Profile detection method and profile detection device

Abstract

A profile detection method according to the present invention includes a detection step and an output step. The detection step is for using a trained model using a training image including a specific form and information pertaining to the specific form included in the training image to detect, from an image to be detected including a specific form, the specific form included in the image to be detected. The output step is for outputting form information about the specific form that has been detected.

Description

Technical Field

[0001] The present disclosure relates to a profile detection method and a profile detection device.

Background Art

[0002] Patent Document 1 discloses a technique for imaging a circuit pattern existing at a desired position on a semiconductor device by a scanning electron microscope (SEM) in order to measure or inspect a semiconductor.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The present disclosure provides a technique for efficiently detecting a specific shape included in a detection target image.

Means for Solving the Problems

[0005] A profile detection method according to an aspect of the present disclosure includes a detection step and an output step. The detection step detects a specific shape included in a detection target image from the detection target image using a learning image including the specific shape and a model that has learned information regarding the specific shape included in the learning image. The output step outputs shape information of the detected specific shape.

Effects of the Invention

[0006] According to the present disclosure, a specific shape included in a detection target image can be efficiently detected.

Brief Description of the Drawings

[0007] [Figure 1]Figure 1 shows an example of the functional configuration of an information processing device according to the embodiment. [Figure 2] Figure 2 shows an example of a cross-sectional image of a substrate according to this embodiment. [Figure 3] Figure 3 shows an example of a binarized image according to the embodiment. [Figure 4] Figure 4 shows an example of the process for generating a boundary detection model according to the embodiment. [Figure 5] Figure 5 shows an example of the flow for detecting each recessed region in the image according to the embodiment. [Figure 6] Figure 6 shows an example of the flow for detecting the boundary of a film included in an image according to the embodiment. [Figure 7] Figure 7 shows an example of the detection results of the membrane boundary according to the present invention. [Figure 8] Figure 8 is a schematic diagram showing an example of the flow of the profile detection method according to the embodiment. [Figure 9] Figure 9 is a schematic diagram illustrating another example of the flow of the profile detection method according to the embodiment. [Modes for carrying out the invention]

[0008] Hereinafter, embodiments of the profile detection method and profile detection apparatus disclosed in this application will be described in detail with reference to the drawings. However, the disclosed profile detection method and profile detection apparatus are not limited by these embodiments.

[0009] Traditionally, process engineers have assisted in optimizing semiconductor manufacturing process recipes. For example, a scanning electron microscope is used to image a cross-section of a semiconductor device with recesses such as trenches and holes. The suitability of the manufacturing process recipe is then determined by measuring the dimensions, such as the Critical Dimension (CD), of the recesses in the image. Process engineers manually specify the range of the recesses in the image and the position of the contours to be measured, making the measurement process human-dependent. As a result, the measurement process is time-consuming. Furthermore, because the measurement process, including the specification of the contour positions, is human-dependent, human-dependent errors may occur in the measured dimensions. In addition, measuring the dimensions of numerous recesses is laborious.

[0010] Therefore, there is a growing demand for technologies that can efficiently detect specific shapes, such as recesses, contained within images.

[0011] [Embodiment] Embodiments will now be described. The following description will use the case where the dimensions of specific shapes, such as recesses, contained in an captured image are measured by the information processing device 10 as an example. Figure 1 is a diagram showing an example of the functional configuration of the information processing device 10 according to the embodiment. The information processing device 10 is a device that provides the function of measuring the dimensions of specific shapes contained in an captured image. The information processing device 10 is a computer, such as a server computer or a personal computer. A process engineer uses the information processing device 10 to measure the dimensions of recesses in an captured image. The information processing device 10 corresponds to the profile detection device of this disclosure.

[0012] The information processing device 10 includes a communication interface unit 20, a display unit 21, an input unit 22, a storage unit 23, and a control unit 24. The information processing device 10 may also include other components of a computer in addition to the above-mentioned components.

[0013] The communication I / F unit 20 is an interface that performs communication control with other devices. The communication I / F unit 20 is connected to a network (not shown) and transmits and receives various information with other devices via the network. For example, the communication I / F unit 20 receives data of a digital image captured by a scanning electron microscope.

[0014] The display unit 21 is a display device that displays various information. Examples of the display unit 21 include display devices such as LCD (Liquid Crystal Display) and CRT (Cathode Ray Tube). The display unit 21 displays various information.

[0015] The input unit 22 is an input device that inputs various information. For example, examples of the input unit 22 include input devices such as a mouse and a keyboard. The input unit 22 receives an operation input from a user such as a process engineer and inputs operation information indicating the received operation content to the control unit 24.

