Abstract
Semiconductor manufacturing is a large scaled process with high complexity. It is a very difficult task to identify the cause of product defects occurred during the manufacturing processes. In order to meet the expectation of a high yield target, quick identification of root cause becomes a vital issue. This means manufacturers must develop a method that enables them to improve yield as fast as they can after production starts. Spatial patterns appearing on the defect map contain information about defect shape and size, and might reveal the root causes of the occurred defects. As a result, this research intends to develop a defect spatial pattern recognition algorithm based on image processing techniques. The objectives of this research are (1) to accurately identify defect spatial patterns, including line type, curve type, half-ring type, ring type, edged type, repeating type and zone type, (2) to develop a recognizing procedure with high processing speed, (3) to deal with multiple defect patterns on a wafer simultaneously regardless of the wafer size, and (4) to standardize the defect classification. The recognition procedure includes the detection of repeating defects, defects denoising, defects clustering, feature extracting, and the fuzzy inference system. A knowledge base will also be built up based on the expertise accumulated from the regular recognition process. The developed methodology is verified with industrial data from a famous semiconductor company. The experimental results demonstrate that the proposed methodology can not only meet a high accuracy target but also save much time on dealing with the defect classification comparing to human operations.