Tabular Neural Network Threads for Faster Defect Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks underperform when applied on tabular data due to their design for learning hierarchical features from unstructured data types like images and text, leading to inefficiencies in defect detection in semiconductor manufacturing.
Innovation Solution
A novel Tabular Neural Network Classifier (TNNC) architecture with a collective input layer, multiple neural network threads, and dropout operations on input features, enabling improved accuracy and efficiency in classifying tabular data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional neural networks are applied on tabular data, then the model can process structured datasets, but the classification accuracy and detection precision deteriorate due to the network architecture being designed for unstructured data types
Solution Approach 1:
The neural network is segmented into multiple independent threads, each thread processing different subsets of input features. This segmentation allows the network to handle tabular data more effectively by distributing the processing load across specialized pathways, thereby improving classification accuracy while maintaining versatility with structured datasets
Solution Approach 2:
The patent introduces a new architectural dimension by organizing neural network threads in a parallel structure that is specifically suited for tabular data. This dimensional reorganization transforms the conventional single-pathway network into a multi-threaded architecture that can simultaneously process different feature combinations, resolving the mismatch between network design and data structure
2Reliability
If conventional neural networks are applied on tabular data, then the model can perform defect detection, but the processing time increases due to inefficiencies in handling structured datasets
Solution Approach 1:
By segmenting the neural network into multiple threads that process different feature subsets in parallel, the system maintains reliable defect detection capability while significantly reducing processing time. Each thread can independently analyze specific feature combinations, enabling faster convergence and more efficient classification of semiconductor defects
Solution Approach 2:
The patent applies partial action by having different neural network threads process only specific subsets of features rather than the entire feature set. This selective processing reduces the computational burden on each thread, allowing parallel execution that maintains detection reliability while decreasing overall processing time
3Productivity
If conventional neural networks are applied on tabular data, then the model can classify samples, but the true positive to false positive ratio deteriorates leading to more false detections
Solution Approach 1:
The segmentation of the neural network into multiple threads, where each thread specializes in processing specific feature subsets, improves the true positive to false positive ratio. This specialization allows each thread to develop more accurate decision boundaries for its assigned features, reducing false positives while maintaining high true positive detection rates across the entire system
Solution Approach 2:
The patent applies local quality by giving different neural network threads different levels of expertise for different feature subsets. Each thread develops specialized processing capabilities tailored to its specific features, creating local optimization that improves overall classification precision and reduces false detections while maintaining high productivity
Data Source
AI summary
The presently disclosed subject matter includes a novel computer-implemented method and computer system for the classification of tabular data using a new neural network classifier model (also referred to herein as “Tabular Neural Network Classifier” or TNNC). The disclosed method and system are characterized by improved accuracy and efficiency, as compared to other existing tabular data classification techniques such as Random Forests, XGBoost, etc. The inventor found that the TNNC exhibits in general a better TP to FP ratio in the classification output and a shorter processing time, as compared to existing tabular data classification techniques.