Extract Parameters from Cv Curves Using A Machine Learning Method

Parameter extraction from CV curves is crucial in the development and optimization of gate stack technology. Traditional ways of obtaining these parameters are time consuming and sometimes require additional measurements. In this paper, we extract the work function (WF), oxide thickness (Tox) and substrate doping concentration (Nsub) from CV curves using a regression-based machine learning method. A dataset containing 2500 curves each for Cgg, Cgb and Cgc is generated by CVC (NCSU) and Ngspice simulation, with 80% data used for training and 20% for testing. The training results show the extracted Tox and WF have maximum errors of ± 1% and ± 0.5%, respectively. Our trained model is further validated in the parameter extraction from silicon CV data, and expected to be used for fast and automated analysis.