[0016] The storage unit 23 is a storage device such as a hard disk, SSD (Solid State Drive), or optical disk. Note that the storage unit 23 may be a semiconductor memory such as RAM (Random Access Memory), flash memory, or NVSRAM (Non Volatile Static Random Access Memory) in which data can be rewritten.

[0017] The storage unit 23 stores an OS (Operating System) executed by the control unit 24 and various programs including a profile detection program described later. Further, the storage unit 23 stores various data used in the programs executed by the control unit 24. For example, the storage unit 23 stores learning data 30, image data 31, and model data 32.

[0018] The learning data 30 is data used for generating a model to be used for detecting a profile. The learning data 30 includes various types of data used for generating a model. For example, the learning data 30 stores data of a learning image including a specific shape to be detected and information regarding the specific shape included in the learning image.

[0019] The image data 31 is data of a detection target image for detecting a profile.

[0020] The model data 32 is data storing a model for detecting a specific shape. The model of the model data 32 is generated by performing machine learning on the learning data 30.

[0021] In the present embodiment, the learning image and the detection target image are images obtained by imaging a cross-section of a semiconductor device with a scanning electron microscope. The semiconductor device is formed on a substrate such as a semiconductor wafer, for example. The learning image and the detection target image are acquired by imaging a cross-section of the substrate on which the semiconductor device is formed with a scanning electron microscope. In the present embodiment, recesses such as trenches and holes are detected as specific shapes from the detection target image.

[0022] FIG. 2 is a diagram showing an example of an image of a cross-section of a substrate according to the embodiment. FIG. 2 is an image obtained by imaging a cross-section of a semiconductor device in which trenches and holes are formed with a scanning electron microscope. The horizontal direction of the image is defined as the x direction, and the vertical direction of the image is defined as the y direction. In the image shown in FIG. 2, a plurality of recesses 50 recessed in the y direction are formed side by side in the x direction. The recess 50 is, for example, a cross-section of a trench or a hole formed in the semiconductor device.

[0023] The learning data 30 stores a plurality of sets of an image of a cross-section of a substrate used as a learning image and information regarding the recess 50 included in the image.

[0024] The image data 31 stores an image of a cross-section of a substrate that is a detection target of a profile.

[0025] Returning to Figure 1, the control unit 24 is a device that controls the information processing device 10. The control unit 24 can be an electronic circuit such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), or GPU (Graphics Processing Unit), or an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). The control unit 24 has an internal memory for storing programs that define various processing procedures and control data, and executes various processes using these.

[0026] The control unit 24 functions as various processing units through the operation of various programs. For example, the control unit 24 includes an operation reception unit 40, a learning unit 41, a detection unit 42, a measurement unit 43, and an output unit 44.

[0027] The operation reception unit 40 accepts various operations. For example, the operation reception unit 40 displays an operation screen on the display unit 21 and accepts various operations on the operation screen from the input unit 22.

[0028] The information processing device 10 generates model data 32 by performing machine learning and stores the model data 32 in the storage unit 23, thereby enabling profile detection.

[0029] For example, the operation reception unit 40 receives instructions from the operation screen to specify training data 30 to be used for machine learning and to start model generation. Also, for example, the operation reception unit 40 receives instructions from the operation screen to specify image data 31 to be targeted for profile detection. The operation reception unit 40 reads the specified image data 31 from the storage unit 23 and displays the image of the read image data 31 on the display unit 21. The operation reception unit 40 receives instructions from the operation screen to start profile detection.

[0030] For example, a process engineer or manager specifies the training data 30 to be used for machine learning and instructs the start of model generation. Also, when a process engineer is determining the suitability of a recipe, they specify the image data 31 of the cross-section of the semiconductor device on which the substrate processing for the recipe to be determined has been performed from the operation screen. Then, the process engineer instructs the start of profile detection from the operation screen.

[0031] The learning unit 41 performs machine learning on the specified training data 30 and generates a model for detecting specific shapes contained in the image. In this embodiment, the learning unit 41 generates a model for detecting recesses 50 contained in the image. Any machine learning method can be used as long as it can produce a model capable of detecting specific shapes. Examples of machine learning methods include image segmentation methods such as U-net.

[0032] The learning unit 41 generates multiple models used to detect specific shapes contained in an image. For example, the learning unit 41 generates a contour detection model to detect the contours of specific shapes contained in an image. The learning unit 41 also generates a boundary detection model used to detect the boundaries of films contained in an image.

