Analysis of trade-off relationships between resolution, line edge roughness, and sensitivity in extreme ultraviolet lithography using lasso regression

Abstract

The high-volume production of semiconductor devices by extreme ultraviolet (EUV) lithography has started since 2019. The trade-off relationships between resolution, line edge roughness (LER), and sensitivity are a significant concern for the extendability of EUV lithography. In this study, the trade-off relationships were analyzed using the least square and lasso regressions. Within the half-pitch range of 5–16 nm, the relationships among resolution, chemical gradient (an indicator for LER), sensitivity, total sensitizer concentration, and effective reaction radius for deprotection were nicely predicted by the fitted parameters. The number of feature values required for describing the chemical gradient was examined. The dependence of fitting accuracy on the number of data used for the analysis was also examined. The lasso regression was effective to not only the reduction of the number of feature values but also the improvement of fitting accuracy for small data sets, compared with the least square regression.