The rigorous coupled-wave analysis (RCWA) is a semi-analytic solver to Maxwell's equation, which is one of the most successful methods for modeling periodic optical structure. The repetitive nature of semiconductors has made RCWA widely applied in the semiconductor metrology industry. However, devices with high aspect ratio units, such as vertical NANDs(V-NANDs), require lengthy computation times, making them difficult to model in practice even with fully parallelized RCWA applications. This is because RCWA involves a time-consuming process of eigendecomposition and matrix inversion for each layer sliced along the vertical axis. In order to circumvent such computations, we propose a neural network based approach: channel-hole approximating network in the electromagnetic aspect (CHANEL). Based on the characteristic that the horizontal cutting plane is topologically consistent along the vertical axis of the channel-hole, CHANEL directly predicts the scattering matrix of each layer from its structural and optical parameters. In the scattering matrix of each layer, we found salient regions for Jones matrix calculation, which enhanced the accuracy of Jones matrix prediction with intensive learning on that area. In this paper, we demonstrate that CHANEL outperforms the traditional CPU-based RCWA implementations in terms of time, performing diffraction simulation more than 10 times faster.
KEYWORDS: Nanostructures, Data modeling, Machine learning, Scanning electron microscopy, 3D modeling, Signal processing, Semiconductors, Performance modeling, Nondestructive evaluation, Image processing
The trend to produce semiconductor devices having more complex nanostructures results in the increasing importance of exquisite systems measuring multiple Critical Dimensions (CDs) of nanostructures. However, from a practical point of view, it is difficult to apply conventional methodologies to mass production because of cost and complexity issues. In this study, we propose an application of machine learning techniques utilizing optical information to measure nanoscale profiles of channel holes in High-Aspect-Ratio (HAR) structure of vertical NAND flash, which is applicable to mass production. By combining the conventional methodologies, the proposed method yields data pairs for supervised learning which include optical spectra obtained with Rotating Polarizer Ellipsometer (RPE) and images obtained with Scanning Electron Microscopy (SEM). Several preprocessing steps and machine learning techniques are introduced to train a model with sufficient performance to be applicable to mass production. In experiments, we obtained a model with coefficient of determination (R2) of 0.8 and Root Mean Square Error (RMSE) of 1.3 nm when predicting hundreds of nanoscale profiles of the channel holes which are measured with SEM. Furthermore, we confirmed that only 500 samples of data are sufficient to achieve the model performance with R2 greater than 0.7 and RMSE less than 1.5 nm. The proposed method is capable of replacing the conventional methods of profile measurement in the mass production stage by reducing the cost of destructive methods and accurately measuring the profiles of complex nanostructures without theoretical modeling.
Conventional semiconductor etching process control has been performed by separated steps: process, metrology, and feedback control. Uniformity of structures such as Critical Dimension (CD) is an important factor in determining completeness of etching process. To achieve better uniformity, several feedback control has been performed. However, it is difficult to give feedback to the process after metrology due to the lack of process knowledge. In this study, we propose a machine learning technique that can create process control commands from the measured structure using a miniaturized Integrated Metrology (IM) of Spectroscopic Ellipsometery (SE) form. And it is possible to learn the physical analysis through machine learning without introducing a physical analysis method. The proposed analysis consists of two machine learning part: the first neural network for CD metrology, and second network for command generation. The first neural network takes a spectrum sampled at 2048 wavelengths obtained from IM as an input, and outputs CDs of structures. Finally, the second artificial neural network takes a changes of temperatures in a wafer and outputs the control commands of powers. As a result, we have improved the CD range of poly mask in a wafer from 1.69 nm to 1.36 nm.
The road network is one of the most important types of information in the Geographic Information System (GIS).
However, automatic extraction of roads is still considered a challenging problem. In this paper, we focus on robust
extraction of main roads. In the proposed algorithm, we first determine the roadness of each pixel using the eigenvalues
of its Hessian matrix. The roadness represents the belongingness of a pixel to a road; and its determination is performed
on a multi-scale basis so that it is robust to various widths of roads. We then perform directional grouping to the
determined initial road map and remove outliers in each group via directionally morphological filtering. Finally, we
determine roads by combining the results from each group. Experimental results show that the proposed algorithm can
automatically extract most main roads in various remote sensing images.
Recent infrared (IR) sensors are mostly based on a focal-plane array (FPA) structure. However, IR images suffer from
the fixed pattern noise (FPN) due to non-uniform response of a FPA structure. Various nonuniformity correction (NUC)
techniques have been developed to alleviate the FPN. They can be categorized into reference-based and scene-based
approaches. In order to deal with a temporal drift, however, a scene-based approach is needed. Among scene-based
algorithms, conventional algorithms compensate only for the offset non-uniformity of IR camera detectors based on the
global motion information. Local motions in a video, however, can introduce inaccurate motion information for NUC.
Considering global and local motions simultaneously, we propose a correction algorithm of gain and offset. Experiment
results using simulated and real IR videos show that the proposed algorithm provides performance improvement on the
FPN reduction.
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.