Chemical mechanical polishing (CMP) has become a necessary processing step in the fabrication of copper interconnects. Copper CMP is recognized to suffer from pattern dependent problems such as dishing and erosion, which cause increased line resistance and non-uniformity within the die. The non-uniformity on one metal level can lead to cumulative non-uniformity on higher metal levels, leading to potential integration and manufacturing problems. Predictive pattern dependent models of copper CMP processes are therefore highly desirable for predicting dishing and erosion on random layouts, assessing the effectiveness of dummy fills in minimizing within-die non-uniformity, aiding in the generation of smart interconnect design rules, and identifying potential bulk copper clearing problems in multi-level metallization designs. In this thesis, the first predictive semi-physical chip-scale pattern dependent model for copper CMP processes is developed. A comprehensive model calibration methodology for any multi-step copper CMP process is also developed. The model takes into account the initial long range electroplated topography, the effective pattern density, and the initial local step heights within the arrays. The model also accounts for the temporal evolution of the bulk copper thickness during CMP, the temporal evolution of dishing and erosion, and the layout dependencies of dishing and erosion. A three step conventional copper CMP process experiment and a single step abrasive-free copper CMP process experiment are performed to test the accuracy of the model and the calibration methodology.
展开▼
