Critical dimension scanning electron microscope image processing method and electronic equipment

Abstract

The present invention discloses a critical dimension scanning electron microscope image processing method and electronic device. The method includes using a pre-trained image processing model to process an input image to obtain an optimized image. The pre-trained image processing model includes a denoising branch and an edge enhancement branch, both of which are constructed based on the U-Net network architecture. The denoising branch includes a first encoder module and a first decoder module connected in sequence, and the first encoder module and the first decoder module both use SC block modules to extract local and global features of the image. The edge enhancement branch includes an edge information extraction module, a second encoder module and a second decoder module connected in sequence. The method also includes a feature fusion module for fusing the outputs of the first encoder module and the second encoder module at each level to serve as the inputs of the first encoder module at each level. Compared with traditional image denoising algorithms, this algorithm solves the edge blurring problem caused by over-smoothing of images while denoising.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a critical dimension scanning electron microscope image processing method and electronic equipment. Background Art

[0002] With the continuous advancement of high numerical aperture extreme ultraviolet lithography (High NA EUVL) technology, extensive research into thinner photoresists has become increasingly urgent. However, photoresist shrinkage remains a significant issue in lithography feature metrology during critical dimension electron scanning microscopy (CD-SEM) measurement. This not only affects CD-SEM measurement accuracy but can also impact process yields due to electron beam damage to the sample. Thinning photoresist exacerbates this shrinkage issue, posing a significant challenge. To minimize the impact of resist shrinkage, SEM images must be acquired using lower voltages or currents and fewer scan frames. However, this inevitably introduces image noise, reducing metrology accuracy. Unbiased line width roughness (LWR) estimates become less accurate when the signal-to-noise ratio (SNR) falls below 2, and errors can reach as high as 30% when the SNR reaches 1. Therefore, a high SNR is crucial for reliable LWR estimation. However, attempting to obtain high SNR images by modifying CD-SEM settings can result in excessively long measurement times, reduced throughput, and material damage. Under extreme conditions, low signal-to-noise ratio CD-SEM images may not be able to be correlated using edge profile-based algorithms. Therefore, designing effective image denoising methods is crucial to obtain accurate metrology while minimizing photoresist shrinkage.

[0003] In recent years, with the rise of deep learning technologies, deep learning methods have become increasingly important in the field of denoising. They overcome the limitations of traditional methods, which rely on prior information and artificial features, while also possessing strong generalization capabilities. Examples include the feedforward denoising convolutional neural network (DnCNN), the convolutional blind denoising network (CBDNet), the U-shaped Transformer network (Uformer), the attention-guided denoising CNN (ADNet), and the multi-stage image denoising convolutional neural network (MWDCNN). While existing CD-SEM image denoising algorithms have achieved promising results, they still face challenges such as over-smoothing of the denoised images and loss of edge information. Summary of the Invention

[0004] The embodiments of the present invention provide a critical dimension scanning electron microscope image processing method and electronic equipment to solve the problems of over-smoothing of denoised images and loss of edge information that still exist in image denoising algorithms in the prior art.

[0005] According to a first aspect of the present invention, a critical dimension scanning electron microscope image processing method is provided, comprising:

[0006] Use a pre-trained image processing model to process the input image to obtain an optimized image;

[0007] The pre-trained image processing model includes:

[0008] The denoising branch is based on the U-Net convolutional neural network architecture and includes a first encoder module and a first decoder module connected in sequence. Both the first encoder module and the first decoder module use the SC block module to extract local and global features of the image.

[0009] The edge enhancement branch is based on the U-Net convolutional neural network architecture, which includes an edge information extraction module, a second encoder module, and a second decoder module connected in sequence;

[0010] A feature fusion module is used to fuse the outputs of each level of the first encoder module and the second encoder module as the inputs of each level of the first encoder module.

[0011] According to a second aspect of the present invention, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the critical dimension scanning electron microscope image processing method of the first aspect mentioned above.

[0012] According to a third aspect of the present invention, there is provided a storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the critical dimension scanning electron microscope image processing method of the first aspect.

