CNN Virtual Metrology with DTW-Aligned Time-Series Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional virtual metrology methods, particularly those using back-propagation neural networks, face limitations in accuracy and require time-consuming feature selection processes, and struggle with data sets of varying lengths and unsimilar temporal distribution profiles, which affect the performance of virtual metrology models.
Innovation Solution
The proposed method employs a convolutional neural network (CNN) with an automated data alignment scheme using dynamic time warping (DTW) to standardize data lengths and profiles, building a virtual metrology model that includes a CNN model and a conjecture model, which automatically extracts features and improves accuracy by aligning and adjusting data sets to ensure consistency and similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional virtual metrology methods using back-propagation neural networks are used, then the system can perform virtual metrology, but the accuracy is limited and manual feature selection is time-consuming
Solution Approach 1:
The patent extracts and removes the manual feature selection step from the conventional virtual metrology process. By using CNNs, the system automatically extracts relevant features from raw process data, eliminating the time-consuming manual feature engineering while improving accuracy through automatic feature learning from the data itself.
Solution Approach 2:
The patent replaces the mechanical/manual feature selection process with an automated neural network-based system. The CNN automatically performs feature extraction and selection, substituting the manual mechanical process with an intelligent automated system that learns optimal features from data.
2Adaptability or versatility
If data sets of varying lengths and unsimilar temporal distribution profiles are used, then the system can handle diverse process data, but the virtual metrology model performance deteriorates
Solution Approach 1:
The patent applies preliminary data alignment and normalization actions before feeding data into the CNN model. By pre-processing the data to align temporal distributions and standardize lengths, the system prepares the data in advance to ensure consistent model performance while maintaining the ability to handle diverse process data.
Solution Approach 2:
The patent transforms the input data parameters through normalization and alignment operations. By changing the temporal distribution parameters and data length parameters through standardization, the system maintains adaptability to diverse data sources while ensuring consistent model performance through parameter uniformity.
3Ease of operation
If manual feature extraction is used, then the system can interpret process data, but the process is time-consuming and requires expert knowledge
Solution Approach 1:
The patent enables the system to perform feature extraction autonomously without human intervention. The CNN automatically identifies and extracts relevant features from raw process data, making the system self-sufficient in the feature extraction task that previously required expert manual intervention.
Solution Approach 2:
The patent replaces the manual mechanical feature extraction process with an automated neural network system. The CNN substitutes human experts' manual feature identification with an automated intelligent system that learns features directly from data, reducing both time and dependency on expert knowledge.
Data Source
AI summary
A virtual metrology method using a convolutional neural network (CNN) is provided. In this method, a dynamic time warping (DTW) algorithm is used to delete unsimilar sets of process data, and adjust the sets of process data to be of the same length, thereby enabling the CNN to be used for virtual metrology. A virtual metrology model of the embodiments of the present invention includes several CNN models and a conjecture model, in which plural inputs of the CNN model are sets of time sequence data of respective parameters, and plural outputs of the CNN models are inputs to the conjecture model.