Deep Learning Models for Materials Simulations and Experiments

Friday, April 12, 2019, 11:00 am - 12:00 pm PDTiCal
This event is open to the public.
AI Seminar
Rajiv Kalia, USC Computer Science Department
Video Recording:


This presentation will focus on the application of deep learning models to massively parallel quantum and reactive molecular dynamics simulations of two-dimensional (2D) layered materials and electron microscopy data.  Electrical and optoelectronic properties of 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 and metallic phases via novel intermediate structures which have been observed by scanning transmission electron microscopy. Quantum simulations show that interfaces synthesized by deep learning models are stable and the devices with those interfaces are suitable for novel nanoelectronics applications.


Rajiv Kalia is a professor of computer science, chemical engineering, and materials science in the USC Viterbi School of Engineering, and a professor of physics and astronomy in the USC Dornsife College of Letters, Arts, and Sciences.  His multidisciplinary research concentrates in the areas of biophysics and materials science.  The Kalia group has performed the largest atomistic simulations,including 109-atom molecular dynamics simulations on IBM Blue Gene/Q. Under the aegis of a DOEMaterials Genome Center, the Kalia group is currently involved in computational synthesis and characterization of novel layered material architectures.

« Return to Upcoming Events