Optical Metrology for Unknown Structure Feature Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional optical metrology techniques struggle to accurately measure structures with unknown or non-periodic sections due to the inability to rigorously model such features, leading to inaccurate data fitting and limited applicability of existing modeling methods.
Innovation Solution
Employ a trained neural network, such as a Domain Adversarial Neural Network (DANN) or multi-target Deep Neural Network (DNN), to determine parameter values for both known and unknown structures within a sample by analyzing combined optical metrology data, enabling separation of contributions from the structure-of-interest (SOI) and unknown sections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional optical metrology modeling techniques are used to measure structures with unknown or non-periodic sections, then the measurement process can be performed with standard equipment, but the measurement accuracy deteriorates due to inability to rigorously model such features
Solution Approach 1:
The patent segments the optical metrology problem by separating the structure into known periodic portions and unknown non-periodic portions. The neural network is trained to specifically identify and characterize the unknown portions while the periodic portions are handled by conventional modeling techniques, thus resolving the contradiction between measurement accuracy and adaptability to unknown structures
Solution Approach 2:
The patent introduces a neural network as an intermediary component between the optical measurement system and the analysis process. This neural network acts as a mediator that can handle both known and unknown structure types, enabling the system to achieve high measurement accuracy across diverse structure types including those with unknown or non-periodic sections
2Measurement precision
If rigorous modeling is attempted for structures with unknown sections, then measurement accuracy may improve, but the device complexity and computational requirements increase significantly
Solution Approach 1:
The patent employs a neural network trained on synthetic data that can rapidly characterize unknown structures without requiring complex rigorous modeling for each measurement. This approach replaces expensive, computationally intensive rigorous modeling with a more efficient neural network-based method that achieves comparable or superior accuracy while reducing computational complexity
Solution Approach 2:
The patent performs preliminary training of the neural network using synthetic data generated from rigorous models. This preliminary action creates a pre-trained system that can then quickly and accurately characterize unknown structures during actual measurements without requiring complex real-time modeling, thus reducing device complexity while maintaining measurement precision
3Ease of manufacture
If conventional modeling methods are used for periodic structures, then the analysis process is straightforward, but the method becomes inapplicable when non-periodic or unknown sections are present
Solution Approach 1:
The patent creates a universal measurement system by training the neural network to handle multiple structure types including periodic, non-periodic, known, and unknown structures. This multi-functional neural network maintains the ease of conventional analysis for periodic structures while extending applicability to non-periodic and unknown sections, thus resolving the contradiction between ease of analysis and adaptability
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 precise measurement of both known and unknown structural parameters with high orthogonality, improving measurement accuracy and expanding the applicability of optical metrology to non-periodic and nuisance features.
Implementation Method 1
a sample under test is illuminated with light, e.g., at a single wavelength or multiple wavelengths. After interacting with the sample, the resulting light is detected and analyzed
Data Source
AI summary
A sample that includes a target structure with a structure-of-interest (SOI) having a set of known parameters optically coupled to an unknown structure, e.g., having unknown parameters, is optically measured using light that is incident on the target structure. Light detected from the target structure in response to the incident illumination light incident is used to obtain metrology data, which is a combination of a from the SOI and a contribution from the unknown structure. A set of parameter values characterizing the SOI and a set of parameter values characterizing the unknown structure are determined from a trained neural network based on the metrology data.