Image noise reduction method

Abstract

To provide an image noise reduction method capable of reducing noise with high accuracy from an image generated by a scanning electro microscope.SOLUTION: The present image noise reduction method generates a reference image of a sample by a scanning electro microscope, generates an artificial noise image by applying an artificial noise to the reference image, generates an artificial low quality image by adopting an aerotropic filter constructed so as to expand the artificial noise to a scanning direction of an electron beam of the scanning electro microscope to the artificial noise image, creates a denoised model by executing mechanical learning by using training data containing the reference image and the artificial low quality image, generates an image of a workpiece as an object of inspection and formation measurement of a semiconductor device by the scanning electro microscope, inputs the image of the workpiece to the denoised model, and outputs a denoised image from the denoised model.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to an image noise reduction method for reducing noise on an image generated by a scanning electron microscope.

Background Art

[0002] Noise may appear in an image generated by a scanning electron microscope due to various factors such as optical or electrical factors. As one solution for reducing such noise on an image, a denoising model may be used. More specifically, an image containing noise is input into the denoising model, and a denoised image with reduced noise is output from the denoising model. Such a denoising model is a trained model created by machine learning.

[0003] FIG. 4 is a schematic diagram showing an example of a conventional method for creating a denoising model by machine learning. Machine learning is executed using training data including a large number of pairs of high-quality images and low-quality images generated by a scanning electron microscope. Each pair of high-quality image and low-quality image is an image of the same location of a sample such as a wafer. The high-quality image is obtained under conditions of a low scan rate, and the low-quality image is obtained under conditions of a high scan rate.

[0004] However, this method requires generating images of the same location of the sample under different conditions (i.e., different scan rates). Such an imaging method is very time-consuming for a scanning electron microscope.

[0005] Figure 5 is a schematic diagram illustrating another example of a conventional method for creating a denoising model using machine learning. The machine learning is performed using training data consisting of numerous pairs of high-resolution and artificially noisy images generated by a scanning electron microscope. High-resolution images are obtained under low scan rate conditions, similar to the method shown in Figure 4. Artificially noisy images are obtained by adding artificial noise (e.g., Gaussian noise) to the obtained high-resolution images. This method allows the scanning electron microscope to generate images under identical conditions (same scan rate).

[0006] However, artificial noise such as Gaussian noise does not resemble the actual noise on images generated by scanning electron microscopes, and as a result, denoising models created by machine learning were sometimes unable to sufficiently reduce the noise. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2021-197144 [Overview of the project] [Problems that the invention aims to solve]

[0008] Therefore, the present invention provides an image noise reduction method that can accurately reduce noise from images generated by a scanning electron microscope. [Means for solving the problem]

[0009] In one embodiment, an image noise reduction method for reducing noise from an image generated by a scanning electron microscope is provided, comprising: generating a reference image of a sample using the scanning electron microscope; generating an artificial noise image by adding artificial noise to the reference image; generating an artificial low-resolution image by applying an anisotropic filter configured to stretch the artificial noise in the scanning direction of the electron beam of the scanning electron microscope to the artificial noise image; and performing machine learning using training data including the reference image and the artificial low-resolution image. , A method for reducing image noise is provided, which involves creating a noise model, generating an image of a workpiece to be inspected and its shape measured using a scanning electron microscope, inputting the image of the workpiece into the denoising model, and outputting a denoised image from the denoising model.

[0010] In one embodiment, the anisotropic filter has parameters of at least the scan rate and scan direction of the scanning electron beam, and the response characteristics of the electron detector of the scanning electron microscope, and these parameters are set to match the parameters used when capturing an image of the workpiece to be denoised. In one embodiment, the anisotropic filter has parameters of at least the material and cross-sectional structure of the sample surface and the acceleration voltage and current of the irradiating electron beam, and these parameters are set to match the parameters used when capturing an image of the workpiece to be denoised. In one embodiment, the artificial noise is noise that follows a statistical distribution. In one embodiment, the artificial noise is Poisson noise. In one embodiment, the artificial noise is noise that follows a normal distribution or a log-normal distribution. [Effects of the Invention]

