Charged particle beam device

Abstract

This invention provides a charged particle beam device that can suppress the increase of data volume and shorten the time required for scheme evaluation. The charged particle beam device comprises: a microscope that scans a sample with a charged particle beam, detects secondary particles emitted from the sample, and outputs a detection signal; and a computer system that generates frame images based on the detection signal and performs image processing based on the frame images. The computer system calculates moment images between multiple frame images and calculates feature data of the frame images based on the moment images.

Description

Technical Field

[0001] This disclosure relates to charged particle beam devices. Background Technology

[0002] Semiconductor devices are becoming increasingly large-scale and highly integrated. On the other hand, with the miniaturization of semiconductor devices, there is a growing demand for improved measurement and inspection accuracy in semiconductor inspection equipment that uses charged particle beams.

[0003] Scanning electron microscopes (SEMs), as one of the semiconductor inspection devices, irradiate samples such as semiconductor devices with an electron beam and generate images by detecting secondary electrons emitted from the sample.

[0004] In addition, other semiconductor inspection devices include, for example, CD-SEM (Critical Dimension SEM) and DR-SEM (Defect Review SEM). In CD-SEM, the dimensions of semiconductor circuit patterns are measured using a scanning electron microscope. In DR-SEM, various defects and foreign objects in semiconductor circuits are inspected and observed.

[0005] For example, Patent Document 1 discloses a method in which a low-magnification reference image is generated based on the average brightness value calculated from a low-magnification defect image, and the difference regions of these images are detected as defect regions. Patent Document 2 discloses a method in which an average brightness image and a brightness standard deviation image are obtained from multiple acceptable product reference images, and a reference image generated based on these images is used, thereby suppressing the detection of virtual defects without being affected by large variations in brightness values.

[0006] Patent document 3 describes a method for inspecting appearance images by calculating average images and standard deviation images. Patent document 4 describes a method for obtaining multiple images of the object to be inspected and updating the reference images of the inspection plan based on the average images and standard deviation images.

[0007] When introducing new semiconductor manufacturing processes, a scheme must be developed to define the processing sequence of the device in order to obtain images of target measurement patterns, foreign objects, and defect patterns. This scheme requires optimization of factors such as the accelerating voltage containing particles (e.g., electrons), probe current, field of view size (frame image size), frame accumulation, and pattern recognition parameters used to determine the position of the target measurement pattern.

[0008] However, in order to optimize the solution, it is necessary to repeatedly execute the solution under various shooting conditions to optimize the shooting conditions and parameters. Therefore, the evaluation of the solution takes a lot of time, and the amount of data involved in the evaluation of the solution also increases.

[0009] Patent Document 1: Japanese Patent Application Publication No. 2014-130026

[0010] Patent Document 2: Japanese Patent Application Publication No. 2005-274157

[0011] Patent Document 3: Japanese Patent Application Publication No. 2004-185259

[0012] Patent Document 4: Japanese Patent Application Publication No. 2011-174757 Summary of the Invention

[0013] Therefore, the purpose of this disclosure is to provide a charged particle beam device that can suppress the increase of data volume and shorten the time required for scheme evaluation.

[0014] The outline of a representative charged particle beam device disclosed in this application is described simply as follows. A representative embodiment of the charged particle beam device of the present invention includes: a microscope that scans a sample with a charged particle beam, detects secondary particles emitted from the sample, and outputs a detection signal; and a computer system that generates frame images based on the detection signal and performs image processing based on the frame images. The computer system calculates moment images between multiple frame images and calculates feature data of the frame images based on the moment images.

[0015] The effects obtained by the representative charged particle beam device disclosed in this application will be explained simply as follows.

[0016] That is, according to the representative embodiments of this disclosure, the increase in data volume can be suppressed and the time required for scheme evaluation can be shortened. Attached Figure Description

[0017] Figure 1 This is a structural diagram illustrating an example of a charged particle beam device.

[0018] Figure 2 Explain the method for generating the weighted average image.

[0019] Figure 3 This is a flowchart illustrating an example of a method for calculating characteristic data.

[0020] Figure 4 This schematically illustrates an example of the intensity distribution of a secondary electronic signal relative to a patterned cross section.

[0021] Figure 5 Explain the method for generating virtual images.

[0022] Figure 6 This is a flowchart illustrating an example of a method for generating virtual images.

[0023] Figure 7 This illustrates the characteristics of skewness and kurtosis.

[0024] Figure 8 Describe the situation when contraction occurs.

[0025] Figure 9 Explain the situation when the phenomenon of drift occurs.

[0026] Figure 10 This represents the second electron intensity distribution of a frame image containing saturated brightness values.

[0027] Figure 11 This is a flowchart illustrating an example of a method for generating a charged drift image.

[0028] Figure 12 This is a flowchart illustrating the inspection and processing flow of a batch in a defect re-inspection SEM.

[0029] Figure 13 This is a flowchart illustrating an example of a parameter adjustment method for defect detection processing.

[0030] Figure 14 This is a flowchart illustrating the inspection process for reproducing low-magnification SEM images.

[0031] Figure 15 This is a flowchart illustrating an example of a parameter adjustment method for defect detection processing using a replicated low-magnification SEM image.

[0032] Figure 16 This is a flowchart illustrating an example of optimized frame rate processing. Detailed Implementation

[0033] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. The embodiments described below are examples for implementing the present disclosure and do not limit the technical scope of the present disclosure. In the embodiments, components with the same function are labeled with the same symbols, and repeated descriptions are omitted except where specifically necessary.

[0034] (Implementation Method 1)

[0035] In this embodiment, a virtual image is generated using feature data calculated from the frame image, and the virtual image is used for scheme evaluation, etc.

[0036] <Structure of a Charged Particle Beam Device>

[0037] Figure 1This is a structural diagram illustrating an example of a charged particle beam device. Here, a scanning electron microscope is used as an example to explain the structure of the charged particle beam device. Figure 1 As shown, the scanning electron microscope (charged particle beam device) 200 includes an electron microscope 201, a computer system 202, and a display device 203.

