Use of adaptive network fuzzy inference system to predict plasma charging damage on electrical MOSFET properties

Abstract

A prediction model of plasma-induced charging damage is presented. The model was constructed using adaptive network fuzzy inference system (ANFIS). The prediction performance of ANFIS model was optimized as a function of training factors, including a step-size, a normalization factor, and type of membership function. Charging damage data were obtained from antenna-structured MOSFET with the variations in process parameters. For a systematic modeling, the experiment was characterized by means of a face-centered Box Wilson experiment. Electrical properties modeled include a threshold voltage ( V), a subthreshold swing ( S), and a transconductance ( G). Both S and G were found to be considerably affected by the normalization factor. For the variations in the type of membership function, either V or S was the most significantly influenced. The optimized root mean square errors are about 0.041 (V), 5.040 (mV/decade), and 12.311 (×10 −6/Ω), respectively. Better predictions were demonstrated against statistical regression models and the improvements were even more than 15% for V and S models.