[0011] According to the present invention, an anisotropic filter that mimics noise specific to scanning electron microscopes is applied to an artificially noisy image, making it possible to create an artificially low-resolution image that resembles an actual image containing noise. By performing machine learning using training data that includes such an artificially low-resolution image and a reference image (high-resolution image), a denoising model with high noise reduction accuracy can be created. [Brief explanation of the drawing]

[0012] [Figure 1] This is a schematic diagram showing one embodiment of an image generation system. [Figure 2] This figure illustrates one embodiment of a method for creating a denoising model using machine learning. [Figure 3] This figure shows examples of a reference image, an artificially noised image, and an artificially low-resolution image. [Figure 4] This diagram illustrates an example of a conventional method for creating a denoising model using machine learning. [Figure 5] This figure illustrates another example of a conventional method for creating a denoising model using machine learning. [Modes for carrying out the invention]

[0013] Embodiments of the present invention will be described below with reference to the drawings. Figure 1 is a schematic diagram showing one embodiment of the image generation system. The image generation system comprises a scanning electron microscope 1 that generates an image of a workpiece W, and a processing system 5 that processes the image generated by the scanning electron microscope 1. Examples of workpiece W include wafers, masks, panels, substrates, etc., which are the targets of inspection and shape measurement of semiconductor devices.

[0014] The processing system 5 consists of at least one computer. The processing system 5 includes a storage device 5a in which a program is stored, and an arithmetic unit 5b that performs calculations according to the instructions contained in the program. The storage device 5a includes a main memory such as random access memory (RAM) and an auxiliary storage device such as a hard disk drive (HDD) or solid state drive (SSD). Examples of arithmetic units 5b include a CPU (central processing unit) and a GPU (graphics processing unit). However, the specific configuration of the processing system 5 is not limited to this embodiment.

[0015] The processing system 5 may be an edge server connected to the scanning electron microscope 1 via a communication line, a cloud server connected to the scanning electron microscope 1 via a communication network such as the internet or a local network, or a fog computing device (gateway, fog server, router, etc.) installed within the network connected to the scanning electron microscope 1. The processing system 5 may be a combination of multiple servers. For example, the processing system 5 may be a combination of edge servers and cloud servers connected to each other via a communication network such as the internet or a local network. In another example, the processing system 5 may consist of multiple servers (computers) that are not connected by a network.

[0016] The scanning electron microscope 1 includes an electron gun 15 that emits an electron beam, a focusing lens 16 that focuses the electron beam emitted from the electron gun 15, an X-deflector 17 that deflects the electron beam in the X direction, a Y-deflector 18 that deflects the electron beam in the Y direction, an objective lens 20 that focuses the electron beam onto a workpiece W, which is an example of a sample, a workpiece stage 31 that supports the workpiece W, and a stage moving device 35 that moves the workpiece stage 31 in translation. The configuration of the electron gun 15 is not particularly limited. For example, a field emitter type electron gun or a semiconductor photocathode type electron gun can be used as the electron gun 15.

[0017] The electron beam emitted from the electron gun 15 is focused by the focusing lens 16, and then deflected by the X deflector 17 and the Y deflector 18, and is focused by the objective lens 20 and irradiated onto the surface of the workpiece W. When the primary electrons of the electron beam irradiate the workpiece W, electrons such as secondary electrons and reflected electrons are emitted from the workpiece W. The electrons emitted from the workpiece W are detected by the electron detector (scintillator) 26. The electron detection signal of the electron detector 26 is input to the image acquisition device 28 and converted into an image. In this way, the scanning electron microscope 1 generates an image of the surface of the workpiece W. The image acquisition device 28 is connected to the processing system 5.

[0018] The processing system 5 has a program in the storage device 5a for creating a denoising model that reduces noise from the image generated by the scanning electron microscope 1. Hereinafter, an embodiment of a method for creating a denoising model by machine learning will be described with reference to FIG. 2.

