ABSTRACT
In recent years, advanced packaging schemes have been demonstrated to achieve greater interconnect density for advanced technology node devices. Hybrid bonding facilitates direct copper to copper interconnection, reducing performance impact of connections in traditional solder bump technologies. Such hybrid bonds require stringent surface smoothness and alignment accuracy to achieve the required bond quality level. Copper CMP is a key enabling technology to achieve the required surface smoothness of the mating surfaces. Monitoring of the surface planarity is challenging due to the struct wafer smoothness requirement (<1 nm) combined with typical pitch size of the interconnect pad array (often greater than 1 um).
Vertical traveling scatterometry (VTS) methodologies, which leverage the SI channel’s unique capability, are known to enhance sensitivity for measuring surface and near-surface parameters.
In this work, vertical traveling scatterometry machine learning (VTS-ML) solutions were trained to successfully monitor copper recess from wafers with different incoming stacks. Reference data from 5 measurement dies across 15 wafers was used in the VTS-ML training. Evaluation of the VTS-ML correlation to reference was studied on 22 blind test wafers. The impact of the VTS filter position was studied in both test-on-train and blind test R2.
Keywords: Machine Learning, scatterometry, modeless, vertical traveling scatterometry, reference metrology, hybrid bonding, copper recess, underneath layer process variation
*Corresponding author: [email protected]