Convolutional-Neural-Network-Based Parameter Extraction for the BSIM-BULK Model

With the advancement of semiconductor technology into the deep-submicron and nanoscale domains, the BSIM-BULK model provides a reliable foundation for device modeling, supporting circuit design in advanced process nodes. However, traditional parameter extraction methods often rely on engineer experience to adjust initial values and optimization steps, posing challenges in efficiency and accuracy, especially for high-dimensional nonlinear optimization tasks. This paper proposes a CNN-based parameter extraction method for the BSIM-BULK model. By leveraging CNN, the method enhances the modeling accuracy, improves the generalization capability, and increases the robustness of extracted models. Experimental results demonstrate that a single CNN model can fast and precisely extract both the I-V and C-V parameters of the BSIM-BULK model.