Atsushi Miyamoto
Atsushi Miyamoto has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
Akira Ito
Hitachi, Ltd, Research and Development Group, 292 Yoshida-cho, Totsuka-ku
2 shared publications- Kenji Nakahira
Tamagawa University
2 shared publications Minoru Harada
Hitachi, Ltd., Research and Development Group , 292 Yoshida-cho, Totsuka-ku, Yokohama, Kanagawa 244-0817, Japan
2 shared publications- Akihiro Tsusaka
Department of Electrical Engineering, Aichi Institute of Technology 1247 Yachigusa Yakusa‐cho Toyota 470‐0391 Japan
1 shared publication - Akira Ito
Hitachi, Ltd., Research and Development Group , 292 Yoshida-cho, Totsuka-ku, Yokohama, Kanagawa 244-0817, Japan
1 shared publication Gou Koutaki
Graduate School of Science and Technology
1 shared publication
Latest Publications
- 26 December 2025
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 75 (4) , 343-353
Semi-supervised semantic segmentation of SEM images considering multi-scale structural consistency loss in semiconductor pattern layouts
In the fabrication of semiconductor devices, increased yield is achieved using Scanning Electron Microscopes (SEM) to measure and inspect circuit patterns. With the recent decreasing scale and increasing complexity of semiconductor circuit patterns, it has become increasingly difficult to recognize patterns accurately using rule-based image processing methods. As such, we propose a method that uses semi-supervised learning for segmentation processing to recognize which pattern level each pixel represents. With existing methods, the pseudo-labels used for training were not accurate enough, and there were issues such as inconsistent recognition of repeated-pattern layouts and mixed-up results in large unmarked areas distant from the pattern contour. Accordingly, the proposed method is able to perform highly accurate segmentation with the design of new types of loss for evaluating consistency in pattern structure at various scales. When compared with Unimatch and CAC, which are well-known high-performance segmentation methods, the accuracy relative to visual identification increased dramatically, from 10–12% to 100%. In quantitative evaluation using mean Intersection-over-Union (mIoU) at the pixel level, mean values also increased from a range between 0.45 and 0.65 to over 0.94, confirming that the proposed method is effective.
- 26 December 2025
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 75 (4) , 343-353
Semi-supervised semantic segmentation of SEM images considering multi-scale structural consistency loss in semiconductor pattern layouts
In the fabrication of semiconductor devices, increased yield is achieved using Scanning Electron Microscopes (SEM) to measure and inspect circuit patterns. With the recent decreasing scale and increasing complexity of semiconductor circuit patterns, it has become increasingly difficult to recognize patterns accurately using rule-based image processing methods. As such, we propose a method that uses semi-supervised learning for segmentation processing to recognize which pattern level each pixel represents. With existing methods, the pseudo-labels used for training were not accurate enough, and there were issues such as inconsistent recognition of repeated-pattern layouts and mixed-up results in large unmarked areas distant from the pattern contour. Accordingly, the proposed method is able to perform highly accurate segmentation with the design of new types of loss for evaluating consistency in pattern structure at various scales. When compared with Unimatch and CAC, which are well-known high-performance segmentation methods, the accuracy relative to visual identification increased dramatically, from 10–12% to 100%. In quantitative evaluation using mean Intersection-over-Union (mIoU) at the pixel level, mean values also increased from a range between 0.45 and 0.65 to over 0.94, confirming that the proposed method is effective.
- 4 February 2023
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 72 (5) , 408-417
Super-resolution method for SEM images based on pixelwise weighted loss function
Scanning electron microscopy (SEM) has realized high-throughput defect monitoring of semiconductor devices. As miniaturization and complexification of semiconductor circuit patterns increase in recent years, so has the number of defects. There is thus a great need to further increase the throughput of SEM defect monitoring. Toward this end, we propose a deep learning–based super-resolution method that reproduces high-resolution (HR) images from corresponding low-resolution images. Image quality factors such as pattern contrast and sharpness are important in SEM HR images in order to evaluate the quality of printed circuit patterns. Our proposed method meets various image quality requirements by changing the loss calculation method pixelwise based on the pattern in the image. It realizes super-resolved images that compare favorably with actual HR images and can improve SEM throughput by 100% or more.
