3D Reconstruction Slice Imaging With Selective Map Acquisition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D reconstruction techniques in charged particle microscopes are time-consuming when acquiring multiple data types, such as images, compositional, and crystalline data, especially for large sample volumes, leading to excessive processing times and potential missed information due to periodic acquisition methods that are agnostic to sample changes.
Innovation Solution
A method using a slice-and-view technique with neural networks to determine significant changes in sample surfaces, acquiring maps only when necessary, and interpolating missing data using reference surfaces and images, allowing for efficient generation of multi-data set 3D reconstructions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If elemental information is acquired at every n surfaces while imaging each surface, then process time is reduced, but the elemental data coverage is limited and may not capture changes to the sample
Solution Approach 1:
The system performs preliminary imaging of all surfaces first, then uses neural networks to analyze the images and identify which surfaces require elemental mapping based on detected changes. This preliminary action allows the system to plan the elemental data acquisition strategy before actually acquiring the elemental data, ensuring both speed and completeness
Solution Approach 2:
The system uses neural networks to provide feedback by analyzing images and determining whether significant changes have occurred between surfaces. This feedback mechanism dynamically decides whether to acquire elemental data at each surface, optimizing both the speed and completeness of data acquisition
2Loss of information
If elemental information is acquired at every surface while imaging, then complete data coverage is achieved, but process time increases into days or weeks
Solution Approach 1:
Instead of acquiring elemental data at every surface (excessive action), the system acquires elemental data only at surfaces where the neural network detects significant changes (partial action). This selective approach maintains complete data coverage for changed regions while dramatically reducing overall process time
Solution Approach 2:
The system transitions from a static, predetermined sampling strategy to a dynamic strategy where the decision to acquire elemental data at each surface is made in real-time based on neural network analysis of image changes. This dynamic adaptation optimizes the balance between data completeness and processing speed
3Productivity
If periodic acquisition of elemental data is used, then process time is reduced, but the method is agnostic to sample changes and may miss information
Solution Approach 1:
The neural network provides continuous feedback by analyzing each image and comparing it to previous surfaces, dynamically determining whether elemental data acquisition is needed. This feedback mechanism ensures that data is acquired reliably at all surfaces where changes occur, regardless of the periodic sampling schedule
Solution Approach 2:
The system uses its own imaging capability and neural network analysis to automatically identify when elemental data is needed, making the determination internally based on actual sample changes rather than relying on external periodic scheduling. This self-service approach ensures both efficiency and reliability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Apparatuses and processes for generating data for three-dimensional reconstruction are disclosed herein. An example method at least includes exposing a subsequent surface of a sample, acquiring an image of the subsequent surface, comparing the image of the subsequent surface to an image of a reference surface, based on the comparison exceeding a threshold, acquiring a compositional or crystalline map of the subsequent surface, and based on the comparison not exceeding the threshold, exposing a next surface. Such method of the type slice-and-view allows to reduce the overall processing time of the reconstruction since the compositional or crystalline map of a subsequent surface is measured only if the corresponding image shows significant changes. Artificial neural networks are used to propagate the compositional or crystalline map information for slices where no map has been measured.