Machine Learning Endpoint Detection for Substrate Metrology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional endpoint detection systems in microelectronics manufacturing are inadequate for meeting the increasing demands of precision and accuracy due to variations in substrate processing, leading to issues like under-processing and over-processing, which result in substandard devices.
Innovation Solution
A machine learning-based system that collects and analyzes spectral data to provide real-time metrology measurements by training models on historical data, allowing for precise endpoint detection and process termination, thereby preventing defects and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional endpoint detection systems are used, then the manufacturing process can be monitored, but the precision and accuracy of endpoint detection are insufficient due to variations in substrate processing
Solution Approach 1:
The system performs preliminary actions by collecting historical spectral data from multiple manufacturing systems and process conditions, then training machine learning models in advance. This pre-training on diverse historical data enables the models to adapt to various process variations without requiring real-time recalibration, thereby improving both precision and reliability of endpoint detection.
Solution Approach 2:
The invention changes the parameter of detection methodology from conventional fixed-threshold methods to adaptive machine learning models that can dynamically adjust to process variations. The ML models learn optimal detection parameters from historical data, enabling accurate endpoint detection across different manufacturing systems and process conditions.
2Manufacturing precision
If real-time monitoring is implemented to improve quality control, then product quality improves, but the system complexity increases
Solution Approach 1:
The invention replaces complex mechanical and optical endpoint detection systems with machine learning-based software models. Instead of relying on sophisticated hardware systems to detect endpoint, the solution uses ML models that analyze spectral data, thereby reducing hardware complexity while maintaining or improving detection accuracy through intelligent algorithms.
Solution Approach 2:
The system creates a virtual model of the manufacturing process using machine learning models trained on historical spectral data. This digital twin or copy of the process enables prediction and monitoring without requiring equivalent physical measurement systems at every manufacturing location, simplifying the overall monitoring infrastructure.
3Adaptability or versatility
If machine learning models are trained on historical spectral data from multiple manufacturing systems, then the model generalization improves, but the data processing complexity increases
Solution Approach 1:
The system performs preliminary data processing and model training using historical spectral data from multiple manufacturing systems before deployment. By pre-processing the data and training models in advance, the system achieves broad generalization across different manufacturing systems without requiring complex real-time data processing during actual substrate processing.
Data Source
AI summary
A machine learning model trained to provide metrology measurements for a substrate is provided. Training data generated for a prior substrate processed according to a prior process is provided to train the model. The training data includes a training input including a subset of historical spectral data extracted from a normalized set of historical spectral data collected for the prior substrate during the prior process. The subset of historical spectral data includes an indication of historical spectral features associated with a particular type of metrology measurement. The training data also includes a training output including a historical metrology measurement obtained for the prior substrate, the historical metrology measurement associated with the particular type of metrology measurement. Spectral data is collected for a current substrate processed according to a current process. A subset of current data extracted from a normalized set of the spectral data for the current substrate is provided as input to the trained model. Metrology measurement data for the current substrate is extracted from one or more outputs of the trained model.