JP7735582B2 - Dimension measurement system, estimation system, and dimension measurement method - Google Patents

Description

This disclosure relates to a technology for measuring dimensions of device structures, etc. from images of samples such as semiconductor devices.

Technologies for measuring (sometimes referred to as length measurement) the dimensions of device structures and other components from images of semiconductor devices and other samples, as well as for observing, analyzing, evaluating, and inspecting samples, include those that use charged particle beam instruments, such as scanning electron microscopes (SEMs) and transmission electron microscopes (TEMs), to perform these processes (sometimes collectively referred to as dimensional measurement). A computer system built into or externally connected to the charged particle beam instrument processes the images captured by the charged particle beam instrument to measure the dimensions of target structures and patterns (e.g., trenches) in device structures and other components.

For example, the computer system detects the contours of a structure from a cross-sectional image of a semiconductor device (e.g., a cross-section created by cleavage) and measures dimensions (e.g., trench width and trench depth) based on the contours.

An example of prior art is Japanese Patent No. 4,865,147 (Patent Document 1). Patent Document 1 describes, as a "pattern measurement method and measurement system using a display microscope image," "a step of setting an edge reference line on the edge of a pattern in a microscope image," "a step of setting multiple edge points along the edge reference line," "a step of integrating the brightness profile near a search line that intersects with the edge reference line at one edge point and calculating the edge position for one edge point," "a step of calculating an approximation line for the edge from the edge positions calculated for each of the multiple edge points," and "a step of measuring the shape of the pattern using the approximation line" (Claim 1).

Patent No. 4,865,147 (corresponding U.S. Patent No. 7,054,506)

Technologies for dimension measurement systems that measure the dimensions of an object based on the detection of the contours of the target structure from images (e.g., cross-sectional images) captured using the charged particle beam device include, first, a technology in which a computer system automatically performs contour detection and dimension measurement using image processing such as template matching. Second, a technology in which a person manually inputs and sets the information required for contour detection and dimension measurement, and the computer system performs contour detection and dimension measurement based on that information.

With the first technique, it may be difficult to properly detect contours using image processing such as template matching, or the computational load may be too high.

The second technology requires a person to set the information necessary for contour detection and dimension measurement. For example, in the case of Patent Document 1, the user sets information such as "edge reference lines" and "edge points." Using information set by a person can facilitate image processing by a computer system. However, the second technology requires a lot of manual operation by a person, which can reduce work efficiency.

The purpose of this disclosure is to provide technology for dimension measurement systems and other systems that use the above-mentioned charged particle beam device to measure the dimensions of target structures from captured images, which allows for some information to be input and set manually, while reducing the amount of manual operation required and improving work efficiency. Another purpose is to provide technology that can reduce the impact of errors caused by manual operation.

A representative embodiment of the present disclosure has the following configuration. A dimensional measurement system according to the embodiment is a dimensional measurement system for measuring the dimensions of a structure in an image, and includes a processor and memory resources for storing data processed by the processor. The processor displays to a user a screen including the image and a graphical user interface (GUI), specifies a reference line on the image on the screen based on manual operation by the user to identify a portion of the area of the structure to be measured, acquires position information of the reference line, specifies a reference point on the image on the screen based on manual operation by the user to be included in the area to identify the area of the structure to be measured, acquires position information of the reference point, detects a contour or edge of the area of the structure to be measured in the image using the reference line and the reference point, and measures the dimensions of the structure to be measured in the image using information about the detected contour or edge.

An estimation system according to an embodiment of the present invention is an estimation system for estimating a structure within an image, and includes a processor and memory resources for storing data processed by the processor. The processor displays to a user a screen including the image and a graphical user interface (GUI), specifies a reference line on the image on the screen based on manual operation by the user to identify a portion of the area of the structure to be measured, acquires position information for the reference line, specifies a reference point within the area to identify the area of the structure to be measured based on manual operation by the user, acquires position information for the reference point, accumulates data including the image, the reference line, and the reference point as learning data, uses the learning data to train a machine learning learning model for estimating the reference line and the reference point from an input image, inputs the input image to be estimated into the learning model, and outputs information on the estimated reference line and the estimated reference point as estimation results from the learning model.

Representative embodiments of the present disclosure relate to technologies such as a dimension measurement system that uses the above-mentioned charged particle beam device to measure the dimensions of an object from a captured image, and while allowing a person to input and set some of the information, the effort and burden of manual operation by a person can be reduced, thereby improving work efficiency. Furthermore, the impact of errors caused by manual operation by a person can be reduced. Issues, configurations, effects, etc. other than those described above are described in the detailed description of the invention.

FIG. 1 is a diagram showing an example of the configuration of a system including a dimension measurement system according to a first embodiment. FIG. 2 is a diagram showing an example of the configuration of the dimension measurement system according to the first embodiment as a computer system. 1 is a diagram showing a configuration example of an SEM in Embodiment 1. FIG. 1A and 1B are diagrams showing the structure, dimensions, etc. of a semiconductor device in the first embodiment; FIG. 3 is a diagram showing a processing flow of the dimension measurement method according to the first embodiment. FIG. 3 is a diagram showing an example of a GUI screen in the first embodiment. FIG. 4 is a diagram showing an example of definition information in the first embodiment. 4A to 4C are diagrams showing examples of cross-sectional images and an interface selection GUI in the first embodiment. 3A to 3C are diagrams showing examples of cross-sectional images and a region selection GUI in the first embodiment. 5A and 5B are diagrams showing examples of algorithms for detecting contours in cross-sectional images in the first embodiment; 5A and 5B are diagrams showing examples of algorithms for detecting contours in cross-sectional images in the first embodiment; FIG. 4 is a diagram showing a line profile graph in the first embodiment. FIG. 10 is a diagram showing an example in which the second rule is applied to edge candidate extraction in the first embodiment. 5A and 5B are diagrams showing examples of combinations of edge candidates in the first embodiment; 5A and 5B are diagrams showing examples of detected contours (edge points and contour lines) in the first embodiment. 5A to 5C are diagrams showing examples of contour detection related to a mask in the first embodiment. 5A and 5B are diagrams showing an example of contour detection related to a pillar in the first embodiment. 4A to 4C are diagrams showing examples of measurement target dimensions in the first embodiment. FIG. 10 is a diagram showing an example of measuring the area of a structure in the first embodiment. FIG. 10 is a diagram showing an example of a GUI screen that displays a dimension measurement result in the first embodiment. 5A to 5C are diagrams showing examples of cross-sectional images etc. when dust is attached in the first embodiment. 4A and 4B are diagrams showing examples of an image of an etching processing target pattern and an image after actual processing in the first embodiment. 10A and 10B are diagrams showing an example in which a plurality of area reference points or the like are collectively designated within a cross-sectional image in the first embodiment. 10A and 10B are diagrams showing a first example of setting two interface reference lines in a cross-sectional image in the first embodiment; FIG. 10 is a diagram showing a second example of setting two interface reference lines in a cross-sectional image in the first embodiment. FIG. 10 is a diagram showing an example of a GUI screen for an interface reference line adjustment function, as a first modification of the first embodiment. 10A to 10C are diagrams showing cross-sectional images etc. in a case where a reference line is set to a point other than an interface, as a second modification of the first embodiment. FIG. 10 is a diagram showing another method for setting a line profile as a third modification of the first embodiment. 13A and 13B are diagrams showing a method for setting a dimension measurement target line using area reference points, as a fourth modification of the first embodiment; FIG. 13 is a diagram showing a method for setting on/off of a dimension measurement target using an area reference point, as a fifth modification of the first embodiment. 10A and 10B are diagrams showing a method for setting a horizontal line profile in the dimension measurement system according to the second embodiment. 10A and 10B are diagrams showing a method for setting a line profile in the radial direction in the dimension measurement system 2 of the third embodiment. FIG. 10 is a diagram showing an example of the configuration of an estimation system according to a fourth embodiment as a computer system. A diagram showing the input/output configuration of a learning model in embodiment 4. A figure showing the configuration of a learning model in embodiment 4. FIG. 13 is a diagram showing a learning process flow in the fourth embodiment. FIG. 13 is a diagram showing an example of a GUI screen for an estimation function in the fourth embodiment. A diagram showing the input/output configuration of a learning model in a variant of embodiment 4. A diagram showing the input/output configuration of a learning model in a variant of embodiment 4. FIG. 13 is a diagram showing an example of a GUI screen in a modification of the fourth embodiment. FIG. 13 is a diagram showing an example of the configuration of a dimension measurement system according to a fifth embodiment as a computer system. FIG. 13 is a diagram showing a dimension measurement processing flow in the fifth embodiment. 10A and 10B are diagrams showing an example of setting points and lines for supporting contour detection in an image in a first comparative example to the embodiment. 10A and 10B are diagrams showing an example of setting points and lines for supporting dimension measurement on an image in a second comparative example to the embodiment.

Embodiments of the present disclosure will be described in detail below with reference to the drawings. In the drawings, identical parts are generally designated by the same reference numerals, and repeated explanations will be omitted. In the drawings, the depiction of components may not represent their actual position, size, shape, range, etc., in order to facilitate understanding of the invention.

For the purpose of explanation, when describing processing by a program, the program, functions, processing units, etc. may be described as the main focus, but the main hardware focus for these is the processor, or a controller, device, computer, system, etc. that is made up of that processor, etc. The computer executes processing according to the program read into memory using resources such as memory and communication interfaces as appropriate through the processor. This realizes specified functions, processing units, etc. The processor is made up of semiconductor devices such as a CPU/MPU or GPU, for example. Processing is not limited to software program processing, but can also be implemented using dedicated circuits. Dedicated circuits such as FPGAs, ASICs, and CPLDs can be used.

The program may be pre-installed as data on the target computer, or may be distributed as data from a program source to the target computer. The program source may be a program distribution server on a communications network, or a non-transitory computer-readable storage medium, such as a memory card or disk. The program may be composed of multiple modules. The computer system may be composed of multiple devices. The computer system may be composed of a client-server system, a cloud computing system, an IoT system, etc. Various data and information may be composed of structures such as tables and lists, for example, but are not limited to these. Expressions such as identification information, identifiers, IDs, names, and numbers are interchangeable.

[Assignments, etc.]
We will provide additional information on the challenges. In recent years, device structures have been miniaturized to improve semiconductor device performance, and device structures, such as Fin-FETs (Field Effect Transistors) and 3D-NAND, have become more three-dimensional and complex. The dimensions of these cutting-edge devices are on the order of nanometers (nm), and the film structures between structures now require processing precision on the order of sub-nm.

Semiconductor devices are generally manufactured through a combination and repetition of processes such as lithography, etching, film formation, and planarization. For example, in the etching process, as processing becomes more difficult, efforts are being made to optimize etching conditions by combining post-processing dimensional data with machine learning, creating a need for technology that can measure dimensions precisely at high speed and in large quantities.

Charged particle beam instruments such as SEM and TEM are often used as a method for precise dimensional measurement. This is because charged particle beam instruments use charged particle beams with wavelengths shorter than those of light, allowing for high-resolution observation of microstructures on the sub-nanometer order that cannot be observed with optical microscopes.

When comparing SEM and TEM as measurement methods, observation with a TEM requires pre-processing using a focused ion beam (FIB). Therefore, when cross-sectional observation and dimensional measurement need to be performed in a shorter time, a method of cleaving a sample after etching and observing the cross-section with an SEM may be chosen.

However, due to damage and contamination caused by cleavage, the cross section of a cleaved sample may partially collapse or become contaminated with dust, resulting in the observed structure being different from the structure formed during the original etching process. If the dimensions of a structure with such cleavage artifacts are measured, there is a risk of unintended feedback being given to the etching conditions when the measurement results are fed back. For this reason, it is necessary to measure only structures that have been cleaved normally, in other words, only structures that have been cleaved without any collapse or dust.

In addition, when measuring dimensions during the etching process, the processing depth of the layer to be etched (e.g., trench depth) is often measured using the interface between the mask layer and the layer to be etched (e.g., a Si layer) as a reference. Therefore, detecting the position of such an interface (in other words, the material interface or reference interface) is also important.

However, it is difficult to determine the normality of structures such as the cross-section of the cleaved sample mentioned above, or to accurately detect the position of interfaces, using conventional image processing methods such as template matching. Therefore, the normality of structures whose dimensions are being measured has traditionally been left to the subjective judgment of the operator performing the measurement.

Even when the correct dimension measurement target structure and the exact position of the reference interface are determined, accuracy issues remain in the measurement process after determination. When observing the cross section of a cleaved sample using an SEM, a white shadow-like area (hereinafter referred to as a "white band") appears in the captured image at the boundary between the device structure (e.g., a pillar) and the open vacuum region, i.e., the area near the contour of the device structure. This white band is a high-brightness pixel area on the image, and is caused by, for example, secondary electrons from a structure located in the depth direction of the captured image. This makes it difficult to automatically detect the contour of the device structure from the image using image processing such as binarization. In other words, there are issues regarding the efficiency and accuracy of detecting the contour or edge of the device structure.

For this reason, conventional dimensional measurement of cross-sectional SEM images is often performed manually and with the operator's judgment, which is time-consuming. Furthermore, there is also the issue that errors in manual operation and judgment occur depending on the operator, resulting in differences in the precision and accuracy of contour detection and dimensional measurement.

For example, Patent Document 1 describes an example in which the cross-sectional structure of a magnetic head device is observed with an SEM to measure dimensions such as the structure's height and sidewall angle. Patent Document 1 also describes how, for a single magnetic head structure, a user can specify an edge reference line or multiple edge points on an SEM image to calculate approximate straight lines for each side of the structure, which are then used to measure dimensions and calculate the angle of the structure's sidewalls.

In the case of a method in which the user inputs and sets some of the information required for contour detection and dimension measurement as described above (the second technique mentioned above), the number of input operations required by the user increases, resulting in significant effort. When a cross-sectional SEM image contains multiple target structures, the number of input operations required by the user is equal to the product of the number of target structures and the number of measurement dimension types (e.g., width, height, etc.). When performing dimension measurements on multiple samples or complex structures, this increases the user's burden and reduces work efficiency.

In particular, when measuring dimensions after etching, it is necessary to measure various dimensions of the remaining mask pattern, various dimensions of the trench portion dug by etching, and various dimensions of the fin or pillar-type structure directly below the mask pattern. In such cases, there is a need for automation of the contour detection and dimensional measurement of the target structure pattern from cross-sectional images, as well as reducing the burden of manual input by the user and improving work efficiency.

[First and second techniques]
In a conventional semiconductor device manufacturing process, structures such as masks, pillars, and trenches (sometimes referred to as structures, patterns, etc.) are formed on wafers by processes such as deposition and etching. In the manufacturing process, it is necessary to confirm and inspect whether the three-dimensional structures obtained as a result of such processing are appropriate. In this case, as described above, a cross section of the structure is prepared using, for example, cleavage or FIB processing, and an image of the cross section is captured using a SEM or other means. A computer system and an operator then perform processing and operations such as length measurement and observation on the cross-sectional image to confirm and inspect whether the shape of the structure is appropriate.

To do this, it is necessary to determine and detect the contours or edges of structures from cross-sectional images and measure their dimensions. To streamline this type of processing and work, automation and efficiency using computer systems is required. However, as mentioned above, it is difficult to fully automate the process of detecting the contours of structures from cross-sectional images and measuring their dimensions.

Conventionally, the first technique described above, in which a computer system is used to fully automatically detect the contours of target structures from cross-sectional SEM images and measure their dimensions, poses challenges in the contour detection process, making efficient and accurate contour detection difficult. Therefore, the second technique described above involves a method in which a user manually inputs and sets some of the information for a cross-sectional SEM image to support the computer system's contour detection and dimension measurement processes. For example, reference lines and points are specified and set for the cross-sectional image. The computer system then performs contour detection and dimension measurement processes using the set information for the lines and points for the cross-sectional image. This type of method combines manual operation and automatic processing (a general term that includes methods, devices, etc.), and for purposes of explanation, is sometimes referred to as a semi-automatic method.

[Comparative Example 1]
FIG. 43 is an explanatory diagram of a technology of Comparative Example 1 in comparison with the embodiment. The technology of Comparative Example 1 corresponds to the technology of Patent Document 1. In the technology of Comparative Example 1, a user manually sets points and lines, etc., that serve as support for contour detection on a cross-sectional image. FIG. 43A is a schematic diagram of an example cross-sectional image, showing a vacuum background (shown in white in the schematic but black in luminance) with a roughly trapezoidal pillar 4301 and trench 4302 region. FIG. 43B shows an example in which a user manually inputs and sets points and lines to support contour detection on the image of FIG. 43A. In this example, the user sets edge points ep1 to ep6 on the sides of the trapezoid of pillar 4301. In other words, the user sets edge lines EL1 to EL3 (corresponding to the "edge reference lines" in Patent Document 1) on the sides of the trapezoid of pillar 4301. For example, a straight line connecting edge point ep1, which is the starting point, and edge point ep2, which is the end point, becomes edge line EL1.

The computer system uses the set edge points ep1 to ep6 or edge lines EL1 to EL3 to determine the edges that make up the contour of pillar 4301. For example, the computer system detects the edge of pillar 4301 by referencing the brightness profile in the vertical direction on edge line EL1 and searching for brightness on the brightness profile. The computer system determines an approximate straight line for the contour by connecting the detected edges. Based on the information about the detected contour, the computer system measures the dimensions of the structure pattern, such as the width and height of pillar 4301.

In Comparative Example 1, the user must manually input and set at least two edge points for each side of a structure in the image, such as a trapezoidal pattern. Therefore, the more structures, such as pillars 4301, and the more types of dimensions to be measured, the more time and effort the user must manually input. Furthermore, the more complex the pattern shape, the greater the effort required.

Comparative Example 1 can be considered a technology in which a user inputs and sets support lines, points, etc. by tracing the contours or edges of a structural pattern in an image. This technology requires a lot of manual operation. Furthermore, because it is a manual operation, it is not always possible for a user to accurately set support lines, points, etc. for the contours or edges of a structural pattern in an image, and errors occur between users.

[Comparative Example 2]
FIG. 44 shows Comparative Example 2 as another example of conventional technology. The technology of Comparative Example 2 involves a user manually setting lines, points, and the like on a cross-sectional image to support dimensional measurement. FIG. 44A shows a schematic diagram of an example cross-sectional image, for example, having four pillars 4401 and four trenches 4402. In Comparative Example 2, the user manually inputs and sets the location where the user wants to measure the dimension, i.e., the position that defines the dimension of the measurement target, for a structural pattern in the cross-sectional image, for example, a pillar 4401 or a trench 4402 that has a roughly trapezoidal shape.