[0038] The electron microscope 201 includes an electron source 204, a focusing lens 205, a deflector 206, an objective lens 207, a sample stage 209 for holding a sample 208, and a detector 210. The electron beam emitted from the electron source 204 is focused by the focusing lens 205 and the objective lens 207 and illuminates the sample 208 on the sample stage 209.

[0039] Deflector 206 deflects the electron beam according to the deflection signal output from computer system 202. Thus, the electron beam scans two-dimensionally across sample 208. Detector 210 detects secondary electrons and backscattered electrons generated from sample 208, converting the detected secondary electrons (secondary particles) and backscattered electrons into detection signals. These detection signals are then output to computer system 202.

[0040] like Figure 1 As shown, the computer system 202 includes a data processing unit 211, an image storage unit 212, an image processing unit 213, and a non-volatile memory 214.

[0041] The data processing unit 211 performs input and output of information between its constituent elements, as well as input and output of information with the electron microscope 201 and the display device 203.

[0042] The data processing unit 211 generates, for example, control signals (such as deflection signals) to control each component of the electron microscope 201. For example, the data processing unit 211 generates signals to adjust the bias and range of the detector 210, so that the detection signal output from the detector 210 converges within a predetermined dynamic range.

[0043] Regarding the adjustment of bias and range, it can be done by using hardware ABC (Auto Bias Control) or by using software ABC.

[0044] In addition, the data processing unit 211 receives various signals from the electron microscope 201, such as detection signals and response signals to control signals. For example, the data processing unit 211 generates frame images based on the detection signals and saves the generated frame images in the image storage unit 212.

[0045] In addition, the data processing unit 211 outputs image data for display on the display device 203. Image data may include, for example, image data containing frame images, or image data processed by the image processing unit 213 (described later).

[0046] The image processing unit 213 performs image processing based on frame images. For example, the image processing unit 213 generates a weighted average image 310 based on multiple frame images stored in the image storage unit 212. Figure 2 The processing performed by the image processing unit 213 will be described later.

[0047] Display device 203 is a device for displaying various information about the scanning electron microscope 200. For example, display device 203 displays setting information of the scanning electron microscope 200, information indicating its operating status, frame images, and images after image processing. Furthermore, display device 203 may also have the function of an input device, such as a touch panel. In this case, a user interface for inputting information is displayed on display device 203.

[0048] Figure 2 The method for generating the weighted average image is explained. The image processing unit 213 adds up all the results obtained by accumulating arbitrarily set weights w1, w2, ..., wk (w1+w2+...+wk=1) for each of the first frame image 310_1, the second frame image 310_2, ..., the kth frame image 310_k, to generate the weighted average image 310. The image processing unit 213 sets the respective weights to w1=w2=...=wk=1 / k, thereby enabling the generation of the average image 102 from multiple frame images.

[0049] The image processing unit 213 generates feature data based on the frame image and stores the generated feature data in the image storage unit 212. The feature data will be described in detail later.

[0050] Furthermore, the image processing unit 213 virtually generates a simulated image by performing a simulation, and stores the generated simulated image in the image storage unit 212. The simulation uses feature data stored in the image storage unit 212. Hereinafter, the simulated image will sometimes be referred to as a virtual image.

[0051] The data processing unit 211 and the image processing unit 213 are implemented, for example, by a processor such as a CPU executing programs stored in the non-volatile memory 214. Alternatively, some functions of the data processing unit 211 and the image processing unit 213 can also be implemented using dedicated hardware. In this case, the image storage unit 212 can be configured as a memory connected to the CPU, for example. Then, when the CPU performs the required processing, it can read the programs from the non-volatile memory 214 and expand them in the internal memory.

[0052] <Calculation of Feature Data>

[0053] Figure 3 This is a flowchart illustrating an example of a method for calculating feature data. The feature data calculated here includes, for example, the mean, standard deviation, kurtosis, and skewness of a frame image. These feature data are calculated in pixels. Figure 3 The process includes steps S501 to S507. In step S501, the electron microscope 201 scans the sample 208 with an electron beam for a predetermined number of times.

[0054] In step S502, the data processing unit 211 generates a frame image each time the electron beam scans the sample 208, and saves the generated frame image in the image storage unit 212. Then, the image processing unit 213 compares the multiple frame images with each other and calculates the positional offset between the frame images. Then, the image processing unit 213 uses the calculated positional offset to correct each frame image.

[0055] In step S503, the image processing unit 213 uses the Nth power average of the frame images to calculate the Nth order moment image.

[0056] First, the image processing unit 213 uses the corrected frame image as an intermediate image and calculates the first power average image <Img1>, the second power average image <Img2>, the third power average image <Img3>, and the fourth power average image <Img4> of the frame image respectively.

[0057] The first power average image of the frame image <Img1> is represented by equation (1) as shown below. The second power average image of the frame image <Img2> is represented by equation (2) as shown below. The third power average image of the frame image <Img3> is represented by equation (3) as shown below. The fourth power average image of the frame image <Img4> is represented by equation (4) as shown below. In equations (1) to (4), [i, j] represents the pixel position on the frame image, N represents the number of frame images, and frame image k represents the frame image of the kth frame.

[0058] <img1>[i, j] = (1 / N) × Σ{k = 1, ..., N} frames of images k[i, j]...(1)

[0059] <img2>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 2 …(2)

[0060] <img3>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 3 …(3)

[0061] <img4>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 4 …(4)

[0062] Hereinafter, the image processing unit 213 can similarly calculate the Nth power average image <ImgN>[i,j] of the frame image N.

