Discipline: Technology and Engineering
Subcategory: Civil/Mechanical/Manufacturing Engineering
Room: Virginia C
Fatahillah Iskandar - Virginia State University
Co-Author(s): Dr. Zhenhua Wu, Virginia State University, VA
Chemical Mechanical Planarization, or CMP, is a manufacturing process that uses both chemical reaction and mechanical abrasion to smoothen and finely remove unwanted material on the workpiece. CMP is primarily used in semiconductor wafers to remove unwanted dielectrics or conductors and to minimize surface defects. One good indication of quantifying the performance of CMP process is the wafer material removal rate. The nature of combining chemical and mechanical work means the material removal rate are dependent on various parameters, including but not limited to pad usage, wafer hardness, and slurry flow rate etc. To quantify material removal rate, either a physics-based model or a data-driven-based model is used. In such models, some parameters are favored more than others, such as Preston’s Equation which calculates material removal rate primarily with pad pressure and rotation velocity. Based on the industrial data on CMP process, the objective of the research is to build a data-driven model and physics model between processed data and material removal rate.
The industrial data given is written in csv files. Each file have several individual wafer data, including but not limited to wafer rotation, stage, and usage of membrane. Each wafer have an average material removal rate. The data is then cleaned from missing data if necessary and organized by a chosen feature. The features that are best correlated with material removal rate are picked using ANOVA. Those features are used to build regression and physics models to display the relationship between the features and material removal rate.
The industrial data is organized and cleaned using the PANDAS library in python. Each dataframe is organized by wafer identification number and stage since the average material removal rate data is listed with relation to the wafer ID number.
We have cleaned and organized over million pieces of CMP data on 1900 wafers. Our future work is to identify which features are correlated to material removal rate and build the cyber-physical models using machine learning.
Funder Acknowledgement(s): This research is partially supported by the grant of NSF-1818655 and Trojan Center for Undergraduate Applied Research (T-CUAR) at Virginia State University.
Faculty Advisor: Dr. Zhenhua Wu, email@example.com
Role: I constructed the python code that takes all of the industrial data and split them by wafer ID number and machining stage. I also constructed the python code to take specific features' data and plot them in a histogram.