Embedded ML Model Segmentation for Wafer Defect Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in efficiently implementing machine learning models on embedded systems for defect recognition in semiconductor wafer imaging datasets, requiring significant programming effort and time, with limited flexibility and high computational resources.
Innovation Solution
A method involving the separation of model architecture from model data, where the architecture is implemented on the embedded system, while the data is dynamically updated, allowing for adaptable and reusable models across varying imaging datasets, utilizing modular architectures like neural networks partitioned into head and tail modules for flexible implementation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are implemented on embedded systems for defect recognition, then defect recognition capability is improved, but programming effort and implementation time increase significantly
Solution Approach 1:
The patent segments the machine learning model into two distinct parts: a fixed model architecture implemented on the embedded system hardware, and dynamic model data (weights and biases) stored in memory. This segmentation allows the time-consuming architecture design to be done once offline, while only the model data needs to be updated and transferred for different defect recognition tasks, significantly reducing on-site programming effort.
Solution Approach 2:
The model architecture is designed and configured in advance during system development, with all computational graphs, layer structures, and operation sequences predetermined. This preliminary action eliminates the need for complex real-time programming during deployment, as the embedded system only needs to load pre-configured architecture definitions and corresponding model data for specific defect recognition applications.
2Reliability
If machine learning models are implemented on embedded systems, then defect recognition is achieved, but device complexity and computational resource requirements increase
Solution Approach 1:
The patent implements dynamic model data loading and switching capabilities, allowing the embedded system to adapt to different defect recognition tasks by loading appropriate pre-trained model weights without hardware reconfiguration. This dynamic approach reduces the need for multiple fixed hardware implementations, thereby reducing overall device complexity while maintaining versatile defect recognition capabilities.
3Reliability
If traditional machine learning implementation approaches are used, then defect recognition can be performed, but flexibility and adaptability to varying imaging datasets are limited
Solution Approach 1:
Multiple model data sets for different defect types and imaging conditions are pre-trained and stored in the system memory beforehand. When a new imaging dataset is encountered, the system can quickly switch between pre-trained models or load the appropriate model data without requiring retraining, thereby achieving high adaptability while maintaining recognition performance.
Data Source
AI summary
A computer implemented method for defect recognition in an imaging dataset of a wafer in a charged particle beam system comprising an embedded system, the method comprising: i) obtaining an imaging dataset of a wafer; ii) obtaining model data for a model architecture of a machine learning model for defect recognition in the imaging dataset of the wafer, the model architecture being implemented in the embedded system; iii) transferring the model data to a programmable memory of the embedded system; and iv) applying the machine learning model to an imaging dataset of a wafer to recognize defects, comprising executing the embedded system implemented model architecture with the transferred model data.