Deep Learning Semiconductor Image Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image measurement systems fail to accurately measure attributes of images with variations due to changes in manufacturing processes and imaging conditions, leading to erroneous measurements or manual intervention, which is time-consuming and prone to errors.
Innovation Solution
A machine learning model is trained to perform image segmentation and measurement on images with variations, using synthetic images generated by deforming original images based on expected changes in manufacturing parameters, allowing for automated measurement without user setup.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a setup is created based on a first image to measure attributes, then the setup can be run on remaining images, but if the remaining images have variations, the setup fails to measure attributes accurately
Solution Approach 1:
The system performs preliminary actions by training the machine learning model on a diverse dataset of images with various variations before actual measurement. This pre-training enables the model to adapt to different image variations without requiring manual setup adjustments for each new image type.
Solution Approach 2:
The system changes parameters by using a machine learning model that can dynamically adjust its measurement parameters based on the input image characteristics. Instead of using fixed parameters from a single setup, the model learns optimal parameters for different image variations during training and applies them automatically during inference.
2Measurement precision
If manual measurement is performed by process engineers, then measurements can be obtained, but errors and variations between engineers occur and significant time is consumed
Solution Approach 1:
The system implements self-service by enabling automated image measurement without requiring manual intervention from process engineers. The machine learning model independently performs measurement tasks that previously required human engineers, eliminating inter-engineer variations and significantly reducing time consumption.
Solution Approach 2:
The system replaces the mechanical system of manual measurement with an automated machine learning-based measurement system. This substitution eliminates human factors such as errors and variations between engineers while dramatically reducing the time required for measurements.
3Reliability
If conventional image measurement systems are used, then simple images can be measured, but images with variations due to manufacturing changes cannot be measured accurately
Solution Approach 1:
The system performs preliminary training on a comprehensive dataset that includes images with various manufacturing variations. This pre-training ensures the model is prepared to handle diverse image structures reliably without requiring reconfiguration for each new manufacturing condition.
Solution Approach 2:
The machine learning model achieves universality by being trained to handle multiple types of images with different variations. A single model can process various image structures and manufacturing conditions that would require multiple specialized conventional measurement systems.
Data Source
AI summary
Methods, systems, and non-transitory computer readable medium are described for automated image measurement for process development and optimization. An example method may include receiving an image of a product associated with a manufacturing process, wherein the product comprises a plurality of structures; identifying, using a trained machine learning model, a segment of the image that comprises a structure of the plurality of structures; determining a plurality of image measurements of the segment that comprises the structure; and storing the plurality of image measurements.