Stochastic Contour Prediction for EUV Lithography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in semiconductor manufacturing lies in predicting the contour of patterns formed by extreme ultraviolet (EUV) lithography, which is prone to patterning defects due to the random distribution of photons, leading to complications in the manufacturing of sub-20 nm dimensions.
Innovation Solution
A stochastic prediction system using a cycle generative adversarial network (GAN) is employed to predict the contour of patterns by generating a contour histogram image based on design layouts, resist images, aerial images, slope maps, density maps, and photon maps, allowing for the verification of optical proximity correction (OPC) and the manufacturing of EUV photomasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If EUV lithography is used to process ultra-fine dimensions less than or equal to 20 nm, then manufacturing precision is improved, but reliability deteriorates due to patterning defects caused by random distribution of photons
Solution Approach 1:
The system performs preliminary stochastic prediction of contour formation before actual EUV lithography manufacturing. By using a trained neural network model to predict contour histograms based on design layouts and process parameters, the system identifies potential patterning defects in advance, allowing for preventive optimization of lithography conditions and mask designs before production begins.
Solution Approach 2:
The system creates a virtual copy of the lithography process through computational simulation. A neural network model is trained on historical SEM images and contour data to generate predicted contour histograms that replicate the statistical behavior of actual EUV patterning. This virtual model enables defect prediction without requiring physical trial runs, reducing the risk of real manufacturing defects.
2Manufacturing precision
If the number of photons per EUV patterning area is reduced, then manufacturing precision is improved for sub-20 nm dimensions, but reliability worsens due to increased susceptibility to random photon distribution effects
Solution Approach 1:
The system replaces physical trial-and-error experimentation with a computational prediction model. Instead of repeatedly running EUV lithography experiments to assess contour formation reliability, the invention uses a neural network-based stochastic prediction system that processes design layouts and process parameters to generate contour histograms, substituting mechanical/physical testing with information-processing-based prediction.
Solution Approach 2:
The system analyzes and optimizes multiple lithography parameters simultaneously to improve reliability. The neural network model processes variations in photon dose, numerical aperture, focus conditions, and mask characteristics to predict their combined effect on contour formation. By evaluating parameter interactions computationally, the system identifies optimal parameter sets that maintain precision while improving contour stability.
3Reliability
If conventional lithography techniques are used, then reliability is maintained with sufficient photons, but manufacturing precision deteriorates for sub-20 nm dimensions
Solution Approach 1:
The system introduces an intermediary prediction layer between design and manufacturing. The stochastic prediction model acts as a mediator that translates design layouts and process parameters into predicted contour outcomes. This intermediary system enables the use of aggressive EUV parameters for sub-20 nm manufacturing by providing computational assurance of contour quality, bridging the gap between high-precision requirements and photon statistics limitations.
Data Source
AI summary
The inventive concepts provide a method of providing a stochastic prediction system. The method includes extracting contours of patterns corresponding to a first design layout from a plurality of scanning electron microscope (SEM) images, respectively, generating a first contour histogram image based on the contours, and training a stochastic prediction model by using the first contour histogram image as an output, and by using the first design layout and a first resist image, a first aerial image, a first slope map, a first density map, and/or a first photo map corresponding to the first design layout as inputs, in which the stochastic prediction model comprises a cycle generative adversarial network (GAN).