Reciprocal Space Interface Detection for Automated Boundary Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for detecting material boundaries in semiconductor samples are often unsuitable for automation, prone to errors, and require significant trial-and-error to adapt to new materials or device types, especially when multiple interfaces need to be identified.
Innovation Solution
The use of reciprocal space images, such as diffraction patterns, analyzed by a neural network to classify materials and determine boundary locations with high accuracy, employing tools like convergent beam electron diffraction for precise mapping and binary search to locate transitions between materials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If manual inspection or conventional machine vision techniques are used for interface detection, then the technique can be simple to implement, but it is unsuitable for workflow automation and prone to errors
Solution Approach 1:
The patent replaces conventional machine vision techniques with reciprocal space imaging and neural network-based material classification. This substitution transforms the detection approach from direct spatial imaging to reciprocal space analysis, enabling automated workflow while improving reliability through physics-based material identification that is less prone to errors compared to heuristic machine vision methods
Solution Approach 2:
The patent changes the detection parameter from direct real-space imaging to reciprocal space representation. By transforming the imaging modality to reciprocal space and using neural networks for material classification, the system achieves both automation suitability and high detection reliability, as the neural network can consistently classify materials based on reciprocal space patterns without manual intervention
2Adaptability or versatility
If a one-size-fits-all imaging modality is sought for detecting multiple interfaces between different material pairs, then versatility is improved, but finding such a modality is challenging and requires recurring research efforts
Solution Approach 1:
The patent implements a universal reciprocal space imaging approach combined with neural network material classification that can identify interfaces between any material pairs. The neural network is trained to recognize material characteristics in reciprocal space, making the system adaptable to multiple material combinations without requiring separate imaging modalities or extensive reconfiguration for each material pair
Solution Approach 2:
The patent replaces the need for multiple specialized imaging modalities with a single reciprocal space imaging system coupled with neural network analysis. This substitution provides universal adaptability across different material pairs while reducing device complexity, as the neural network automatically learns to distinguish between different materials based on their reciprocal space signatures
3Measurement precision
If machine vision techniques are used to find edges, then edge detection can be effective, but identifying which edge is the desired edge becomes error-prone and requires significant heuristics configuration
Solution Approach 1:
The patent replaces machine vision edge detection with reciprocal space imaging and neural network material classification. This substitution eliminates the need for complex heuristics to identify desired edges, as the neural network directly identifies material interfaces based on reciprocal space patterns, providing both precision and simplicity
Solution Approach 2:
The patent introduces reciprocal space representation as an intermediary between imaging and material identification. This intermediary transformation allows the neural network to classify materials and identify interfaces without relying on direct spatial edge detection heuristics, simplifying the process while maintaining or improving measurement precision
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
Enables robust, automated, and versatile detection of material boundaries with an accuracy of 100 nm or better, facilitating workflow automation and precise analysis of semiconductor structures.
Implementation Method 1
A reciprocal space image can be obtained directly, e.g. by a diffractometer
Implementation Method 2
tools such as convergent beam electron diffraction (CBED)
Data Source
AI summary
Methods and apparatus determine a location of a boundary of a given material in a sample, based on classification of reciprocal space images at respective positions. A diffraction image at a given position is classified to identify a material at that position. Imaging and classification at multiple positions can quickly and reliably find the boundary with sub-micron accuracy. Binary search provides speed-up. Classification is performed by neural network. The technique is suitable for distinguishing monocrystalline, polycrystalline, and amorphous silicon, high-Z materials (tungsten), or low-Z materials (carbon), among others. The technique integrates into automated workflows, with precise positioning of further imaging, probing, or milling operations based on the located boundary. Examples and variations are disclosed.