Reducing defectivity in 90nm BEOL metal CMP using geometrybased segmented inspection strategy

Abstract

In the early development and ramp stage of a product's lifecycle, defect densities are comprised of defects of interest (DOI) and non-critical nuisance defects. Nuisance defects can overwhelm the inspection results, limiting the overall usefulness of inspection for routine line monitoring and DOI reduction efforts. In an attempt to accelerate yield ramp and optimize the capture rate of critical area defects, a dynamic approach to recipe sensitivity using a full die inspection has been investigated. In this current methodology, a simple auto-threshold (AT) inspection has been replaced by a 4-segment segmented auto-threshold (SAT) inspection, which defines regions and their respective thresholds based upon the critical area within back-end-of-line (BEOL) Cu CMP layers. The strategy employed is to separate critical minimum geometry lines from larger, less critical lines, and further separate open regions susceptible to prior level defects (i.e. hillocks). Inline automatic defect classification (iADC) has also been used to further reduce the remaining prior-level defects not segmented out via SAT. This approach allows us to have a tunable critical area weighted inspection. As the defect densities improve over the product's lifecycle, the respective segments can further be tuned to optimize capture rates. In addition, each segment can be modulated, allowing for high purity segment specific DOI recipes to drive defect reduction activities (i.e. minimum geometry missing pattern)