The impact of the reticle and wafer alignment mark placement accuracy on the intra-field mask-to-mask overlay

Abstract

The mask-to-mask writing error contribution as part of the on-wafer intra-field overlay performance has been extensively studied over the past few years. An excellent correlation (R2 > 0.96) was found between the off-line registration measurements by the PROVE tool and the on-wafer intra-field overlay results. The residual mismatch between the offline registration measurements and the on-wafer intra-field overlay was around 0.58-nm. This value is approximately 30% of the dedicated chuck overlay performance of the scanner that was used. A careful analysis was performed to understand and quantify the two dominant underlying contributors that are responsible for the 0.58-nm mismatch. The first contributor could be attributed to the reproducibility of the reticle alignment of the scanner (~0.43-nm after 10 wafers averaging). The second contributor was assigned to the sampling difference between the PROVE registration measurement and that of the alignment sensor inside the scanner (~0.39-nm). The sampling difference is a direct result of the relatively large metrology feature (alignment mark diffraction grating) in combination with older generation e-beam mask writing tools that were used in the experiments. Local grating placement variations are averaged out when the scanner alignment sensor is used for an overlay measurement. This is due to the large spot size and the scanning principle to obtain a position. This is fundamentally different for a mask registration tool since it has been designed to perform dedicated measurements on single features (globally or in-die) across the entire mask. Previous investigations used only two sampling points for each individual alignment mark diffraction grating in order to keep the total number of measurements and time under control. It is expected that the sampling difference will significantly decrease if state-of-the-art mask e-beam writers are used and/or if the number of sampling points as measured by the PROVE will be increased. It might be obvious that the ability to perform dense off-line local registration measurements has large value to reveal local mask writing errors. The new local registration map (LRM) mode of PROVE can be used to average out local reticle writing errors enabling a more accurate placement determination of large metrology features like reticle and/or wafer alignment marks. The application of LRM can be used to further improve the accuracy between the scanner and the PROVE mask registration tool if required. So far, all published correlation studies between off-line mask registration measurements and on-wafer overlay measurements were based on TIS (Transmission Image Sensor) reticle alignment marks. In this paper, we have applied LRM to improve the placement accuracy of more advanced PARIS (Parallel ILIAS) reticle alignment marks. A comparison with on-wafer measurements is made. In addition, the placement accuracy of a wafer alignment mark is considered as well. The impact of a wafer alignment mark placement error due to reticle writing errors on the intra-field overlay is experimentally determined and discussed. This includes the effect of an applied intra-field scanner (reticle alignment) correction on the wafer alignment mark placement.