MOSFET statistical parameter extraction using multivariate statistics

Abstract

A methodology for the generation of MOSFET device model parameter sets which reflect measured device performance variations is described and assessed for its accuracy and suitability in predicting actual circuit variations. The proposed scheme is based on the principal component method of multivariate statistical techniques and utilizes Monte Carlo simulations. Comparisons between the predictions of device and circuit characteristics and measured characteristics over a wafer lot are shown and discussed. It is suggested that the techniques used are most suitable for the prediction of the measured distributions of precision analog circuits rather than large digital circuits where 25 or more circuit simulations may be totally unacceptable because of the amount of CPU time required.<>