As AI for Science (AI4SCI) advances, more industries are adopting AI to improve traditional methods and explore new approaches. In semiconductor technology, the silicon carbide (SiC) manufacturing process, particularly annealing, has become more prominent and can now be effectively simulated using deep learning. However, conventional deep learning models often struggle with uncertainty quantification and overfitting, especially with small or imbalanced datasets. Bayesian Neural Networks (BNNs) offer a robust framework for capturing uncertainty in regression predictions, making them ideal for scenarios with limited data and high uncertainty. This study systematically compares the predictive performance of BNNs with traditional networks through generalization verification.