[0033] Here, we will explain the models generated by machine learning. First, we will explain the contour detection model. When generating a contour detection model, the training data 30 stores multiple data sets that associate images containing the recesses 50 shown in Figure 2 with information about the contours of the recesses 50 contained in those images. For example, as information about the contours of the recesses 50, a binarized image obtained by binarizing the image containing the recesses 50 is stored.

[0034] Figure 3 shows an example of a binarized image according to the embodiment. Figure 3 is a binarized image obtained by binarizing the image in Figure 2, where the portion of the film with the substrate and recess 50 is formed is assigned a first value (e.g., 0), and the empty space is assigned a second value (e.g., 1). In Figure 3, the portion of the binarized image with the first value is shown in black, and the portion with the second value is shown in white. For example, the training data 30 stores multiple images, each containing a recess 50 and its binarized version.

[0035] The learning unit 41 reads multiple data sets that associate images containing recesses 50 stored in the training data 30 with information about the contours of the recesses 50 contained in those images, and generates a contour detection model using machine learning. For example, the learning unit 41 generates a contour detection model using machine learning from multiple data sets that associate images containing recesses 50 shown in Figure 2 with binarized images of images containing recesses 50 shown in Figure 3. The generated contour detection model takes an image containing a recess 50 as input and performs calculations to output information about the contours of the recesses 50. For example, the contour detection model outputs a binarized image of an image containing a recess 50.

[0036] Next, the boundary detection model will be described. When generating the boundary detection model, the training data 30 stores data for each image containing the recess 50 shown in Figure 2, associating the image of a predetermined size region with information on whether or not that region contains the membrane boundary. For example, to indicate whether or not the membrane boundary is included, 1 is stored if the membrane boundary is included, and 0 is stored if the membrane boundary is not included.

[0037] The learning unit 41 reads data that associates images of regions obtained by dividing each image containing the recess 50 stored in the training data 30 into predetermined size groups with information on whether or not the image of the region contains the boundary of the film, and generates a boundary detection model using machine learning.

[0038] Figure 4 shows an example of the process for generating a boundary detection model according to the embodiment. For example, for each image containing a recess 50, an image of a predetermined size region of the image is randomly extracted, and data is created that associates the image of that region with information on whether or not that region contains the membrane boundary. In Figure 4, each image containing a recess 50 is divided into predetermined size regions, and the images of each region are shown as patch images 60. In addition, whether or not each patch image 60 contains the membrane boundary is shown as a label 61. Label 61 stores 1 if the corresponding patch image 60 contains the membrane boundary, and 0 if it does not. The training data 30 stores data that associates each patch image 60 with information on whether or not that patch image 60 contains the membrane boundary.

[0039] The learning unit 41 reads the patch images 60 stored in the training data 30 and the values ​​of the labels 61 corresponding to the patch images 60, and generates a boundary detection model using machine learning. The generated boundary detection model takes an image of a predetermined size as input and performs calculations to output information on whether or not the image contains the membrane boundary. For example, the boundary detection model outputs 1 if it is estimated that the image contains the membrane boundary, and 0 if it is estimated that the image does not contain the membrane boundary.

[0040] The learning unit 41 stores the data of each generated model in the model data 32. For example, the learning unit 41 stores the data of the generated contour detection model and boundary detection model in the model data 32.

[0041] When the detection unit 42 is instructed to start profile detection, it uses the model stored in the model data 32 to detect a specific shape from the image data 31. For example, the detection unit 42 uses the contour detection model and boundary detection model stored in the model data 32 to detect a recess 50 from the image data 31.

[0042] The detection unit 42 includes a contour detection unit 42a, a region detection unit 42b, and a boundary detection unit 42c.

[0043] The contour detection unit 42a uses the contour detection model stored in the model data 32 to detect the contour of the recess 50 contained in the specified image data 31. For example, the contour detection unit 42a inputs the image data 31 to the contour detection model and performs calculations. The contour detection model outputs a binarized image of the input image data 31. For example, if the contour detection model is input to an image containing the recess 50 shown in Figure 2, it outputs a binarized image of the image containing the recess 50 shown in Figure 3. The contour detection unit 42a detects the contour of the recess 50 from the binarized image output by the contour detection model. For example, the contour detection unit 42a detects the boundary portion where the pixel value changes between adjacent pixels in the binarized image as a contour. For example, the contour detection unit 42a generates an image from the binarized image by increasing or decreasing the black region of the boundary portion by 1 pixel. Then, the contour detection unit 42a calculates a difference image for each pixel at the corresponding position, which is the difference between the original binarized image and the image with the black region of the boundary portion increased or decreased by 1 pixel. Because the boundary area is increased or decreased by one pixel, the difference image retains only the boundary area as a black region. The contour detection unit 42a detects the black region of the difference image as the contour of the recess 50.

