Semiconductor Inspection Image Alignment via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods face challenges in accurately aligning optical images with CAD images for semiconductor devices that have undergone various process rules, leading to difficulties in inspection accuracy.
Innovation Solution
A semiconductor inspection device and method utilizing a photodetector, optical system, image generation unit, reception unit, image conversion unit via machine learning, and alignment unit to convert and align optical and CAD images, ensuring accurate alignment by making the CAD image patterns resemble the optical images or vice versa.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional alignment methods are used for semiconductor devices with various process rules, then the inspection process remains simple, but the alignment accuracy between optical images and CAD images deteriorates
Solution Approach 1:
The patent applies preliminary action by performing image conversion on the CAD image before the alignment process. The image conversion unit converts the CAD image into a format that resembles the optical image using machine learning models trained on process rule-specific data. This preliminary conversion prepares the images for accurate alignment by accounting for process rule variations in advance, thereby improving alignment accuracy without requiring complex real-time adjustments during the alignment process itself.
Solution Approach 2:
The patent utilizes parameter changes by employing different machine learning models trained on specific process rules to convert CAD images. Each process rule has its own conversion parameters and characteristics, and the system selects and applies the appropriate parameters based on the semiconductor device being inspected. This allows the alignment system to adapt to various process rules while maintaining a relatively simple overall process structure.
2Measurement precision
If machine learning-based image conversion is applied to improve alignment accuracy, then the alignment precision improves, but the processing time and computational complexity increase
Solution Approach 1:
The machine learning models for image conversion are trained in advance on datasets containing optical images and corresponding CAD images for various process rules. This preliminary training phase allows the system to learn the complex relationships between CAD representations and actual optical appearances. During actual inspection, the pre-trained models can quickly convert new CAD images without requiring time-consuming real-time training, thus improving alignment precision while minimizing additional processing time.
Solution Approach 2:
The patent creates converted CAD images that copy the visual characteristics of optical images through machine learning. Instead of directly processing and comparing raw optical images which may contain noise and variations, the system generates synthetic versions of optical images from CAD data using learned patterns. This copying approach preserves the essential alignment information while reducing the impact of optical variations, thereby improving precision without proportionally increasing processing time.
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
The solution significantly improves the accuracy of image alignment for semiconductor devices with various process rules by converting image patterns to match each other, enhancing inspection precision.
Implementation Method 1
a photodetector configured to detect light from a semiconductor device and output a detection signal
Data Source
AI summary
An observation system includes a detector that detects light from a semiconductor device and outputs a detection signal, a 2D camera, an optical device that guides light to the detector and the 2D camera, an image processing unit that generates a first optical image of the semiconductor device based on the detection signal and receives an input of a first CAD image, an image analysis unit that learns a conversion process of the first CAD image by machine learning using the first optical image as training data, and converts the first CAD image into a second CAD image resembling the first optical image by the conversion process based on a result of the learning, and an alignment unit that performs alignment based on a second optical image and the second CAD image.