LCDU is a known measurable metric for studying imaging contrast in semiconductor lithography. It is rather common to separate it into systematic and stochastic subcomponents using ANOVA decomposition technique. In this work we present several experiments performed on an ASML NXE scanner to showcase what practical learnings this type of decomposition can bring, and how it could potentially be useful in process setup, monitoring and diagnostics. Firstly, we study reticle impact on the LCDU systematics and local MEEF, and demonstrate the consequences for process correctability using standard scanner tuning knobs such as dose and focus. Secondly, we highlight experimentally the difference between how strongly resist parameters, such as layer thickness, affect systematic and stochastic LCDU terms, and compare our data with scaling laws from an established model. Finally, we make an observation on how a classical ANOVA-based method to determine metrology contribution has limitations due to random destructive resist-SEM beam interaction, and discuss ramifications for interpreting the overall measured LCDU values.
|