[0044] The region detection unit 42b detects the region of each recess 50 in the specified image data 31. For example, the region detection unit 42b uses the contour detection result from the contour detection unit 42a to detect the region of each recess 50 in the image.

[0045] Figure 5 is a diagram illustrating an example of the process for detecting the region of each recess 50 in an image according to the embodiment. Figure 5 shows an image in which the contours of the recesses 50 have been detected. The horizontal direction of the image is defined as the x-direction, and the vertical direction of the image is defined as the y-direction. For example, the region detection unit 42b identifies the range in the y-direction of the image where the contours of the recesses 50 detected by the contour detection unit 42a exist. For example, the region detection unit 42b obtains the minimum and maximum values ​​in the y-direction from the coordinates of each pixel constituting the contour of the recess 50, and identifies the range between the minimum and maximum values ​​as the range in the y-direction that includes multiple recesses 50. In Figure 5, the range including the recesses 50 in the y-direction of the image is shown as the Y Range.

[0046] The region detection unit 42b detects the boundaries of each recess 50 region in the x-direction from a specified range of the image. For example, the region detection unit 42b calculates the average value of the brightness of each pixel in the y-direction for each position in the x-direction of the image from a specified range of the image. Based on the calculated average value for each position in the x-direction, the region detection unit 42b detects the regions of each recess 50 from a specified range of the image.

[0047] For example, the region detection unit 42b extracts the Y Range range in the identified y direction of the image, and calculates the average value of the brightness of each pixel in the y direction for each position in the x direction of the image from the image within the extracted Y Range range. The region detection unit 42b arranges the average values ​​for each position in the x direction in order of position in the x direction to obtain a profile of the average values. Figure 5 shows the profile AP of the average values ​​obtained by arranging the average values ​​for each position in the x direction in order of position in the x direction. The region detection unit 42b binarizes each value of the profile AP of the average values ​​in the x direction. For example, the region detection unit 42b calculates the average value of the profile AP of the average values ​​in the x direction, and uses the calculated average value as a threshold to binarize each value of the profile AP. For example, if the value of the profile AP of the average values ​​is greater than or equal to the threshold, the region detection unit 42b uses it as the first value, and if the value of the profile AP of the average values ​​is less than the threshold, it uses it as the second value, and binarizes each value of the profile AP. Figure 5 shows the binarized profile BP, where each value of the profile AP is set to "0" if it is greater than or equal to the threshold (average value), and to "1" if it is less than the threshold. The region detection unit 42b detects the position of the center of each continuous portion where the second value is consecutive in the binarized profile BP as the pattern boundary of the recess 50 in the x direction. For example, the region detection unit 42b detects the position of the center of each continuous portion where "1" is consecutive in the binarized profile BP as the pattern boundary of the recess 50 in the x direction. For images within the Y Range, the region detection unit 42b detects the regions between the detected pattern boundaries as regions of the recess 50. In Figure 5, each region of the recess 50 detected from the image in which the contour of the recess 50 has been detected is shown as a rectangle S1.

[0048] The boundary detection unit 42c uses the boundary detection model stored in the model data 32 to detect the boundaries of the film contained in the specified image data 31. For example, for the image of the specified image data 31, the boundary detection unit 42c uses the boundary detection model to derive information on whether or not each region of the image contains a film boundary, for each predetermined size used to generate the boundary detection model. The boundary detection unit 42c then uses the derived information for each region to detect the boundaries of the film contained in the specified image data 31.

[0049] Figure 6 is a diagram illustrating an example of the process for detecting the boundary of a membrane contained in an image according to the embodiment. Figure 6 shows an image of the specified image data 31. For example, the boundary detection unit 42c divides the image of the specified image data 31 into patch images 60, and inputs the divided patch images 60 into a boundary detection model for calculation. The boundary detection model outputs information on whether the input patch image 60 contains the boundary of a membrane. For example, the boundary detection model outputs 1 if it is estimated to contain the boundary of a membrane, and outputs 0 if it is estimated not to contain the boundary of a membrane. For each position in the y-direction of the image data 31, the boundary detection unit 42c calculates the average value of the output values ​​of the boundary detection model for each patch image 60 at the same position in the y-direction. The boundary detection unit 42c arranges the average values ​​for each position in the y-direction in order of position in the y-direction to obtain a profile of the average value. Based on the average value at each position in the y-direction, the boundary detection unit 42c detects the position of the membrane boundary. The average value approaches 1 at positions containing the boundary of a membrane, and represents the probability that it is a boundary. The boundary detection unit 42c detects a position in the y-direction where the average value is close to 1 as the boundary of the film. For example, the boundary detection unit 42c detects a position where the average value is greater than or equal to a predetermined threshold (e.g., 0.8) as the boundary of the film.

