Hybrid OPC Model Segmentation for Semiconductor Processing Time
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity of semiconductor device layouts at smaller technology nodes leads to exponential increases in optical proximity correction (OPC) processing time, elongating tape-out cycle time and straining hardware and software resource utilization.
Innovation Solution
The OPC method involves grouping original design patterns into larger and smaller-sized groups, establishing simple and complex OPC models respectively using different numbers of convolution kernels, and combining these models to generate a hybrid OPC model for efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a single complex OPC model with many convolution kernels is used to maintain high accuracy for all patterns, then manufacturing precision is improved, but OPC processing time increases exponentially
Solution Approach 1:
The patent segments the original design patterns into multiple groups based on their sizes. Different OPC models with different numbers of convolution kernels are established for different pattern groups. Specifically, patterns are divided into groups where smaller patterns use models with more convolution kernels for higher accuracy, while larger patterns use models with fewer kernels to reduce processing time. This segmentation resolves the contradiction by applying appropriate model complexity to each pattern type rather than using a single complex model for all patterns.
Solution Approach 2:
The patent applies local quality by using different OPC model complexities for different spatial regions or pattern types. Instead of uniformly applying a complex model everywhere, the system tailors the model complexity to local requirements: high-accuracy complex models are applied only where needed (for smaller, more critical patterns), while simpler models are used for less critical larger patterns. This resolves the contradiction by optimizing the balance between accuracy and processing time locally rather than globally.
2Manufacturing precision
If more convolution kernels are used in the OPC model to improve compensation accuracy, then manufacturing precision is improved, but hardware and software resource utilization deteriorates
Solution Approach 1:
The patent segments the OPC modeling process into multiple models with different resource requirements. Instead of creating one comprehensive complex model that requires extensive hardware and software resources, the system creates multiple specialized models with varying kernel counts. Each model is optimized for specific pattern types, reducing the overall resource burden while maintaining necessary accuracy for each pattern category.
Solution Approach 2:
The patent applies local quality by assigning different model complexities to different pattern groups based on their specific requirements. Smaller, more critical patterns receive complex models with more convolution kernels for high accuracy, while larger patterns use simpler models with fewer kernels. This localized approach optimizes resource utilization by avoiding the deployment of complex models where they are not needed, thus resolving the contradiction between accuracy and resource consumption.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces OPC processing time by 5% to 12% while maintaining high accuracy, enhancing resource utilization flexibility.
Implementation Method 1
establishing a simple OPC model for patterns of the first group with a relatively small number of convolution kernels; establishing a complex OPC model for patterns of the second group with a relatively large number of convolution kernels
Data Source
AI summary
An optical proximity correction (OPC) method is disclosed, in which original design patterns are first grouped into a first group and a second group, wherein each pattern of the first group has a size greater than a size of any pattern of the second group. Next, a simple OPC model and a complex OPC model are individually established for the two groups using different numbers of convolution kernels. After that, the simple OPC model and the complex OPC model are combined together to generate a hybrid OPC model which is thereafter used to perform an OPC treatment on the original design patterns. This method is capable of shortening the OPC processing time and increasing the flexibility in utilizing OPC software and hardware resources.