Date of Award:

5-2009

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Electrical and Computer Engineering

Advisor/Chair:

Brandon Eames

Abstract

The world of nano-science is an ever-changing field. Molecular Dynamics (MD) is a computational suite of tools that is useful for analyzing and predicting behaviors of substances on the molecular level. The nature of MD is such that only a few types of computations are repeated thousands or sometimes millions of times over. Even a small increase speedup or efficiency of an MD simulator can compound itself over the life of the simulation and have a positive and observable effect. This thesis is the end result of an attempted speedup of the MD problem. Two types of MD architectures are developed: a dynamic architecture that is able to change along with the computational demands of the system, and a static architecture that is configured in terms of processing elements to be best suited to a variety of computational demands. The efficiency, throughput, area, and speed of the dynamic and static architectures are presented, highlighting the improvement that the dynamic architecture presents in its ability to provide load balancing.

Share

COinS