Synthetic Metrology Data for Semiconductor Process Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in obtaining sufficient training data for models used in process control, leading to increased metrology effort and reduced productivity due to the time-consuming nature of metrology processes.
Innovation Solution
A method for generating synthetic metrology data based on known behavioral properties and actual metrology results to create a training data set for training models, allowing for more efficient process control strategies without the need for extensive real-time metrology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive real-time metrology is performed to obtain sufficient training data for process control models, then model training quality improves, but productivity decreases due to time-consuming metrology processes
Solution Approach 1:
The patent creates synthetic copies of real metrology data by generating artificial training datasets that replicate the statistical properties and behavioral characteristics of actual process measurements. This copying approach enables model training without requiring extensive real-time metrology measurements, thus maintaining training data quality while avoiding productivity loss
Solution Approach 2:
The patent performs preliminary data preparation by pre-generating synthetic training datasets before actual production runs. This preliminary action creates a ready-to-use training corpus that eliminates the need for time-consuming real-time metrology during production, thereby improving both model training quality and manufacturing throughput
2Manufacturing precision
If more metrology measurements are taken to improve process control, then manufacturing precision improves, but loss of time increases due to extended measurement periods
Solution Approach 1:
The patent replaces time-consuming real metrology measurements with pre-generated synthetic data copies that capture the essential variability and behavioral patterns of the manufacturing process. This enables comprehensive process control modeling without the time penalty of extensive real-time measurements
Solution Approach 2:
The patent performs preliminary generation of synthetic measurement data that encompasses a wide range of process variations. This preliminary action provides a comprehensive training dataset that would otherwise require extensive real-time metrology to collect, thereby achieving high manufacturing precision without time loss
Data Source
AI summary
A method for obtaining a training data set including synthetic metrology data, the training data set being configured for training of a model relating to a manufacturing process for manufacturing an integrated circuit. The method includes obtaining behavioral property data describing a behavior of a process parameter resultant from the manufacturing process and/or a related tool or effect. Additionally, or alternatively metrology data performed on a structure formed by the manufacturing process and/or a similar manufacturing process may be obtained. Using the behavioral property data and/or metrology data, synthetic metrology data is determined, which describes the effect of variations in the manufacturing process, and/or a related tool or effect on the process parameter. The model is trained using the training data set including the synthetic metrology data.