Detecting defects in the inspection stage of semiconductor manufacturing process is a crucial task to improve yield and productivity as well as wafer quality. Recent Advances in semiconductor process ...technology have greatly increased the transistor density. As a result, an increasingly high number of defects inevitably emerge and we need a more accurate and efficient detection method to manage them. In this paper, we propose a deep-learning-based defect detection model to expedite the process. It adopts an adversarial network architecture of conditional GAN. The discriminator of an adversarial network architecture helps the detection model learn to detect and classify defects accurately. The high performance is achieved by using Focal Loss, PixelGAN and multi-scale level features, which is shown to be better than the baseline model, CenterNet, when tested for a real industrial dataset.
Alignment between the reference layout (or target pattern) and the corresponding scanning electron microscope (SEM) image is a crucial task for the die-to-database (D2DB) inspection in the ...semiconductor industry. However, it is challenging to align them accurately because the style and quality of reference layouts represented as a computer-aided design (CAD) are quite different from those of grayscale SEM images with noise. Direct application of conventional cross-correlation-based matching methods often leads to misalignment. Here, we propose a new method enabling the precise pattern alignment. The main idea is to transform SEM images into target-like CAD images using a generative adversarial network (GAN). As the generated and real layout images have similar style and quality, they can be precisely aligned using conventional matching methods. Then, the SEM image can be located correctly on the corresponding reference layout for inspection. A polygon-based clustering algorithm for target patterns is developed to avoid manual selection and minimize the number of training data.
Since the invention of transistors and integrated circuits, the development of semiconductor processes has advanced rapidly. Current microchips contain hundreds of millions of transistors. The ...remarkable development of semiconductors thus far has also led to difficulties in designing tightly packed lithography patterns without unwanted defects called hotspots in the manufacturing process. Therefore, research areas focusing on these problems have received much attention. In particular, predicting hotspots during the design stage is essential for high productivity in the semiconductor industry. In this study, we developed a deep learning-based SEM image generation model to predict hotspots from layout patterns at the design stage. Our model combines a segmentation network and an image-to-image translation network based on a conditional generative adversarial network in parallel. Our proposed model can predict and display potential hotspots in scanning electron microscopy images generated from given layouts. Additionally, the model leverages prior knowledge of the optical diameter to predict patterns that are prone to hotspots. Our model shows improved performance over baseline models when evaluated on real-world industrial data.
The aim of this paper is to propose a hybrid framework that combines a data-driven pose estimation with model-based force calculation in order to predict the ski jumping force from a recorded motion ...video. A skeletal model consisting of five joints (ear, hip, knee, ankle, and toe) and four rigid segments (head/arm/trunk or HAT, thigh, shank, and foot) connecting each joint is developed. The joint forces are calculated from the dynamic equilibrium equations, which requires the time history of joint coordinates. They are estimated from a recorded motion video using a deep neural network for pose estimation trained with human motion data. Joint coordinates can be obtained by the proposed deep neural network directly from images of jumping motion without using any markers. The validity and usefulness of the proposed method are confirmed in lab experiments. Further, our method is practically applicable to the study in a real competition environment because it is not required to attach any sensor or marker to athletes.
Highlights
A method to predict the ski jumping force from a recorded motion video is proposed.
It combines a data-driven pose estimation with a model-based force calculation.
The proposed method does not require any markers and sensors to be attached to athletes.
In a laboratory environment, the relative error in the maximum jumping force is less than 7%.
The method can be easily applied to a field study in a real competition environment.