Intermediates of Forming Transition Metal Dichalcogenide Heterostructures Revealed by Machine Learning Simulations

1. Problem Statement

Two-dimensional (2D) transition metal dichalcogenide (TMD) van der Waals heterostructures (vdWHs) are promising for next-generation electronics, optoelectronics, and nonlinear optics. However, their large-scale synthesis faces two critical challenges:

• Size Constraints: Mechanical assembly methods produce high-quality interfaces but cannot achieve wafer-scale dimensions.
• Alloying Contamination: Chemical Vapor Deposition (CVD) methods can grow large areas but often suffer from unintended alloying (mixing of metal atoms, e.g., Mo and W) at the interfaces, degrading the electronic properties of the heterostructure.

Recent experimental studies reported a two-step vapor deposition method to grow wafer-sized TMD vdWHs with minimal impurities. However, the atomic-scale mechanism governing the growth, specifically the stability of intermediate states and the origin of alloying, remained unclear. Traditional Density Functional Theory (DFT) is too computationally expensive for the large-scale, long-time simulations required to model growth dynamics, while classical Molecular Dynamics (MD) lacks the accuracy to describe complex bond breaking and formation.

2. Methodology

The authors developed a hybrid computational approach combining Machine Learning Potentials (MLP) with Molecular Dynamics (MD) to bridge the gap between accuracy and scale.

• MLP Development:

  • Architecture: Utilized NequIP, an E(3)-equivariant graph neural network, trained on extensive DFT data.
  • Dataset: Constructed a diverse dataset of ~26,000 configurations, including pristine TMD layers, MoS2/WS2 vdWHs, metal clusters, alloy configurations, and growth intermediates.
  • Training Strategy: Employed an iterative learning process. The initial set was generated via "on-the-fly" MD simulations, followed by successive rounds of sampling specific structural motifs (e.g., metal-sulfur exchanges, alloying, sulfurization) to refine the model.
  • Validation: The MLP achieved high accuracy with a Root Mean Square Error (RMSE) of 10.6 meV/atom for energies and 151 meV/Å for forces, comparable to DFT but at a fraction of the computational cost.

• Simulation Setup:

  • Software: Large-scale Atomic/Molecular Massive Parallel Simulator (LAMMPS) coupled with the trained MLP.
  • Conditions: Simulations were conducted at growth-relevant temperatures (900 K – 1100 K) using the NVT ensemble.
  • Processes Modeled:
    1. Deposition of bare metal atoms (Mo) onto TMD surfaces (MoS2 and WS2).
    2. Formation and stability of intermediate structures.
    3. Sulfurization of these intermediates to form final bilayer structures.
    4. Electronic property calculations (DFT) of the resulting interfaces.

3. Key Contributions

• Discovery of Metastable Intermediates: Identified a crucial, previously unobserved intermediate structure termed SMMS (where M = Mo or W). In this structure, a layer of bare metal atoms spontaneously sinks beneath the top sulfur layer of the TMD substrate, forming a sandwich-like configuration (S-M-M-S).
• Mechanism of Alloying: Revealed that the SMMS intermediate facilitates the exchange of metal atoms (Mo and W) between the upper and lower layers. This atomic swapping is driven by configurational entropy and stress relief, leading to the formation of alloyed TMDs rather than pristine heterostructures.
• Prevention Strategy: Demonstrated that bare metal atoms are the root cause of sinking and alloying. Conversely, metal-sulfur (Mo-S) clusters remain on the surface and do not sink. This explains why sulfur-rich conditions in experimental CVD/MOCVD processes successfully prevent alloying.
• Device Application: Characterized the electronic properties of the SMMS intermediate, identifying it as a metallic phase with a low Schottky barrier height (SBH) suitable for use as an electrode in MoS2 field-effect transistors (FETs).

4. Key Results

A. Deposition Dynamics and Intermediate Formation

• Instability of Bare Metals: When bare Mo atoms are deposited on MoS2 or WS2, they are thermodynamically unstable on the surface. Within picoseconds, they sink into the sulfur layer, releasing significant energy (~1.45 eV for Mo on MoS2).
• SMoMoS and SMMS Structures:
  • On MoS2, sinking Mo atoms form SMoMoS.
  • On WS2, sinking Mo atoms form SMoWS, which subsequently undergoes Mo-W exchange to form the alloyed SMMS structure.
  • Phonon Analysis: Confirmed that SMoMoS and SMMS are dynamically stable (no imaginary frequencies), whereas the surface-adsorbed metal layer is unstable.
• Alloying Mechanism: In the SMMS structure, Mo and W atoms exchange positions. Free energy calculations show that a uniform distribution of Mo and W (alloyed state) minimizes free energy due to mixing entropy and stress alleviation.

B. Sulfurization and Growth Control

• Sulfurization of SMoMoS: When sulfur is deposited on the SMoMoS intermediate, it pulls Mo atoms back to the surface, reforming a clean bilayer MoS2.
• Sulfurization of SMMS (Alloyed): When sulfur is deposited on the alloyed SMMS, it pulls out both Mo and W atoms indiscriminately, resulting in an alloyed bilayer (MoxW1-xS2) rather than a clean heterostructure.
• Role of Sulfur-Rich Conditions: Simulations of Mo-S clusters (Mo-S, Mo-S2, Mo-S3) showed that these clusters do not sink into the substrate. They remain on the surface and diffuse, facilitating nucleation.
  • Conclusion: The success of the experimental two-step growth method is attributed to residual sulfur in the chamber reacting with deposited metal atoms to form Mo-S clusters before they can sink, thereby suppressing the formation of the alloying-prone SMMS intermediate.

C. Electronic Properties

• Metallic Nature: Both SMoMoS and SMMS exhibit metallic band structures.
• Schottky Barrier Height (SBH):
  • The interface between these metallic intermediates and semiconductor MoS2 forms a p-type Schottky contact.
  • SBH Values: SMoMoS-MoS2 has an SBH of 0.55 eV, and alloyed SMMS-MoS2 has an SBH of 0.69 eV.
  • Comparison: These values are significantly lower than those of conventional metal contacts (Ti, Au, etc.) which often suffer from strong Fermi-level pinning. The SMMS interface shows reduced Fermi-level pinning, suggesting it acts as an ideal low-resistance contact.

5. Significance

• Theoretical Insight: The study provides the first atomic-scale explanation for the formation of alloyed TMDs during vapor deposition and clarifies why specific experimental conditions (sulfur-rich) are necessary to prevent it.
• Process Optimization: It offers a clear guideline for synthesizing high-quality, non-alloyed TMD vdWHs: prevent the adsorption of bare metal atoms by ensuring a sulfur-rich environment during deposition.
• Device Engineering: The discovery that the SMMS intermediate is a stable, metallic phase with low Schottky barriers suggests a new strategy for creating atomic-layer bonding (ALB) contacts in 2D electronics. This could overcome the contact resistance limitations in TMD-based transistors, enabling high-performance integrated circuits.
• Methodological Advance: The successful application of NequIP-based MLPs to simulate complex, non-equilibrium growth processes of 2D materials demonstrates the viability of machine learning for solving materials synthesis problems that are intractable for DFT alone.