Meaning
Statistical equations for wafer yield estimation utilize probability distributions to account for the non-random clustering of semiconductor defects. The defect density negative binomial model calculates the likelihood of a die being functional by assuming that defects are not distributed uniformly across the wafer surface but instead cluster in specific regions. This distribution represents a more realistic scenario than simpler Poisson models, which tend to underestimate yield by assuming random defect spacing.
By incorporating a clustering parameter, this model provides semiconductor foundries with a reliable tool for forecasting production volumes and setting contract prices for silicon wafers.
Mathematical Foundation
The equation relies on two primary parameters, which are the average number of defects per unit area and a clustering parameter that measures the degree of defect grouping. A lower clustering value indicates highly concentrated defects, which leaves larger areas of the wafer entirely defect-free. When the clustering parameter grows, the distribution approaches a Poisson distribution where defects are random.
Foundries use historical wafer scans to calibrate these parameters before finalizing customer agreements.
Distribution Strategy
Applying this model allows manufacturers to negotiate volume commitments with realistic delivery schedules. The defect density negative binomial model predicts how many wafers must be processed to satisfy a specific order. This prediction minimizes the risk of under-delivery or over-production, both of which disrupt the supply chain.
Contracts often incorporate these yield projections to define the baseline cost per good die.
Yield Prediction
Accurate yield forecasting prevents the financial losses associated with unexpected wafer failures. By utilizing the defect density negative binomial model, product managers can estimate the break-even point for new silicon architectures. This calculation directly influences the decision to transition from established nodes to advanced nodes.
