Dimension measuring devices, semiconductor manufacturing equipment, and semiconductor device manufacturing systems

Abstract

The present invention provides a dimension measuring device that can automatically correct contour offsets without operator judgment. The dimension measuring device comprises: a model learning unit (103) that acquires a learning cross-sectional image and learning labels attached to different regions of the learning cross-sectional image, and generates a deep learning model for image region segmentation using the learning cross-sectional image and the learning labels; a model estimation unit (104) that applies the model to a newly input object image and attaches labels to each independent region; a contour correction unit (105) that uses the object image and the labels attached by the model estimation unit to detect the contours of each region, sets representative points for the contours of the regions, moves each representative point according to a movement rule, and repeatedly moves the representative points until the correction completion condition is met; and a dimension measuring unit (106) that uses the contour corrected by the contour correction unit to measure the dimensions of the device's cross-sectional structure.

Description

Technical Field

[0001] The present invention relates to a dimension measuring device, a semiconductor manufacturing apparatus, and a semiconductor device manufacturing system capable of automatically measuring the dimensions of a device structure based on a cross-sectional image of the device obtained by an observation device. Background Technology

[0002] In semiconductor manufacturing process development, it is necessary to obtain the shape and dimensions of semiconductor devices from cross-sectional images, leading to the development of automated dimensional measurement technology. In the dimensional measurement process, to determine the measurement start and end points, a process is involved in detecting the contour of the measurement object area. This contour detection method includes methods that manually draw boundary lines, edge detection methods based on local brightness variations, and detection methods utilizing deep learning.

[0003] In any contour detection method, the detected contour may deviate from the actual contour. Therefore, the dimensions of the device structure measured from the profilometer, which includes this deviation, will have errors. To reduce this measurement error, a process of correcting the detected contour is considered effective.

[0004] Patent Document 1 discloses a technique in which movable points and reference points are set for the generated contour in a medical image of the object being tested, and the movable points are corrected according to the positional relationship between the movable points and the reference points, thereby correcting the contour.

[0005] Prior art literature

[0006] Patent documents

[0007] Patent Document 1: Japanese Patent Application Publication No. 2013-5983 Summary of the Invention

[0008] The problem that the invention aims to solve

[0009] If the technique disclosed in Patent Document 1 is used, the contour can be easily corrected by manually making adjustments bit by bit. However, in this technique, an initial setting of the object region is required for contour correction. In medical applications, since the image contains only one object region, there are cases where an initial setting is not required. However, in cases where there are multiple different object regions, such as a cross-sectional image of a semiconductor device, correction cannot be performed.

[0010] Furthermore, in the technology disclosed in Patent Document 1, the contour is corrected by moving a movable point set on the contour. However, the method of moving the movable point, such as the direction and distance of the movement, requires operation by an operator. Therefore, there is a problem that the effect of contour correction is considered to depend on the operator's subjective judgment, and if the correction process takes time, there will be errors due to the operator's experience and subjective judgment.

[0011] The purpose of this invention is to solve the above-mentioned problems and provide a dimension measuring device, a semiconductor manufacturing apparatus, and a semiconductor device manufacturing system that can automatically correct contour offsets without operator judgment during the correction process.

[0012] Methods for solving problems

[0013] To achieve the above objectives, the present invention provides a size measuring device that uses an image to measure the size of a measured object. In the size measuring device, based on a given point in the contour of the measured object, the contour is repeatedly moved until a given necessary condition is met. The size of the measured object is measured using the repeatedly moved contour. The given necessary condition is a necessary condition for correcting the offset of the contour estimated using machine learning.

[0014] In the dimension measuring device of the present invention, the movement of the contour is performed after the contour is estimated by machine learning.

[0015] Invention Effects

[0016] By correcting the contours detected from the cross-sectional images, the measurement accuracy of the automatic dimensioning process can be improved. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating a structural example of a semiconductor manufacturing apparatus according to various embodiments.

[0018] Figure 2 This is a diagram illustrating an example of the dimensional measurement process involved in Embodiment 1.

[0019] Figure 3 This is a diagram illustrating an example of the contour correction effect involved in Embodiment 1.

[0020] Figure 4 This is a diagram illustrating an example of contour correction processing involved in Embodiment 1.

