Patterned Substrate 3D Metrology From a Single SEM Image
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing metrology methods for determining three-dimensional structure of functional elements in semiconductor manufacturing are time-consuming and inaccurate, negatively affecting the yield of the patterning process.
Innovation Solution
A system using a scanning electron microscope (SEM) with electron beam optics and a trained model to extract depth information from a single captured image of a patterned substrate, employing a convolutional neural network (CNN) for unsupervised training to infer depth data from disparity between images captured at different angles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional stereo image methods are used to determine three-dimensional structure, then measurement precision is improved, but productivity deteriorates due to time-consuming processing
Solution Approach 1:
The patent replaces traditional mechanical/stereo-image-based 3D measurement methods with a machine learning model that processes single images. The CNN model substitutes the complex mechanical process of capturing and processing multiple stereo images with a computational model that directly extracts 3D information from single images, thereby improving throughput while maintaining measurement precision.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between the single image capture and 3D structure determination. This intermediary model learns the complex mapping between 2D image features and 3D structural parameters, enabling accurate 3D measurement without requiring actual stereo image pairs or complex mechanical measurement systems.
2Measurement precision
If multiple images are captured at different angles for depth information, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent extracts the essential 3D measurement capability from the complex stereo imaging system and encapsulates it within a machine learning model. By taking out the core function of 3D measurement and implementing it through trained neural networks, the system achieves accurate depth information from simple single images without requiring complex multi-angle capture hardware or processing systems.
Solution Approach 2:
The patent creates a computational copy of the 3D measurement function through machine learning models trained on stereo image data. Instead of physically capturing multiple angles, the system uses a copied knowledge representation (the trained model) that can infer 3D structure from single images, thereby reducing device complexity while maintaining measurement precision.
3Manufacturing precision
If conventional metrology methods are used, then manufacturing precision is maintained, but loss of time increases
Solution Approach 1:
The patent performs preliminary action by pre-training machine learning models on extensive stereo image datasets before actual measurement. This preliminary training phase captures the essential patterns and relationships between 2D images and 3D structures, enabling the model to make rapid predictions during actual metrology measurements without requiring time-consuming real-time analysis, thus reducing measurement time while maintaining manufacturing precision.
Solution Approach 2:
The patent substitutes conventional time-consuming mechanical measurement and image processing systems with accelerated machine learning inference. The trained models perform 3D structure determination much faster than traditional methods by using computational patterns recognition rather than complex image processing algorithms, thereby reducing metrology measurement time while preserving the precision needed for quality control.
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 accurate and efficient determination of three-dimensional structure from a single image, increasing throughput and reducing computational resources compared to traditional stereo image methods.
Implementation Method 1
a scanning electron microscope (SEM) with electron beam optics configured to capture an image of a patterned substrate
Data Source
AI summary
A system, method, and apparatus for determining three-dimensional (3D) information of a structure of a patterned substrate. The 3D information can be determined using one or more models configured to generate 3D information (e.g., depth information) using only a single image of a patterned substrate. In a method, the model is trained by obtaining a pair of stereo images of a structure of a patterned substrate. The model generates, using a first image of the pair of stereo images as input, disparity data between the first image and a second image, the disparity data being indicative of depth information associated with the first image. The disparity data is combined with the second image to generate a reconstructed image corresponding to the first image. Further, one or more model parameters are adjusted based on the disparity data, the reconstructed image, and the first image.