Decision Tree Substrate Partitioning for Semiconductor Process Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, it is challenging to accurately predict and control post-processing data based on pre-processing data due to the complexity of relating numerous pre-processing parameters to post-processing characteristics, especially in high-volume manufacturing where many parameters and configurations are involved.
Innovation Solution
A method using a recursive decision tree algorithm to partition pre-processing data into subsets based on characteristics of post-processing data, enabling the creation of a reliable model for predicting post-processing outcomes and optimizing process corrections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a reliable model linking pre-processing data to post-processing data is constructed, then prediction accuracy of post-processing characteristics is improved, but the complexity of data processing and model construction increases due to the large amount of pre-processing parameters
Solution Approach 1:
The patent segments the large set of pre-processing parameters into multiple subsets based on their relevance to specific post-processing characteristics. By dividing the parameter space into manageable groups, the system can construct multiple specialized models for different parameter subsets rather than attempting to model all parameters simultaneously, thereby reducing the complexity of individual models while maintaining overall prediction accuracy.
Solution Approach 2:
The patent introduces a new dimension of analysis by creating a hierarchical model structure where parameters are organized across multiple levels of abstraction. Instead of treating all parameters at the same level, the system creates a multi-dimensional framework that separates parameters by their influence on different post-processing characteristics, enabling more efficient model construction and reducing the effective dimensionality of each individual prediction task.
2Manufacturing precision
If control parameters are customized for individual substrates based on pre-processing data, then manufacturing precision is improved, but the processing time and complexity increase
Solution Approach 1:
The patent performs preliminary analysis and model construction during the pre-processing phase, before actual manufacturing begins. By pre-segmenting parameters, pre-constructing models for different parameter subsets, and pre-determining the relationships between pre-processing data and post-processing characteristics, the system eliminates the need for time-consuming real-time analysis during manufacturing, thus maintaining high precision while reducing processing time.
Solution Approach 2:
The patent creates simplified representative models (copies) of the complex relationships between pre-processing parameters and post-processing characteristics. These simplified models can be quickly applied to individual substrates without requiring full analysis of all parameters, enabling rapid customization of control parameters while maintaining manufacturing precision.
3Adaptability or versatility
If all pre-processing parameters are considered in the control model, then the comprehensiveness of the model is improved, but the difficulty of model construction and data management increases
Solution Approach 1:
The patent segments the comprehensive set of pre-processing parameters into multiple focused subsets, each associated with specific post-processing characteristics. This segmentation allows the system to maintain model comprehensiveness by covering all parameters across multiple subsets while reducing the construction difficulty of each individual subset model, as each model deals with a smaller, more manageable set of parameters.
Solution Approach 2:
The patent creates a universal framework and methodology that can be applied to any set of pre-processing parameters and post-processing characteristics. The segmentation approach and model construction method are designed to be adaptable to different parameter types and manufacturing processes, providing a multi-functional solution that maintains comprehensiveness while simplifying construction through standardized procedures.
Data Source
AI summary
Substrates to be processed are partitioned based on pre-processing data that is associated with substrates before a process step. The data is partitioned using a partition rule and the substrates are partitioned into subsets in accordance with subsets of the data obtained by the partitioning. Corrections are applied, specific to each subset. The partition rule is obtained using decision tree analysis on a training set of substrates. The decision tree analysis uses pre-processing data associated with the training substrates before they were processed, and post-processing data associated with the training substrates after being subject to the process step. The partition rule that defines the decision tree is selected from a plurality of partition rules based on a characteristic of subsets of the post-processing data. The associated corrections are obtained implicitly at the same time.