Author
Don Kent, Senior Product Engineer, IM Flash Technologies dkent
In semiconductor high-volume manufacturing, some wafers each week are classified as low yield. Often these wafers have distinct spatial signatures. Similarly, many yield-improvement opportunities are linked to wafers with subtle spatial signatures. In both of these cases, the spatial patterns offer clues to improving the process. Efforts to review and classify these spatial signatures typically bear fruit; however, these efforts take the longest time and are most open to interpretation. A wafer spatial pattern classification script has been created using JSL. Spatial statistics and machine learning have been combined with high-dimension data analysis techniques in an adaptive clustering and wafer visualization application. The JSL solution utilizes several of the platforms within JMP, including PCA and Clustering, and uses the Graph Builder for visualizing the resulting spatial clusters by referencing custom wafer map files. This add-in, deployed across the IM Flash Network, permits engineers and technicians to quickly classify wafers into similar groups based on their spatial signatures. This allows quicker root-cause determination by providing a reliable metric for data mining that is faster and more accurate than manual scoring methods.
