Adaptive Wafer Misregistration Modeling With NUSV Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring misregistration in semiconductor device manufacturing are complex due to multiple contributions from different sources, leading to non-trivial correlation of misregistration data between units, which results in low accuracy and high computing costs.
Innovation Solution
An adaptive modeling method and system that calculates a fitting function for a group of semiconductor device wafers, removes non-unit-specific values, and generates misregistration data by correlating it with fabrication and metrology tools, using techniques like principal component analysis to separate unit-specific and non-unit-specific contributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing methods are used to measure misregistration with multiple contributions from different sources, then comprehensive measurement coverage is achieved, but measurement precision deteriorates due to non-trivial correlation between units
Solution Approach 1:
The patent segments misregistration measurements into unit-specific contributions and non-unit-specific contributions. By separating these components, the method eliminates correlated errors that affect multiple units, thereby improving measurement precision without requiring a fundamentally more complex measurement system.
Solution Approach 2:
The patent extracts and removes non-unit-specific contributions from the measurement data. This extraction process isolates the unit-specific misregistration measurements from systematic errors, improving accuracy while maintaining the existing measurement infrastructure.
2Measurement precision
If existing methods are used to measure misregistration with multiple contributions from different sources, then comprehensive measurement coverage is achieved, but computing costs increase due to complex data correlation
Solution Approach 1:
The patent extracts non-unit-specific contributions from measurement data, removing the need for complex correlation computations across multiple units. This extraction approach reduces computing costs while maintaining comprehensive measurement coverage and improving precision.
Solution Approach 2:
The patent uses a fitting function to model and replicate the expected non-unit-specific contributions across units. By copying this fitted model rather than computing actual correlations, the method significantly reduces computational requirements while maintaining measurement accuracy.
3Loss of information
If comprehensive measurement of multiple contributions is performed, then complete misregistration data is obtained, but measurement precision deteriorates due to non-trivial correlation between units
Solution Approach 1:
The patent segments complete misregistration data into unit-specific and non-unit-specific components. This segmentation preserves all measurement information while separating correlated errors from actual unit variations, thereby maintaining data completeness while improving measurement precision.
Solution Approach 2:
The patent extracts non-unit-specific contributions from the complete measurement data set. This extraction removes systematic correlations that degrade precision while retaining all unit-specific information, achieving both complete data utilization and improved measurement accuracy.
Data Source
AI summary
An adaptive modeling method for generating misregistration data for a semiconductor device wafer (SDW) including calculating a fitting function for a group of SDWs (GSDW) having units, including measuring an SDW in said GSDW, thereby generating test data sets corresponding to the units, removing non-unit-specific values (NUSVs) from the test data sets, thereby generating cleaned test data sets, and analyzing the cleaned test data sets, thereby generating the fitting function, and generating misregistration data for at least one additional SDW (ASDW) in the GSDW, including measuring the ASDW, thereby generating run data sets, removing NUSVs from the run data sets, thereby generating cleaned run data sets, fitting each of the cleaned run data sets to the fitting function, thereby generating coefficient sets, and calculating misregistration data for the ASDW, at least partially based on the fitting function and the coefficient sets.