Semiconductor Structure Metrology With Pre-Process Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing metrology methods struggle to accurately measure critical dimensions of complex semiconductor structures due to limited reference data and inadequate capture of pre-process variations, leading to inaccurate and time-consuming measurements.
Innovation Solution
Employ pre-process measurement data to train a post-process machine learning (ML) based measurement model, augmenting reference data with estimated values to improve measurement accuracy and robustness, using a combined measurement model that leverages correlations between structural characteristics before and after process steps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical model based metrology methods are used to measure complex semiconductor structures, then measurement capability is provided, but measurement time and computation time increase significantly
Solution Approach 1:
The patent applies preliminary action by training machine learning models in advance using pre-process measurement data and reference data. The trained models are then deployed for rapid post-process measurements, eliminating the need for time-consuming physical model-based analysis during actual measurements. This pre-computation approach significantly reduces measurement time while maintaining accuracy.
Solution Approach 2:
The patent creates a virtual copy of the measurement process through machine learning models that replicate the functionality of physical model-based metrology. These ML models are trained to predict structural parameters from measurement data, providing a faster alternative to actual physical model-based calculations while preserving measurement capability.
2Ease of manufacture
If machine learning models are trained only with limited actual reference data, then model training is feasible, but measurement accuracy deteriorates due to inadequate capture of pre-process variations
Solution Approach 1:
The patent merges pre-process measurement data with post-process reference data to create an augmented training dataset. This combination allows the ML model to learn correlations between pre-process and post-process structural characteristics, capturing process variations that would be missed with reference data alone. The merged dataset significantly improves measurement accuracy while maintaining training feasibility.
Solution Approach 2:
The patent performs preliminary data preparation by collecting and processing pre-process measurement data before model training. This pre-processing step creates an augmented reference dataset that captures process variations, which is then used to train more accurate ML models. The preliminary action of data augmentation directly addresses the accuracy problem.
3Productivity
If complex three-dimensional semiconductor structures are fabricated with new geometries, then device performance is improved, but characterization difficulty increases
Solution Approach 1:
The patent introduces machine learning models as an intermediary between complex semiconductor structures and measurement systems. These ML models learn the complex relationships between measurement data and structural parameters from augmented training data, enabling accurate characterization of 3D structures without requiring complex physical models. The intermediary ML model simplifies the characterization process while handling the complexity of new geometries.
Solution Approach 2:
The patent changes the approach from physics-based parameter calculation to data-driven parameter prediction. By training ML models on augmented datasets that include pre-process measurements, the system adapts to new 3D structures and geometries through parameter learning rather than relying on fixed physical models. This flexibility enables accurate measurement of emerging device architectures.
Data Source
AI summary
Methods and systems for using pre-process measurement data to train post-process, machine learning (ML) based measurement models are described herein. In one aspect, a post-process, ML based measurement model is trained using reference data derived from actual reference measurements and estimated reference data generated by a trained mapping model. The trained mapping model maps measured values of parameters of interest at a pre-process state to estimated reference values at the post-process state. In this manner, the reference data employed to train a ML based measurement model is augmented based on pre-process measurement data. In another aspect, measurements of complex semiconductor structures are based on a combined measurement model including trained pre-process and post-process measurement models. Pre-process measurement data is employed directly as part of a combined ML based measurement and indirectly as part of the training data set for the combined ML based measurement model.