Kinetic Monte Carlo Simulations on Self-organization of Nanostructures Accelerated by Massive Parallelization

Modern graphics processing units (GPUs) currently provide the most peek processing performance regarding both cost and energy consumption. This is achieved by mounting large numbers of simple cores rather than a few complex ones. The characteristics that come with this design demand a large degree of data-parallelism from applications. Thus, new approaches are needed for parallelizing tasks that are not by nature data-parallel.
The 3D kinetic lattice Monte Carlo (KLMC) method is a means of performing atomistic simulations of self-organization processes in solids at by far larger scales than those accessible via molecular dynamics (MD). This method has been implemented for GPUs, achieving up to 70 times higher performance than the sequential reference implementation on a single core of a modern CPU. This enables atomistic simulations at even larger scales, even putting space and time scales comparable to the experiment within range.
The new program has been shown to be useful to study the phase separation in large binary systems. This was done with an application for third generation photovoltaics cells in mind which is a subject of a current BMBF project. It was also applied to out-of-equilibrium problems, backing up a theory of inverse Ostwald ripening (IOR) from an angle that was not previously looked at.