Stochastic Defect Criticality Prediction Enabled By Physical Stochastic Modeling And Massive Metrology

With the adoption of extreme ultraviolet lithography for high volume production in the advanced chip manufacturing fabs, the defects resulting from the stochastic effects could become one of the major yield killers and draw increasing interest from the industry. The industry has been looking for a way to mitigate the stochastic effects through measuring, modeling, predicting and validating. In this paper, we will present a flow for the full chip stochastic failure probability prediction which is enabled by physical stochastic modeling and massive metrology.First, a new physical stochastic model has been developed by addressing multiple factors affecting final contour uncertainty. This model has demonstrated significant and consistent improvements for multiple cases. This model can be used to predict the stochastic variation sigma(sepe) at the full chip level.Using computational lithography inspection tool, the lithographic simulations are run across a full chip design, and the stochastic variation sigma(sepe) is generated at the full chip level. From the simulated CT sigma(sepe), a failure probability is calculated for each critical cut-line location. The failure probability of each pattern group is defined as the multiplication of the population and the defect probability. The pattern failure probability is used to identify the top hotspots by ranking defect criticality. Then the locations of the top hotspots are used to guide an inspection tool to find defects on a wafer and validate the failure probability prediction.With the identified hotspot locations, massive large field of view images can be collected at different dose and focus conditions on a focus exposure matrix wafer through a fast e-beam metrology. Up to 300 thousands data points are measured on these images for each hotspot at each process condition. Then the stochastic failure probability is measured at each condition for each hotspot, and the stochastic aware process window is calculated for each hotspot. The measured stochastic failure probability has been correlated with the predicted failure probability for multiple cases. A reasonable agreement has been achieved.