- 4 February 2023
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 72 (5) , 408-417
Super-resolution method for SEM images based on pixelwise weighted loss function
Scanning electron microscopy (SEM) has realized high-throughput defect monitoring of semiconductor devices. As miniaturization and complexification of semiconductor circuit patterns increase in recent years, so has the number of defects. There is thus a great need to further increase the throughput of SEM defect monitoring. Toward this end, we propose a deep learning–based super-resolution method that reproduces high-resolution (HR) images from corresponding low-resolution images. Image quality factors such as pattern contrast and sharpness are important in SEM HR images in order to evaluate the quality of printed circuit patterns. Our proposed method meets various image quality requirements by changing the loss calculation method pixelwise based on the pattern in the image. It realizes super-resolved images that compare favorably with actual HR images and can improve SEM throughput by 100% or more.
- Open Access
- 19 April 2022
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 71 (4) , 214-221
Machine-learning-based quality-level-estimation system for inspecting steel microstructures
Quality control of special steel is accomplished through visual inspection of its microstructure based on microscopic images. This study proposes an ‘automatic-quality-level-estimation system’ based on machine learning. Visual inspection of this type is sensory-based, so training data may include variations in judgments and training errors due to individual differences between inspectors, which makes it easy for a drop in generalization performance to occur due to overfitting. To deal with this issue, we here propose the preprocessing of inspection images and a data augmentation technique. Preprocessing reduces variation in images by extracting features that are highly related to the level of quality from inspection images. Data augmentation, meanwhile, suppresses the problem of overfitting when training with a small number of images by taking into account information on variation in judgment values obtained from on-site experience. While the correct-answer rate for judging the quality level by an inspector was about 90%, the proposed method achieved a correct-answer rate of 92.5%, which indicates that the method shows promise for practical applications.
- 19 April 2022
- Published by Oxford University Press (OUP) in Microscopy
- Vol. 71 (4) , 214-221
Open AccessMachine-learning-based quality-level-estimation system for inspecting steel microstructures
Quality control of special steel is accomplished through visual inspection of its microstructure based on microscopic images. This study proposes an ‘automatic-quality-level-estimation system’ based on machine learning. Visual inspection of this type is sensory-based, so training data may include variations in judgments and training errors due to individual differences between inspectors, which makes it easy for a drop in generalization performance to occur due to overfitting. To deal with this issue, we here propose the preprocessing of inspection images and a data augmentation technique. Preprocessing reduces variation in images by extracting features that are highly related to the level of quality from inspection images. Data augmentation, meanwhile, suppresses the problem of overfitting when training with a small number of images by taking into account information on variation in judgment values obtained from on-site experience. While the correct-answer rate for judging the quality level by an inspector was about 90%, the proposed method achieved a correct-answer rate of 92.5%, which indicates that the method shows promise for practical applications.
- 3 September 2021
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 32 (12) , 125403
A technique for high-precision drift correction in electron microscopes
We propose a technique for improving the image quality of electron microscopes, which are widely used for the observation and analysis of microscopic structures. To improve the visual recognition of frame addition images of electron microscopes, it is necessary to perform correlation calculations in order to line up the frame images and eliminate the small positional offsets (drifts) between them. However, since the individual frames have a very low signal-to-noise ratio, it is difficult to estimate these drifts stably. To address this problem, we propose a phase-only correlation method that uses a combination of noise suppression and highly accurate estimation of phase images based on complex wavelet transform. By implementing this method in a parallel processor, we have made it possible to observe fine structures with high definition in real time (30 frame s −1 ).
- 3 September 2021
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 32 (12) , 125403
A technique for high-precision drift correction in electron microscopes
We propose a technique for improving the image quality of electron microscopes, which are widely used for the observation and analysis of microscopic structures. To improve the visual recognition of frame addition images of electron microscopes, it is necessary to perform correlation calculations in order to line up the frame images and eliminate the small positional offsets (drifts) between them. However, since the individual frames have a very low signal-to-noise ratio, it is difficult to estimate these drifts stably. To address this problem, we propose a phase-only correlation method that uses a combination of noise suppression and highly accurate estimation of phase images based on complex wavelet transform. By implementing this method in a parallel processor, we have made it possible to observe fine structures with high definition in real time (30 frame s −1 ).