[0013] The critical dimension scanning electron microscope (CD-SEM) image processing method of this invention transmits the input image in parallel to an edge enhancement module and a denoising module. The denoising module, which includes an SC block, enhances the ability to extract global and local features of the input image. The edge enhancement branch first obtains edge information from the input image through the edge information extraction module, then extracts global and local features of this edge information. A feature fusion module also fuses the features of the encoder and decoder modules of the two branches to compensate for the shortcomings of the denoising branch in edge feature extraction. Compared with traditional image denoising algorithms, this algorithm not only removes noise but also solves the edge blurring problem caused by over-smoothing of the image. Experimental results demonstrate that the algorithm achieves good denoising results in test images and improves the metrology accuracy of critical dimension scanning electron microscope (CD-SEM). Furthermore, the proposed CD-SEM image processing method provides an effective solution to the challenges posed by photoresist shrinkage and is of great significance for promoting the development of high numerical aperture extreme ultraviolet (EUV) lithography. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 A schematic diagram of the network structure of an image processing model used in a critical dimension scanning electron microscope image processing method according to one embodiment of the present invention;

[0016] Figure 2 This is a block diagram of the SC block module in the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention;

[0017] Figure 3 This is a flow chart of processing an input feature map by an SC block module in a critical dimension scanning electron microscope image processing method according to one embodiment of the present invention;

[0018] Figure 4 Schematic diagram of four trainable Soble operators in horizontal, vertical and diagonal directions used in edge enhancement processing in a critical dimension scanning electron microscope image processing method according to one embodiment of the present invention;

[0019] Figure 5 1. A block diagram of a cross-attention fusion module in a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention;

[0020] Figure 6A process for processing input features of a cross-attention fusion module in a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention;

[0021] Figure 7 This is a flow chart of the steps of performing noise simulation processing on an original SEM image in a critical dimension scanning electron microscope image processing method according to one embodiment of the present invention;

[0022] Figure 8 A diagram comparing the processing results of a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and the processing results of an existing traditional processing method;

[0023] Figure 9 A diagram comparing the visual effects of a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and the visual effects of a conventional processing method;

[0024] Figure 10 1. A comparison chart of CD values ​​of three randomly selected CD-SEM images processed by a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and by an existing traditional processing method;

[0025] Figure 11 1. A comparison chart of LER values ​​of three randomly selected CD-SEM images processed by a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and by an existing traditional processing method;

[0026] Figure 12 1. A comparison chart of LWR values ​​of three randomly selected CD-SEM images processed by a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and by an existing traditional processing method;

[0027] Figure 13 for Figures 10 to 12 A line chart comparing the CD, LWR, and LER values ​​under each treatment method;

[0028] Figure 14 FIG. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0030] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0031] In the description of the present invention, it should be understood that if the terms "center", "middle", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "axial", "radial", "circumferential" and the like appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The features defined as "first" and "second" are used to distinguish the feature names, and do not have special meanings. In addition, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0032] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0033] It should also be noted that, in this document, the terms "include" and "comprising" include not only those elements, but also other elements not explicitly listed, or elements inherent to such processes, methods, articles, or devices. In the absence of further limitations, elements defined by the phrase "include..." do not exclude the presence of other identical elements in the process, method, article, or device that includes the elements. The terms used in this document are generally commonly used by those skilled in the art. If there is any inconsistency with commonly used terms, the terms in this document shall prevail.

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0035] The present invention will be further described in detail below with reference to the accompanying drawings.

[0036] Specifically, in the present invention, a pre-trained image processing model is used to process the input image to obtain an optimized image, thereby achieving denoising of the image and effectively reducing the problem of edge information loss. In the present invention, the pre-trained image processing model used includes a denoising branch and an edge enhancement branch. The denoising branch and the edge enhancement branch are both based on the U-Net convolutional neural network architecture. The denoising branch includes a first encoder module and a first decoder module connected in sequence, and the first encoder module and the first decoder module both use an SCblock module to extract local and global features of the image. The edge enhancement branch includes an edge information extraction module, a second encoder module and a second decoder module connected in sequence, and the edge information extraction module is used to obtain the edge information of the input image, and then extract the global and local features of the edge information. The image processing model also includes a feature fusion module for fusing the outputs of the first encoder module and the second encoder module at each level as the input of the first encoder at each level.

[0037] Specifically, Figure 1 The network structure diagram of the image processing model used in the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention is schematically shown. Figure 1 As shown in the figure, the image processing model used in the present invention includes a denoising branch and an edge enhancement branch. The denoising branch is the main branch, and the edge enhancement branch is the auxiliary branch. The outputs of the encoder and decoder modules of the edge enhancement branch are fused with the outputs of the encoder and decoder modules of the denoising branch through a feature fusion module to improve the denoising branch's shortcomings in edge extraction.