[0063] Next, the image processing unit 213 calculates the Nth-order moment image of the frame image based on the 1st to 4th power average images and the Nth power average image calculated by equations (1) to (4). The 1st-order moment image of the frame image is represented by equation (5) shown below. The 2nd-order moment image of the frame image is represented by equation (6) shown below. The 3rd-order moment image of the frame image is represented by equation (7) shown below. The 4th-order moment image of the frame image is represented by equation (8) shown below.

[0064] The first-order moment graph [i, j] = <Img1> [i, j]…(5)

[0065] The second-order moment graph [i, j] = <Img2>[i, j] - <Img1>[i, j] 2 …(6)

[0066] 3rd order moment graph [i, j] = <img3>[i,j]﹣3 <img2>[i,j]× <img1>[i,j]+2 <img1>[i, j] 3 …(7)

[0067] 4th order moment graph [i, j] = <img4>[i,j]﹣4× <img3>[i,j]× <img1>[i,j]+6× <img2>[i,j]× <img1>[i,j] 2 ﹣3 <img1>[i, j] 4 …(8)

[0068] The Nth-order moment graph can be calculated using the expansion formula (9) of the binomial theorem shown below.

[0069] (a+b) N =Σ{k=0,…,N} N C k ×a k ×b (N-k) …(9)

[0070] Next, the image processing unit 213 uses each moment image to perform calculations for the average image (S504), the standard deviation image (S505), the kurtosis image (S506), and the skewness image (S507).

[0071] The average image in step S504 is calculated using the following equation (10).

[0072] The average image [i, j] = the first-order moment image [i, j]…(10)

[0073] The standard deviation image of step S505 is calculated by the following equation (11).

[0074] Standard deviation image

[0075] The skewness image in step S507 is calculated using equation (12) shown below.

[0076] Skewness image [i, j] = 3rd moment image [i, j] ÷ standard deviation image [i, j] 3 …(12)

[0077] The kurtosis image of step S506 is calculated by the following equation (13).

[0078] Kurtosis image [i, j] = 4th order moment image [i, j] ÷ standard deviation image [i, j] 4 -3…(13)

[0079] The order of steps S504 to S507 can be appropriately changed. When the processing of steps S504 to S507 is completed, the calculation and processing of the feature data is finished.

[0080] Here, an example of calculating a 4th-order moment image is shown. By using a 4th-order moment image, the load on the image processing unit 213 can be reduced, and the accuracy of feature calculation can be ensured.

[0081] <The signal generation process in SEM images>

[0082] This section explains the signal generation process in SEM images. Figure 4 This schematically illustrates an example of the intensity distribution of a secondary electronic signal relative to a patterned cross section. Figure 4 (a) represents the patterned cross-section of the specimen. In Figure 4 In (a), 401 represents the non-patterned edge and 402 represents the patterned edge. Figure 4 (b) represents the intensity distribution of the second-order electronic signal. Figure 4 In (b), the horizontal axis represents the intensity of the second-order electron signal, and the vertical axis represents the frequency. Figure 4 In (b), 403 is the intensity distribution of the secondary electronic signal of the non-patterned edge portion 401, and 404 is the intensity distribution of the secondary electronic signal of the patterned edge portion 402.

[0083] Due to edge effects, secondary electrons in the SEM are easily generated from the unpatterned edge portion 401. On the other hand, the SEM secondary electrons generated from the patterned edge portion 402 are relatively fewer compared to those generated from the unpatterned edge portion 401. Assuming that the probability of secondary electron generation is a random process, and that the frequency of secondary electron generation follows a normal distribution.

[0084] In this embodiment, as described above, feature data (mean, standard deviation, skewness, kurtosis) are obtained by extracting the characteristics of the probability distribution from N frame images on a pixel-by-pixel basis. For example, the pixels on the average image corresponding to the pattern edge 402 are... Figure 2 The pixels 320 at the edge shown. The pixels on the average image corresponding to the non-patterned edge 401 become Figure 2 The non-edge pixel 330 is shown.

[0085] Similarly, in the first frame, the pixel corresponding to the non-patterned edge 401 is pixel 330_1, and the pixel corresponding to the patterned edge 402 is pixel 320_1. In the second frame, the pixel corresponding to the non-patterned edge 401 is pixel 330_2, and the pixel corresponding to the patterned edge 402 is pixel 320_2. In the kth frame, the pixel corresponding to the non-patterned edge 401 is pixel 330_k, and the pixel corresponding to the patterned edge 402 is pixel 320_k.

[0086] If there are enough frame images, the intensity distribution 403 of the secondary electronic signal of the non-patterned edge portion 401 and the intensity distribution 404 of the secondary electronic signal of the patterned edge portion 402 on the frame image follow a normal distribution.

[0087] If we assume that the intensity distribution of the second electron signal follows a normal distribution, then we can virtually generate any K frames of SEM images as simulated images (virtual images) based on the central limit theorem. Furthermore, we can also virtually generate a cumulative image of multiple frames from the virtual images of each frame.

[0088] The Central Limit Theorem states that when calculating the average of sampled data following a continuous probability distribution, the deviation of the average converges to the value obtained by dividing the population standard deviation by the square root of the number of samples.

[0089] <Methods for Generating Virtual Images>

[0090] Next, the method for generating virtual images will be explained. Figure 5 Explain the method for generating virtual images. Figure 5 (a) represents the concept map of virtual image generation. Figure 5 In (a), 101 is the virtual image. 102 is the average image. 103 is the standard deviation image. 104 is the random number image. 105 is the correction coefficient. For example, as... Figure 5 As shown, the correction factor 105 is a value specified by the reciprocal of the square root of the number of frames (K) used to generate the virtual image. Figure 5 (b) represents the distribution of normally distributed random numbers 106. Normally distributed random numbers have the characteristics of mean = 0 and standard deviation = 1.

