Electron Microscope Aberration Estimation From Diverse Ronchigrams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating aberration values in electron microscopes, particularly in scanning transmission electron microscopes, face challenges in precision due to variations in Ronchigram content caused by aberrations, defocus, illumination center position, and magnification, leading to inconsistent calculation accuracy.
Innovation Solution
A method involving machine learning is employed to generate an estimation model by simulating various Ronchigram conditions and supplying training data to improve the model's accuracy, allowing for precise estimation of aberration values, including intrinsic aberrations that cannot be corrected by the aberration corrector.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods (SRAM method) are used to calculate aberration values from Ronchigrams, then the calculation process can be completed, but the estimation precision is insufficient due to significant variations in Ronchigram content caused by different aberrations, defocus, illumination center position, and magnification
Solution Approach 1:
The patent applies preliminary action by generating a large number of simulated Ronchigrams covering various aberration conditions, defocus values, illumination center positions, and magnifications before actual measurement. These pre-generated training data enable the machine learning model to learn the relationships between Ronchigram patterns and aberration values across diverse conditions, thereby improving estimation precision without requiring complex real-time calculations.
Solution Approach 2:
The patent utilizes parameter changes by systematically varying multiple parameters during simulation (aberration values, defocus, illumination center position, magnification) to generate diverse training data. The machine learning model learns to estimate aberration values by recognizing patterns across these varying parameters, enabling it to handle various Ronchigrams under different conditions with consistent precision.
2Measurement precision
If machine learning methods are applied to improve aberration value estimation precision, then the estimation model can handle various Ronchigrams, but the complexity of the system increases due to the need for simulation and training data generation
Solution Approach 1:
The patent applies copying by creating simulated Ronchigrams that replicate the characteristics of actual Ronchigrams under various conditions. These copied training data allow the machine learning model to learn from synthetic examples that mirror real measurement scenarios, improving estimation precision without requiring complex physical experimentation or additional hardware.
Solution Approach 2:
The patent implements self-service by using the electron microscope system itself to generate the training data through simulation. The system creates its own training dataset by modeling the physical processes and generating corresponding Ronchigrams, eliminating the need for external data sources or complex experimental setups.
Data Source
AI summary
An aberration value estimator has a learned estimation model for estimating an aberration value set based on a Ronchigram. In a machine learning sub-system, a simulation is repeatedly executed while changing a simulation condition, and calculated Ronchigrams are generated in a wide variety and in a large number. By machine learning using the calculated Ronchigrams, the learned estimation model is generated.