Contour line analysis device, processing condition decision system, shape estimation system, semiconductor device manufacturing system, search device, and data structure used in these

Abstract

High-precision size extraction can be performed on a complicated shape generated in semiconductor processing. A contour line of an object structure is described by setting a start point (110) and an end point (111) on the circumference of a pattern formed by combining a plurality of ellipses (100), and using a curve (120) of one stroke on the circumference connecting the two points as a shape model.

Description

TECHNICAL FIELD

[0001] The present application relates to a contour line analysis device, a processing condition decision system, a shape estimation system, a semiconductor device manufacturing system, a search device, and a data structure used in these. BACKGROUND

[0002] By processing a semiconductor sample according to a suitable processing condition in a semiconductor process, a desired semiconductor processing can be implemented. In recent years, with the introduction of new materials constituting a device, the device structure is complicated, the control range of the semiconductor processing device is expanded, and many control parameters are added. The process is multi-staged, and a fine and complicated processing is implemented. In order to produce a high-performance device using the semiconductor processing device, process development is required to derive a suitable processing condition that achieves a target processing shape of the semiconductor sample.

[0003] In order to fully mobilize the performance of the semiconductor processing device, optimization of a large number of control parameters is indispensable, and for its implementation, technical know-how of process development, high device operation skills, and a large number of trial and error of processing experiments are required. Therefore, in the process development, a large number of times of size measurement is required. For example, in the case of a sample that considers a line / space (L / S) pattern as a processing object, if the size of each line pattern CD (Critical Dimension), depth, etc. is set to 10 places, the number of line patterns to be measured is set to 10, then 100 measurements are required for each 1 sample. Therefore, assuming that 100 samples are processed, a total of 10,000 times of size measurement is required.

[0004] The more complicated the structure of the device is, the more the number of measurement sites increases, and therefore, the delay of the process development accompanied by the elongation of the size measurement time becomes a problem. In addition, these sizes are reduced every year with the miniaturization of the structure, and it becomes difficult to extract the size of the target structure by hand. Therefore, a technology is required to extract the size of the target structure from the image of the semiconductor sample at high speed and high accuracy without depending on the hand. Patent Literature 1 discloses such a technology.

[0005] In Patent Literature 1, a hypothetical processing shape is generated using a shape model, and a database of processing shapes and SEM signal waveforms is created using SEM simulation. By comparing the actual signal waveform obtained in the SEM with the database, the processing shape close to the signal waveform is determined, and it is estimated as the processing shape in the observation. Thereby, contour line detection (edge detection) of the SEM image, and extraction of the size of the target structure can be performed.

[0006] PRIOR ART DOCUMENT

[0007] PATENT LITERATURE

[0008] Patent Literature 1: JP-A No. 2009-198339 SUMMARY

[0009] PROBLEMS TO BE SOLVED BY THE INVENTION

[0010] In the case of Patent Literature 1, since the shape model shown in the example is simplified, it is considered difficult to recognize a complex shape. Further, since estimation is performed by database comparison, it is considered difficult to perform estimation for a shape not in the database. Currently, miniaturization and 3Dization are advancing in semiconductor devices, and further, various structures such as quantum computers are being proposed. Along with this, it is considered that size extraction by hand will be more difficult in the future, and it is necessary to extract a complex and various shape in a short time without manual operation.

[0011] Due to the problem of difficulty in recognizing a complex and various shape, for example, it is difficult to recognize a shape having different curvatures at a side wall and a bottom in a line / space pattern. In general etching, since the effects of radical-based isotropic etching and ion-assisted anisotropic etching overlap, such a processing result of different curvatures frequently occurs. Further, if a database of a large scale is used in order to avoid the problem of difficulty in recognizing an unknown shape, there is a possibility that estimation in real time becomes difficult.

[0012] MEANS FOR SOLVING THE PROBLEMS

[0013] By providing a start point and an end point on the circumference of a figure formed by combining a plurality of ellipses, a curve of one stroke on the circumference connecting the two points is used as a shape model, and a contour line of an object structure is described.

[0014] EFFECT OF THE INVENTION

[0015] It is possible to perform high-precision size extraction for a complex shape that can occur in semiconductor processing.

[0016] Other problems and new features will be apparent from the description of the present specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a shape model using a plurality of ellipses.

[0018] Figure 2A is a cross-sectional view of a sample with a mask pattern.

[0019] Figure 2B is a cross-sectional view after etching processing of the sample with a mask pattern.

[0020] Figure 2C is a cross-sectional view of Figure 2B illustrating a typical shape feature quantity.

[0021] Figure 3 is an example of a hypothetical shape used in scatterometry.

[0022] Figure 4A is a configuration example of the size extraction system of Embodiment 1.

[0023] Figure 4B is a hardware configuration example of the profile analysis device.

[0024] Figure 5 is a flowchart of the size extraction in Embodiment 1.

[0025] Figure 6A is Figure 2B profile data of a sample with a mask pattern shown in FIG. 6.

[0026] Figure 6B is a graph showing a shape model fitted to a profile of Figure 6A .

[0027] Figure 6C is a graph showing a shape model fitted to a profile of Figure 6A .

[0028] Figure 7 is a graph for explaining a method of calculating a size using a shape model.

[0029] Figure 8 is an example of a data structure of a shape model database.

[0030] Figure 9 is a configuration example of the processing condition decision system of Embodiment 2.

[0031] Figure 10 is a flowchart of the processing condition decision of the semiconductor processing device in Embodiment 2.

[0032] Figure 11 is an example of an input GUI.

[0033] Figure 12 is an example of an output GUI.

[0034] Figure 13 is a configuration example of the structure shape estimation system of Embodiment 3.

[0035] Figure 14 is a flowchart of the shape estimation in Embodiment 3.

[0036] Figure 15 is a configuration example of the profile detection system of Embodiment 4.

[0037] Figure 16 is a flowchart of the profile detection in Embodiment 4.

[0038] Figure 17 is an example of an input GUI.

[0039] Figure 18 is an example of an output GUI. DETAILED DESCRIPTION

[0040] Embodiments of the present application will be described below with reference to the accompanying drawings. The present application is not limited to the description of the embodiments shown below. It will be readily apparent to those skilled in the art that the specific structural

[0041] Further, the position, size, shape, and range of each structure shown in the drawings and the like are not indicative of actual position, size, shape, and range, and the like, but are made to be easy to understand the present application. Therefore, in the present application, the position, size, shape, and range, and the like disclosed in the drawings and the like are not limited.

[0042] Embodiment 1

[0043] Figure 1 and Figure 4A are diagrams each showing a structural example of a shape model and a dimension extraction system in Embodiment 1. The dimension extraction system of Embodiment 1 extracts a dimension desired by a user (a desired dimension) from an image acquired from a measurement device such as a SEM image, using the shape model.

