Conclusions

Abstract

Three main conclusions are drawn from this work:

• First of all, the EA has a very large potential within VLSI CAD. The problems encountered in this field are extremely complex, which is exactly the situation in which the performance of the EA compares best to that of other methods. For some of the VLSI CAD problems considered in this book the presented EA approaches define the state-of-the-art.

• To make an impact in the CAD field, it is crucial that certain practical guidelines for performance evaluation are followed. It will always be difficult, and often impossible, to provide an absolute fair comparison of two algorithms with very different characteristics. However, applying the guidelines from Section 6.5 is likely to eliminate at least some of the problems often occurring, and will in any case improve the reputation of the EA in the CAD world.

• The main characteristic of the algorithms reviewed in this book is that they apply problem-specific knowledge in one way or another. Combinations with other heuristics, problem-specific representations and/or operators are the rule rather than the exception. This strongly suggests that incorporation of problem-specific knowledge is necessary for the EA to be competitive to the best existing approaches. Therefore, the traditional “pure” EA will most often have to be abandoned when competitive performance is the main objective.