Molecular Diffusion Simulation on ARUZ – Massively-parallel FPGA-based Machine

In this paper, the implementation of a molecular diffusion model named Dynamic Lattice Liquid (DLL) is discussed. Two algorithms, sequential and parallel, implementing this model are examined. The presented theoretical considerations indicate some limitations of the sequential approach. The proposed parallel algorithm overcomes these limitations, but it requires dedicated hardware to function. An example of such hardware, the FPGA-based HPC system named ARUZ, is presented.

I. Introduction – DLL Model

Simulations based on DLL (Dynamic Lattice Liquid) model are devoted and widely used to the analysis of phenomena in molecular environments [1] , [2] . DLL quite precisely predicts the results of the investigated phenomena, but also properly models the dynamics of simulated processes by accurately imitating the behaviour of the analysed matter. The simulated physical substance (e.g.: a polymer immersed in a solvent) is modeled by a set of nodes representing elements of this substance (e.g.: atoms, particles, monomers) or empty regions, where such elements can enter during simulation. It can be metaphorically said that the elements are “placed in” nodes and can “jump” among them. The nodes are logically positioned in the sites of face-centered cubic (FCC) lattice, chosen due to the highest packing factor among regular lattices [3] . The FCC lattice has a coordination number Z = 12, which means that each node is connected to 12 neighboring nodes. As a consequence, in DLL simulations, each node directly interacts with the 12 nearest nodes. DLL does not involve long-distance interconnections. As a consequence, in DLL simulations, each node directly interacts with the 12 nearest nodes. This limitation is actually the source of DLL efficiency, giving – for a wide range of applications – a satisfactory modelling precision.