Neural Network Yield Prediction for Semiconductor Manufacturing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor manufacturing yield prediction methods rely solely on main process measurement data, leading to reduced accuracy due to the complexity of influencing factors, necessitating a more comprehensive approach.
Innovation Solution
A system and method utilizing machine learning with different neural network models classified by data types, including production, equipment, fault, and measurement data, to predict semiconductor manufacturing yield, allowing for real-time data input and integration of various data types for accurate prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only measurement data from main processes is used for yield prediction, then the prediction system remains simple, but prediction accuracy deteriorates due to the complexity of influencing factors
Solution Approach 1:
The patent segments the prediction system into multiple specialized neural network models, each dedicated to processing a specific data type (production data, equipment data, fault data, measurement data). This segmentation allows each model to optimize for its specific data characteristics while collectively achieving comprehensive yield prediction, resolving the contradiction between accuracy and complexity.
Solution Approach 2:
The patent creates a universal prediction framework that integrates multiple data types through a common architecture. The plurality of neural network models work together within a unified system that can handle diverse data formats and characteristics, achieving multi-functionality while maintaining system coherence and managing complexity.
2Measurement precision
If multiple types of data are integrated for yield prediction, then prediction accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent applies local quality by assigning different neural network models to different data types based on their specific characteristics. Each model is tailored to process its designated data type with appropriate algorithms and parameters, ensuring optimal processing quality for each local data category while simplifying the overall processing architecture.
Data Source
AI summary
Provided is a machine learning-based semiconductor manufacturing yield prediction system and method. A result prediction method according to an embodiment of the present invention comprises: learning different neural network models by classifying different types of data according to their types and respectively inputting the classified different types of data to the different neural network models; and predicting result values by classifying input data according to their types and respectively inputting the classified input data to different neural network models. Therefore, it is possible to apply different neural network models to respective data according to their types, thereby ensuring a neural network model having a structure appropriate for the characteristics of each type of data and thus accurately predicting a result value.