K-Nearest Neighbor Tool Process Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The semiconductor industry faces challenges in managing and analyzing vast amounts of sensor data from tools running recipes, as existing statistical approaches like PCA struggle with non-linearity and variable duration process steps, making it difficult to transform raw data into meaningful information for users.
Innovation Solution
A k-nearest neighbor-based method is employed to create a process index, which compares tool performance to a reference fingerprint, identifying deviations and classifying tool/recipe combinations as normal or abnormal, and providing detailed sensor and process step contributions to deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If PCA-based statistical approach is used to transform raw sensor data, then data transformation capability is improved, but inability to handle non-linearity and variable duration process steps worsens
Solution Approach 1:
The patent transitions from PCA-based statistical parameters to k-nearest neighbor distance-based parameters. The process index is calculated as the average distance to k-nearest reference samples, fundamentally changing the parameter representation from linear statistical transformations to non-linear distance-based metrics that can capture complex process variations and handle variable duration steps effectively
Solution Approach 2:
The patent replaces the mechanical PCA transformation system with a k-nearest neighbor classification system. Instead of using linear algebra-based PCA to reduce dimensions, the system uses distance calculation in the original feature space, substituting a mathematical transformation approach with a pattern recognition approach that naturally handles non-linearity
2Quantity of substance
If vast amounts of sensor data are collected for analysis, then data completeness is improved, but difficulty in managing and analyzing the data worsens
Solution Approach 1:
The patent extracts only the essential information needed for process assessment by calculating a single process index value that represents the overall similarity to reference processes. Instead of managing and analyzing all raw sensor data points, the system extracts the key discriminative feature (distance to k-nearest neighbors) that captures process quality, significantly reducing data management complexity
Solution Approach 2:
The patent segments the analysis process into distinct components: reference data collection, k-nearest neighbor identification, distance calculation, and process index computation. This segmentation allows the system to handle large volumes of sensor data by processing them through modular steps, where each step operates on manageable data subsets rather than the complete dataset simultaneously
3Loss of information
If detailed sensor data is provided for analysis, then information completeness is improved, but user understanding of complex processes worsens
Solution Approach 1:
The patent introduces the process index as an intermediary between detailed sensor data and user understanding. The process index serves as a simplified metric that mediates the complex relationship between multiple sensor readings and process quality, providing users with an intuitive single-value assessment that preserves the essential information from detailed sensor data without requiring users to interpret complex multi-parameter datasets
Data Source
AI summary
A server computer system creates a reference fingerprint for a first tool running a recipe. The server computer system uses reference data that pertains to the first tool performing within parameters to create the reference fingerprint. The reference fingerprint includes a target baseline and allowable ranges based on the target baseline. The server computer system determines whether sample data that is associated with a second tool running the recipe is performing within the parameters based on a comparison of the sample data to the reference fingerprint. The second tool can be the first tool or another tool of a same type as the first tool. The server computer system provides a classification of the second tool to at least one of a system or a user based on the comparison of the sample data to the reference fingerprint. The classification indicates whether the second tool is performing within the parameters.