For example, (B) in Figure 44 shows the case of measuring trench depth. When measuring trench depth, the user determines and specifies two end points (points a1 and a2) that define the trench depth (and the corresponding dimension measurement target line 4403) for each trench 4402 as the measurement target points. For example, this is done by clicking the mouse for each point. Point a1 indicates the top of trench 4402, and point a2 indicates the bottom of trench 4402. One end of dimension measurement target line 4403 is point a1, and the other end is point a2. The computer system then measures the portion of dimension measurement target line 4403 defined by the two specified points (points a1 and a2) as the trench depth. In other words, the two points are the starting and ending points of the dimension.

Similarly, in (C) of Figure 44, when pillar width is to be measured, the user determines and specifies two end points (points b1 and b2) that define the pillar width (and the corresponding dimension measurement target line 4404) for each pillar 4401 as the measurement target points. Point b1 is a point on the left sidewall of pillar 4401, and point b2 is a point on the right sidewall of pillar 4401. One end of dimension measurement target line 4404 is point b1, and the other end is point b2. The computer system then measures the portion of dimension measurement target line 4404 defined by the two specified points (points b1 and b2) as the pillar width.

As such, in Comparative Example 2, the user must input at least two points for each target structure pattern, and for each measurement location and dimension type. Therefore, as with Comparative Example 1, the greater the number of target structures and dimension types, the greater the manual effort required by the user in Comparative Example 2. Furthermore, errors may occur due to manual human operation (such as mouse operation) or subjective judgment. The two points specified by the user for the measurement location and dimension (for example, trench depth) may deviate from the actual situation, and individual differences may also occur. Such errors affect the accuracy of the dimension measurement.

[Means of solution, etc.]
In consideration of the above-mentioned problems, the dimension measurement system and the like according to the embodiment have the following means for solving the problems.

The dimension measurement system of the embodiment has a processor and memory resources, and performs processing related to contour detection and dimension measurement on images (in other words, cross-sectional images, target images, captured images, etc.) captured by a charged particle beam device such as an electron microscope so as to capture the cross-sectional structure of the semiconductor device sample. The processor performs processing according to the dimension measurement method of the embodiment on data and information such as images stored in the memory resources.

The dimension measurement system of the embodiment is a system that employs a semi-automatic method that combines a method in which a user or other person manually inputs, specifies, and sets predetermined information, such as reference lines and reference points, to support contour detection and dimension measurement for cross-sectional images, and an automatic method in which a computer automatically processes the cross-sectional images for which the predetermined information has been set. The embodiment incorporates innovations in this semi-automatic method. Note that systems and methods that manually specify and select material interfaces or structural areas to be measured and then automatically perform contour detection and dimension measurement are sometimes referred to as "semi-automatic methods."

The dimension measurement method of the embodiment includes the steps of: designating, based on manual operation by a person such as a user, a location such as a material interface that is difficult to automatically detect or recognize, as a reference line (i.e., an interface reference line, a support line, etc.) on a cross-sectional image; and designating, based on manual operation by a user, an area of a structure pattern to be measured that is difficult to automatically detect or recognize, as a reference point (i.e., an area reference point, a support point, etc.) on the cross-sectional image. These steps are steps related to manual operation by a person. The dimension measurement method of the embodiment further includes the steps of automatically detecting the contour of the structure pattern to be measured for the image in which the reference line and reference point are set, and automatically measuring the dimensions of the measurement object based on information about the detected contour. These steps are steps related to automatic computer processing.

The dimension measurement system and method of the embodiment are characterized by a manual setting method in which, for cross-sectional images, the user manually operates a GUI screen to specify an interface reference line, a reference line for indicating the material interface of the target structure, and also to specify an area reference point, a reference point for indicating the area of the structure to be measured. These manual operation steps are realized as a new user interface that differs from conventional technology examples. This reduces the amount of input, etc., reducing the user's effort and burden, and also reducing errors due to manual user operation.

An automatic feature of the dimension measurement system and method of the embodiment is that, for a cross-sectional image in which the specified information has been set as described above, a computer system processes the cross-sectional image using a specified algorithm to identify the area of the target structure pattern based on the reference lines and reference points, and detects the edges that make up the contour of the target structure pattern.

The computer system then measures the specified type of dimension in the region of the structure pattern to be measured based on the reference lines and reference points and the detected contour information. The type of dimension to be measured (e.g., trench width, trench depth, etc.) can be set and specified by the user on the GUI screen.

The computer system displays at least one of the detected contour information and the dimension measurement results to the user on a GUI screen.

The algorithm for the above-mentioned contour detection process is new and different from conventional examples (see Figure 10 below, etc.), and can detect the contours of structures more efficiently and accurately. The computer system can then measure the dimensions of the object more efficiently and accurately based on the information on the detected contours.

First Embodiment
The dimension measurement system and method according to the first embodiment will be described with reference to FIG.

[Semiconductor dimension measurement system]
FIG. 1 shows an example of the configuration of a semiconductor dimension measurement system, which is a system including the dimension measurement system of the first embodiment. The entire system of FIG. 1 is a semiconductor dimension measurement system, and is configured to include a scanning electron microscope (SEM) 2 and a computer system 1, which is the dimension measurement system 1 of the first embodiment. The example of the system configuration of FIG. 1 also includes a database (DB) 3, semiconductor device manufacturing equipment 4, a manufacturing execution system (MES) 5, a client terminal (operation PC) 6, and the like, on a communication network 9. These components are interconnected via, for example, a LAN as the communication network 9, and can communicate with each other. The computer system 1, which is the dimension measurement system 1 of the first embodiment, is configured as a server on the communication network 9, and is connected via communication with devices such as the scanning electron microscope (SEM) 2 and the operation PC 6.

The dimension measurement system 1 of embodiment 1 is configured by a computer system 1 described below (Figure 2). This computer system 1 realizes the functions of the dimension measurement system 1 based on program processing by a processor. The dimension measurement system 1 has at least a contour detection function F1 and a dimension measurement function F2 as its functions. The contour detection function F1 is a function that detects the contour of a structure (in other words, a pattern) from an image 7, which is a cross-sectional image of the target sample. The dimension measurement function F2 is a function that measures the dimensions of the target based on information about the contour detected by the contour detection function F1 and outputs the dimension measurement result 8.

SEM2 is an example of a cross-sectional image capturing device. SEM2 has the function of capturing a cross-section of a semiconductor device (e.g., a wafer) that is a target sample, and obtaining the cross-sectional image as image 7. As shown in Figure 3 below, SEM2 is configured with a main body 301 and a controller 302 connected to main body 301.

The SEM2 or other components may have the function of creating a cross section on the target sample. For example, the SEM2 may be a device that has the function of creating a cross section on the target semiconductor device. For example, the SEM2 may be an FIB-SEM device equipped with a focused ion beam mechanism. The FIB-SEM device creates a cross section on a part of the sample by FIB processing. A device that has the function of creating a cross section may exist separately from the SEM2. For example, a cleaving device that creates a cross section by cleaving the sample may be provided on the communication network 9.

The SEM 2 and computer system 1 may communicate with external devices as appropriate to input and output necessary data and information. For example, the SEM 2 may transmit image 7 data obtained by photographing to the computer system 1 or the operating PC 6. Alternatively, the SEM 2 may store the image 7 data in DB 3. In this case, the computer system 1 retrieves the image 7 from DB 3. DB 3 may be, for example, a DB server, and may store image 7 data and various other information such as design data.

Semiconductor device manufacturing equipment 4 is equipment that manufactures semiconductor devices, which are target samples, and includes equipment for each manufacturing process, such as etching equipment. MES 5 is a system that executes and manages the manufacturing of semiconductor devices using semiconductor device manufacturing equipment 4, and contains sample design data, manufacturing process information, etc. Computer system 1 may reference design data, manufacturing process information, etc. from MES 5, semiconductor device manufacturing equipment 4, etc.

The client terminal 6 is an operating PC used by the user U1, and is an information processing terminal device that has the function of accessing the dimension measurement system 1 (particularly the server function) via the communication network 9. The user U1, such as an operator, may operate the dimension measurement system 1 directly, or may use the dimension measurement system 1 by accessing it from the client terminal 6. The client terminal 6 can be a general PC, etc., and may have an input device for input operations and an output device for display, etc. built-in, or these may be externally connected. The user U1 may access and use the SEM2, etc. from the client terminal 6.

In embodiment 1, the dimension measurement system 1, which is a server, is configured as a client-server system with the client terminal 6. The dimension measurement system 1 is responsible for the main processing, and the client terminal 6 is responsible for GUI functions. The computer system 1 and the client terminal 6 may also be integrated into a system. In other words, the computer system 1 may have input/output and GUI functions.

The dimension measurement system 1 of embodiment 1 is not limited to the configuration example shown in Figure 1. For example, the dimension measurement system 1 may be integrated into the controller 302 (Figure 3) of the SEM 2. The dimension measurement system 1 may also be integrated into other components of Figure 1. The computer system 1 is not limited to being implemented by a single computer, and may be realized by a cloud computing system, etc.

Not limited to the example shown in Figure 1, an example of an external device, which is another component, may include a cross-section observation device that observes the cross-section of a sample and performs analysis, inspection, etc. The cross-section observation device may be, for example, a TEM device. Alternatively, the SEM2 may be the cross-section observation device. The SEM2 may be an STEM (scanning transmission electron microscope), another type of microscope, or an imaging device.

A program distribution server (not shown) or the like may be provided in the communication network 9. The program distribution server distributes data such as programs (software 110 in FIG. 2) for realizing the dimension measurement system 1 of embodiment 1 to the computer system 1. Furthermore, the data such as images 7 and programs may be input and output in a form stored on a storage medium such as a memory card, rather than being input and output via communication.

[Computer System]
2 shows an example of the configuration of the dimension measurement system 1 according to the first embodiment as a computer system 1. The computer system 1 includes predetermined hardware and software that work together to realize functions. The computer system 1 shown in FIG. 2 includes a processor 101, memory resources 102, a communication interface 103, an input/output interface device 104, and the like. These components are connected to a bus and can communicate with each other. The memory resources 102 include a ROM 105, a RAM 106, a storage 107, and the like.

The processor 101 is composed of a semiconductor device such as a CPU, MPU, or GPU. The processor 101 may also include ROM, RAM, various peripheral functions, etc. The processor 101 executes processing in accordance with software 110, including a control program stored in memory resources 102, such as a dimension measurement program, which is read from ROM 106 or storage 107 to RAM 105 and deployed. This realizes predetermined functions such as the contour detection function F1 and dimension measurement function F2 in Figure 1. The software 110 controls all functions and operations described in embodiment 1. When the computer system 1 starts up, an execution module based on the processor 101 and software 110 operates.

Storage 107 stores various data and information such as image 7, shooting information 111, definition information 112, and dimension measurement results 8. Image 7 is image data obtained from SEM 2. Dimension measurement results 8 are data such as dimension measurement results obtained by dimension measurement function F2. Shooting information 111 is shooting information when image 7 is captured by SEM 2, and includes, for example, shooting conditions and optical setting information for SEM 2, attribute information for image 7, etc. Shooting information 111 includes at least information such as pixel size, which is the size in real space per pixel in image 7, or the magnification at which that pixel size can be calculated.

As will be described later (see Figure 7, etc.), definition information 112 is the necessary definition information and setting information related to the functions of the dimension measurement system 1, and includes system setting information, user setting information, etc. In addition, the memory resource 102 stores data and the like that arises during the processing process as appropriate. Note that the dimension measurement system 1 may also use external memory resources (e.g., DB3 or an external storage device) as appropriate.

In the computer system 1, for example, data such as an image 7 acquired in advance from the SEM 2, DB 3, or operating PC 6 in Figure 1 is stored in the storage 107. The image 7 is associated with and managed by the shooting information 111, definition information 112, dimension measurement results 8, etc.

The communication interface 103 is a device that implements a communication interface for the SEM 2, the communication network 9, etc. The input/output interface 104 is a device that implements an input/output interface, and allows external connection of input and output devices. Examples of input devices include a keyboard, mouse, and microphone. Examples of output devices include a display, printer, and speaker. Note that the input and output devices may be built into the computer system 1.

Based on user U1's input of instructions, etc. on the GUI screen 60, the computer system 1 uses the input image 7 of the dimension measurement target, shooting information 111, and definition information 112 to detect the contour of the structure pattern in the target image 7, measure the specified type of dimension, and generate and store the dimension measurement result 8. The computer system 1 then transmits the dimension measurement result 8 to the operation PC 6. The operation PC 6 displays the image 7 and dimension measurement result 8 on the GUI screen 60 along with the GUI. The user U1 can check the dimension measurement result 8, etc. on the GUI screen 60.

A user U1, such as an operator, uses the computer system 1 by operating the input devices and displaying the screen of the output device on the client terminal 6, which is an operation PC. The operation PC 6 includes a processor, memory, a communication interface, a monitor display, a mouse, a keyboard, etc., all of which are not shown. A GUI screen 60 is displayed on the screen of the monitor display.

Client-server communication takes place between the computer system 1 and the client terminal 6, and the following communications and operations take place, for example: User U1 operates the client terminal 6 to access the computer system 1. The client terminal 6 accesses the server function of the computer system 1 via the communication network 9 and sends a request. In response to the request, the server function of the computer system 1 generates screen data including a GUI (graphical user interface) and sends it to the client terminal 6. The screen data may be in the form of a web page, etc. Alternatively, the target data and information may be sent and received directly, rather than screen data. Based on the received screen data, the client terminal 6 displays a GUI screen 60 in the form of a web page, etc., on its display screen. User U1 looks at the GUI screen 60 to check information about the functions of the dimension measurement system 1 and inputs settings, instructions, etc. as necessary.

The client terminal 6 sends requests and information based on user-input information to the computer system 1. The computer system 1 executes processing related to the function based on the received requests and information and saves the results. The computer system 1 sends screen data including, for example, an image 7 and dimensional measurement results 8 to the client terminal 6. The client terminal 6 displays a GUI screen 60 including the image 7 and dimensional measurement results 8 based on the received data. The user U1 can view and confirm the image 7 and dimensional measurement results 8 on the GUI screen 60.

While Figure 1 shows one user U1 and one client terminal 6, the system may have multiple users and multiple client terminals 6. The computer system 1 and processor 101 are not limited to being physically one, and may exist in multiple numbers.

[SEM]
3 shows an example of the configuration of the SEM 2. The SEM 2 is broadly divided into an SEM main body 301 and a controller 302 connected to the main body 301. The main body 301 further comprises an electron optical column (hereinafter referred to as the "column") and a sample chamber provided below the column. The controller 302 is a system that controls imaging by the main body 301. The controller 302 includes an overall control unit 320, a signal processing unit 321, a storage unit 322, a communication interface 323, etc., and is externally connected to an input device 324 and an output device 325.

The column of the main body 301 includes, as its components, an electron gun 311, an acceleration electrode 312, a focusing lens 313, a deflection lens 314, an objective lens 315, and a detector 317. The electron gun 311 emits an electron beam b1, which is a charged particle beam. The acceleration electrode 312 accelerates the electron beam b1 emitted from the electron gun 311. The focusing lens 313 focuses the electron beam b1. The deflection lens 314 deflects the trajectory of the electron beam b1. The objective lens 315 controls the focusing height of the electron beam b1. The sample chamber is a room in which samples such as wafers and coupons (broken pieces of wafers) are stored, and includes a stage 316 for moving the observation position on the sample to the irradiation point of the electron beam.

The stage 316 is a sample stage on which the target sample 30, a semiconductor device, is placed. The stage 316 has the ability to move not only in the X and Y directions but also in the Z direction, and to rotate about the XY, YX, or Z axes, allowing the captured field of view to be moved horizontally and vertically relative to the front-facing image, or in a rotational direction within the field of view. This allows the field of view for imaging to be set.

The detector 317 detects particles b2, such as secondary electrons and backscattered electrons, generated from the sample 30 irradiated with the electron beam b1 as an electrical signal. The detector 317 outputs a detection signal, which is an electrical signal.

The overall control unit 320 controls the operation of the controller 302 and the main body 301. The overall control unit 320 issues instructions such as drive control to each unit. Each unit, such as the overall control unit 320, can be implemented as a computer or dedicated circuit. The signal processing unit 321 receives the detection signal from the detector 317, performs processing such as analog/digital conversion to generate an image signal, and stores it as image data in the memory unit 322. The memory unit 322 can be implemented as a non-volatile memory device, etc. The overall control unit 320 also stores shooting information related to the image in association with the memory unit 322.

The communication interface 323 is a device that implements a communication interface for the communication network 9 and the computer system 1. In response to a request from the computer system 1, for example, the overall control unit 320 transmits data such as images and shooting information stored in the memory unit 322 to the computer system 1 via the communication interface 323. The computer system 1 stores data such as images 7 received from the controller 302 of the SEM 2 in the memory resource 102.

[Semiconductor devices]
4A and 4B show examples of a structure pattern of a semiconductor device, which is a target sample in the first embodiment. In (A) of FIG. 4A, the shape of a design pattern of a three-dimensional structure is shown in a perspective view, e.g., a cross section in the X-Z plane. This cross section is formed by, e.g., cleavage or FIB processing. This structure has a mask 401, pillars 402, and trenches 403. The mask 401 and pillars 402 are structures with trapezoidal cross sections, and the trench 403 is a structure with an inverted trapezoidal cross section, with the pillars 402 formed below the portion where the mask 401 is present and the trench 403 formed below the portion where the mask 401 is not present.

Figure 4(B) is a cross-sectional view corresponding to the cross-section in the X-Z plane of (A), showing an example of the dimension to be measured. The semiconductor device has, in the Z direction, a mask layer 411 made of a first material on a layer to be etched 412 made of a second material. A mask 401 is formed on the mask layer 411, and pillars 402 and trenches 403 are formed in the layer to be etched 412. The material interface 413 between the mask 401 and pillars 402 in the Z direction is shown by a dashed line.

Dimension A1 is the trench width at the top surface (in other words, the top) 414 of the trench 403. The position of the top surface 414 of the trench 403 corresponds to the position of the material interface 413 between the mask 401 and the pillar 402. Dimension A2 is the trench width at the bottom surface (in other words, the bottom) 415 of the trench 403. Dimension A3 is the trench width in the height/depth direction (Z direction) of the trench 403, for example, at an intermediate position (for example, the 50% position). The position of dimension A3 is not limited to being defined as a percentage (%), but may also be defined by the distance from the top 414, the distance from the bottom 415, etc. Dimensions A1 to A3 are lengths in the X direction.