[0091] Figure 6 This is a flowchart illustrating an example of a method for generating virtual images. Figure 6 In the example, when generating the virtual image, steps S1001 to S1004 are executed. In step S1001, the image processing unit 213 uses the distribution 106 of normally distributed random numbers to generate a random number image 104 representing the frequency distribution of secondary electron generation. In step S1002, the image processing unit 213 calculates a correction coefficient 105.

[0092] Then, the image processing unit 213 multiplies the standard deviation image 103, the random number image 104, and the correction coefficient 105 (step S1003), and adds the average image 102 to the result calculated in step S1003 to generate the simulated image 101 in the K frames (S1004). The result obtained by multiplying the standard deviation image 103 and the correction coefficient 105 is the deviation (standard deviation) of the cumulative image in the K frames.

[0093] The same process could be performed by randomly selecting K frames of images for calculating feature data and averaging them, but the storage capacity required to store the image data would be enormous, making it impractical. Furthermore, the image processing would become cumbersome, placing a greater burden on the image processing unit 213.

[0094] The virtual images generated in this way can be used, for example, as images for pattern matching or for scheme evaluation.

[0095] Virtual images are generated from the average image, standard deviation image, and random number image obtained from multiple frame images. Based on this structure, images for scheme evaluation are generated using only a smaller number (3) of images. Therefore, the generation time for images for scheme evaluation can be shortened. This helps to suppress the increase in data volume and reduces the time required for scheme evaluation.

[0096] Furthermore, according to this embodiment, a virtual image of the SEM image with altered shooting conditions based on feature data can be generated, thus enabling efficient evaluation of different scenarios without the need for a device. Additionally, since damage to the sample is suppressed and reproducibility is improved, cumulative evaluation of multiple frames can be performed.

[0097] Furthermore, by utilizing virtual images, various parameters, including measurement parameters, can be optimally set without relying on the experience of application engineers. Additionally, when using design data, virtual images can be used to predict the reproducibility of noise causes, eliminating the need for measurements to verify the sample.

[0098] (Implementation Method 2)

[0099] Next, implementation method 2 will be described.

[0100] Reducing the frame rate to shorten the SEM image capture time can speed up defect re-inspection. However, it is difficult to predict the decrease in defect detection rate when the frame rate is reduced, requiring the program to be run again to monitor the defect detection rate. Therefore, there is a problem of time-consuming image processing parameter optimization.

[0101] Furthermore, the detection position of the measured edge is visually confirmed to determine whether the frame rate of the captured image is optimal and whether the signal-to-noise ratio (S / N) is sufficient. In the case of detecting minute defects or foreign objects, it is necessary to visually confirm their existence.

[0102] Furthermore, optimizing image processing parameters for detecting defects from SEM images requires acquiring numerous SEM images, which capture defects and contain significant noise. Repeatedly acquiring SEM images not only increases the time required for optimization but also alters the visual recognizability of defects caused by contamination traces (contamination), thus presenting a challenge in achieving reproducible optimization of image processing parameters.

[0103] Therefore, in this embodiment, skewness and kurtosis are used to optimize the cumulative number of frames. Figure 7 This illustrates the characteristics of skewness and kurtosis. Figure 7 (a) represents the characteristic of skewness 601. Figure 7 (b) represents the kurtosis characteristic 605. In Figure 7 of (a), Figure 7 In (b), the horizontal axis represents the intensity of the second electronic signal, and the vertical axis represents the frequency.

[0104] In a normal distribution, the skewness and kurtosis are usually known to be zero. Figure 7 In (a), when the frequency distribution of skewness becomes a distribution 603 that is biased towards the positive from the normal distribution 602, the kurtosis becomes a negative value. On the other hand, when the frequency distribution of skewness becomes a distribution 604 that is biased towards the negative from the normal distribution 602, the kurtosis becomes a positive value.

[0105] exist Figure 7 In (b), when the frequency distribution of kurtosis changes from a normal distribution 606 to a distribution with a bulging hem 607, the skewness becomes a positive value. On the other hand, when the frequency distribution of kurtosis changes from a normal distribution 606 to a distribution with a contracting hem 608, the skewness becomes a negative value.

[0106] Next, we will explain how to optimize the frame accumulation using these characteristics of skewness and kurtosis. It is known that the intensity distribution of the second-order electron signal, which corresponds to the frame accumulation, roughly follows a Poisson distribution, and when the frame accumulation is sufficient, it approaches a normal distribution according to the central limit theorem. When the intensity distribution of the second-order electron signal follows a normal distribution, the kurtosis and skewness are zero, as mentioned above. Therefore, by determining whether the kurtosis and skewness are close to zero, the frame accumulation can be optimized.

[0107] Specifically, the image processing unit 213 can determine whether the cumulative number of frames is optimal by determining whether the kurtosis and skewness are below predetermined thresholds set in advance.

[0108] For example, if the kurtosis and skewness are greater than a predetermined threshold, the frame accumulation is considered insufficient, and frame acquisition continues. Conversely, if the kurtosis and skewness are below the predetermined threshold, the frame accumulation is considered sufficient, and frame acquisition stops. In this way, by determining the kurtosis and skewness, the frame accumulation can be optimized.

[0109] Furthermore, the optimization of the cumulative frame count can also be performed during the execution of the scheme. During the execution of the scheme, the image processing unit 213 calculates the kurtosis and skewness of the accumulated image each time a frame image is acquired, and compares the calculated kurtosis and skewness with a predetermined threshold, thereby determining whether the cumulative frame count is optimal and changing the cumulative frame count.

[0110] According to this embodiment, the frame accumulation number can be optimized within a scanning electron microscope (charged particle beam device) 200 using skewness and kurtosis. Therefore, frame accumulation number optimization can be performed in a short time without operator confirmation.

[0111] Furthermore, it is possible to confirm whether the frame rate of the captured images is optimal or the signal-to-noise ratio is sufficient during the registration process. Additionally, it eliminates the need for visually verifying the detection location of the measured edges or visually confirming whether minor defects or foreign objects are false alarms.