[0044] In Figure 2A , a cross-sectional view of a sample with a mask pattern is shown as a typical example of a semiconductor sample. A mask 200 is formed on an etched film 201. Figure 2B is a cross-sectional view after etching processing of the sample with the mask pattern. The etched film 201 is etched more preferentially than the mask 200, and a groove is formed between the mask patterns. If the cross-sectional shape targeted is a groove as shown in Figure 3 , the user needs to grasp the shape of the object structure of Figure 2B , and adjust the control parameters of the semiconductor processing device. However, since the processed shape formed in a state where the control parameters of the semiconductor processing device are not optimized has a complex shape, dimension measurement is a laborious job, and as described above, becomes one of the causes of delay in process development.

[0045] In the present embodiment, by using a shape model that utilizes Figure 1The shape model of multiple ellipses as shown can describe complex and various shapes with high accuracy. The ellipses can describe from a straight line to a curve by changing the lengths of the major axis and the minor axis, and can describe shapes with different curvatures for each portion by combining multiple ellipses. Further, instead of performing database matching as in Patent Literature 1, the parameters of the shape model are estimated by fitting the shape model to the contour line of the object structure. If a plausible parameter value is obtained by fitting, a plausible shape model that describes the contour line of the object structure with high accuracy is obtained by substituting the parameter value into the shape model. Various dimensions of the contour line are analytically extracted using the plausible shape model.

[0046] In the present embodiment, as shown in Figure 1 , a start point 110 and an end point 111 are provided on the circumference of a figure formed by combining multiple ellipses 100a to 100e, and a one-stroke curved line connecting the two points on the circumference is used as a shape model 120. In Figure 1 , a line-symmetrical shape model combining five ellipses is shown, but the number of ellipses can be several as long as it is multiple, and the model can be a non-line-symmetrical model.

[0047] Hereinafter, the model parameters of the shape model 120 will be referred to as shape model parameters. The shape model parameters include: first parameters related to the shapes and arrangement of the ellipses, such as the center coordinates 130, the minor axis length 131, the major axis length 132, and the inclination of the minor axis 133 of each ellipse as exemplified in Figure 1 ; second parameters related to the one-stroke method (whether to connect the inner circumference or the outer circumference of the ellipses, etc.); and the like. Figure 1 The specific examples of the first parameters shown in are exemplified, and at least one or more of these are included. The ellipses 100 can adjust the flatness by changing the balance between the minor axis length 131 and the major axis length 132. Thus, the shape model 120 can describe a complex contour line formed by combining contours close to a straight line and contours close to a curve. Further, the concave-convex of the contour line can be changed by changing the one-stroke method.

[0048] In this way, various shapes can be described by adjusting the shape model parameters. In addition, since multiple ellipses are used, appropriate numbers such as the first ellipse, the second ellipse, and the like are assigned to each ellipse, and each shape model parameter is referred to as the major axis length of the first ellipse, the major axis length of the second ellipse, and the like.

[0049] The shape model 120 is used to Figure 4AThe size extraction system using the shape model 120 will be described. The measurement device 400 is a measurement device using a charged particle line device such as a scanning electron microscope, which outputs information of an object structure of a sample as image data. The measurement device 400 is a measurement device which acquires information of a sample as an image using a phenomenon such as reflection, transmission, interference, or the like caused when an electron or the like is incident on the sample. Specifically, it is a scanning electron microscope (SEM), a transmission electron microscope (TEM), a scanning transmission electron microscope (STEM), or the like, which is a measurement device using an electron microscope. The image data acquired in the measurement device 400 can be image data obtained in any charged particle line device.

[0050] The contour line detection device 410 is a device which detects an edge (contour line) from an image input from the measurement device 400. For example, for an input SEM (Scanning Electron Microscopy) image, a TEM image, contour line data thereof is output. As a method of detecting a contour line, there are a Sobel method, a Canny method, a Laplacian method, or the like, which detects based on a change in pixel value, an Open CV method which detects using machine learning, and the like, and any method can be used. In addition, the contour line detection device 410 can be implemented as one function of the contour line analysis device 420 described later.

[0051] In Figure 4BA hardware configuration example of the contour line analysis device 420 is shown. The contour line analysis device 420 includes a processor 431, a main storage 432, an auxiliary storage device 433, an input / output interface 434, a display interface 435, a network interface 436, an input / output (I / O) port 437, which are combined by a bus 438. The input / output interface 434 is connected to an input device 440 such as a keyboard and a mouse, and the display interface 435 is connected to a display 439, thereby realizing a GUI (Graphical User Interface). The network interface 436 is an interface for connecting to an external device via a network. The auxiliary storage device 433 is typically constituted by a nonvolatile memory such as a HDD (Hard Disk Drive), a SSD (Solid State Drive), a flash memory, and stores a program executed by the contour line analysis device 420, data to be processed by the program, and the like. The main storage 432 is constituted by a RAM (Random Access Memory), and temporarily stores a program, data required for execution of the program, and the like by a command of the processor 431. The processor 431 executes the program loaded from the auxiliary storage device 433 to the main storage 432. The functions of the contour line analysis device 420 are realized by the processor 431 executing the program stored in the auxiliary storage device 433, in cooperation with other hardware, to perform a determined process. There are cases where a program executed by a computer and the like, a function thereof, or a unit realizing the function are referred to as a "function", a "unit", and the like. The contour line analysis device 420 can be realized, for example, on a PC (Personal Computer), a server, or a cloud.

[0052] The contour line analysis device 420 has an analysis unit 421, a shape model database 422, a shape model setting unit 423, a shape model fitting setting unit 424, a dimension extraction setting unit 425, a dimension calculation method setting unit 426, and is a device that calculates a value of a shape model parameter as a parameter of a shape model, a value of a dimension desired to be extracted, from contour line data input from the contour line detection device 410. Hereinafter, the shape model parameter calculated by the contour line analysis device 420 and the dimension extracted are referred to as a plausible shape model parameter and a plausible dimension, respectively.

[0053] The analysis unit 421 performs analysis such as fitting of a shape model, calculation of a value of a plausible shape model parameter, and calculation of a value of a plausible dimension, with respect to the input contour line data. The calculated value of the plausible shape model parameter and the value of the plausible dimension are stored in the shape model database 422.

[0054] The user sets the specifications of the shape model (shape model specifications) by the shape model setting section 423, and sets the fitting method of the shape model by the shape model fitting setting section 424. The analysis section 421 performs fitting of the shape model using the profile line data input from the profile line detection device 410 based on these settings, and calculates the values of the shape model parameters that appear to be reasonable. The shape model obtained by substituting the values of the shape model parameters that appear to be reasonable is the shape model that appears to be reasonable.

[0055] Further, the user sets the dimensions desired to be extracted by the dimension extraction setting section 425, and sets the calculation method of the values of the dimensions desired to be extracted by the dimension calculation method setting section 426. The analysis section 421 calculates the values of the dimensions desired to be extracted that appear to be reasonable using the shape model that appears to be reasonable based on the above settings.

[0056] In Figure 5 A flowchart of the dimension extraction system shown in Figure 4A will be described below. Figure 5 The method of dimension extraction will be described below.

