Fault Clustering in deep-submicron CMOS Processes

The fraction of ICs that pass all production tests but fail in the application is called the defect level. Defect levels depend on the average number of defects per IC, and also on the clustering of these defects. High clustering leads to a higher yield and a lower defect level. This paper compiles the coefficients for defect clustering using research findings from 1970 until 2001. Because recent data for deep submicron processes are missing in the literature, the clustering coefficient has been calculated using scan fail distributions of ICs in a 180 nm process. Clustering coefficients show a steady trend towards higher defect clustering. This is beneficial, but it is probably not sufficient to achieve today´s ambitious target of ´zero defects´.