Application of machine learning to stochastic effect analysis of chemically amplified resists used for extreme ultraviolet lithography

Abstract

Chemically amplified resists will be used in the high numerical aperture (NA) tools of extreme ultraviolet lithography. However, stochastic defects are a serious problem for their application to the high NA tools. In this study, the stochastic defect generation was simulated on the basis of the sensitization mechanisms and analyzed to clarify the contribution of process and material parameters using machine learning. The half-pitch HP, the sensitivity s, the total sensitizer concentration C s, the effective reaction radius for deprotection R eff, and the initial standard deviation of the number of protected units per polymer molecule σ i were used as variables. As a result, the exponential function reproduced the simulation results well. s and HP had dominant effects in LER formation. For pinching, s and HP were dominant. σ i had a major effect. For bridging, s and HP were also dominant, the effect of σ i was not major and C s and R eff effects were major.