[0057] First, the image data of the SEM image and the like is acquired using the measurement device 400 (S101). Next, the profile line data is acquired from the image data using the profile line detection device 410 (S102). As an example of the profile line data, the profile line data 600 representing the cross section of the sample on which the patterned mask has been etched is shown in Figure 6A The profile line data 600 representing the cross section of the sample on which the patterned mask has been etched is shown. The etching film 201 is etched more preferentially than the mask 200, and a groove is formed between the mask patterns.

[0058] The shape model specifications are set by the shape model setting section 423 (S103). For example, the kind of model is specified as "a line-symmetrical shape model composed of five ellipses" as shown in Figure 1 The kind of shape model parameters that describe the shape model is specified.

[0059] Next, the method of fitting the shape model to the profile line data is set by the shape model fitting setting section 424 (S104). For example, the least squares method, the weighted least squares method, or the regularized least squares method is used in the fitting, and the values of the shape model parameters can be estimated by a nonlinear optimization method that uses an iterative solution method, a combinatorial optimization method. Further, settings related to the end condition of the fitting, the random number or the initial value generation method at the time of the optimization processing are performed.

[0060] Fitting of the shape model is performed with respect to the profile line data input from the profile line detection device 410 based on the settings of steps S103 and S104 (S105). The values of the shape model parameters that appear to be reasonable are calculated using the profile line data 600 shown in Figure 6Bto illustrate how the fitting looks. The shape model 601 before fitting depicts the shape model for the initial values of the shape model parameters. The shape model parameters values are improved by fitting using the contour data 600, as the shape model 602 in the fitting, the shape model is close to the contour 600. If the fitting is normally completed, as shown in Figure 6C the contour 600, the seemingly reasonable shape model parameters values are obtained for the contour 600, and the seemingly reasonable shape model 603 becomes a model that well reproduces the contour data 600.

[0061] Next, it is determined whether there is a failure in the fitting (S106), and in the case where a failure is confirmed, the process returns to step S103, and the re-setting of the specifications of the shape model, and the re-setting related to the fitting method of the shape model are implemented. Here, the failure related to the fitting is that the fitting does not end, the shape model parameter values fall into local values in the fitting, an abnormality is observed in the shape of the seemingly reasonable shape model obtained by the fitting, and the like. An example of the abnormality of the shape is a case where there is a deviation in the shape recognized by the user visually measuring the image acquired by the measurement device 400 and the shape of the seemingly reasonable shape model. In the case where a failure is not confirmed in step S106, the fitting is ended, and the obtained seemingly reasonable shape parameter values are stored in the shape model database 422 (S107).

[0062] Next, the kind of the dimension that the user wants to extract is set by the dimension extraction setting section 425 (S108). As the kind of the dimension that the user wants to extract, for example, a series of shape features as shown in Figure 2C is set. Figure 2C Shape features that can be set for the cross-sectional view of Figure 2B are exemplified. In the case where machine learning is used in the process development, a correlation model Y = f(X) of regression or the like is obtained with the processing conditions as the explanatory variable X, and the feature quantity related to the processed shape (referred to as a shape feature quantity) as the target variable Y, and the processing conditions that give the shape feature quantity of the target are estimated using the correlation model. In addition, there is no quantitative policy for what is selected as the shape feature quantity of the semiconductor sample, and the dimension data selected and extracted by the user can be used. In the case where the shape of the groove is Figure 2C , various kinds of shape feature quantities are considered, such as the maximum width 210, the depth 211, the width of the bottom 212, the width of the upper portion 213, the width of the middle portion 214, the taper angle 215, the angle of the bow 216, the eccentricity of the bottom 217, and the like.

[0063] Next, the calculation method of the dimension that the user wants to extract is set by the dimension calculation method setting section 426 (S109). For example, the calculation method of the dimension is considered Figure 2CThe method for calculating the maximum width 210, depth 211, bottom width 212, and bottom eccentricity 217 of the shape features shown. This is achieved through the process up to step S108. Figure 2B The contour data of the sample shown was obtained from the SEM image. Figure 6C The seemingly reasonable shape model 603 shown.

[0064] use Figure 7 Here is an example illustrating the method for calculating dimensions using shape model 603. For the coordinate axes, the x-axis is taken along the boundary between mask 200 and etched film 201, and the y-axis is taken in the direction perpendicular to it. The width 702 of the bottom can be estimated using the length of the major axis of each ellipse 604. Since the eccentricity can be calculated from the lengths of the major and minor axes of each ellipse 604, the eccentricity 703 of the bottom can be estimated. The depth 701 can be estimated from the center coordinates, major axis length, minor axis length, and minor axis inclination of each ellipse 604. Furthermore, for each ellipse 604, as... Figure 7 As shown, by taking multiple differential values ​​of 0 about the y-axis and calculating the distance between their x-coordinates, the maximum width of 700 can be estimated. Thus, the desired dimension can be estimated by using singular points in the shape model 603, such as extreme points where the differential value is 0, points where differentiation is impossible, and inflection points where the differential value changes from positive to negative. This method of calculating dimensions using singular points of the shape model is called the singularity search method.

[0065] The calculation method is not limited to the singularity search method. For example, the x-coordinates on the shape model 603 can be extracted along the y-axis in a certain step, and the maximum width 700 can be used as the maximum value of the distance between the x-coordinates. This method of calculating dimensions by extracting coordinates along a specific axis is called the strip search method.

[0066] For the desired size set in step S108, the calculation method set in step S109 is used, for example, to calculate a seemingly reasonable size such as a maximum width of 700 (S110). The extracted seemingly reasonable size data is stored in the shape model database 422 (S111), and the process ends. Figure 8 This illustrates an example of the data structure of the shape model database 422 obtained by applying the above process to SEM images (corresponding to data numbers) of N samples.

[0067] Example 2

[0068] Figure 9 This is a diagram illustrating a structural example of the processing condition determination system in Embodiment 2. In the processing condition determination system of Embodiment 2, suitable processing conditions for the semiconductor processing device are determined.

[0069] Embodiment 2 aims at speeding up the process development using machine learning. As described above, in the process development using machine learning, a search for a processing condition giving a target shape feature quantity is generally performed using a correlation model with a processing condition as an explanatory variable X and a shape feature quantity related to a processed shape as a target variable Y. However, since there is no quantitative guideline on what to use as a shape feature quantity, the following cases can occur: a redundant shape feature quantity is adopted in the target variable, or an important shape feature quantity describing a processed shape is lacking. In the former case, since a large amount of experimental data is required in learning of the correlation model with an increase in the target variable, the number of processing times of the semiconductor processing apparatus increases, and there is a concern about a prolongation of the process development period. In the latter case, since the expressiveness of the correlation model is weakened, it becomes difficult to predict a processing condition achieving a target processed shape, and there is also a concern about a delay in the process development.

[0070] In this embodiment, by setting a shape model parameter suitably describing a processed shape as a target variable of the correlation model, the adoption of a redundant shape feature quantity and the lack of an important shape feature quantity described above can be avoided. Thus, the process development can be sped up.

[0071] Here, the measurement apparatus 400, the profile line detection apparatus 410, the profile line analysis apparatus 420, the analysis section 421, the shape model database 422, the shape model setting section 423, and the shape model fitting setting section 424 are defined as the same as in Embodiment 1.

