Multi-frame Image Estimation for Electron Microscopy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Charged particle beam apparatuses, such as electron microscopes, face challenges in generating high-resolution images while minimizing sample damage due to excessive beam scanning, which can cause structure shrinkage, and existing methods do not effectively address image estimation from such apparatuses.
Innovation Solution
A method involving multi-frame scanning followed by low-frame scanning, where images from both processes are used to train a learning device to estimate a multi-frame image from a low-frame image, reducing beam irradiation and associated sample damage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the electron beam is excessively scanned to generate high-resolution image, then the image quality is improved, but the sample structure may shrink due to beam irradiation
Solution Approach 1:
The patent performs preliminary multi-frame scanning to acquire reference image data before the actual measurement scanning. This preliminary action creates a dataset that captures the sample's initial state, enabling subsequent estimation algorithms to compensate for beam-induced shrinkage during the actual measurement process.
Solution Approach 2:
The patent creates a copy of the sample's image information through multi-frame scanning and uses this copied data to train estimation algorithms. These algorithms then generate estimated images that replicate the quality of multi-frame scans without requiring excessive beam irradiation during actual measurements.
2Object-affected harmful factors
If the beam scanning is performed with a small number of frames to suppress sample damage, then the sample damage is reduced, but the image quality deteriorates
Solution Approach 1:
The patent introduces an intermediary estimation algorithm that acts as a mediator between low-frame scan data and high-quality image output. This algorithm processes the limited data from low-frame scanning and generates images with quality comparable to multi-frame scans, eliminating the need to choose between sample safety and image quality.
Solution Approach 2:
The patent changes the parameter of frame number from a fixed value to a variable that can be optimized through machine learning. By training estimation algorithms on multi-frame data and applying them to low-frame data, the system achieves high image quality with reduced frame counts, effectively changing the relationship between scanning parameters and output quality.
3Measurement precision
If the multi-frame scanning is performed to generate high-quality image, then the image quality is improved, but the throughput is reduced
Solution Approach 1:
The patent performs preliminary multi-frame scanning to build training datasets and train estimation algorithms during off-hours or on reference samples. This preliminary action allows the system to develop high-quality image generation capabilities without impacting the throughput of actual production measurements, as the trained models can then rapidly generate images from low-frame scans.
Solution Approach 2:
The patent creates copies of high-quality image characteristics through estimation algorithms trained on multi-frame data. Once trained, these algorithms can rapidly generate high-quality images from low-frame input during production, copying the quality benefits of multi-frame scanning without requiring the time-consuming multi-frame acquisition process for each measurement.
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 high-precision measurement and inspection with reduced throughput impact, suppressing the influence of beam irradiation-induced shrinkage and maintaining high-speed performance.
Implementation Method 1
performing two-dimensionally scanning on an object on a sample with a beam a plurality of times
Implementation Method 2
detecting charged particles obtained by the plurality of times of two-dimensional scanning by using a detector
Implementation Method 3
training a learning device by using teacher data with the second image as an input and the first image as an output
Data Source
AI summary
The present invention relates to an image generation method for an objective for generating an image corresponding to a multi-frame image from image signals obtained by scanning a small number of frames are proposed. To achieve the above objective, there is proposed a method of performing two-dimensionally scanning on an object on a sample with a beam a plurality of times, generating a first image by integrating image signals obtained by a plurality of times of scanning at a first timing among the image signals generated based on the plurality of times of the two-dimensional scanning (S103), generating a second image based on the smaller number of times of scanning than the number of times of scanning at the first timing including scanning after the first timing (S105), training a learning device by using teacher data with the second image as an input and the first image as an output (S108), and inputting input image signals obtained by the smaller number of times of scanning than the number of times of scanning at the first timing to the trained learning device to output an estimated image.