Dimension A4 is the trench depth, which is the length from the top 414 to the bottom 415 of the trench 403 in the Z direction. Dimension A5 is the trench sidewall angle, which is the angle that the sidewall 416 (the sloped portion of the trapezoid) makes with, for example, the bottom 415 of the trench 403. The trench sidewall angle may be defined as the angle at a given position in the Z direction.

Dimension B1 is the pillar width at the top surface (i.e., the top) of pillar 402. The position of the top of pillar 402 corresponds to the position of material interface 413. Dimension B2 is the pillar width at the bottom surface (i.e., the bottom) of pillar 402. The position of the bottom of pillar 402 corresponds to the position of the bottom of trench 403. Dimension B3 is the pillar width in the height direction (Z direction) of pillar 402, for example, at an intermediate position (e.g., the 50% position). The position of dimension B3 is not limited to being defined as a percentage (%), but may also be defined by the distance from the top, the distance from the bottom, etc. Dimensions B1 to B3 are lengths in the X direction.

Dimension B4 is the pillar height, which is the length from the top to the bottom of the pillar 402 in the Z direction. Dimension B5 is the pillar sidewall angle, which is, for example, the angle that the sidewall makes with respect to the bottom surface of the pillar 402. The pillar sidewall angle may be defined as the angle at a given position in the Z direction.

The structures and dimensions to be measured are not limited to the example shown in Figure 4. Other examples of structures include holes and fins. Other examples of dimensions include the area of trench 403, etc.

[Dimension measurement method]
A dimension measurement method according to the first embodiment will now be described. FIG. 5 shows a flow diagram of a dimension measurement process executed by a computer system 1, which is the dimension measurement system 1 according to the first embodiment. This flow includes steps S1 to S6. The dimension measurement system 1 executes the dimension measurement method according to the first embodiment according to the flow shown in FIG. 5. In summary, a user U1 selects and displays a target image 70 on a GUI screen 60 (FIG. 6) (step S1), and manually specifies and sets an interface reference line 11 (FIG. 8) and an area reference point 12 (FIG. 9) (step S3). Based on the set information, the computer system 1 identifies the area of the structure to be measured and detects its contour (step S4), measures the specified type of dimension based on the detected contour information (step S5), and displays the dimension measurement results 8 on the GUI screen 60 (FIG. 20) (step S6).

In step S1, the computer system 1 specifies the image 7 to be measured. In this example, the user U1 operates the mouse of the operating PC 6 to move the cursor 609 on the GUI screen 60 (Figure 6), and operates the GUI such as menus and buttons to select the image file of the target image 7.

In step S2, the computer system 1 specifies the definition information 112 (Figure 2) to be applied to this dimension measurement. If the definition information 112 to be applied has already been specified and set, it is sufficient to simply refer to that setting. In this example, user U1 selects the file of definition information 112 on the GUI screen 60. The definition information 112 may be prepared in advance by the business operator as part of the software 110 (Figure 2), or may be set and saved by user U1 on the GUI screen 60. As will be described later (Figure 7), the definition information 112 defines the structure pattern of the dimension measurement target, the dimensions of the measurement target (in other words, the type of dimension), and details of the measurement of those dimensions to be applied to the target image 7.

In step S3, the computer system 1 specifies an interface reference line 11 and an area reference point 12 for the target image 70 on the GUI screen 60 (Figures 6 and 9, etc.) based on manual operation by the user U1. Multiple area reference points 12 can be specified at once as needed. Information such as the position coordinates of the interface reference line 11 and area reference point 12 set for the image 7 (also referred to as reference information) is appropriately saved in the memory resource 102 (Figure 2). For example, this reference information is saved as part of the dimension measurement result 6.

In step S4, the computer system 1 detects the contour of the region of the structure pattern to be measured in the target image 70 based on the interface reference line 11 and region reference point 12 specified in step S3 and the applied definition information 112. In embodiment 1, a distinction is made between contours and edges, and edges that make up contours are particularly detected. Edges are represented by data structures such as edge points or contour lines. Information about the detected contours (information about edge points and contour lines; also referred to as contour information) is also appropriately stored in the memory resource 102 (Figure 2). For example, the contour information is stored as part of the dimension measurement results 8.

In step S5, the computer system 1 measures the specified type of dimension in the target structure using the information on the interface reference line 11 and area reference point 12 specified in step S3 and the contour information detected in step S4. The computer system 1 stores the resulting dimension measurement result 8 data in the memory resource 102.

In step S6, the computer system 1 outputs the dimension measurement result 8 to the user U1. The computer system 1 transmits the dimension measurement result 8 to the operation PC 6, thereby displaying the image 7 and the dimension measurement result 8 on the GUI screen 60.

Below, we will explain each step in detail and provide specific examples using diagrams.

[Step S1: Target Image]
Initially, a GUI screen 60 such as the example in Fig. 6 is displayed on the display of the operation PC 6. The upper part of Fig. 6 shows the initial display state of the GUI screen 60, and the lower part shows the state in which a target image 70 is displayed in an image field 601. The GUI screen 60 includes the image field 601, a toolbar 602, tabs 603, etc. In the initial display state, a target image has not yet been specified, so no image is displayed in the image field 601. Tabs 603 such as "Semi-auto" and "Auto" are also not activated.

User U1 clicks the "Semi-auto" tab 604 and uses the "File" button 605 on the toolbar 602 to select the image file of the image to be measured. The target image is image 7 that is to be displayed in the image field 601. Operations such as clicking and selecting on the GUI screen 60 can be performed using input devices such as the mouse and keyboard of the operating PC 6 and the GUI corresponding to those operations.

The "Semi-auto" tab 604 is used to select the "semi-automatic" mode. This "semi-automatic" mode corresponds to the dimension measurement method of embodiment 1, and as described above, is a method in which the interface reference line 11 and area reference point 12 are manually specified. The "Auto" and "train" tabs are modes used in embodiment 4, etc., which will be described later.

When user U1 wishes to measure the dimensions of the target image, he or she presses the "Measure" button 606. In response to this operation, the computer system 1 transitions the screen to the state shown in the lower part of Figure 6. In the screen in the lower state, the specified target image 70 is displayed in the image field 601. In this screen, buttons such as the "Mask" button 611, "Pillar" button 612, and "Trench" button 613 are also activated. The lower state corresponds to the interface reference line setting mode described below. The "Mask" button 611 and the like are buttons provided for each type of structure.

User U1 clicks one of the "Mask" button 611, "Pillar" button 612, or "Trench" button 613 depending on the structure pattern to be measured in the target image 70. These buttons are used to select the type of structure pattern for which the interface reference line 11 and area reference point 12 are to be specified, such as mask 401, pillar 402, trench 403, etc., as shown in Figure 4. By specifying the type of target structure pattern in this way, the processing target and processing content by computer system 1 can be limited, making the process easier. Note that if there are other types of structure patterns, other buttons are similarly provided.

For example, if user U1 in Figure 1 wants to measure a trench in the target image 70, in other words, if user U1 wants to set an interface reference line 11 or an area reference line 12 for the trench, user U1 first presses the "Trench" button 613. This allows the computer system 1 to determine the type of structure pattern to be processed.

[Step S2: Definition information]
In step S2, a file of definition information 112 to be applied to the target image 70 is selected. The file of definition information 112 is a file that stores, for each target structure pattern, definition information, setting information, etc., required for dimensional measurement of the target structure pattern.

The top of Figure 7 shows an example of a file for definition information 112 in table format. In the table shown, this file has the following column items: row number (#), "pattern type," "measurement name," "measurement type," and "measurement definition information." The "pattern type" item indicates the type of structure pattern to be measured (such as the trench mentioned above). The "measurement name" item indicates the name of the measurement to be performed. This measurement name represents the target dimension as shown in Figure 4 mentioned above, and can be given any name by user U1. The "measurement type" item indicates the type of measurement to be performed. For example, the value "CD from interface" means "CD (Critical Dimension) at a position a specified distance from the interface."

The "Measurement Definition Information" item indicates the measurement definition information for each "Measurement Type." For example, a value of "0 nm from reference line" means that the measurement is performed at a position of 0 nm from the interface reference line. This corresponds to the information specifying the vertical position of dimensions A1 to A3 in Figure 4. The specified distance corresponds, for example, to distance d in Figure 18 (C) described below.

The definition information 112 required for measurement can be set and stored in a file in advance. For example, the definition information 112 file can be set on the GUI screen 60, the file to be applied can be selected, and the same information can be reused for each measurement. Alternatively, the definition information 112 can be input in step S2 each time a measurement is performed. In dimensional measurements using images of semiconductor devices, it is common to repeatedly measure the dimensions of a specific location on a sample with the same structure. When performing such measurements, if the measurement position, measurement type, and other measurement-related setting information are set once as definition information 112, there is no need to set the setting information each time a measurement is performed.

The lower part of Figure 7 shows an example of the configuration of a GUI screen for selecting definition information 112. After specifying a trench for the target image 70, for example, using the "Trench" button 613, a definition information selection GUI 701 such as a list box is displayed on the GUI screen, and user U1 can select and apply one or more definition information 112 files to be applied—in this example, the definition information for the specified row in the table shown at the top. This function is realized by cooperation between the processor 101 and software 110. Alternatively, a program for selecting definition information 112 may be stored in storage provided in the client terminal 6, and the operation shown in the lower part of Figure 7 may be performed by a processor provided in the client terminal 6. Furthermore, if a program for editing definition information 112 is stored in the client terminal 6, the definition information 112 can be pre-configured on the client terminal 6 side.

If there are multiple candidate definition information associated with a specified target structure pattern (e.g., a trench), user U1 can select one or more definition information to apply by operating the GUI screen 60. For example, in the example of definition information 112 in Figure 7, there are three candidate types for trench 403, as shown in rows numbered #3 to #5. User U1 selects one or more of these candidates to apply. In this way, it is possible to measure multiple types of dimensions simultaneously, improving processing and work efficiency.

[Specifying and setting target dimensions using definition information]
In the first embodiment, as described above, the definition information 112 (in other words, dimension measurement setting information, dimension setting information) can be used by the user U1 to specify and set the measurement target dimensions (including the dimension type and measurement position as shown in FIG. 4) for each target structure pattern in the target image. In summary, the definition information 112 is information that determines what types of dimensions are to be measured and how they are to be measured for the interface reference line 11, the area reference point 12, and the detected contour information.

The target dimensions and definition information 112 may also be specified and set on the GUI screen 60 as follows. In a modified example, the computer system 1 displays an image related to dimensions, such as that shown in FIG. 4B, as a GUI image on the GUI screen 60 based on design pattern information, etc., and also displays a table of definition information 112, such as that shown in FIG. 7. The computer system 1 displays, on the GUI screen 60, an image (e.g., an arrow image) of a dimension measurement target line representing a candidate dimension (e.g., dimension A1) for each structure pattern, such as trench 403. The computer system 1 accepts cursor or other operation by user U1 on the GUI screen 60 to select a target dimension, and sets detailed definition information for each selected target dimension in the table of definition information 112. For example, the vertical position defining dimension A3 can be set using a ratio, distance, etc.

[Step S3: (1) Interface Reference Line]
In step S3, on the GUI screen 60, the user U1 manually operates a mouse or the like on the target image 70 to first specify and set an interface reference line 11 for the structure pattern to be measured, for example, for a trench or the like specified by the "Trench" button 613 in FIG. 6. This interface reference line 11 is specified in accordance with the material interface 413 as shown in FIG. 4. The bottom of the mask 401, the top of the pillar 402, and the top of the trench 403 are located in the same vertical position as the material interface 413.

In the interface reference line setting mode, user U1 operates the interface selection GUI 61 displayed in the image field 601 to specify the interface reference line 11 at the desired position aligned with the material interface 413 in the target image 70. The computer system 1 sets the interface reference line 11 at the specified vertical (Z-direction) position in the target image 70.

The interface selection method and interface reference line setting method in embodiment 1 will be explained using Figure 8 and other figures. Figure 8 (A) shows an example of a target image 70 in the image field 601 corresponding to Figure 6, and Figure 8 (B) shows the state in which the interface reference line 11 is displayed as an interface selection GUI 61 on the target image 70 in (A).

The example of target image 70 in (A) includes a mask 401, pillar 402, and trench 403. Note that depending on the image, only some of these may be included. Note that in the drawings, the color brightness of the image is represented as a black and white schematic diagram. In this example, the vacuum region, which is the background region including trench 403, has the lowest brightness and is darkest, and is shown with a black pattern. Compared to trench 403, the brightness increases and becomes brighter in the order of the pillar 402 region and the mask 401 region. In addition, a white band 406 appears near the boundary between the background region including trench 403 and the mask 401 and pillar 402 regions. This region 406 has the highest brightness and is brightest, and is shown with a white pattern.

In the example of (B), in the initial state, the interface reference line 11 is displayed as the interface selection GUI 61 at an arbitrary position within the target image 70, for example, at the midpoint in the Z direction. The interface selection GUI 61 is displayed as a dotted straight line extending in the X direction and corresponds to the interface reference line 11. User U1 operates an input device such as a mouse to operate the cursor 609 in the image field 601 to operate the interface reference line 11 of the interface selection GUI 61. As a result, user U1 specifies the interface reference line 11 at a vertical (Z direction) position aligned with the material interface 413 (Figure 4), in this example, the interface between the mask 401 and the pillar 402.

The interface selection GUI 61 allows the user to change the position of the interface reference line 11, indicated by a dotted straight line, in the Z direction (vertical direction), for example, by clicking and dragging the cursor 609. For example, user U1 positions the cursor 609 on the interface reference line 11, clicks, and drags it to move the cursor 609 up and down, thereby changing the position of the interface reference line 11 up and down. User U1 can roughly determine the position of the material interface 413 from the difference in brightness between different materials (e.g., mask 401 and pillar 402) in image 70, and places the interface reference line 11 to align with the position of the material interface 413. User U1 determines the position of the interface reference line 11 with a predetermined operation (e.g., releasing the click button).

In this example, the interface reference line 11 is a straight line extending in the X direction as shown in the figure, and specifying the interface reference line 11 can be achieved, at a minimum, by simply specifying the height position of one point within the target image 70, reducing the amount of input required.

The computer system 1 sets the interface reference line 11 at the position specified by the above operation on the target image 70. The setting information for the interface reference line 11 includes, for example, position coordinate information in the vertical direction (Z direction) as shown in the figure. The computer system 1 displays the set interface reference line 11 in a predetermined representation. For example, the interface reference line 11 in the interface selection GUI 61 may change from a dotted line of a predetermined color to a solid line of a predetermined color. The interface selection GUI 61 is not limited to the above example. For example, up and down buttons or a vertical slide bar may be used, or a form for inputting vertical position coordinates may be used.

In this example, since the material interface 413 extends in the horizontal direction (X direction) in the cross-sectional image, the interface reference line 11 is set to extend in the horizontal direction (X direction) to correspond to the X direction. This is not limited to this, and the interface reference line 11 may extend in a direction other than horizontal depending on the actual direction of the material interface in the cross-sectional image. As a variant, the GUI screen 60 may allow the user U1 to specify and set the extension direction of the interface reference line 11.

After setting the interface reference line 11, user U1 operates the "Next" button 610 in the GUI screen 60 (Figure 6). In response to this operation, computer system 1 transitions the GUI screen 60 from the interface reference line setting mode shown in Figure 6 to the area reference point setting mode shown in Figure 9.

[Function of the interface reference line]
In the first embodiment, the interface reference line 11 is a line that serves as a support and reference for the contour detection and dimension measurement processes. In particular, the interface reference line 11 is a line that allows the user U1 to determine and specify the material interface 413. In particular, the interface reference line 11 is a line that can also be used as a reference for detecting and determining the region of a structure to be measured and the dimensions of the measurement target.

In embodiment 1, as described below, when measuring dimensions, the measurement target dimension can be determined based on the detected contour line and definition information 112, with the specified interface reference line 11 as the reference. For example, in Figure 4, the trench width, which is dimension A1, and the trench depth, which is dimension A4, are defined based on the detected contour line and in accordance with the Z-direction position of the specified interface reference line 11. For example, the position of the top edge, which is one end of the trench depth, is defined by the interface reference line 11. Furthermore, dimension A3 is defined as a position at a predetermined distance or rate below the position of the specified interface reference line 11.

[Step S3: (2) Area Reference Point]
Furthermore, in step S3, the user U1 manually specifies and sets area reference points 12 that roughly represent the area of the structure pattern to be measured for the target image 70 in the image field 601 on the GUI screen 60. At this time, one or more area reference points 12 are specified for the structure pattern of the type specified by the button using the area selection GUI 62 as a predetermined GUI.

With reference to Figure 9, a method for selecting the area reference point 12 of the area of the structure to be measured will be described. Figure 9 shows the state of the area reference point setting mode on the GUI screen 60. The bottom of Figure 9 shows an enlarged view of the target image 70 in the image field 601. In this mode, in addition to the interface reference line 11 set above, the area reference point 12 and other elements are displayed as the area selection GUI 62 on the target image 70 in the image field 601. In the initial state, the area reference point 12 is not displayed. User U1 operates the area selection GUI 62 by operating a cursor 609 or the like based on a mouse or the like, and specifies the area reference point 12 within the area, such as a trench, to be measured.

In a modified example, the area reference point 12 and the perpendicular line 13 may be displayed at any position within the target image 70 from the initial state.

In this example, region reference points 901, 902, and 903 are displayed at positions specified by user U1 as region reference points 12 in the region selection GUI 62. In this example, region reference points 901 and 902 are specified within the regions of the two masks 401 after clicking the "Mask" button 611, and region reference point 903 is specified within the region of the trench 403 after clicking the "Trench" button 613. The computer system 1 displays the region reference points 12 in a predetermined manner (e.g., circle, predetermined color, etc.) at the specified positions. As shown, the regions of the masks 401 and trench 403 are partially separated by interface reference lines 11.

Furthermore, upon designation of the area reference point 12, the computer system 1 draws a perpendicular line 13, which is a straight line perpendicular to the interface reference line 11 in the vertical direction (Z direction), from the position of the designated area reference point 12, and displays it in a predetermined manner (e.g., a dotted line). In this example, a perpendicular line 904 is drawn upward from the area reference point 903, and perpendicular lines 905 are drawn downward from the area reference points 901 and 902, respectively. In this example, the display of the perpendicular line 13 is included as part of the area selection GUI 62, but the display of the perpendicular line 13 may be omitted.

Not limited to the above example, in other examples of the area selection GUI 60, the user U1 may move and place the initially displayed area reference point 12 to the desired position by clicking, dragging, etc. with the cursor 609. Not limited to the above example, the area reference point 12 can also be specified for the pillar 402 in a similar manner. Also, if necessary, it is possible to return to the interface reference line setting mode and redo the setting of the interface reference line 11. In a modified example, the interface reference line 11 and area reference point 12 may be set simultaneously in parallel within the target image 70 in the image field 601.

