Lithographic Hotspot Detection Using Multiple ML Kernels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hotspot detection methods in advanced process technology face challenges such as high computational complexity, limited flexibility in recognizing unseen hotspots, and high false alarm rates, particularly in lithography processes, where subwavelength lithography gaps cause unwanted shape distortions.
Innovation Solution
A computer-implemented method using multiple machine learning kernels trained to identify different hotspot topologies, where layout clips are evaluated to combine results and classify data into hotspot and non-hotspot clusters, extracting topological and non-topological critical features to construct kernels that focus on specific hotspot types, balancing data populations to reduce false alarms and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full lithography simulation is used for hotspot detection, then detection accuracy is improved, but computational complexity and runtime increase extremely
Solution Approach 1:
The patent segments hotspot detection into multiple specialized machine learning kernels, each trained to detect specific hotspot patterns. This segmentation allows the system to achieve high detection accuracy for different hotspot types without requiring full lithography simulation for every case, thereby reducing overall computational complexity while maintaining precision.
Solution Approach 2:
The patent changes the detection parameters by training multiple kernels with different parameter sets optimized for specific hotspot patterns. Each kernel uses parameters tuned for its target pattern type, enabling accurate detection without the computational burden of full simulation across all possible patterns.
2Productivity
If pattern matching is used for hotspot detection, then runtime is reduced, but flexibility to recognize unseen hotspots is limited
Solution Approach 1:
The patent introduces dynamic adaptability by training multiple machine learning kernels that can be selectively applied based on the input pattern characteristics. This dynamic approach allows the system to maintain fast detection speeds for known patterns while adapting to recognize unseen hotspot types, bridging the gap between speed and flexibility.
Solution Approach 2:
The patent uses machine learning kernels as intermediaries between rigid pattern matching and full simulation. These kernels provide the flexibility to recognize diverse hotspot patterns while maintaining computational efficiency, acting as an intermediary layer that combines advantages of both approaches.
3Adaptability or versatility
If machine learning is used for hotspot detection, then unknown hotspots are detected well, but false alarm rate increases
Solution Approach 1:
The patent segments the machine learning detection into multiple specialized kernels, each focused on specific hotspot patterns. This segmentation reduces false alarms by ensuring that each kernel only makes decisions about patterns it is specifically trained to recognize, rather than a single general-purpose kernel making broad classifications that may lead to false positives.
Solution Approach 2:
The patent implements feedback mechanisms where kernel results are combined and analyzed to reduce false alarms. The system uses feedback from multiple kernel evaluations to verify detections, cross-checking results to distinguish true hotspots from false alarms while maintaining the ability to detect unknown patterns.
4Measurement precision
If hybrid approach combining pattern matching and machine learning is used, then accuracy and false alarm reduction are improved, but runtime increases
Solution Approach 1:
The patent creates a dynamic hybrid system where the choice between pattern matching and machine learning kernel evaluation is determined by the input characteristics. This dynamic selection allows the system to achieve high accuracy when needed while maintaining fast runtime by using simpler methods when appropriate, avoiding the constant overhead of a full hybrid approach.
Data Source
AI summary
A hotspot detection system that classifies a set of hotspot training data into a plurality of hotspot clusters according to their topologies, where the hotspot clusters are associated with different hotspot topologies, and classifies a set of non-hotspot training data into a plurality of non-hotspot clusters according to their topologies, where the non-hotspot clusters are associated with different topologies. The system extracts topological and non-topological critical features from the hotspot clusters and centroids of the non-hotspot clusters. The system also creates a plurality of kernels configured to identify hotspots, where each kernel is constructed using the extracted critical features of the non-hotspot clusters and the extracted critical features from one of the hotspot clusters, and each kernel is configured to identify hotspot topologies different from hotspot topologies that the other kernels are configured to identify.