Semiconductor Process Monitoring with Dissimilarity Regression Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor process management technologies, such as MAPC and R2R control, fail to provide intuitive guidance for process optimization and accurately consider the complex interactions between process factors, leading to difficulties in detecting and addressing abnormalities effectively.
Innovation Solution
A method and apparatus that generate a reference pattern for a normal state, calculate dissimilarity with observed data, and construct a regression tree to identify influential process factors, providing intuitive guide information for process management and optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If MAPC technology uses a neural network model to infer relationships between process factors, then the system can process complex data, but the administrator cannot obtain intuitive guidance for process optimization due to the black box nature of the model
Solution Approach 1:
The patent introduces a decision tree model as an intermediary between the complex neural network processing and the administrator. The decision tree translates the black box neural network outputs into interpretable if-then rules that provide intuitive guidance while maintaining the ability to process complex multivariate process data through the underlying neural network
2Ease of manufacture
If R2R control technology uses a simple relational expression to obtain target values, then the system is easy to implement, but it cannot accurately obtain target values and consider complex interactions between process factors
Solution Approach 1:
The patent merges the simplicity of rule-based systems with the power of neural networks by combining them into a hybrid system. The neural network processes complex multivariate relationships to generate predictions, while rule-based logic provides interpretable guidance and constraints, achieving both accuracy and ease of implementation
3Reliability
If the system observes multiple process factors to detect abnormalities, then the detection capability improves, but the complexity of analyzing the interactions between process factors increases
Solution Approach 1:
The patent segments the complex analysis task into two parts: the neural network handles the complex multivariate analysis and pattern recognition for detecting abnormalities, while the decision tree segments the results into interpretable rules that identify which specific process factors are causing issues, reducing the perceived complexity for the administrator
Data Source
AI summary
Provided is a method of managing a target process. The method performed by a process management apparatus includes: generating a reference pattern indicating a normal state based on reference observed data on a process factor measured while the target process is maintained in the normal state; obtaining observed data on the process factor measured for a specified observation period; calculating a dissimilarity between the reference pattern and the observed data; and constructing a regression tree for the target process by using the observed data and the dissimilarity, wherein the process factor is set as an independent variable of the regression tree, and the dissimilarity is set as a dependent variable of the regression tree.