[0072] The processing condition decision apparatus 900 has a processing condition database 901, a learning section 902, a processing condition estimation section 903, and a target dimension value setting section 904, and is an apparatus for deciding a suitable processing condition based on data of a seemingly reasonable shape model parameter in the shape model database 422 and a processing condition in the processing condition database 901. The hardware structure of the processing condition decision apparatus 900 is also the same as that of the profile line analysis apparatus 420 shown in Embodiment 1, and thus repeated description is omitted. Figure 4B The profile line analysis apparatus 420 is also the same as that shown in Embodiment 1, and thus repeated description is omitted.

[0073] The processing condition database 901 stores the existing processing conditions and the processing conditions estimated by the processing condition estimation unit 903. In the learning unit 902, the relevant models of seemingly reasonable shape model parameters in the shape model database 422 and the processing conditions in the processing condition database 901 are learned. In the target size value setting unit 904, the user's desired shape model parameter values ​​(target values) are set for the seemingly reasonable shape models. Alternatively, the shape model parameter values ​​(target values) of the seemingly reasonable shape models can be calculated based on the user's desired shape size. In the processing condition estimation unit 903, the relevant models obtained in the learning unit 902 are used to estimate the processing conditions for issuing the shape model parameter values ​​set in the target size value setting unit 904.

[0074] Semiconductor processing apparatus 910 is an apparatus for processing semiconductor samples, using processing conditions determined by processing condition determining device 900. Semiconductor processing apparatus 910 includes a photolithography apparatus, a film forming apparatus, a patterning apparatus, an ion implantation apparatus, a heating apparatus, and a cleaning apparatus, which are semiconductor manufacturing apparatuses. Photolithography apparatuses include exposure apparatuses, electron line drawing apparatuses, and X-ray drawing apparatuses. Film forming apparatuses include CVD (Chemical Vapor Deposition) apparatuses, PVD (Physical Vapor Deposition) apparatuses, evaporation apparatuses, sputtering apparatuses, and thermal oxidation apparatuses. Patterning apparatuses include wet etching apparatuses, dry etching apparatuses, electron beam processing apparatuses, and laser processing apparatuses. Ion implantation apparatuses include plasma doping apparatuses and ion beam doping apparatuses. Heating apparatuses include resistance heating apparatuses, lamp heating apparatuses, and laser heating apparatuses.

[0075] exist Figure 10 Showing through Figure 9 The flowchart below shows how the system determines the processing conditions of the semiconductor processing device. Figure 10 This explains the method for determining the processing conditions of the semiconductor processing device 910.

[0076] Figure 10 The processes of steps S201 to S207 are respectively related to Figure 5 Steps S101 to S107 are the same, so the explanation is omitted.

[0077] After the plausible shape model parameter is stored in the shape model database 422 in step S207, the shape model parameter value of the target of the user is set by the target size value setting section 904 (S208). Next, it is determined whether the plausible shape model parameter value estimated in step S207 is close to the shape model parameter value set in the target size value setting section 904 (S209). Here, the distance of the closeness of the evaluation value is calculated using any one of the Euclidean distance, the Manhattan distance, the Chebyshev distance, and the Mahalanobis distance. It is determined whether it is close or not based on whether the calculated value is larger or smaller than the reference value determined by the user.

[0078] In step S209, in the case where it is determined that the plausible shape model parameter value estimated in step S207 is close to the shape model parameter value set in the target size value setting section 904, the process ends. On the other hand, in the case where it is determined that it is not close in step S209, the learning section 902 learns the correlation model of the processing condition in the processing condition database 901 and the plausible shape model parameter in the shape model database 422 (S210). Here, the correlation model represents a model of regression, classification, a model using a kernel method, a model using a neural network, a model using a decision tree, and the like.

[0079] Next, the processing condition estimation section 903 estimates the processing condition that gives the shape model parameter value set in the target size value setting section 904 using the correlation model obtained in the learning section 902 (S211). The estimated processing condition is added to the processing condition database 901, and the database is updated (S212). The estimated processing condition is used in the semiconductor processing device 910 to perform processing on a new sample (S213). The processed sample is taken out from the semiconductor processing device 910 and transferred to the process of step S201. The above series of processes are repeated until the process ends.

[0080] Next, the GUI related to Embodiment 1 and Embodiment 2 will be described. Figure 11 、 12

[0081] Figure 11 The input GUI 1100 illustrated in FIG. 11 is an example of an input screen for setting the input profile line analysis device 420 of Embodiments 1 and 2. This input screen is prompted after the image is obtained in the process of step S101 in the case of Embodiment 1 and after the image is obtained in the process of step S201 in the case of Embodiment 2.

[0082] ​The input GUI 1100 has a contour line detection setting frame 1110, a shape model selection frame 1120, a fitting setting frame 1130, a dimension extraction setting frame 1140, a calculation method setting frame 1150, a valid / invalid display section 1160, and a decision button 1170. The contour line detection setting frame 1110, the shape model selection frame 1120, the fitting setting frame 1130, the dimension extraction setting frame 1140, and the calculation method setting frame 1150 perform settings related to the contour line detection device 410, the shape model setting section 423, the shape model fitting setting section 424, the dimension extraction setting section 425, and the dimension calculation method setting section 426, respectively.

[0083] The contour line detection setting frame 1110 has a detection method input section 1111 and an image input section 1112. For example, in the detection method input section 1111, as a detection method of a contour line, a method that performs detection based on a change in pixel value, such as the Sobel method, the Canny method, the Laplacian method, or the like, a method that utilizes machine learning, such as Open CV, or the like, can be selected. Further, in the case of Embodiment 1, by dragging the image data acquired in step S101 to the image input section 1112, it is possible to input the image data to the contour line detection device 410, and in the case of Embodiment 2, by dragging the image data acquired in step S201 to the image input section 1112, it is possible to input the image data to the contour line detection device 410.

[0084] The shape model selection frame 1120 has a model input section 1121. For example, a line-symmetrical shape model composed of a specific number of ellipses, a shape model that does not assume line symmetry, or the like is selected.

[0085] The fitting setting frame 1130 has a fitting method input section 1131. For example, a method that optimizes shape model parameters by the Levenberg-Marquardt method using the least squares method (LS-LM method), a method that optimizes shape model parameters by an annealing method using the least squares method (LS-annealing method), or a method that optimizes shape model parameters by the Levenberg-Marquardt method using the weighted least squares method (WS-LM method) is selected. In addition, in the case of Embodiment 1, a method that optimizes shape model parameters by the Levenberg-Marquardt method using the least squares method (LS-LM method) is selected. Figure 11

[0086] ​The size extraction setting frame 1140 has a size input section 1141 that sets a desired size to be extracted by the user. The calculation method setting frame 1150 has a calculation method input section 1151 that sets a calculation method for the desired size to be extracted input in the size input section 1141. For example, the singular point search method and the strip search method described in Embodiment 1 can be selected. In the case of Embodiment 2, the size extraction setting frame 1140 and the calculation method setting frame 1150 are not needed in the input GUI 1100.

