Transfer Learning Metrology Model Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional metrology techniques face challenges in maintaining accuracy and robustness when process variations occur, requiring frequent re-generation of libraries and re-collection of reference data, especially for complex structures like 3D-NAND devices, due to assumptions about fixed and coupled parameters.
Innovation Solution
The implementation of transfer learning for machine learning models, where a first set of metrology data is used to train a model that can be adapted to predict key parameters for a second set of structures, reducing the need for extensive reference data and improving recipe robustness and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional metrology techniques are used with fixed parameter assumptions, then the analysis process is simple, but the accuracy degrades when process variations occur
Solution Approach 1:
The patent applies dynamics by transitioning from fixed, static parameter assumptions to dynamic parameter adaptation through transfer learning. The system continuously adapts model parameters when process variations are detected, allowing the metrology analysis to remain accurate across different process conditions without requiring complete re-calibration.
Solution Approach 2:
The patent implements parameter changes by using transfer learning to adjust and update model parameters based on new process data. Instead of maintaining fixed parameters, the system dynamically modifies parameters to account for process variations, thereby maintaining measurement precision across different manufacturing conditions.
2Measurement precision
If libraries are re-generated frequently to maintain accuracy under process variations, then the measurement precision is maintained, but the time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-training metrology models on diverse process variations during the initial library generation phase. This preliminary exposure to various process conditions enables the model to adapt more efficiently later, reducing the need for frequent complete re-generations and minimizing time loss when process variations occur.
Solution Approach 2:
The patent extracts and transfers only the essential learned parameters and features from the pre-trained model to new process conditions, rather than re-generating entire libraries. This selective extraction and transfer of critical information maintains measurement precision while significantly reducing the time and computational resources required compared to complete re-generation.
3Productivity
If traditional metrology techniques are used with fixed parameters, then the device complexity is low, but the productivity decreases due to frequent re-calibration
Solution Approach 1:
The patent applies copying by replicating and adapting pre-trained model components to new process conditions through transfer learning. Instead of developing new models from scratch for each process variation, the system copies and adjusts existing model parameters, thereby increasing productivity while managing complexity through reuse of proven model structures.
4Measurement precision
If extensive reference data is collected for each new process condition, then the measurement precision is maintained, but the loss of time and resources increase
Solution Approach 1:
The patent introduces transfer learning as an intermediary mechanism that bridges different process conditions. Instead of requiring extensive reference data for each new condition, the transfer learning model acts as an intermediary that transfers knowledge from existing data to new conditions, significantly reducing the quantity of reference data needed while maintaining predictive accuracy.
Data Source
AI summary
Non-contact measurements, such as optical measurements or X-ray measurements, of a structure are supported using transfer learning for training a machine learning (ML) model for predicting key parameters. A first set of metrology data for a first set of structures is obtained and used to train a first ML model. A second set of metrology data for a second one or more structures is obtained. Transfer learning from the first ML model to the second set of metrology data is performed to produce a second ML model for predicting key parameters of the second one or more structures. Domain adaptation may be used in which metrology data is selected from the first set of metrology data and the second set of metrology data using a feature extractor and used to train a ML model for predicting key parameters of the second one or more structures.