Approximation effects and user-controllable design space exploration

Abstract

System Design Space Exploration (DSE) is useful in finding new designs or architectures however often generates an untenably large space that is both unintuitive and difficult to analyze. When adding restrictive rules with human experience and approximations, some generated designs may not be realizable, but users cannot distinguish them from the realizable ones. The extent of the negative influences of these approximation effects is unknown in the systems engineering area. This paper provides a thorough analysis to this problem and studies the intrinsic meaning of approximation in DSE, theoretically. We provide a rigorous design method for complex DSE systems allowing quantitative evaluation of the impact from approximations. Traditional DSE methods have been reconstructed for generation and navigation of the design space of user-controllable sizes without unknowingly generating erroneous designs. Results, including theory, algorithms, and a tool are demonstrated using an example electrical power system design.