ML Lithography Prediction for Edge Placement Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in accurately predicting and correcting edge placement errors and pattern failures due to stochastic variability, especially in extreme ultraviolet lithography, where existing models are inadequate for full-chip simulations and fail to predict low-frequency pattern failures.
Innovation Solution
A machine learning prediction model is employed to predict multi-dimensional substrate geometry and edge placement errors by analyzing geometry and process information, using asymmetrically distributed training data to determine symmetric or asymmetric edge placement error metrics, and adjusting manufacturing parameters based on these predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional lithography models are used for full-chip simulations, then computational speed is maintained, but prediction accuracy of edge placement errors and pattern failures deteriorates
Solution Approach 1:
The patent segments the lithography prediction task into multiple components: training data generation from aerial images, machine learning model training phase, and prediction phase. This segmentation allows using complex models only during training while enabling fast predictions during actual use, resolving the contradiction between accuracy and computational complexity.
Solution Approach 2:
The patent performs preliminary actions by generating training data from aerial images and training the machine learning model before actual lithography predictions are needed. This pre-computation approach stores learned patterns in the trained model, enabling fast and accurate predictions without requiring complex computations during the prediction phase itself.
2Manufacturing precision
If extensive defect inspection is performed to detect pattern failures, then manufacturing precision is improved, but productivity decreases
Solution Approach 1:
The patent performs preliminary defect prediction using the trained machine learning model on design data before manufacturing. This early prediction identifies potential pattern failures without requiring extensive post-manufacturing inspection, thereby maintaining high manufacturing precision while improving productivity by reducing inspection time.
Solution Approach 2:
The patent creates a computational copy of the lithography process through the machine learning model trained on aerial images. This model copy can predict edge placement errors and pattern failures from design data without requiring physical manufacturing and inspection, enabling virtual verification that replaces time-consuming physical inspections.
3Reliability
If low-frequency pattern failures are detected through comprehensive inspection, then reliability is improved, but loss of time increases
Solution Approach 1:
The patent creates a virtual copy of the lithography process through the trained machine learning model that can predict low-frequency pattern failures from design data instantaneously. This computational copy replaces time-consuming comprehensive physical inspections while maintaining reliable detection of rare failure modes.
Solution Approach 2:
The patent performs preliminary prediction of low-frequency pattern failures using the trained model before manufacturing occurs. By identifying potential failures in advance from design data, the system ensures reliable defect detection without requiring time-consuming post-manufacturing inspection to catch rare failure modes.
Data Source
AI summary
Systems and methods for predicting substrate geometry associated with a patterning process are described. Input information including geometry information and/or process information for a pattern is received and, using a machine learning prediction model, multi-dimensional output substrate geometry is predicted. The multi-dimensional output information may include pattern probability images. A stochastic edge placement error band and/or a stochastic failure rate may be predicted. The input information can include simulated aerial images, simulated resist images, target substrate dimensions, and/or data from a lithography apparatus associated with device manufacturing. Different aerial images may correspond to different heights in resist layers associated with the patterning process, for example.