A new method supporting the nominal design of analog integrated circuits with regard to constraints

The essential feature of the presented method is the characteristic boundary curve (CBC), which represents the potential error reduction as a function of the parameter correction for one iteration step. The CBC in the unconstrained case has been described by Antreich et al. (1988) but most practical optimization problems include constraints. Therefore, this paper presents a definition of the CBC for problems with constraints and provides a new algorithm for the calculation of the CBC. This algorithm is more accurate and allows the calculation of the CBC for unconstrained problems as well as for constrained problems. A second new algorithm is also described. It calculates a point on the CBC corresponding to a given parameter correction norm. This new algorithm is faster, more accurate, and additionally deals with constraints. Another new property of the CBC is described: The CBC shows the influence of constraints on the optimization process. The algorithms presented in the paper have been implemented in the program system CANDI (Circuit level Analog Design system for Integrated circuits), which is now tested in several installations in industry.<
>