[0038] The denoising branch is built based on the U-Net convolutional neural network architecture, which includes the SC block module, the SConv module (strided convolution), and the TConv module (transposed convolution). Figure 1As shown in the figure, the first encoder module is composed of a repeated SC block module and an SConv module, and the first decoder module is composed of a repeated SC block module and a TConv module, and the SConv module and the TConv module are connected via a jump connection. SC block is a new type of neural network module that enhances the feature extraction capability by combining Swin Transformer (SwinT) and residual convolution (RConv). The design of this module aims to effectively fuse the features of SwinT block and RConvblock through convolution, Split function, Concat function and residual connection (Residual Connection) to improve the performance of the network when processing images. Specifically, Figure 2 The module diagram of the SC block module in the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention is schematically shown. Figure 3 The schematic diagram shows the process of the SC block module processing the input feature map, refer to Figure 2 and Figure 3 As shown, the process can be specifically implemented as follows:

[0039] Step S11: perform 1×1 convolution on the input feature map and split it into the first feature map and the second feature map through the Split function;

[0040] Step S12: Input the first feature map into the SwinT block to obtain the third feature map, and input the second feature map into the RConv block to obtain the fourth feature map;

[0041] Step S13: The third feature map and the fourth feature map are combined by the Concat function to obtain a fifth feature map, and the number of channels of the fifth feature map is adjusted by a 1×1 convolution layer to match the acquired image;

[0042] Step S14: Perform a residual connection on the fifth feature map with adjusted channel number and the input feature map to obtain an output feature map.

[0043] Specifically, when executing step S11, first input feature map X Perform 1×1 convolution and then divide it into two parts through the Split function, which are respectively X The corresponding first and second feature maps are expressed as follows:

[0044]

[0045] Among them, the obtained X 1 is the first feature map obtained in step S11,X 2 is the second feature map obtained in step S11.

[0046] Then, step S12 is executed to obtain the first feature map X 1 and the second figure X 2 are input to SwinTblock and RConv block respectively, and the first feature map is processed by SwinT block X 1 is processed to obtain the third feature map, and the second feature map is processed by RConv block. X 2 is processed to obtain the fourth feature map. Its expression is as follows:

[0047]

[0048] Among them, the obtained Y 1 is the third feature map obtained in step S12, Y 2 is the fourth feature map obtained in step S12.

[0049] Then, step S13 and step S14 are executed to obtain the third feature map. Y 1 and the fourth characteristic map Y 2 are merged to obtain the fifth feature map, and the number of channels of the fifth feature map is adjusted through a 1×1 convolution layer. Finally, a residual connection is performed to obtain the final SCblock output, that is, the output feature map. Its expression is as follows:

[0050]

[0051] The edge enhancement branch is built based on the U-Net convolutional neural network architecture, which includes the Sobel convolution module, the RConvblock module (residual convolution), the SConv module (strided convolution), and the TConv module (transposed convolution). Figure 1As shown, the edge information extraction module is set to a pre-trained Sobel convolution module, the second encoder module is composed of repeatedly set RConv block modules and SConv modules, and the second decoder module is composed of repeatedly set RConv block modules and TConv modules, and the SConv module and the TConv module are connected by a jump connection. The Sobel convolution adopted can adaptively adjust the value of the Sobel factor α during the network model training process, so that the network model can dynamically adjust the sensitivity of edge detection for different training data and tasks during the training process, thereby being able to extract edge information of different intensities. In the process of edge enhancement processing of the acquired image using the Sobel convolution module, the acquired image is convolved with the Sobel operator in the Sobel convolution module to obtain an edge information feature map. Specifically, in order to retain the edge features of the processed image as much as possible, in some possible implementations, four trainable Sobel operators including horizontal, vertical and diagonal directions (two diagonal directions) can be selected to convolve the acquired image. As shown Figure 4 As shown, Figure 4 The first one on the left is a trainable Sobel operator for the horizontal direction, the second one on the left is a trainable Sobel operator for the vertical direction, and the two on the right are trainable Sobel operators for different diagonal directions. The input image is passed through the edge information map module to obtain a four-channel edge information feature map corresponding to the four Sobel operators mentioned above.

