3D-NAND memory will continue to increase in the aspect ratio of channel holes. High throughput and in-line monitoring solutions for 3D profiling of high aspect ratio (HAR) features are the key for yield improvement. A deep learning (DL) model has been developed to improve the 3D profiling accuracy of the HAR features. In this work, the HAR holes with different bowing geometries were fabricated and a high-voltage CD-SEM was used to evaluate the performance of the DL model. The accuracy and the sensitivity of the DL model was evaluated by comparing the predicted cross-sections with the X-SEM measurement. The results show that the DL model enables the maximum CD (MCD) of the bowing features to be predicted with a sensitivity of 0.93 and its depth position to be predicted with a sensitivity of 0.91. The DL learning model reduced the absolute error of the predicted MCD depth position from several hundreds of nanometers, the error occurring when using the exponential model, to within 100 nm.
We applied deep learning techniques to improve the accuracy of 3D-profiling for high aspect ratio (HAR) holes. As deep learning requires big data for training, we developed a method for generating a large amount of BSE line-profiles by a numerical calculation in which the aperture angle and the aberration effects of the electron beam are considered. We then utilized these numerically calculated datasets to train the deep learning model to learn the mapping from the BSE line-profiles to the target cross-sectional profiles of the HAR holes. Two different one-dimensional neural network architectures: convolutional neural network (CNN) and multi-scale convolutional neural network (MS-CNN) were trained, and different loss functions were investigated to optimize the networks. The test results show that the MS-CNN model with a defined loss function of weighted mean square error (WMSE) provided higher accuracy than the others. The mean absolute percentage error (MAPE) distribution was narrow and the typical MAPE was 4% over 2810 items of test data. This model enables us to predict the cross-section of the HAR holes with different sidewall profiles more accurately than our previously proposed exponential model. These results demonstrate the effectiveness of the learning approach for improving the accuracy of 3D-profiling of the HAR features.
A depth measurement technique for extremely deep holes (such as channel holes in 3D flash memory devices)—by using back-scattered-electron (BSE) images obtained by a high voltage critical dimension scanning electron microscope (CDSEM)— was developed. A high voltage CD-SEM can detect BSEs that penetrate solids surrounding deep holes. These BSE images include rich information concerning the bottom structures of deep holes. As the BSEs lose their energies according to the distance they travel in solids, it is deduced that the BSE image intensity at hole bottoms depends on hole depth. In a feasibility study on depth measurement using an SEM simulator, it was found that the intensity also depends on hole diameter. The relationship between BSE intensity, hole depth, and hole diameter was modeled by simplifying a backscattering model and approximating the target medium by volume density. Based on this model, a depth measurement technique using only a top-view BSE image is proposed. Measurement error of the technique for channel holes of a 3D flash memory device with depths of a few microns was evaluated by using a high voltage CD-SEM. According to the results of the evaluation, error range was 62 nm and measurement repeatability was ± 18 nm. It is concluded that these values are sufficient for detecting depth defects. This technique achieves fast and non-destructive depth measurement of individual extremely deep holes.