[0050] Figure 7 shows an example of the detection result of the film boundary according to the embodiment. Figure 7 shows an example of an image of a cross-section of the substrate. In Figure 7, the films constituting the side walls of the recess 50 are shown with different patterns, and the detected film boundaries in the y direction are indicated by lines L1 and L2. The film above the recess 50 in the image is, for example, a mask. The boundary detection unit 42c is able to detect the film boundary. By detecting the film boundary in this way, it can be used for automatic adjustment of the rotational misalignment of the image. For example, the boundary detection unit 42c may use the film boundary line L2 to perform rotational correction of the image so that line L2 becomes horizontal. This corrects the rotational misalignment of the image and makes it easier to understand the positional relationship and film thickness of the films from the image.

[0051] The information processing device 10 according to this embodiment can automatically detect the extent of the recess 50 in the image and the contour of the recess 50, thereby making dimensional measurement more efficient.

[0052] The measurement unit 43 measures dimensions. For example, the operation reception unit 40 displays an image of the contour of the recess 50 detected by the contour detection unit 42a on the display unit 21 and receives a specification from the input unit 22 for the position of the contour of the recess 50 to be measured. The measurement unit 43 measures the dimensions of the CD or other dimensions of the recess 50 at the specified contour position.

[0053] The measurement unit 43 may automatically measure dimensions such as CD of the recess 50 at predetermined positions on the contour without requiring a specified position. The positions for measuring dimensions may be set in advance, or they may be set based on the detection results of the boundary detection unit 42c or the contour detection unit 42a. For example, the measurement unit 43 may measure dimensions such as CD at the film boundary from the contour of each recess 50 detected by the contour detection unit 42a at the height of the film boundary detected by the boundary detection unit 42c. Alternatively, the measurement unit 43 may automatically measure dimensions such as CD at predetermined positions on the recess 50, such as the top of the recess 50, the middle of the side wall inside the recess 50, or the bottom of the recess 50. Furthermore, the measurement unit 43 may measure dimensions such as CD at various positions in the y-direction from the edge profile of the contour of each recess 50 detected by the contour detection unit 42a.

[0054] The output unit 44 outputs shape information of a specific shape detected by the detection unit 42. For example, the output unit 44 displays the contour of the recess 50, the area of ​​the recess 50, and the boundary of the film detected by the detection unit 42 on the display unit 21, along with the image of the specified image data 31. The output unit 44 also outputs the measurement results measured by the measurement unit 43 from the specific shape detected by the detection unit 42. For example, the output unit 44 displays the measured dimensions along with the measurement position on the display unit 21. The output unit 44 may store the shape information of the specific shape detected by the detection unit 42 and the measurement result data in the storage unit 23, or it may transmit them to another device via the communication I / F unit 20.

[0055] The output unit 44 may output only shape information that best represents the features of interest by selecting a target region from multiple regions containing a specific shape on the image. For example, the input unit 22 accepts the selection of a region of interest from multiple regions of a specific shape displayed on the display unit 21. For example, the input unit 22 accepts the selection of a region of interest of a recess 50 from multiple regions of recesses 50 displayed on the display unit 21. The output unit 44 may output only the shape information of the selected region of recess 50. For example, the output unit 44 may output only the feature quantities of the recess 50 that represent the features of the recess 50 of interest, such as the dimensions of the selected recess 50, such as the CD. This allows the process engineer to efficiently grasp the feature quantities of the recess 50 of interest.

[0056] Furthermore, the output unit 44 may perform profile selection, such as outlier removal and maximum CD selection, on the measurement results measured by the measurement unit 43, and output the selected measurement results. For example, there may be areas of recesses 50 that are unsuitable for measurement due to factors such as collapse of the side walls. By applying an outlier detection method such as the 3σ rule to the TOP CD of all measured recesses 50 areas, it is possible to remove abnormal values ​​due to collapse of the side walls, etc. The output unit 44 may apply an outlier detection method such as the 3σ rule to the TOP CD of all measured recesses 50 areas to remove abnormal values ​​and output the selected measurement results. Also, for example, there may be cases where it is desired to select the recess 50 with the largest CD as the target of measurement. The output unit 44 may output the shape information and measurement results of the recess 50 with the largest CD. For example, the output unit 44 may output the maximum value among the TOP CDs of recesses 50 that were not removed by the above outlier detection. Alternatively, the output unit 44 may select based on the median or the score of an unsupervised learning model such as the Local Outlier Factor, rather than the maximum value. Furthermore, the output unit 44 may not only select and output the shape information and measurement results of one recess 50, but may also calculate and output the average or median values ​​of the shape information and measurement results of multiple recesses 50.

