Skip to Main Content
white paper

Modeling electrochemical deposition with machine learning for chemical mechanical polishing simulation

Using machine learning to model post-electrochemical deposition surfaces can determine the optimum chemical mechanical polishing process | Scatter plots of predictions vs. targets for ZNT and ZT using XGBoost ML model
Accurate modeling of post-electrochemical deposition surface topography variation is crucial for correct and optimum chemical mechanical polishing simulations. Siemens EDA and the American University of Armenia collaborated to investigate and evaluate the use of machine learning (ML) modeling techniques to predict these complicated topography variations. Using various ML methods to model post-ECD surface profiles and comparing the results enabled them to determine which architectures and models provided the best combination of running time and accuracy.

Using machine learning to model post-electrochemical deposition surfaces can determine the optimum chemical mechanical polishing process

CMP simulation is a valuable tool for determining the optimum CMP process to use during chip manufacturing. Large surface topography variations generated after ECD affect the post-CMP surface profile. In addition, weak long-range interactions of patterns on the design are inherent for post-ECD surface profiles, which means that the surface height above the given pattern is not defined solely by the pattern itself, but is affected by neighbor patterns. The ability to create an accurate post-ECD model is essential to successful CMP simulation.

Siemens EDA and the American University of Armenia used different ML methods to model surface profiles after electrochemical deposition of copper over patterned wafers for the creation of interconnection wires for chips. Using physics and chemistry-based electrochemical deposition simulation, they generated training, validation, and test data for the ML model building. The ML models they evaluated included DNNLSTM RNN, combined CNN-DNN, and XGBoost-based models. This process enabled them to determine which model provided the best accuracy in surface height prediction, as well as correct data trends and high correlation with simulated linescans.

Share