Wafer Defect Analysis

Patterns on semiconductor wafers are often complex and hard to understand. With so many measurements made across so many chips, it is arduous to track down how and when a pattern starts to form. This Spotfire App helps you navigate through this challenge. A combination of visual analytics and data science aids root cause analysis by comparing patterns on wafers at different points in the manufacturing process.

To explore this dashboard yourself, check it out on our demo gallery.

What is a wafer? You can think of a wafer as a circular disk that contains chips, or “dies”, that might eventually go into a phone, car, computer, etc.
 

When discovering patterns on BIN data, which are recorded at the end, you naturally want to identify the first instances of the pattern. In our solution, we compute correlations across all possible relationships on a wafer using a Spotfire data function. Analyzing the results in our template points us in the right direction towards understanding our wafer.

Take the following example:

By marking the highest correlation in the top left table, we are pointed to a wafer with a clear pattern, along with a snapshot of earlier measurements with a similar pattern. From there, we can observe the time series of correlations and determine if this is the correct origin. Now there is a more complete picture of each wafer—I see its pattern, when it started, how significant it might be.

This dashboard is interactive with all sorts of filters and markings available. You can ignore sparse wafers, set a minimum defect size, etc. These customizations are offered to help with the end goal of finding significant patterns and their origins. Doing this over a longer period of time, you can begin to develop a catalog of patterns that you may want to track. If that sounds interesting, be sure to check out our work on Wafermap Pattern Recognition!