- 1 October 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 354-357
An Experimental Study on Arc Discharge Occured by Breaking of Line in Low-Voltage PV Systems
In recent years, the installation of photovoltaic power generation systems (PV-system) has been increasing because of the promotion of the use of renewable energy. However, a fire accident has occurred by an arc discharge due to a disconnection or short circuit of the photovoltaic systems. Therefore, in order to prevent the fire accident, it is necessary to promptly extinguish the arc discharge. In this paper, as a first step, we focused on DC arc caused by disconnection in PV systems and experimentally investigated the process of its occurrence and disappearance. As a result, it was confirmed that the arcing time occurred by the solar power supply is longer than that generated by the DC constant-voltage supply.
- 1 October 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 354-357
An Experimental Study on Arc Discharge Occured by Breaking of Line in Low-Voltage PV Systems
In recent years, the installation of photovoltaic power generation systems (PV-system) has been increasing because of the promotion of the use of renewable energy. However, a fire accident has occurred by an arc discharge due to a disconnection or short circuit of the photovoltaic systems. Therefore, in order to prevent the fire accident, it is necessary to promptly extinguish the arc discharge. In this paper, as a first step, we focused on DC arc caused by disconnection in PV systems and experimentally investigated the process of its occurrence and disappearance. As a result, it was confirmed that the arcing time occurred by the solar power supply is longer than that generated by the DC constant-voltage supply.
- 8 January 2019
- Published by Wiley in Electronics and Communications in Japan
- Vol. 102 (3) , 36-44
Automatic extraction technique of CD‐SEM evaluation points to measure semiconductor overlay error
- 8 January 2019
- Published by Wiley in Electronics and Communications in Japan
- Vol. 102 (3) , 36-44
Automatic extraction technique of CD‐SEM evaluation points to measure semiconductor overlay error
- 16 February 2017
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 28 (4) , 045405
Robust surface reconstruction by design-guided SEM photometric stereo
We present a novel approach that addresses the blind reconstruction problem in scanning electron microscope (SEM) photometric stereo for complicated semiconductor patterns to be measured. In our previous work, we developed a bootstrapping de-shadowing and self-calibration (BDS) method, which automatically calibrates the parameter of the gradient measurement formulas and resolves shadowing errors for estimating an accurate three-dimensional (3D) shape and underlying shadowless images. Experimental results on 3D surface reconstruction demonstrated the significance of the BDS method for simple shapes, such as an isolated line pattern. However, we found that complicated shapes, such as line-and-space (L&S) and multilayered patterns, produce deformed and inaccurate measurement results. This problem is due to brightness fluctuations in the SEM images, which are mainly caused by the energy fluctuations of the primary electron beam, variations in the electronic expanse inside a specimen, and electrical charging of specimens. Despite these being essential difficulties encountered in SEM photometric stereo, it is difficult to model accurately all the complicated physical phenomena of electronic behavior. We improved the robustness of the surface reconstruction in order to deal with these practical difficulties with complicated shapes. Here, design data are useful clues as to the pattern layout and layer information of integrated semiconductors. We used the design data as a guide of the measured shape and incorporated a geometrical constraint term to evaluate the difference between the measured and designed shapes into the objective function of the BDS method. Because the true shape does not necessarily correspond to the designed one, we use an iterative scheme to develop proper guide patterns and a 3D surface that provides both a less distorted and more accurate 3D shape after convergence. Extensive experiments on real image data demonstrate the robustness and effectiveness of our method.
