SEM Depth Resolution via Blind Source Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current SEM techniques struggle to generate depth-resolved imagery for samples with unknown composition and geometry, particularly in biological applications, due to the lack of a priori knowledge about the sample's characteristics, limiting their application in life sciences.
Innovation Solution
Employing a statistical Blind Source Separation technique, specifically Principal Component Analysis (PCA), to automatically process SEM data and spatially resolve it into depth-resolved imaging pairs, leveraging the linearity of the Point Spread Function of backscattered electrons to disentangle signals from different depth layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If inter-image contrast adjustment is performed using prior knowledge of sample composition and geometry, then depth-resolved imagery can be generated, but the method cannot be applied to samples with unknown composition such as biological samples
Solution Approach 1:
The system performs self-adjustment by automatically determining scaling factors through statistical analysis of the measured data itself, without requiring external knowledge about sample composition or geometry. The Blind Source Separation technique enables the system to extract depth information purely from the multi-energy SEM data, making the method self-sufficient and applicable to unknown samples.
Solution Approach 2:
The method varies the primary beam energy parameter across multiple measurement sessions to acquire SEM images at different penetration depths. By collecting data at multiple energy levels and applying statistical processing, the system transforms the dependency on prior knowledge into an advantage, using the energy-dependent signal variations to automatically resolve depth information through Blind Source Separation.
2Loss of information
If manual contrast adjustment is performed based on user knowledge, then some depth information can be revealed, but the process requires extensive user input and cannot handle complex samples with unknown structures
Solution Approach 1:
The manual mechanical process of user-based contrast adjustment is replaced with an automated computational system using Blind Source Separation and statistical analysis. The system automatically processes the multi-energy data set to extract depth-resolved information, eliminating the need for manual user input while recovering complete depth information that would otherwise be lost.
3Length of stationary object
If multiple SEM images are acquired with increasing primary beam energy, then deeper layers can be probed, but automatic de-convolution is not possible without prior compositional knowledge
Solution Approach 1:
The method segments the complex task of depth resolution into multiple measurement sessions at different beam energies, then further segments the data processing into statistical analysis steps using Blind Source Separation. This segmentation transforms the intractable problem of automatic de-convolution into a series of manageable computational steps that can be performed without prior sample knowledge.
Solution Approach 2:
Statistical analysis and Blind Source Separation techniques serve as intermediaries between the raw multi-energy SEM data and the final depth-resolved imagery. These intermediary processing steps automatically extract the depth information embedded in the energy-dependent signal variations, bridging the gap between measurement and interpretation without requiring user knowledge.
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 automatic de-convolution of SEM data and generation of depth-resolved imagery for complex samples, providing high depth resolution and overcoming the limitations of prior techniques by not requiring user-input signal adjustments, thus expanding SEM applications to biological samples.
Implementation Method 1
Detecting stimulated radiation emitted by the sample during each measurement session... Signals associated with backscattered (BS) electrons generally yield sufficient information from all generation depths within their detectable range
Implementation Method 2
The mathematical Point Spread Function (PSF) of BS electrons in several types of samples, including stained bio-samples and polymers, is generally (highly) linear
Data Source
Figure 1
Figure 2A~2B
Figure 3(A)~4
AI summary
A method of investigating a sample using Scanning Electron Microscopy (SEM), comprising the following steps: - Irradiating a surface (S) of the sample using a probing electron beam in a plurality (N) of measurement sessions, each measurement session having an associated beam parameter (P) value that is chosen from a range of such values and that differs between measurement sessions; - Detecting stimulated radiation emitted by the sample during each measurement session, associating a measurand (M) therewith and noting the value of this measurand for each measurement session, thus allowing compilation of a data set (D) of data pairs (Pi, Mi), where 1 ≤ i ≤ N, wherein: - A statistical Blind Source Separation (BSS) technique is employed to automatically process the data set (D) and spatially resolve it into a result set (R) of imaging pairs (Qk, Lk), in which an imaging quantity (Q) having value Qk is associated with a discrete depth level Lk referenced to the surface S. A suitable example of such a BSS technique is Principal Component Analysis (PCA), e.g. employing a Karhunen-Loeve transform operation. This technique allows high-resolution 3D volume reconstruction from a sequence of backscattered images acquired by a SEM. The method differs from known techniques in that it can be used on complex samples with unknown structure. With this method, one can compute compensation factors between high- and low-energy images using second-order (or higher-order) multivariate statistics, which allows for the effective separation of different depth layers in a sample without using a priori knowledge of sample structure. The method has a wide range of applications in life-science and material science imaging.