Abnormality Detection for Semiconductor Shape Prediction Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a discrepancy between predicted and actual processing shapes in semiconductor manufacturing due to potential inaccuracies in prediction models or processing abnormalities, leading to increased costs and development time.
Innovation Solution
An abnormality detection apparatus that uses a processing shape prediction model to identify discrepancies between predicted and actual processing results, incorporating observation data and control parameters to detect abnormalities, and a degree-of-contribution analysis to refine the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training data including experimental results with processing abnormalities is used for training the prediction model, then the model can be trained with more data, but the accuracy of the prediction model deteriorates due to irregular training data
Solution Approach 1:
The patent extracts and removes abnormal processing results from the training data set. The abnormality detection apparatus identifies processing results that deviate from normal patterns and excludes them from training data, ensuring that only normal processing results are used for model training. This resolves the contradiction by maintaining data quantity while improving accuracy through selective extraction of valid training samples.
Solution Approach 2:
The patent implements a feedback mechanism where the prediction model's output is compared with actual processing results, and abnormal deviations are detected and fed back to filter the training data. This closed-loop feedback system continuously identifies and removes abnormal data points, enabling the model to learn from clean, normal processing results while maintaining sufficient training data volume.
2Manufacturing precision
If processing tests are repeated to improve prediction model accuracy, then the model accuracy improves, but the cost and development period increase
Solution Approach 1:
The patent applies preliminary action by detecting and removing abnormalities from training data before model training begins. The abnormality detection apparatus pre-processes the training data set to eliminate problematic samples, ensuring that the model training starts with clean data. This preliminary filtering prevents the need for repeated processing tests, reducing both time and cost while maintaining high prediction accuracy.
3Adaptability or versatility
If more training data is collected to account for potential abnormalities, then the model can handle more cases, but the training period and computational resources increase
Solution Approach 1:
The patent extracts only the necessary normal processing results from the available data, creating a refined training data set that maintains model coverage while reducing size. By removing abnormal data points, the system achieves the same adaptability with fewer training samples, thereby shortening the training period and reducing computational resource requirements.
Data Source
AI summary
An abnormality detection apparatus including: a processing shape prediction unit configured to predict, using a processing result prediction model in which a control parameter value of the processing apparatus and an observation parameter value obtained by observing a phenomenon occurring in the processing apparatus during the processing are set as an independent variable and an evaluation value of the processing by the processing apparatus is set as a dependent variable, the evaluation value of the processing by the processing apparatus; and a first abnormality detection unit configured to detect, based on a difference between an evaluation value of determination target processing and a prediction evaluation value of the determination target processing predicted by inputting a control parameter value used in the determination target processing and an observation parameter value observed in the determination target processing to the processing result prediction model, an abnormality in the processing result.