[0021] Figure 5 This is a diagram illustrating an example of the contour correction process involved in Embodiment 1.

[0022] Figure 6 This is a diagram illustrating an example of the contour detection processing flow involved in Embodiment 1.

[0023] Figure 7 This is a diagram illustrating an example of the representative point movement processing flow involved in Embodiment 1.

[0024] Figure 8 This is a diagram illustrating an example of the luminance profile involved in Embodiment 1.

[0025] Figure 9 This is a diagram illustrating an example of the process of dividing the complex region involved in Embodiment 2 into simple shapes.

[0026] Figure 10 This is a diagram illustrating an example of a dimensional measurement process involving object detection according to Embodiment 2. Detailed Implementation

[0027] An embodiment of the present invention is a plasma processing apparatus having a dimensional measuring device. The dimensional measuring device of the plasma processing apparatus is a general-purpose computer equipped with a processor and memory; it can be an installation of software that processes data according to a program, or it can be a dedicated hardware installation rather than a general-purpose computer.

[0028] Alternatively, it can be installed using dedicated hardware assembled on a computer, combining software installation and hardware installation. The dimensional measuring device can be externally connected, or it can be externally connected as a module that can be used in conjunction with other data processing. Various embodiments will be described below using the accompanying drawings.

[0029] Furthermore, this specification utilizes movement rules, which include multiple rules. A rule includes a brightness distribution, a condition for determining the movement direction corresponding to the brightness distribution, and a condition for determining the movement distance corresponding to the brightness distribution. These movement rules can be preset based on the relationship between the SEM image of the dimensional measuring device and the material and shape.

[0030] Brightness distribution is information used to select an appropriate rule from multiple rules. The direction of movement determines whether the representative point moves towards or away from the center of gravity, and the distance of movement determines the amount by which the representative point is moved, corresponding to several pixels. When no movement is needed, it becomes 0.

[0031] Example 1

[0032] Example 1 is an example of a plasma processing apparatus as a semiconductor manufacturing apparatus equipped with a dimensional measuring device.

[0033] Figure 1 The plasma processing apparatus 1 shown includes a size measuring device 10, an input / output device 11, an input device 12, an output device 13, a processing condition search device 14, a processing device 15, and an evaluation device 16.

[0034] In the plasma processing apparatus 1, the target processing result, such as the processing shape, and the parameters used in the processing apparatus are selected, input from the input device 12, and the optimal processing conditions for the target processing shape are output.

[0035] The processing condition search device 14 is a device that receives the target processing shape from the input device 12 and outputs the optimal processing conditions to the output device 13.

[0036] The input device 12 includes an input interface such as a GUI and a storage medium reading device such as a card reader, and inputs various data to the processing condition search device 14. Furthermore, it receives dimensional measurement values ​​not only from the user but also from the input / output device 11 and inputs them to the processing condition search device 14. The input device 12 may include, for example, a keyboard, mouse, touch panel, and storage medium reading device.

[0037] The output device 13 displays to the user the processing conditions transmitted from the processing condition search device 14 as the optimal processing conditions. The means of display includes showing to a monitor or writing to a document. The output device 13 may include, for example, a monitor, a printer, or a storage medium writing device.

[0038] Processing apparatus 15 is an apparatus for processing semiconductors or semiconductor devices containing semiconductors. The processing performed by the processing apparatus is not particularly limited. For example, it may include photolithography apparatus, film deposition apparatus, and patterning apparatus. Photolithography apparatus may include, for example, exposure apparatus, electron beam mapping apparatus, and X-ray mapping apparatus. Film deposition apparatus may include, for example, CVD (Chemical Vapor Deposition), PVD (Physical Vapor Deposition), evaporation apparatus, sputtering apparatus, and thermal oxidation apparatus. Patterning apparatus may include, for example, wet etching apparatus, dry etching apparatus, electron beam processing apparatus, and laser processing apparatus.

[0039] The processing device 15 processes the semiconductor or semiconductor device based on the processing conditions input from the processing condition search device and transmits the results to the evaluation device 16.

