FPGA-accelerated molecular dynamics

Abstract

Molecular dynamics simulation (MD) is one of the most important applications in computational science and engineering. Despite its widespread use, there exists a many order-of-magnitude gap between the demand and the performance currently achieved. Acceleration of MD has therefore received much attention. In this chapter, we discuss the progress made in accelerating MD using Field-Programmable Gate Arrays (FPGAs). We first introduce the algorithms and computational methods used in MD and describe the general issues in accelerating MD. In the core of this chapter, we show how to design an efficient force computation pipeline for the range-limited force computation, the most time-consuming part of MD and the most mature topic in FPGA acceleration of MD. We discuss computational techniques and simulation quality and present efficient filtering and mapping schemes. We also discuss overall design, host-accelerator interaction and other board-level issues. We conclude with future challenges and the potential of production FPGA-accelerated MD.