Dynamic defect-limited yield prediction by criticality factor

In this paper, we present a new methodology for inline yield prediction based on defect inspection and design data. We derive a new metric called criticality factor (CF), which is essentially a fractional critical area for a defect of the reported size in a small layout window around the reported defect location. CF would be a good predictor of yield if geometrical considerations alone determined whether an electrical fail will result. Since other properties of the defect affect the electrical outcome (such as material properties), we employ a Training Set of wafers where the functional relation between CF and die yield is learned for each critical inspection step. From that point on these curves are used to predict the yield impact of in-line defects for new wafers. In addition, we show that highly-systematic defects (i.e. layout dependent) deviate from the CF functional curves, and hence add noise to the calculation. We suggest a technique to separate these defects from the random population, and calculate a corrected CF value for them.