[0052] The feature fusion module includes the Cross-Attention Fusion Module (CAFM), refer to Figure 1 As shown, the cross attention fusion module fuses the outputs of the first-level SConv modules in the first encoder module and the second encoder module. Specifically, as the two input features Fc1 and Fc2 of the cross attention fusion module, the shapes of both are C × H × W ,in, C is the number of channels of the feature map, H is the height of the feature map, W is the width of the feature map. Specifically, Figure 5 The module diagram of the cross attention fusion module in the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention is schematically shown. Figure 6 The schematic diagram shows the process of processing the input features of the cross attention fusion module, referring to Figure 5 and Figure 6 As shown, the process can be specifically implemented as follows:

[0053] Step S21: Processing the first input feature and the second input feature according to the channel attention mechanism of the cross attention fusion module to obtain a first channel weight coefficient corresponding to the first input feature and a second channel weight coefficient corresponding to the second input feature respectively;

[0054] Step S22: obtaining a cross matrix according to the first channel weight coefficient and the second channel weight coefficient;

[0055] Step S23: obtaining a first feature corresponding to the first input feature and a second feature corresponding to the second input feature according to the cross matrix, the first input feature, and the second input feature;

[0056] Step S24: Processing the first feature and the second feature according to the spatial attention mechanism of the cross attention fusion module to obtain a first spatial weight coefficient corresponding to the first feature and a second spatial weight coefficient corresponding to the second feature respectively;

[0057] Step S25: obtaining a first fused feature corresponding to the first input feature and a second fused feature corresponding to the second input feature according to the first spatial weight coefficient, the second spatial weight coefficient, the first input feature, and the second input feature;

[0058] Step S26: Output the first fusion feature and the second fusion feature through the fully connected layer to obtain a fusion feature.

[0059] Specifically, when executing step S21, the channel attention mechanism of the cross attention fusion module is used to respectively perform the first input feature Fc 1 and the second input feature Fc 2 is processed to obtain the first input feature Fc The first channel weight coefficient corresponding to 1 Mc 1, and get the second input feature Fc 2 corresponds to the second channel weight coefficient Mc 2.

[0060] Then, step S22 is executed to set the first channel weight coefficient Mc 1 and the second channel weight coefficient Mc 2 multiply to get the cross matrix M cross The resulting cross matrix M cross The shape is C ×1×1.

[0061] Then execute step S23, according to the cross matrix M cross, the first input feature Fc 1 and the second input feature Fc 2. Get the first input feature Fc 1 corresponds to the first feature Fs 1 and the second input feature Fc 2 corresponds to the second feature Fs 2. Specifically, its expression is as follows:

[0062]

[0063] Get the first feature Fs 1 and second characteristics Fs 2, step S24 is executed to respectively perform the first feature Fs 1 and second characteristics Fs 2 is processed to obtain the first input feature Fc The first spatial weight coefficient corresponding to 1 Ms 1, and get the second input feature Fc The second spatial weight coefficient corresponding to 2 Ms 2.

[0064] Then execute step S25, the first spatial weight coefficient Ms 1 and the first input feature Fc 1 and processed to get the same as the first input feature Fc 2 corresponds to the first fusion feature Fs out1 , the second spatial weight coefficient Ms 2 and the second input feature Fc 2 and processed to get the same as the second input feature Fc 2 corresponds to the second fusion feature Fs out2 , add the residual connection to the fusion module. Specifically, its expression is as follows:

[0065]

[0066] Finally, step S26 is executed to obtain the first fusion feature Fs out1 and the second fusion feature Fs out2 Output through the fully connected layer and finally get the optimized image Y Specifically, its expression is as follows:

[0067]

[0068] in, W is the weight of the output fully connected layer, Bis the bias term of the output fully connected layer, W and B The value of is first obtained by random initialization, and then back-propagation B and W Update to confirm.

[0069] The image processing model of the present invention is trained using a dataset of raw SEM images generated using the Thorsos method and the ARTIMAGEN library. This dataset includes not only the raw SEM images but also noise images corresponding to these images, obtained by noise-simulating the raw SEM images. Furthermore, when training these model modules, they can be trained based on both the raw SEM images and the corresponding noise images, making each model module more suitable for denoising and optimizing SEM images, thereby improving the accuracy of critical dimension scanning electron microscopy (CD-SEM) metrology. For example, when generating raw SEM images using the Thorsos method and the ARTIMAGEN library, 10,800 raw SEM images can be generated. When preprocessing these original SEM images, the dataset can be divided into three parts: training set, test set and validation set. The training set is used to train the image processing model, the validation set is used for tuning during the training process, guiding the adjustment of model hyperparameters and preventing overfitting, and the test set is used to finally evaluate the performance of the model after the model training and tuning are completed, and to test the performance of the model on unseen data. The training set, validation set and test set account for 70%, 15% and 15% of the entire dataset respectively.

