Circuit Hotspot Down Selection Using Layout-Based Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current hotspot detection processes in circuit design suffer from low precision in identifying candidate hotspot locations, leading to high false positive rates and inefficient use of high precision verification techniques due to inadequate down selection strategies.
Innovation Solution
Implementing a machine learning-based system that extracts feature vectors from layout data and correlates them with detected hotspot locations, using data balancing techniques to enhance the accuracy and efficiency of down selecting candidate hotspot locations for further verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high precision verification techniques are applied to all candidate hotspot locations, then defect detection accuracy is improved, but processing time and resource consumption increase significantly
Solution Approach 1:
The verification process is segmented into two stages: (1) initial screening using low precision but high throughput imaging to generate candidate hotspot locations, and (2) focused verification using high precision techniques only on down-selected candidates. This segmentation resolves the contradiction by applying different precision levels to different subsets of data, achieving both efficiency and accuracy.
Solution Approach 2:
Different verification qualities are applied to different locations: high precision verification is applied locally only to selected candidate hotspot locations, while low precision screening is applied globally to all potential locations. This local quality approach allows the system to maintain high defect detection accuracy for critical areas while avoiding unnecessary processing time on non-critical areas.
2Measurement precision
If high precision imaging processes are used to inspect entire chip wafers, then defect detection precision is improved, but bandwidth and resource requirements become prohibitive
Solution Approach 1:
A preliminary low precision imaging step is performed before the high precision imaging step to identify and down-select candidate hotspot locations. This preliminary action filters out the majority of non-defective areas, allowing the subsequent high precision imaging to focus only on promising candidates, thereby maintaining high detection precision while improving overall inspection throughput.
Solution Approach 2:
Instead of applying high precision imaging to the entire wafer (excessive action), the system applies it partially only to selected candidate locations. The low precision imaging performs the screening function that would otherwise require full high precision coverage, enabling the high precision step to be performed on a limited subset and thus improving productivity while maintaining precision where needed.
3Productivity
If current down selection strategies are used, then processing efficiency is maintained, but false positive rates remain high leading to inadequate verification
Solution Approach 1:
The system uses feedback from the low precision imaging results to guide the high precision verification process. Candidate locations identified by the low precision imaging are fed into the high precision verification step, creating a feedback loop that continuously refines the selection process. This feedback mechanism improves verification reliability by ensuring that high precision resources are directed to locations most likely to contain defects, while maintaining processing efficiency through the initial filtering stage.
Data Source
AI summary
A method may include the steps of accessing an input data set of hotspot locations on manufactured circuits of a circuit design. The hotspot locations may be confirmed through a high precision imaging process from a set of candidate locations of the circuit design determined by a low precision imaging process. The method may further include correlating the hotspot locations to layout data for the circuit design, extracting fragment feature vectors for the hotspot locations from optical proximity correction (OPC) fragments of the layout data, processing the fragment feature vectors, providing the processed fragment feature vectors as a training set for training a machine-learning model, and applying the machine-learning model to down select a different set of candidate locations determined by the low precision imaging process.