[0019] The processing system 5 issues a command to the scanning electron microscope 1 to generate a reference image of the sample. The sample may have the same structure as the workpiece W shown in FIG. 1 or may have a different structure. Examples of the sample include a wafer, a mask, a panel, a substrate, etc. that are targets for inspection and shape measurement of semiconductor devices. The reference image is an image used as the correct label for machine learning and is a high-quality image. More specifically, the scanning electron microscope 1 can generate a reference image, which is a high-quality image, by scanning the surface of the sample with an electron beam at a low scan rate.

[0020] The processing system 5 acquires the reference image from the scanning electron microscope 1. Further, the processing system 5 adds artificial noise to the reference image to generate an artificial noise image. The artificial noise is noise that is statistically independent between the pixels of the reference image. In other words, the artificial noise is noise that has no correlation between the pixels of the reference image. More specifically, the artificial noise is noise that follows a statistical distribution, and specific examples thereof include Poisson noise, Gaussian noise, and noise that follows a lognormal distribution.

[0021] In this embodiment, Poisson noise is used as the artificial noise. This is because Poisson noise is similar to the noise that appears on images generated by the scanning electron microscope 1. Therefore, the denoising model constructed by machine learning, as described later, is expected to be able to accurately reduce the noise on images generated by the scanning electron microscope 1.

[0022] The processing system 5 generates an artificial low-resolution image by applying an anisotropic filter to the artificial noise image generated as described above. The anisotropic filter is configured to make the artificial noise on the artificial noise image more similar to the actual noise on the image generated by the scanning electron microscope 1. The actual noise on the image is affected by the scanning operation of the electron beam, the afterglow caused by the electron detector (scintillator) 26, and the charge of the object being imaged. The anisotropic filter is a filter designed to mimic such noise specific to the scanning electron microscope 1.

[0023] An anisotropic filter is configured to stretch artificial noise in the scanning direction of the electron beam of the scanning electron microscope 1. In one embodiment, the anisotropic filter is configured by the following equation:

number

[0024] In one embodiment, the anisotropic filter has parameters such as the scan rate and scan direction of the electron beam being scanned, and the response characteristics of the electron detector 26, and these parameters are set to match the parameters used when capturing the image to be denoised.

[0025] In one embodiment, the anisotropic filter has parameters such as the material and cross-sectional structure of the sample surface, and the acceleration voltage and current of the irradiating electron beam, and these parameters are set to match the parameters used when capturing the image to be denoised.

[0026] Figure 3 shows a reference image, an artificial noise image created by adding artificial noise to the reference image, and an example of an artificial low-resolution image created by applying an anisotropic filter to the artificial noise image. The artificial low-resolution image is close to the actual low-resolution image generated by the scanning electron microscope 1.

[0027] Anisotropy in the real image (SEM image) generated by the scanning electron microscope 1 can also be caused by the charging of the sample. In this case, instead of equation (1) above, a transfer function that depends on the material of the sample surface, the cross-sectional structure of the sample, the acceleration voltage of the electron beam, the current, and the scan rate of the electron beam may be used.

[0028] The artificially low-resolution image generated by applying an anisotropic filter to the artificially noisy image, and the reference image generated by the scanning electron microscope 1, are used as training data for machine learning. That is, the artificially low-resolution image is used as an explanatory variable, and the reference image is used as the target variable (ground truth label). The processing system 5 performs machine learning of the denoising model so that the difference between the image output from the denoising model and the reference image is minimized when the artificially low-resolution image is input to the denoising model.

[0029] Examples of machine learning methods include SVR (Support Vector Regression), PLS (Partial Least Squares), deep learning, random forests, and decision trees. For example, a denoising model consists of a neural network constructed using deep learning.

[0030] The denoising model created by machine learning using training data including reference images and artificially low-resolution images is stored as a trained model in the storage device 5a of the processing system 5.