In the example of Figure 9, a single point at any position contained within the area of a certain trench 403 (the inverted trapezoidal area enclosed by the interface reference line 11 as shown) is specified as the area reference point 903. This area reference point 903 indicates the area of that trench 403. A perpendicular line 13 (904) is drawn vertically upward from the area reference point 903, i.e., in the direction of the interface reference line 11. Since the perpendicular line 13 is automatically determined once the area reference point 11 is specified, there is no need for user U1 to draw the perpendicular line 13. As described above, the area selection GUI 62 requires only the operation of specifying a single point at a rough position by clicking, etc. for each structure pattern; there is no need to specify a precise position, so the operation is minimal.

In addition, multiple area reference points 12 and associated perpendicular lines 13 can be specified together within the target image 70 as needed. In this example, when two masks 401 and one trench 403 are to be measured, three area reference points 12 are specified. The computer system 1 determines the type of structure associated with the specified area reference point 12 in accordance with the type of structure specified using the "Mask" button 611, etc. The computer system 1 sets the specified area reference point 12 for each structure pattern in the target image 70.

When user U1 has finished specifying and setting the interface reference line 11 and area reference point 12, he or she presses the "Finish" button 614 on the GUI screen 60. In response to this operation, the computer system 1 ends the area reference point setting mode, thereby completing step S3. Thereafter, based on this setting information, the computer system 1 automatically performs contour detection and dimension measurement (steps S4 and S5).

[Regional reference point function]
In the first embodiment, the area reference point 12 is a point for supporting the contour detection and dimension measurement processes, and in particular, the area reference point 12 is a point for the user U1 to roughly specify the area of the structure pattern to be measured.

For example, the structure to be measured is a trench 403, and the dimension (or dimension type) to be measured is the trench width (e.g., dimension A3 in Figure 4, or dimension 1803 in Figure 18(C) described below). In this case, the area reference point 12 merely roughly indicates the area of the target trench 403. By processing using the algorithm described below based on this area reference point 12 and the interface reference line 11, the area of the trench 403 is identified and its contour is detected. The area reference point 12 does not specify the height position at which the trench width is measured, nor does it specify the two points at both ends of the dimension measurement target line that defines the trench width. Therefore, the effort required to operate the area reference point 12 is minimal.

In addition, in embodiment 1, one of the functions of the area reference point 12 is not only to indicate the area of the structure to be measured, but also to perform the following functions. That is, as shown in Figure 10 (described later), the position of the area reference point 12 in the Z direction separates an upper rectangular area from which a line profile 20 is extracted in the X direction, and a lower semicircular area from which a line profile 20 is extracted in the radial direction. The method for generating these line profiles 20 changes depending on the position of the area reference point 12.

In addition, in a modified example (described below), the area reference point 12 can also be used as a reference for determining the dimensions to be measured by the computer system 1.

[Material interfaces and dimension measurement areas]
The material interface of a structure pattern in a cross-sectional image (e.g., material interface 413 in FIG. 4 ) can serve as a reference for dimensional measurement. For example, there are cases where it is desired to measure a dimension (e.g., trench depth) starting from a material interface. Therefore, it is effective to grasp the material interface as accurately as possible. However, as described above, conventional image processing and other techniques have difficulty in determining and detecting the material interface from a cross-sectional image, and are inefficient. Therefore, in the first embodiment, a semi-automatic method is adopted, in which user U1 specifies and sets an interface reference line 11 indicating the material interface in the cross-sectional image through subjective judgment and manual operation. This allows the computer system 1 to perform dimensional measurement, etc., based on the interface reference line 11, enabling efficient processing.

Furthermore, with conventional technology, it is difficult for a computer to automatically determine and detect the area of a structure (e.g., a trench) whose dimensions are to be measured from a cross-sectional image. In typical conventional technology (e.g., Comparative Example 2), the user manually specifies the area to be measured. However, such conventional technology requires a lot of effort from the user. Therefore, in embodiment 1, a semi-automatic method is used, in which user U1 manually specifies and sets an area reference point 12 that roughly indicates the area to be measured in the cross-sectional image, and then the computer system 1 automatically identifies the area of the target structure and measures its dimensions.

[Step S4: Contour detection part 1]
In step S4, a contour detection process is performed on the structure pattern whose dimensions are to be measured, using the position coordinate information of the selected/specified interface reference line 11 and the position coordinate information of the area reference point 12. Below, the algorithm and the like of the process when contour detection is performed on the trench 403 will be described with reference to FIG.

Figure 10 is a schematic explanatory diagram of contour detection using the interface reference line 11 and area reference point 12 in the target image 70. In this example, contour detection is shown using the area reference point 903 and perpendicular line 904 for the trench 403. Figure 11 is an explanatory diagram corresponding to Figure 10, in which the image content is simplified for ease of viewing, for example by making the background area white, and supplementary information about the line profile 20, etc. is shown.

[Line profile]
The computer system 1 acquires a line profile 20 in the X direction (horizontal direction in the image), which is perpendicular to a perpendicular line 13 (904) extending upward in the Z direction (vertical direction in the image) from the area reference point 12 (903), starting from the perpendicular line 13. The line profile 20 is data configured from the luminance values of pixels on that line in the pixel data of the image 7. In the example of Fig. 10, 15 line profiles 20 numbered #1 to #15 have been acquired as the multiple line profiles 20.

Multiple line profiles 20 are obtained, for example, as follows. In Figures 10 and 11, the computer system 1 draws a perpendicular line 13 from the area reference point 12 to the upper interface reference line 11 and extracts the intersection 14 between the perpendicular line 13 and the interface reference line 11. Starting from the intersection 14, the computer system 1 first extracts line profiles 20 (#1, #15) on the interface reference line 11 in both the left and right directions in the X direction. In this example, the line segment extending from the intersection 14 to the left to the edge of the range 23 is the line profile 20 numbered #1, and the line segment extending from the intersection 14 to the right to the edge of the range 23 is the line profile 20 numbered #15. Note that these left and right line profiles 20 are set as two lines, not a single line.

Next, the computer system 1 sets multiple points (division points 15) on the perpendicular line 13 connecting the intersection point 14 on the interface reference line 11 and the area reference point 12, for example, at a predetermined interval or number. These division points 15 define the positions for extracting multiple X-direction line profiles 20. The computer system 1 similarly extracts line profiles 20 extending in the X direction within the range 23 on both the left and right sides of the X direction, starting from each division point 15. In this example, three division points 15 (Figure 11) are added, and line profiles 20 numbered #2 to #4 are obtained on the left side, and line profiles 20 numbered #12 to #14 are obtained on the right side. Through the above process, eight line profiles 20 numbered #1 to #4 and #12 to #15 are first obtained in the range above the area reference point 12.

Range 23 is the range for extracting line profile 20. Range 23 is a rectangular area extending from the area reference point 12 in the Z direction above the area reference point 12 to a predetermined length on both sides of the perpendicular line 13, and a semicircular area extending from the area reference point 12 to a predetermined radius below the area reference point 12 in the Z direction.

In the algorithm for the contour detection process in the dimension measurement method of embodiment 1, as shown in the figure, a line profile 20 extending in the X direction is set in the rectangular region above the region reference point 12, and a line profile 20 extending in the radial direction at each position in the radial direction is set in the semicircular region below the region reference point 12. In embodiment 1, these two types of line profiles 20 (in other words, brightness profile information) are used to detect the contour (particularly the edges that make up the contour) of the target structural pattern.

Next, the computer system 1 acquires multiple line profiles 20 extending in radial directions within the semicircular range 23 below the area reference point 12 (903). First, the computer system 1 acquires the line profile 20 numbered #5, which extends to the left in the X direction from the position of the area reference point 12, and the line profile 20 numbered #11, which extends to the right in the X direction. The computer system 1 then sets multiple radial directions by dividing the semicircular region from the line profile 20 numbered #5 to the line profile 20 numbered #11, with the area reference point 12 as the center point, at predetermined angular intervals or numbers. The computer system 1 then acquires line profiles 20 extending in each radial direction (in other words, the direction of the rotation angle). In this example, five line profiles 20, numbered #6 to #10, are acquired.

In addition, the two line profiles 20, numbered #5 and #11, extend in the X direction and can be said to belong to both the upper and lower ranges, so it can be said that a total of 10 line profiles 20 are obtained in the upper range, including these two.

In the example of Figure 11, when setting multiple line profiles 20, identification numbers (IDs) are assigned in the order starting from the left and wrapping around to the right, as shown by the dashed arrows, but this is not limited to this. Furthermore, the processing order of multiple line profiles 20 does not have to be limited to the order of the numbers.

The range 23 (e.g., length from the area reference point 12) and quantity (e.g., division interval and number) for obtaining the line profile 20 may be set as fixed settings in advance in the implementation of the software 110, or may be user-settable as setting values according to the structure pattern in the definition information 112.

The method of using two directional line profiles 20 in embodiment 1 corresponds to the characteristics of the target structure pattern, such as its shape. For example, a structure pattern such as trench 403 has a roughly concave or convex or trapezoidal shape. For such structure pattern shapes, extracting line profiles 20 in the X direction or the radial direction is appropriate. In the example of Figure 10, when detecting the contour of the trench 403, which has a concave, inverted trapezoidal shape, the line profile 20 is set in the radial direction (i.e., the radial direction) centered on the region reference point 12 in the range below the region reference point 12. The radial line profiles 20 (set for numbers 6 to 10 in Figure 11) pass near the sidewalls and bottom of trench 403, as shown, making it possible to extract edge candidate points 21 near the sidewalls and bottom of trench 403 (edge candidate points present on each line profile numbered #6 to #10).

In embodiment 1, two types of methods for setting the line profile 20 can be automatically selected depending on the specification of the target structure pattern or the positional relationship, such as whether the area reference point 12 is above or below the interface reference line 11 in the Z direction.

[Edge candidate extraction]
Next, the computer system 1 extracts, from each of the acquired multiple line profiles 20, edge candidate points 21 that are edge candidates related to the contour of the region of the structure whose dimensions are to be measured. The computer system 1 extracts edge candidate points 21 from one line profile 20 based on a predetermined rule. The edge candidate points 21 are points that are candidates for becoming edges that form the contour of the target structure pattern. There are cases where no edge candidate points 21 can be extracted from each line profile 20, and cases where two or more edge candidate points 21 are extracted.

Note that a distinction is made between contours and edges. In Figure 10, edge candidate points 21 extracted from the line profile 20 are shown as white and black squares. In this example, when two edge candidate points 21 are extracted, the first point is designated as edge candidate point 21a, indicated by a white square, and the second point is designated as edge candidate point 21b, indicated by a black square, for the sake of distinction. The contours of trenches 402, etc., can be ambiguous, as in the case of white bands 406. Here, an edge (end point) refers to, for example, the pixel (or the position or coordinates of that pixel) that is most likely to be the contour of a structure among the pixels that make up the white band 406, or a line formed by a set of pixels at the most likely positions. Note that the coordinates between pixels can also be defined as an edge.

The computer system 1 determines the brightness change on each line profile 20 and extracts edge candidate points 21 according to a predetermined rule. This processing for each line profile 20 only determines the change in brightness value in a single direction, for example, the X direction. This makes it easier to detect contours or edges compared to the prior art, which determines the brightness change in each direction within a two-dimensional image.

The predetermined rules are regulations regarding the direction and amount of brightness change used to determine edge candidate points 21. For example, the predetermined rules may include that the points must be areas where brightness increases from low to high, that the amount of brightness change must be greater than or equal to a threshold, and that the median value of the difference in brightness change must be used. The details of the predetermined rules are not limited.

[Extraction of edge candidate points on line profile]
12 shows a graph plotting the line profile 20 numbered #1 in FIG. 10 , with the horizontal axis representing the position coordinate in the horizontal direction (X direction) within the target image 70 and the vertical axis representing the luminance value of a pixel on the target image 70. In this example, the left side of the horizontal axis represents the side closer to the perpendicular line 13 (positive X direction), and the right side represents the side farther from the perpendicular line 13 (negative X direction). The position of the origin of the graph (coordinate x0) corresponds to the position of the intersection 14, and the position of the right end of the graph (coordinate x5) corresponds to the position where the end of the range 23 is reached.

The bottom of Figure 12 shows a graph in which the positive and negative X-axis directions of the graph at the top are reversed, and this graph has the same direction as Figure 10, etc.

On the line profile 20 numbered #1, moving left from the intersection point 14 (coordinate x0), the luminance value increases as the line passes from the trench 403 region, which corresponds to the background, to the pillar 402 or mask 401 region. At this point, the luminance value increases significantly as the line passes through the white band 406. In this example, the luminance value increases from v1 to v2. The area from coordinate x1 to coordinate x2 corresponds to the white band 406. After that, near coordinate x2, the luminance value decreases as the line enters the pillar 402 or mask 401 region. In this example, the luminance value decreases from v2 to v4. The area from coordinate x2 to coordinate x3 corresponds to the pillar 402 or mask 401 region (particularly the material interface 413). Near coordinate x3, the line passes through the white band 406 again, resulting in a significant increase in the luminance value, in this example increasing from v4 to v2. The area from coordinate x3 to coordinate x4 corresponds to the white band 406. Around coordinate x4, the area leaves the white band 406 and enters the background area again, so the brightness value decreases significantly, decreasing from brightness v2 to brightness v1 in this example.

In embodiment 1, the following rules are applied as an example of the process for extracting edge candidate points 21 on a line profile 20. In a graph such as that shown in Figure 12, the position of the edge candidate point 21 to be extracted is defined as a position between the minimum and maximum brightness values. In this example, near coordinate x1, the edge candidate point 21 is defined as a position between brightness value v1, which is the minimum value, and brightness value v2, which is the maximum value. For example, the coordinate x1 of the position corresponding to brightness value v3, which is midway between the minimum and maximum values, becomes the edge candidate point 21 indicated by the white square dot (first edge candidate point 21a). The position from which the edge candidate point 21 is extracted is defined, for example, as the midpoint between the minimum and maximum values, but is not limited to this.

The definition of which position among the brightness change portions is to be extracted as the edge candidate point 21 is not limited to the example of using the intermediate value. For example, a threshold value Th [%] relating to the rate of brightness value is set. Here, if the brightness value is B, the maximum value in the brightness change around the edge candidate is B _max , the minimum value is B _min , and the brightness value to be used as the edge candidate is B _th , then B _th is calculated using, for example, the following equation 1. The position at _{B th} is set as the edge candidate point 21.

Equation 1: B _th = (B _max - B _min )*Th/100+B _min

In Equation 1, the position of the brightness value corresponding to the ratio of the threshold value Th to the difference between the maximum and minimum values is determined as the position of the edge candidate point 21. When the threshold value Th is set to 50%, this corresponds to the example of the intermediate value in Figure 12.

The above rule for determining edge candidate points 21 is merely one example. Other examples of rules for determining edge candidate points 21 include a method of determining edge candidate points 21 at positions that are shifted a certain amount from the positions of the maximum or minimum values.

In addition, in embodiment 1, one of the rules for extracting edge candidate points 21 is that, in contour detection of the trench 403 region in FIG. 10 and other figures, edge candidate points 21 are extracted from areas where the side closer to the perpendicular line 13 is dark and the side farther from the perpendicular line 13 is bright. In other words, the rule is to determine edge candidate points 21 from areas where the line profile 20 improves and brightness increases. For example, in the graph of FIG. 12, areas with large brightness changes include near coordinates x1 and x2. Near coordinate x1, brightness value v1 in the positive X direction near the perpendicular line 13 is small, while brightness value v2 in the negative X direction farther from the perpendicular line 13 is large, resulting in an increase in brightness from low to high. Near coordinate x2, brightness value v2 in the positive X direction near the perpendicular line 13 is large, while brightness value v4 in the negative X direction farther from the perpendicular line 13 is large, resulting in a decrease in brightness from high to low.

Therefore, under this rule, of the two locations with large brightness changes, edge candidate point 21 (21a) is extracted near coordinate x1, for example, where the brightness change increases from low to high. Edge candidate point 21 is not extracted near coordinate x2, for example, where the brightness change decreases from high to low. Based on this rule, in this example, edge candidate point 21a (white square point) is extracted at the position of coordinate x1, and edge candidate point 21b (black square point) is extracted at the position of coordinate x3.

For example, near coordinate x1, the computer system 1 determines the amount of change in brightness at the point where brightness value v1, which is the minimum value, increases to brightness value v2, and calculates brightness value v3, which is the intermediate value, and extracts edge candidate point 21a at coordinate x1 corresponding to brightness value v3. The computer system 1 determines and extracts edge candidate points 21 for each line profile 20 using the same rules.

In a modified example, the rules for extracting the edge candidate point 21 may be defined as a first rule for extracting the edge candidate point 21 at a location where brightness increases and a second rule for extracting the edge candidate point 21 at a location where brightness decreases. When the first rule is applied, the edge candidate point 21 is extracted on the side of the white band 406 closer to the perpendicular line 13 (e.g., near coordinate x1), as described above. However, when the second rule is applied, the edge candidate point 21 (e.g., the position of brightness value v5) is extracted on the opposite side of the white band 406, i.e., the side farther from the perpendicular line 13 (e.g., near coordinate x2). For example, near coordinate x2, the brightness value decreases from the maximum value v2 to the minimum value v4, and the coordinate x2 corresponding to the intermediate brightness value v5 is extracted as the edge candidate point 21.

Figure 13 shows, in a modified example, edge candidate points 21 extracted for trench 403 when the applied rule is changed from the first rule to the second rule for a target image 70 similar to that of Figure 10. These edge candidate points 21 are positioned generally further outward (closer to pillar 402) than the edge candidate points 21 under the first rule in Figure 11.

Different rules may be applied depending on the type of structure pattern. For example, when the target is trench 403, a first rule that extracts points where brightness increases may be applied, and when the target is mask 401 or pillar 402, a second rule that extracts points where brightness decreases may be applied. Different edge candidate points 21 are extracted depending on the rule applied. For example, the contour line of trench 403 and the contour line of pillar 402 can be detected as separate contour lines with different position coordinates, or these contour lines can be detected as contour lines with similar position coordinates.

Which of the various rules and methods, including the rules for extracting the edge candidate points 21, to apply may be a fixed setting in advance during implementation of the software 110, or the computer system 1 may allow the user U1 to set it on the GUI screen 60. Based on the definition information 112, a rule corresponding to the structure pattern may be selected and applied.

