Combined simplex-trust-region optimization algorithm for automated IC design

One of the basic algorithms employed in IC design automation is parametric optimization. It is used when a minimum of a so-called cost function is sought. There exist many optimization algorithms. One of the most successful ones is the Box simplex algorithm. The main drawback of this algorithm that was observed in practice is its slow progress in the last stages of the search. The paper addresses this issue by combining it with a trust-region based algorithm. The combined algorithm was implemented in the SPICE OPUS tool for circuit optimization and tested on several real-world IC design problems. The results show that the combined algorithm exhibits faster convergence and in most cases results in a lower final cost function value than the Box simplex algorithm.