Electromigration Failure Mode Detection via Segmented Log-Normal Fitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for associating electromigration failure data points with failure modes in semiconductor devices are inaccurate due to reliance on a single log-normal distribution fit, which fails to distinguish between multiple failure modes effectively.
Innovation Solution
A computer-implemented method that sorts data points by time to failure, determines overall and group-specific log-normal distribution fits, and calculates weighted R-squares to identify and separate multiple failure modes through iterative grouping and recombination of data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single log-normal distribution fit is used for all data points, then the analysis is simple, but the association between each data point and each failure mode is inaccurate
Solution Approach 1:
The patent segments the electromigration failure data into multiple groups, where each group corresponds to a specific failure mode. Instead of applying a single log-normal distribution fit to all data points, the method divides the data set and applies separate distribution fits to each segment, thereby improving the accuracy of failure mode association while maintaining analytical tractability.
2Measurement precision
If data points are divided into multiple groups for separate log-normal distribution fits, then the failure mode association accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent employs an iterative feedback mechanism where initial groupings are made based on preliminary analysis, then the log-normal distribution fits are calculated for each group. The results feed back into refining the group assignments, with data points being re-evaluated and re-assigned to optimize the overall fit. This feedback loop continues until convergence, ensuring accurate failure mode association while systematically managing computational complexity.
Data Source
AI summary
According to one exemplary embodiment, a computer implemented method for detecting multiple failure modes in a set of electromigration failure data points includes sorting the data points by time to failure and dividing the data points to form first and second groups of data points to determine a first combination of first and second seed groups of data points providing an initial highest weighted R-square. The method further includes defining an intermediate group of data points shared between the first and second seed groups of data points and grouping the intermediate group of data points with the first and second seed groups of data points to determine a second combination of the first and second seed groups of data points providing a final highest weighted R-square. The initial highest weighted R-square is then compared to the final highest weighted R-square.