Statistcal modeling of large analog integrated circuits and its application to design for quality and manufacturability

Abstract

The objective of this dissertation is to develop new and efficient methodologies for statistical modeling of analog integrated circuits and apply them to practical problems. Through appropriate statistical modeling, the Design for Quality (DFQ) of large analog integrated circuits can be realized. Various problems related to the statistical modeling of analog integrated circuits are discussed and solutions are presented. Several topics are included: statistical IC model parameter extraction, statistical CMOS device modeling, statistical modeling of analog functional blocks, global optimization for nominal parameter extraction, and related applications. Three major contributions of the dissertation are: (1) A new and very accurate methodology is developed for the statistical IC model parameter extraction. (2) A new adaptive, multi-level strategy for the statistical modeling of analog functional blocks is proposed. (3) An efficient heuristic global optimization algorithm is implemented for the nominal parameter extraction and circuit optimization. Several applications using the proposed methods are discussed, together with the software implementation. The proposed methodology for statistical IC model parameter extraction solves the fundamental problems and difficulties existing in the traditional optimization- based method. In the proposed method, the accuracy of statistical parameter extraction is well under control, and the desired features of "uniqueness" and "repeatability" are realized through the developed technique called recursive inverse approximation. The proposed strategy for the statistical modeling of analog functional blocks uses an adaptive selection method to construct the statistical models with increasing complexity. In the proposed global optimization algorithm, heuristic cluster analysis and adaptive random reflection are combined to reduce the number of function evaluations and to maintain the global searching ability. A program called SMIC has been developed by implementing the methods proposed in the dissertation. SMIC can be used for various statistical IC modeling problems. Some application examples are given to demonstrate the proposed methods.