FPGA-based acceleration of CHARMM-potential minimization

Abstract

Energy minimization is an important step in molecular modeling, with applications in molecular docking and in mapping binding sites. Minimization involves repeated evaluation of various bonded and non-bonded energies of a protein complex. It is a computationally expensive process, with runtimes typically being many hours on a desktop system. In the current article, we present acceleration of the energy evaluation phase of minimization using Field Programmable Gate Arrays. We project a multiple orders-of-magnitude speed-up over a single CPU core and a factor of 8 speed-up over our previous acceleration using an NVIDIA Tesla 1060 GPU.