Deep learning of CVD growth and phase-transition pathways in layered materials

Conference Dates

March 10-15, 2019


Electrical and optoelectronic properties of two-dimensional (2D) transition metal dichalcogenides (TMDCs) can be tuned by exploiting their polymorphism. Here, polymorphism in TMDCs and their growth by chemical vapor deposition (CVD) are examined using deep generative models namely, the variational autoencoder (VAE) and Restricted Boltzmann Machine (RBM), trained with molecular dynamics (MD) simulation data. The VAE correctly identifies pathways connecting the semiconducting (2H) and metallic (1T) phases via novel intermediate structures called a and bin MoWSe2 alloy. These intermediate structures are similar to those observed in a 2D MoS2 by scanning transmission electron microscopy. Structures generated solely by VAE are combined to form novel MoWSe2 devices with a and b interfaces. Quantum simulations based on density functional theory show that interfaces synthesized by VAE are stable and the devices with those interfaces are suitable for novel nanoelectronics applications. A novel deterministic RBM algorithm is used to identify 2H and 1T phases and defects in reactive MD simulations of the synthesis of a MoS2 monolayer by CVD. Applications of the deterministic RBM approach to modeling of experimental data will also be presented.

* This work was supported as a part of the Computational Materials Sciences Program funded by the U.S Department of Energy, Office of Science, Basic Energy Sciences, under Award Number DE-SC0014607.

This document is currently not available here.