Automatic Device Model Parameter Extractions via Hybrid Intelligent Methodology

Abstract

We report an advanced hybrid intelligent methodology for device model parameter extractions combining multiobjective evolutionary algorithms, numerical optimization methods, and unsupervised learning neural networks on a unified optimization framework. The results between experimentally measured data and the calculation from industrial standard compact models are accurate, stable and convergent rapidly for all I-V curves. Verifications from diodes, bipolar transistors, MOSFETs, FinFETs, to nanowire MOSFETs confirm the robustness of the developed prototype, where the extraction is within 5% of accuracy.