Design for LDMOS Device Based on Machine Learning

The LDMOS design involves physical-based TCAD simulations and multiple rounds of experiment to calibrate models and optimize parameters. However, due to limited experimental data, the calibration of models and tuning of parameters become complex and time-consuming. To address these challenges, we propose a neural network (NN) model trained via transfer learning to replace and refine traditional design routine. The NN is trained on TCAD simulation data and fine-tuned with a small set of measured data to ensure accurate performance predictions. This AI-TCAD methodology significantly shortens design cycles, reduces costs, and is adaptable for optimizing other IC devices.