A rule-based computing approach for the segmentation of semiconductor defects

This paper presents a rule-based approach to detect defect patterns and to classify the defect patterns that appear on the semiconductor wafer surfaces. To obtain a general and modular defect pattern detection technique, the proposed approach adopts a hierarchical perspective. A formal analogy has been drawn between the structure of defect patterns and the symptom of disease in clinical practice. The defect patterns to be recognized are viewed as decision made to a particular disease. Design goals include detection of flaws and correlation of defect features based on co-occurrence matrix. The system is capable of identifying the defects on the wafers after die sawing. Each unique defect structure is defined as an object. Objects are grouped into user-defined categories such as chipping, metallization peel off, silicon dust contamination, etc. after die sawing and micro-crack, scratch, ink dot being washed off, bridging, etc. from the wafer.