ML-Based EM Response Model for X-Ray Scatterometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current metrology techniques face challenges in accurately measuring complex semiconductor structures due to increasing complexity, small resolution requirements, and the use of opaque materials, leading to high computational effort and accuracy issues in characterizing parameters like those in FinFETs and 3D FLASH devices.
Innovation Solution
A machine learning-based electromagnetic response model is used to estimate parameter values from X-ray scatterometry measurements, replacing traditional three-dimensional structure discretization and physics-based simulations, reducing computational effort while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional physics-based electromagnetic simulation is used for X-ray scatterometry measurements, then measurement accuracy is maintained, but computational effort becomes excessively high for complex semiconductor structures
Solution Approach 1:
The patent pre-computes electromagnetic response metrics for a comprehensive set of parameter combinations before actual measurements. This creates a lookup table or training dataset that stores pre-calculated scattering patterns for various structural parameters, eliminating the need for real-time physics-based simulations during measurement and estimation processes.
Solution Approach 2:
The patent creates simplified representations or surrogate models that copy the essential input-output relationships of complex physics-based simulations. These surrogate models capture the mapping between structural parameters and electromagnetic responses without requiring full physics-based computation, enabling fast parameter estimation while maintaining accuracy.
2Loss of information
If more parameters are used to characterize complex structures like FinFETs, then measurement completeness improves, but parameter correlation increases making reliable decoupling difficult
Solution Approach 1:
The patent segments the complex structure into distinct regions or layers (e.g., different material layers in FinFETs or 3D FLASH structures) and associates specific electromagnetic response metrics with each segment. This segmentation allows independent characterization of different structural components, reducing parameter correlation and enabling more reliable decoupling of structural parameters.
Solution Approach 2:
The patent transforms the parameter space by selecting and emphasizing specific parameters that have strong influence on measurable electromagnetic responses. By changing the parameter representation and focusing on independently influential parameters, the patent reduces correlation between parameters while maintaining complete structural characterization.
3Length of stationary object
If longer wavelengths are used to penetrate deep into 3D FLASH structures, then depth of penetration improves, but sensitivity to small anomalies and light intensity decrease
Solution Approach 1:
The patent employs a multi-wavelength approach where multiple illumination wavelengths are used to probe different depths and features of the structure. Shorter wavelengths provide sensitivity to surface and near-surface features, while longer wavelengths penetrate deeper into the structure. By combining measurements at multiple wavelengths, the patent achieves both deep penetration and high sensitivity to small anomalies simultaneously.
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
This approach enables accurate X-ray scatterometry measurements of complex semiconductor structures with significantly less computational effort, improving measurement efficiency and accuracy.
Implementation Method 1
X-ray scatterometry measurements Employing A Machine Learning Based Electromagnetic Response Model
Data Source
AI summary
Methods and systems for estimating values of parameters of interest from X-ray scatterometry measurements with reduced computational effort are described herein. Values of parameters of interest are estimated by regression using a trained, machine learning (ML) based electromagnetic (EM) response model. A training data set includes sets of Design Of Experiments (DOE) values of parameters of interest and corresponding DOE values of a plurality of electromagnetic response metrics. In some examples, values of parameters of interest are determined from measured images based on regression using a sequence of trained ML based electromagnetic response models. In some examples, input values employed to train the ML based EM response model are scaled based on model output variation.