Design space exploration for embedded systems using co-simulation

Abstract

In the development of embedded systems, co-simulation permits the systematic exploration of the design space with the aim of selecting models of optimal solutions, and allowing trade-offs between computing and physical elements of alternative designs. We present an approach to such design space exploration using experiment design. Beginning from a classification for the parameters that one may wish to vary during an experiment, we discuss methods for reducing the number of experiments performed, from screening experiments to the use of orthogonal matrices or a space-filling search. Two case studies which illustrate the importance of careful experiment design in this context. We finally discuss ways in which experimental results may be ranked to permit automatic selection of the best designs analysed.