Using Machine Learning to Predict the Durability of a Mold for Producing Nanostructures in Ultraviolet Nanoimprint Lithography

Performing the minimum number of ultraviolet nanoimprint lithography imprints is important for ensuring high throughput and low costs. The stamp lifetime can be estimated quickly from little information during imprint processing. We proposed two methods to predict the stamp lifetime from durability test data with a line-patterned mold. We characterized the contact angle, concentration of release resin, and the number of imprints from the data. Both of the proposed methods used machine learning. One method was binary classification, and the other was regression analysis. Under the binary classification method, the recall was 50% and the prediction showed that the recall becomes 0% when Gaussian random noise is added. Under the regression analysis, the prediction did not drastically change, with an approximately 100-fold increase in the mean absolute error. The results show that regression analysis is useful for predicting the lifetime of a line-patterned mold.