Machine Learning OPC Mask Prediction from Lithography Layouts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of building a physical model for optical proximity correction (OPC) is labor-intensive and time-consuming, and existing methods require rebuilding for different vendors, necessitating a more efficient method to produce OPC mask designs from lithography target data.
Innovation Solution
A computer-implemented method using machine learning to create photolithographic masks by obtaining lithography and mask polygon coordinates, performing linear regression to build a machine learning model, and predicting output masks from input designs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a physical model is built for optical proximity correction (OPC), then manufacturing precision is improved, but loss of time increases due to labor-intensive and time-consuming model building
Solution Approach 1:
The patent creates a simplified computational model that copies the essential behavior of the physical OPC model without replicating its full complexity. This computational model uses trained parameters to predict optical proximity effects, achieving comparable accuracy to physical models but with significantly reduced computational time and resource requirements.
Solution Approach 2:
The patent transforms the complex physical model into a computational model by changing the parameters from physical measurements to trained computational parameters. This involves extracting key features from training data and using machine learning algorithms to learn the relationship between mask patterns and optical effects, thereby reducing model building time while maintaining precision.
2Manufacturing precision
If a physical model is built for OPC, then manufacturing precision is improved, but device complexity increases due to labor-intensive processes
Solution Approach 1:
The patent replaces the manual, labor-intensive physical model building process with an automated computational system. Machine learning algorithms automatically train the computational model using training data, eliminating the need for manual model construction and reducing complexity in the model building process while maintaining or improving accuracy.
Solution Approach 2:
The computational model system performs self-training through automated machine learning processes. The system automatically processes training data, extracts features, and optimizes parameters without requiring extensive manual intervention, thereby reducing the complexity and labor intensity associated with traditional physical model building.
3Manufacturing precision
If traditional methods are used to produce OPC mask designs, then manufacturing precision is maintained, but productivity decreases due to time-consuming processes
Solution Approach 1:
The patent performs preliminary training of the computational model using extensive training data before actual mask design production. This pre-training phase establishes the model's predictive capabilities, allowing for rapid generation of OPC mask designs during production without repeating the full model building process, thereby increasing productivity while maintaining precision.
Solution Approach 2:
The computational model efficiently copies and applies learned patterns from training data to new mask design scenarios. This allows for rapid prediction of optical effects and generation of corrected mask designs without repeating the full physical modeling process, significantly improving productivity while maintaining manufacturing precision.
Data Source
AI summary
A computer-implemented method is provided for creating a photolithographic mask. The method includes, in a model building stage, obtaining lithography polygon coordinates from an input lithography target layout. The method further includes, in the model building stage, obtaining mask polygon coordinates from an input mask layout from a test mask. The method also includes, in the model building stage, obtaining correlated mask to lithography features from the lithography polygon coordinates and the mask polygon coordinates. The method additionally includes, in the model building stage, performing linear regression on the correlated mask to lithography features to obtain a machine learning model for predicting an output mask from an input lithography target design. The method further includes, in an inference stage, predicting a given output mask from a given input lithography target design using the machine learning model.