[0057] [Processing flow] Next, the flow of the profile detection method according to the embodiment will be described. The information processing device 10 according to the embodiment performs the profile detection method by executing a profile detection program. Figure 8 is a schematic diagram showing an example of the flow of the profile detection method according to the embodiment.

[0058] When a process engineer determines whether a recipe is suitable, they specify image data 31 and instruct the system to start detecting the profile.

[0059] When the detection unit 42 is instructed to start profile detection by specifying image data 31, it uses the model stored in the model data 32 to detect a specific shape from the image of the specified image data 31. For example, the contour detection unit 42a uses the contour detection model stored in the model data 32 to detect the contour of a recess 50 included in the image of the specified image data 31 (step S10). The region detection unit 42b uses the contour detection result by the contour detection unit 42a to detect the region of each recess 50 in the image (step S11). The boundary detection unit 42c uses the boundary detection model stored in the model data 32 to detect the boundary of a film included in the image of the specified image data 31 (step S12). Note that the processing in steps S10 and S11 and the processing in step S12 may be performed in reverse order or in parallel.

[0060] The measuring unit 43 measures dimensions (step S13). For example, the measuring unit 43 measures the dimensions of the CD of the recess 50 at a predetermined position on the contour.

[0061] The output unit 44 outputs shape information of a specific shape detected by the detection unit 42 (step S14). For example, the output unit 44 displays the contour of the recess 50, the area of ​​the recess 50, and the boundary of the film detected by the detection unit 42 on the display unit 21 along with the image of the specified image data 31. The output unit 44 also outputs the measurement results of the measurement unit 43. The output unit 44 may perform profile selection, such as outlier removal and maximum CD selection, on the measurement results of the measurement unit 43 and output the selected measurement results.

[0062] The information processing device 10 according to this embodiment can measure the dimensions of recesses 50 in an image in this way, thereby streamlining dimension measurement. As a result, the time required for dimension measurement can be reduced. Furthermore, since the information processing device 10 can detect the contour that will be used to measure dimensions, human-dependent errors in the measured dimensions can be reduced. In addition, the information processing device 10 can efficiently measure the dimensions of a large number of recesses 50. For example, by automatically measuring the dimensions of each recess 50 included in an image, a large number of measurement values ​​can be collected for use in data analysis. Furthermore, by automatically measuring the dimensions of each recess 50 included in an image and analyzing the measured dimensions of each recess 50, abnormal recesses 50 can be detected.

[0063] In the above embodiment, the case in which the region detection unit 42b detects the region of the recess 50 using the contour detection result by the contour detection unit 42a was described as an example. However, it is not limited to this. The region detection unit 42b may detect the region of the recess 50 without using the contour detection result by the contour detection unit 42a. For example, the learning unit 41 generates a region detection model by machine learning from a plurality of data that associates an image containing the recess 50 with an image showing the region of the recess 50, and stores the region detection model in the model data 32. The region detection unit 42b may detect the region of the recess 50 using the region detection model stored in the model data 32.

[0064] Furthermore, the contour detection model described in the above embodiment is merely an example and is not limited thereto. Any contour detection model is acceptable as long as it can detect contours. The learning unit 41 may generate a contour detection model by machine learning from a plurality of data sets that associate an image containing a recess 50 with a binarized image obtained by binarizing the contour portion of the image containing the recess 50. The contour detection unit 42a may input the image of the specified image data 31 into the contour detection model, perform calculations, and detect the contour of the recess 50 from the binarized image output from the contour detection model. The boundary detection model and region detection model are also merely examples and are not limited thereto. Any boundary detection model is acceptable as long as it can detect contours. Any region detection model is acceptable as long as it can detect the region of the recess 50.

[0065] Furthermore, the above embodiment described an example in which the contour of the recess 50 and the boundary between the region of the recess 50 and the film are detected individually. However, it is not limited to this. Any two or all of the contour of the recess 50 and the boundary between the region of the recess 50 and the film may be detected using one model. For example, the contour of the recess 50 and the boundary between the film may be detected using one model. In this case, for example, the learning unit 41 may generate a contour boundary detection model by machine learning from multiple data sets that associate an image containing the recess 50 with binarized images obtained by binarizing the contour portion and the boundary portion of the film in the image containing the recess 50. The detection unit 42 may input the specified image data 31 to the generated contour boundary detection model and perform calculations, and detect the contour of the recess 50 and the boundary between the film from the binarized image output from the contour boundary detection model. Figure 9 is a schematic diagram showing another example of the flow of the profile detection method according to the embodiment. Figure 9 shows the case in which the contour of the recess 50 and the boundary between the film are detected simultaneously. In Figure 9, instead of steps S10 and S12 in Figure 8, the detection unit 42 inputs the specified image data 31 into the contour boundary detection model and performs calculations, detecting the contour of the recess 50 and the boundary of the film from the binarized image output from the contour boundary detection model (step S20).

