Equipment Parameter Model via Semi-Best Subset Algorithm
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex electric device manufacturing, identifying key process steps and equipment parameters to improve yield rates is challenging due to the numerous variables involved, requiring an effective method for creating equipment parameter models that accurately represent relationships between process steps and outputs.
Innovation Solution
A method using a semi-best subset (SBS) algorithm to analyze process outputs by iteratively selecting and updating regression models based on correlation rankings until convergence, creating equipment parameter models that represent relationships between equipment parameters and outputs, thereby identifying key process and equipment parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all process steps and equipment parameters are analyzed to identify key factors, then the accuracy of yield rate improvement is improved, but the computational complexity and time required increase significantly
Solution Approach 1:
The patent segments the analysis process into multiple iterative steps, where in each step only a subset of process steps or equipment parameters is selected for regression model creation. The subset selection is based on correlation calculations from previous steps, allowing the system to gradually focus on the most significant factors without analyzing all variables simultaneously, thus reducing computational complexity while maintaining identification accuracy.
Solution Approach 2:
The patent applies partial action by creating regression models for only a selected subset of process steps or equipment parameters in each iteration, rather than analyzing all variables. The subset is determined by correlation rankings from previous iterations, ensuring that the most influential factors are captured while avoiding the computational burden of complete analysis.
2Reliability
If regression models are created for all combinations of process steps and equipment parameters, then the completeness of the equipment parameter model is improved, but the manufacturing time and computational resources increase
Solution Approach 1:
The patent performs preliminary correlation calculations on all process steps and equipment parameters before creating regression models. Based on these preliminary results, a subset of high-correlation variables is selected for model creation in each iteration. This preliminary filtering action reduces the number of regression models needed while ensuring that the most significant factors are included, thus maintaining model completeness with reduced manufacturing time.
Solution Approach 2:
The patent employs a dynamic iterative approach where the set of variables included in regression models changes across iterations. In each iteration, new variables are added to the model based on correlation rankings, and the model is re-evaluated. This dynamic process allows the system to progressively build a complete model while controlling computational resources by focusing on high-impact variables at each stage.
3Measurement precision
If multiple regression models are created to ensure comprehensive coverage of process variables, then the accuracy of process output analysis is improved, but the convergence time of the iterative process increases
Solution Approach 1:
The patent incorporates feedback mechanisms where the correlation calculations from each iteration inform the variable selection for the next iteration. The system uses the results of previous regression models to update the correlation rankings and select new variables for inclusion. This feedback loop ensures that each iteration builds upon previous insights, converging to an accurate model more efficiently by avoiding redundant analysis of low-impact variables.
Data Source
AI summary
A method for analyzing a process output and a method for creating an equipment parameter model are provided. The method for analyzing the process output includes the following steps: A plurality of process steps are obtained. A processor obtains a step model set including a plurality of first step regression models, each of which represents a relationship between N of the process steps and a process output. The processor calculates a correlation of each of the first step regression models. The processor picks up at least two of the first step regression models to be a plurality of second step regression models whose correlations are ranked at top among the correlations of the first step regression models. The processor updates the step model set by a plurality of third step regression models, each of which represents a relationship between M of the process steps and the process output.