3D Modeling Using Adaptive Grid Segmentation for Semiconductor Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Simulating three-dimensional structures in semiconductor processes is costly and time-consuming, often requiring high computing resources and resulting in decreased accuracy due to performance limitations of existing simulators.
Innovation Solution
A method and system for 3D modeling that involves obtaining geometric data, generating grid data, down-sampling, and using machine learning models to create attribute profile data and state data, allowing for efficient and accurate modeling of 3D structures by preprocessing and post-processing input parameters and feature maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional simulation methods are used to model 3D structures, then accuracy can be maintained, but computing resources and time are excessively consumed
Solution Approach 1:
The patent segments the 3D structure modeling process into multiple stages with different grid resolutions. Coarse grids are used for initial modeling and large-scale structures, while fine grids are applied only to critical regions requiring high precision. This segmentation allows the system to maintain accuracy where needed while reducing overall computing time and resources.
Solution Approach 2:
The patent applies partial action by using high-resolution simulation only for specific critical regions rather than the entire 3D structure. The adaptive mesh refinement technique selectively refines grids in areas where precision is most important, avoiding the excessive computational cost of uniform high-resolution modeling across all regions.
2Measurement precision
If high computing resources are allocated for 3D simulation, then accuracy may be maintained, but cost increases significantly
Solution Approach 1:
The patent divides the computational domain into multiple regions with different resource allocation levels. Critical regions receive high computing resources with fine grid resolution, while non-critical regions use coarser grids with lower resource consumption. This segmentation strategy optimizes the distribution of computing resources to maintain accuracy only where necessary.
Solution Approach 2:
The patent dynamically changes grid resolution parameters based on local feature importance and computational constraints. By adjusting the grid density parameter adaptively across different regions of the 3D structure, the system optimizes computing resource usage while maintaining simulation accuracy in critical areas.
3Manufacturing precision
If detailed 3D modeling is performed across the entire structure, then precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the 3D structure into multiple modeling domains with different levels of detail. Each segment is modeled with appropriate precision based on its importance, avoiding the need to apply maximum detail uniformly across the entire structure. This reduces overall model complexity while maintaining manufacturing precision for critical components.
Solution Approach 2:
The patent applies local quality by assigning different modeling precision levels to different regions of the 3D structure. Critical regions requiring high manufacturing precision use fine-grained detailed models, while non-critical regions use coarser models. This localized approach maintains necessary precision without increasing overall device complexity.
Data Source
AI summary
A three-dimensional (3D) modeling method includes obtaining geometric data representing a 3D structure and input parameters including factors determining an attribute of the 3D structure, generating grid data from the geometric data, sequentially generating at least one piece of down-sampled data from the grid data, pre-processing the input parameters to generate a 3D feature map, and generating attribute profile data, representing a profile of the attribute in the 3D structure, from the at least one piece of down-sampled grid data and the 3D feature map based on at least one machine learning model respectively corresponding to at least one stage.