[0087] Whether the above input has been made validly is displayed by the valid / invalid display section 1160 of each of the setting frames described above. If all of the valid / invalid display section 1160 become valid, the process of step S102 in the case of Embodiment 1 or the process of step S202 in the case of Embodiment 2 is started by pressing the decision button 1170 of the input GUI 1100.

[0088] The process of step S102 in the case of Embodiment 1 or the process of step S202 in the case of Embodiment 2 is started by pressing the decision button 1170 of the input GUI 1100. Figure 5 or Figure 10 The process of step S106 in the case of Embodiment 1 or the process of step S206 in the case of Embodiment 2 is started. Figure 12 The output GUI 1200 is shown, and the process of step S106 in the case of Embodiment 1 or the process of step S206 in the case of Embodiment 2 is prompted. This GUI displays the current situation, and the user selects whether to proceed to the next process. The size extraction result display section 1210 has a plausible shape model display section 1211 and a size calculation result display section 1212. The user is prompted about the appearance of the shape model fitted in step S105 or S205 by the plausible shape model display section 1211. Furthermore, the user is prompted about the plausible shape model parameter value and the plausible size value in the size calculation result display section 1212. The user can select completion or re-setting in the completion / re-setting selection section 1220 based on the information of the shape model displayed in the size extraction result display section 1210. When the user judges that there is no defect in the fitting, completion is selected and the decision button 1230 is pressed, whereby the process of step S107 in the case of Embodiment 1 or the process of step S207 in the case of Embodiment 2 is advanced. When it is judged that there is a defect, re-setting is selected and the decision button 1230 is pressed, whereby the screen of the input GUI 1100 is returned, and re-setting can be performed.

[0089] Embodiment 3

[0090] Figure 13 is a diagram showing a configuration example of a structure shape estimation system in Embodiment 3. In the structure shape estimation system of Embodiment 3, the shape of an object structure of a sample is estimated from a spectroscopic spectrum.

[0091] Example 3 uses a shape model using a plurality of ellipses to achieve high precision of optical shape measurement such as scatterometry. In scatterometry, optical simulation such as RCWA (Rigorous Coupled Wave Analysis) is performed on CAD (Computer-aided Design) data of a shape model generated hypothetically, and data of a hypothetical spectroscopic spectrum is generated. A correlation model of the CAD data of the obtained hypothetical shape model and the hypothetical spectroscopic spectrum data is sought, and the correlation model is used to estimate the CAD data of the hypothetical shape model that gives the hypothetical spectroscopic spectrum data closest to the spectroscopic spectrum actually obtained in a spectrometer. Thus, shape estimation is performed from the spectroscopic spectrum obtained in the spectrometer. In the past, the hypothetical shape was a simple shape composed of a combination of rectangles and the like as shown in FIG. 8, and there was a problem that the estimation precision of a shape having a curvature was low. In the present embodiment, a complex shape having a curvature that differs for each portion can be generated by using a shape model using a plurality of ellipses, and shape estimation can be made highly precise. Figure 3

[0092] The hypothetical shape data generating apparatus 1300 has a shape model setting section 1301, a generation method setting section 1302, a parameter generating section 1303, a hypothetical dimension database 1304, a CAD section 1305, and a hypothetical shape database 1306, and is an apparatus that generates hypothetical shape data.

[0093] The user sets the shape model specifications through the shape model setting section 1301, and sets the generation method of the set of shape model parameter values through the generation method setting section 1302. The parameter generating section 1303 stores the set of shape model parameter values generated based on these settings in the hypothetical dimension database 1304. The CAD section 1305 outputs a shape obtained by substituting the shape model parameters in the hypothetical dimension database 1304 into the shape model as CAD data. The output CAD data is stored in the hypothetical shape database 1306.

[0094] The optical simulator 1310 is a simulator that performs optical simulation such as RCWA on the CAD data (hypothetical shape model) in the hypothetical shape database 1306. It is particularly configured to be a simulator that can calculate a theoretical value of a spectroscopic spectrum obtained by scatterometry for the geometry represented by the CAD data. Hereinafter, this theoretical value of the spectroscopic spectrum is referred to as a hypothetical spectroscopic spectrum.

[0095] The spectroscopic spectrum measurement apparatus 1330 is an apparatus that acquires a spectroscopic spectrum from scattered light, reflected light, interference light, and the like generated by the object structure of the semiconductor sample.

[0096] ​The optical shape estimation device 1320 is a device for estimating the shape of an object structure of a semiconductor sample from a measured spectroscopic spectrum. It includes a hypothetical spectroscopic spectrum database 1321, a learning unit 1322, and a shape estimation unit 1323. The hypothetical spectroscopic spectrum database 1321 is a database of hypothetical spectroscopic spectra calculated by the optical simulator 1310. The learning unit 1322 learns the correlation model between the shape model parameter values ​​in the hypothetical size database 1304 and the hypothetical spectroscopic spectra in the hypothetical spectroscopic spectrum database 1321. The shape estimation unit 1323 uses the correlation model obtained in the learning unit 1322 to estimate shape model parameter values ​​that best approximate the hypothetical spectroscopic spectrum obtained in the spectroscopic spectrum measurement device 1330. Furthermore, it outputs the shape obtained by substituting these shape model parameter values ​​into the shape model.

[0097] Furthermore, the hardware structures of the hypothetical shape data generation device 1300, the optical simulator 1310, and the optical shape estimation device 1320 are respectively related to... Figure 4B The outline resolution device 420 shown is similar, so repeated descriptions are omitted.

[0098] exist Figure 14 Showing through Figure 13 The flowchart shown illustrates a structural shape estimation system that estimates the shape from a spectroscopic spectrum measured by a spectroscopic measuring device 1330. The following uses... Figure 14 This will illustrate the shape estimation method.

[0099] The shape model specifications are set via the shape model setting unit 1301 (S301). For example, the type of model can be specified as "a line-symmetric shape model composed of 5 ellipses", or the type of shape model parameters describing the shape model can be specified.

[0100] Next, the generation method (S302) sets the group of shape model parameter values ​​by the generation method setting unit 1302. For example, by specifying a range of shape model parameters, multiple values ​​can be generated by dividing the range at equal intervals. Hereafter, this interval is referred to as the step size. Here, the range and step size can also be set to different values ​​according to the type of shape model parameter. Alternatively, random numbers can also be used for generation.

[0101] The parameter generation unit 1303 uses the generation method set in the generation method setting unit 1302 to generate a set of shape model parameter values, and stores the generated set in the imaginary dimension database 1304 (S303). The CAD unit 1305 outputs the imaginary shape model obtained by substituting the shape model parameter values ​​in the imaginary dimension database 1304 into the shape model as CAD data (S304).

[0102] The CAD section 1305 stores the generated CAD data (hypothesis shape model) in a hypothesis shape database 1306 (S305), and calculates a hypothesis spectral distribution using an optical simulator 1310. The calculated hypothesis spectral distribution is stored in a hypothesis spectral distribution database 1321 (S306). A learning section 1322 learns a correlation model of the hypothesis spectral distribution in the hypothesis spectral distribution database 1321 and the shape model parameters in the hypothesis size database 1304 (S307). Here, the correlation model represents a model of regression, classification, a model using a kernel method, a model using a neural network, a model using a decision tree, and the like.

