The purpose of running experiments is to find out something that is not known, for example how the electrical characteristics of a semiconductor device behaves when the dose of an implant dose is changed. In general, experiments are designed to show the dependencies of the input and output parameters based on a few observations. Design of Experiments methods [29] are a methods for observing the system behavior with only a small number of required experiment points.
In nature experiments are not perfectly repeatable, because they are usually effected by factors not considered in in the inputs, like environmental properties. For these experiments the results vary from trial to trial, so each point should be observed several times.
Simulations instead of real experiments usually give reproducible results so one observation does not have to be repeated. In some kind of simulators random numbers are very important like Monte Carlo methods [22] or some grid generators [46]. If these simulations are well justified the numerical noise in the results, using the same start values, is negligible.
During the planing stage of a set of experiments the ranges in which the input parameters, also called factors, are changed have to be defined. The layout of the design specifies how to choose the sample points in the selected subspace of the input parameters. Depending on the ranges the observed model and the analysis of the results different algorithms, also called experimental designs, can be used.