Semiconductor Wafer Classification With Human-Guided Model Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying semiconductor wafers based on parameter data, such as yield data, face challenges due to the lack of ground truth data, leading to uncertain and noisy clusters, and the need for improved accuracy in diagnostics and monitoring.
Innovation Solution
A method involving a model, potentially using machine learning algorithms, that sorts semiconductor wafers into subsets based on common characteristics, identifies wafers with high probability of correct allocation, compares parameter data to reference data, and reconfigures the model based on human input to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a model sorts wafers into subsets based on parameter data, then wafer classification is achieved, but accuracy is reduced due to lack of ground truth data and noisy clusters
Solution Approach 1:
The patent implements feedback by identifying wafers with high probability of misclassification and using human expert input to correct these assignments. The model continuously learns from these corrections, improving its classification accuracy over time while reducing the noise and uncertainty in clusters.
Solution Approach 2:
The patent performs preliminary classification using the model to sort wafers into subsets based on parameter data. This preliminary sorting provides a structured framework that can then be refined through human expert review of uncertain cases, combining automated efficiency with human judgment.
2Productivity
If automated model-based classification is used, then productivity is improved, but measurement precision deteriorates due to uncertain clusters
Solution Approach 1:
The patent applies partial action by having human experts review only the subset of wafers with high probability of misclassification, rather than manually reviewing all wafers. This maintains high productivity for clearly classified wafers while improving accuracy for uncertain cases through targeted human intervention.
Solution Approach 2:
Human expert corrections to misclassified wafers provide feedback that retrains the model, progressively improving its accuracy. This creates a self-improving system that maintains high productivity while reducing classification errors over time.
Data Source
AI summary
Methods and apparatus for classifying semiconductor wafers. The method can include: sorting a set of semiconductor wafers, using a model, into a plurality of sub-sets based on parameter data corresponding to one or more parameters of the set of semiconductor wafers, wherein the parameter data for semiconductor wafers in a sub-set include one or more common characteristics; identifying one or more semiconductor wafers within a sub-set based on a probability of the one or more semiconductor wafers being correctly allocated to the sub-set; comparing the parameter data of the one or more identified semiconductor wafers to reference parameter data; and reconfiguring the model based on the comparison. The comparison is undertaken by a human to provide constraints for the model. The apparatus can be configured to undertake the method.