Molecular dynamics (MD) is an important technique in computational nanotechnology. It allows investigation of the dynamical behaviour of nanostrutures and nanodevices using a fully atomistic approach. However, MD is computationally intensive, particularly for large atom counts and long time scales. We report results of our efforts to accelerate and optimize the available Brenner reactive hydrocarbon potential implementation and our Stillinger-Weber potential implementation. Acceleration techniques investigated include neighbour lists, cell-based spatial decomposition (link-cells), fixed bond networks, vectorization using the SSE2 instructions available on Pentium 4 CPU, interaction cutoff distance tuning for the Stillinger-Weber potential and parallelisation using MPI. We have been using these codes to simulate diamond rods of up to 100,000 atoms for 200,000 timesteps (100 ps).
Department of Computer Science and Information Technology, RMIT
GPO Box 2476V, Melbourne, VIC 3001 AUSTRALIA
Phone: +61 3 9925 3207 Fax: +61 3 96621671