Semiconductor Lot Anomaly Detection Using ML-Based Hold Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional statistical methods for detecting anomalies in semiconductor manufacturing are time-consuming, prone to errors, and fail to identify defects within acceptable limits until a significant time has passed, leading to delayed detection and resource wastage.
Innovation Solution
Implementing machine learning techniques, specifically using algorithms like isolation forest, kernel density estimation, and local outlier factor, to dynamically set lower anomaly detection limits, enabling earlier identification of defective lots and automating the holding of such lots for real-time defect addressing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If traditional statistical methods are used for anomaly detection, then the detection process is simple and easy to implement, but the detection time is delayed and defects are not identified until a significant time has passed
Solution Approach 1:
The patent replaces traditional statistical methods with machine learning models (isolation forest, kernel density estimation, local outlier factor algorithms) to detect anomalies. This substitution enables earlier identification of defective lots by learning complex patterns from historical data, reducing detection time while accepting increased system complexity through automated ML pipelines
Solution Approach 2:
The system performs preliminary anomaly detection using machine learning models before final product completion. By training models on historical data and applying them to detect deviations early in the manufacturing process, the system identifies defects at an earlier point in time, preventing waste of resources on obviously defective products
2Reliability
If traditional statistical limits are used, then the manufacturing process remains stable and predictable, but defective lots are not identified until they exceed acceptable limits
Solution Approach 1:
The patent changes the detection parameters by using multiple machine learning algorithms (isolation forest, kernel density estimation, local outlier factor) instead of single statistical limits. Each algorithm provides different perspectives on anomaly detection, and their results are combined to improve both reliability and precision in identifying defective lots before they exceed traditional acceptable limits
Solution Approach 2:
The system implements feedback mechanisms where machine learning models continuously learn from detected anomalies and adjust their detection thresholds. The models are trained on historical data and refined based on actual defect patterns, improving their ability to reliably identify defective lots with high precision while maintaining manufacturing stability
3Productivity
If machine learning models are implemented for early anomaly detection, then defective lots are identified earlier and more accurately, but the system complexity and computational resources increase
Solution Approach 1:
The patent segments the anomaly detection task into multiple independent machine learning models (isolation forest for outlier detection, kernel density estimation for probability assessment, local outlier factor for neighborhood-based analysis). Each model handles specific aspects of defect detection, improving overall productivity through specialized detection while managing complexity by dividing the system into modular, independently trainable components
Solution Approach 2:
The machine learning models are designed to automatically train on historical data and perform detection without continuous manual intervention. Once trained, the models self-service by continuously monitoring manufacturing data and flagging anomalies, improving productivity while the initial complexity investment pays off through automated operation
Data Source
AI summary
According to certain aspects, one or more processors can be configured to: determine a limit for detecting a lot associated with a specified product as an anomaly based on one or more machine learning models, the limit for detecting a lot associated with the specified product as an anomaly enabling a semiconductor manufacturing system to identify one or more defective lots at an earlier point in time than using another limit associated with the specified product determined based on a statistical method, and to identify one or more defective lots that do not satisfy the other limit based on the statistical method; in response to a failure rate of a first lot in connection with the parameter satisfying the limit, identify the first lot as an anomaly and automatically hold the first lot in order to address defects associated with the first lot in real time.