[0112] Furthermore, in pattern matching with the registered guide pattern image, by using a virtual image instead of the SEM image (frame image), the minimum cumulative number of frames required for pattern detection can be easily calculated.

[0113] (Implementation Method 3)

[0114] Next, implementation method 3 will be described.

[0115] For example, pattern size measurements using CD-SEM are automatically and continuously performed according to the following scheme, which consists of the position of the measurement point on the semiconductor wafer, SEM observation conditions, a pattern image that guides the determination of the measurement position, and measurement parameters. Specifically, the CD-SEM moves to the measurement pattern position according to the scheme information, scans the pattern under the registered SEM observation conditions, and obtains a SEM image. After processing the obtained SEM image based on measurement parameters such as the following threshold, the distance between the two edges of the measurement pattern is measured, wherein the threshold is used to determine the smoothing size and edge position of the noise removal filter.

[0116] The smoothing size and edge position thresholds used to determine the noise removal filter depend on the noise level and edge shape of the SEM image. Therefore, the optimization of measurement parameters largely depends on the experience of the application engineer. Furthermore, it is difficult to explain the basis for previously set measurement parameters.

[0117] Furthermore, since measurement parameters need to be pre-registered, even if the position of the target pattern is determined using design data, it is still necessary to confirm the reproducibility when repeatedly measuring the edge detection of the pattern.

[0118] After the CD-SEM moves the worktable to the measurement pattern position according to the scheme information, it performs pattern matching between the registered guide pattern image and the acquired image to detect the measurement position with high accuracy. In order to improve throughput, it is preferable to acquire SEM images in a short time. However, due to the influence of image contrast and noise, the optimization of measurement parameters takes a long time.

[0119] Therefore, in this embodiment, when generating virtual images, multiple random number images 104 that conform to the distribution of standard deviation images 103 are prepared, thereby generating multiple virtual images with different white noise.

[0120] By using multiple virtual images, the thresholds used to determine spatial smoothing dimensions and edge detection locations in pattern measurement processing can be easily optimized offline, i.e., within the device.

[0121] In addition, by using multiple virtual images, such as performing pattern matching to determine the pattern measurement location and determining whether the pattern matching is successful, these thresholds can be optimized.

[0122] (Implementation Method 4)

[0123] Next, Embodiment 4 will be described. Due to prolonged irradiation with an electron beam during frame accumulation, a shrinkage phenomenon, known as pattern shrinkage, sometimes occurs in the sample 208. If this shrinkage occurs, the reproducibility of the execution results decreases. This shrinkage phenomenon is significant, for example, in resist materials. Therefore, in this embodiment, the skewness and kurtosis of the characteristic data are used to detect the shrinkage phenomenon.

[0124] Figure 8 Describe the situation when contraction occurs. Figure 8 (a) represents the change in the pattern cross section when shrinkage occurs. Figure 8 (b) represents the frequency histogram change of the second electronic signal intensity of the pattern outline portion when contraction occurs.

[0125] like Figure 8 As shown in (a), when shrinkage occurs, the pattern cross section of the sample 208 changes from the pattern cross section 701 at the beginning of the electron beam scan to the pattern cross section 702 in the middle stage of the electron beam scan and the pattern cross section 703 at the end of the electron beam scan, and the pattern width gradually decreases.

[0126] At this point, the frequency of secondary electron generation in the pattern outline portion 704 becomes a composite distribution 708, obtained by combining the secondary electron generation probability distribution (normal distribution) 705 of the edge portion of the pattern section 701 at the beginning of the electron beam scan, the secondary electron generation probability distribution (normal distribution) 706 of the edge portion of the pattern section 702 in the middle stage of the electron beam scan, and the secondary electron generation probability distribution (normal distribution) 707 of the edge portion of the pattern section 703 at the end of the electron beam scan. Thus, the probability distribution at the time of contraction deviates from the normal distribution. Figure 8 (b)

[0127] In the synthetic distribution 708, which deviates from the normal distribution, skewness and kurtosis become non-zero values, thus enabling the detection of which part of the pattern within the image is shrinking. Consequently, if shrinkage is detected during registration, the following measures can be taken: reducing the cumulative number of frames, or calculating the amount of shrinkage and switching to a mode that reflects the amount of shrinkage in the measurement value.

[0128] According to this embodiment, the shrinkage phenomenon of the pattern can be automatically detected, and appropriate processing can be performed when registering the scheme or executing the scheme.

[0129] (Implementation Method 5)

[0130] Next, implementation method 5 will be described.

[0131] Because the sample 208 becomes charged due to prolonged electron beam irradiation during frame accumulation, either charge drift or stage drift occurs. Charge drift refers to image drift, while stage drift refers to image drift caused by the inertia of the sample stage 209 immediately after it stops. When image drift occurs, the reproducibility of the execution results decreases. In this embodiment, a method for detecting these image drift phenomena is described.

[0132] Figure 9 Explain the situation when the phenomenon of drift occurs. Figure 9 (a) represents the frame image that produces the drift phenomenon. Figure 9 (b) represents the frequency histogram variation of the second electronic signal intensity at the left edge. Figure 9 (c) represents the frequency histogram variation of the second electronic signal intensity at the right edge.

[0133] like Figure 9 As shown in (a), when the image drift phenomenon occurs, it is observed that the pattern moves in parallel from the frame image 801 at the beginning of the electron beam scan, in the order of the frame image 802 in the middle stage of the scan and the frame image 803 at the end of the scan.

[0134] On the other hand, in the left edge pixel 811 and right edge pixel 821 specified in the frame image 801 at the start time of electron beam scanning, the brightness values ​​change sequentially as shown in the frame images 802 and 803 as the pattern moves.

