Molecular Dynamics Computing System Based On FPGA

FPGA-based acceleration of Molecular Dynamics (MD) simulation, which is widely used in biology, materials science, has been the subject of several recent studies due to its combination of the advantages of both ASICs and general processors. According to the property of MD simulation, this paper gives an overview on FPGA-based MD simulations, and to explore the feasibility of parallel computing MD simulations system based on FPGA.In the simulation of MD system, the force between particles, which are often partitioned into short-range and long-range components, consume longer time that directly impact efficiency of simulations, so these forces can be accelerated by hardware. Since the long-range component can neglect in some situations such as simple fluid calculation. This paper proposes the acceleration of Lennard-Jones (short-range) force calculation, and presents a parallel system using FPGA. The overall design of MD system consists of a PC working as a host and FPGA configured as a coprocessor. The LJ force computation is performed on the FPGA with all units organized as a pipeline and the remaining computations are computed by the host. The data is transferred between the host and the FPGA by a PCI communication board which is specially designed. The LJ force decrease when the distance between two particles gets longer and longer. Because of this property, this system use the following methods: cell-list method, combining discrete function and interpolation for computing high power to avoid square root operation, and using Filter and LJ force Calculators to improve the performance of the system.The whole design is developed using Verilog HDL language under the simulating software environments ISE 10.0 from Xilinx. This system is implemented on a Xilinx FPGA Virtex-II Pro hardware platform, and achieves substantial speedup over production software.