[0040] The evaluation device 16 captures a cross-section of the semiconductor or semiconductor device processed by the processing device 15, acquiring a cross-sectional image 17 as a processing result. The evaluation device 16 includes an SEM (Scanning Electron Microscope), a TEM (Transmission Electron Microscope), and a processing dimension measuring device using an optical monitor. Alternatively, a portion of the semiconductor or semiconductor device processed by the processing device 15 can be removed as a fragment and transported to the evaluation device 16 for measurement. The acquired cross-sectional image 17 is transmitted to the input / output device 11.

[0041] The dimension measuring device 10 that uses images to measure the size of an object is a dimension measuring device that repeatedly moves the contour based on a given point in the contour of the object until a given necessary condition is met, and uses the repeatedly moved contour to measure the size of the object. The given necessary condition is a necessary condition for correcting the offset of the contour estimated by machine learning.

[0042] In this dimensional measuring device, the movement of the contour is performed after the contour is estimated through machine learning. Then, the dimensions of the object being measured are measured using the contour estimated by machine learning by using the contour that has been moved repeatedly as learning data.

[0043] The device for measuring the dimensions of the cross-sectional structure of a device being measured includes: a model learning unit that acquires a learning cross-sectional image and learning labels attached to different regions of the learning cross-sectional image, and generates a model using the learning cross-sectional image and the learning labels; a model estimation unit that applies the model to an object image and attaches labels to each independent region; a contour correction unit that uses the object image and the labels attached by the model estimation unit to detect the contour of each region, sets representative points for the contour of the region, moves each representative point according to a movement rule, and repeatedly moves the contour until the correction completion condition is met; and a dimension measuring unit that measures the dimensions of the cross-sectional structure of the device using the contour corrected by the repeated movement by the contour correction unit.

[0044] like Figure 1 As shown, the dimensional measurement device 10 includes a central processing unit 101 and a database 102. The central processing unit 101 functions as a model learning unit, a model estimation unit, a contour correction unit, and a dimensional measurement unit for machine learning. The central processing unit 101 receives feature points and dimensional definitions, magnification, learning datasets, and profile images input via the input / output device 11, measures the given dimensions based on the profile images, and outputs them to the input / output device 11.

[0045] The input / output device 11 includes an input / output interface such as a GUI and a storage medium reading device such as a card reader, and inputs feature points and size definitions, magnification, and training datasets to the dimensional measuring device 10. Furthermore, it receives cross-sectional images from the evaluation device 16 and transmits them to the central processing unit 101. The input / output device 11 includes, for example, a keyboard, mouse, display, touch panel, and storage medium reading device. It also displays the size values ​​transmitted from the dimensional measuring device 10 to the user or directly transmits them to the input device 12. When displaying to the user, it does so through methods such as displaying on a monitor or writing to a document.

[0046] Figure 2This is a diagram showing an example of the dimension measurement process in the dimension measuring device 10. The process of the dimension measuring device 10 includes the processes of the model learning unit 103, the model estimation unit 104, the contour correction unit 105, the dimension measuring unit 106, etc.

[0047] The model learning unit 103 uses the input training SEM image and training labels to generate a deep learning model for image region segmentation and sends it to the model estimation unit 104. The training SEM image is an image prepared in advance by the user, such as a 1280×960 grayscale pixel matrix. The number of prepared images can be determined according to the deep learning algorithm and the construction of the deep network. The training labels are labels that the user has pre-attached to each pixel of the training SEM image, such as numbers like 0, 1, and 2.

[0048] The model learning unit 103 inputs the learning profile image and the learning label into the deep learning method for image segmentation to generate a deep learning model. The model estimation unit applies the model to the object image and adds labels.

[0049] For example, such as Figure 3 As shown, when three regions—vacuum 30, mask 31, and silicon 32—exist in the SEM image, they are labeled using the numbers 0 (vacuum), 1 (mask), and 2 (silicon), respectively. However, the number of independent regions in the SEM image is not limited to three; multiple regions of various types may exist. In this example, a 1280×960 label matrix is ​​generated for a single SEM image, and the storage and input / output format of the labels can follow the requirements of deep learning algorithms.

[0050] The model estimation unit 104 applies the learning model generated by the model learning unit 103 to the input SEM image, generates a label for the image (e.g., a 1280×960 digital matrix), and sends the label to the contour correction unit 105.

[0051] The contour correction unit 105 extracts the contour of the independent region from the label information, corrects the contour using the brightness information present in the SEM image, and sends the corrected contour to the dimension measurement unit 106.

