EDA Hotspot Prediction Using Balanced Layout Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic design automation (EDA) applications struggle to accurately predict circuit hotspots and defects in modern circuit designs due to limitations in optical rule check processes and physical inspection methods, leading to inefficiencies and reduced manufacturing yields.
Innovation Solution
Implementing a machine learning-based hotspot prediction system that processes feature vectors from layout data and correlates them with detected defects, using a hotspot processing engine to extract and balance training data for improved hotspot prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional optical rule check processes are used for hotspot detection, then the process is simple and fast, but the accuracy and reliability of defect detection deteriorates
Solution Approach 1:
The patent replaces traditional optical rule check processes with a machine learning-based detection system. The ML model processes feature vectors extracted from layout data to predict hotspot locations, substituting the mechanical/optical inspection method with an intelligent computational approach that achieves higher accuracy.
Solution Approach 2:
The patent introduces feature vectors as an intermediary representation between the circuit layout data and the machine learning model. These feature vectors capture essential characteristics of circuit fragments and serve as the bridge for the ML model to make accurate hotspot predictions without directly processing the complex layout data.
2Reliability
If physical inspection methods are used to detect manufacturing defects, then detection reliability improves, but productivity and inspection time deteriorate
Solution Approach 1:
The patent performs preliminary hotspot prediction during the design phase using machine learning models that process layout data before manufacturing occurs. By identifying potential hotspots in advance through feature vector analysis, the system enables preventive measures to be taken during manufacturing, avoiding the need for time-consuming physical inspection of the entire wafer.
Solution Approach 2:
The patent creates a computational model that copies and analyzes the essential features of circuit layouts through feature vectors. This virtual representation allows for rapid prediction of hotspot locations without requiring physical inspection equipment to examine each circuit, thereby maintaining high reliability while significantly improving inspection throughput.
3Measurement precision
If machine learning models are trained with balanced training data, then prediction accuracy improves, but data processing complexity and time deteriorate
Solution Approach 1:
The patent extracts relevant features from the complex layout data to create simplified feature vectors that capture the essential characteristics needed for hotspot prediction. By taking out only the critical features rather than processing the entire layout data, the system reduces data processing time while maintaining the accuracy needed for effective training.
Solution Approach 2:
The patent applies partial action by focusing the data processing on specific regions or features that are most relevant to hotspot detection, rather than uniformly processing the entire dataset. This selective approach to data processing reduces the time required to create balanced training data while still achieving sufficient accuracy for reliable hotspot prediction.
Data Source
AI summary
A computing system may include a hotspot processing engine and a hotspot prediction engine. The hotspot processing engine may be configured to access an input data set of hotspot locations on manufactured circuits of a circuit design, correlate the hotspot locations to layout data for the circuit design, and extract fragment feature vectors for the hotspot locations. The hotspot processing engine may further be configured to process the fragment feature vectors such that hotspot fragment feature vectors are a threshold percentage of the total number of feature vectors in the fragment feature vectors and provide the processed fragment feature vectors as a training set for training a machine-learning model. The hotspot prediction engine may be configured to apply the machine-learning model to characterize locations of the circuit design as a hotspot location or a non-hotspot location.