Wafer Metrology Dimensionality Reduction for Predictive Corrections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional manufacturing processes for producing substrates are time-consuming and wasteful, often resulting in sub-optimal products due to the high dimensionality of metrology data, which leads to increased energy consumption, processor overhead, and bandwidth usage, as well as the need to process large amounts of data from numerous manufacturing parameters.
Innovation Solution
A method involving the use of a trained machine learning model to reduce the dimensionality of metrology data, generating compressed data that allows for predictive actions to be taken on manufacturing equipment, thereby optimizing parameter selection and reducing the number of features processed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional manufacturing processes process all metrology data from numerous manufacturing parameters, then manufacturing precision is maintained, but processor overhead and energy consumption increase significantly
Solution Approach 1:
The patent extracts only the most relevant features from the high-dimensional metrology data using machine learning models. Instead of processing all manufacturing parameters, the system identifies and processes only the critical features that significantly impact substrate properties, thereby reducing energy consumption while maintaining manufacturing precision.
Solution Approach 2:
The patent segments the high-dimensional metrology data into multiple feature groups or layers. The machine learning model processes different feature groups at different levels of abstraction, allowing the system to handle complex manufacturing parameters in a structured manner that reduces overall computational energy requirements.
2Manufacturing precision
If conventional manufacturing processes process large amounts of metrology data, then manufacturing precision is maintained, but processor overhead increases
Solution Approach 1:
The patent extracts essential features from the complete metrology data set using trained machine learning models. By identifying and processing only the critical features that determine substrate quality, the system significantly reduces processor overhead while preserving the information necessary for maintaining manufacturing precision.
Solution Approach 2:
The patent transforms high-dimensional metrology data into lower-dimensional feature representations through machine learning dimensionality reduction techniques. This dimensional transformation maintains the essential information needed for quality control while reducing the computational complexity and processor overhead required for data processing.
3Measurement precision
If conventional manufacturing processes perform all metrology operations, then measurement precision is ensured, but bandwidth usage and energy consumption increase
Solution Approach 1:
The patent performs preliminary feature extraction and dimensionality reduction on metrology data before the complete data processing pipeline. By pre-processing the data to identify and retain only the most relevant features, the system reduces subsequent bandwidth usage and energy consumption while preserving measurement precision for critical parameters.
Solution Approach 2:
The patent extracts only the essential metrology features that are critical for quality assessment, rather than processing all metrology operations. This selective extraction approach maintains measurement precision for key parameters while significantly reducing the energy consumption and bandwidth usage associated with processing complete metrology data sets.
4Manufacturing precision
If conventional manufacturing processes use high dimensionality metrology data, then manufacturing precision is maintained, but the number of substrates needed for analysis increases
Solution Approach 1:
The patent transforms high-dimensional metrology data from multiple substrates into lower-dimensional feature representations that capture the essential variations. This dimensional reduction allows the system to achieve the same manufacturing precision with fewer substrates, as the machine learning model learns to identify critical patterns from reduced feature sets rather than requiring extensive data from numerous substrates.
Solution Approach 2:
The patent extracts the most informative features from metrology data across substrates, enabling the system to achieve reliable manufacturing precision with a smaller number of substrates. By focusing on critical features rather than processing all dimensions from many substrates, the system reduces the quantity of substrates needed for effective analysis and process control.
Data Source
AI summary
A method includes receiving first metrology data associated with first substrates produced by manufacturing equipment. The method further includes training a first machine learning model with data input including the first metrology data to generate a first trained machine learning model. The first trained machine learning model is capable of reducing dimensionality of second metrology data associated with second substrates produced by second manufacturing equipment to perform corrective actions associated with the second manufacturing equipment.