Reticle Inspection Using Deep Learning and Physics Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor reticle inspection techniques, particularly for EUV photomasks, face challenges in efficiently detecting defects due to the lack of actinic EUV photomask inspectors and the computational complexity of using physics-based methods like the Hopkins method, which are time-consuming and impractical for full mask inspection.
Innovation Solution
A deep learning-based approach using a convolutional neural network (CNN) is employed to map reticle patterns to diffracted fields, combined with a physics-based model to simulate far field reticle images, allowing for efficient and accurate defect detection by generating near field reticle images from design databases and aligning them with actual inspection tool images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based methods like the Hopkins method are used for reticle inspection, then measurement precision is improved, but productivity deteriorates due to time-consuming computations
Solution Approach 1:
The patent pre-calculates and stores transmission cross coefficients (TCC) for the inspection optical system before actual inspection. This preliminary computation of the physics-based model parameters allows rapid defect detection during inspection without performing time-consuming Hopkins method calculations in real-time, thus improving throughput while maintaining measurement precision
Solution Approach 2:
The patent creates a computational model (copy) of the inspection system's optical transfer characteristics using the Hopkins method, storing the TCC values for reuse. This model copy enables rapid defect detection by comparing actual images against pre-computed reference patterns, avoiding repeated complex physics calculations while maintaining accurate defect detection
2Measurement precision
If physics-based methods are used for full mask inspection, then measurement precision is improved, but loss of time increases due to computational complexity
Solution Approach 1:
The TCC values representing the physics-based optical model are pre-calculated and stored before inspection begins. This preliminary action separates the time-consuming physics computation from the actual inspection process, enabling rapid defect detection across the full mask without sacrificing measurement precision
Solution Approach 2:
The patent applies the physics-based Hopkins method selectively to pre-calculate only the essential optical transfer characteristics (TCC) needed for defect detection, rather than performing complete physics simulations for every inspection point. This partial application of the physics method reduces overall computation time while maintaining sufficient precision for defect detection
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method significantly improves defect sensitivity and inspection throughput by simplifying the training process, reducing computational costs, and achieving accurate defect detection in EUV photomask inspection, outperforming traditional physics-based approaches in terms of speed and accuracy.
Implementation Method 1
A plurality of reference far field images are simulated by inputting a plurality of reference near field images into a physics-based model
Data Source
AI summary
Disclosed are methods and apparatus for inspecting a photolithographic reticle. A plurality of reference far field images are simulated by inputting a plurality of reference near field images into a physics-based model, and the plurality of reference near field images are generated by a trained deep learning model from a test portion of the design database that was used to fabricate a test area of a test reticle. The test area of a test reticle, which was fabricated from the design database, is inspected for defects via a die-to-database process that includes comparing the plurality of reference far field reticle images simulated by the physic-based model to a plurality of test images acquired by the inspection system from the test area of the test reticle.