[0070] Figure 7 The following schematically illustrates the steps of performing noise simulation on the original SEM image in the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention. Figure 7 As shown, this step process can be specifically implemented as including the following steps:

[0071] Step S31: cropping the original SEM image into non-overlapping images of a preset pixel size;

[0072] Step S32: calculating the mean of the cropped original SEM image;

[0073] Step S33: adjusting the contrast of the original image according to the calculated mean value to generate an image with reduced contrast;

[0074] Step S34: Generate Gaussian noise with a mean of 0 and a standard deviation of σ;

[0075] Step S35: Add the generated Gaussian noise to the contrast-reduced image to obtain a noisy image.

[0076] Since the size of the partially generated original SEM image is 64×1024, when performing step S31, the 64×1024 SEM image is first cropped into non-overlapping SEM images of size 64×64, and then steps S32 to S35 are further performed to obtain a noise image.

[0077] Step S32, calculate the mean of the original SEM image. Taking a single original SEM image as an example, its mean μ The calculation formula is as follows:

[0078]

[0079] in, H is the height of the original SEM image, W is the width of the original SEM image, I(i , j ) is the pixel value of the original SEM image, i and j are the horizontal and vertical coordinate values ​​of the pixel respectively.

[0080] Calculate the mean of the SEM image μ After that, step S33 is continued to be executed to adjust the contrast of the original SEM image to generate an image with reduced contrast. Specifically, when reducing the contrast of the original SEM image, the pixel value of the SEM image can be adjusted. I ( i , j ), applying a contrast factor β To calculate the new pixel value I’ ( i , j The specific formula is as follows:

[0081]

[0082] in, β is the contrast factor, and its value range is [0,1]. Preferably, the contrast factor β The value is 0.35, and the visual effect displayed at this time can better simulate the low-dose SEM image.

[0083] Then, step S34 is executed to generate Gaussian noise. The generated Gaussian noise has a mean of 0 and a standard deviation of σ Gaussian noise with standard deviation σThe value of can be set to [40,70], which can better simulate the low-dose SEM image. The specific distribution expression of the generated Gaussian noise is as follows:

[0084]

[0085] Where N represents Gaussian distribution.

[0086] Finally, step S35 is performed to add the generated Gaussian noise to the contrast-reduced image (pixel values ​​are I’ ) to obtain a simulated noise image (pixel value is I’’ The specific formula is as follows:

[0087]

[0088] After that, it is only necessary to generate a corresponding noise image for each SEM image to form a data set for training the edge enhancement module, denoising module and cross-attention fusion module.

[0089] Figure 8 The following diagram schematically illustrates the comparison between the processing results of a critical dimension scanning electron microscope image processing method according to an embodiment of the present invention and the processing results of an existing conventional processing method. Specifically, the operating system used for the comparison is Linux, the CPU model is Core i9-9900k, and the GPU model is NVIDIA GeForce RTX3090. During the training process, the contrast factor β Set to 0.35, standard deviation σ The value is set to [40, 70], the MSE loss function is used as the loss function, the Adam optimizer is selected for gradient descent, the epoch is set to 150, the initial learning rate is set to 0.0001, and the learning rate is reduced to 0.00001 after 130 epochs. The batch size is 64, and the model network is trained based on the Pytorch framework.

[0090] Figure 8 The PSNR shown in the figure is the peak signal-to-noise ratio, and SSIM is the structural similarity. PSNR is used to indicate the quality of the reconstructed image. The unit of PSNR is decibel. The larger the PSNR value, the smaller the distortion of the reconstructed image. Generally, when PSNR>30dB, the reconstructed image is considered to be of high quality. Its calculation formula is as follows:

[0091]

[0092] Among them, MAX represents the maximum possible value of the pixel value in the image, and MSE represents the size H ×W The mean square error between the original SEM image and the optimized image is expressed as follows:

[0093]

[0094] in, I ( i , j ) is the pixel value of the original image, K ( i , j ) is the pixel value of the denoised image.

