# Fundamentals of Modeling

 Presenter: Colin McAndrew NXP Semiconductors Start: 10:20 (30 minutes)

## Abstract

Compact models (aka SPICE models) are the “tunnel” through which designers, especially of analog, mixed-signal, and RF circuits, “see” the target manufacturing technology for a product. This tutorial gives a 10,000 m overview of how circuit simulation works, which sets the stage to understand how models should be defined. Best practices for model formulation and implementation are summarized. The 5 basic formulations for MOS transistor models are reviewed. What design textbooks teach you about transistor capacitances is wrong: this tutorial explains why, and presents the most useful way to conceptualize, and understand, capacitances. Characterization of relative (i.e. %) difference is a cornerstone of parameter extraction, and underlies quantification of statistical variability both for process spread and for mismatch (which is modeled via the standard deviation of difference). How you were taught to calculate % difference is wrong: this tutorial presents proper metrics, both bounded (for model parameter extraction) and unbounded (for process spreads and mismatch).

## Biography

Compact models (aka SPICE models) are the “tunnel” through which designers, especially of analog, mixed-signal, and RF circuits, “see” the target manufacturing technology for a product. This tutorial gives a 10,000 m overview of how circuit simulation works, which sets the stage to understand how models should be defined. Best practices for model formulation and implementation are summarized. The 5 basic formulations for MOS transistor models are reviewed. What design textbooks teach you about transistor capacitances is wrong: this tutorial explains why, and presents the most useful way to conceptualize, and understand, capacitances. Characterization of relative (i.e. %) difference is a cornerstone of parameter extraction, and underlies quantification of statistical variability both for process spread and for mismatch (which is modeled via the standard deviation of difference). How you were taught to calculate % difference is wrong: this tutorial presents proper metrics, both bounded (for model parameter extraction) and unbounded (for process spreads and mismatch).