Exploring Phonon Dynamics of Defective Transition Metal Dichalcogenides (TMDs) using Quantum Mechanics, ReaxFF and Machine Learning Methods
PI: Adri van Duin (Mechanical Engineering, Chemistry, Chemical Engineering, Material Science & Engineering and Engineering Science & Mechanics)
Other senior team member: Vin Crespi, Departments of Physics, Chemistry and Material Science & Engineering
The pre-exist or introduced structural defects in two-dimensional transition metal dichalcogenides (TMDs)1,2, such as vacancies, dislocations, and grain boundaries, dopants, govern the electronic and thermal transport properties of these materials and influence their applications in novel nanoelectronic devices.3,4 Simulating defective systems requires large-scale atomic models at the nanometer scale, making first-principles based quantum mechanical methods computationally prohibitive. Therefore, computational methods capable of accurately simulating lattice dynamics and thermal transport at such scales — especially those that can also handle the reactive dynamics involved in the formation of structural imperfections that may limit thermal transport─are valuable. In semiconducting 2D TMDs, where thermal transport is primarily governed by phonons, empirical molecular dynamics (MD) simulations using a validated force field offer an effective means to study the phonon dynamics and thermal properties of these defective systems. Previously developed ReaxFF reactive force field for MoS2 system mainly focus on reproducing structural and mechanical properties,5–7 synthesis and crystallization dynamics8–10; they do not target phonon dispersion. Recently we developed a new reactive force field for MoS2 using a relatively small set of quantum mechanical calculations with the density functional theory (DFT) method for frozen phonons of pristine 2D MoS2 that significantly improves the accuracy of its phonon dispersion relations while retaining the ability to describe reactive dynamics.11 However, this force field can not accurately reproduce the Raman peak shifts in MoS2 with sulfur vacancies, likely due to the lack of quantum mechanical data for defective states near equilibrium. To address this limitation, we aim to further optimize this force field by incorporating DFT data of frozen phonons of defective MoS2 system. We also plan to extend this force field to include oxygen element, enabling the study of surface oxidation effects. This enhanced force field will allow us to explore the phonon dynamics in defective MoS2 systems, such as those with sulfur vacancies, grain boundaries, oxygen dopants, etc. Later we will also transfer this methodology to other emerging TMDs12–14 and their heterostructures, including defective systems. In this project, graduate students/postdocs/junior researchers would be participating in the ReaxFF force field development, sampling defective systems using machine learning methods,15 and conducting MD simulations for phonon dynamics for defective 2D TMDs.
Expertise/skills of interest:
- Computational experience using VASP/QE, LAMMPS/AMS /JAX.
- Programming experience using python, C++ and/or JAX.
Expectations:
- Post-comps graduate student or postdocs with at least some experience and/or training in: (1) Solid-state Physics/Chemistry or a related field; and (2) Computational physics/chemistry, Data Sciences or a related field.
- Weekly meeting and project updates with faculty advisor(s). Participate in group meetings (~1 hour weekly) and present the results in seminar.
Goal:
The preliminary results could support a future funding proposal to DOE or NSF call. Our longterm goal is to combine reactive MD simulations with machine learning techniques for exploring the distributions of defective systems in experimental measurements, such as Raman spectra and high-resolution transmission electron microscopy to advance the understand of phonon dynamics and structural evolutions of defective TMD systems and pave the way for their applications in nanodevices.
Specific Objectives:
1. Conducting first-principles calculations for force field training and train the force field.
2. Generating defective systems using machine-learning method
3. Performing MD simulations for phonon dynamics using AMS software or JAX
4. Wrapping up the results and publish a peer-review paper
Engagement:
van Duin serves as the director of the Material Computation Center at Penn State and has, in that position, regularly participated in ICDS activities and has served on numerous committees and in leadership roles related to ICDS.