Diffraction Based Overlay Target Design for On-Product Overlay Measurement

• Main Author: Anifatul Faricha
• Other Authors: Hung-Fei Kuo
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/t6ec6a

碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 104 === To acquire the advanced-node technology, overlay becomes the key challenges for the nano-lithography in the semiconductor manufacturing. Overlay needs to evolve to accurately represent product device patterns for the control of overlay errors. The diffraction based overlay (DBO) is considered as the promising method, it is fast, non-destructive, stable, and also well-suited to being integrated with a 193-immersion lithography system for controlling the lithography process variations from die-to-die, wafer-to-wafer, and lot-to-lot. The DBO target is the candidate for the on-product overlay control since the DBO’s accuracy and resolution are beyond the diffraction limit. The calculation is more complicated and involved many sophisticated overlay target designs which should be robust against symmetry and asymmetry variation. Several resolution enhancement techniques (RET) such as sub-resolution assist feature (SRAF) and segmentation, are employed to meet the optimal overlay product. In this study, the segmented with SRAF model provided the sharpest aerial image contrast and the lowest critical (CD) error among other models. The feed-forward ANN model was integrated with the scatterometry to assess the resist CD profile in the DBO measurement. Based on the integrated ANN-scatterometry technique, the segmented with SRAF model had the closest result of resist CD profile to the device target among other models. The ANN-scatterometry technique, which was also used for predicting displacement offset of the DBO target with various SWAs, was successfully demonstrated. The numerical result of the minimum square error and the mean of residual in the flat region showed that the ANN model provided more effective displacement offset prediction than the linear and high order regression models, which implemented in the current scatterometry system.