EUV Stochastic Defect Model Calibration for Complex Layouts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stochastic defect prediction models in EUV lithography are impractical for complex design layouts, requiring extensive experimental data for calibration and prone to under/over prediction, especially in off-nominal operating conditions.
Innovation Solution
A method and system using a processor to receive and tune stochastic defect models by grouping defect locations into probability bins, determining consistency with observed results, and adjusting model parameters for improved accuracy using statistical metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If generalized empirical models are built with extensive experimental data for calibration, then prediction accuracy is improved, but device complexity and data collection requirements worsen
Solution Approach 1:
The patent segments the complex calibration process into two distinct stages: a training stage that builds the model using simplified assumptions, and a calibration stage that uses a small amount of experimental data to adjust specific parameters. This segmentation allows the model to achieve high prediction accuracy without requiring extensive experimental data for the entire calibration process, thereby reducing overall device complexity.
Solution Approach 2:
The patent applies preliminary action by first building the empirical model using theoretical calculations and simplified experimental data before the actual calibration process. This preliminary model structure is then refined using a small amount of targeted experimental data, avoiding the need to collect extensive data from the beginning and reducing the complexity of the complete system.
2Reliability
If models are calibrated using experimental data at nominal conditions, then prediction reliability is improved, but adaptability to off-nominal conditions worsens
Solution Approach 1:
The patent changes the approach by calibrationting model parameters not just at nominal conditions but across a range of process conditions. The calibration process adjusts parameters to account for variations in focus, dose, and other process parameters, enabling the model to maintain high prediction reliability across both nominal and off-nominal operating conditions, thereby improving adaptability without sacrificing reliability.
3Productivity
If defect review is performed at discrete locations only, then inspection time is reduced, but measurement precision for defect characterization worsens
Solution Approach 1:
The patent segments the inspection process into rapid screening at discrete locations using standard inspection tools, followed by targeted high-resolution analysis only at locations where defects are detected. This segmentation allows the system to maintain high throughput by not performing expensive high-resolution analysis on all locations, while still achieving precise defect characterization at the necessary locations.
Solution Approach 2:
The patent introduces an intermediary step using machine learning predictions as a mediator between the initial inspection and the final high-resolution analysis. The ML model predicts defect locations and characteristics, guiding the high-resolution inspection to only those critical locations, thus maintaining both high productivity and measurement precision.
Data Source
AI summary
Using an initial probability of occurrence of a stochastic defect over an inspection area of a workpiece, one or more defects within the inspection area are imaged using an optical tool or an electron beam tool. A probability of occurrence of a stochastic defect at each of the defect locations is generated using the model. The defect locations are grouped into probability bins. A consistency between the initial probability and observed results is determined and the model can be tuned based on the consistency.