[0052] At this time, a movement rule corresponding to at least one brightness distribution is pre-specified. The contour correction unit 105 compares the brightness distribution extracted from the straight line passing through the representative point and centroid of the movement with the pre-specified brightness distribution and applies the movement rule corresponding to the specified brightness distribution with the highest similarity. Furthermore, the contour correction unit extracts contours from the independent regions segmented by the model estimation unit and outputs them sequentially. The dimension measurement unit 106 uses the corrected contours to calculate and output dimensional information such as the width, height, and area of ​​the measurement object region.

[0053] use Figure 3 The contour correction effect is explained. Additionally, in Figure 3 Only a portion of the 1280×960 SEM image is shown. In this embodiment's SEM image, there are three regions: vacuum 30, mask 31, and silicon 32. Furthermore, regarding the mask type, there are multiple independent, unconnected regions. The white lines 311, 312, and 313 surrounding the three mask regions are contours extracted based on label information generated by the model estimation unit. The difference between the extracted contours and the actual contours is shown as an offset. Figure 3 As shown in the contour 311 before correction and the contour 314 after correction, the processing of the contour correction unit 105 is a process of correcting the contour offset.

[0054] In addition, Figure 3 In SEM images, although there are three types of regions, the number of regions is not limited to three; it can be n types of regions. That is, when there are n different types of regions in the object image, the system learns to attach n different numbers to each pixel of the object image using labels, thereby marking them.

[0055] Figure 4 This diagram illustrates an example of contour correction processing, where contour correction of a mask region is performed through a series of processes consisting of an initial contour, representative point setting, representative point movement, and movement completion. Specifically, a mask contour 40 is extracted based on the label information generated by the model estimation unit 104. This contour 40 is referred to as the initial contour. As shown in contour 40, the initial contour is offset from the actual mask contour displayed in gray. The necessity of processing the contour of such a separate region as the contour of a mask will be described later. Furthermore, the method for extracting the contour of this separate region will be described later.

[0056] Next, for contour correction, representative points 41 are first set for the initial contour. The contour during and after correction is formed by straight lines connecting the representative points 41. Figure 4 In the representative point setting, 12 representative points are shown as an example, but the actual number of representative points can be more than 12. Comparing the contour displayed using the lines between these representative points with the initial contour 40, fewer representative points result in a coarser contour. On the other hand, more representative points increase the correction processing time, and the positional relationships of adjacent representative points may become disordered. In actual processing, representative points can also be set at given intervals. The given interval can use an empirical value of 5 pixels or be adjusted by the user. Alternatively, a given percentage (e.g., 25%) of the total number of pixels in the contour can be randomly selected and set as representative points.

[0057] Next, to move the representative points, the centroid of the contour is calculated using the established representative points. The centroid is calculated by averaging the coordinates (x, y) of the representative points. The centroid is a crucial reference point for subsequent correction processing, and the centroid of each independent region must be calculated separately. Therefore, independent regions need to be extracted, requiring image region segmentation for this purpose. While there are many methods for directly detecting contours (edges) without image segmentation, this embodiment performs region segmentation.

[0058] like Figure 4 As shown in the diagram, contour correction is performed by moving each representative point 41. For example, P1 is moved along a straight line passing through representative point P1 and the centroid 42. The purpose of this movement is to correct the offset by bringing the representative point closer to the actual contour. Then, the representative points are moved sequentially. For example, all the representative points are moved in the order of P1, P2, P3, P4. To move all the representative points to the optimal position, in other words, to the actual contour, the contour movement process is repeated. The final contour 43, resulting from the completed movement, is displayed. Figure 4 The movement is in progress. The decision regarding the direction and distance of the representative point's movement will be described later.

[0059] Figure 5 This diagram illustrates an example of the contour correction process flow. The data input to the contour correction unit is the region segmentation result of the SEM image (S50). In this embodiment, the result includes a 1280×960 SEM image and a 1280×960 label matrix. The possible values ​​of the labels are related to... Figure 3 The corresponding numbers are 0 (background), 1 (mask), and 2 (silicon).

