Our work in this area is concerned with the study of basic mechanisms of diffusion, reaction, and aggregation in complex environments such as external fields and compositional variations. The aim is to be able to accurately describe these processes in simulation tools. While molecular dynamics (MD) simulations only require an interatomic force-field as input and therefore are able to predict mechanisms, coarser approaches, such as continuum rate equation (CRE) modeling and kinetic Monte Carlo (KMC), require that all the relevant physical and chemical mechanisms be explicitly specified a priori. Coarse-grain simulations are necessary to access length and time scales of practical interest.
We have developed an “overlap” multiscale computational framework in which molecular dynamics simualtions are used to generate “experimental data” to which coarser simulations can be fitted. The advantage of this approach is that the MD-generated data comes from a material that is fully specified – i.e. we know exactly all of its physical properties becuase they can be computed using the same force-field used to generate the experimental data! In this way, the only uncertainties in the coarse model are mechanistic, which can be subsequently probed in detail by the two-scale comparison. A schematic of this approach is shown in Figure 1:
Figure 1: Multiscale overlap approach for investigating nanoscale phenomena.
One application example is shown in Figure 2, which demonstrates how a continuum model of vacancy aggregation in silicon can be constructed to quantitatively reproduce the dynamics predcited by a large-scale MD simulation, in which 1000 vacancies in a 216,000-atom lattice were allowed to evolve under conditions of strong supersaturation. The parametrically consistent comparison led to the discovery of previously ignored microscopic physical phenomena such as surprisingly large cluster mobilities and cluster shape fluctuations that dramatically increase the capture radius for aggregation. In subsequent comparisons to experimental data, these additional physical components were found to greatly enhance the quantitative accuracy and predictive capability of continuum models for void evolution during silicon crystal growth.
Figure 2: Parametrically consistent comparison between MD and continuum model predictions for vacancy aggregation.