Total-Reflection X-Ray Fluorescence for Rapid Contamination Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing total reflection X-ray fluorescence spectrometers struggle to accurately determine the presence of minute contamination on substrates due to low signal-to-noise ratios, especially when measurement times are reduced, making it difficult to achieve highly accurate determinations.
Innovation Solution
A total reflection X-ray fluorescence spectrometer and estimation method using machine learning to analyze spectra quickly by acquiring and learning from two types of spectra, one for a shorter duration, allowing for rapid determination of contamination presence and quantity using a convolutional neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If measurement time is reduced to increase inspection throughput, then productivity is improved, but signal-to-noise ratio decreases and measurement precision deteriorates
Solution Approach 1:
The system performs preliminary action by acquiring a first spectrum for a shorter time and using it to generate estimation data through machine learning, rather than directly making determination. This preliminary estimation enables quick screening while maintaining accuracy, resolving the contradiction between fast inspection and precise determination.
Solution Approach 2:
The machine learning estimation unit acts as an intermediary between the acquired spectrum and the final determination. It processes the spectral data and generates estimation data that facilitates accurate contamination detection even with reduced measurement time, serving as a mediator that preserves precision while enabling faster throughput.
2Productivity
If measurement time is reduced, then productivity is improved, but signal-to-noise ratio of the spectrum decreases
Solution Approach 1:
The system creates a copy of the spectral information through estimation data generated by the machine learning model. The estimation unit processes the first spectrum and generates estimation data that preserves the essential contamination information even when the original spectrum has low signal-to-noise ratio due to short measurement time.
3Productivity
If conventional analysis methods are used with short measurement time, then productivity is improved, but determination accuracy of contamination presence deteriorates
Solution Approach 1:
The patent replaces the conventional mechanical/chemical analysis system with a machine learning-based estimation system. The estimation unit uses algorithms to analyze spectral data and generate determination results, substituting traditional analytical methods with intelligent processing that maintains reliability even with reduced measurement time and lower signal quality.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables easy and quick detection of contamination on substrates with improved detection sensitivity, reducing measurement time while maintaining accuracy by utilizing machine learning to analyze spectra efficiently.
Implementation Method 1
irradiating a surface of a substrate with primary X-rays at a total reflection critical angle or less
Implementation Method 2
acquire a spectrum representing a relationship between intensities, and energies, of emitted fluorescent X-rays
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided are a total reflection X-ray fluorescence spectrometer and an estimation method which are capable of easily and quickly determining whether contamination exists on a substrate through use of a machine learning device. The total reflection X-ray fluorescence spectrometer includes: a spectrum acquisition unit configured to acquire a spectrum representing a relationship between intensities, and energies, of emitted fluorescent X-rays by irradiating a surface of a substrate with primary X-rays at a total reflection critical angle or less; and a learning unit which includes an estimation unit configured to generate estimation data on an element contained in contamination on the surface of the substrate in response to input of the spectrum, and for which learning by the estimation unit has been executed based on teacher data including the spectrum for learning and data on the element contained in the contamination on the surface of the substrate which has been used to acquire the spectrum for learning and the estimation data generated when the spectrum for learning is input to the estimation unit.