Authors: N. Lu
Affilation: IBM, United States
Pages: 566 - 569
Keywords: statistical model, skewing in compact model, corner model
We present an innovative approach to generate an optimal skew corner of a compact device model (e.g., a Spice model) for a single performance target, and to generate a common/optimal skew corner of a compact model for multiple performance targets. Each statistical parameter of the model is properly skewed within its tolerance range in such a way that the model will reproduce the fast or slow corner results of one or multiple performance targets while maximizing the joint probability density. The existance conditions of a common/optimal corner are analyzed. The corner model problem of N targets and M statistical model parameters is analyzed based on the sensitivities in the first-order approximation. Explict equations are given, and explicit corner model solution are provided. After obtaining the first-order solution, we further construct a 2nd-order (i.e., quadratic response surface) approximation to the performance targets. Then, we use the above first-order solution as the starting point of an iteration solution process to solve a set of non-linear (say, quadratic) equations. In this approach, we obtain a set of more accurate corner solution.