Furthermore, in the example of Figure 10, for example, in addition to the first edge candidate point 21a (white square point), a second edge candidate point 21b (black square point) is also extracted on the line profile 20 numbered #1. This point will be explained. In the example rule above, the brightness change on the line profile 20 is examined in the direction from the perpendicular line 13 to the end of the range 23, and a location where the brightness change is sufficiently large, for example, a location where there is a significant brightness change near the white band 406, and a position corresponding to the intermediate value, is extracted as the edge candidate point 21. Depending on such rules and methods, the second edge candidate point 21b (black square point) may also be extracted, as shown in the figure.

In this example, the first edge candidate point 21a is an edge that constitutes the contour of the target trench 403, but the second edge candidate point 21b does not correspond to an edge that constitutes the actual contour of the trench 403. However, in other examples, the second edge candidate point 21b may also be an edge that constitutes the contour of the target structure pattern.

In embodiment 1, the computer system 1 obtains multiple edge candidate points 21 by extracting edge candidate points 21 on the line profile 20 in the same manner for all of the multiple line profiles 20 as described above. For example, as shown in Figure 10, multiple (e.g., 15) edge candidate points 21a indicated by white square points and multiple (e.g., 13) edge candidate points 21b indicated by black square points are obtained. The computer system 1 in embodiment 1 makes a comprehensive judgment from these multiple edge candidate points 21 and detects the edges that make up the contour of the target structure pattern.

Line profiles 20 numbered #1 to #5 and #11 to #15 correspond to the transition from trench 403 to pillar 402, and edge candidate points 21a and 21b corresponding to the sidewalls of trench 403 and pillar 402 are obtained. Note that line profiles 20 numbered #1 and #15 are on the interface reference line 11. Therefore, for example, brightness value v4 in Figure 12 is the brightness value near the material interface, and although it is a constant value in the example of Figure 12, it may actually be a fluctuating value.

For the radial line profiles 20 numbered #6 to #10, edge candidate points 21 can be similarly extracted in the direction of each line profile 20. For example, for the line profile 20 numbered #6, only one edge candidate point 21 (white square point) is extracted. For the line profile 20 numbered #7, edge candidate point 21a near the bottom of trench 403 and edge candidate point 21b corresponding to the boundary with the underlying material layer are extracted.

In addition, when extracting edge candidate points 21 for mask 401 or pillar 402, the same processing as in the case of trench 403 can be applied.

[Step S4: Contour detection part 2]
Next, an embodiment of a method for determining edges constituting the contour of a structure pattern of a dimension measurement target, for example, a trench 403, based on the plurality of edge candidate points 21 of the plurality of line profiles 20 will be described.

For each line profile 20, the computer system 1 selects arbitrary edge candidate points 21 and creates combinations of multiple edge candidate points 21. In doing so, the computer system 1 calculates an evaluation value for the contour edge-likeliness of each combination of edge candidate points 21. The computer system 1 determines the combination of edge candidate points 21 with the highest evaluation value as the contour edge (in other words, an edge point). The computer system 1 determines which edge point to select based on factors such as the positional relationship between the multiple edge candidate points 21 in the combination. The software 110 also implements an algorithm for making such a determination.

Figure 14 shows examples of combinations of edge candidate points 21. Combination #1 is a correct example. Small white circles indicate one edge point 22 selected from the edge candidate points 21 for each line profile 20. By connecting the selected edge points 22, a contour line is formed that constitutes the contour of the sidewall of the trench 403. Combination #2 is an incorrect example, in which the second edge candidate point 21b is selected in the line profile 20 numbered #1. Combination #3 is an incorrect example, in which the second edge candidate point 21b is selected in the line profile 20 numbered #2. Combination #4 is an incorrect example, in which the second edge candidate point 21b is selected in the line profiles 20 numbered #1 and #2.

Note that there are various cases where there are no edge candidate points 21 on one line profile 20, where there is only one edge candidate point 21, or where there are three or more edge candidate points 21. Even in these cases, combinations can be created in the same way.

[Evaluation of edge-likeness]
Here, there are at least two evaluation criteria for evaluating edge-likeness. The first evaluation criteria is the amount of change in brightness value. Because FIG. 10 and other figures are shown as schematic diagrams, only clear edges are visible, but in actual cross-sectional images taken with an electron microscope, the edges may be unclear. For example, due to the aforementioned collapse and contamination, minute brightness changes may occur near the contours of structures in cross-sectional images caused by dust adhesion or noise during imaging. It would be erroneous to determine an edge based on such brightness changes as edge candidate points 21.

In the first embodiment, in order to reduce the influence of these factors, it is defined that the greater the amount of change in brightness during the above evaluation, the higher the edge-likeness, i.e., the larger the evaluation value. A specific calculation of the evaluation value may be defined as the difference between the maximum and minimum brightness values in the brightness change region around the edge candidate point 21 (ΔB=(B _max - B _min )). Alternatively, it may be defined as a brightness gradient value. As an example of the evaluation of the amount of change in brightness, in the example of FIG. 12, the difference ΔB near the coordinate x1 can be calculated.

The second evaluation aspect is the distance between edge candidate points 21. Since the contours of a structure should be continuously connected, it is defined that the closer the distance between edge candidate points 21, the higher the edge-likeness. If the distance between edge candidate points 21 is far, there is a high possibility that the edge of something other than the target object, such as dust attached to the cross section of the structure, has been detected. For example, in the examples of Figures 10 and 14, the edge candidate point 21a indicated by the white square point in line profile 20 numbered #1 is close to the edge candidate point 21a indicated by the white square point in line profile 20 numbered #2, while the edge candidate point 21b indicated by the black square point in line profile 20 numbered #2 is far away. These correspond to combinations #1 and #3 in Figure 14.

Therefore, it is considered that the edge-likeness of the contour is higher when the edge candidate point 21a of the line profile 20 numbered #2 is connected to the edge candidate point 21a of the line profile 20 numbered #1 (combination #1, etc.). In other words, the computer system 1 calculates the combination so that the smaller the distance between the edge candidate points 21, the higher the evaluation value.

When calculating evaluation values quantitatively, the greater the amount of brightness change in an edge location, as in the first perspective above, the more edge-like the location is. However, based on that perspective alone, evaluation values for edge candidates in locations other than the currently focused structure may also be high. Therefore, in embodiment 1, in addition to the first perspective above, the second perspective above is also applied to calculate evaluation values for edge-likeness, in order to include contour continuity in the evaluation. Based on the evaluation results, computer system 1 selects multiple edge points 22 that are considered to have the highest edge-likeness. For example, combination #1 is selected.

To calculate the specific evaluation value for the distance between edge candidate points 21, any edge candidate points 21 are selected from each line profile 20 to create a combination as shown in Figure 14, and the evaluation value E of the contour line consisting of these edge candidate points 21 is calculated. The evaluation value E is an evaluation value of the continuity of the contour, and can be defined, for example, by the following equation 2.

Equation 2: E=Σ^(N-1)_(n=1) {ΔB _n /D _n,n+1 }

In Equation 2, ΔB _n is the difference between the maximum and minimum brightness values around the edge candidate point 21 selected in the nth line profile 20. _{D n,n+1} is the distance between the edge candidate point 21 selected in the nth line profile 20 and the edge candidate point 21 selected in the (n+1)th line profile 20. The computer system 1 selects and determines, from among all the combinations of edge candidate points 21, the combination that maximizes the evaluation value E as the edge point 22. The computer system 1 may create a contour formed by those edge points 22, for example, a contour line connecting the edge points 22 in order, and use this contour line as the contour detection result.

However, the calculation using the method for determining the edge with the maximum evaluation value E may be NP-difficult. For example, if there are 100 line profiles 20 and each line profile 20 has 10 edge candidate points 21, there are 10 to the power of 100 combinations, making it difficult to calculate the evaluation value E for all combinations. Therefore, an approximation method may be used to determine the combination of edge candidate points 21 with approximately the maximum evaluation value E. For example, an arbitrary edge candidate point 21 in the nth line profile 20 is denoted as _Pn , and an arbitrary edge candidate point 21 in the (n+1)th line profile 20 adjacent to the nth line profile 20 is denoted as Pn ₊₁ . The distance between _Pn and Pn ₊₁ is denoted as D, and the difference between the maximum and minimum brightness values around the edge candidate point 21 in Pn ₊₁ is denoted as _ΔBn . In this case, the temporary evaluation value is denoted as _Etemp and defined by the following Equation 3.

Equation 3: E _temp =ΔB/D

The computer system 1 then selects one edge candidate point 21 from the edge candidate points 21 in the first line profile 20 and determines the edge candidate point 21 in the second line profile 20 that maximizes the evaluation value E _temp for that edge candidate point 21. The computer system 1 determines the edge candidate points 21 in the third and subsequent line profiles 20 in a similar manner. The computer system 1 calculates the evaluation value E from the resulting combination of edge candidate points 21. The computer system 1 then selects another edge candidate point 21 in the first line profile 20 and performs the same process to calculate the evaluation value E from the resulting combination of edge candidates. The computer system 1 determines the combination with the highest evaluation value E among the combinations of edge candidate points 21 obtained from the edge candidate points 21 selected in the first line profile 20 as the combination of edge candidate points 21 that approximately maximizes the evaluation value E. This approximation method significantly reduces calculation time and enables automatic contour detection within a finite time. The method for determining edge points 22 is not limited to the above method.

[Edge point and contour line detection]
In the example of trench 403 specified by region reference point 12 in Figure 10, edge points 22 corresponding to 15 edge candidate points 21 (21a) indicated by white square points are obtained as the contour of trench 403. Figure 15 shows an example of the contour detection result. As shown in Figure 15, computer system 1 may form a contour line 24 of trench 403 by connecting these edge points 22 with lines. This contour line 24 corresponds to part of the contour detection result.

In the example of Figure 15, a contour line 24 is formed in the target image 70 by connecting the detected edge points 22 (shown as white circles) with straight lines. In the example of Figure 15, the contour line 24 represents the contours of the sidewalls and bottom of the trench 403 region. The top surface of the trench 403 region is open and is separated by the interface reference line 11.

Without being limited to this, the computer system 1 may, for example, create an approximate curve as a smoothing process based on multiple edge points 22, and use this as the contour line 22. Also, the contour line 24 may be omitted, and the contour may be represented using only multiple edge points 22. If the number of edge points 22 is sufficiently large, the contour can still be represented.

The computer system 1 may also display the edge points 22 or contour lines 24, which are the contour detection results, as shown in Figure 15, on the GUI screen 60, for example, on the target image in the image field 601. The computer system 1 may also display the dimension values of the dimension measurement results 8 described below, together with the contour lines 24, etc., on the GUI screen 60.

[Contour detection for masks and pillars]
While the above example shows the case of trench 403, the same is possible for pillar 402, etc. FIG. 16 similarly shows the case of contour detection (particularly edge candidate point extraction) for mask 401, and FIG. 17 shows the case of contour detection (particularly edge candidate point extraction) for pillar 402. Note that the examples of FIGS. 16 and 17 show the case where the first rule of extraction at a location of increased brightness, described above, is applied as the rule for extracting edge candidate points 21. In this case, for example, the edge of the contour of trench 403 described above and the edge of the contour of pillar 402 will result in different results.

In the example of Figure 16, a region reference point 12 is specified within a region of a certain mask 401. The computer system 1 draws a perpendicular line 13 downward from the region reference point 12 to the interface reference line 11 and extracts the intersection point 14. The computer system 1 extracts multiple X-direction line profiles 20 in the region from the region reference point 12 to the intersection point 14. The computer system 1 extracts multiple radial direction line profiles 20 in the region above the region reference point 12. The computer system 1 extracts edge candidate points 21 from each line profile 20 based on the first rule described above, etc. The computer system 1 determines edge points 22 for the mask 401 based on a combination of the multiple edge candidate points 21.

As shown in the figure, in the case of mask 401, a radial line profile 20 is applied to the area above the area reference point 12. As a result, the radial line profile 20 passes through the contour of the mask 401, which has a convex, trapezoidal shape, and edge candidate points 21 of the contour of the mask 401 can be extracted.

In the example of Figure 17, a region reference point 12 is specified within the region of a certain pillar 402. The computer system 1 draws a perpendicular line 13 upward from the region reference point 12 to the interface reference line 11 and extracts the intersection point 14. The computer system 1 extracts multiple X-direction line profiles 20 in the region from the region reference point 12 to the intersection point 14. The computer system 1 extracts multiple radial direction line profiles 20 in the region below the region reference point 12. The computer system 1 extracts edge candidate points 21 from each line profile 20 based on the first rule described above, etc. The computer system 1 determines an edge point 22 for the pillar 402 based on a combination of the multiple edge candidate points 21.

[Step S5: Dimension measurement]
In step S5, the computer system 1 performs dimension measurement processing on the contour information detected in step S4 by referring to the definition information 112 selected in step S2, obtains dimension measurement results 8, and displays them on the GUI screen 60.

Figure 18 shows a schematic explanatory diagram of an example of measuring the dimensions of a trench 403 based on the contour information of the trench 403. As a premise, the contour information in the two-dimensional target image 70 includes information on the edge points 22 and contour lines 24 as described above.

Figure 18 (A) shows the case of measuring trench depth (dimension A4 in Figure 4). Computer system 1 sets a dimension measurement target line 1801 for measuring trench depth dimension A4. First, trench depth dimension measurement target line 1801 can be defined as a line drawn perpendicular to the edge of the contour that is furthest vertically from interface reference line 11 specified in step S1 (edge point p8 in this example). Computer system 1 measures dimension A4 using the set dimension measurement target line 1801.

Figure 18 (B) shows the case where the trench width (dimension A1 in Figure 4) is measured at the position of the interface reference line 11 corresponding to the material interface 413. The computer system 1 sets a dimension measurement target line 1802 for dimension A1. The computer system 1 takes edge points p1 and p15, which are the two intersections between the interface reference line 11 specified in step S3 and the contour of the trench 403, and can define the dimension measurement target line 1802 as a line drawn between these edge points.

Figure 18 (C) shows the case where the trench width is measured at a predetermined distance d, 30 nm below the interface reference line 11 corresponding to the material interface 413. Here, d = 30 nm is the distance specified and set by user U1 in the definition information 112. As with dimension A3 in Figure 4, this can also be specified as a percentage (%) rather than a distance. Computer system 1 sets a dimension measurement target line 1803 at a predetermined distance. Based on information such as the pixel size of the target image, computer system 1 calculates how many pixels away from the interface reference line 11 the distance 30 nm corresponds to on the target image. Computer system 1 extracts two intersections (end points 1811 and 1812 in this example) between a horizontal line (a straight line extending in the X direction) at the calculated position and the contour (contour line 24) of trench 403. Computer system 1 can define dimension measurement target line 1803 as a line drawn between these two intersections.

The dimensions of the pillars 304 and the mask 303 can be measured in a similar manner. Other types of dimensions can also be measured in a similar manner. For example, the computer system 1 may measure an angle calculated from a partial area of the contour of the structure (e.g., the angle of a sidewall).

Figure 18 (D) shows the case of measuring the sidewall angle 1804 of trench 403. In this example, based on the definition information 112, the angle 1804 formed by the left sidewall is measured using a position specified a distance (e.g., 30 nm) below the interface reference line 11 as a reference. The computer system 1 calculates an approximate straight line for the sidewall from edge points (in this example, edge points p1 to p3) in a partial region of the trench 403's contour, in this example, the left sidewall, and in a region 1820 extending 30 nm below the interface reference line 11. Alternatively, the computer system 1 may refer to a portion of the contour line 24 already calculated from the aforementioned edge point 22. The computer system 1 then calculates the angle 1804 formed by the approximate straight line or contour line 24 with respect to the horizontal line at the specified position. In this way, the sidewall angle at the specified position can be measured.

As another example of another type of dimension, the computer system 1 may measure the area of the region enclosed by the structure's contour (contour line 24) and the material interface 413 (interface reference line 11), such as the trench area. The computer system 1 can measure the area of the region of the target structure based on the interface reference line 11, the region reference point 12, the definition information 112, the detected contour line 24, etc.

Figure 19 shows an example of measuring the area of a trench 403. On the GUI screen 60, user U1 specifies area as the type of measurement target dimension based on the definition information 112, and specifies the interface reference line 11 and area reference point 12 for the target image 70. Based on the specified interface reference line 11 and area reference point 12, the computer system 1 detects edge points 22 and contour lines 24 as described above. The computer system 1 extracts a closed area including the area reference point 12, which is composed of the interface reference line 11 and the contour line 24 formed by multiple edge points 22. This closed area corresponds to the area of the trench 403. The computer system 1 counts the number of pixels included in the extracted area of the trench 403, and calculates the area based on pixel information, etc. This results in an area measurement result.

[Step S6: Output of dimension measurement results]
The computer system 1 stores the data of the dimension measurement result 8, which is the result of step S5, in the memory resource 102 and displays it to the user U1 on the GUI screen 60. The dimension measurement result 8 may include position coordinate information of the interface reference line 11 and position coordinate information of the area reference point 12 specified in step S3.

Figure 20 shows an example of a GUI screen 60 displaying dimension measurement results 8. In the GUI screen 60 of Figure 20, the measurement target dimension (e.g., trench width dimension A4) of the measurement target structure (e.g., trench 403) is superimposed on the target image 70 in the image field 601. For example, a dimension measurement target line (indicated by a white arrow) for the target dimension is displayed, and the target dimension (e.g., "A4"/"Depth") and dimension measurement value (e.g., "100 nm") are displayed with extension lines added to the dimension measurement target line as necessary. Additionally, the results field 2001 below displays information such as the dimension measurement value for each target dimension in a table. As another example, information such as the position coordinates of two points defining the dimension measurement target line may also be displayed.

[Effects of the First Embodiment]
According to the first embodiment, a user U1 manually specifies and sets the interface reference line 11 and the region reference point 12, which are difficult to automatically detect, and the rough location of the structure region to be measured, in a cross-sectional image. The computer system 1 then automatically detects the contour and measures the dimensions based on the interface reference line 11 and the region reference point 12. This semi-automatic method reduces the time required for processing and performing dimension measurement on the cross-sectional image, thereby reducing the effort required by the user U1. It also reduces the impact of errors due to manual human operation. In the case of the prior art examples (Comparative Examples 1 and 2), the contour of the structure or the target dimension location is specified by hand, resulting in many errors, which affect the accuracy of the dimension measurement. In contrast, according to the first embodiment, after the contour is detected using the semi-automatic method, automatic dimension measurement is performed, resulting in fewer errors and higher accuracy of the dimension measurement.

