—consistent with observed nucleation patterns.
• Temperature and geometry dependence: High ALD temperatures (200–170 °C) suppress undesired MoS2 deposition; lower temperatures (140–110 °C) reduce diffusion lengths, increase coverage and collision rates, and cause deposition on MoS2. Increasing the MoS2 stripe width (longer diffusion distance) increases undesired deposition on MoS2.

The findings demonstrate a new AS-ALD paradigm on 2D lateral superlattices where selectivity arises from precursor physisorption and surface diffusion kinetics rather than chemisorption reactions. The MoSe2 valleys, created by periodic compressive strain and curvature, act as kinetic basins that concentrate TMA via stronger adsorption, faster diffusion into the valleys, and slower desorption, leading to preferential nucleation and growth only on the MoSe2 stripes. This mechanism explains the observed periodic initial nucleation, the strict selectivity even for highly reactive TMA, and the dependence on temperature and MoS2 width. The approach achieves sub-10 nm half-pitch patterns and is broadly compatible with multiple materials and TMD systems (including WS2–WSe2), offering a bottom-up route to nanoscale patterning beyond the limits of CAS-ALD and optical lithography. These results are relevant to advanced nanoelectronic device integration where precise placement of dielectrics and metals on 2D semiconductors at extreme scales is required.

This work introduces superlattice-based AS-ALD (SAS-ALD) on monolayer 2D lateral TMD superlattices, achieving selective deposition on MoSe2 with sub-10 nm half-pitch resolution. The selectivity mechanism is governed by differences in adsorption, diffusion, and desorption of ALD precursors driven by periodic strain and curvature, rather than surface reactions or defects. The method supports selective deposition of oxides and metals (Al2O3, HfO2, Ru, Sb2Se3, Te) and is extendable to other lateral superlattice systems (e.g., WS2–WSe2). The authors anticipate further downscaling and envisage integration of 2D/3D nanoscale device architectures, such as TMD nanoribbons and short-channel devices with selectively deposited gate oxides and contacts. Future work should focus on optimizing the materials properties of AS-ALD films, improving global alignment (e.g., via epitaxy), expanding precursor sets, and refining process windows for robust manufacturability.

Selectivity degrades at lower ALD temperatures (e.g., 140–110 °C) due to reduced diffusion lengths and increased residence/collision on MoS2, leading to undesired deposition. Wider MoS2 stripes increase the diffusion distance and promote nucleation on MoS2. On amorphous Si/SiO2 substrates, superlattice flakes are randomly oriented, yielding non-aligned line patterns; epitaxial sapphire is needed for aligned arrays. CVD-grown TMDs contain defects (e.g., chalcogen vacancies), though defect-mediated reactions were shown not to govern selectivity. Simulation constraints include MD conducted at low temperatures and DFT models on subsets/idealized strain fields, which may not capture all experimental complexities, though trends align with observations.