A linear statistical FET model using principal component analysis

An important issue in statistical circuit design, other than the algorithms themselves, is the development of efficient, statistically valid element models. The authors first discuss what features are needed for a good statistical model. The standard FET model is shown to be difficult to use in a statistical simulation, due to the nonlinear relation between FET S-parameters and model parameters. A linear statistical FET model is then proposed that is based on principal component analysis. This linear model gives uncorrelated model parameters. In an example using measured S-parameter data from ninety 0.5-μm GaAs FETs, 13 uncorrelated model parameters were needed to model the data from 1 to 11 GHz and at one bias. Simulation using this linear model and issues relating to bias are discussed