Robust model parameter extraction using large-scale optimization concepts

Abstract

A robust approach to FET model parameter extraction is presented. By introducing DC constraints and formulating the modeling process as a complete and integrated optimization problem, the uniqueness and reliability of the extracted model parameters is improved. The approach uses multibias measurements and DC device characteristics in a sequential model building approach based on a decomposition dictionary that can be used to arrive at a suitable compromise between the simplicity and adequacy of the model. Novel automatic decomposition concepts for large-scale optimization are used to detect possible model topology deficiencies. A powerful l/sub 1/ optimization technique is used in the algorithm, and all the required gradients are provided through efficient adjoint analyses for both DC and AC sensitivities. A FET modeling example is described in detail to demonstrate the approach.<>