Outlier Detection for OCD Metrology ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning (ML) models for optical critical dimension (OCD) metrology rely on training data that may contain anomalies and measurement errors due to equipment or system calibration issues, leading to non-repeatable and hard-to-identify measurement errors.
Innovation Solution
A system and method for OCD metrology that involves receiving training data, calculating outlier metrics to identify outlier pairs, and removing these pairs from the training data before training an ML model. This process includes using metrics such as Cook's distance, merit function deviations, and comparisons to mean or median values, with corresponding outlier thresholds applied to determine outlier pairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are trained using all available scatterometric data, then the model can capture more patterns and improve prediction coverage, but the model accuracy deteriorates due to inclusion of anomalous data points and measurement errors
Solution Approach 1:
The patent extracts and removes anomalous data points from the training set before model training. Outlier detection algorithms identify and exclude measurements that deviate significantly from expected patterns, ensuring that only high-quality data is used for training. This extraction process resolves the contradiction by maintaining comprehensive prediction coverage through the retained data while eliminating the harmful effect of anomalous points on model accuracy.
Solution Approach 2:
The patent applies preliminary data cleaning and outlier detection actions before the actual model training process. By pre-processing the data to remove anomalies and measurement errors in advance, the system ensures that the training model receives only clean, reliable data. This preliminary action prevents anomalous data from degrading model accuracy while preserving the ability to make predictions across the full range of valid patterns.
2Measurement precision
If outlier detection and data cleaning processes are implemented, then model accuracy improves by removing anomalous data points, but the processing time and computational complexity increase
Solution Approach 1:
The patent employs parameter-based outlier detection methods that evaluate statistical parameters such as z-scores, inter-quartile ranges, and deviation from mean values. By transforming the data into standardized parameter spaces and applying threshold criteria, the system efficiently identifies outliers without requiring complex computational processes. This parameter change approach resolves the time-accuracy contradiction by providing rapid outlier detection with minimal computational overhead while maintaining high model accuracy.
3Reliability
If multiple outlier detection metrics and thresholds are applied, then the reliability of outlier identification improves, but the complexity of the data processing system increases
Solution Approach 1:
The patent segments the outlier detection process into distinct, modular stages: data preprocessing, outlier detection, and data cleaning. Each stage handles specific aspects of the problem independently, making the overall system more manageable and easier to implement. By dividing the complex task into segmented steps with clear inputs and outputs, the system achieves reliable outlier identification while reducing the perceived complexity through structured organization.
Solution Approach 2:
The patent introduces statistical parameters and standardized metrics as intermediary elements between the raw data and the final cleaned dataset. These intermediaries (such as z-scores, standardized deviations, and threshold values) serve as mediators that simplify the detection process and make the system more transparent. By using these intermediary parameters, the system achieves reliable outlier identification through objective criteria while reducing complexity through standardization and formalization of the detection logic.
Data Source
AI summary
A system and methods for OCD metrology are provided including receiving training data for training an OCD machine learning (ML) model, including multiple pairs of corresponding sets of scatterometric data and reference parameters. For each of the pairs, one or more corresponding outlier metrics are by calculated and corresponding outlier thresholds are applied whether a given pair is an outlier pair. The OCD ML model is then trained with the training data less the outlier pairs.