Machine Learning Enhances Model-Based Optical Proximity Correction Framework in Advanced Semiconductor Manufacturing

Strategic Insights -

Researchers at Fudan University in Shanghai, China have proposed a machine learning-enhanced model-based optical proximity correction (MBOPC) framework that employs a convolutional neural network (CNN) to predict mask edge imaging thresholds. This innovation addresses accuracy bottlenecks in traditional MBOPC techniques due to physical modeling errors. The CNN-based variable threshold strategy significantly improves threshold prediction accuracy and critical dimension (CD) simulation consistency by integrating geometric features of local patterns and optical intensity distribution parameters.

Key Takeaways:

  • The proposed framework integrates physical models with data-driven methods for the OPC model, providing a new paradigm for next-generation MBOPC technologies that balance physical interpretability and high precision in advanced nodes.
  • The convolutional neural network (CNN) module reduces the root mean square error (RMSE) by approximately 78% on training datasets and 69% on testing datasets versus the constant threshold strategy.
  • The hybrid model reduces error medians and confines the statistical upper and lower limits of the distribution ranges to ±5 nm in CD simulations for typical patterns.
  • The simulation results demonstrate tighter error distributions across typical 1D and 2D patterns in the metal layer of 28 nm node, highlighting the method's robustness.
  • The framework is designed to be applied near the diffraction limit of lithography, where traditional MBOPC techniques face accuracy bottlenecks.
  • The research emphasizes the potential for next-generation MBOPC technologies that combine physical models with data-driven methods for improved accuracy and precision.

Statistics:

  • 78% reduction in root mean square error (RMSE) on training datasets compared to the constant threshold strategy.
  • 69% reduction in root mean square error (RMSE) on testing datasets compared to the constant threshold strategy.
  • ±5 nm upper and lower limits of distribution ranges for error medians in CD simulations for typical patterns.
  • 28 nm node simulated using the proposed MBOPC framework.
  • 1D and 2D patterns in the metal layer of 28 nm node demonstrated tighter error distributions.

Sources:

  • Machine Learning-Enhanced Model-Based Optical Proximity Correction Framework With Convolutional Neural Network-Based Variable Threshold Method Near the Diffraction Limit. IEEE Access, 2025, 13():136985-136994.
  • Fudan University, Shanghai, People's Republic of China
  • Jinhao Zhu, Liwan Yue, Ying Li, Xianhe Liu, Qiang Wu, Qi Wang, Yanli Li
  • IEEE.