[0031] The denoising model created in this way can accurately reduce the noise on the image generated by the scanning electron microscope 1. Specifically, the scanning electron microscope 1 generates an image of the workpiece W that is the target of inspection and shape measurement of semiconductor devices, the processing system 5 acquires an image (SEM image) of the workpiece W from the scanning electron microscope 1, inputs the image of the workpiece W into the denoising model, and outputs a denoised image, which is an image with reduced noise, from the denoising model.

[0032] According to this embodiment, an anisotropic filter that mimics the noise specific to the scanning electron microscope 1 is applied to the artificial noise image, making it possible to create an artificial low-resolution image that resembles an actual image containing noise. By performing machine learning using training data that includes such an artificial low-resolution image and a reference image (high-resolution image), a highly accurate denoising model can be created.

[0033] This section describes an example of applying the effects of the denoising process described above to the measurement of semiconductor devices. Here, we will focus on a simple stripe-like pattern, such as a series of linear patterns, and examine a case where the edge roughness of each line is measured.

[0034] The scanning electron microscope 1 generates multiple SEM images scanned at a scan rate of 3 MHz. The processing system 5 applies an anisotropic filter represented by the above formula (1) to these images and creates multiple artificially low-resolution images with artificial noise equivalent to a scan rate of 100 MHz.

[0035] Next, the processing system 5 uses the aforementioned pair of SEM images (reference images) with a scan rate of 3 MHz and artificially low-resolution images as training data to perform machine learning that generates SEM images (reference images) from artificially low-resolution images, thereby creating a denoising model.

[0036] The processing system 5 uses the obtained denoising model to denoise the SEM image actually acquired at a scan rate of 100 MHz. Using the edge roughness measured on a high-quality image at a scan rate of 3 MHz as a reference, the correlation with the edge roughness measured on the denoised image was approximately 0.8. This is equivalent to the result obtained at a scan rate of 25 MHz, indicating that this denoising process can increase the imaging speed by four times.

[0037] The embodiments described above are intended to enable persons with ordinary skill in the art to implement the present invention. Various modifications of the above embodiments can be made naturally by those skilled in the art, and the technical idea of ​​the present invention can be applied to other embodiments as well. Therefore, the present invention is not limited to the embodiments described, but is to be interpreted in the broadest sense according to the technical idea defined by the claims. [Explanation of symbols]

[0038] 1. Scanning electron microscope 5 Processing System 15. Electron gun 16 Focusing lens 17X deflector 18 Y deflector 20 Objective lenses 26 Electron detectors 28 Image acquisition device 31 Workpiece Stage 35 Stage Moving Device

Claims

1. An image noise reduction method for reducing noise from images generated by a scanning electron microscope, A reference image of the sample is generated using the aforementioned scanning electron microscope. Artificial noise is added to the aforementioned reference image to generate an artificial noise image. By applying an anisotropic filter configured to stretch the artificial noise in the scanning direction of the electron beam of the scanning electron microscope to the artificial noise image, an artificial low-resolution image is generated. A denoising model is created by performing machine learning using training data including the aforementioned reference image and the aforementioned artificial low-resolution image. Images of the workpiece to be inspected and its shape measured for semiconductor devices are generated by the scanning electron microscope. The image of the workpiece is input to the denoising model, An image noise reduction method that outputs a denoised image from the aforementioned denoising model.

2. The image noise reduction method according to claim 1, wherein the anisotropic filter has parameters of at least the scan rate and scan direction of the electron beam to be scanned, and the response characteristics of the electron detector of the scanning electron microscope, and the parameters are set to match the parameters when an image of the workpiece to be denoised is captured.

3. The image noise reduction method according to claim 1, wherein the anisotropic filter has parameters of at least the material and cross-sectional structure of the sample surface and the acceleration voltage and current of the irradiating electron beam, and the parameters are set to match the parameters when an image of the workpiece to be denoised is captured.

4. The image noise reduction method according to claim 1, wherein the artificial noise is noise that follows a statistical distribution.

5. The image noise reduction method according to claim 4, wherein the artificial noise is Poisson noise.

6. The image noise reduction method according to claim 4, wherein the artificial noise is noise that follows a normal distribution or a log-normal distribution.

Images