In embodiment 1, user U1 simply specifies an interface reference line 11 for the material interface in the cross-sectional image through subjective judgment and manual operation, and roughly specifies a region reference point 12 for the region of the structure to be measured. Based on the specified interface reference line 11 and region reference point 12, computer system 1 detects edges that form the contour of the structure pattern in the cross-sectional image through processing of a predetermined algorithm. Based on the detected edge information, computer system 1 measures the specified type of dimension. Thus, according to embodiment 1, user U1's main manual operation is limited to specifying the interface reference line 11 and region reference point 12, reducing the amount of input required and the effort required by user U1. Furthermore, because computer system 1 accurately detects contours and measures dimensions based on the interface reference line 11 and region reference point 12, human error is less likely to be reflected in the dimension measurement results.

The semi-automatic method in embodiment 1 has at least the following first and second advantages compared to conventional technology examples such as automatic dimension measurement methods.

[Regarding the first advantage]
The first advantage is that it is possible to measure dimensions while excluding structural patterns that are not suitable for dimension measurement. Examples of "structural patterns that are not suitable for dimension measurement" include structures that are collapsed or have dust attached, as described above.

As an example of conventional technology, there is a method of recognizing the cross-sectional structure of a semiconductor based on pattern matching/template matching and detecting the contours/edges of the recognized cross-sectional structure. However, with this method, dimensions are measured for all objects, including dust and other contaminants, that are recognized and detected within the target image based on pattern matching/template matching. Areas caused by distortion, dust, noise during photography, etc. may also be recognized and detected as contours/edges based on pattern matching/template matching and included in the dimension measurement. Areas with such dust or other contaminants are inappropriate and should be excluded from dimension measurement.

Figure 21 shows an example of contour detection and dimension measurement results for a target image 2100 containing a structure pattern with dust attached. Assume that dust due to contamination has adhered to the cross section of the target sample. This type of dust adhesion can occur in an SEM used to observe cleaved substrates. In the example of Figure 21, a dust portion 2110 corresponding to the dust is located on the right sidewall of a trench 2102 in the target image 2100.

In (A) at the top of Figure 21, the contour edges (edge points 22 and the contour lines connecting them) of three trenches 2101-2103 are detected, and the trench widths are measured, for example, at a position 30 nm below the interface reference line 11 corresponding to the material interface. The dimension measurement target lines 2111-2113 for each trench 2101-2103 are indicated by arrows.

In this example, the dust portion 2110 is also contour-detected and measured. The dust portion 2110 attached to the right sidewall of trench 2102 protrudes into the trench 2102 area, and a contour including edge point 2105 is detected at the protruding portion. The dimension measurement target line 2112 for trench 2102 is set between the contour of the left sidewall and the contour of the dust portion 2110. Therefore, the measurement value for the dimension measurement target line 2112 is shorter than the measurement value for a trench structure (e.g., trench 2101) without dust.

In contrast, according to embodiment 1, even in the case of a target image 2100 that includes dust 2110 as shown in (A), it is possible to remove the dust 2110. (B) at the bottom of Figure 21 shows the response in embodiment 1. In the manual operation step described above, user U1 visually checks the contents of target image 2110, and if he determines that trench 2102 containing dust 2110 is inappropriate for length measurement, user U1 does not specify a region reference point 12 for that trench 2102. As a result, for example, length measurements are performed for trenches 2101 and 2103, for which region reference point 12 has been specified, but length measurements are not performed for trench 2102. In this way, portions that would result in inappropriate length measurements can be removed.

Another variation may be as follows. As in (A) of Figure 21, area reference points 12 (not shown) are specified for three trenches, and once dimensional measurement results have been obtained, user U1 checks those dimensional measurement results on GUI screen 60. If user U1 wants to exclude the dimensional measurement results for trench 403 containing dust portion 2110, he or she changes the area reference point 12 displayed in the target image from on to off by clicking or otherwise operating it. Following this operation, computer system 1 rejects the dimensional measurement results for that trench 403, and, as in (B), erases the display of the dimensional measurement results.

[Regarding the second advantage]
A second advantage is that it is possible to measure the dimensions of a structure pattern that is different from the design pattern or template pattern. An example of a "structure pattern that is different from the design pattern or template pattern" is a structure pattern such as that shown in FIG.

Figure 22 (A) shows an image 2201 of the cross-sectional structure of the target shape of etching, for example, and Figure 22 (B) shows an image 2202 of the processed shape after actual etching. (A) corresponds to the design pattern, etc. As can be seen by comparing Figure 22 (A) and (B), the structure of trench 403B shown in (B) has a shallower trench depth and deviates from the structure of trench 403A in (A). In other words, trench 403B has a larger difference in trench depth than trench 403A.

In such a case, it is difficult to match a template image using image 2201 in Figure 22(A) with image 2202 in Figure 22(B) using the template matching method of the conventional technology. In this case, the computer system cannot match, and will determine that there is no object for dimension measurement.

It is also difficult to prepare a template image equivalent to image 2202 in Figure 22(B) in advance. Generally, when observing the cross section of a semiconductor processed under certain processing conditions to confirm its structure, the target structure can be determined, but the structure that results from the actual processing cannot be known until it is examined under an electron microscope. Of course, there are many cases where an electron microscope image under optimal etching conditions is not readily available.

In contrast, according to embodiment 1, as shown in (C) of Figure 22, even without a template image, the dimensions of trench 403B (e.g., trench depth 2203) can be measured simply by specifying interface reference line 11 and area reference line 12 on image 2202.

[Multiple target structures]
According to the first embodiment, in step S3, on the GUI screen 60, the user U1 can also set the interface reference line 11 and the area reference point 12 collectively for a plurality of structure patterns in the cross-sectional image 70.

Figure 23 shows an example of setting multiple area reference points 12, etc., for multiple structure patterns in a target image 70. In the example of Figure 23, the target image 70 includes four masks 401, four pillars 402, and three trenches 403. User U1 performs an operation to specify one interface reference line 11 at the material interface between the masks 401 and the pillars 402. User U1 also performs an operation to specify area reference points 12 in each structure area. For example, user U1 operates a mouse or the like to specify multiple area reference points 12 by clicking the position of each area sequentially with the cursor 609. In this example, the area reference point 12 for the mask 401 is displayed as a triangle, the area reference point 12 for the pillars 402 is displayed as a square, and the area reference point 12 for the trenches 403 is displayed as a circle.

As such, according to embodiment 1, measuring the lengths of multiple structures requires less manual effort than conventional techniques. For multiple material interfaces that are at the same height, it is sufficient to specify a single interface reference line 11. For multiple structure regions, any position within each region can be roughly specified, and the position of the region reference point 12 does not need to be precise.

[Two or more interface reference lines]
In the first embodiment, a case where one interface reference line 11 is set for one material interface in a cross-sectional image has been described, but the present invention is not limited to this. It is also possible to set two or more interface reference lines 11 for two or more material interfaces in a cross-sectional image.

Figure 24 shows the setting of two interface reference lines 11 (11A, 11B) within a cross-sectional image 2400. In this example, the cross-sectional image 2400 has three layers of structures on a base layer: a first layer structure 2401, a second layer structure 2402, and a third layer structure 2403. Each layer is made of a different material and has a different brightness. The material interfaces include a first interface between the first layer structure 2401 and the second layer structure 2402, and a second interface between the second layer structure 2402 and the third layer structure 2403. User U1 sets interface reference lines 11 at each of the two material interfaces for this cross-sectional image 2400 on the GUI screen 60. An interface reference line 11A is specified for the first interface, and an interface reference line 11B is specified for the second interface.

User U1 can also specify region reference points 12 for regions separated by interface reference lines 11 corresponding to each material interface. In this example, region reference points 12A, 12B, and 12C are specified for the first layer structure 2401, the second layer structure 2402, and the third layer structure 2403. Contour detection and dimension measurement are possible for each region specified by the region reference points 12. For example, in the case of region reference point 12B, as shown in the lower part of Figure 24, perpendicular lines 13 are drawn from region reference point 12B to the upper and lower interface reference lines 11A and 11B, respectively. In this example, computer system 1 applies an X-direction line profile 20 to the region including region reference point 12B. Computer system 1 extracts edge candidate points 21 on each of the multiple X-direction line profiles 20 within the region. Based on this information, the computer system 1 can detect the contour of the second layer structure 2402 including the area reference point 12B and measure a specified dimension (for example, width).

Figure 25 also shows an example of setting two interface reference lines 11 (11a, 11b) at different heights when there are two material interfaces at different heights in the cross-sectional image 2500. In this example, the cross-sectional image 2500 has a structure 2501 at a first height, a structure 2502 at a second height, a structure 2503 at a third height, and a structure 2504 at a fourth height. The material interfaces include a first material interface between structure 2501 and structure 2502, and a second material interface between structure 2503 and structure 2504. The heights of the first material interface and the second material interface are different.

Using the interface selection GUI 61 on the GUI screen 60, user U1 specifies interface reference lines 11 according to the height position of each material interface. Interface reference line 11a is specified for the first material interface, and interface reference line 11b is specified for the second material interface. In this example, the interface selection GUI 61 is a GUI that allows each location within the cross-sectional image 2500 to be specified as a line segment. For example, user U1 operates the cursor 609 with a mouse or the like to specify the start and end points for each interface reference line 11. This allows the interface reference lines 11a and 11b to be drawn at the desired positions. The computer system 1 can then detect the contours and measure the dimensions of the structures 2501-2504 based on the respective interface reference lines 11 and area reference points 12.

In this way, if the height positions of the material interfaces at multiple locations within the cross-sectional image are different, it is possible to specify interface reference lines 11 at different height positions.

In the first embodiment, we have described the case where an interface reference line 11 is set at the material interface between different materials. However, this is not limited to this. In other variations, reference lines to support contour detection and the like may be set at locations other than the material interface. For example, in Figures 4 and 10, an X-direction reference line may be set to point to the bottom of pillar 402. Furthermore, if it is desired to divide an area of an arbitrary structure into multiple sections within an image, reference lines may be set to point to those sections. The computer system 1 can use these reference lines as boundaries or edges for predetermined processes such as contour detection.

[Contour detection system]
In the dimension measurement system and method of the first embodiment, the functional portion up to step S4 in FIG. 5 can be regarded as a contour detection system and method, and may be implemented as such a contour detection system and method. In this contour detection system and method, computer system 1 stores and outputs information about the contour or edge of the structure pattern detected from the target image in step S4. For example, computer system 1 displays an image including edge points 22 and contour lines 24, as shown in FIG. 15, to user U1 on GUI screen 60. Any system (e.g., observation device, inspection device, etc.) can use the contour or edge information generated and output by this contour detection system.

[Modification]
The first embodiment can be modified in various ways as follows.

In the first embodiment, the interface reference line 11 is a straight line extending in the horizontal direction (X direction) depending on the direction of the material interface in the cross-sectional image, but this is not limited to this. In a modified example, the interface reference line 11 can also be a line extending in the vertical direction (Z direction), for example, depending on the content of the cross-sectional image of the target sample.

[Modification 1: Interface Reference Line Adjustment Function]
The dimension measurement system and method of variant 1 of embodiment 1 may have an adjustment function to adjust the positional deviation if there is a positional deviation between the material interface of the structure pattern and the interface reference line 11 specified by user U1.

Figure 26 shows an explanatory diagram of the adjustment function in variant example 1 (in other words, the interface reference line adjustment function). In Figure 26, an example of a GUI screen 60 has an image field 601 as well as an adjustment field 2601 for the interface reference line 11.

The position of the interface reference line 11 specified and set on the GUI screen 60 in Figure 6 can serve as a reference for defining the region of the structure to be measured and the dimensions to be measured. This interface reference line 11 can be set by manual selection by user U1, as shown in Figure 8, or by estimation by AI, as described below. However, in cases where the boundaries between the regions that make up the cross-sectional image are unclear, the interface reference line set by a human eye or estimated by AI may not necessarily coincide with the actual material interface. If this deviation is greater than a certain level, it can have a significant impact on the accuracy of the dimension measurement.

Therefore, in variant 1, the computer system 1 adjusts the positional deviation of the interface reference line 11 using vertical brightness information of the pixels of the target image. In variant 1, when an image file is selected in step S1 of Figure 2, a GUI screen 60 like that shown in Figure 26 is displayed. The computer system 1 displays an average brightness graph 2604 (in other words, an average profile of brightness values of the y coordinate) in the adjustment field 2601 to the right of the target image 70 in the image field 601. This average brightness graph 2604 is a graph of the average brightness values for each y coordinate (vertical coordinate within the image) of the pixel data that make up the target image 70. In this example, this graph is displayed with the vertical axis on the screen representing the y coordinate of the target image 70 and the horizontal axis on the screen representing the average brightness value at each y coordinate.

In step S3 of Figure 2, assume that user U1 manually sets an interface reference line 11c on the target image 70. This interface reference line 11c has a certain degree of deviation from the actual interface between the mask and the pillar, for example, it may extend into the area on the pillar side.

If user U1 wishes to make an adjustment in this state, he or she presses the "Correct" button 2605. The computer system 1 then detects the position (position 2606) corresponding to the interface on the average profile (average graph 2604) of y-coordinate brightness values. This detection is possible based on a determination of brightness changes, as shown. The processor then corrects the interface reference line 11 to align it with position 2606. In this example, the corrected interface reference line 11d is displayed as having been moved vertically upward from the pre-correction interface reference line 11c.

As an algorithm for detecting the interface position using the average graph 2604, a method can be applied that calculates the first derivative of the profile, since the average brightness values of the profile differ between the mask 401 and pillar 402 parts. Alternatively, a method can be applied that finds the local minimum of the profile. One method for setting the search range for the interface is to limit the range to an appropriate number of pixels above and below the interface reference line 11c set by user U1.

The adjustment function in variant 1 may be included in the software 110 in Figure 2, or may be prepared as separate adjustment software and stored in storage 107, etc. According to variant 1, the impact on dimensional measurement accuracy when there is a large deviation in the interface reference line 11 can be reduced.

As another variant, in the interface selection GUI 61 (Figure 6) for setting the interface reference line 11, when displaying the interface reference line 11 as the initial state, the above adjustment function can be used from the beginning for the target image to automatically determine and display the interface reference line 11.

[Modification 2: Reference lines other than interfaces]
Modification 2 has a function of specifying and setting a location other than an interface in a structure in a cross-sectional image as a reference line.

Figure 27 (A) shows an example of setting a reference line 11C for a cross-sectional image 2701 in variant example 2, and (B) shows an example of setting a reference line 11D for a cross-sectional image 2701. In (A), reference line 11C is set below the material interface 413 in the vertical direction (Z direction), within the area of pillar 402. In (B), reference line 11D is set above the material interface 413 in the vertical direction (Z direction), within the area of mask 401.

For example, if the location where user U1 wants to measure dimensions is a location that is not based on the material interface 413, there is no need to align the position of the reference line (the aforementioned interface reference line 11) with the material interface 413.

As a specific example, in Figure 27 (A), dimension measurement is performed at a position a predetermined distance above the bottom (bottom 2705) of trench 403, for example 20 nm. The dimension measurement target line 2710 at that position is shown. In this case, reference line 11C is specified at a position below the material interface 413. Also, although not shown, an area reference point 12 is specified below reference line 11C. Based on these reference line 11C and area reference point 12, computer system 1 detects the outline of the area corresponding to trench 403, and measures the dimension at the position of dimension measurement target line 2710 based on definition information 112.

Similarly, in Figure 27 (B), when measuring dimensions at the same position as in (A), reference line 11D is specified at a position above material interface 413. Similarly, in (B), dimensions can be measured at the position of dimension measurement target line 2710. In both (A) and (B), the measurement position is the same, and the dimension value at dimension measurement target line 2710 is the same. In either case, there is no problem because the dimension at the position of dimension measurement target line 2710 is not determined based on the positions of material interface 413 and reference lines 11C and 11D.

In the example of Figure 27, the bottom (bottom 2705) of trench 403 can be determined based on the position coordinate information of the contour edge detected by the method described above. For example, of the edge points 22 described above (Figure 15), the edge point 22 with the lowest position in the Z direction may be determined as the bottom 2705 of trench 403. Alternatively, the average value of multiple edge points 22 that are close in position in the Z direction may be determined as the bottom 2705. A dimension measurement target line 2710 can be set at a position, for example, 20 nm above the bottom 2705 in accordance with the definition information 112.

As in variant example 2, the reference line specified in step S3 may be significantly deviated from the material interface 413, and may also be intentionally specified at a location other than the material interface 413. In variant example 2, reference lines 11C and 11D do not have the function of indicating the material interface 413, but rather have the function of roughly indicating a portion of the target trench 403 region (e.g., the top).

Incidentally, (C) at the bottom of Figure 27 shows an example of setting measurement target dimensions such as those shown in (A) and (B) of Figure 27 in the definition information 112. Such dimensions can also be set and selected on the GUI screen 60 mentioned above.

[Modification 3: Area Reference Point]
The following modifications are possible regarding the function of the area reference point 12.

Figure 28 is an explanatory diagram of variant example 3. In Figure 28, when specifying a region reference point 12 for trench 403 in a target image 70 similar to that in Figure 10, the region reference point 12 is specified at a higher vertical position than in the example in Figure 10. This region reference point 12 roughly indicates the region of trench 403.

In Variation 3, rather than setting an X-direction line profile 20 and a radial-direction line profile 20 with the position of the area reference point 12 as the boundary as in Figure 10, these two types of line profiles 20 are set using a different predetermined method. In Figure 28, the computer system 1 not only draws a perpendicular line 13 from the area reference point 12 toward the interface reference line 11, but also draws a vertical line 2801 (shown as a dashed line) in the opposite negative Z direction. For example, as a pre-defined setting, the computer system 1 sets a reference point 16 (shown as a diamond dot) on the perpendicular line 13 and the vertical line 2801 at a predetermined distance below the intersection 14 of the interface reference line 11 in the Z direction. The rectangular area 2810 from the intersection 14 to the reference point 16 is determined as the first area from which the X-direction line profile 20 is extracted, and the semicircular area 2820 further below the reference point 16 is determined as the second area from which the radial-direction line profile 20 is extracted.

The distance from the intersection 14 to the reference point 16 may be set by the software 110 or the definition information 112. Furthermore, the interval and number of line profiles 20 extracted in the first and second regions may also be set by the software 110 or the definition information 112.

In variant example 3, two types of line profiles 20 are generated regardless of the Z-direction position at which the area reference point 12 is specified. In the example of Figure 23, the reference point 16 is set below the area reference point 12, but this is not limited to this, and the same applies even if the reference point 16 is set above the area reference point 12.

[Modification 4: Area Reference Point]
In the fourth modification, the area reference point 12 has the function of not only roughly indicating the area of the structure to be measured, but also of specifying the position that defines the dimensions of the object to be measured.

