A die-based defect-limited yield methodology for line control

Defect monitoring and control in the semiconductor fab has been well documented over the years. The methodologies typically described in the literature involve controls through full-wafer defect counts, or defect densities, with attempts to correlate defects to electrical fail modes in order to predict the yield impact. These wafer-based methodologies are not adequate for determining the impact of defects on yield. Most notably, severe complications arise when applying wafer-based methods on wafers with mixed distributions (mix of random and clustered defects). This paper describes the proper statistical treatment of defect data to estimate yield impact for mixed-distribution wafer maps. This die-based, defect-limited yield (DLY) methodology properly addresses random and clustered defects, and applies a die-based multi-stage sampling method to select defects for review. The estimated yield impact of defects on the die can then be determined. Additionally, a die normalization technique is described that permits application of this die-based methodology on multiple products with different die sizes.