[0135] At this point, the frequency of secondary electron generation at the left edge of the pattern becomes a composite distribution 834, obtained by combining the secondary electron generation probability distribution (normal distribution) 831 at the beginning of the electron beam scan, the secondary electron generation probability distribution (normal distribution) 832 during the middle stage of the scan, and the secondary electron generation probability distribution (normal distribution) 833 at the end of the scan. Thus, the probability distribution at the left edge during image drift deviates from the normal distribution. Figure 9 (b)

[0136] Similarly, the frequency of secondary electron generation at the right edge of the pattern becomes a composite distribution 844, which is the result of combining the probability distributions of secondary electron generation (normal distribution) 841 at the beginning of the electron beam scan, the probability distributions of secondary electron generation (normal distribution) 842 during the middle stage of the scan, and the probability distributions of secondary electron generation (normal distribution) 843 at the end of the scan. Thus, the probability distribution at the right edge during image drift also deviates from the normal distribution. Figure 9 (c)).

[0137] At this point, similar to the detection of the aforementioned shrinkage phenomenon, the composite distribution 834 at the left edge and the composite distribution 844 at the right edge deviate from the normal distribution, with non-zero values ​​for skewness and kurtosis. Therefore, it is possible to detect brightness changes in the pattern within the frame image.

[0138] Furthermore, by comparing the changes in brightness values ​​at the left and right edges, it is possible to distinguish between image drift and shrinkage. The data used for comparison here include the variance, skewness, and kurtosis of a normal distribution.

[0139] Furthermore, if image drift is detected during the execution of the scheme, the image processing unit 213 calculates the drift amount and generates an image drift image after correcting the frame image based on the drift amount. Additionally, the image processing unit 213 uses the image drift image to recalculate the average image. Then, the image processing unit 213 calculates a virtual image using the recalculated average image.

[0140] Methods for generating charged drift images

[0141] Here, a method for generating charged drift images will be explained as an example of a drift image. Figure 11 This is a flowchart illustrating an example of a method for generating charged drift images. Figure 11 This includes steps S1101 to S1107.

[0142] In step S1101, the image processing unit 213 generates a cumulative image based on multiple frame images. However, since the initial frame images have already been generated at this stage, the brightness of each pixel in the cumulative image is zero.

[0143] In steps S1102 to S1106, each frame image is processed. In step S1103, the data processing unit 211 generates a frame image and saves the generated frame image in the image storage unit 212. In step S1104, the image processing unit 213 calculates feature data from the frame image generated in step S1103. Then, the image processing unit 213 compares the calculated feature data with the feature data when no charge drift phenomenon occurs, and calculates the feature data deviation in that frame image. The image processing unit 213 may also calculate the feature data deviation by comparing it with the feature data in the immediately preceding frame image.

[0144] In step S1105, the image processing unit 213 performs image translation based on the calculated feature data offset to generate a translated image.

[0145] In step S1106, the accumulated image up to the previous frame is accumulated with the translation image generated in the current frame to generate a new accumulated image. Then, the accumulated image generated in step S1106 relative to the last frame image becomes the charged drift image (step S1107).

[0146] According to this embodiment, it is possible to automatically detect image drift caused by electrification phenomena, such as drift when the worktable stops, and to perform appropriate processing when registering a scheme or executing a scheme.

[0147] (Implementation Method 6)

[0148] Next, implementation method 6 will be described.

[0149] In a typical approach, a frame image is acquired after the detection signal converges within a predetermined dynamic range by controlling the bias and range of detector 210 via hardware ABC. To improve throughput, another approach exists: omitting the hardware ABC step by setting a pre-registered bias and range on detector 210.

[0150] However, the amount of secondary electrons released can vary significantly depending on the material of the sample. Therefore, sometimes the secondary electron detection signal does not converge within the system's dynamic range, resulting in saturation of brightness values ​​in certain areas (pixels) of the frame image, making it impossible to obtain a suitable image. Therefore, in this embodiment, the bias and range are reset.

[0151] Figure 10 This represents the second electron intensity distribution of a frame image containing saturated brightness values. In cases where some brightness values ​​within the image have saturated but not converged to the dynamic range, the second electron intensity distribution becomes, for example... Figure 6 Such distributions. 902 is the second-order electron signal intensity distribution in the non-edge region. 903 is the second-order electron signal intensity distribution in the edge region. These distributions deviate from the normal distribution, therefore their skewness and kurtosis are non-zero values.

[0152] Furthermore, by calculating the distribution 902 in the non-edge region and the distribution 903 in the edge region respectively, it is possible to determine whether these distributions saturate at the minimum value of the dynamic range or at the maximum value of the dynamic range.

[0153] According to this embodiment, when executing the scheme, if saturation of brightness value is detected in the frame image obtained when a pre-registered bias and range are set in the detector 210, a frame image with unsaturated brightness value can be obtained by additionally using hardware ABC to reset the bias and range.

[0154] (Implementation Method 7)

[0155] Next, implementation method 7 will be described.

[0156] In this embodiment, a scheme condition evaluation method using SEM images is described, wherein the defect re-inspection object is captured in the SEM image. First, the defect inspection process in defect re-inspection, the definition of defect detection rate, and an overview of parameter adjustments for defect detection processing are explained. Then, as a specific example, a scheme condition evaluation method related to defect inspection using virtual images is described.

[0157] Figure 12 This is a flowchart illustrating the inspection and processing flow of one batch in a defect re-inspection SEM. Figure 12 The process includes steps S1201 to S1209. During inspection, after the semiconductor wafer loading process (S1201), an alignment process (S1202) is performed. The alignment process is a process of detecting alignment marks located on the semiconductor wafer and performing coordinate correction. In the alignment process, alignment marks of at least two points are detected.

[0158] Next, the following steps S1203 to S1209 are repeatedly performed on each defect re-inspection target on the semiconductor wafer. Assume that the location and size of the defect re-inspection target on the semiconductor wafer are known from the inspection by the pre-inspection device. First, the field of view is moved towards the defect re-inspection location (S1203).