Figure 29 shows an example of the target image 70, interface reference line 11, area reference point 12, dimension measurement target line 2901, etc. in variant example 4. The specified area reference point 12 not only roughly indicates the area of trench 403, but also specifies the vertical position for measuring the trench width (dimension A3 in Figure 4) as shown by the dimension measurement target line 2901. In other words, in this example, the vertical position for measuring the trench width (dimension A3) as shown by the dimension measurement target line 2901 is specified not by the distance or rate in the definition information 112 described above, but by the position of the area reference point 12 manually operated by user U1.

The computer system 1 detects the contour of the trench 403 based on the specified interface reference line 11 and area reference point 12, sets a dimension measurement target line 2901 for measuring the trench width (dimension A3) at the position of the specified area reference point 12, and measures the dimension A3.

In another example, the area reference point 12 can also specify an X-direction position relative to the measurement target dimension, such as trench depth (dimension A4 in Figure 4).

In this variant example 4, the position of the area reference point 12 is specified precisely, which increases the burden on the user slightly, but compared to the method of specifying two points for the measurement target dimension as in comparison example 2, only one point needs to be specified.

[Modification 5]
21, in step S3 of Fig. 5, the user U1 looks at the cross-sectional image on the GUI screen 60 to check for inappropriate areas such as dust, and excludes such areas from the target by not specifying the area reference point 12. Thereafter, the computer system 1 automatically executes contour detection and dimension measurement, and outputs the results.

In variant 5, an additional step of confirmation by user U1 may be provided between the contour detection process of step S4 in Figure 5 and the dimension measurement process of step S5. This variant 5 has the following processing and operation. In variant 5, after the contour detection process of step S4 in Figure 5, a step is provided in which user U1 confirms the contour detection results on the GUI screen 60. In this additional step, the computer system 1 displays the contour detection results of step S4 superimposed on the cross-sectional image on the GUI screen 60.

Figure 30 shows an example of displaying contour detection results for the target image 3000 in an additional step in variant example 5. In Figure 30, for example, if area reference points 12 (12c, 12d, 12e) are specified for three trenches 3001 to 3003, the contours (edge points 22, etc.) of those three trenches are detected and displayed as contour detection results. User U1 checks and decides whether to perform dimensional measurement on these contour detection results as they are, and if so, presses a dimension measurement execution button (not shown) on the GUI screen 60. This causes the computer system 1 to perform dimension measurement processing based on the contour detection results and display the dimension measurement results.

Alternatively, user U1 can specify a portion of the contour detection results to be excluded from dimension measurement. For example, by re-specifying the displayed area reference point 12 with cursor 609 or the like, the area reference point 12 can be changed from an ON state to an OFF state. For example, trench 3002 contains a dust area 3010, which is reflected in the contour detection results. To exclude trench 3002 from the measurement target, user U1 clicks or otherwise operates area reference point 12d to turn it OFF. For example, the display of area reference point 12d changes from an ON state to an OFF state. User U1 then presses the execute dimension measurement button. This causes the computer system 1 to exclude trench 3002, including the OFF-state area reference point 12d, from the measurement target, execute dimension measurement, and display the dimension measurement results.

The dimension measurement results are displayed, for example, as shown in (B) at the bottom of Figure 30. For trenches 3001 and 3003, the trench width, for example, is displayed as the dimension measurement result. In this way, in variant example 5, user U1 can determine the dimension measurement target after viewing the contour detection results.

<Second Embodiment>
A dimension measurement system and the like according to the second embodiment will be described with reference to FIG. 31 and the like. The basic configuration of the second embodiment and the like is the same as that of the first embodiment, and the following mainly describes the components of the second embodiment and the like that are different from those of the first embodiment. In the first embodiment, as shown in FIG. 10 and the like, a method for detecting the edge of the contour of a structure was described using two types of line profiles 20, namely, a line profile 20 extending in the X direction and a line profile 20 in the radial direction. However, this is not limiting, and there are cases where edge detection using the line profile 20 in the radial direction is not necessary. In the second embodiment, a method for detecting the edge of the contour is described using only the line profile 20 extending in the X direction (lateral direction) without using the line profile 20 in the radial direction.

[Contour detection using only horizontal line profiles]
31 shows an example of detecting edges constituting the outline of a pillar 402 using only a line profile 20 extending in the horizontal direction according to the second embodiment. (A) of Fig. 31 shows an example of an interface reference line 11, an area reference point 12, a dimension measurement target line 3101, etc. in a certain target image 3100. (B) of Fig. 31 shows an example of setting a line profile 20 in the X direction based on (A).

As shown in (A), in the structure of pillar 402, the pillar width (dimension B3 in Figure 4) is measured at a position a predetermined distance, for example 30 nm, below the interface reference line 11 corresponding to the material interface in the Z direction. In this case, the dimension measurement target line 3101 for that pillar width is shown. User U1 specifies the interface reference line 11 in line with the material interface, and specifies the area reference point 12 within the area of pillar 402 to be measured.

In this example, user U1 specifies the area reference point 12 at any position in the Z direction below the dimension measurement target line 3101. However, here, the Z direction position of the area reference point 12 is assumed to be lower than the Z direction position at which the pillar width, which is the measurement target dimension, is measured (condition A).

In this case, the computer system 1 draws a perpendicular line 13 from the specified area reference point 12 to the interface reference line 11 and extracts the intersection point 14, as shown in (B). The computer system 1 extracts multiple line profiles 20 extending in the X direction in the area 3110 between the area reference point 12 and the intersection point 14. The line profiles 20 are straight lines perpendicular to the perpendicular line 13 and the direction of a predetermined distance (30 nm). The method for generating the X-direction line profiles 20 is the same as in embodiment 1. In this example, ten line profiles 20 numbered #1 to #10 are extracted.

The computer system 1 uses the acquired multiple X-direction line profiles 20 to detect the edges that make up the contour of the pillar 402 in a manner similar to that of embodiment 1. In this example, multiple edge candidate points 21 indicated by white square points and edge candidate points 21 indicated by black square points are detected in areas where brightness decreases on the line profile 20. Edge points 22 are determined for the multiple edge candidate points 21 based on the evaluation described above. For example, the multiple edge candidate points 21 indicated by white square points are determined as edge points 22 that make up the contour line of the pillar 402.

Based on this detected edge information and definition information 112, the computer system 1 can measure the pillar width (dimension B3) for the dimension measurement target line 3101 at a position on the pillar 402 a predetermined distance (30 nm) below the interface reference line 11 corresponding to the material interface. Thus, in this example, edge detection using the radial line profile 20 is not required.

Condition A can also be satisfied as follows: The computer system 1 displays, within the target image 3100, a guideline image corresponding to the dimension measurement target line 3101 at the position where the pillar width, which is the measurement target dimension, is measured based on the definition information 112 and the interface reference line 11, and accepts the specification of an area reference point 12 at a position below the dimension measurement target line 3101. By looking at the guideline image of the dimension measurement target line 3101, user U1 can specify the area reference point 12 at a position below it.

Alternatively, the following may be done to accommodate cases where user U1 specifies an area reference point 12 above the position where the pillar width is measured. The computer system 1 not only draws a perpendicular line 13 upward from the specified area reference point 12, but also draws a vertical line downward from the area reference point 12. The computer system 1 sets a range on the perpendicular line 13 and that perpendicular line for extracting the X-direction line profile 20. For example, the computer system 1 may set a range for extracting the X-direction line profile 20 from the intersection point 14 to a position a set distance below, or from the area reference point 12 to a position a set distance below. This satisfies condition A.

As in the above example, in embodiment 2, contour detection and dimension measurement can be achieved by omitting the radial line profile 20 and performing edge detection using only the X-direction line profile 20, depending on the area reference point 12, the structure to be measured, and its dimensions.

Embodiment 2 may be implemented with only the function of edge detection using only the horizontal line profile 20. Alternatively, embodiment 2 may be implemented with both the function of edge detection in the X direction and the function of edge detection in the radial direction (described in embodiment 3 below), with the computer system 1 or user U1 selecting which function to use. For example, the computer system 1 may receive specification of definition information 112 and a measurement target structure (e.g., a pillar) on the GUI screen 60, as well as specification of the interface reference line 11 and area reference point 12, and automatically determine which function to use based on the relative positions of these, etc.

<Third Embodiment>
The third embodiment shows a method for detecting the edge of a contour using only the line profile 20 in the radial direction, without using the line profile 20 in the horizontal direction.

[Contour detection using only radial line profiles]
32 shows an example of detecting edges constituting the contour of trench 403 using only radial line profile 20 in embodiment 3. Fig. 32 shows an example of interface reference line 11, area reference point 12, dimension measurement target line 3201, radial line profile 20, etc. in a certain target image 3200.

For example, when detecting the contour and measuring the dimensions of a shallow trench structure (trench 403 with a relatively small trench depth) as shown in Figure 32, the region reference point 12 may be specified and placed on the interface reference line 11 corresponding to the material interface. In this case, as shown, the aforementioned perpendicular line 13 does not exist. Therefore, in embodiment 3, the computer system 1 sets only radial line profiles 20 in the semicircular range 3210 below the region reference point 12. In this example, seven line profiles 20, numbered #1 to #7, are set. The computer system 1 detects the contour edges of the trench 403 on the radial line profiles 20. In this example, edge candidate points 21 are detected on each line profile 20. The aforementioned edge points 22 are determined based on the edge candidate points 21. The contour line of the trench 403 can be formed based on the edge points 22.

In addition, in Figure 32, the computer system 1 may consider the area reference point 12 to be specified on the interface reference line 11 and apply edge detection in the radial direction, even if the vertical positions of the interface reference line 11 and the area reference point 12 specified by the user U1 do not match, if the difference between their positions is small within a specified range.

In this example, the trench depth (dimension A4 in Figure 4) is used as an example of the measurement target dimension, and the corresponding dimension measurement target line 3201 (indicated by the white arrow) is shown. Based on the detected edge information and definition information 112, the computer system 1 can measure the trench depth for the dimension measurement target line 3201 set between the interface reference line 11 corresponding to the top of the trench 403 and the bottom of the trench 403 based on the contour line. Thus, in this example, edge detection using the X-direction line profile 20 is not necessary.

As with embodiment 2, embodiment 3 may be implemented in a form that only implements the function of edge detection in the radial direction, or it may be implemented in a form that has both functions and allows the user to select the function to use, or it may be implemented in a form in which the computer system 1 automatically determines the function to use.

<Fourth Embodiment>
An estimation system and method according to a fourth embodiment will be described with reference to Figure 33 and the like. In the fourth embodiment and the fifth embodiment described below, machine learning using AI is added to the dimension measurement system and method according to the first embodiment and the like. In the fourth and fifth embodiments, AI machine learning is applied to the step of specifying and setting the interface reference line 11 and the area reference point 12 by manual operation by the user U1 in step S3 of Figure 5.

In particular, the estimation system and method of embodiment 4 is a system and method that separates only the machine learning and estimation parts for the interface reference line 11 and area reference point 12 in the cross-sectional image. The estimation system and method of embodiment 4 additionally includes a step in which computer system 1 performs machine learning on the interface reference line 11 and area reference point 12 for the cross-sectional image, and estimates the interface reference line 11 and area reference point 12 from the input target image based on the learned model and presents them to user U1. Embodiment 5, described below, corresponds to a form in which the functional parts of embodiment 4 are added to the dimension measurement system and method of embodiment 1.

This makes it easier to identify material interfaces and structure pattern regions from cross-sectional images in embodiment 4. Embodiment 5 further reduces the effort required for the user U1 to manually input, specify, and set the interface reference line 11 and region reference point 12, thereby further improving the efficiency of the entire process.

In other words, the system in embodiment 4 is an interface reference line and area reference point estimation system, a material interface estimation system, a dimension measurement target structure area estimation system, a measurement target dimension estimation system, etc.

[Machine learning target]
In the fourth embodiment, the following three forms are possible for the target of machine learning. In the fourth embodiment, the following case 3 (FIG. 34) will be particularly described, but the present invention is not limited to this, and the following cases 1 and 2 are also possible, as shown in a modified example (FIG. 38).
1. A form in which machine learning and estimation are performed only on the interface reference line 11.
2. A form in which machine learning and estimation are performed only on the area reference points 12.
3. A form in which machine learning and estimation are performed on both the interface reference line 11 and the area reference point 12.

In embodiment 4, computer system 1 constructs a DB containing a machine learning model using cross-sectional images and information contained in the dimensional measurement results of embodiment 1, including position coordinate information for interface reference lines 11 and information including position coordinate information for area reference points 12. By utilizing image recognition technology using machine learning, etc., computer system 1 realizes estimation of material interfaces (and corresponding interface references 11) and area of a structure to be measured (and corresponding area reference points 12), which are difficult to detect using ordinary image recognition.

In addition, the dimension measurement result 8 (Figure 2) in embodiment 1 includes information on the interface reference line 11 and area reference point 12 set in step S3. Alternatively, the information on the interface reference line 11 and area reference point 12 may be stored in a form associated with the information on the dimension values, etc. in the dimension measurement result 8. In embodiment 4, this information is used.

[Computer System]
Fig. 33 shows the configuration of a computer system 1D, which is an estimation system according to embodiment 4. The configuration of Fig. 33 is roughly the same as that of Fig. 2, but differs as follows: The computer system 1D has a function of estimating the interface reference line 11 and the area reference point 12 in a cross-sectional image using AI machine learning.

Software 110D including a learning and estimation program is stored in memory resources 102 of computer system 1D, for example RAM 106. This software 110D is data that realizes processing functions including AI machine learning and estimation in embodiment 4. Computer system 1D stores DB 140 for AI machine learning, etc. in storage 107. DB 140 is data handled by software 110D, and includes learning data 141 related to machine learning, etc., and learning model data 142.

The learning data 141 is a dataset for training AI, in other words, a dataset necessary for machine learning, a dataset for training a model, and includes data such as the aforementioned image 7. The learning model data 142 is a dataset for a machine learning model, is data for a model generated based on the learning data 141, and includes data on learning parameters optimized through learning. The contents of the learning data are as shown in Figure 34, etc.

[Machine Learning and Inference]
FIG. 34 shows the configuration of the input/output relationship for the machine learning learning model 200 in the fourth embodiment. In the learning phase, the computer system 1D constructs and trains the learning model 200 using learning data 141. The learning model 200 is a model in which learning parameters included in learning model data 142 are set. The learning data 141 is data including pixel data of the cross-sectional image, coordinates of the interface reference line 11, coordinates of the area reference point 12 (if there are multiple points, they are designated as points #1, #2, ...), etc. The learning data 141 can be data in which the user U1 has specified the interface reference line 11 and the area reference point 12 in the target image, which has been used and saved in the system of the first embodiment in the past.

In the estimation phase, computer system 1D inputs input data 201, including cross-sectional images that are actually the subject of dimensional measurement, etc., into the latest trained learning model 200. Computer system 1D performs estimation processing on the input data 201 using learning model 200, and outputs the estimation results as output data 202. Output data 202 is data on features estimated from the input target image, and includes the coordinates of the estimated interface reference line 11 (referred to as estimated interface reference line 11X), the coordinates of the estimated area reference point 12 (referred to as estimated area reference point 12X), etc.

The computer system 1D or any other system can optionally use the estimation result 3302 (Figure 33), which is the output data 202. For example, the computer system 1D may display the estimated interface reference line 11X and estimated area reference point 12X, which are the estimation result 3302, on the target image on the GUI screen 60 and present them to the user U1 as suggested or recommended information. The user U1 can check the estimated interface reference line 11X and estimated area reference point 12X on the GUI screen 60 and use them as is or modify them.

The processor 101 of the computer system 1D in Figure 33 executes all processing in accordance with the software 110D deployed from the storage 107 to the RAM 106. This realizes the processing functions related to learning and estimation as shown in Figure 34. The image data 3301 is image data newly input to the AI. For example, image 7 is acquired from the SEM 2 to the operation PC 6, and the image 7 is set as the image data 3301. The operation PC 6 sends the image data 3301 to the computer system 1D. The computer system 1D stores the image data 3301 as part of the learning data 141. Without being limited to this, for example, the image data 3301 of image 7 acquired by the computer system 1D from the SEM 2 may be used as learning data. Image data stored in another external storage device or DB server, etc., external to the computer system 1D may also be used as learning data.

The estimation result 3302 is data of the estimation result by the AI learning model 200. The computer system 1D generates, stores, and outputs the estimation result 3302 as output data 202 of the learning model 200 for the target image of the input data 201 (for example, an image specified from the operation PC 6). The computer system 1D transmits the estimation result 3302 to the operation PC 6. The operation PC 6 displays the estimated interface reference line 11X and estimated area reference point 12X as the estimation result 3302 on the cross-sectional image corresponding to the input target image on the GUI screen 60 (Figure 37 described below).

[Accumulation of learning data]
The computer system 1D stores learning data 141. To this end, related information such as image data and shooting information 111, coordinates of interface reference lines 11 included in past dimension measurement results 8, and coordinates of area reference points 12 are stored in the learning data 141 of the DB 140 from the SEM 2 and the operation PC 6. In the following, it is assumed that the learning data 141 has already been stored in the DB 140 of the storage 107.

[Generation of learning model]
The computer system 1D constructs and generates a learning model 200 based on the learning data 141 in the DB 140. FIG. 35 shows an example of the configuration of the machine learning learning model 200. In this example, a CNN (convolutional neural network), an example of a neural network, is used as the learning model 200. This CNN propagates pixel information 3501 of an image input to the input layer through the intermediate layer and the output layer in order, and performs calculations. As a result, this CNN outputs estimated interface coordinates (coordinates of the estimated interface reference line 11X) and one or more structure area coordinates (coordinates of the estimated area reference point 12X) from the output layer. In the intermediate layer, multiple convolutional layers and the like are repeated. The interface coordinates are estimated values of position coordinates of the interface reference line 11 representing the material interface 413. The structure area coordinates are estimated values of position coordinates of the area reference point 12 representing the area of the structure pattern that is the target of dimension measurement, etc.

In the learning phase of Figure 34, the parameters of the intermediate layer (included in the learning model parameters 142) are adjusted so that the error between the interface coordinates and structure area coordinates output by the learning model 200, which is a CNN, and the correct interface coordinates and structure area coordinates is minimized.

The specific layer structure of a CNN varies depending on the learning model used, and details are not limited. In embodiment 4 (Figure 35), a CNN is shown as the learning model 200 to be applied, but the application is not limited to this. Any machine learning model that can output interface coordinates and structure area coordinates in response to input image information such as pixel data is applicable.

