Feature-Value Clustering for Semiconductor Defect Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual inspection methods for defect detection in semiconductor manufacturing vary in accuracy among individuals and are time-consuming when dealing with large volumes of image data, and existing neural network-based fault identification techniques suffer from reduced accuracy when multiple faults overlap.
Innovation Solution
A classification device employing a memory unit, processing unit, and classifier, utilizing a discriminative model with convolutional and fully connected layers to extract feature values from image data, followed by clustering to improve defect identification and classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual inspection is performed by individuals, then defect detection can be performed, but accuracy varies among individuals and time consumption increases with large volumes of image data
Solution Approach 1:
The patent replaces manual visual inspection with an automated machine learning system. A neural network model processes image data to identify defects, eliminating human variability and significantly reducing processing time while maintaining consistent accuracy across large volumes of data.
Solution Approach 2:
The system enables self-service defect detection through automated machine learning models that process and analyze images without human intervention. The model independently identifies defects, determines their types, and provides results, making the inspection process autonomous and efficient.
2Productivity
If neural network-based fault identification is used, then defect identification is automated, but accuracy is reduced when multiple faults overlap
Solution Approach 1:
The patent segments the fault identification process into multiple specialized neural network models. Each model is trained to detect specific defect types or characteristics, allowing the system to handle overlapping faults more effectively by dividing the complex task into manageable, specialized components.
Solution Approach 2:
The system changes parameters by using multiple neural network models with different configurations and training focuses. By varying the model parameters and approaches, the system can accurately identify different types of defects even when they overlap, overcoming the limitations of a single generic model.
3Ease of manufacture
If information on faults is obtained using traditional image processing, then fault identification can be performed, but accuracy is lowered when faults overlap or exist in the same region
Solution Approach 1:
The patent replaces traditional image processing methods with machine learning-based neural network models. This substitution enables the system to automatically learn and identify fault patterns, including overlapping faults, without requiring complex manual processing rules, thereby maintaining simplicity while improving accuracy.
Data Source
AI summary
A novel classification device is provided. The classification device includes a memory unit, a processing unit, and a classifier. A plurality of pieces of image data and a discriminative model are stored in the memory unit. Each of the plurality of pieces of image data is image data determined to contain a defect. The discriminative model includes an input layer, an intermediate layer, and an output layer. First to n-th (n is an integer greater than or equal to 2) image data of the plurality of pieces of image data are supplied to the processing unit. The processing unit has a function of outputting feature values of the first to the n-th image data (a first to an n-th feature value) on the basis of the discriminative model. A feature value output from the processing unit is a numerical value of a neuron included in the intermediate layer. The first to the n-th feature value output from the processing unit are supplied to the classifier. The classifier has a function of performing clustering of the first to the n-th image data on the basis of the first to the n-th feature value.