Process Variation Index for Wafer Structural Change Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately monitor and predict structural changes in semiconductor processes due to variations in process conditions, leading to potential defects in semiconductor devices.
Innovation Solution
A method involving the use of a machine learning model trained with structural parameters to calculate a process variation index, which accounts for changes in structural properties by comparing measured and predicted parameters, allowing for advanced recognition of process condition changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Goodness of Fit (GOF) methods are used to determine structural parameters, then the measurement process is simple, but the accuracy in reflecting structural changes due to process variations is insufficient
Solution Approach 1:
The patent introduces a process variation index as an intermediary metric that bridges the gap between simple GOF measurements and accurate structural change detection. This index is calculated by comparing predicted structural parameters from a reference model with actual measured parameters, enabling accurate detection of process-induced structural variations without requiring complex direct measurement systems
Solution Approach 2:
The patent establishes a reference model containing predicted structural parameters before actual manufacturing processes occur. This preliminary model serves as a baseline for comparing subsequent measurements, allowing the system to detect deviations caused by process variations. The reference model is created in advance through simulations and stored for later comparison with actual production data
2Measurement precision
If machine learning models are trained with extensive structural analysis data, then the prediction accuracy of structural parameters improves, but the data processing time and computational resources increase
Solution Approach 1:
The patent trains machine learning models in advance using comprehensive training data that includes measurement data and structural parameters from multiple processes. This preliminary training phase allows the model to learn complex relationships between process conditions and structural outcomes. Once trained, the model can quickly predict structural parameters for new measurements without requiring extensive real-time computation, thus reducing processing time during actual manufacturing
Solution Approach 2:
The patent creates a virtual copy of the manufacturing process through simulation to generate training data. Instead of requiring extensive real-world measurements and structural analyses for training, the system uses simulated process data that replicates actual manufacturing conditions. This copying approach provides sufficient training data without the time and resource costs of collecting equivalent real-world data
Data Source
AI summary
A method for monitoring a process variation index includes operations of: obtaining a target parameter to be monitored and a reference parameter used to increase goodness of fit among structural parameters predicted by measuring a structure in a specific location of a wafer; obtaining a reference parameter set in a reference model; and calculating a process variation index capable of confirming a structural change of the structure according to a change in process conditions using the structural parameter and the reference parameter.