[0095] SSIM measures the similarity between two images by comparing brightness, contrast, and structure. Its value range is [-1, 1]. The closer it is to 1, the more similar the two images are. It can be expressed as follows:

[0096]

[0097] in, x and y is the pixel value of the original SEM image and the optimized image, μ x is an image x The average value of μ y is an image y The average value of σ x 2 is an image x The covariance of σ y 2 is an image y The covariance of σ xy is an image x and y The covariance of , represents the structural information. c 1 and c 2 are two constants introduced to avoid the denominator being zero, usually c 1=( k 1 l ) 2 and c 2=( k 2 l ) 2 ,in l is the dynamic range of pixel values, k 1 and k 2 are two very small constants, such as k 1=0.01, k 2=0.03.

[0098] Figure 8 The processing result comparison includes 7 existing denoising algorithms: BM3D, WNNM, DnCNN, FFDNet, ADNet, SwinIR and MWDCNN, and the contrast factor β is 0.35, the noise level (standard deviation σ ) are 40, 50, and 60 respectively, and these algorithms are quantitatively compared. Figure 7 The peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) of the algorithms in the paper show that the traditional denoising methods BM3D and WNNM have significantly lower PSNR and SSIM than the deep learning methods at all noise levels, demonstrating the significant advantages of deep learning methods in processing high-noise images. Compared with the other five deep learning methods, the proposed method exhibits the best denoising effect at all noise levels. Specifically, when the noise level is 40, the proposed method improves the PSNR by 0.5186 dB compared to MWDCNN and the SSIM by 0.0012 compared to SwinIR. When the noise level is 50, the proposed method improves the PSNR by 0.4224 dB compared to MWDCNN and the SSIM by 0.0006 compared to SwinIR. When the noise level is 60, the proposed method improves the PSNR by 0.327 dB compared to MWDCNN and the SSIM by 0.0002 compared to SwinIR. In summary, the method proposed in this paper can achieve higher PSNR and SSIM when processing images with different noise levels, verifying its effectiveness in the denoising task of low-dose CD-SEM images with low contrast and high noise.

[0099] Figure 9 The following schematically illustrates the comparison between the visual effects of the processing results of the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention and the visual effects of the processing results of the conventional processing method. Figure 9 (a) is the original image, and (b) is the noise level (standard deviation σ ) is 30 and the contrast factor β Figures (c) and (d) show the denoising effects of the BM3D and WNNM algorithms, respectively. It can be observed that these two traditional algorithms cause severe image edge blurring and detail loss during the denoising process, while also reducing the overall image contrast.

[0100] Figure 9 Figure (e) shows the visual effect of the processing results of the feedforward denoising convolutional neural network (DnCNN). Figure 9Figure (f) shows the visual effect of the processing results of the fast and flexible denoising network (FFDNet). Figure 9 Figure (g) shows the visual effect of the processing results of the attention-guided denoising network (ADNet). Figure 9 Figure (h) shows the visual effect of the processing results of the SwinIR network. Figure 9 Figure (i) shows the visual effect of the processing results of the multi-stage image denoising convolutional neural network (MWDCNN). Figure 9 Figure (j) shows the processing results of the method of the present invention. It can be seen that the method of the present invention effectively removes noise while better preserving image edge details and maintaining high image contrast. Compared with other denoising methods, the method of the present invention demonstrates significant advantages in edge restoration accuracy, which is particularly important for accurately obtaining edge information in CD-SEM images.

[0101] To verify the effectiveness of the proposed denoising method for improving metrology in low-dose CD-SEM images, the open-source software SMILE was used to obtain image critical dimension (CD), line edge roughness (LER), and line width roughness (LWR). Comparison of CD, LER, and LWR values ​​obtained from different methods confirmed the effectiveness of the proposed method in improving CD-SEM metrology accuracy. CD typically refers to the minimum line or space width in a lithographic feature pattern. It is a crucial parameter in semiconductor manufacturing because it directly impacts device performance and functionality. LER refers to the random, minute variations or irregularities at the edge of a feature pattern. LER is typically calculated by statistically analyzing the deviation of the edge position from its ideal position. It is caused by non-uniformity in the lithography and etching processes and is a significant factor affecting device performance. LWR can represent the variation or irregularity of the line width of a feature pattern. It is closely related to LER because the variation in line width is usually caused by the variation of the edges on both sides. LWR is quantified by measuring the width of multiple locations on a line and calculating the standard deviation or root mean square value of these widths.

