IC Process Yield Killer Identification via Data Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Identifying the major yield killer in IC and semiconductor processes becomes increasingly difficult as yield approaches 80%, due to close correlations and large noise in test data, leading to time-consuming and labor-intensive identification processes.
Innovation Solution
A data analysis method that categorizes products or dies into normal and unqualified groups based on in-line quality and yield tests, defining problematic groups and performing statistical analysis to identify the major yield killer, thereby reducing the complexity and time required to improve process yield.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional data analysis methods are used to identify yield killers, then comprehensive test coverage is achieved, but the analysis time and labor increase significantly when yield is high (above 80%)
Solution Approach 1:
The patent segments the test data analysis by dividing products into normal and abnormal groups based on in-line quality test results, then further segments abnormal products into first problematic group (normal quality test but failed yield test) and second problematic group (abnormal quality test and failed yield test). This segmentation reduces the complexity of analyzing all test data comprehensively while maintaining identification accuracy.
Solution Approach 2:
The patent applies local quality analysis by focusing statistical analysis on specific problematic groups rather than all products. The first problematic group receives analysis of product test and yield test data, while the second problematic group receives analysis of in-line quality test data, allowing targeted identification of yield killers without unnecessary comprehensive analysis.
2Measurement precision
If comprehensive statistical analysis is performed on all test data, then accurate yield killer identification is achieved, but the complexity of the analysis process increases
Solution Approach 1:
The patent simplifies the analysis process complexity by segmenting products into distinct groups based on test results. By dividing abnormal products into first and second problematic groups with different analysis approaches, the method reduces the complexity of handling all test data uniformly while maintaining comprehensive coverage where needed.
Solution Approach 2:
The patent applies partial action by performing statistical analysis only on relevant subsets of data rather than all test data. The first problematic group is analyzed using product test and yield test data, while the second problematic group is analyzed using in-line quality test data, avoiding unnecessary analysis of data that would not contribute to yield killer identification.
3Productivity
If yield is increased to high levels (above 80%), then product quality improves, but the difficulty of identifying remaining yield killers increases due to close correlations and noise in test data
Solution Approach 1:
The patent addresses the difficulty of detecting yield killers at high yield levels by applying local quality analysis to specific problematic groups. By focusing statistical analysis on the first problematic group (products with normal quality test but failed yield test) and second problematic group (products with abnormal quality test and failed yield test), the method isolates signals from noise in the broader dataset, making yield killer detection feasible even when overall yield is above 80%.
Solution Approach 2:
The patent segments the product population to isolate yield killers from the majority of合格 products. By dividing products into normal and abnormal groups based on in-line quality tests, and further dividing abnormal products into first and second problematic groups, the method creates focused subsets where yield killers can be detected despite close correlations and noise present in the complete dataset.
Data Source
AI summary
A data analysis method for an integrated circuit process is described, for analyzing the results of at least an in-line quality test, a product test and a yield test done to the products of the IC process. The products are divided into a normal group and an abnormal group based on the result of the in-line quality test, and are divided into a qualified group and an unqualified group based on the result of the yield test. A categorization step is performed to define the intersection of the unqualified group and the normal group as a first problematic group and to define the intersection of the unqualified group and the abnormal group as a second problematic group. By analyzing one or both of the two problematic groups, the major yield killer can be identified so that process modification can be made accordingly to improve the yield.