Enhanced Cu CMP Process Control by Machine Learning Enabled Measurement on E-Test Macro

The utilization of machine learning algorithms for OCD model generation has been proven to provide high quality measurement solutions with faster time to solution and good correlation to relevant accuracy metrics. Direct trench height monitoring on BEOL e-test macro by supervised machine learning is described in this paper utilizing spectra acquired from integrated metrology units on board the CMP process tool. The improved correlation of the e-test macro measurement to the final e-test value is compared for two different product type scenarios. Preliminary data is presented in the paper from utilizing the ML e-test macro measurement for automated process control (APC) of the CMP process.