Semiconductor Machine Quality Prediction Using Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional quality assurance methods, such as Statistical Process Control (SPC), are inadequate for predicting defects in complex manufacturing processes due to their reliance on linear models and manual threshold setups, and AI-based solutions are limited by the need for balanced data, leading to a lack of predictive management in semiconductor manufacturing.
Innovation Solution
An AI-based system using neural networks trained through multi-task and transfer learning to model semiconductor production machines, providing real-time defect predictions by integrating production, maintenance, and quality data, leveraging cloud and on-premises deployment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional Statistical Process Control (SPC) methods are used for quality assurance, then the system is simple to implement, but it cannot reliably anticipate defects in complex manufacturing processes
Solution Approach 1:
The patent transforms the quality assurance approach by changing from traditional SPC parameters to AI-based parameters including neural network models, multi-task learning frameworks, and transfer learning techniques. This enables the system to capture complex non-linear relationships in manufacturing data, significantly improving defect prediction accuracy while managing complexity through automated threshold generation and cloud-based model deployment.
2Reliability
If advanced AI techniques such as neural networks are implemented for quality prediction, then modeling accuracy for complex systems improves, but the requirement for vast amounts of balanced data increases
Solution Approach 1:
The patent applies preliminary action by implementing data augmentation techniques and synthetic data generation methods before model training. This prepares sufficient training data in advance, enabling neural networks to achieve high modeling accuracy even when actual production data is limited or imbalanced. The system pre-processes and enriches available data to meet the requirements of advanced AI techniques.
Solution Approach 2:
The patent introduces transfer learning as an intermediary mechanism that bridges the data gap. By training models on data from similar machines or processes and then adapting them to the target system, the approach enables effective modeling with limited direct production data, reducing the quantity requirement while maintaining high accuracy.
3Ease of operation
If manual threshold setup is used in traditional process control, then the system is easier to understand, but it complicates practical use and reduces automation
Solution Approach 1:
The patent implements self-service by enabling the AI system to automatically generate, optimize, and adjust prediction thresholds without manual intervention. The neural network models autonomously learn optimal decision boundaries from training data, and the system automatically updates thresholds as new data becomes available. This maintains high ease of operation through automated model deployment and continuous learning, eliminating the need for manual threshold configuration while preserving interpretability through automated explanation mechanisms.
Data Source
AI summary
The subject matter herein provides for AI-based prediction of production defects in association with a production system, such as a semiconductor manufacturing machine. In one embodiment, a method begins by receiving production data from the production system. The production data typically comprises non-homogeneous machine parameters and maintenance data, quality test data, and product and process data. Using the production data, a neural network is trained to model an operation of a given machine in the production system. Preferably, the training involves multi-task learning, transfer learning (e.g., using knowledge obtained with respect to a machine of the same type as the given machine), and a combination of multi-task learning and transfer learning. Once the model is trained, it is associated with the given machine operating environment, wherein it is used to provide quality assurance predictions.