Characterization and prediction of overlay signatures in WxW bonding processes for backside power distribution networks (BS-PDN)

Wafer-to-wafer bonding processes have found widespread adoption in image sensor, 3D NAND, logic and DRAM applications. The mechanical deformations during the bonding process result in significant distortions at the interface between the two bonded wafers. In this work we will first discuss the in- plane distortions introduced by the wafer bonding process and how changes in bonding process parameters such as plasma activation parameters, bow and gap modify the distortions. Next, we will discuss a shape-metrology based overlay prediction methodology to characterize the distortions introduced by the bonding process ahead of the lithography exposure. We demonstrate that the overlay prediction model captures overlay changes induced by intended changes in the bonding process as well as uncontrolled variations i.e. those that modulate distortions even with nominally fixed bonder recipe parameters. We have used these results to generate correction per exposure (CPE) corrections in feed forward mode. The wafers were then exposed with these CPE corrections to assess the resulting residual overlay. As we will show using this shape-based feedforward process approximately 5 nm 3 sigma post lithographic overlay performance can be achieved even with significant DOE variations used. Finally, we will point out a methodology whereby a systematic analysis of overlay variability is possible for the purpose of identifying mechanisms contributing to the overlay variability.