Polycrystalline Crystal Mapping With 4D-STEM and Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing crystal structure of ultrafine thin films and distinguishing between slightly different crystal grains are limited in resolution and accuracy, particularly in polycrystalline materials.
Innovation Solution
A method using a four-dimensional scanning transmission electron microscope (4D-STEM) with a parallel deep convolutional neural network (DCNN) algorithm to analyze and map crystal information of polycrystalline materials, employing unsupervised and supervised learning to generate a two-dimensional image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional EBSD or NED methods are used to analyze crystal structure, then analysis of samples with tens of nanometers thickness is possible, but samples with ultrafine thin film thickness of 10 nm or less cannot be analyzed
Solution Approach 1:
The patent changes the measurement parameters by using 4D-STEM with PACBED patterns instead of conventional EBSD or NED methods. This parameter change enables analysis of ultrafine thin films with thickness of 10 nm or less, resolving the contradiction between thin film thickness analysis capability and analysis reliability
2Manufacturing precision
If 4D-STEM method with PACBED pattern is used to analyze ultrafine thin films, then analysis of samples with thickness of 10 nm or less becomes possible, but resolution is reduced due to positional averaging
Solution Approach 1:
The patent applies segmentation by dividing the sample into multiple regions of interest (ROIs) and performing individual analysis on each ROI. This segmentation approach maintains high spatial resolution while still enabling analysis of ultrafine thin films, resolving the contradiction between thin film analysis capability and spatial resolution
3Reliability
If conventional diffraction methods are used, then crystal structure analysis is possible, but it is difficult to distinguish between same crystal grains with slightly different crystal orientations
Solution Approach 1:
The patent introduces an intermediary approach by using deep learning algorithms (CNN and GMM) as mediators between the raw diffraction data and crystal orientation determination. This intermediary processing enables precise differentiation between crystal grains with slightly different orientations while maintaining reliable crystal structure analysis
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-resolution two-dimensional mapping of crystal information in polycrystalline materials, overcoming resolution reduction and grain differentiation challenges, facilitating precise analysis of ultrafine thin films.
Implementation Method 1
acquiring a diffraction pattern by scanning an electron beam to the polycrystalline material
Data Source
AI summary
A method for two-dimensional mapping of crystal information of a polycrystalline material may include acquiring a diffraction pattern acquired by scanning an electron beam to a polycrystalline material, generating a plurality of clusters by applying a clustering algorithm to the acquired diffraction pattern based on unsupervised learning, acquiring crystal information of the polycrystalline material by applying a parallel deep convolutional neural network (DCNN) algorithm to each of the plurality of generated clusters based on supervised learning, and generating a two-dimensional image in which the acquired crystal information is mapped.