Defect detection in patterned wafers using multichannel Scanning Electron Microscope

In this paper, we propose an accurate automatic defect detection method using the quadtree decomposition on scanning electron microscope (SEM) images. The proposed method consists of five steps including alignment between the inspection and reference images, edge relaxation, defect candidate detection, multichannel integration, and defect scoring. Since the periodicity of semiconductor SEM images is usually high, alignment can be failed due to a lot of local minima. To resolve this problem, we propose the enhancement method of non-periodic areas in SEM images. Then, we propose an edge relaxation technique to minimize the falsely remained edges in the aligned and subtracted image due to different focusing and exposure between two images. Then, the defect candidates are detected by analyzing the inhomogeneity based on quadtree decomposition, which is enhanced by our outlier-robust decomposition rule and adaptive decomposition thresholding to prevent over/under decomposition. Whereas the conventional defect detection system focuses only on the intensity of defects, our method considers both the inhomogeneity and intensity of defects. Field test results demonstrated that the proposed method provided more accurate defect detection with faster speed as compared to the conventional system. The proposed method can much reduce the automatic defect review rework compared with the conventional system.

With the progression of deep learning algorithms in computer vision, a lot of research is taking place in the semiconductor industry towards improving real-time defect detection and classification analysis. An Automated Defect Classification and Detection (ADCD) framework not only enables rapid measurement of dimensions and classification of defects, but also helps minimize production costs, engineering time as well as tool cycle time associated with the defect inspection process. As we continue to shrink the pitch (below 36nm), defect characterization at wafer scale becomes a key issue as it demands rapid measurement but without losing accuracy and repeatability. Also, in the context of high NA lithography (thin resist), accurate metrology becomes difficult with very noisy as well as low contrast images (No BKM exists till now). Human eyes generally demonstrate close to the Bayesian Error limit in detecting smaller objects (for example, extracting contextual information instantaneously from nanoscale defects in SEM images). However, for most One-stage and Two-stage object detectors, this is still a very challenging task due to variable image resolution and SEM (scanning electron microscope) image quality (low SNR). In this research work, we have experimented with different modified YOLOv5 object detectors to improve challenging stochastic defect detection precision. In this work, we have proposed an ensemble strategy by empirically combining multiple custom-trained models (YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x) together at the test and inference time. We have noticed four YOLOv5 architecture variants are outperforming against our previous Ensemble ResNets model with improvements of the average precision metric (AP) of the most difficult defect classes as p gap and microbridges as well as overall mAP accuracy. With Ensemble YOLOv5, the p gap AP and microbridge AP metrices have been improved by 35% and 25.33%, respectively, whereas the overall mAP has been improved by 6.25%. The proposed Automated Defect Classification and Detection (ADCD) framework can also be used for high resolution and high-speed metrology, providing rapid identification of defects with improved certainty and further root cause investigation.

Recent computational methods of wafer defect detection often rely on the difference image between an inspected image and its reference image, and highly depend on registration accuracy. In this paper, we present a novel method for defect detection in patterned wafers, based on reconstruction of the inspected image from the reference image using anisotropic kernels. This method avoids registration between the inspected and reference image and compensates for pattern variations, thus reducing the false detection rate. Experimental results demonstrate the advantages and robustness of the proposed method. Efficient implementation of the algorithm makes it be suitable for industrial use. We also demonstrate extension of the kernel-based similarity concept to the multichannel Scanning Electron Microscope (SEM) images.

Scanning Electron Microscope (SEM) image analysis plays a vital role in semiconductor thin-film characterisation. In particular, defect detection and classification are performed using SEM images. However, conventional methods rely on labelled datasets or handcrafted features for classification, which limits their generalisation in real-world industrial inspection settings. The present work proposes a self-supervised multi-resolution learning framework for label-agnostic morphology representation learning and clustering of semiconductor thin-film defects. It uses a SEM image Dataset (4591 images) obtained from industrial wafer inspection. The framework starts with image pre-processing to remove acquisition artifacts. It employs a multi-resolution image pyramid for capturing surface morphologies at fine, intermediate, and coarse spatial scales. A shared-weight convolutional encoder that ensures alignment between embeddings across the three resolutions is trained (on an unlabelled dataset) and utilised for unsupervised defect morphology grouping. The framework learns morphological representations without defect labels. However, defect labels are used only for post-hoc evaluation by normalized mutual information (NMI) and visualization. Intrinsic clustering metrics and low-dimensional visualization are used to assess the algorithm's efficacy. Experimental results reveal that the proposed method, gray level co-occurrence matrix (GLCM), local binary patterns (LBP), wavelet-based features, and principal component analysis (PCA) on raw pixels obtained a silhouette score of 0.50, 0.43, 0.31, 0.45, and 0.22, respectively. While normalized mutual information (NMI) values remained moderate across the models. These results reflect the label-agnostic nature of the proposed SSL framework. Further, UMAP and t-SNE visualizations confirm the coherent manifold structure and the effectiveness of morphology-driven grouping. These results demonstrate the robust, scale-invariant quality of the proposed self-supervised multi-resolution learning framework for defect clustering.

