A Hybrid Intelligent Computational Methodology for Semiconductor Device Equivalent Circuit Model Parameter Extraction

Abstract

In this paper, a hybrid intelligent computational methodology is presented for the parameter extraction of compact models. This solution technique integrates the genetic algorithm (GA), the neural network (NN), and the Levenberg-Marquardt (LM) method for current-voltage (I-V) curves characterization, optimization, and parameter extraction of deep-submicron metal-oxide-semiconductor field effect transistors (MOSFETs). For a specified compact model, this unified optimization technique extracts a set of corresponding parameters with respect to measured data. The GA is performed to search solutions according to the feedback of the NN, where the LM solves a local optimization problem with the input of the GA. The well-known BSIM and EKV compact models of MOSFETs have been studied and implemented for automatic parameters extraction. In terms of accuracy and convergence of score, the proposed optimization technique is computationally verified to show its advantages for parameter extraction of MOSFETs. Comparisons among pure GA approach, solution with GA and NN, solution with GA and LM, and the proposed method are also discussed.