Charged Particle Beam Aberration Modeling From Defocus Spot Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Charged particle beam systems, such as electron microscopes, face challenges in accurately determining and correcting beam aberrations, which limit resolution due to unknown aberration coefficients and require time-consuming methods like UNO that are limited to specific correctors.
Innovation Solution
A computational model is generated using machine learning to determine actual aberration coefficients by obtaining beam spot information at various defocus settings, creating a training data set, and producing a model to accurately calculate aberration coefficients for beam correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional aberration measurement methods (e.g., UNO method) are used, then aberration coefficients can be determined, but the process is time-consuming and limited to specific aberration correctors
Solution Approach 1:
The patent applies preliminary action by pre-training a neural network model with simulated aberration data before actual measurement. The model is trained offline with various aberration scenarios, enabling rapid prediction during actual use without time-consuming traditional measurements. This transforms the time-consuming online measurement process into a fast inference process while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical/optical measurement system (traditional aberration measurement methods requiring physical adjustments and measurements) with an information processing system (neural network-based computational model). This substitution eliminates the need for time-consuming physical measurements and specific corrector hardware, allowing universal application across different systems.
2Reliability
If aberration correctors with many controls are provided, then beam aberrations can be compensated, but it becomes challenging to appropriately adjust the settings due to unknown aberration values
Solution Approach 1:
The patent implements feedback by using the neural network model to predict aberration coefficients based on actual beam spot measurements, then using these predicted values to automatically adjust the aberration corrector settings. This closed-loop feedback system eliminates the need for manual adjustment of multiple controls, as the system automatically determines the appropriate settings based on the predicted aberrations.
Solution Approach 2:
The system applies self-service by automatically determining aberration coefficients and adjusting corrector settings without requiring operator intervention. The neural network model serves itself by taking beam spot information as input and directly outputting the corrected aberration values, enabling the system to self-correct without human expertise in adjusting multiple controls.
3Measurement precision
If multiple measurements are carried out to determine aberration dependencies, then accurate correction can be achieved, but the process becomes time-consuming
Solution Approach 1:
The patent applies preliminary action by performing comprehensive simulations and training the neural network model in advance with multiple aberration scenarios and measurement conditions. This offline preparation replaces the need for multiple online measurements, as the model has already learned from diverse training data and can provide accurate predictions from a single measurement set.
Solution Approach 2:
The patent uses copying by creating a virtual model (neural network) that replicates the complex relationships between aberration parameters and beam spot characteristics. This computational copy of the physical system allows rapid prediction without repeatedly performing physical measurements, maintaining accuracy while dramatically improving efficiency.
Data Source
AI summary
A method of generating a computational model for determining one or more aberration coefficients of a charged particle beam focused by a focusing lens towards a specimen is described. The method includes obtaining a plurality of input data associated with a plurality of beam spot information for a plurality of defocus settings; obtaining a plurality of one or more aberrations coefficients associated with corresponding input data of the plurality of input data; providing a training data set including the plurality of aberrations coefficients and the corresponding input data; and producing or generating the computational model by machine learning from the training data set. The computational model is also used for determining one or more aberration coefficients.