[0159] Since the defect re-inspection location measured by the preceding inspection device contains measurement error, a low-magnification SEM image of the defect re-inspection location entering the field of view is first captured (S1204), and the captured image is stored, for example, in the image storage unit 212 of the computer system 202 (S1205).

[0160] The image processing unit 213 performs image processing (defect detection processing) (S1206) on the low-magnification SEM image for detecting defect re-inspection objects. Based on the image processing results, the image processing unit 213 calculates the size of the defect re-inspection object and the positional offset determined by the difference between the detected defect re-inspection object's position and the center of the SEM image.

[0161] Then, the computer system 202 adjusts the deflection of the electron beam based on the position offset, so that the defect re-inspection object appears in the center of the image (S1207). After adjusting the deflection of the electron beam, a high-magnification SEM image is captured (S1208), and the captured image is saved in the image storage unit 212 (S1209). When the processing of all defect re-inspection objects is completed, the inspection process ends.

[0162] The probability frequency of identifying the location of a defect (the object of defect re-inspection) in the inspection and processing flow is called the defect detection rate. If the defect detection location is contained within a defect area, it is considered that the defect detection location has been accurately identified.

[0163] In the inspection process of semiconductor equipment, a defect detection rate of over 95% is required. The imaging method of SEM images (e.g., the pattern shape of non-defective areas, the sharpness of defects, or the signal-to-noise ratio) varies depending on the type of inspection process, materials, and manufacturing equipment. Therefore, appropriately adjusting the parameters of the defect detection process is essential to improving the accuracy of the defect detection rate.

[0164] Next, the parameter adjustment method for defect detection and processing will be explained. Figure 13 This is a flowchart illustrating an example of a parameter adjustment method for defect detection processing. Figure 13 This includes steps S1301 to S1305. Here, it is assumed that the inspection process has been performed in advance, and a low-magnification SEM image and a defect depiction area on the SEM image have been shown.

[0165] First, the defect detection parameters are initially set (S1301). For the initial parameters, for example, parameters adjusted using an SEM image close to the process being inspected are used. Next, the low-magnification SEM image shown is processed using the current parameters. Figure 12 Defect detection processing (S1302). Image processing unit 213 calculates defect detection rate based on defect detection results and taught defect depiction areas (S1303).

[0166] In step S1304, the image processing unit 213 determines whether to end the parameter adjustment of the defect detection process. If the defect detection rate is above a predetermined value (e.g., 95%) (Yes), the image processing unit 213 ends the parameter adjustment of the defect detection process. On the other hand, if the defect detection rate is less than 95% (No), the image processing unit 213 updates the parameters using a low-magnification defect image and the defect detection rate (S1305).

[0167] For parameter updates, a parameter search method using the experimental planning approach is employed, selectively changing parameters that are highly sensitive to defect detection rates. For example, in the case of defect detection processing using deep learning, parameters that are highly sensitive to defect detection rates can be selectively changed by using inverse error propagation or stochastic gradient methods.

[0168] According to this embodiment, defect detection parameters can be updated based on the defect detection rate.

[0169] (Implementation Method 8)

[0170] Next, implementation method 8 will be described.

[0171] In this embodiment, a defect inspection method using feature data will be described. Specifically, the feature data described in Embodiment 1 is used to reproduce the low-magnification SEM image shown. This allows for the adjustment of defect detection parameters using a small number of images.

[0172] In this embodiment, the extension Figure 12 Defect inspection process and Figure 13 The process for adjusting defect detection parameters.

[0173] Figure 14 This is a flowchart illustrating the inspection process for reproducing low-magnification SEM images. Figure 14 and Figure 12 Similarly, with Figure 12 The main difference is that step S1410 is inserted between step S1204 and step S1205.

[0174] In step S1410, the image processing unit 213 calculates feature data related to each frame image using the method described in Embodiment 1. Then, in Figure 14 In step S1205, the feature data (feature data image) calculated in step S1410 is stored in the image storage unit 212. At this time, the corresponding frame image can also be stored together with the feature data image.

[0175] Figure 15 This is a flowchart illustrating an example of a parameter adjustment method for defect detection processing using a copied low-magnification SEM image. Figure 15 and Figure 13 Similarly, with Figure 13 The main difference is that step S1506 is inserted between step S1301 and step S1302.

[0176] In step S1506, the image processing unit 213 uses a method for processing the low-magnification SEM image shown in the example. Figure 14 The feature data calculated in step S1410 is used to virtually replicate the low-magnification SEM image.

[0177] Then, in step S1302, the image processing unit 213 uses the original SEM image, the SEM image copied in step S1506, and the currently set parameters to execute... Figure 14 Defect detection and processing.

[0178] Because SEM images are prone to noise, when adjusting defect detection parameters using teaching data that contains a lot of noise, it is necessary to obtain a large number of similar defect images during the defect inspection process.

[0179] In contrast, in this embodiment, similar defect images with the same S / N ratio can be reproduced, thus reducing the teaching data for parameter adjustment in defect detection and enabling the collection of teaching data in a short period of time.

[0180] According to this embodiment, a virtual SEM image is generated when searching for parameters to identify defects from the acquired SEM image. This allows for the addition of teaching data to optimize the parameters.

[0181] (Implementation Method 9)

[0182] Next, implementation method 9 will be described.

[0183] In this embodiment, a method for optimizing shooting conditions to maintain the defect detection rate above a predetermined value (e.g., 95%) using the feature data described in Embodiment 1 will be explained.

[0184] In defect inspection and re-inspection, a faster inspection process is required while maintaining the defect detection rate. To speed up the defect inspection process, shortening the acquisition time of low-magnification SEM images is essential.

[0185] Therefore, in this embodiment, the number of shooting frames is optimized in order to shorten the shooting time of low-magnification SEM images.

