| Summary: | 碩士 === 國立清華大學 === 工業工程與工程管理學系 === 100 === With the rapid development of semiconductor manufacturing technology, it is critical to control the production process effectively and minimize process variation for yield enhancement in semiconductor manufacturing industry. To ensure the assignable cause of process variation, one of the most effective ways is to analyze the spatial defect patterns exhibiting on the wafers. Wafer bin map (WBM) can provide important rules for engineers to rapidly find the potential root cause by identifying patterns correctly. As the driven force for semiconductor manufacturing technology, classification of WBM to the correct pattern becomes more difficult because the same pattern may have different size, density, rotation angle and noise degree on the WBM. Nowadays, most companies still rely on engineers’ experiences of visual inspections and personal judgments in the map patterns. This manual approach is not only subjective, lack of justice and consistent standard, but also very time consuming and inefficient.
This study proposes an approach that integrates spatial corrlegram and decision tree to classify the pattern of WBM. First, each map is testing for spatial randomness test to classified into Random, Repeat and Clustered. For the clustered map, applying a data preparation procedure to enhance the signal of cluster and remove the noise. With a spatial correlogram to detect defect pattern, extracting the wave features based on it, to build an classification tree to classify the pattern of WBM. An simulated data sets was conducted for validation. The experimental results show that our method is robust to random noise and has a robust catching rate regardless the pattern size, density and rotation angle of WBM.
|
|---|
