TSV (Through-Silicon Via) is a vertical interconnection structure achieved by creating holes in the silicon substrate of a chip. This paper explores the subtle differences in the reflection characteristics of light across various TSV structures. Electromagnetic simulation software, namely, Lumerical FDTD, based on the Finite-Difference Time-Domain (FDTD) method, was employed to establish electromagnetic models for TSV structures. 3D-FDTD simulations calculate near-field electromagnetic field data and derive near-field reflection spectra. The dataset comprises different TSV structures with varying critical dimensions (CD) and their corresponding far-field reflection spectra obtained through simulation. Our approach introduces two deep learning networks: the forward network and the inverse network. Given the time-consuming nature of FDTD electromagnetic simulation tools when dealing with complex structures, the forward network is trained to rapidly predict reflection spectrum signal data corresponding to CD parameters, effectively replacing simulation software. Using a threshold of 1% to determine the accuracy of the spectral predictions, forward prediction achieves 100% accuracy. Similarly, the inverse network is trained to predict the CD parameters of TSV structures from the reflection spectrum signal, allowing for fast and accurate inference of the dimensions of TSV structures and the extraction of geometric parameters. In reverse prediction, the MAPE (Mean Absolute Percentage Error) for Rtop, height, and Rbot remains consistently below 5%.
KEYWORDS: Education and training, Data modeling, Defect detection, Inspection, Titanium, Holmium, Light sources and illumination, Performance modeling, Cameras, Windows
In the field of automatic defect detection, a major challenge in training accurate classifiers using supervised learning is the insufficient and limited diversity of datasets. Obtaining an adequate amount of image data depicting defective surfaces in an industrial setting can be costly and time-consuming. Furthermore, the collected dataset may suffer from selection bias, resulting in an underrepresentation of certain defect classes. Our research aims to address surface defect detection in titanium metal spacer rings by introducing an approach leveraging a digital twin framework. The behavior of the digital twin is optimized using a reinforcement learning (RL) algorithm. The optimized digital twin is then used to generate synthetic data, which is employed to train a spacer defect detection classifier. The classifier’s performance is evaluated using real-world data. The results show that the model trained with synthetic data outperforms the one trained on a limited amount of real data. Our work highlights the potential of digital twin-based synthetic data generation and RL optimization in enhancing spacer surface defect detection and addressing the data scarcity challenge in the field. When the inspection network is trained solely using generated synthetic data, it achieves an inspection precision of 96.10%, with a recall of 85.77%. Incorporating real data alongside synthetic data for training further enhances performance, yielding an inspection precision of 93.07% and a recall of 94.20%. This surpasses the defect detection capabilities observed when training the inspection network exclusively with real data.
In the field of automatic defect detection, one of the major challenges for training accurate classifiers using supervised learning is the insufficient and limited diversity of datasets. Obtaining an adequate amount of image data depicting defective surfaces in an industrial setting is costly and time-consuming. Furthermore, the collected dataset may suffer from selection bias, resulting in underrepresentation of certain defect classes. This research aims to tackle the issue of surface defect detection in titanium metal spacer rings by introducing a novel approach that leverages a digital twin framework. The behavior of the digital representation is optimized using a reinforcement learning algorithm. Subsequently, the optimized digital twin is utilized to generate synthetic data, which is then employed to train a spacer defect detection classifier. The performance of this classifier is evaluated using real-world data. The results illustrate that the model trained with synthetic data outperforms the one trained on a limited amount of real data. This work emphasizes the potential of digital twin-based synthetic data generation and reinforcement learning optimization in enhancing spacer surface defect detection and addressing the data scarcity challenge in the field. When the generated synthetic data and real data combined is used to train inspection network, the inspection background accuracy reaches 93.07% and defect detection accuracy reaches 94.2% surpassing the defect detection performance of inspection network trained only using real data.
This study aims to improve defect detection and optimize deep learning models for wafer surfaces. Deep learning methods for high-yield wafer processes often suffer from insufficient sample size. The challenge of generating enough defective samples hinders the model's ability to accurately detect defects through training. This study resolves the issue by using a generative adversarial network style transfer method to generate synthetic data in batches and flexibly determine the location of defect generation. The optimized model achieved an accuracy reduction of only 2% and a 55% reduction in inference time. Edge computing was also applied for online, real-time detection of defects.
Wafer defects are caused by defects on the surface of the Wafer grinding jig. Therefore, this research is aimed at optical inspection of the self-made dataset of the Wafer grinding jig by using the UNet++ network model. Because of the different background distribution of the inner and outer edges in Wafer grinding jig, it is difficult to detect defects solely by Deep learning. Hence, this study uses Image processing to separate the inner and outer edge and then uses UNet++ with transfer learning to detect the location of the defects. In addition, false detection is reduced by tracking the defects as the jig moves regularly. The proposed deep network layer reduction method can decrease the detection time to 48.27% compared to the original network model. The recall rate and IoU of the outer edge are increased by 13.8% and 16.33% through transfer learning. The recall rate and IoU of the inner edge are increased by 2.01% and 1.16% respectively through transfer learning. And the final defect tracking recall rate reached over 75.81%, while the F1 score reached over 77.08% in every frame from 112 to 192 frames.
This paper proposes a machine-vision-based system for the inspection of the geometrical features of cell phone rubber gaskets. The system consists of two industrial cameras and an industrial computer. To begin the inspection process, the cell phone gaskets inside the camera field were identified and a novel transparent fixture were employed to guide the lighting to highlight the defects. To characterize the metrological features of cell phone gaskets, image preprocessing procedure was then implemented and defect classification algorithmic strategies was described in detail. Finally, experimental results on images of different type of the cell phone gaskets are reported together with the metrological classification of the proposed measurement system. The geometrical quality of every individual silicone rubber gasket, which could be evaluated and guaranteed to the full grade of assessment which the proposed system can achieve.
KEYWORDS: Spindles, Acoustics, Signal processing, Finite element methods, Signal detection, Surface roughness, Feedback control, Surface finishing, Signal analysis, Sensors
Chatter suppression in milling machines, using acoustic signals, is examined for preventing poor finishes and for
upgrading production quality. The technique of on-line acoustic signal analysis for suppressing chatter has been
examined in realistic milling experiments. The natural frequency was determined for end mills, based on finite element
analysis, so that beam equations could be used for chatter detection. Frequency domain analysis was employed to
monitor the shift in the dominant frequency. In this study, the acoustic chatter signal frequency and resonant peak were
applied using a feedback control loop to quantify the acoustic signal and to compensate for spindle speed, respectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.