[0066] Furthermore, the training data 30 and image data 31 may contain out-of-focus images. Therefore, the learning unit 41 may remove out-of-focus images before training. The detection unit 42 may remove out-of-focus images to detect specific shapes. For example, the learning unit 41 and the detection unit 42 apply a Fast Fourier Transform (FFT) to the entire image and determine that an image is out of focus if the magnitude of the high-frequency power is below a threshold.

[0067] As described above, the profile detection method according to the embodiment includes a detection step (steps S10 to S12) and an output step (step S14). The detection step uses a model that has learned a training image containing a specific shape and information about the specific shape contained in the training image to detect a target image containing a specific shape from the target image containing the specific shape. The output step outputs the shape information of the detected specific shape. As a result, the profile detection method according to the embodiment can efficiently detect a specific shape contained in a target image. As a result, the profile detection method according to the embodiment can efficiently measure the dimensions of a specific shape. For example, the profile detection method according to the embodiment can efficiently detect recesses 50 contained in an image and reduce the time required to measure the dimensions of the recesses 50. In addition, the profile detection method according to the embodiment can reduce human-dependent errors that occur in the dimensions to be measured. Furthermore, the profile detection method according to the embodiment can efficiently measure the dimensions of a large number of recesses 50.

[0068] Furthermore, the detection process includes a contour detection process (step S10), a boundary detection process (step S12), and a region detection process (step S11). The contour detection process detects the contours of specific shapes contained in the target image from the target image. The boundary detection process detects the boundaries of films contained in the target image from the target image. The region detection process detects regions of specific shapes contained in the target image from the target image. At least one of the contour detection process, region detection process, and boundary detection process is performed using a model. As a result, the profile detection method according to the embodiment can efficiently measure the dimensions of a specific shape from the detected contours of the specific shape, the boundaries of the films, and the regions of the specific shape.

[0069] Furthermore, the training image and the image to be detected are images of a cross-section of a semiconductor substrate in which multiple recesses 50, which represent via or trench cross-sections, are arranged as a specific shape. As a result, the profile detection method according to the embodiment can efficiently detect the recesses 50 included in the image to be detected. As a result, the profile detection method according to the embodiment can efficiently measure the dimensions of the recesses 50.

[0070] Furthermore, the profile detection method according to the embodiment further includes a measurement step (step S13). The measurement step measures the dimensions of the detected specific shape. This allows the profile detection method according to the embodiment to efficiently measure the dimensions of the specific shape.

[0071] Furthermore, the model learns information about training images and the contours of specific shapes contained in those training images. The contour detection step uses the model to detect the contours of specific shapes contained in the target image. As a result, the profile detection method according to this embodiment can accurately detect the contours of specific shapes.

[0072] Furthermore, the model learns information about each region of a predetermined size in the training image, including the image of that region and whether or not that region contains the membrane boundary. In the boundary detection step, for each region of the target image, the model is used to derive information about whether or not the image of each region contains the membrane boundary, and the membrane boundaries included in the target image are detected from the derived information for each region. As a result, the profile detection method according to the embodiment can detect membrane boundaries with high accuracy.

[0073] Furthermore, the region detection step identifies the range of a specific shape from the contour of the specific shape detected by the contour detection step, in one direction and in the intersecting direction to that direction of the image to be detected, and detects the identified range as the region of the specific shape. As a result, the profile detection method according to the embodiment can detect the region of a specific shape with high accuracy.

[0074] Furthermore, the output process selects a detection target region from multiple regions containing a specific shape on the image, thereby outputting only the shape information representing the feature of interest. As a result, the profile detection method according to this embodiment can output only the shape information representing the feature of interest once the detection target region is selected.

[0075] Furthermore, the output process selects a detection target region from multiple regions including the recess shape 50 on the image, thereby outputting only the recess feature quantities that represent the feature of interest. As a result, the profile detection method according to this embodiment can output only the feature quantities of the recess 50 that represent the feature of interest once the detection target region is selected.

[0076] While embodiments have been described above, it should be understood that the embodiments disclosed herein are illustrative and not restrictive in all respects. Indeed, the embodiments described above can be embodied in a variety of forms. Furthermore, the embodiments described above may be omitted, replaced, or modified in various ways without departing from the scope and spirit of the claims.