[0186] Figure 16 This is a flowchart illustrating an example of frame rate optimization processing. Figure 16 This includes steps S1601 to S1606. As... Figure 16 The prerequisite for the process is that the defect inspection process of Implementation Method 8 has been executed beforehand. Figure 14 ) and parameter adjustment for defect detection ( Figure 15 ).

[0187] First, the image processing unit 213 obtains the frame number from the shooting conditions of the captured image obtained in the defect inspection process and sets the shooting frame number as the current frame number (S1601). Then, the image processing unit 213 virtually generates a SEM image (virtual image) that accumulates the current frame number using the virtual image generation method described in Embodiment 1 (S1602).

[0188] The image processing unit 213 performs defect detection processing using the virtually generated SEM image in step S1602 (S1603). The image processing unit 213 calculates the defect detection rate using the defect detection results and the pre-taught defect region (S1604).

[0189] In step S1605, the image processing unit 213 determines whether to end the adjustment of the number of shooting frames. If the defect detection rate is above a predetermined value (e.g., 95%) (No), the image processing unit 213 updates the number of shooting frames to half of the current number (S1606) and continues the adjustment of the number of shooting frames. At this time, the image processing unit 213 maintains the previously set number of shooting frames. On the other hand, if the defect detection rate is less than 95% (Yes), the image processing unit 213 restores the number of shooting frames to the previously set number of shooting frames and ends the adjustment of the number of shooting frames.

[0190] According to this embodiment, the number of shooting frames can be set to the minimum number of shooting frames with a defect detection rate of a predetermined value (e.g., 95%) or higher, which can shorten the shooting time.

[0191] Explanation of reference numerals in the attached figures

[0192] 200 Scanning electron microscope (charged particle beam device), 201 Electron microscope, 202 Computer system, 203 Display device, 211 Data processing unit, 212 Image storage unit, 213 Image processing unit, 214 Non-volatile memory.

Claims

1. A charged particle beam device, characterized in that, have: A microscope that scans a sample with a beam of charged particles, detects secondary particles emitted from the sample, and outputs a detection signal; and A computer system generates frame images based on the detection signals and performs image processing based on the frame images. The computer system calculates moment images among multiple frame images, and calculates feature data of the frame images based on the moment images. The feature data includes mean, variance, skewness, and kurtosis. The computer system calculates the first-order moment images to the fourth-order moment images of the multiple frame images, represented by the following equations (5) to (8), based on the first-order to fourth-order average images of the multiple frame images calculated by equations (1) to (4) as shown below: <img1> [i,j]=(1 / N)×Σ{k=1,…,N}frames of image k[i,j]…Equation (1), <img2>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 2 …Equation (2), <img3>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 3 …Equation (3), <img4>[i, j] = (1 / N) × ∑{k = 1, ..., N} frames of images k[i, j] 4 …Equation (4), First-order moment graph [i,j] = <Img1>[i,j]...Equation (5), The second-order moment graph [i, j] = <Img2>[i, j] - <Img1>[i, j] 2 …Equation (6), 3rd order moment graph [i, j] = <img3>[i,j]﹣3 <img2>[i,j]× <img1>[i,j]+2 <img1>[i, j] 3 …Equation (7), 4th order moment graph [i, j] = <img4>[i,j]﹣4× <img3>[i,j]× <img1>[i,j]+6× <img2>[i,j]× <img1>[i,j] 2 ﹣3 <img1>[i, j] 4 …Equation (8), In equations (1) to (8) above, [i, j] represents the pixel position on the frame image, N represents the number of frame images, and frame image k represents the frame image of the k-th frame. The computer system uses the feature data and multiple random number images following the distribution of standard deviation images to generate multiple simulated images with different white noise levels, and uses the simulated images to evaluate the scheme.

2. The charged particle beam device according to claim 1, characterized in that, The computer system calculates the positional offset between the multiple frame images and corrects the frame images based on the positional offset.

3. The charged particle beam device according to claim 1, characterized in that, The computer system optimizes the cumulative number of the frame images based on the kurtosis and skewness values ​​of the frame images when the energy distribution of the secondary particles released from the sample follows a normal distribution.

4. The charged particle beam device according to claim 1, characterized in that, The computer system detects the shrinkage phenomenon of the sample based on the kurtosis and skewness values ​​of the frame image when the energy distribution of the secondary particles released from the sample follows a normal distribution.

5. The charged particle beam device according to claim 1, characterized in that, The computer system detects image drift of the sample based on the kurtosis and skewness values ​​of the frame image when the energy distribution of the secondary particles emitted from the sample follows a normal distribution.

6. The charged particle beam device according to claim 1, characterized in that, When the energy distribution of the secondary particles emitted from the sample follows a normal distribution and the brightness values ​​of some pixels in the frame image are saturated, the computer system resets the bias and range of the detector used to detect the secondary particles based on the kurtosis and skewness values ​​of the frame image.

7. The charged particle beam device according to claim 1, characterized in that, The computer system evaluates the scheme conditions by using the defect detection processing of the frame image of the defect re-inspection object. When the defect detection rate is less than a predetermined value, the defect detection parameters set in the scheme conditions are updated.

8. The charged particle beam device according to claim 7, characterized in that, The computer system calculates the feature data of the frame image of the defect re-inspection object, uses the feature data to generate a simulated image for defect detection, and uses the simulated image for defect detection to perform the defect detection processing.

9. The charged particle beam device according to claim 1, characterized in that, The computer system evaluates the scheme conditions by using defect detection processing of the frame images of the defect re-inspection object, and updates the number of shooting frames set in the scheme conditions according to the defect detection rate.

10. The charged particle beam device according to claim 9, characterized in that, When the defect detection rate exceeds a predetermined value, the computer system updates the number of shooting frames set in the scheme conditions to half of the original number.

11. The charged particle beam device according to claim 9, characterized in that, When the defect detection rate is less than a predetermined value, the computer system restores the number of shooting frames set in the scheme conditions to the previously set number.

Images