Spectra Machine Learning for the Prediction of Sheet Resistance and Capacity

Optical spectra based on ellipsometry and reflectometry provide valuable information about the geometric and dielectric properties of wafers, and thus the correlated electric performance. This work presents a machine learning based method to directly extract the electric information from optical spectra for the prediction of sheet resistance and capacity of metal layers. Both training and testing data shows excellent consistency between predicted results by spectra machine learning and actual results from wafer acceptance test. The optical methodology enables non-destructive and in-situ prediction of electric properties at correlated processes with high inline throughput and sampling rate. Additionally, the model-free machine learning methodology provides the ability to measure the complex structures in wafers with less time to solution and more reserved resources.