[0060] First, in the contour detection process, pixels belonging to the contour are detected by labels (S51). Here, for example, the contours of independent regions that function as mask regions are detected and processed. To extract the contours of multiple independent regions from a SEM image, correction processing is performed one by one. The method for extracting independent regions will be described later.

[0061] The contour correction unit calculates the centroid of the contour based on the coordinates of each representative point. It then selects a movement rule and determines the movement method of the representative points based on the positional relationship between the moving representative points and the centroid, and the brightness distribution extracted from the straight line connecting the moving representative points and the centroid. Furthermore, the contour correction unit determines the movement direction based on the selected movement rule, selects a movement rule based on the brightness distribution, and determines the movement distance based on the selected movement rule.

[0062] The following describes the contour correction process for an independent region. As described earlier regarding the setting of representative points, representative points are set at equal intervals (5 pixels) (S52). Then, the centroid of the contour is calculated based on the average coordinates of the representative points (S53). The coordinates are in pixels, and the centroid coordinates are the calculated average value rounded to the nearest integer. Any representative point is selected. The direction and distance of movement for the selected representative point are determined (S55), and the representative point is moved (S56). If there are no remaining representative points, the next representative point is selected (S54).

[0063] If all representative points have been moved (S57), it is determined whether the correction completion condition, i.e., the stopping condition, is met (S58). The stopping condition is determined by the sum of the moving distances of the representative points. For example, if the sum of the moving distances of the representative points is less than 10 pixels, then the contour correction is completed (S59). If the sum of the moving distances of the representative points is greater than 10 pixels, the process returns to the centroid calculation step, and the representative points are moved again.

[0064] Figure 6 This diagram illustrates an example of the contour detection process. The method for extracting contours from independent regions is explained using this diagram. The region segmentation result is received (S60). In the label (a 1280×960 number matrix), each pixel is checked sequentially, for example, starting from the top left of the matrix, and adjacent pixels with different numbers are detected as contours (S61). All contour pixels are set to an "unprocessed" state (S62). Any pixel from the unprocessed contour pixels is taken and set to a "processing" state (S63).

[0065] Next, if there are unprocessed contour pixels among the adjacent contour pixels of all processed contour pixels, the state is set to "processing" (S64). This setting is repeated until there are no unprocessed contour pixels among the adjacent contour pixels of all processed contour pixels (S65). As a result, it is possible to extract the contour of an independent region containing the initially extracted contour pixels.

[0066] The extracted contour information of the independent regions is output to the next aforementioned representative point setting step, and the contour status is set to "processed" (S67). If there are still unprocessed contours (S68), return to the previous steps and extract the contours of the independent regions again; otherwise, the process is completed (S69).

[0067] Figure 7 This diagram illustrates an example of the processing flow for moving representative points. The centroid is calculated based on the result of the representative point setting (S70) (S71). Then, the representative points are moved sequentially, but the following steps are performed on a single representative point (e.g., Figure 8 The movement of P1 will be explained.

[0068] like Figure 8 As shown, the representative point P1 is set as the moving representative point (S72). The brightness distribution is extracted along a straight line passing through the representative point P1 and the centroid (S73). The length of the extracted brightness distribution is as shown in the figure. Figure 8 As shown by the dashed lines, the length is set to include the actual contour pixels. For example, for P1, the actual contour is between P1 and the centroid, but there are also cases like P4 and P5 where the actual contour is on the extension line between the centroid and the representative point.

[0069] The extracted brightness distribution is compared with a pre-set brightness distribution, and the corresponding movement rule is selected (S74). For example, in a brightness distribution such as... Figure 8 In the case of a brightness distribution like Example 1, where the center pixel is bright, the moving direction is set towards the pixel with 90% center brightness, and the moving distance is set to 0.25 of the distance between the representative point and the pixel with 90% center brightness. For example, in a brightness distribution like... Figure 8 In the case of a dark center pixel as in Example 2 of the brightness distribution, the moving direction is set to the pixel with the maximum brightness, and the moving distance is set to 0.25 of the distance between the representative point and the pixel with the maximum brightness.

[0070] Furthermore, other rules and parameters can be set based on the characteristics of SEM images. Brightness distribution can also be compared using pattern matching methods from general image and signal processing techniques.

