SEM Depth Estimation Using Simulated Images and ANN Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing semiconductor structures using scanning electron microscopes (SEMs) provide only two-dimensional planar images, which are inadequate for understanding the structure, and existing systems for obtaining three-dimensional depth information are limited by the availability and cost of training data.
Innovation Solution
A method involving an artificial neural network (ANN) model trained using simulated scanning electron microscope (SEM) images and data augmentation techniques to generate accurate depth maps, supplemented by transmission electron microscope (TEM) and atomic force microscope (AFM) images, to estimate reliable depth information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulated SEM images and data augmentation are used for training, then training data availability is improved, but measurement precision may deteriorate
Solution Approach 1:
The patent creates synthetic training data by generating simulated SEM images from 3D structural models. These simulated images serve as copies that mimic real SEM images but can be produced in unlimited quantities through computational methods, thereby solving the data availability problem without requiring extensive physical sample preparation and imaging
Solution Approach 2:
The patent performs data augmentation and simulation operations in advance to create a comprehensive training dataset before actual depth estimation is needed. By pre-generating diverse training examples with known ground truth depth information, the system prepares high-quality training data that improves model precision while avoiding the need for extensive real-time data collection
2Loss of information
If traditional stereo image methods are used, then depth information can be obtained, but device complexity and cost increase
Solution Approach 1:
The patent extracts the depth estimation function from complex multi-camera stereo systems and implements it within a single SEM apparatus. By using a single imaging device combined with simulated image processing, the system eliminates the need for additional cameras, synchronization hardware, and complex calibration systems while still achieving depth information extraction
Solution Approach 2:
The patent introduces simulated SEM images as an intermediary between the actual SEM image and the depth information extraction process. These simulated images serve as a computational bridge that enables depth estimation without requiring physical stereo imaging hardware, thereby simplifying the overall system architecture
3Measurement precision
If more training data is collected from real samples, then model accuracy improves, but time and resource consumption increase
Solution Approach 1:
The patent generates synthetic training data by computationally simulating SEM images from 3D models, creating unlimited training examples without physical sample preparation. This copying approach provides diverse training data with known ground truth depth information, improving model accuracy while avoiding the time-consuming process of collecting, preparing, and imaging real samples
Solution Approach 2:
The patent varies parameters in the simulated images (such as imaging conditions, sample structures, and depth values) to generate diverse training examples. By systematically changing simulation parameters, the system creates a comprehensive training dataset that improves model generalization and accuracy without requiring extensive physical data collection
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of estimating depth information includes generating a first simulated image using a simulator provided with a first depth map, training an artificial neural network model based on the first depth map and the first simulated image, generating a second depth map by inputting an actual image into the trained artificial neural network model, and generating a second simulated image using the simulator provided with the second depth map.