Dual-Band Neural Network for Accurate Lithography Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lithography simulation methods are computationally expensive and limited in accuracy, especially when seeking high resolution and multiple shape contour generation, and existing machine learning solutions are restricted in usage scenarios or low in modeling accuracy.
Innovation Solution
A dual-band neural network system with two perception paths is employed, where a first branch extracts low-frequency global information using an optimized Fourier unit and a second branch captures high-frequency local information, enabling efficient and accurate lithography simulation at 1 nm2/pixel resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional rigorous simulation methods are used, then manufacturing precision is improved, but productivity deteriorates due to computational expense and slow processing speed
Solution Approach 1:
The patent replaces conventional rigorous optical simulation methods (which are computationally expensive and slow) with a neural network-based system that has been trained on simulation data. The neural network learns the mapping between mask patterns and resist contours through training, allowing for fast inference without performing the full rigorous simulation each time, thus achieving both high accuracy and high speed
Solution Approach 2:
The patent creates a digital copy of the rigorous simulation process in the form of a trained neural network model. The neural network is trained on input-output pairs from rigorous simulations, effectively copying the simulation's computational results in a compressed, learnable representation. During inference, the network produces similar results much faster by applying learned patterns rather than performing the full simulation
2Productivity
If machine learning models like LithoGAN are used, then productivity is improved through faster computation, but manufacturing precision deteriorates due to limited usage scenarios and low modeling accuracy
Solution Approach 1:
The patent segments the lithography simulation task into multiple sub-tasks handled by specialized components: a U-Net architecture handles the main contour generation with high precision, while separate modules handle specific features like contacts and vias. This segmentation allows each component to be optimized for its specific function, achieving both speed and accuracy that monolithic models cannot attain
Solution Approach 2:
The patent applies local quality by using different processing approaches for different regions and features within the lithography pattern. The U-Net architecture processes different parts of the image with different levels of detail, and separate specialized modules handle specific feature types (contacts, vias, metal layers) with appropriate precision for each feature's requirements
3Manufacturing precision
If deep lithography simulator with multiple image channels is used, then manufacturing precision is improved, but device complexity increases due to multiple input requirements
Solution Approach 1:
The patent implements universality by designing a single neural network architecture that can process multiple types of input features (mask patterns, optical parameters, process conditions) through a unified U-Net framework. The network learns to handle various input configurations and produces consistent high-quality contours across different lithography scenarios, eliminating the need for separate processing pipelines for each feature type
Data Source
AI summary
As integrated circuit geometries have shrunk, lithography simulation has developed to ensure that the masks used to fabricate the circuits satisfy the chip yield and fabrication turnaround time targets. To manufacture an integrated circuit (chip), an initial layout for the integrated circuit design is processed to compute a wafer image (e.g., resist material “printed” on the wafer using photomasks). Lithography simulation processes the initial layout according to optical physics to compute an estimated wafer image without actually constructing the physical masks or consuming any wafer fabrication resources and may be used to confirm manufacturability of the design layout before it is fabricated. Performing lithography simulation using a dual-band neural network produces accurate results efficiently. Dual-band refers to a dual frequency band processing whereby the input layout (mask image) is separately processed by both a first and second branch to extract low-frequency (global) features and high-frequency (local) features, respectively.