Statistical Circuit Design: The Optimum Assignment of Component Tolerances for Electrical Networks

The prediction of manufacturing yield of an electric circuit, given the tolerances and statistics of the components, is a straightforward procedure using Monte Carlo tolerance analysis. The inverse problem, namely, specifying the component tolerances that produce the cheapest network, is more difficult and has received little attention in the literature. In this paper an algorithm is presented that solves this problem for a significant class of networks. The algorithm is limited to circuits for which one hundred percent yield is sought and whose components are statistically independent, i.e., discrete circuits. The one hundred percent yield assumption is used to reduce the number of possible tolerance choices. A variation of the branch and bound strategy that is shown to be particularly efficient is then used to make an optimum selection. Two-dimensional performance contours, worst-case, and Monte Carlo computations are also used as part of the procedure. An example circuit is demonstrated for which the algorithm produced a tolerance assignment substantially cheaper than that projected by the designer.