Maverick Control Limits for Semiconductor Device Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional maverick control methods in semiconductor manufacturing assume a normal probability density function for parametric data, leading to incorrect disposal of satisfactory devices and failure to identify high-risk ones, and neglect correlation information between variables, resulting in reduced yield and unnecessary computational resources.
Innovation Solution
A method and apparatus for calculating enhanced maverick control limits that select critical parameters, identify the actual probability density function, remove outliers, and determine disposition based on 1-dimensional or 2-dimensional parametric data sets, accounting for correlations between variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional maverick control methods assume a normal probability density function for parametric data, then the statistical screening process is simplified, but significant errors are introduced leading to incorrect disposal of satisfactory devices and failure to identify high-risk ones
Solution Approach 1:
The patent applies parameter changes by transforming the parametric data through statistical methods (such as taking logarithms or other transformations) to convert non-normal distributions into normal distributions. This allows the use of simpler normal distribution-based statistical screening while maintaining accuracy in device identification, thus resolving the contradiction between simplicity and reliability.
2Use of energy by moving object
If conventional maverick control methods neglect correlation information between parametric variables, then the computational resources are reduced, but the screening performance is degraded
Solution Approach 1:
The patent segments the analysis into two stages: first performing univariate analysis on individual parameters, then selectively applying multivariate correlation analysis only when needed. This segmentation allows the system to maintain high screening performance by capturing correlation information when present, while avoiding unnecessary computational overhead when correlations are absent, thus resolving the contradiction between computational efficiency and screening performance.
3Reliability
If comprehensive testing is conducted on each semiconductor feature, then 100% test coverage is achieved, but the testing complexity and time increase drastically due to nanometer processes
Solution Approach 1:
The patent extracts and utilizes correlation information between parametric variables to infer the status of untested features. By identifying strong correlations between measured and unmeasured parameters, the system can predict the behavior of features that were not directly tested, thereby achieving near-complete effective coverage without physically testing every single feature, thus resolving the contradiction between test coverage and testing time.
Data Source
AI summary
A method and apparatus is provided for the calculation of maverick control limits. The maverick control limit method selects the correct parameter(s) as critical parameters to be utilized by the maverick control limit method. Next, the maverick control limit method identifies the probability density function that is associated with the parametric data set(s) that are associated with the critical parameter(s). Next, abnormal data points within the measured parametric data set(s) are removed. Maverick control limits are then calculated to properly disposition semiconductor die into pass/fail categories.