MACHINE LEARNING-BASED PERFORMANCE PREDICTION MODEL OPTIMIZATION FOR SOI LDMOS USING ADAPTIVE SMALL SPACE DATASET

This work proposes an adaptive small space dataset generation approach to improve the ML-based model. By leveraging radial basis function(RBF)kernel to calculate similarity to the target structure, small space training data sets with high similarity are filtered for model training. The numerical results show that the small space model can not only improve the performance prediction accuracy, but also accelerate the model testing process. Meanwhile, the memory requirements during the training process are significantly reduced. In addition, the accuracies of small space models using different algorithms also demonstrate the robustness of the proposed method.