A study of statistical variability-aware methods

Abstract

Conventionally circuit performance variability is analyzed with Monte-Carlo simulation and design corner analysis. On the other hand, statistical methods such as design of experiments (DoEs), response surface modeling (RSM) and compact modeling (CM) can achieve a better trade-off between simulation efficiency and accuracy. This paper investigates these variability-aware analysis methodologies. Based on industry standard BSIM4 compact model, selected physical parameters are applied to DoE-RSM and CM methods. Methodologies are validated with both analog (op-amp) and digital circuits (flip-flop) at 65 nm node. A 3X speed up is achieved with DoE-RSM. A proper selection of CM parameters is critical to model accuracy.