K-means clustering with ANN based classification to predict current-voltage characteristics of advanced FETs

Abstract

In this work, we proposed a novel data-based methodology using artificial neural network (ANN) based classifier to predict current-voltage (I-V) characteristics of advanced FETs. The K-means clustering is employed to cluster and map the transistor drain current samples to centroids. This flexible and data dependent clustering enables accurate prediction over a wide parameter space for all regions of transistor operation. The classifier along with Savitzky-Golay filter predicts the I-V characteristics and the derivatives of I-V characteristics with an accuracy of 98%, outperforming the ANN regressor on a common test set. By utilizing the proposed model, an I-V characteristics can be predicted 8000 times faster as compared to an industry-standard TCAD tool.