Free-Form Kernel Lithography Model Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lithography model simulations using Gaussian kernels face challenges in achieving consistent results and efficient processing time, particularly in optical proximity correction (OPC) for semiconductor device fabrication, as they require extensive computational resources and time.
Innovation Solution
The implementation of a lithography model simulation method utilizing free-form kernels within a convolutional neural network framework, which generates resist images by convolving mask images with independent arbitrary matrices, improving model consistency and reducing modeling time by leveraging machine learning and upsampling operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a Gaussian function is used as the kernel in a resist model, then the model structure is simple and easy to implement, but the model consistency is poor and requires extensive computational resources for optical proximity correction
Solution Approach 1:
The patent changes the kernel function from a fixed Gaussian form to a free-form kernel with learnable parameters. The convolutional neural network learns optimal kernel parameters during training, allowing the kernel to adapt its shape and characteristics to better match the actual lithography process, thereby improving model consistency while maintaining computational efficiency through parameterized representation
Solution Approach 2:
The patent replaces the traditional physics-based Gaussian kernel approach with a machine learning-based convolutional neural network. This substitution allows the system to learn complex patterns from training data without being constrained by predefined physical models, achieving better consistency while reducing the need for extensive computational resources in OPC processes
2Manufacturing precision
If a free-form kernel is used to improve model consistency, then the modeling accuracy is improved, but the modeling time and computational cost increase significantly
Solution Approach 1:
The patent performs preliminary training of the convolutional neural network with free-form kernels using training data before actual lithography simulation. This pre-training phase establishes the optimal kernel parameters and model weights in advance, so that during actual OPC operations, the already-trained model can quickly process new patterns without requiring extensive real-time computation, thus reducing modeling time while maintaining high accuracy
Solution Approach 2:
The patent uses training data (copies of actual lithography patterns and results) to train the convolutional neural network. By learning from these copies, the model captures the essential characteristics of the lithography process without needing to perform computationally intensive simulations during actual production, thereby reducing modeling time while preserving accuracy
Data Source
AI summary
Provided is a lithography model simulation method. The method comprises receiving a first mask image, generating a second mask image by simulating an optical model on the first mask image, generating at least one third mask image by simulating a quenching model on the second mask image, and generating a resist image by performing machine learning on the first mask image, the second mask image, and the third mask image. The generating of the resist image comprises outputting first output data by convolving the first mask image with a first kernel, outputting second output data by convolving the second mask image with a second kernel, outputting third output data by convolving the third mask image with a third kernel, and adding together the first to third output data. Each of the first to third kernels is or includes a free-form kernel.