TEM Image Segmentation for Faster Semiconductor Core Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional transmission electron microscope (TEM) image processing for semiconductor devices relies heavily on manual measurements, which are time-consuming, inaccurate, and prone to deviations, especially due to unclear boundaries between materials in TEM images.
Innovation Solution
A TEM image processing apparatus and method that uses weak labeling and artificial intelligence to automatically generate ground truth and perform image segmentation, reducing manual input and improving accuracy by iteratively refining the segmentation model with user feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual measurement methods are used for TEM image analysis, then measurement accuracy can be maintained through expert judgment, but the process is time-consuming and labor-intensive
Solution Approach 1:
The system enables self-service automation where the AI model independently performs image segmentation and measurement tasks that were previously requiring manual expert intervention. The model processes TEM images automatically, generating measurements without continuous human oversight, thus resolving the contradiction between maintaining accuracy and reducing time consumption.
Solution Approach 2:
The patent replaces the mechanical manual measurement process with an automated AI-based image processing system. The segmentation model and measurement algorithms substitute human experts' manual operations, achieving both high accuracy through sophisticated image analysis and significant time reduction through automation.
2Measurement precision
If strong supervision with manual ground truth generation is used for training AI models, then segmentation accuracy can be improved, but the burden of manual work increases significantly
Solution Approach 1:
The system applies partial supervision by using only a small subset of manually labeled data for training the segmentation model. Instead of requiring comprehensive manual ground truth for all training samples, the model is trained on limited annotated data and then applied to segment numerous unlabeled TEM images, significantly reducing manual work burden while maintaining segmentation accuracy.
Solution Approach 2:
The system creates synthetic training data by copying and transforming limited ground truth examples. The segmentation model learns from a small set of manually labeled images and generalizes this knowledge to accurately segment similar structures in new TEM images without requiring manual annotation of each individual image.
3Reliability
If complete manual labeling of TEM images is performed, then ground truth quality can be ensured, but the process becomes excessively time-consuming and complex
Solution Approach 1:
The system performs preliminary action by pre-processing TEM images through automatic preprocessing steps before segmentation. The preprocessing pipeline prepares images in advance with standard operations, reducing the need for complex manual intervention during the segmentation and measurement phases, thus simplifying the overall process while ensuring reliable results.
Data Source
AI summary
A method of operating a transmission electron microscope (TEM) image processing apparatus, includes acquiring a TEM image from a TEM facility, performing weak labeling of the TEM image, generating ground truth for a partially labeled TEM image using a guide model, performing segmentation using training data consisting of a pair of the TEM image and the ground truth; measuring a device core structure according to a result of the segmentation, and visualizing a measurement result according to the device core structure, and storing the same in a database.