Wafer Conveyance Abnormality Prediction Using Multi-Sensor ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing conveyance abnormality prediction systems in semiconductor manufacturing have low detection probabilities due to simple threshold-based methods, which fail to accurately identify abnormalities in sensor data from wafer handling machines, leading to potential equipment failures and productivity losses.
Innovation Solution
A conveyance abnormality prediction system utilizing a learned model that estimates abnormality based on historical sensor data from multiple sensors, including vibration, sound, and image data, using machine learning techniques such as neural networks and LSTM to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a simple threshold-based method is used to detect conveyance abnormalities, then the system is easy to operate and implement, but the detection probability of abnormalities is low
Solution Approach 1:
The patent replaces the simple threshold-based mechanical detection method with a machine learning-based system that uses neural networks and LSTM to analyze sensor data patterns, thereby improving detection probability while maintaining system operability through automated learning
Solution Approach 2:
The patent changes the detection parameters from simple threshold values to complex multi-parameter sensor data sets including vibration, sound, and image data, processed through machine learning models to improve abnormality detection accuracy
2Reliability
If multiple sensor data types are analyzed using machine learning, then the detection probability of conveyance abnormalities is improved, but the device complexity increases
Solution Approach 1:
The patent implements a multi-functional system where a single machine learning framework processes multiple sensor data types (vibration, sound, image) simultaneously, allowing the system to perform comprehensive abnormality detection without proportionally increasing complexity
Solution Approach 2:
The system uses self-service through automated machine learning models that automatically learn from historical sensor data and improve detection capabilities without requiring manual intervention, thereby managing complexity through automation
Data Source
AI summary
A conveyance abnormality prediction system includes an estimation unit that has a learned model having machine learned a relationship between a data set including sensor data outputted, at a time of substrate transport in the past, from each of a plurality of sensors provided on a substrate transport unit and a degree of conveyance abnormality at the time of the substrate transport, estimates a degree of conveyance abnormality at a time of new substrate transport by using, as an input, a data set including sensor data outputted from each of the plurality of sensors at the time of the new substrate transport, and outputs the estimated degree of conveyance abnormality.