Accurate identification and classification of atmospheric particulates can provide the basis for their source apportionment. Most current research studies mainly focus on the classification of atmospheric particles based on the energy spectrum of particles, which has the problems of low accuracy and being time-consuming. It is necessary to study the classification method of atmospheric particles with higher accuracy. In this paper, a convolutional neural network (CNN) model with attention mechanism is proposed to identify and classify the scanning electron microscopy (SEM) images of atmospheric particles. First, this work established a database, Qingdao 2016–2018, for atmospheric particles classification research. This database consists of 3469 SEM images of single particulates. Secondly, by analyzing the morphological characteristics of single particle SEM images, it can be divided into four categories: fibrous particles, flocculent particles, spherical particles, and mineral particles. Thirdly, by introducing attention mechanism into convolutional neural network, an Attention-CNN model for the identification and classification of the four types of atmospheric particles based on the SEM images is established. Finally, the Attention-CNN model is trained and tested based on the SEM images database, and the results of identification and classification for four types of particles are obtained. Under the same SEM images database, the classification results from Attention-CNN are compared with those of CNN and SVM. It is found that Attention-CNN has higher classification accuracy and reduces significantly the misclassification number of particles, which shows the focusing effect of attention mechanism.

In semiconductor device manufacturing, critical dimension (CD) metrology provides a measurement for precise line-width control during the lithographic process. Currently scanning electron microscope (SEM) tools are typically used for this measurement, because the resolution requirements for the CD measurements are outside the range of optical microscopes. While CD has been a good feedback control for the lithographic process, line-widths continue to shrink and a more precise measurement of the printed lines is needed. With decreasing line widths, the entire sidewall structure must be monitored for precise process control. Sidewall structure is typically acquired by performing a destructive cross sectioning of the device, which is then imaged with a SEM tool. Since cross sectioning is destructive and slow, this is an undesirable method for testing product wafers and only a small sampling of the wafers can be tested. We have developed a technique in which historical cross section/top down image pairs are used to predict sidewall shape from top down SEM images. Features extracted from a new top down SEM image are used to locate similar top downs within the historical database and the corresponding cross sections in the database are combined to create a sidewall estimate for the new top down. Testing with field test data has shown the feasibility of this approach and that it will allow CD SEM tools to provide cross section estimates with no change in hardware or complex modeling.

In order to perform nano-manipulation more accurately, it is necessary to study the relationship between the scanning electron microscope (SEM) image and the actual length. In this paper, based on the secondary electron image, the displacement of 1.81nm in SEM was measured the actual displacement was obtained according to the change of the pixels, so as to realize the nanometer displacement calibration and real-time measurement of the nano robot manipulator in SEM. According to the magnification, resolution and scale bar of SEM image, the mapping relationship between pixels and actual length was established. The obvious feature point and area in the SEM image were selected to be marked. The driving signal was applied to the manipulator of the nano robot, and the template matching algorithm was used to track the feature point and area in real time. The change of the feature point in the SEM image was obtained, and the actual displacement of the manipulator was calculated according to the mapping relationship between the pixel points and the actual displacement. We calibrated the displacement of the nanorobots and measured the actual displacement in real time, which laid a foundation for the precise control of the nanorobots in the future.

Image noise is an important factor that influences the accuracy of strain field measurements by means of digital image correlation and scanning electron microscope (SEM) imaging. We propose a new model to quantify the SEM image noise, which extends the classical photon noise model by taking into account the brightness setup in SEM imaging. Furthermore, we apply this model to investigate the impact of different SEM setting parameters on image noise, such as detector, dwell time, spot size, and pressure in the SEM chamber in the context of low vacuum imaging.

As the technology node shrinks, the systematic defects such as the missing and extra pattern are generated in the process of cell patterning. The defect such as the extra and missing pattern can lead to a critical failure on device. Subsequently, the systematic defect in the UBE could be an etch-gas path on proceeding the etch process and cause the quality degradation of device. In this paper, we propose an automated system to inspect the systematic defect in the cell array with the scanning electron microscope (SEM) image and the physical design. The method consists of conversion, matching, and detection. In the conversion method, the SEM image is converted into the layout using image processing such as noise reduction, segmentation and contour tracing. In the following matching method, gradient descent optimization is used to match the coordinate of layout converted from the SEM image with the coordinate of physical design. In the final detection method, the defect is detected by inspecting the patterns of two layouts. On the test of 2,500 data, we confirmed that the accuracy improved from 97% to 100% as a result of comparing the engineer's visual inspection method with the proposed method. In addition, the turnaround time (TAT) is improved by approximately 40 times. The proposed system is currently applied to DRAM products and used for the field of design for manufacturing (DFM) and manufacturing process.