[0103] A spectral distribution is obtained using a spectral distribution measurement device 1330 for a semiconductor sample (S308). A shape estimation section 1323 estimates the shape model parameters that give a hypothesis spectral distribution closest to the obtained spectral distribution using the correlation model learned by the learning section 1322 (S309). Here, the distance of the closeness of the evaluation value is calculated using any one of Euclidean distance, Manhattan distance, Chebyshev distance, and Mahalanobis distance.

[0104] Next, the shape estimation section 1323 substitutes the estimated shape model parameters into the shape model, and outputs the obtained shape (S310). This shape is hereinafter referred to as an estimated shape. Whether or not there is a defect with respect to the estimated shape is determined (S311), and in the case where a defect is confirmed, the process returns to step S301, and the specification of the shape model is redefined, and the generation method of the shape model parameters is redefined. Here, the defect with respect to the estimated shape is a case where the estimated shape model parameter value departs from the range of the shape model parameters set in the generation method setting section 1302, a case where a defective portion is generated in the estimated shape, and the like. In the case where a defect is not confirmed in step S311, the process ends.

[0105] Embodiment 4

[0106] Figure 15 is a diagram showing a configuration example of a contour line detection system in Embodiment 4. Embodiment 4 detects a contour line of an image obtained by an inspection device such as a SEM image.

[0107] Embodiment 4 aims at high precision of contour line detection from an image obtained by an inspection device. In particular, in a detection method using machine learning, the detection precision greatly depends on the quality and quantity of learning data of images and contour lines. In this embodiment, by generating various shapes and contour lines thereof using a shape model using a plurality of ellipses, the contour line detection is made highly precise.

[0108] Here, the measurement device 400 is defined in the same manner as the measurement device of Embodiment 1, and the hypothetical shape data generating device 1300 is defined in the same manner as the hypothetical shape data generating device of Embodiment 3. In this embodiment, the CAD data generated by the CAD section 1305 is hypothetical contour line data 1500 and hypothetical image data 1501. The hypothetical contour line data 1500 is contour line data of the object structure obtained by the shape model, which is set as CAD data by the CAD section 1305. The hypothetical image data 1501 is CAD data obtained by the CAD section 1305 performing correction of color, hue, brightness, noise amount, magnification, screen size, and the like on the shape obtained by the shape model, and is an image simulating an actual image obtained by the measurement device 400.

[0109] The contour line estimating device 1510 is a device that detects contour line data from the image data input from the measurement device 400 based on the hypothetical shape database 1306 of the hypothetical shape data generating device 1300. The contour line estimating device 1510 has a learning section 1511 and a contour line estimating section 1512. The hardware structure of the contour line estimating device 1510 is the same as that of the contour line analysis device 420 shown in FIG. 15, and thus repeated description is omitted. Figure 4B The contour line analysis device 420 shown in FIG. 15 is the same, and thus repeated description is omitted.

[0110] The learning section 1511 learns a correlation model of the hypothetical contour line data 1500 and the hypothetical image data in the hypothetical shape database 1306. The contour line estimating section 1512 estimates the hypothetical contour line data 1500 that gives the hypothetical image data 1501 closest to the image obtained by the measurement device 400 using the correlation model obtained in the learning section 1511. Further, this hypothetical contour line data is output.

[0111] The process of detecting the contour line from the image data obtained by the measurement device 400 by the contour line detecting system shown in FIG. 15 is shown in a flowchart in FIG. 16. The contour line detecting method is described below using the flowchart of FIG. 16. Figure 16 Figure 15 The processes of steps S401 to S403 of the method of FIG. 16 are the same as those of steps S101 to S103 of the method of FIG. 10, respectively. Figure 16

[0112] Figure 16 Figure 14 ​​​S301 to S303 are the same, so the explanation is omitted. The CAD unit 1305 takes the shape obtained by substituting the shape model parameter values ​​in the hypothetical dimension database 1304 into the shape model as CAD data, and outputs hypothetical image data 1501 and hypothetical contour data 1500 (S404). This CAD data is stored in the hypothetical shape database 1306 (S405). The learning unit 1511 learns the correlation model between the hypothetical image data 1501 and the hypothetical contour data 1500 in the hypothetical shape database 1306 (S406). Here, the correlation model represents regression, classification models, and models using kernel methods, neural networks, decision trees, etc.

[0113] Next, image data such as SEM images are acquired in the measuring device 400 (S407). The contour estimation unit 1512 uses the correlation model obtained in the learning unit 1511 to estimate imaginary contour data that is closest to the acquired image data (S408). Here, the distance for evaluating the closeness of the image is obtained by calculating any one of Euclidean distance, Manhattan distance, Chebyshev distance, and Mahalanobis distance based on values ​​related to color, hue, and brightness in each pixel. Next, the contour estimation unit 1512 outputs the estimated imaginary contour data (S409). The estimated imaginary contour data is then judged to be defective (S410). If a defect is confirmed, the process returns to step S401 to reset the shape model specifications and the method related to the generation of shape model parameters. Here, defects in the estimated imaginary contour data include situations where the shape model parameter values ​​of the imaginary contour data deviate from the range of the shape model parameters set in the generation method setting unit 1302, and situations where missing parts are generated in the imaginary contour data. If no defects are detected in step S410, the process ends.

[0114] Next, use Figure 17 , 18 To illustrate the GUI related to Examples 3 and 4.

[0115] Figure 17 The input GUI 1700 shown is an example of the input screen for setting the hypothetical shape data generation device 1300 in Embodiments 3 and 4. In Embodiment 3, this input screen prompts the user before step S301, and in Embodiment 4, it prompts the user before step S401.

[0116] The input GUI 1700 includes a shape model selection box 1710, a shape model parameter generation method setting box 1720, a valid / invalid display section 1730, and a confirmation button 1740. The shape model selection box 1710 and the shape model parameter generation method setting box 1720 respectively perform settings related to the shape model setting section 1301 and the generation method setting section 1302.

[0117] The shape model selection box 1710 has a model input section 1711. For example, it can select a line-symmetric shape model composed of a specific number of ellipses, or a shape model that does not assume line symmetry.

[0118] The shape model parameter generation method setting box 1720 has a range input section 1721 and a step size input section 1722. The range of shape model parameters is specified through the range input section 1721. This range can be different for each shape parameter. The step size input section 1722 specifies the step size to be divided within the range input section 1721. Figure 17 In the example shown, the ranges of shape model parameters Parameter1, 2, and 3 are divided into 100÷1=100, 10÷1=10, and 1÷0.1=10, respectively. That is, in this case, the generated shape model parameter values ​​are Parameter 1=(0, 1, 2, …100), Parameter 2=(0, 1, 2, …10), and Parameter 3=(0, 0.1, 0.2, …1).