[0071] According to the movement rules, the direction and distance of movement are determined (S75-S76), and the representative point is moved (S77). Then, to maintain good accuracy and continuity of the contour, the insertion and deletion of representative points are performed after the representative point is moved (S78). For example, if the distance between the moved representative point and its adjacent representative point is less than 2 pixels, the representative point is deleted. If the distance between the moved representative point and its adjacent representative point is greater than 10 pixels, a new representative point is inserted in the middle between the moved representative point and its adjacent representative point (S79). Since the new representative point is moved in the next correction process, it does not matter if it is offset from the actual contour during insertion.

[0072] The above processing is applied to all representative points to complete one contour correction (S80, S81).

[0073] Furthermore, the representative point moves in the direction of the coordinates of the maximum brightness value after the brightness distribution is smoothed. Additionally, the moving distance of the representative point is set to half the distance between the representative point and the maximum brightness value after the brightness distribution is smoothed.

[0074] Example 2

[0075] In this embodiment, the processing for regions with complex shapes will be described. Parts not described are the same as in Embodiment 1. In this embodiment, after segmenting the object image, the model estimation unit uses an object detection deep learning method to segment the complex-shaped regions into regions with multiple simple shapes when complex-shaped regions exist.

[0076] Here, the object detection deep learning method inputs the learning profile image and the position information of objects with simple shapes into the deep learning method to generate an object detection model. This model is then applied to the input object image to detect the position of objects with simple shapes, and the contours of the aforementioned complex-shaped regions are segmented based on the detected object positions.

[0077] Figure 9 This diagram illustrates an example of the process of dividing a complex region into simple shapes. (Example) Figure 9 As shown in the upper section of the complex-shaped object region, even calculating the centroid of the lower silicon portion cannot properly correct the contour. It is possible that multiple contours exist between the representative point and the centroid. In contrast, as... Figure 9 The result of dividing the lower segment into simple shapes is shown, and contour correction is performed on each simple shape separately.

[0078] Figure 10 This diagram illustrates an example of a dimensional measuring device including an object detection unit. The object detection unit 107 generates an object detection model using a deep learning method, employing a learning profile SEM image and information about the object's position. Here, a simple shape is defined as an object. The object position is the coordinates (in pixels) representing the location of each simple shape. For example, it can be displayed using the coordinates of the top-left and bottom-right corners of a rectangle surrounding a simple shape. The specific form follows the requirements of the object detection deep learning algorithm.

[0079] The object detection model is applied to the input SEM image, outputting the coordinate information of all simple-shaped regions. This coordinate information is combined with the contours of the detected individual regions to further segment the contours of complex regions. Then, contour correction is performed on these regions in the same manner as on the other individual regions.

[0080] This invention is not limited to the embodiments described above, and includes various modifications. For example, the embodiments described above have been detailed for ease of understanding of the invention, but are not limited to having all the described structures. Furthermore, for a part of the structure of each embodiment, structures included in other embodiments can be added, deleted, or replaced.

[0081] Explanation of reference numerals in the attached figures

[0082] 1: Plasma processing device;

[0083] 10: Dimensioning measuring device;

[0084] 101: Central Processing Department;

[0085] 102: Database;

[0086] 103: Model Learning Department;

[0087] 104: Model estimation section;

[0088] 105: Contour correction section;

[0089] 106: Dimensioning and Measurement Department;

[0090] 107: Object Detection Unit;

[0091] 11: Input / output devices;

[0092] 12: Input device;

[0093] 13: Output device;

[0094] 14: Processing condition search device;

[0095] 15: Processing device;

[0096] 16: Evaluation device;

[0097] 17: Image.

Claims

1. A dimension measuring device that uses an image to measure the dimension of an object being measured, characterized in that, Based on given points in the contour of the object being measured, the contour is moved repeatedly until a given necessary condition is met. The dimensions of the object being measured are determined using the contour that has been repeatedly moved. The centroid of the contour is determined using each of the given points. The movement rule is selected based on the positional relationship between the given point of movement and the center of gravity, and the brightness distribution obtained by drawing a straight line through the given point of movement and the center of gravity. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. The movement rule is a specification for moving the given point.