The MeX™ software is a useful tool for tridimensional data collection for surface evaluation and could be relevant to evaluate the same specimen in different phases of the study, assuming repeated measures of dental enamel roughness. The aim of this study was to evaluate the influence of sample metallization for dental enamel roughness analysis with 3D images reconstructed using MeX™ software from Scanning Electron Microscopy (SEM) images. The influence of 74.98% (%mol/mol) argon-oxygen plasma for carbon layer removal on surface roughness of the metallized specimen was also evaluated. Dental enamel specimens were prepared for SEM analysis with and without carbon metallization using conventional or environmental modes. Argon-oxygen plasma for carbon layer removal was used and surface roughness was re-evaluated. Roughness obtained by SEM and MeX™ reconstructed images, with or without metallization, did not differ. No significant alteration on surface roughness after carbon layer removal using plasma was found. SEM baseline evaluation using conventional mode without sample preparation and in environmental mode were not comparable. Roughness of enamel 3D images reconstructed with MeX™ software from SEM images, with or without metallization was similar. The 74.98% (%mol/mol) argon-oxygen plasma removed the carbon layer with no effect on enamel roughness.

Evaluating the ability of energy dispersive X-ray analysis to monitor binding oil content of carbon paste electrodes exposed to biofouling agents

Nanomaterials, owing to their distinctive features, are crucial across numerous scientific domains, especially in materials science and nanotechnology. Precise segmentation of Scanning Electron Microscope (SEM) images is essential for evaluating attributes such as nanoparticle dimensions, morphology, and distribution. Conventional image segmentation techniques frequently prove insufficient for managing the intricate textures of SEM images, resulting in a laborious and imprecise process. In this research, a modified U-Net architecture is presented to tackle this challenge, utilizing a ResNet50 backbone pre-trained on ImageNet. This model utilizes the robust feature extraction abilities of ResNet50 alongside the effective segmentation performance of U-Net, hence improving both accuracy and computational efficiency in TiO2 nanoparticle segmentation. The suggested model was assessed using performance metrics including accuracy, precision, recall, IoU, and Dice Coefficient. The results indicated a high segmentation accuracy, demonstrated by a Dice score of 0.946 and an IoU of 0.897, with little variability reflected in standard deviations of 0.002071 and 0.003696, respectively, over 200 epochs. The comparison with existing methods demonstrates that the proposed model surpasses previous approaches by attaining enhanced segmentation accuracy. The modified U-Net design serves as an excellent technique for accurate nanoparticle segmentation in SEM images, providing substantial enhancements compared to traditional approaches. This progress indicates the model's potential for wider applications in nanomaterial research and characterization, where precise and efficient segmentation is essential for analysis.

This paper presents a study of micro-scale deformation of materials utilizing scanning electron microscopy (SEM) images and the digital image correlation (DIC) technique. A loading stage was integrated into the SEM imaging system. During the experiment, a series of SEM images of the specimen were acquired in situ. The DIC technique was then applied to these SEM images to calculate the displacement and strain field at the area of interest. Additional surface preparation may be needed in order to have an effective pattern for DIC analysis. Two applications are presented in the paper. Using small tensile specimens, the mechanical properties of electrodeposited nickel-based LIGA (an acronym from German words for lithography, electroplating, and moulding) specimens were characterized. In this case, the natural microstructural feature of the specimen surface was used directly as the pattern for DIC analysis. This method was also applied to study the strain concentration around the crack tip during the ductile fracture test of Al 6061-T651. In contrast to the previous case, the DIC patterns were generated by sputtering a thin layer of gold film on to the specimen surface through the copper mesh grid.

A methodology for the automated analysis of metal powder scanning electron microscope (SEM) images towards material characterization is developed and presented. This software-based tool takes advantage of a combination of recent artificial intelligence advances (mask R-CNN), conventional image processing techniques, and SEM characterization domain knowledge to assess metal powder quality for additive manufacturing applications. SEM is being used for characterizing metal powder alloys, specifically by quantifying the diameter and number of spherical particles, which are key characteristics for assessing the quality of the analyzed powder. Usually, SEM images are manually analyzed using third-party analysis software, which can be time-consuming and often introduces user bias into the measurements. In addition, only a few non-statistically significant samples are taken into consideration for the material characterization. Thus, a method that can overcome the above challenges utilizing state-of-the-art instance segmentation models is introduced. The final proposed model achieved a total mask average precision (mAP50) 67.2 at an intersection over union of 0.5 and with prediction confidence threshold of 0.4. Finally, the predicted instance masks are further used to provide a statistical analysis that includes important metrics such as the particle size distinction.

The advent of small-scale testing procedures coupled with scanning electron microscopy (SEM) imaging allow for high-resolution digital image correlation (DIC) studies to examine strain localization at the grain size length scale. A systematic study was performed to determine how speckle patterning parameters (speckle density and shape) affect strain resolution of DIC using SEM imaging. Strain resolution increased with increased speckle density from 23 to 58 % area fraction. Patterns with less than 23 % area fraction exhibited significant signal noise, and a loss in strain resolution due to inadequate correlation. It was also observed that when the edges of square speckles were aligned with SEM rastering directions, the noise in the e yy data was double the noise in the e xx data. Rotating the speckle to eliminate edge alignment with the rastering direction significantly decreased the e yy strain noise. Competing optimization requirements for the correlation parameters were needed to minimize strain intensity noise or maximize spatial resolution. Application of the optimization techniques to high temperature in-situ studies of Ni-based superalloys will also be presented.