[0102] Figures 10 to 12 The results of CD, LWR and LER values ​​of three randomly selected CD-SEM images processed by the critical dimension scanning electron microscope image processing method according to one embodiment of the present invention and the existing traditional processing method are schematically shown. It should be pointed out that in this study, the original pixel setting of N. Chaudhary et al. is abandoned, and the pixel size is set to 1nm per pixel, which is conducive to data statistics when using SMILE. Figure 10 、 Figure 11 and Figure 12 The results clearly show that noise has a significant impact on the metrology of CD-SEM images. Compared with deep learning methods, traditional methods show obvious disadvantages in accurate metrology of low-dose CD-SEM images. Figure 13 Schematically shows Figures 10 to 12 The line chart comparing the CD, LWR and LER values ​​under each treatment method. Figure 13 It can be clearly seen that the image restored using the DENet method is more consistently close to the original image data. This demonstrates that the proposed method is effective in obtaining low-dose CD-SEM images, thus providing a new solution to the photoresist shrinkage problem.

[0103] The critical dimension scanning electron microscope (CD-SEM) image processing method of this invention transmits the input image in parallel to an edge enhancement module and a denoising module. The denoising module, which includes an SC block, enhances the ability to extract global and local features of the input image. The edge enhancement branch first acquires edge information from the input image through Sobel convolution and then extracts global and local features of this edge information. A feature fusion module also fuses the features of the encoder and decoder modules of the two branches to compensate for the shortcomings of the denoising branch in edge feature extraction. Compared with traditional image denoising algorithms, this algorithm not only removes noise but also solves the edge blurring problem caused by over-smoothing of the image. Experimental results demonstrate that the algorithm achieves good denoising results in test images and improves the metrology accuracy of critical dimension scanning electron microscope (CD-SEM). Furthermore, the proposed CD-SEM image processing method provides an effective solution to the challenges posed by photoresist shrinkage and is of great significance for promoting the development of high numerical aperture extreme ultraviolet (EUV) lithography.

[0104] In some embodiments, an embodiment of the present invention provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the critical dimension scanning electron microscope image processing method of any of the above embodiments of the present invention.

[0105] In some embodiments, an embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the critical dimension scanning electron microscope image processing method of any one of the above embodiments.

[0106] In some embodiments, an embodiment of the present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the critical dimension scanning electron microscope image processing method of any of the above embodiments.

[0107] In some embodiments, an embodiment of the present invention further provides a storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the critical dimension scanning electron microscope image processing method of any of the above embodiments.

[0108] Figure 14 FIG. 1 is a schematic diagram of the hardware structure of an electronic device for executing a critical dimension scanning electron microscope image processing method according to another embodiment of the present invention. Figure 14 As shown, the device includes:

[0109] One or more processors 110 and memory 120, Figure 14 A processor 110 is taken as an example.

[0110] The apparatus for performing the critical dimension scanning electron microscope image processing method may further include: an input device 130 and an output device 140 .

[0111] The processor 110, the memory 120, the input device 130 and the output device 140 may be connected via a bus or other means. Figure 14 The bus connection is taken as an example.

[0112] Memory 120, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the critical dimension scanning electron microscope image processing method according to the embodiments of the present invention. Processor 110 executes the non-volatile software programs, instructions, and modules stored in memory 120 to execute various server functional applications and data processing, thereby implementing the critical dimension scanning electron microscope image processing method according to the aforementioned method embodiment.

[0113] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the critical dimension scanning electron microscope image processing method, etc. In addition, the memory 120 may include a high-speed random access memory and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely located relative to the processor 110, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0114] The input device 130 may receive input digital or character information and generate signals related to user settings and function control of the image processing device. The output device 140 may include a display device such as a display screen.

[0115] The one or more modules are stored in the memory 120 , and when executed by the one or more processors 110 , perform the critical dimension scanning electron microscope image processing method in any of the above method embodiments.

[0116] The above-mentioned product can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0117] The electronic devices of the embodiments of the present invention exist in various forms, including but not limited to:

[0118] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and are primarily designed to provide voice and data communications. These terminals include smartphones (e.g., iPhones), multimedia phones, feature phones, and low-end phones.

