Spectral Machine Learning Training for Non-Destructive CD Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional predictive modeling algorithms in manufacturing processes are incapable of accurately predicting spatial variations in feature profiles, such as critical dimensions, and require destructive metrology measurements, which are costly and time-consuming.
Innovation Solution
A method involving machine learning models trained with spectral and metrology data to generate synthetic microscopy images and CD profiles, utilizing feature model configurations and combinations, and generative adversarial networks to produce accurate, non-destructive predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional predictive modeling algorithms are used, then manufacturing processes can be controlled, but they are incapable of accurately predicting spatial variations in feature profiles
Solution Approach 1:
The patent creates synthetic microscopy images that copy the essential features and spatial variations of real substrate images without requiring physical samples. The generative model learns from training data to generate realistic synthetic images that preserve the statistical and spatial characteristics of actual feature profiles, enabling accurate prediction without destructive measurement.
Solution Approach 2:
The patent transforms the prediction approach by changing from direct physical measurement to generating images through controlled parameter manipulation. The system uses input parameters (spectral data, process conditions) to generate output images representing feature profiles, fundamentally changing how spatial variations are predicted and measured.
2Measurement precision
If destructive metrology measurements are performed, then accurate feature profile data can be obtained, but it is costly and time-consuming
Solution Approach 1:
The system generates synthetic copies of microscopy images that replace the need for expensive and time-consuming destructive metrology measurements. These synthetic images provide equivalent measurement information without requiring physical sample destruction, dramatically improving processing efficiency while maintaining data accuracy.
Solution Approach 2:
The patent replaces physical destructive measurement systems with a computational image generation system. Instead of using physical metrology equipment that requires sample destruction, the system uses machine learning models to generate measurement-equivalent images computationally, eliminating the bottlenecks of physical measurement.
3Measurement precision
If multiple machine learning models are trained with different feature model configurations, then model selection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the model training process into multiple parallel configurations, each trained on different feature model combinations. This segmentation allows systematic exploration of model variations while organizing the computational workload into manageable, independent training tasks that can be evaluated and selected based on performance metrics.
Solution Approach 2:
The system trains multiple models with different configurations (excessive action) to ensure optimal model selection, then selects the best-performing model for deployment. This approach accepts the computational overhead of training multiple models as a necessary investment to guarantee the selection of the most accurate model for production use.
Data Source
AI summary
A method includes receiving spectral data of a substrate and metrology data corresponding to the spectral data of the substrate. The method further includes determining a plurality of feature model configurations for each of a plurality of feature models, each of the plurality of feature model configurations including one or more feature model conditions. The method further includes determining a plurality of feature model combinations, where each feature model combination of the plurality of feature model combinations includes a subset of the plurality of feature model configurations. The method further includes generating a plurality of input datasets, where each input dataset of the plurality of input datasets is generated based on application of the spectral data to a respective feature model combination of the plurality of feature model combinations. The method further includes training a plurality of machine learning models, where each machine learning model is trained to generate an output using an input dataset of the plurality of input datasets and the metrology data. The method further includes selecting a trained machine learning model from the plurality of trained machine learning models satisfying one or more selection criteria.