Speaker
Description
We present a multiscale method for the simulation of thin film deposition processes, which couples atomistic simulation methods with event-based stochastic modelling. Our approach aims at physical vapor deposition (PVD) processes, but extends to chemical vapor deposition (CVD), atomic layer deposition (ALD) processes, and possibly liquid-phase deposition and etching processes.
Popular methods for the simulation of thin film deposition processes include Molecular Dynamics (MD) and Kinetic Monte Carlo (KMC) methods. MD offers atomistic accuracy in nanometer-scale simulation spaces, while KMC allows for large-scale simulation at reduced accuracy, usually on fixed grids with a limited set of transition events.
Our approach couples KMC and MD methods, yielding a multiscale method for with atomistic accuracy in micrometer-scale structures. The overarching KMC simulation locates and selects adsorption events in a global domain, which are then realized by concurrent MD simulations acting on local subcomains of the global domain. Efficiency is further improved by applying homotopy theory and graph theory to allow for thousands of active MD workers.
Our method is composed of three main components:
-
A general-purpose direct KMC algorithm, which we have designed to be efficient for large numbers of adsorption events with low acceptance/rejection ratios
-
A hierarchical invocation of MD simulations of individual adsorption events in exclusive subdomains in order to preserve chronology and handle rejections
-
A surface-affine domain decomposition, which represents the surface and active simulation regions. We maintain the global surface as non-overlapping partial Alpha shape boundaries on overlapping subdomains. This allows for graph-based handling of adsorption events by the KMC algorithm while maintaining the exact atom positions for the MD simulations
We demonstrate the benefits of our method on micrometer-scale film deposition simulations on a nanostructured substrate, which was previously impossible using existing methods.
| Speaker Country | Germany |
|---|