Parametric Yield Maximization

Abstract

As indicated in the previous chapter, parametric yield is a function of the nominal values of the process controls and or the layout. Thus we can consider choosing a nominal design point that maximizes the number of circuits which meet a given performance specification. This approach to yield maximization is called design centering. Prior to 1977, Monte Carlo methods [48], [28], albeit crude and expensive, had been the technique primarily used for design centering. Bandler et al.. [6] then proposed a design centering procedure that was directly related to classical nonlinear programming methods [33]. This approach was based upon selecting for each individual problem, a scalar objective functional which when minimized, inscribed a hypercube in the “feasible region” R of the n-dimensional parameter space.² However, the nonlinear programming approach does not attempt to approximate the boundary of the feasible region, but only determines points on this boundary. As will be seen below, an explicit approximation of the boundary of the feasible region is useful for solving a variety of other statistical design problems [89, 121, 16] as well as allowing one to obtain a Monte Carlo estimate of the yield at a reduced computational cost.