2. A dimension measuring device that uses images to measure the dimensions of an object, characterized in that... Based on given points in the contour of the object being measured, the contour is moved repeatedly until a given necessary condition is met. The dimensions of the object being measured are determined using the contour that has been repeatedly moved. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. After the measurement object is segmented, if there is a first region with a complex shape of the measurement object, the first region is segmented into a second region with multiple simple shapes of the measurement object.

3. The size measuring device according to claim 1 or 2, characterized in that, The movement of the contour is performed after the contour is inferred through machine learning.

4. The size measuring device according to claim 1 or 2, characterized in that, The size of the object is measured using a contour inferred by machine learning through repeated movements as learning data.

5. The size measuring device according to claim 1, characterized in that, The movement direction of the given point is determined according to the selected movement rule.

6. The size measuring device according to claim 1, characterized in that, The movement distance of the given point is determined according to the selected movement rule.

7. The size measuring device according to claim 5, characterized in that, The given point is moved based on the movement rule corresponding to the pre-calculated brightness distribution that is most similar to the calculated brightness distribution.

8. The size measuring device according to claim 6, characterized in that, The given point is moved based on the movement rule corresponding to the pre-calculated brightness distribution that is most similar to the calculated brightness distribution.

9. The dimensional measuring device according to claim 1 or 2, characterized in that, The contour is determined based on the various regions of the segmented measurement object.

10. The size measuring device according to claim 2, characterized in that, The contour of the first region is segmented based on the position of the simple shape estimated by inputting an image of the measured object into a model generated using information about the position of the simple shape and the image of the learning object.

11. A semiconductor manufacturing apparatus comprising a dimension measuring device for measuring the size of an object using an image, and a semiconductor manufacturing apparatus, characterized in that... Using the dimensional measuring device, the contour is repeatedly moved based on given points in the contour of the object being measured until given necessary conditions are met. The dimensions of the object being measured are determined using the contour that has been repeatedly moved. The centroid of the contour is determined using each of the given points. The movement rule is selected based on the positional relationship between the given point of movement and the center of gravity, and the brightness distribution obtained by drawing a straight line through the given point of movement and the center of gravity. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. The movement rule is a specification for moving the given point.

12. A semiconductor manufacturing apparatus comprising a dimension measuring device for measuring the size of an object using an image, and a semiconductor manufacturing apparatus, characterized in that... Using the dimensional measuring device, the contour is repeatedly moved based on given points in the contour of the object being measured until given necessary conditions are met. The dimensions of the object being measured are determined using the contour that has been repeatedly moved. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. After the measurement object is segmented, if there is a first region with a complex shape of the measurement object, the first region is segmented into a second region with multiple simple shapes of the measurement object.

13. The semiconductor manufacturing apparatus according to claim 11 or 12, characterized in that, The sample forming the semiconductor device is subjected to plasma treatment.

14. A semiconductor device manufacturing system, comprising: a semiconductor manufacturing apparatus; The semiconductor device manufacturing system is characterized by having a platform connected to the semiconductor manufacturing apparatus via a network, and performing dimensional measurement processing that uses images to measure the dimensions of the object being measured. The dimensional measurement process has the following characteristics: Based on a given point in the contour of the object being measured, the process of moving the contour is repeated until a given necessary condition is met. The step of measuring the dimensions of the object being measured using the contour that has been repeatedly moved; The step of finding the centroid of the contour using each of the given points; as well as The step of selecting a movement rule based on the positional relationship between the given point of movement and the center of gravity, and the brightness distribution obtained by traversing the given point of movement and the center of gravity. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. The movement rule is a specification for moving the given point.

15. A semiconductor device manufacturing system comprising: a semiconductor manufacturing apparatus; and a platform connected to the semiconductor manufacturing apparatus via a network, and performing dimensional measurement processing using images to measure the dimensions of a measurement object, characterized in that... The dimensional measurement process has the following characteristics: The step of repeatedly moving the contour based on given points in the contour of the measured object until given necessary conditions are met; and The step of measuring the dimensions of the object being measured using the repeatedly moved contour. The given necessary condition is a necessary condition for correcting the offset of the contour inferred using machine learning. After the measurement object is segmented, if there is a first region with a complex shape of the measurement object, the first region is segmented into a second region with multiple simple shapes of the measurement object.

16. The semiconductor device manufacturing system according to claim 14 or 15, characterized in that, The dimensional measurement process is executed as an application of the platform.

Images