Semiconductor Outlier Identification Using Serial Data Transform Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current outlier screening techniques for semiconductor devices are insufficient in predicting failures in sub-micron semiconductor devices, leading to increased scrap and defect rates, as they rely on time-zero failures and do not effectively identify statistical outliers that can cause early failures.
Innovation Solution
The method combines multiple data transform processing methodologies, such as statistical, mathematical, and spatial transformations, to identify outlier semiconductor devices by generating processed data that is then analyzed using a second transform processing method to define outliers, thereby reducing the number of hidden outliers in the good device population and improving correlation with electrical failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If time-zero failures are used to predict early failures without burn-in, then test cost is reduced, but prediction accuracy deteriorates leading to increased scrap of good devices
Solution Approach 1:
The patent applies preliminary action by performing multiple data transformations on test data before final outlier identification. The system transforms data through statistical methods, mathematical operations, and spatial transformations in sequence, preparing the data progressively to enhance prediction accuracy before making the final outlier determination, thereby improving early failure prediction without requiring burn-in testing
Solution Approach 2:
The patent uses intermediary processing by introducing multiple data transformation methodologies as intermediate steps between raw test data and final outlier identification. These intermediate transformations (statistical, mathematical, spatial) act as mediators that progressively refine the data to reveal hidden outliers, enabling accurate prediction without direct burn-in testing
2Measurement precision
If burn-in process is used to identify early failures, then prediction accuracy is improved, but test cost increases and chip life expectancy is reduced
Solution Approach 1:
The patent performs preliminary data transformations on standard test data to extract hidden outlier information before final disposition. By applying statistical transformations, mathematical operations, and spatial transformations in sequence, the system prepares the data to reveal early failure indicators that would otherwise require expensive burn-in testing to detect
Solution Approach 2:
The patent creates transformed copies of the original test data through multiple data transformation methodologies. These transformed data sets (statistical, mathematical, spatial) serve as alternative representations that reveal outlier patterns without requiring physical burn-in testing, thereby achieving accurate prediction at lower cost
3Device complexity
If single data transformation method is used for outlier identification, then processing complexity is reduced, but outlier detection accuracy deteriorates
Solution Approach 1:
The patent segments the outlier detection process into multiple distinct transformation stages: statistical transformations, mathematical transformations, and spatial transformations. Each segment processes the data in a specific way, and the segments are combined in sequence to achieve comprehensive outlier identification that is more accurate than any single method alone
Data Source
AI summary
A method for identifying outlier semiconductor devices from a plurality of semiconductor devices includes performing at least one electrical test to obtain electrical test data including at least one test parameter, applying at least a first data transform processing methodology to the electrical test data to generate processed test data, and applying a second data transform processing methodology that is different from the first data transform processing methodology to process the processed test data. The second data transform processing methodology applies an outlier test limit to identify non-outlier devices that comprise semiconductor devices from the semiconductor devices that conform to the outlier test limit and outlier devices that do not conform to the outlier test limit. The semiconductor devices are dispositioned using the outlier identification results. At least one of the data transform processing methodologies can include statistics.