[0119] The valid / invalid display section 1730 of each of the above setting boxes displays whether the above input has been valid. If all valid / invalid display sections 1730 are valid, pressing the decision button 1740 of the input GUI 1700 starts the process of step S303 in the case of embodiment 3, and starts the process of step S404 in the case of embodiment 4.

[0120] Press the Decision button 1740 in the GUI 1700 input method to execute. Figure 14 or Figure 16 In the process, Figure 18 The output GUI 1800 is displayed, and prompts are given at step S304 in Embodiment 3 and at step S404 in Embodiment 4. The GUI displays the current status, allowing the user to choose whether to proceed to the next step. The hypothetical shape generation result display unit 1810 includes a hypothetical shape data number display unit 1811, a CAD data display unit 1812, a shape model parameter display unit 1813, and a completion / resetting selection unit 1820.

[0121] The hypothetical shape data number display part 1811 displays the serial number of the generated shape model parameter. The shape model parameter display part 1813 displays the group of shape model parameters having the serial number displayed in the hypothetical shape data number display part 1811. That is, in the example of Figure 18 the 23rd generated shape model parameter is displayed. The CAD data display part 1812 displays the CAD data generated by the CAD part 1305 based on the group of shape model parameters.

[0122] The user can select completion or re-setting in the completion / re-setting selection part 1820 based on the information of the hypothetical shape displayed in the hypothetical shape generation result display part 1810. When the user judges that there is no defect in the CAD data, the shape model parameter, by selecting completion and pressing the decision button 1830, the process advances to step S305 in the case of Embodiment 3, and step S405 in the case of Embodiment 4. When it is judged that there is a defect, by selecting re-setting and pressing the decision button 1830, the screen of the input GUI 1700 is returned, and re-setting can be performed.

[0123] In the above, as the implementation of the illustrative embodiments 1 to 4, the semiconductor device manufacturing system in which the application for performing the operation management of the production line including the semiconductor processing device, the measuring device is executed on the platform is considered. In this case, by making the profile line detection device 410, the profile line analysis device 420, the processing condition decision device 900, the hypothetical shape data generation device 1300, the optical shape estimation device 1320, the profile line estimation device 1510 as the application on the platform, the respective processes are executed, and thus the embodiments 1 to 4 can be implemented in the semiconductor device manufacturing system.

[0124] Explanation of Reference Numerals

[0125] 100: ellipse, 110: start point, 111: end point, 120: shape model, 130: center coordinate, 131: length of short axis, 132: length of long axis, 133: inclination of short axis, 200: mask, 201: etched film, 210: maximum width, 211: depth, 212: width of bottom, 213: width of upper portion, 214: width of middle portion, 215: taper angle, 216: angle of arc, 217: eccentricity of bottom, 400: measuring device, 410: profile line detecting device, 420: profile line analyzing device, 421: analysis section, 422: shape model database, 423: shape model setting section, 424: shape model fitting setting section, 425: dimension extraction setting section, 426: dimension calculation method setting section, 600: profile line, 601, 602: shape model, 603: plausible shape model, 604: ellipse, 700: plausible maximum width, 701: plausible depth, 702: plausible width of bottom, 703: plausible eccentricity of bottom, 900: processing condition determining device, 901: processing condition database, 902: learning section, 903: processing condition estimating section, 904: target dimension value setting section, 910: semiconductor processing device, 1100: input GUI, 1110: profile line detection setting frame, 1111: detection method input section, 1112: image input section, 1120: shape model selection frame, 1121: model input section, 1130: fitting setting frame, 1131: fitting method input section, 1140: dimension extraction setting frame, 1141: dimension input section, 1150: calculation method setting frame, 1151: calculation method input section, 1160: valid / invalid display section, 1170: determination button, 1200: output GUI, 1210: dimension extraction result display section, 1211: plausible shape model display section, 1212: dimension calculation result display section, 1220: completion / re-setting selection section, 1230: determination button, 1300: hypothetical shape data generating device, 1301: shape model setting section, 1302: generation method setting section, 1303: parameter generating section, 1304: hypothetical dimension database, 1305: CAD section, 1306: hypothetical shape database, 1310: optical simulator, 1320: optical shape estimating device, 1321: hypothetical spectroscopic database, 1322: learning section, 1323: shape estimating section, 1330: spectroscopic measuring device, 1500: hypothetical profile line data, 1501: hypothetical image data, 1510: profile line estimating device, 1511: learning section, 1512: profile line estimating section, 1700: input GUI, 1710: shape model selection frame, 1711: model input section, 1720: shape model parameter generation method setting frame, 1721: range input section, 1722: step width input section,1730: valid / invalid display section, 1740: decision button, 1800: output GUI, 1810: hypothetical shape generation result display section, 1811: hypothetical shape data number display section, 1812: CAD data display section, 1813: shape model parameter display section, 1820: completion / re-setting selection section, 1830: decision button.

Claims

1. A contour line analysis apparatus that analyzes contour line data of an object structure detected from image data of a semiconductor sample acquired by a measuring apparatus that utilizes a charged particle beam device, the contour line analysis apparatus characterized by comprising: a shape model specification section that specifies a specification of a shape model, i.e., a shape model specification, the shape model being a one-stroke curve on a perimeter of a figure that is a combination of a plurality of ellipses; a shape model fitting specification section that specifies a fitting method of the shape model; an analysis section that causes the shape model of the shape model specification to be fitted to the contour line data by the fitting method specified in the shape model fitting specification section, thereby obtaining a plausible shape model of the contour line data; a shape model database that stores values of shape model parameters of the plausible shape model; a dimension extraction specification section that specifies a dimension desired to be extracted based on the contour line data; and a dimension calculation method specification section that specifies a calculation method of the dimension desired to be extracted specified in the dimension extraction specification section, the analysis section using the plausible shape model to calculate a value of the dimension desired to be extracted specified in the dimension extraction specification section by the calculation method specified in the dimension calculation method specification section.

2. The contour line analysis apparatus according to claim 1, characterized in that: the shape model parameters of the shape model include parameters related to shapes of the ellipses that constitute the figure and arrangement, and parameters related to the one-stroke method.

3. The contour line analysis apparatus according to claim 1, characterized in that: the shape model fitting specification section is capable of selecting a least squares method, a weighted least squares method, or a regularized least squares method for fitting, and is capable of selecting a nonlinear optimization method that utilizes an iterative solution method or a combinatorial optimization method in order to obtain the shape model parameters of the plausible shape model.

