Reconstruction of the Failing Chips Per Wafer Distribution from Clustering Measurements

Average yield is the typical and most generally accepted metric for the semiconductor manufacturing process. Here we show that statistical analysis of the same data used to measure average yield can be used to extract additional useful information that reveals a more detailed picture of the manufacturing process. We measure the number of failing chips for each wafer in a large data set and find that there is remarkable consistency in the shape of the wafer yield histogram (or distribution) across products manufactured in different locations or with different technologies, despite sometimes significant yield variability from wafer to wafer. We further show that the clustering of failing chips on a wafer is an important determinant of the distribution shape and propose a method to reconstruct the distribution of the number of failing chips per wafer from clustering measurements. The findings here demonstrate that there is useful information in wafer yield data beyond average yield, which can for example lead to the identification of wafer outliers. Furthermore, the consistency of the distributions across products and technologies indicates the existence of common mechanisms for yield loss. Similarly, our cluster analysis method shows that the clustering of failed chips is an important characteristic of the manufacturing process, suggesting that significant yield loss may be due to macroscopic processes (and not necessarily to point defects) that extend beyond the area of a single chip.