Improved objective function for device model parameter extraction

Curve-fitting is a reliable strategy for the automated extraction of the parameter set of an analytical device model from measured data. This strategy always implies the minimisation of a nonlinear objective function that comprises the differences between a set of measured data points and the model. A new type of objective function is proposed that is not based on the common concept of a division between dependent and independent terminal variables, but on a concept where all terminal variables are equally treated. This objective function, which is derived from the principle of maximum likelihood estimation, can be expressed as a sum of squares of nonlinear functions, so that the computationally efficient class of Gauss-Newton methods can be used for its optimisation. Results show that, for strongly nonlinear models in particular, the convergence properties are significantly improved