[0119] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.

[0120] (3) Portable entertainment devices: These devices can display and play multimedia content. These devices include audio and video players (such as iPods), handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0121] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0122] (5) Other electronic devices with data interaction functions.

[0123] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0124] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a general hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the relevant technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A critical dimension scanning electron microscope image processing method, characterized in that: include: Inputting an input image into a pre-trained image processing model, processing the input image using the pre-trained image processing model, and obtaining an output result of the image processing model to obtain an optimized image; Wherein, the image processing model includes: A denoising branch, based on a U-Net convolutional neural network architecture, comprises a first encoder module and a first decoder module connected in sequence, wherein both the first encoder module and the first decoder module employ SC block modules to extract local and global features of the image, and the first encoder module is composed of alternating SC block modules and SConv modules; The edge enhancement branch is based on the U-Net convolutional neural network architecture and includes an edge information extraction module, a second encoder module, and a second decoder module connected in sequence. The second encoder module is composed of an alternating RConv block module and an SConv module. A feature fusion module is used to fuse the outputs of the SConv modules at each level of the first encoder module and the second encoder module as the input of the SC block modules at each level of the first encoder module. The output of the SC block module is used as the input of the next level SConv module of the first encoder module. The cross-attention fusion module is used to fuse the outputs of the first level SConv modules in the first encoder module and the second encoder module.

2. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The edge information extraction module includes a pre-trained Sobel convolution module, and the Sobel operators used in the Sobel convolution module include a horizontal Sobel operator, a vertical Sobel operator and two diagonal Sobel operators.

3. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The SCblock settings are: Perform 1×1 convolution on the input feature map and split it into the first feature map and the second feature map through the Split function; The first feature map is input into the SwinT block to obtain the third feature map, and the second feature map is input into the RConv block to obtain the fourth feature map; The third and fourth feature maps are combined through the Concat function to obtain the fifth feature map, and the number of channels of the fifth feature map is adjusted through a 1×1 convolution layer to match the acquired image; The fifth feature map with adjusted channel number is residually connected to the input feature map to obtain the output feature map.

4. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The cross-attention fusion module is used to fuse the first-level outputs of the first encoder module and the second encoder module, comprising: Processing the first input feature and the second input feature according to the channel attention mechanism of the cross attention fusion module to obtain a first channel weight coefficient corresponding to the first input feature and a second channel weight coefficient corresponding to the second input feature respectively; Obtain a cross matrix according to the first channel weight coefficient and the second channel weight coefficient; Obtaining, according to the cross matrix, the first input feature, and the second input feature, a first feature corresponding to the first input feature and a second feature corresponding to the second input feature; Processing the first feature and the second feature according to the spatial attention mechanism of the cross attention fusion module to obtain a first spatial weight coefficient corresponding to the first feature and a second spatial weight coefficient corresponding to the second feature respectively; Obtaining a first fused feature corresponding to the first input feature and a second fused feature corresponding to the second input feature according to the first spatial weight coefficient, the second spatial weight coefficient, the first input feature, and the second input feature; The first fusion feature and the second fusion feature are output through the fully connected layer to obtain the fusion feature.

5. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The first decoder module is composed of SC block modules and TConv modules that are alternately arranged. The SConv modules at each stage in the first encoder module are connected to the TConv modules at each stage in the first decoder using skip connections.

6. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The second decoder module is composed of RConv block modules and TConv modules that are alternately arranged, and the SConv modules at each stage in the second encoder module are connected to the TConv modules at each stage in the second decoder using skip connections.

7. The critical dimension scanning electron microscope image processing method according to claim 1, characterized in that: The data set used in the training of the image processing model includes the original SEM image and the noise image corresponding to the original SEM image after the original SEM image is subjected to noise simulation processing.

8. The critical dimension scanning electron microscope image processing method according to claim 7, characterized in that: The noise simulation process includes: The original SEM image is cropped into non-overlapping images of a preset pixel size; Calculate the mean of the cropped original SEM image; Adjust the contrast of the original image according to the calculated mean to generate an image with reduced contrast; Generate Gaussian noise with mean 0 and standard deviation σ; The generated Gaussian noise is added to the contrast-reduced image to obtain a noisy image.

9. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the critical dimension scanning electron microscope image processing method according to any one of claims 1 to 8.

Images