Multivariate Part Average Testing for Semiconductor Outlier Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional part average testing methods rely on univariate statistical approaches, which are limited in detecting outliers and often result in overkill, as they use fixed outlier limits based on individual test parameters without considering multiple parameters simultaneously.
Innovation Solution
Implementing a multivariate approach for part average testing that calculates statistics using delta measurements, ratios, and principal component analysis to dynamically update testing limits based on historic data, identifying outliers through mean or median values and eigenvectors, and applying non-linear transformations to amplify outlier detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If univariate statistical methods are used for part average testing, then the testing process is simple and easy to implement, but the detection accuracy of outliers is insufficient and overkill occurs
Solution Approach 1:
The patent transitions from univariate statistical methods to multivariate statistical methods, adding dimensional analysis by simultaneously examining multiple test parameters. This dimensional expansion enables the detection of outliers that may not be apparent in single-parameter analysis, thereby improving outlier detection accuracy while managing complexity through structured multivariate approaches.
Solution Approach 2:
The patent changes the statistical parameters from single-parameter univariate statistics to multi-parameter multivariate statistics. By transforming the statistical framework to include multiple dimensions of test parameters simultaneously, the system achieves better outlier detection capability while maintaining controlled complexity through systematic parameter transformation and analysis.
2Reliability
If fixed outlier limits are used based on individual test parameters, then the testing method is simple, but false positives increase and overkill occurs
Solution Approach 1:
The patent implements dynamic limit adjustment using feedback from historical testing data. The system continuously updates outlier limits based on accumulated test results, allowing the limits to adapt to changing conditions and reduce false positives. This feedback mechanism improves reliability by making outlier identification more accurate while maintaining testing efficiency through automated limit recalibration.
Solution Approach 2:
The patent transforms fixed static outlier limits into dynamic adaptive limits that evolve with accumulated testing data. By making the limits dynamic rather than fixed, the system improves outlier identification accuracy by adapting to actual device variations while maintaining productivity through automated updates that eliminate the need for manual limit recalibration.
3Measurement precision
If multivariate statistical methods are used, then outlier detection accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent segments the multivariate analysis into manageable components, breaking down complex multivariate statistics into individual test parameter analyses that are then integrated. This segmentation approach maintains high outlier detection accuracy by preserving the multivariate perspective while reducing computational complexity through modular processing of individual parameters and their relationships.
Data Source
AI summary
Embodiments of the present invention provide systems and methods for multidimensional parts average testing for testing devices and analyzing testing results to detect outliers according to embodiments of the present invention. The testing can include calculating multivariate (e.g., bivariate) statistics using delta measurements of like devices, a ratio of measurements, or principal component analysis that identifies eigenvectors and eigenvalues to define meta parameters, for example. Raw test result data can be converted to residual space and robust regression can be performed to prevent outlier results from influencing regression, thereby reducing overkill advantageously.