In this paper, we present a new computational technique to extract semiconductor device model parameters. This solution methodology bases on global optimization technique and genetic algorithm with an exponential type weight function, renew operator, and adaptive sampling rule. The proposed approach extracts automatically complete parameters set from a famous BSIM3 and 4 compact models for deep-submicron and nanoscale CMOS devices. Compared to conventional artificial optimization approaches, the proposed extraction methodology tracks the shape variation of I-V curves, therefore, highly accurate result can be obtained directly. Applying the renew operator will keep the evolutionary trend improving by removing the individuals without mainly features. The sampling strategy will speed up the evolution process and still maintain the extraction accuracy in a reasonable range. The approach not only provides a novel alternative for optimal device modeling and VLSI design but also has its merit in applications to System-on-a-Chip (SoC) functional design.