A Unified Treatment of Yield Analysis, Worst-Case Design and Yield Optimization

Abstract

<p>This thesis addresses itself to what is considered to be one of the most general theoretical problems associated with the art of engineering design. A unified treatment is presented of production yield evaluation, worst-case design and yield optimization. The formulation is suited to nonlinear programming methods of solution.</p> <p>Viewed in its entirety the approach integrates the following concepts: design centering, assignment of component tolerances, post-production tuning, yield estimation for realistic distributions and modeling of response functions. Many of the ideas can also be used separately depending on the type of design evaluation required, the number of degrees of freedom involved and the availability and properties of suitable simulation programs.</p> <p>The thesis presents an analytical approach to yield and yield sensitivity evaluation. Basic to the approach is the discretization of the distributions by use of orthotopic cells to which suitable uniform distributions are applied. Multidimensional polynomials provide approximations to actual functions, which may be expensive to compute. Algorithms for updating and evaluating these polynomials are developed to permit efficient use of gradient optimization methods.</p> <p>Industrially oriented design examples are furnished to justify the theory. A telephone channel (lossy) bandpass filter is considered with relative insertion loss specifications to illustrate the analysis of yield. The cascade connection of nonideal, inhomogeneous sections of rectangular waveguides is considered from the worst-case design point of view. A current switch emitter follower involving transistors, a diode and a transmission line provides a challenging example for yield optimization including parameter correlations.</p>