Semiconductor Anomaly Drift Monitoring for Wafer Test Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying defective semiconductor chips during wafer level testing are inadequate, as univariate measurements fail to detect all abnormal chips, leading to higher disposal costs in later manufacturing stages.
Innovation Solution
A method using a weighted area under the curve (WAUC) of cumulative distribution functions to detect data distribution drift in multivariate sensor measurements, flagging ensembles of components for further inspection based on ensemble anomaly values, and setting a drift threshold adapted to semiconductor manufacturing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If univariate measurements are used in WLT, then the measurement process is simple, but defective chips cannot be fully identified
Solution Approach 1:
The patent transitions from univariate to multivariate analysis by examining multiple sensor measurements simultaneously. The anomaly value is calculated based on deviations across multiple measurement dimensions (sensor readings) rather than single measurements, enabling comprehensive defect detection while maintaining operational simplicity through automated statistical analysis.
Solution Approach 2:
The patent changes the measurement parameters from individual sensor readings to a composite anomaly value that captures deviations across multiple parameters. By monitoring the distribution of anomaly values across芯片 batches and detecting shifts in this distribution, the system achieves high-precision defect identification without complicating the measurement process.
2Measurement precision
If multivariate anomaly values are calculated for each chip, then defect detection precision improves, but the complexity of data processing increases
Solution Approach 1:
The patent merges individual chip anomaly values into batch-level distribution analysis. Instead of processing each chip's multivariate data separately for defect determination, the system combines anomaly values from multiple chips into a distribution, then detects defects through distributional shifts. This merging approach maintains high detection precision while significantly reducing processing complexity.
Solution Approach 2:
The patent shifts from analyzing individual chip data to analyzing the distributional dimension across batches. By monitoring how the anomaly value distribution changes between batches rather than examining each chip's raw multivariate measurements, the system achieves efficient defect detection with reduced computational complexity.
3Measurement precision
If individual chips are flagged based on anomaly thresholds, then defect identification is precise, but the number of false positives and user workload increases
Solution Approach 1:
The patent merges individual chip-level decisions into batch-level quality assessments. By analyzing whether the entire batch's anomaly distribution has shifted rather than flagging individual chips, the system maintains precise defect identification while dramatically reducing the number of false positives and the associated user verification workload.
Solution Approach 2:
The patent introduces the batch anomaly distribution as an intermediary between individual chip measurements and final defect decisions. This intermediary layer aggregates individual variations and provides a clearer signal for batch-level quality assessment, reducing noise and false positives while maintaining detection precision.
4Loss of substance
If defective chips are identified early in WLT, then disposal costs are reduced, but the risk of false positives affecting good chips increases
Solution Approach 1:
The patent changes from static anomaly thresholds to dynamic distribution-based detection. By monitoring shifts in the anomaly value distribution across batches rather than using fixed thresholds, the system adapts to normal variations in manufacturing processes, reducing false positives while maintaining early defect detection capability that prevents costly disposal of good chips.
Solution Approach 2:
The patent implements feedback through continuous monitoring of batch anomaly distributions. By comparing each batch's distribution against reference distributions and adjusting for systematic shifts, the system learns from historical data to distinguish between normal process variations and actual defects, thereby reducing false positives while maintaining high detection accuracy for early defect identification.
Data Source
AI summary
A computer-implemented method is disclosed for production quality testing in component manufacturing, in particular semiconductor manufacturing, based on a detection of a drift of data points in a data distribution over a reference data distribution, wherein the two data distributions each comprise frequency distributions of anomaly values in an ensemble of components or reference elements, wherein a drift detection value is obtained as a weighted area under the curve, i.e. by the product of at least the determined cumulative distribution function of the reference data distribution and the data distribution integrated over the range from the smallest occurring anomaly value to the largest occurring anomaly value, and the drift detection value is compared to a predetermined drift threshold value. If the drift detection value exceeds the drift threshold value, the ensembles of components are flagged for further checking.