[Learning Phase Flow]
36 shows a learning process flow for the learning phase in the fourth embodiment, and includes steps S41 to S45. In step S41, user U1 operates the GUI on GUI screen 60 on the display of operation PC 6 to load data of a target image and the dimension measurement results 8 of the target image into computer system 1D. In step S42, computer system 1D registers and saves data, including the coordinates of interface reference line 11 and the coordinates of one or more area reference points 12, contained in the data of the target image and dimension measurement results 8 loaded in step S41 in DB 140 as learning data 141.

In order to train the AI learning model 200, sufficient training data is required to achieve accuracy. Therefore, the computer system 1D does not train the learning model 200 until the amount of registered and accumulated training data reaches or exceeds a predetermined amount. That is, in step S43, the computer system 1D determines whether the amount of training data accumulated as training data 141 has reached or exceeds a predetermined amount. If the amount has not yet reached or exceeds the predetermined amount (N), the computer system 1D returns to step S41 and repeats the process; if the amount has reached or exceeds the predetermined amount (Y), the computer system 1D proceeds to step S44.

In step S44, the computer system 1D acquires the pixel data of the target image, the interface reference line coordinates, and one or more area reference point coordinates in the learning data 141 stored in storage 107 as training data for the learning model 200 (learning data 141 in Figure 34).

Then, in step S45, the computer system 1D uses the teacher data acquired in step S44 (learning data 141 in Figure 34) to learn and train the learning model 200, and generates and updates the learning model. As described above, in learning, pixel data of the target image of the teacher data is input as input data 201 to the learning model 200 (CNN in Figure 35), and learning is performed so that the interface coordinates and structure area coordinates in the output data 202 output as the estimation result are close to the coordinates of the correct interface reference point 11 and area reference point 12 in the teacher data.

Examples of GUIs on the GUI screen 60 related to the learning phase are as follows: In the GUI screen 60 of Figure 6 described above, a "train" tab is provided as one of the tabs 603. To perform model learning, user U1 clicks on the "train" tab or performs other operations. In response to this operation, computer system 1D switches the content/mode of GUI screen 60 to the model learning GUI screen. On the model learning GUI screen, user U1 operates the "File" button and, using a mouse or other device, selects the file of the target image to be used for learning, and then operates the "Enter" button to load the data from the selected file. In addition to the pixel data of the selected target image, computer system 1D also loads various data associated with the image, such as the interface reference line 11 and area reference point 12 in the dimension measurement results 8.

[GUI screen: estimation phase]
FIG. 37 shows an example of the GUI screen 60 displayed in the estimation phase. FIG. 37 is an example of a screen for the estimation function. This screen has an estimation type field 3701, in which interface coordinates (in other words, interface reference line coordinates) and structure area coordinates (in other words, area reference point coordinates) can be selected as the estimation target by turning them on or off. In the estimation target image field 3702, the input estimation target image is selected and displayed. In the estimation result image field 3703, the output estimation result information is displayed superimposed on the estimation target image. In this example, the estimation result image field 3703 displays an estimated interface reference line 11X and an estimated area reference point 12X related to the mask and trench. The type of structure to be estimated may also be selectable on this screen.

[Effects of the Fourth Embodiment]
According to the estimation system and method of the fourth embodiment, machine learning can be used to automate the manual setting steps in the semi-automatic method of the first embodiment, thereby reducing the manual operation effort required by the user U1.

Of course, it is also possible to add various functions and steps from embodiments 1 to 3 to the system and method of embodiment 4.

[Modification (1)]
The following modifications of the fourth embodiment are possible. In the fourth embodiment, as shown in Fig. 34, the learning by the learning model 200 is performed for both the interface reference line 11 and the area reference point 12. In contrast, in the modification, the learning is divided into two parts: the learning of the interface reference line 11 and the learning of the area reference point 12.

Figure 38 shows learning models etc. in a modified example. (A) shows the first learning model 200A, and (B) shows the second learning model 200B. In this modified example, for example, the coordinates of the interface reference line 11 representing the material interface are learned using the first learning model 200A, and the coordinates of the area reference point 12 representing the area of the structure pattern that is the target of dimensional measurement etc. are learned using the second learning model 200B. The structure of learning models 200A and 200B can be applied to a CNN, for example, as in Figure 35.

In the learning phase, the first learning model 200A (A) is learned using learning data 141A including the coordinates of the interface reference line 11. The second learning model 200B (B) is learned using learning data 141B including the coordinates of one or more area reference points 12.

In the estimation phase, a target image is input as input data 201A to the first learning model 200A (A), and data including the coordinates of the estimated interface reference line 11X is output as output data 202A. A target image is input as input data 201B to the second learning model 200B (B), and data including the coordinates of the estimated area reference point 12X is output as output data 202B.

As a variant, the system and method may be one that only performs learning and estimation using the first learning model 200A (A), or one that only performs learning and estimation using the second learning model 200B (B).

[Modification (2)]
As another modification of the fourth embodiment, the following is possible. In this modification, a function of estimating base point coordinates for dimension measurement is provided. In this modification, the estimation target is not the coordinates of the area reference point 12 of the structure pattern area, but the position of the measurement target dimension in the structure pattern area. In particular, the estimation target is the start point coordinates and end point coordinates of the measurement target dimension, in other words, the coordinates of two points that define the dimension measurement target line.

Figure 39 shows the learning model 200C etc. in this modified example. The learning data 141C in the learning phase includes data on the coordinates of the start and end points of the measurement target dimensions of the target image. The output data 202C output as an estimation result from the learning model 200C in the estimation phase includes data on the coordinates of the start and end points of the measurement target dimensions.

Figure 40 shows an example of displaying estimation results on the GUI screen 60 in this modified example. In this example, "Coordinates of start and end points of dimension" is selected in the estimation type field 4001. In addition, in the dimension type field 4004, for example, trench width (middle CD) is selected as the dimension type of the estimation target. In the estimation target image field 4002, the estimation target image is displayed. In the estimation result image field 4003, the coordinates of the start point 4011 and end point 4012 of the measurement target dimension as the estimation result are plotted on the estimation target image and displayed together with the dimension measurement target line 4010.

<Fifth Embodiment>
A dimension measurement system according to the fifth embodiment will be described with reference to Fig. 41 etc. The fifth embodiment is a combination of the first and fourth embodiments, and is provided with both the semi-automatic contour detection and dimension measurement functions described above and an estimation function based on machine learning.

In the learning phase, the system and method of embodiment 5 use the functions of the system of embodiment 1 to allow user U1 to specify the interface reference line 11 and area reference point 12 for the target image, perform contour detection and dimension measurement semi-automatically, and accumulate the data including the interface reference line 11 and area reference point 12 obtained during this process as learning data 141. The system and method of embodiment 5 uses this learning data 141 to train the learning model 200.

In the estimation phase, the system and method of embodiment 5 use the functions of the system of embodiment 4 to output, as estimation results, an estimated interface reference line 11X representing the material interface and one or more estimated area reference points 12X representing the area of the structure to be measured, for the input target image using the learning model 200. The system and method of embodiment 5 then performs contour detection and dimensional measurement of the target image based on the estimation results. The system and method of embodiment 5 may also display the estimated interface reference line 11X and estimated area reference points 12X to user U1 on the GUI screen 60, allowing user U1 to confirm and correct them.

[Dimension measurement system]
41 shows a configuration example of a computer system 1E, which is a dimension measurement system according to a fifth embodiment. This computer system 1E is a system in which the computer system 1 in FIG. 2 and the computer system 1D in FIG. 33 are combined into one. This computer system 1E automatically accumulates dimension measurement results obtained by a semi-automatic method as learning data 141, and uses a learning and estimation function to learn and estimate the interface reference line 11 and the area reference point 12. This computer system 1E displays estimation results 4102 (estimated interface reference line 11X and estimated area reference point 12X) obtained by the learning model on a GUI screen 60, performs automatic dimension measurement using the estimation results 4102, and displays the dimension measurement results 4102(8) on the GUI screen 60.

In the memory resource 102 of the computer system 1E, for example, in RAM 106, software 110 including the dimension measurement program of embodiment 1 and software 110D including the learning and estimation program of embodiment 4 are stored. The processor 101 realizes each function by performing processing in accordance with software 110 and software 110D. Software 110 also realizes the function of automatically measuring dimensions using the estimation results 4102 by software 110D. Furthermore, in embodiment 5, depending on the mode setting, user U1 can select whether to use manual setting using the semi-automatic method of embodiment 1 or automatic estimation using the estimation function of embodiment 4.

In addition to the same data and information as in embodiment 1, storage 107 also stores the estimation results 4102 generated by software 110D. Another storage 107D is equipped with DB 140 similar to that in embodiment 4. Storage 107 and storage 107D may be combined into one.

As in embodiment 4, the target image 7 and its associated shooting information 111 and definition information 112 can be input to the computer system 1E from the SEM 2 or the operation PC 6 and saved in the storage 107 or storage 107D. The computer system 1E transmits the estimation results 4102 and the dimension measurement results 4101(8) to the operation PC 6, and the operation PC 6 displays the target image 7, the estimation results 4102, the dimension measurement results 4101(8), etc. on the GUI screen 60.

[Automatic dimension measurement]
42 shows a process flow for automatic dimension measurement in embodiment 5, and includes steps S51 to S59. In step S51, the computer system 1E inputs the image 7 of the measurement target, the photographing information 111, and the definition information 112 based on the input operation of the user U1 from the operation PC 6, performs dimension measurement using the semi-automatic dimension measurement function in embodiment 1, and obtains the dimension measurement result 8. The image 7 and the dimension measurement result 8 are displayed on the GUI screen 60 of the operation PC 6.

In step S52, the computer system 1E automatically generates learning data 141 for learning from the dimensional measurement results 8 of step S51 (data including the interface reference line 11 and area reference point 12) and registers it in DB 140. In step S53, the computer system 1E uses the selected learning data 141 to construct and train a learning model 200 (e.g., similar to that shown in Figure 34).

In step S54, the computer system 1E checks whether the accuracy of the estimation by the learning model is sufficient. If it is sufficient (Y), proceed to step S57; if it is insufficient (N), proceed to step S55. In step S55, the computer system 1E checks whether the learning model is overfitted, and if it is overfitted (Y), proceed to step S56; if it is not overfitted (N), return to step S52. In step S56, the computer system 1E reselects the learning data 141 to be used for learning and returns to step S53.

In step S57, the computer system 1E registers a learning model with sufficient accuracy as a trained model. In step S58, the computer system 1E inputs the image 7 of the object to be measured based on the operation of the user U1 on the operation PC 6. In addition, the user U1 selects the fully automatic dimension measurement mode on the GUI screen 60. In step S59, the computer system 1E performs dimension measurement on the object image, which is the input image, in the fully automatic dimension measurement mode. Details of step S59 may be, for example, as follows:

In a first processing example, computer system 1E automatically performs contour detection and dimension measurement without presenting the estimation result 4102 to user U1. In this case, computer system 1E inputs pixel data of the input image into a learning model of the estimation function and obtains the estimated interface reference line 11X and estimated area reference point 12X as the estimation result 4102. Computer system 1E detects the contour of the target structure from the target image using the position coordinate information of the estimated interface reference line 11X and estimated area reference point 12X, measures the specified dimensions based on the detected contour, etc., and obtains dimension measurement result 4101. Computer system 1E transmits dimension measurement result 4101(8) to operation PC 6 and displays dimension measurement result 4101(8) on GUI screen 60.

In a second processing example, computer system 1E presents estimation result 4102 to user U1, who then confirms the estimation result 4102 and performs contour detection and dimension measurement. In this case, computer system 1E inputs pixel data of the input image into a learning model of the estimation function and obtains estimated interface reference line 11X and estimated area reference point 12X as estimation result 4102. Computer system 1E transmits estimation result 4102, including estimated interface reference line 11X and estimated area reference point 12X, to operation PC 6. Operation PC 6 displays estimated interface reference line 11X and estimated area reference point 12X of estimation result 4102 on the target image on GUI screen 60. User U1 confirms estimation result 4102 on GUI screen 60, and applies it as is by operating a predetermined button, etc. if applicable; manually modifies it before applying; or rejects (does not apply) it by operating a predetermined button, etc. When the user U1 confirms and applies the estimation result 4102, the computer system 1E detects a contour from the target image using the applied estimated interface reference line 11X and estimated area reference point 12X, measures the specified dimensions based on the detected contour, etc., and obtains the dimension measurement result 4101. The computer system 1E transmits the dimension measurement result 4101(8) to the operation PC 6 and displays the dimension measurement result 4101(8) on the GUI screen 60.

[Effects of the Fifth Embodiment]
The dimension measurement system and method of embodiment 5 have the following advantages. By performing length measurement using the semi-automatic method of embodiment 1, user U1 can accumulate learning data without particularly thinking about it. According to embodiment 5, the accumulated learning data can be used to train the learning model of embodiment 4, thereby improving the accuracy of estimation of material interfaces, structure regions, etc. Then, using the estimation results from the learning model, the interface reference line 11 and region reference point 12 can be set more easily and accurately, making length measurement work more efficient.

The above provides a specific description of the embodiments of the present disclosure, but the present disclosure is not limited to the above-described embodiments and various modifications are possible without departing from the spirit of the present disclosure. Except for essential components, components can be added, deleted, or replaced in each embodiment. Unless otherwise specified, each component may be singular or plural. Forms that combine each embodiment and its variations are also possible.

1...Dimension measurement system (computer system), 2...SEM, 6...Client terminal device (operating PC), 7...Image, 8...Dimension measurement results, 11...Interface reference line, 12...Area reference point, 13...Perpendicular line, 14...Intersection point, 20...Line profile, 21...Edge candidate point, 60...GUI screen, 70...Target image, 101...Processor, 102...Memory resources, 111...Photographing information, 112...Definition information, 401...Mask, 402...Pillar, 403...Trench, U1...User.

Claims (19)

A dimension measurement system for measuring dimensions of a structure displayed as an image including a plurality of regions with different brightness,
a computer system including a processor and a memory resource for storing data to be processed by the processor;
a display on which a graphical user interface (GUI) is displayed, which is manually operated by a user;
The processor:
determining a reference line for identifying a portion of a boundary between the plurality of regions and position information of the reference line, and a reference point set within any one of the plurality of regions and position information of the reference point, based on a manual operation by the user performed on the GUI;
Detecting a contour or edge of the structure in the image using the reference line and the reference point;
measuring the dimensions of the structure using the detected contour or edge information;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
The processor:
On a perpendicular line set between the reference point and the reference line, a plurality of line profiles are set extending in a direction parallel to the reference line;
extracting candidates for the contour or edge from the line profile based on a change in luminance of pixels that make up the line profile;
detecting the contour or edge based on the extracted candidates;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
The processor:
A plurality of line profiles are set on positions extending in a radial direction from the reference point;
extracting candidates for the contour or edge from the line profile based on a change in luminance of pixels that make up the line profile;
detecting the contour or edge based on the extracted candidates;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
The processor:
setting a plurality of first line profiles extending in a direction parallel to the reference line on a perpendicular line set between the reference point and the reference line;
setting a plurality of second line profiles at positions extending in a radial direction from the reference point;
extracting candidates for the contour or edge on the first and second line profiles based on luminance changes of pixels constituting the first and second line profiles;
Detecting the contour or edge based on the extracted candidates.
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
The processor:
determining the dimensions of the structure using information on the detected contour or edge and definition information that defines the dimensions of the measurement object;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
the reference line is designated relative to a material interface within the image;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
The computer system includes:
a dimension measurement system that provides an adjustment function for adjusting the reference line set on the GUI to a probable reference line position on the structure; 2. The dimension measurement system according to claim 1,
The image is a cross-sectional image of a semiconductor device in which at least one of a mask, a pillar, or a trench is formed.
Dimensional measurement system. The dimension measurement system according to any one of claims 2 to 4,
the processor creates combinations of the candidates for the contour or edge in a plurality of the line profiles;
detecting the contour or edge by selecting the contour or edge based on an evaluation value based on a distance between the candidates or an amount of change in brightness;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
the processor causes the specified reference line and the specified reference point to be superimposed on the image displayed on the GUI;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
the processor causes the detected contour or edge to be superimposed on the image displayed on the GUI;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
the processor causes the measured dimensions to be superimposed on the image displayed on the GUI;
Dimensional measurement system. 2. The dimension measurement system according to claim 1,
the computer system includes a storage device in which learning data including the image, the reference line, and the reference point is stored;
a learning model is implemented in the memory resource, which inputs the image and outputs the reference line and the reference point for the image;
the processor detects a contour or an edge of the structure included in a new input image using information on the reference lines and the reference points for the new input image estimated by the learning model;
measuring the dimensions of the structure using the detected contour or edge information;
Dimensional measurement system. 14. The dimension measurement system according to claim 13,
The processor trains the learning model using the learning data stored in the storage.
Dimensional measurement system. 14. The dimension measurement system according to claim 13,
the processor stores the image, and the position information of the reference line and the position information of the reference point calculated for the image based on the user's manual operation, in the storage as the learning data;
Dimensional measurement system. A dimension measurement method for measuring dimensions of a structure displayed as an image including a plurality of regions with different brightness, using a dimension measurement system including a processor, a memory resource for storing data processed by the processor, and a display on which a graphical user interface (GUI) is displayed, the method comprising:
setting a reference line that identifies a part of a boundary between the plurality of regions and a reference point that is designated within one of the plurality of regions based on a manual operation by a user of the dimension measurement system performed on the image of the structure displayed on the GUI;
Acquire position information of the reference line and the reference point;
detecting a contour or an edge of the structure using the reference line and the reference point, and measuring the dimensions of the structure using information on the detected contour or edge;
Dimension measurement method. 17. The dimension measurement method according to claim 16,
On a perpendicular line set between the reference point and the reference line, a plurality of line profiles are set extending in a direction parallel to the reference line;
extracting candidates for the contour or edge on the line profile based on a change in luminance of pixels constituting each line profile;
detecting the contour or edge based on the extracted candidates;
Dimension measurement method. 17. The dimension measurement method according to claim 16,
A plurality of line profiles are set on positions extending in a radial direction from the reference point;
extracting candidates for the contour or edge on the line profile based on a change in luminance of pixels constituting each line profile;
detecting the contour or edge based on the extracted candidates;
Dimension measurement method. A non-volatile recording medium storing a program for causing a computer equipped with a display that displays a graphical user interface (GUI) to execute the following steps in order to measure the dimensions of a structure displayed as an image including multiple regions of different brightness.
setting a reference line that identifies a portion of a boundary between the plurality of regions and a reference point that is designated within any one of the plurality of regions based on a manual operation by a user of the dimension measurement system performed on the image of the structure displayed on the GUI;
acquiring position information of the reference line and the reference point;
detecting a contour or edge of the structure using the reference line and the reference points;
A step of measuring dimensions of the structure using information about the detected contour or edge.