Semiconductor Outlier Identification via Statistical Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting failures in semiconductor chips, especially 'statistical outliers' that fail after passing production tests, are inadequate due to scarcity of failure data and reliance on time-zero failures, which are less reliable for sub-micron devices and increase in 'infant-mortality' failures.
Innovation Solution
A data-driven statistical characterization method that identifies correlations between parameters in both failing and non-failing semiconductor chips, using success data to define parameter constraints and segregate statistical outliers from production chips, without relying on burn-in testing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If burn-in testing is used to identify early failures, then reliability prediction accuracy is improved, but testing cost and time increase
Solution Approach 1:
The patent performs preliminary statistical characterization of good chips during production testing to establish parameter constraints and correlations before the chips are shipped. This preliminary data collection and analysis enables reliable reliability predictions without requiring time-consuming burn-in testing, as the statistical models are built in advance using production test data from a representative sample of good chips.
Solution Approach 2:
The patent creates a statistical copy or model of chip behavior based on production test data from good chips. Instead of physically subjecting chips to burn-in testing, the system builds a probabilistic model that replicates and predicts failure patterns. This statistical copy allows reliability assessment without the time and cost of actual burn-in testing.
2Ease of manufacture
If time-zero failure data is used to predict early failures, then testing cost is reduced, but prediction reliability decreases
Solution Approach 1:
The patent changes the approach from using binary pass/fail data to analyzing continuous parameter variations and correlations among multiple chip parameters. By examining the distribution and relationships of parameters (such as electrical characteristics, dimensional measurements, and process variables) in good chips, the system identifies statistical patterns that predict early failures more reliably than simple time-zero failure data alone.
3Reliability
If failure data is collected to improve statistical confidence, then prediction accuracy is improved, but testing time increases
Solution Approach 1:
The patent makes the production testing process serve dual purposes: it both validates chip functionality and collects the statistical data needed for reliability predictions. The same production test data that ensures chips meet specifications is simultaneously used to build the statistical models for predicting early failures, eliminating the need for separate burn-in testing phases.
Data Source
AI summary
Systems and methods for identification of outlier semiconductor devices using data-driven statistical characterization are described herein. At least some preferred embodiments include a method that includes identifying a plurality of sample semiconductor chips that fail a production test as a result of subjecting the plurality of sample semiconductor chips to a stress inducing process, identifying at least one correlation between variations in a first sample parameter and variations in a second sample parameter (the sample parameters associated with the plurality of sample semiconductor chips) identifying as a statistical outlier chip any of a plurality of production semiconductor chips that pass the production test and that further do not conform to a parameter constraint generated based upon the at least one correlation identified and upon data associated with at least some of the plurality of production semiconductor chips, and segregating the statistical outlier chip from the plurality of production semiconductor chip.