Machine Learning Model for Photolithography Defect Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity of photolithographic masks and the miniaturization of semiconductor structures make traditional methods for defect evaluation in photolithography processes time-consuming and inefficient, particularly when using rigorous simulations or simplified models like the Kirchhoff model, which fail to accurately image all mask effects.
Innovation Solution
A method utilizing a machine learning model to establish a relationship between images and design data associated with defects, allowing for the prediction of unknown defect effects by training the model with a multiplicity of images and design data, enabling efficient defect assessment without the need for complex reference image production or comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rigorous simulations (ab-initio methods solving Maxwell's equations) are used to generate reference aerial images, then the accuracy of defect evaluation is improved, but the time required for processing increases significantly making it impractical for manufacturing environments
Solution Approach 1:
The patent pre-calculates and stores correction values for various mask structures and defect types in lookup tables before actual defect evaluation. During manufacturing, the system simply retrieves pre-computed correction values based on matching the measured aerial image with design data, avoiding time-consuming rigorous simulations while maintaining accuracy.
Solution Approach 2:
The patent creates simplified reference aerial images from design data using fast computational methods, then applies pre-computed correction values to compensate for inaccuracies. This copying approach with corrections achieves rigorous simulation accuracy without the computational burden.
2Productivity
If simplified models like the Kirchhoff model are used to generate reference aerial images, then the processing speed is improved, but the accuracy of imaging mask effects deteriorates
Solution Approach 1:
The patent modifies the parameters and characteristics of the simplified reference aerial image by applying pre-computed correction values that account for the limitations of the Kirchhoff model. This transforms the inaccurate simplified image into a corrected version that accurately represents actual mask effects.
Solution Approach 2:
The patent replaces the need for complex rigorous physical simulations with a data-driven approach using machine learning models and pre-computed correction values, substituting computational physics with statistical patterns learned from training data.
3Measurement precision
If traditional die-to-die method is used for defect inspection, then the evaluation accuracy is improved through comparison with reference images, but the complexity and time required increase particularly for single die masks where reference positions are difficult to find
Solution Approach 1:
The patent extracts and removes the requirement for physical reference images or mask repetitions from the defect inspection process. Instead of comparing against reference positions, the system directly evaluates defects by comparing measured aerial images with design data using pre-computed correction values.
Solution Approach 2:
The patent introduces pre-computed correction values and machine learning models as intermediaries between the measured aerial image and the final defect evaluation. This intermediary layer eliminates the need for direct comparison with reference images, simplifying the process especially for single die masks.
Data Source
AI summary
The present invention relates to a method and an apparatus for determining at least one unknown effect of defects of an element of a photolithography process. The method comprises the steps of: (a) providing a model of machine learning for a relationship between an image, design data associated with the image and at least one effect of the defects of the element of the photolithography process arising from the image; (b) training the model of machine learning using a multiplicity of images used for training purposes, design data associated with the images used for training purposes and corresponding effects of the defects; and (c) determining the at least one unknown effect of the defects by applying the trained model to a measured image and the design data associated with the measured image.