- 16 February 2017
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 28 (4) , 045405
Robust surface reconstruction by design-guided SEM photometric stereo
We present a novel approach that addresses the blind reconstruction problem in scanning electron microscope (SEM) photometric stereo for complicated semiconductor patterns to be measured. In our previous work, we developed a bootstrapping de-shadowing and self-calibration (BDS) method, which automatically calibrates the parameter of the gradient measurement formulas and resolves shadowing errors for estimating an accurate three-dimensional (3D) shape and underlying shadowless images. Experimental results on 3D surface reconstruction demonstrated the significance of the BDS method for simple shapes, such as an isolated line pattern. However, we found that complicated shapes, such as line-and-space (L&S) and multilayered patterns, produce deformed and inaccurate measurement results. This problem is due to brightness fluctuations in the SEM images, which are mainly caused by the energy fluctuations of the primary electron beam, variations in the electronic expanse inside a specimen, and electrical charging of specimens. Despite these being essential difficulties encountered in SEM photometric stereo, it is difficult to model accurately all the complicated physical phenomena of electronic behavior. We improved the robustness of the surface reconstruction in order to deal with these practical difficulties with complicated shapes. Here, design data are useful clues as to the pattern layout and layer information of integrated semiconductors. We used the design data as a guide of the measured shape and incorporated a geometrical constraint term to evaluate the difference between the measured and designed shapes into the objective function of the BDS method. Because the true shape does not necessarily correspond to the designed one, we use an iterative scheme to develop proper guide patterns and a 3D surface that provides both a less distorted and more accurate 3D shape after convergence. Extensive experiments on real image data demonstrate the robustness and effectiveness of our method.
- 13 January 2016
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 27 (2) , 025407
Generation of large field SEM image by panorama composition technology for nano-order measurement
Semiconductor manufacturing has a pressing need for a method to accurately evaluate the global shape deformation of a photomask pattern. We thus propose a novel composition technique for a large field panorama image of scanning electron microscopy (SEM). The proposed method optimises the arrangement of segmented imaging regions (SIRs), which are components of a panorama image, on the basis of the design data of the photomask pattern layout. The quantity of the line pattern segment, which is a clue to the connection in an overlapping region between adjoining SIRs and the connectability of any two SIRs, is evaluated. As a result of the optimisation, it is guaranteed that all SIR images can be connected theoretically. For 30 evaluation points, the maximum connection error of the SIR images was 1.5 nm in a simulation using pseudo-SEM images. The maximum total measurement error, which includes the connection error and CD measurement error from the panorama image, is estimated at 2.5 nm. This error was equivalent to about 1.4% of the photomask line width (target: 3%). The experiments using real SEM images demonstrate the effectiveness of the proposed method. It was visually confirmed that a large field, high-resolution and seamless panorama image can be generated.
- 13 January 2016
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 27 (2) , 025407
Generation of large field SEM image by panorama composition technology for nano-order measurement
Semiconductor manufacturing has a pressing need for a method to accurately evaluate the global shape deformation of a photomask pattern. We thus propose a novel composition technique for a large field panorama image of scanning electron microscopy (SEM). The proposed method optimises the arrangement of segmented imaging regions (SIRs), which are components of a panorama image, on the basis of the design data of the photomask pattern layout. The quantity of the line pattern segment, which is a clue to the connection in an overlapping region between adjoining SIRs and the connectability of any two SIRs, is evaluated. As a result of the optimisation, it is guaranteed that all SIR images can be connected theoretically. For 30 evaluation points, the maximum connection error of the SIR images was 1.5 nm in a simulation using pseudo-SEM images. The maximum total measurement error, which includes the connection error and CD measurement error from the panorama image, is estimated at 2.5 nm. This error was equivalent to about 1.4% of the photomask line width (target: 3%). The experiments using real SEM images demonstrate the effectiveness of the proposed method. It was visually confirmed that a large field, high-resolution and seamless panorama image can be generated.
- Open Access
- 1 January 2015
- Published by Japan Society for Precision Engineering in Journal of the Japan Society for Precision Engineering
- Vol. 81 (12) , 1102-1106
Automatic Generation of Measurement Recipe for CD-SEM Using Design Data
- 1 January 2015
- Published by Japan Society for Precision Engineering in Journal of the Japan Society for Precision Engineering
- Vol. 81 (12) , 1102-1106
Open AccessAutomatic Generation of Measurement Recipe for CD-SEM Using Design Data
- 1 July 2014
- Published by Elsevier in Applied and Computational Harmonic Analysis
- Vol. 37 (1) , 1-11
Multi-link wavelets on hierarchical graphs
- 1 July 2014
- Published by Elsevier in Applied and Computational Harmonic Analysis
- Vol. 37 (1) , 1-11
Multi-link wavelets on hierarchical graphs