Decision Tree Prediction Model Signal-to-Noise Ratio
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Developing robust and generalized prediction models in semiconductor manufacturing is challenging due to high noise levels, correlated data, and a large number of input variables, which makes it difficult to identify key variables affecting yield and device performance.
Innovation Solution
A method and system that iteratively execute a decision tree-based prediction model with varying weights assigned to input variables, using a Rules Ensemble algorithm to filter out non-influential variables and identify key input variables, thereby increasing the signal-to-noise ratio and reducing the complexity of the variable list.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If advanced statistical methods are used to build high fidelity prediction models with all input variables, then model accuracy may be improved, but the complexity of the model increases and it becomes difficult to identify key variables due to high noise levels and multicollinearity
Solution Approach 1:
The patent segments the large set of input variables into two distinct groups: key variables (signal) and non-key variables (noise). This segmentation is achieved through iterative execution of prediction models with varying weights, allowing the system to identify and separate influential variables from non-influential ones. The segmentation resolves the contradiction by maintaining model accuracy through inclusion of all variables while organizing them into manageable groups that reduce apparent complexity.
Solution Approach 2:
The patent extracts key variables from the complete set of input variables by iteratively executing prediction models with different weight assignments. This extraction process isolates the signal (key variables) from the noise (non-key variables), creating a reduced variable list that maintains prediction accuracy while reducing model complexity and improving interpretability.
2Device complexity
If heuristics and past knowledge are used to reduce the input variable list, then model complexity is reduced, but the process becomes time consuming and prone to errors
Solution Approach 1:
The patent implements a self-service mechanism where the prediction model automatically identifies and ranks key variables through iterative execution with varying weights. This automated process eliminates the need for manual heuristic-based variable selection, reducing both time consumption and human error while systematically reducing variable list complexity.
3Adaptability or versatility
If all input variables are included in the prediction model, then comprehensive coverage is achieved, but the signal to noise ratio decreases making it difficult to identify key variables
Solution Approach 1:
The patent employs feedback through iterative execution of the prediction model with varying weights assigned to input variables. Each iteration provides feedback on variable importance, allowing the system to accumulate information about which variables are key and which are non-key. This feedback mechanism maintains comprehensive variable coverage while progressively improving the signal to noise ratio by identifying influential variables.
Data Source
AI summary
A computer system iteratively executes a decision tree-based prediction model using a set of input variables. The iterations create corresponding rankings of the input variables. The computer system generates overall variables contribution data using the rankings of the input variables and identifies key input variables based on the overall variables contribution data.