TEM SADP Image Interconversion for Realistic Diffraction Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional simulation programs for generating TEM SADP images fail to account for the effects of beam stoppers, electron beam misalignment, optical system errors, and variations between TEM manufacturers, leading to poor image quality and errors in diffraction pattern representation.
Innovation Solution
A system and method using deep learning to interconvert synthetic and real TEM SADP images, refining diffraction patterns to remove unnecessary information, and generating adaptive images that prevent ringing effects and blurred diffraction points, while simulating errors and manufacturer-specific variations using input parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional simulation programs are used to generate TEM SADP images, then the generation process is simple and fast, but the image quality is poor and does not reflect real experimental conditions
Solution Approach 1:
A deep learning model is introduced as an intermediary between the simple simulation program and the real experimental conditions. The model learns the mapping from synthetic images to real images by training on paired datasets, effectively mediating the transformation while keeping the original simple simulation program intact.
Solution Approach 2:
Instead of modifying the complex physical simulation processes, the invention creates a learned copy or representation of the transformation from synthetic to real images. The deep learning model captures the essential differences and applies them as a post-processing step, avoiding the need to replicate complex physical phenomena.
2Manufacturing precision
If beam stopper is used to block the brightest point in real TEM SADP image, then the dynamic range is improved, but conventional simulation programs cannot simulate this effect
Solution Approach 1:
The beam stopper effect is pre-applied to the synthetic images during the training phase. By incorporating this preprocessing step that simulates the beam stopper's physical effect, the model learns to generate images that already account for this important experimental condition.
3Manufacturing precision
If multiple errors and variations are considered in simulation, then the image quality and realism are improved, but the computational complexity and processing time increase
Solution Approach 1:
All the complex error simulations and variations are performed in advance during the training phase to create the ground truth real images. During actual use, the model applies these learned transformations instantly through a single forward pass, avoiding repeated complex computations.
Solution Approach 2:
The invention replaces the mechanical/computational process of simulating multiple physical errors and variations with a learned neural network model. The complex physical simulations are substituted by pattern recognition and transformation through the deep learning architecture.
4Measurement precision
If deep learning is used for interconversion, then the image quality and accuracy are improved, but the computational resources and processing time are increased
Solution Approach 1:
The computationally intensive deep learning model is trained in advance on a dataset of paired synthetic and real images. Once trained, the model can perform interconversions efficiently with minimal computational resources, as the heavy lifting of learning the transformation has already been completed.
Data Source
AI summary
A system and a method of generating adaptively a TEM SADP image with high discernment according to inputted parameters are disclosed. The system for converting a diffraction pattern image includes a real diffraction pattern image refining unit configured to remove unnecessary information from a real diffraction pattern image; a synthetic diffraction pattern generating unit configured to obtain a synthetic diffraction pattern image corresponding to the real diffraction pattern image in which the unnecessary information is removed; and a real-synthetic interconversion algorithm learning unit configured to generate an image belonging to a real diffraction pattern domain from an image belonging to a synthetic diffraction pattern domain or generate an image belonging to the synthetic diffraction pattern domain from an image belonging to the real diffraction pattern domain by using at least one of the real diffraction pattern image in which the unnecessary information is removed and the synthetic diffraction pattern image.