Resist Image Estimation Using Segmented Neural Network Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current photolithography processes face limitations in pattern detail due to diffraction effects and varying chemical properties of photoresist materials, which are difficult to predict and correct, especially in semiconductor manufacturing, where achieving high accuracy and precision is crucial.
Innovation Solution
A processor-implemented method using multiple models, including a compact model, an additional kernel model, and a contour model, to generate and refine resist images and contour images through neural networks, minimizing differences between predicted and measured images, thereby improving pattern accuracy and precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional photolithography processes are used, then manufacturing simplicity is maintained, but pattern accuracy and precision deteriorate due to diffraction effects and chemical variations
Solution Approach 1:
The photolithography simulation process is divided into three separate models: a first model for generating aerial images, a second model for generating resist images, and a third model for generating contour images. Each model specializes in a specific stage of the photolithography process, allowing for more accurate and precise predictions while maintaining manageable complexity in each individual model.
Solution Approach 2:
The patent introduces a multi-dimensional approach by combining multiple models operating at different stages of the photolithography process. Instead of using a single complex model, the system uses a sequence of models that process information through different dimensional representations (aerial image → resist image → contour image), thereby improving overall accuracy without excessive complexity.
2Manufacturing precision
If single model approaches are used, then device complexity is low, but manufacturing precision deteriorates due to inability to capture chemical variations
Solution Approach 1:
The simulation process is segmented into three distinct models, each handling a specific aspect of the photolithography process. The first model handles optical imaging, the second model handles resist chemistry and development, and the third model handles contour extraction. This segmentation allows each model to be optimized for its specific function, improving overall precision while keeping individual model complexity manageable.
Solution Approach 2:
The patent introduces intermediate representations (aerial images and resist images) that serve as mediators between the input mask image and the final contour image. These intermediate representations allow the system to capture and process chemical variations and physical effects at appropriate stages, improving pattern precision without requiring a single overly complex model.
3Reliability
If conventional models are used, then ease of manufacture is maintained, but reliability deteriorates due to difficulty in predicting and correcting variations
Solution Approach 1:
The prediction process is segmented into three models that progressively refine the estimation. The first model predicts aerial images with high reliability, the second model predicts resist images capturing chemical effects, and the third model predicts final contour images. This segmentation improves overall prediction accuracy and reliability while maintaining reasonable complexity in each individual model.
Solution Approach 2:
The system uses feedback mechanisms where each model's output serves as input to the next model, allowing for iterative refinement of predictions. The multi-model architecture enables feedback loops that improve reliability by cross-validating predictions at different stages of the photolithography process, making the system more robust against variations.
Data Source
AI summary
A method and apparatus for estimating a resist image (RI) are disclosed. The method includes obtaining an aerial image (AI) and a first RI from a mask image (MI), obtaining a second RI from the AI, and obtaining a third RI based on the first RI and the second RI.