Image-Based Overlay Metrology Model for Semiconductor Wafer Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image-based overlay metrology algorithms are limited in their ability to perform reliably with arbitrary overlay targets or device structures, often ignoring contributions from non-specific pixels and being sensitive to process variations and optical system errors, which complicates accurate measurement of overlay errors in semiconductor manufacturing.
Innovation Solution
The development of an image-based measurement model trained on raw image data from a Design of Experiments (DOE) wafer, allowing for the direct calculation of overlay errors and other parameters from measured image data, reducing errors and sensitivity to systematic errors, and enabling measurements on both same and different layers of a substrate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional image-based overlay metrology algorithms are used, then measurement process is simplified, but measurement precision deteriorates due to sensitivity to process variations and optical system errors
Solution Approach 1:
The patent introduces a trained measurement model as an intermediary between the image data and overlay error calculation. This model, trained on DOE wafer data, acts as a mediator that processes raw image data and outputs accurate overlay measurements while compensating for process variations and optical errors, thus maintaining simplicity while improving precision
Solution Approach 2:
The measurement model is trained in advance using Design of Experiments (DOE) wafer data that captures various process conditions and optical variations. This preliminary training enables the model to automatically compensate for these variations during actual measurements, improving precision without complicating the measurement process
2Device complexity
If traditional image-based algorithms focus on specific target features, then algorithm complexity is reduced, but measurement precision deteriorates due to loss of information from other image areas
Solution Approach 1:
The patent merges information from the entire image by using all pixels as signals in the measurement model, rather than focusing only on specific target features. The trained model integrates contributions from all image areas, preserving information that would otherwise be lost, thereby improving measurement precision while maintaining manageable algorithm complexity
Solution Approach 2:
The measurement model is designed to process arbitrary overlay targets and device structures universally, not just specific target types. This universal approach allows the model to utilize information from any image area regardless of the specific target structure, improving precision without requiring complex target-specific algorithms
3Ease of operation
If specialized target structures are used for traditional image-based overlay measurement, then measurement process is simplified, but adaptability deteriorates because algorithms cannot perform reliably with arbitrary overlay targets
Solution Approach 1:
The trained measurement model achieves universality by being able to process any overlay target structure and device pattern. The model was trained on diverse DOE wafer data that includes various target types and process conditions, enabling it to adapt to arbitrary targets while maintaining a simple, unified measurement process
Solution Approach 2:
The measurement model adapts to different target structures by utilizing the parameter variations captured during training on DOE wafers. The model learns to extract overlay information from diverse structural parameters and target configurations, achieving high adaptability while keeping the measurement process simple through automated model-based analysis
Data Source
AI summary
Methods and systems for creating an image-based measurement model based only on measured, image-based training data are presented. The trained, image-based measurement model is then used to calculate values of one or more parameters of interest directly from measured image data collected from other wafers. The image-based measurement models receive image data directly as input and provide values of parameters of interest as output. In some embodiments, the image-based measurement model enables the direct measurement of overlay error. In some embodiments, overlay error is determined from images of on-device structures. In some other embodiments, overlay error is determined from images of specialized target structures. In some embodiments, image data from multiple targets, image data collected by multiple metrologies, or both, is used for model building, training, and measurement. In some embodiments, an optimization algorithm automates the image-based measurement model building and training process.