Reliable worst-case tolerance design of feedback regulated DC-DC converters by evolutionary algorithms and interval arithmetic

In this paper a reliable approach to the tolerance design of electronic circuits is presented. Firstly, the tolerances of parameters are maximized by means of an evolutionary algorithm according to the design constraints. This optimization step is based on vertex analysis: each candidate solution is considered as feasible if the design constraints are fulfilled in all of the vertices of the tolerance region corresponding to that solution. This approach ensures fast computation and good results provided that the region of acceptability is convex and simply connected. Otherwise, the actual feasibility of the designed tolerance region is verified. To this purpose, a recursive divide-and-conquer algorithm based on interval arithmetic is used. It allows detecting possible infeasibility pockets included in the designed tolerance region and neglected during the optimization step. Any portion of the boundaries of the region of acceptability that is included in the tolerance region is identified and the analysis is refined across it. The technique proposed in the paper has been applied to the tolerance design of two different possible realizations of the feedback control network of a voltage-mode regulated DC-DC converter. The results show that the method guarantees a considerable enlargement of the parameters' tolerances up to the boundary of the region of acceptability. Moreover, a reliable and efficient validation of the designed tolerance region is performed even in presence of nonconvex and not simply connected region of acceptability.