Real-time data enables a variety of yield improvement applications
Statistical software methods can be used to improve quality and reliability of tested parts by rejecting outliers. Outliers are parts that test within the design limits but behave differently than the majority of parts in their lot. Quality experts have known for decades that such devices are most prone to field failure.
While the Automotive Electronics Council (AEC) recommends running dynamic parts average testing (DPAT) at probe for the high quality automotive market, DPAT can produce unnecessary yield losses. Many papers have been published in the industry describing more efficient outlier detection algorithms. See Hakeman diagram.
Following over one year of joint development work with Avago Technologies, Pintail is pleased to announce that SwifTest-FOX incorporates many of the most efficient algorithms for outlier detection and rejection. Furthermore, SwifTest-FOX can be easily programmed by user to incorporate their own proprietary or experimental algorithms.
SwifTest-FOX operates on-line at both probe and final test. It supports three scopes of outlier detection:
- On-DUT: The ability to compare measurements on the same die, e.g. look for outliers in Iddq across the die.
- DUT-DUT: Looking for outliers across multiple devices being tested using simple univariate methods like DPAT.
- Wafer-End: The ability to immediately process data for an entire wafer. This might mean nearest neighbor rules, but SwifTest-FOX also supports more robust methods like multi-variate models including principal component analysis (PCA).
- Maximum effectiveness in finding true outliers
- Maximum efficiency avoids unnecessary yield loss
- Advanced Outlier Algorithms
- DPAT, Robust PAT, Box-plot PAT
- Adjacent Rank Delta
- Inter-Percentile Range
- Single Regression Residual
- Principal Component Analysis
- User may easily add personalized algorithms in Java
- Effective at wafer probe and final test
- On-line operation is more efficient for test floor operations
Defects Based Testing
Pete O’Neil of Avago describes Defects Based Testing (DBT) as follows:
- Definition – Test methods using changes in any measurable circuit behavior to indicate the presence of a defect. The defect-indicating behaviors do not need to be out of specification to indicate a defect nor do they need to be intended functions of the circuit.
- Assumption – Failures are caused by defects, therefore find the defects to find the failures.
- Distinguishing features – Novel stimuli & responses, contrasting conditions, stress, outlier identification.
- Test generation – Automatic based on fault models & models of defect activation mechanisms, confounding conditions, & response statistics.
DBT is in contrast to the traditional test methods of functional testing and structural testing. Pintail supports DBT with SwifTest-FOX. If you are interested in achieving extreme quality and reliability, contact us for more details on this state-of-the-art technique and the product that makes it possible.
Client Resources
Login Required
Click here to login now.
If it’s out of the ordinary, it must be bad.
