Ensemble Learning strategy in modeling of future generation nanoscale devices using Machine Learning

In semiconductor industry, conventional method of device characteristics analysis using TCAD based simulation is sometimes difficult. TCAD based device simulation gives the insight of the device with inclusion of detailed device physics with required analytical calculations. But with similar device structure having change in technology node, performance prediction using TCAD simulation is time consuming and proper expertise is necessary to run the simulations. However, machine learning based prediction algorithms may outperform traditional methods in this regard. But machine learning models should be trained with sufficient data. As the TCAD simulation generated dataset is relatively small in size, performance of the model based on only one algorithm may yield unsatisfactory result. One of the solution is to use ensemble learning strategy. Though several ensemble learning methods exists, this paper discusses the opportunities of using Stacking as the Ensemble Learning method. Here, the model uses heterogeneous algorithms in base-models and regression algorithm in metamodel state, for accurate prediction of performance in Junctionless FinFETs. Results obtained clearly indicates the reduction in prediction error.