Detecting systematic defects on semiconductor wafers through statistical analysis and data mining on production test data: development, implementation and application of automatic defect cluster analysis system

Abstract

Semiconductor manufacturing test has traditionally been seen as a simple task that segregates good Devices-Under-Test (DUTs) from the defective ones (entirely or partially failed). Since recent years, this paradigm has been rapidly changing. As integrated circuit technology advances towards nano-scale geometry, high defect and fault rates are experienced throughout the semiconductor production process. The semiconductor industry has approached an inflection point whereby test-enabled diagnostics and yield learning have become crucial for further progress in Integrated Circuit (IC) manufacturing. The latest International Technology Roadmap for Semiconductors (ITRS) report on Test and Test Equipment has identified Detecting Systematic Defects as one of the industry’s most difficult challenges in the modern test technology area. Production test data was identified as an essential element to overcome these challenges in the manufacturing process feedback loop. One of the observations made from the high-volume production test data is that dies that fail due to systematic failure have a tendency to form certain unique patterns that manifest as defect clusters at the wafer level. Identifying such cluster patterns is a crucial step towards improving the fabrication process and implementing real-time statistical process control. To address the industry’s needs, this research proposes an automatic defect cluster identification and recognition method. Its practical implementation utilises statistical analysis and data mining to discover defect cluster patterns from production test data. Statistical analysis is performed to distinguish between systematic and random failure patterns while a novel Segmentation, Detection and Cluster Extraction (SDC) algorithm is proposed to extract these defect clusters. The research proposes a novel complex number Alternating Decision Tree (ADTree) that incorporates features extracted using Rotational Moment Invariants with customised geometrical dimensions for cluster recognition. Experimental results show that the proposed approach, when implemented as a complete Automatic Defect Cluster Analysis System (ADCAS), offers the required high cluster detection and classification accuracy that is expected from the industry. The application can be done either on-line or off-line. The on-line industrial application is characterised by a short computational time and is well suitable for crucial areas such as defect-oriented testing, real-time statistical process control and fast fault diagnosis. The off-line applications are useful in a periodic manufacturing process performance review or for a general post-analysis of the manufacturing process. The outcome of this research has been presented in several journal and conference publications and transferred to the industry for implementation is mass-manufacturing of modern semiconductor products.