4. The contour line analysis apparatus according to claim 1, characterized in that: the dimension extraction specification section is capable of specifying a singular point search method in which a singular point on the shape model is searched for, and a value of the dimension desired to be extracted specified in the dimension extraction specification section is calculated based on coordinates of the singular point, the singular point including an extreme point in which a differential value is 0, a point at which differentiation cannot be performed, and an inflection point in which a positive or negative change in a differential value occurs in the shape model. comprising:

5. A processing condition decision system characterized by comprising: the contour line analysis apparatus according to claim 1; and a processing condition decision apparatus that decides a processing condition of a semiconductor processing apparatus, the semiconductor sample being a semiconductor sample that has been processed by the semiconductor processing apparatus, the processing condition decision apparatus comprising: a learning section that learns a correlation model of a processing condition of the semiconductor processing apparatus and values of shape model parameters of a plausible shape model of an object structure detected from image data of the semiconductor sample that has been processed by the semiconductor processing apparatus under the processing condition; a target dimension value specification section that specifies a target value of the shape model parameters of the plausible shape model; and a processing condition decision section that decides the processing condition of the semiconductor processing apparatus based on the correlation model learned by the learning section and the target value of the shape model parameters specified by the target dimension value specification section. ​ a processing condition estimation section that estimates a processing condition of the semiconductor processing apparatus that gives a target value of a shape model parameter of the plausible shape model set in the target size value setting section, using the correlation model.

6. A shape estimation system, comprising: a hypothetical shape data generation device that creates hypothetical shape data relating to an object structure possessed by a semiconductor sample; and a shape estimation device that estimates a shape of the object structure from data of the semiconductor sample acquired by a measurement device, the shape estimation system being characterized in that the hypothetical shape data generation device has: a shape model setting section that sets a specification of a shape model, i.e., a shape model specification, wherein the shape model is a one-stroke curve on a circumference of a figure that is a combination of a plurality of ellipses, and describes a shape of the object structure of the semiconductor sample; a generation method setting section that sets a generation method of generating a group of values of shape model parameters, with respect to a shape model specification set in the shape model setting section; a parameter generation section that generates a group of values of shape model parameters by the generation method set in the generation method setting section, with respect to the shape model specification set in the shape model setting section; and a CAD section that generates a hypothetical shape model of the object structure, based on a shape model obtained by substituting an arbitrary one of the group of values of shape model parameters generated in the parameter generation section into the shape model specification, the shape estimation device has: a learning section that learns a correlation model of values of shape model parameters of a shape model used in generation of a hypothetical shape model of the object structure, and hypothetical data obtained for the hypothetical shape model of the object structure as expected from data of the semiconductor sample acquired by the measurement device; and a shape estimation section that estimates values of shape model parameters that give hypothetical data closest to the data of the semiconductor sample acquired by the measurement device, using the correlation model.

7. The shape estimation system according to claim 6, characterized in that the measurement device is a spectroscopic measurement device, the shape estimation system further has: an optical simulator that calculates, for a hypothetical shape model of the object structure, a theoretical value of a spectrogram obtained by scatterometry for the object structure, that is, a hypothetical spectrogram, the learning section of the shape estimation device sets the hypothetical spectrogram calculated by the optical simulator as the hypothetical data.

8. A shape estimation system, comprising: a hypothetical shape data generation device that creates hypothetical shape data relating to an object structure possessed by a semiconductor sample; and a shape estimation device that estimates a shape of the object structure from data of the semiconductor sample acquired by a measurement device, the shape estimation system being characterized in that the hypothetical shape data generation device has: a shape model setting section that sets a specification of a shape model, i.e., a shape model specification, wherein the shape model is a one-stroke curve on a circumference of a figure that is a combination of a plurality of ellipses, and describes a shape of the object structure of the semiconductor sample; a generation method setting section that sets a generation method of generating a group of values of shape model parameters, with respect to a shape model specification set in the shape model setting section; a parameter generation section that generates a set of values of shape model parameters by a generation method set in the generation method setting section, for a shape model specification set in the shape model setting section; and a CAD section that generates first hypothetical data that simulates data of the semiconductor sample expected to be acquired by the measurement device, and second hypothetical data expected for a shape of the object structure from a shape model obtained by substituting any one of the set of values of the shape model parameters generated in the parameter generation section into the shape model specification, the shape estimation device has: a learning section that learns a correlation model of the first hypothetical data and the second hypothetical data generated in the CAD section; and a shape estimation section that estimates the second hypothetical data that gives the first hypothetical data closest to the data of the semiconductor sample acquired by the measurement device, using the correlation model.

9. The shape estimation system according to claim 8, wherein the measurement device is a measurement device that uses a charged particle beam device, the data of the semiconductor sample acquired by the measurement device is image data of the object structure, a contour line of the object structure is estimated as the shape of the object structure.

10. A data structure for describing a contour line of an object structure possessed by a semiconductor sample as a shape model that is a one-stroke curve on a circumference of a figure that is a combination of a plurality of ellipses, the data structure characterized by including: a first parameter related to a shape and arrangement of the ellipses that constitute the figure; and a second parameter related to a method of the one stroke, the data structure is used in a process of calculating a value of a dimension of the shape of the object structure, the first parameter includes at least one or more of a center coordinate, a major axis length, a minor axis length, and a tilt of a major axis of the ellipses that constitute the figure, the second parameter indicates whether to connect an inner circumference or an outer circumference of the ellipses.

11. A data structure for describing a contour line of an object structure possessed by a semiconductor sample as a shape model that is a one-stroke curve on a circumference of a figure that is a combination of a plurality of ellipses, the data structure characterized by including: a first parameter related to a shape and arrangement of the ellipses that constitute the figure; and a second parameter related to a method of the one stroke, the semiconductor sample is a semiconductor sample that has been processed by a semiconductor processing device, the data structure is used in a process of estimating a processing condition of the semiconductor processing device, the first parameter includes at least one or more of a center coordinate, a major axis length, a minor axis length, and a tilt of a major axis of the ellipses that constitute the figure, the second parameter indicates whether to connect an inner circumference or an outer circumference of the ellipses.

12. The data structure according to claim 11, wherein The data structure is used in a learning process of a correlation model of a process condition of the semiconductor processing device and a value of a shape model parameter of a plausible shape model of the object structure from profile line data detected from image data of the semiconductor sample processed by the semiconductor processing device under the process condition.

13. A semiconductor device manufacturing system, comprising: a semiconductor processing device; and a platform connected to the semiconductor processing device via a network, which performs a dimension measurement process of measuring a dimension of a measurement object processed by the semiconductor processing device using an image, the semiconductor device manufacturing system characterized in that the dimension measurement process has the following steps: a specification of a shape model, i.e., a shape model specification, is set, the shape model being a one-stroke curve on a periphery of a figure that is a combination of a plurality of ellipses; a fitting method of a shape model is set; the shape model of the shape model specification is fitted to profile line data of the measurement object by the set fitting method, thereby obtaining a plausible shape model of the profile line data; a value of a shape model parameter of the plausible shape model is stored; a dimension desired to be extracted is set from the profile line data; an arithmetic method of the set dimension desired to be extracted is set; and a value of the set dimension desired to be extracted is calculated by the set arithmetic method using the plausible shape model.

14. The semiconductor device manufacturing system according to claim 13, characterized in that the dimension measurement process is executed as an application provided in the platform.

15. A search device that searches for a shape model parameter value for obtaining a target processing shape in a semiconductor processing device, the search device characterized in that the estimation model is obtained using dimension data measured based on a profile line of a shape estimated using a shape function as learning data, The input parameter values are searched for by estimating a model, wherein the shape function is a function that utilizes a plurality of ellipses. ​

Images