CDSEM Deep Structure Signal Separation for Blind Self-Supervised Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting deep structure signals in semiconductor layers face inefficiencies and damage risks due to high integration times and noise, particularly in high aspect ratio structures, leading to low throughput and inaccurate measurements.
Innovation Solution
The process involves dividing an image into separate upper and lower regions using self-tunable masking, applying spatial domain filtering and stretching for the upper region, and employing blind denoising by self-supervision (BDSS) for the lower region, followed by local enhancement and combination of the processed images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If penetration energy is increased to detect deep structure signals, then measurement precision is improved, but device complexity and capital spending increase
Solution Approach 1:
The patent segments the image into multiple depth layers (surface layer, intermediate layer, deep layer) and processes each layer separately with appropriate denoising algorithms. This allows detection of deep structure signals without requiring increased penetration energy, thereby avoiding capital spending on new high-energy tools while maintaining measurement precision.
Solution Approach 2:
The patent applies preliminary denoising processing to separate layers before combining them. By pre-processing each layer with appropriate algorithms (e.g., non-local means for surface, bilateral filtering for deep layers), the system achieves improved signal detection without needing higher energy penetration, thus resolving the contradiction between measurement precision and device complexity.
2Measurement precision
If integration time is increased to detect deep structure signals, then measurement precision is improved, but productivity decreases
Solution Approach 1:
The patent divides the image collection process into parallel processing of multiple depth layers with different integration times. Surface layers use short integration times for high throughput, while deep layers use longer integration times only where needed. This segmentation allows the system to maintain high overall productivity while achieving precise deep structure signal detection in specific regions.
Solution Approach 2:
The patent applies different integration times and denoising algorithms to different spatial regions and depth layers based on their specific requirements. Surface areas with strong signals use rapid processing, while deep contact regions use enhanced integration and denoising. This local optimization maintains high throughput overall while achieving the necessary measurement precision for deep structures.
3Measurement precision
If integration time is increased to reduce noise, then measurement precision is improved, but object damage increases
Solution Approach 1:
The patent segments the measurement process into multiple passes with different integration times for different depth layers. Instead of using one long integration time for the entire image, the system uses short integration times for surface layers and only applies longer integration times to specific deep regions where signals are weak. This reduces cumulative energy exposure and minimizes damage to contacts and trenches while maintaining measurement precision.
Solution Approach 2:
The patent applies the principle of partial action by using extended integration times only partially - specifically for deep layer detection where signals are weakest - rather than applying it uniformly across the entire image. This selective application reduces overall energy exposure and minimizes damage to the semiconductor structure while achieving sufficient signal detection accuracy for deep contacts.
4Device complexity
If deep structures and surrounding areas are treated as one instance, then device complexity is reduced, but measurement precision decreases
Solution Approach 1:
The patent segments the image into multiple depth layers (surface, intermediate, deep) and applies different denoising algorithms and processing parameters to each layer. Surface layers use non-local means filtering, while deep layers use bilateral filtering and histogram equalization. This segmentation improves measurement precision for deep structures without significantly increasing device complexity, as the processing is automated through layer identification algorithms.
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
This approach enhances the detection of deep structure signals accurately without causing damage, improving throughput and reducing noise-related inaccuracies.
Implementation Method 1
obtaining an image of an area of a semiconductor structure having at least two layers using a charged particle imager
Data Source
AI summary
A method of deep structure detection of deep structure signals for a critical dimension scanning electron microscope (CDSEM) image by dividing the image into separate upper and lower regions using self-tunable masking and then denoising and enhancing the upper and lower regions separately, wherein the lower deep structure region is denoised by applying blind denoising by self-supervision (BDSS) thereto.