Hybrid evolutionary algorithm in MOSFET parameter extraction

In this paper we report an on going research on applying evolutionary computation to the identification of technological parameters of MOS transistors (MOSFETs) using the current-voltage measurements. The identification consists in approximating the observed values of the current with the values generated by the transistor model. Values of parameters for which the smallest approximation error is observed are assumed to be the best estimations to the real values. The model is nonlinear and nondifferentiable, and the error function takes multiple local minima with respect to the parameter values. We apply a combination of an evolutionary algorithm together with the Nelder-Mead method to minimize the error function and we experimentally investigate the effectiveness of the proposed approach.