Semiconductor Process Prediction Model With Defect-Filtered Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, detecting abnormalities in electrical measurement data early in the process is challenging, which affects the yield of final products.
Innovation Solution
A training method for a semiconductor process prediction model that filters out samples with physical defects during the training process, ensuring that only data from defect-free samples is used to improve the accuracy of the model for predicting electrical measurement data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If all samples including those with physical defects are used for training, then the training data volume is maximized, but the prediction accuracy for defect-free wafers deteriorates
Solution Approach 1:
The patent extracts and removes samples with physical defects from the training dataset. The filtering unit identifies defective samples based on process data and excludes them from training the prediction model, ensuring that only defect-free samples contribute to model learning, thereby resolving the contradiction between data volume and prediction accuracy.
2Reliability
If the prediction model is trained to accurately predict electrical measurement data, then the ability to detect abnormalities improves, but the model complexity increases
Solution Approach 1:
The patent applies preliminary action by filtering out defective samples before training the prediction model. This pre-processing step ensures that the model trains on clean, defect-free data, improving abnormality detection capability while avoiding the need for complex mechanisms to handle defective data during training.
3Productivity
If physical defect detection is performed earlier in the process, then yield improvement is maximized, but the detection system complexity increases
Solution Approach 1:
The patent implements self-service by using the prediction model to automatically identify wafers with physical defects based on process data. The system performs self-detection without requiring additional complex external detection equipment, thereby achieving early defect detection and yield improvement while maintaining system simplicity.
Data Source
AI summary
A training method of a semiconductor process prediction model, a semiconductor process prediction device, and a semiconductor process prediction method are provided. The training method of the semiconductor process prediction model includes the following steps. The semiconductor process was performed on several samples. A plurality of process data of the samples are obtained. A plurality of electrical measurement data of the samples are obtained. Some of the samples having physical defects are filtered out according to the process data. The semiconductor process prediction model is trained according to the process data and the electrical measurement data of the filtered samples.