3D Imaging Of Dual Secondary Electron Detectors In SEM

Since the beginning of the 21st century,with the continuous development of nanotechnology,nanomaterials have been widely used in the fields of electronic chip manufacturing,packaging,biomedicine,energy and transportation.Compared with the traditional semiconductor manufacturing technology,nanotechnology has the characteristics of high efficiency,reliability and stability in the fields of 3D nano-device manufacturing,so it has a better development prospect.Scanning Electron Microscope(SEM)is usually used as an essential tool for studying nanotechnology due to its characteristics of real-time visualization and nano precision two-dimensional observation.The scanning electron microscope imprints the two-dimensional plane of the object on the surface of the electron,which lacks the depth information,therefore,it brings a lot of inconvenience to 3D nanometer operation.In this thesis,secondary electron detector and Inlens detector under scanning electron microscope will be combined.Firstly,the principle of secondary electron imaging under the scanning electron microscope is analyzed.Secondly,the time-shift distortion of the SEM image is removed by the time drift distortion model,and the dual detectors are calibrated by Zhang Zhengyou's calibration method.Thirdly,Super-resolution Convolutional Neural Networks(SRCNN)are used to enhance the resolution of secondary electronic images.Finally,the depth calculation of SEM image is realized by binocular vision.The main research contents of this thesis are as follows:(1)The SEM used in this project is MERLIN Compact field emission electron microscope produced by Zeiss company in Germany,and the principle of secondary electron imaging is mainly studied and analyzed.For the issue of inconsistent imaging models at different magnifications,a general imaging model suitable for different magnifications was established.When the magnification was higher than 1000×,the parallel projection model was used,while the magnification was lower than 1000×,the transmission projection model was used.To solve the issue of time-shift distortion in SEM images,the time drift distortion was modeled and removed.In order to solve the issue of spatial distortion in the image taken by the dual secondary electron detectors(SE2 detector and Inlens detector),the calibration method of Zhang zhengyou is adopted to obtain the inner and outer parameters of the detector and remove the distortion.For the issue of different pixel errors under different work distance(WD),the calibration program found that when the work distance was 6.4mm?6.5mm,the pixel error was 0.48pixel and the error reached the minimum.The subsequent 3D depth information calculation was carried out under this work distance.(2)For the modification of the cabin line in the SEM,there is signal interference in the captured images,which causes the image clarity to be low,and the parallax problem between the secondary electronic image pairs cannot be found.Combined with the super-resolution convolutional neural network algorithm.Firstly,the sample image data sets were built for the SE2 detector and the Inlens detector,respectively.Secondly,for the issue of noise in SEM images,gaussian filter is adopted to remove the gaussian noise in SEM images.Thirdly,in view of the issue of long network training time,the Region of Interest(ROI)of image features is selected to compress the image size to reduce the training time.Through the trained image data set,the SEM image can be processed online in real time.Finally,the Peak Signal to Noise Ratio(PSNR)of the evaluation function is introduced.The measured PSNR value of the trained SEM image is 45.46dB,which achieves the purpose of enhancing the resolution of the SEM image and obtains the parallax between the secondary electron image pairs.(3)Based on the above calibrated dual secondary electron detectors and the enhancement of image resolution,the depth information of SEM image is calculated by binocular vision method.At the work distance with the minimum pixel error,the image of the sample was acquired simultaneously by SE2 detector and Inlens detector.For the issue that the difference between the brightness information of the secondary electronic image captured by the SE2 detector and the Inlens detector makes it difficult to match feature point,the SIFT(Scale-invariant feature transform)algorithm is adopted to extract feature points on the secondary electronic image and match them,and the accuracy of the matching result is high.In order to solve the issue that the parallax between the pairs of secondary electronic images is not obvious.Firstly,the polar line correction method is used to correct the SE2 image and the Inlens image,so that the coordinates of the image on the vertical axis are equal,and the parallax of the image on the horizontal axis is retained.Then,in order to obtain the horizontal parallax between the dual secondary electronic image pairs,a method based on DIC(Digital image correlation)was proposed to calculate the parallax of SEM image pairs.Finally,the depth information of the SEM image is calculated by the parallax and depth mapping relation of the SEM image.By comparing with the actual value,it is found that the average calculation error is about 20.5%.