Neural Network Latent-Space Training for Semiconductor Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semiconductor manufacturing processes face challenges in efficiently detecting defects due to the need for high precision and uniformity in ultra large-scale integration, requiring improved automated examination methods for defect detection and classification.
Innovation Solution
A system is developed to train a neural network model that transforms elements from an input space with N dimensions to a latent space with M dimensions, using a desired probability function to allocate elements into clusters, with a training loss value L determined by distance and statistical distance between the desired and actual probability functions, facilitating efficient defect detection and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional defect detection methods are used in semiconductor manufacturing, then examination can be performed on semiconductor wafers, but the detection efficiency and accuracy are insufficient for ultra large-scale integration with submicron features
Solution Approach 1:
The patent replaces traditional mechanical/optical examination systems with an artificial intelligence-based neural network system. The neural network model automatically analyzes semiconductor wafer images, substituting manual or conventional automated image processing methods. This substitution enables higher detection accuracy for submicron features while maintaining efficient processing throughput, resolving the contradiction between measurement precision and productivity.
Solution Approach 2:
The patent transforms the defect detection problem by changing the parameter space through neural network processing. Input images with N dimensions are transformed into a latent space with M dimensions through learned transformations, allowing the system to capture subtle defect patterns that traditional methods miss. This parameter transformation enables superior detection accuracy without proportionally increasing processing time.
2Reliability
If automated examination systems are implemented for defect detection, then quality standards can be maintained, but the complexity of the examination process increases
Solution Approach 1:
The neural network examination system performs multiple functions within a single unified framework: defect detection, defect classification, and quality assessment. The system can handle various semiconductor structures and defect types using the same core architecture, reducing the need for multiple specialized examination systems while maintaining comprehensive quality control.
Solution Approach 2:
The neural network system operates autonomously to maintain quality standards without requiring complex manual intervention. The model automatically adjusts its analysis based on learned patterns, performing self-optimization during operation. This self-service capability reduces the operational complexity compared to systems requiring extensive human oversight and manual parameter tuning.
3Measurement precision
If multiple examination steps are performed on semiconductor wafers, then defect detection coverage is improved, but the examination time and processing cost increase
Solution Approach 1:
The neural network model performs preliminary analysis by transforming input images into a latent space that captures essential defect characteristics. This preliminary transformation enables subsequent rapid classification and detection without requiring multiple sequential examination steps. The pre-processing in latent space prepares the data for efficient downstream analysis, reducing total examination time while maintaining comprehensive defect coverage.
Solution Approach 2:
The patent transitions examination from the original N-dimensional image space to an M-dimensional latent space, where defect patterns are more efficiently separated and analyzed. This dimensional transformation consolidates multiple examination perspectives into a unified representation, achieving comprehensive defect detection coverage in a single processing pass rather than requiring multiple sequential steps.
Data Source
AI summary
A system for training a model representing elements in the input space, each having N dimensions and being associated with images of a semiconductor specimen, to a latent space representing an equal number of elements each having M (M≤N) dimensions. The system includes a processor configured to obtain a desired probability function for transformation of the elements in the input space cluster(s) s of elements in the latent space. Then, using the desired probability function to repeatedly transform, until a specified criterion is met, elements in the input space to equal elements in the latent space in compliance with an actual probability function that is indicative of an actual allocation of the elements to the cluster(s). Lastly, determining a training loss value L associated with the elements in the latent space and testing if the training loss value L meets the specified criterion.