Implementation of machine learning for high volume manufacturing metrology challenges

With more advanced technology nodes and device architectures being used in high volume semiconductor manufacturing, process control and metrology are facing ever-greater challenges. More advanced metrology approaches have been emerging to cope with these challenges. This study investigates machine learning predication as a complementary method to optical critical dimension (OCD) measurements. We have evaluated the suitability of machine learning to address high volume manufacturing metrology requirements for applications in both front end of line (FEOL) and back end of line (BEOL) sectors from advanced technology nodes. In the FEOL sector, we have demonstrated initial feasibility to predict the fin CD values from an inline measurement using machine learning. In the BEOL sector, machine learning is shown to provide direct prediction of electrical resistance using spectra collected from both OCD measurement sites and electrical test (e-test) sites. The predicted resistance correlation to the actual e-test value is improved in comparison with OCD results for multiple metal levels of various products. It is also shown that e-test predictions by machine learning may assist to alleviate the challenges that conventional metrology approaches are facing when the metal linewidth becomes smaller. Furthermore, impact of number of samples in training set was investigated and it was found that reducing number of samples in training set does not degrade the e-test correlation and Total Measurement Uncertainty (TMU) significantly. This paper demonstrates that predictive metrology based on machine learning is an advantageous and complementary technique for high volume semiconductor manufacturing. Together with conventional OCD measurements, a machine learning approach could successfully overcome complex metrology challenges in advanced semiconductor manufacturing.