[0077] For example, the above embodiment described an example in which contour detection (step S10), region detection (step S11), and boundary detection (step S12) are performed in order. However, it is not limited to this. The order of contour detection, region detection, and boundary detection may be different. For example, they may be performed in the order of boundary detection, contour detection, and region detection.

[0078] Furthermore, the above embodiment was described using the example of measuring the dimensions of recesses in a semiconductor device formed on a substrate such as a semiconductor wafer. However, it is not limited to this. The substrate may be any substrate, such as a glass substrate. The profile detection method according to the embodiment may be applied to measuring the dimensions of recesses in any substrate. For example, the profile detection method according to the embodiment may be applied to measuring the dimensions of recesses formed on a substrate for an FPD.

[0079] It should be noted that the embodiments disclosed herein are illustrative and not restrictive in all respects. Indeed, the embodiments described above can be embodied in a variety of forms. Furthermore, the embodiments described above may be omitted, replaced, or modified in various ways without departing from the scope and spirit of the appended claims. [Explanation of Symbols]

[0080] 10 Information Processing Devices 20 Communication I / F Section 21 Display section 22 Input section 23 Memory section 24 Control Unit 30 training data 31 Image data 32 Model Data 40 Operation reception unit 41 Learning Department 42 Detection unit 42a Contour detection unit 42b Region detection unit 42c Boundary detection unit 43 Measurement Unit 44 Output section 50 recesses

Claims

1. A detection step of detecting the specific shape contained in a target image containing the specific shape using a training image containing the specific shape and a model that has learned information about the specific shape contained in the training image, An output step that outputs shape information of the detected specific shape, It has, The aforementioned detection step is A contour detection step for detecting the contour of a specific shape contained in the target image from the target image, A boundary detection step for detecting the boundaries of a film included in the target image from the target image, A region detection step for detecting a region of a specific shape included in the target image from the target image, Includes, At least one of the contour detection step and the boundary detection step is performed using the model, The region detection step identifies the range of the specific shape from the contour of the specific shape detected by the contour detection step in one direction and in the direction intersecting the one direction of the detection target image, and detects the identified range as the region of the specific shape. Profile detection method.

2. The aforementioned training image and the aforementioned detection target image are images of a cross-section of a semiconductor substrate in which multiple recesses representing vias or trenches are arranged, as the specific shape. The profile detection method according to claim 1.

3. The method further includes a measurement step for measuring the dimensions of the detected specific shape, The output process outputs the measured dimensions. The profile detection method according to claim 1.

4. The aforementioned model learns information about the training images and the contours of the specific shapes contained in the training images. The contour detection step uses the model to detect contours of the specific shape contained in the target image. The profile detection method according to claim 1.

5. The model learns information about whether the image of a given region contains the boundary of a membrane, for each region of a predetermined size in the training image. The boundary detection step involves using the model to derive information on whether the image of each region of the target image contains the boundary of the film, for each region of the target image, and then detecting the boundary of the film contained in the target image from the derived information for each region. The profile detection method according to claim 1.

6. The model outputs different output values ​​depending on whether the membrane boundary is included or not. The boundary detection step involves using the model to derive an output value for each region of the target image of the predetermined size, indicating whether the image of each region includes the boundary of the film, calculating the average value of the model's output values ​​at the same vertical position for each vertical position of the target image of the target image, and detecting the vertical position closest to the output value when the average value includes the boundary of the film as the position of the film boundary. The profile detection method according to claim 5.

7. The output step selects a target region from multiple regions containing the specific shape on the image, thereby outputting only the shape information representing the feature of interest. The profile detection method according to claim 1.

8. The output step selects the region to be detected from multiple regions including the recess shape on the image, and outputs only the recess feature quantities that represent the feature of interest. The profile detection method according to claim 1.

9. A detection unit that uses a training image containing a specific shape and a model that has learned information about the specific shape contained in the training image to detect the specific shape contained in the target image containing the specific shape, An output unit that outputs shape information of the specific shape detected by the detection unit, It has, The detection unit is A contour detection unit that detects the contour of the specific shape contained in the target image from the target image, A boundary detection unit that detects the boundaries of a film included in the target image from the target image, A region detection unit that detects a region of a specific shape included in the target image from the target image, Includes, At least one of the contour detection unit and the boundary detection unit performs detection using the model, The region detection unit identifies the range of the specific shape from the contour of the specific shape detected by the contour detection unit in one direction and in the direction intersecting the one direction of the detection target image, and detects the identified range as the region of the specific shape. Profile detection device.

Images