TWI910791B - A scanning electron microscope for examining measurement models and a method for correcting measurement models. - Google Patents

Abstract

A scanning electron microscope with an inspection measurement model and a method for correcting the inspection measurement model are provided, which allows users to recognize the decrease in the reliability of the inspection or measurement results and make adjustments efficiently. A scanning electron microscope with an inspection and measurement model includes: a feature quantity analysis unit that takes as input an electron microscope image or waveform of the sample to be inspected or measured and calculates a first feature quantity, and simultaneously takes as input an electron microscope image or waveform of a calibration sample and calculates a second feature quantity; a learning unit that generates an inspection and measurement model based on the aforementioned second feature quantity; and a calibration unit that compares the aforementioned first feature quantity with a predetermined calibration range based on the aforementioned first feature quantity and the aforementioned inspection and measurement model, and issues an alarm when the aforementioned first feature quantity deviates from the aforementioned calibration range.

Description

A scanning electron microscope for examining measurement models and a method for correcting measurement models.

This invention relates to scanning electron microscopes, and more particularly to scanning electron microscopes with inspection measurement models and methods for correcting inspection measurement models.

Prior art in this field is known in Patent 1. Patent 1 describes an image processing system that pre-stores the detection range of a detector in a charged particle beam device in a memory device, uses this detection range to generate a simulated image of a 3D shape pattern, and pre-learns the relationship between the simulated image and the 3D shape pattern. [Prior Art Documents] [Patent Documents]

[Patent Document 1] Japanese Patent Application Publication No. 2022-29505

[Problem to be Solved by the Invention] Patent Document 1 describes image processing for presumed 3D shapes, and describes the use of images with known correction patterns of 3D shapes as output in 3D shapes. However, Patent Document 1 has the following drawback: it cannot detect whether the features of the actual sample being measured have deviated from the correction range, resulting in errors in the output of 3D shapes due to the inability to perform correct correction.

Therefore, this invention provides a scanning electron microscope for inspecting measurement models and a method for correcting these models, allowing users to recognize the decrease in the reliability of inspection or measurement results and to make efficient adjustments. [Technical Means for Solving the Problem]

To solve the above problems, the scanning electron microscope with an inspection and measurement model associated with the present invention comprises: a feature quantity analysis unit, which inputs an electron microscope image or waveform of the sample to be inspected or measured and calculates a first feature quantity, and simultaneously inputs an electron microscope image or waveform of a calibration sample and calculates a second feature quantity; a learning unit, which generates an inspection and measurement model based on the aforementioned second feature quantity; and a calibration unit, which compares the aforementioned first feature quantity with a predetermined calibration range based on the aforementioned first feature quantity and the aforementioned inspection and measurement model, and issues an alarm when the aforementioned first feature quantity deviates from the aforementioned calibration range.

Furthermore, the method for correcting the inspection and measurement model associated with this invention includes the following steps: a feature quantity analysis unit, which inputs an electron microscope image or waveform of the sample to be inspected or measured and calculates a first feature quantity, and simultaneously inputs an electron microscope image or waveform of a calibration sample and calculates a second feature quantity; a learning unit, which generates an inspection and measurement model based on the aforementioned second feature quantity; and a correction unit, which corrects the aforementioned inspection and measurement model by ensuring that the output of the aforementioned inspection and measurement model is consistent with the aforementioned second feature quantity. [Invention Benefits]

According to the present invention, a scanning electron microscope with an inspection measurement model and a method for correcting the inspection measurement model can be provided, allowing the user to recognize the decrease in the reliability of the inspection or measurement results and to make adjustments efficiently. Other problems, structures, and effects not mentioned above will be explained by the following description of embodiments.

The embodiments are described with reference to the drawings. Furthermore, the embodiments described below are not limited to the invention within the scope of the patent application, and not all elements and combinations described in the embodiments are necessarily necessary for the solutions of the invention. The following describes embodiments of the invention using drawings. [Implement 1]

Figure 1 is an overall schematic diagram of a scanning electron microscope with an inspection measurement model associated with Embodiment 1 of the present invention. As shown in Figure 1, the scanning electron microscope 1 with an inspection measurement model includes a scanning electron microscope 100, a control device 120, a power supply device 121, a power supply device 122, and a computer 200.

A scanning electron microscope 100 irradiates a sample 108 with an electron beam (electron beam) 103. The scanning electron microscope 100 outputs a detection signal obtained based on the irradiation of the electron beam 103. The scanning electron microscope 100, having an inspection measurement model, possesses the necessary components for forming signal waveforms and images based on the detection signal from the scanning electron microscope 100. First, referring to FIG. 1, an example of the scanning electron microscope 100 will be specifically discussed.

An electron beam 103, drawn from an electron source 101 using an extraction electrode 102, is accelerated using an accelerating electrode (not shown). The accelerated electron beam 103 is focused using a focusing lens 104, which functions as a focusing lens. The focused electron beam 103 is then used to scan the sample 108 in one or two dimensions using a scanning electrode 105. The electron beam 103 is decelerated by applying a negative voltage to an electrode located within the sample stage 109, and simultaneously focused using the lens effect of the objective lens 106, illuminating the sample 108.

When the electron beam 103 is irradiated onto the sample 108, it is scattered from the irradiated area into the interior of the sample 108, and emitted as electrons 110, such as secondary electrons (SE) and backscattered electrons (BSE). These emitted electrons 110 are accelerated in the direction of the electron source 101 due to the retarding voltage applied to the sample 108, and collide with the switching electrode 112, generating secondary electrons 111. The secondary electrons 111 emitted from the switching electrode 112 are captured by a detector 113, and the detection signal (output of the detector 113) changes according to the amount of secondary electrons 111 captured.

The detection signal output from detector 113 is supplied to computer 200 via control device 120. Computer 200 has a display device described later. The brightness of the image displayed on this display device varies due to the detection signal. That is, the amount of electrons (electron quantity) captured by detector 113 is displayed on the display device in the form of brightness. For example, in the case of displaying a two-dimensional image, synchronization is achieved between the deflection signal supplied to scanning electrode 105 and the detection signal output from detector 113, and the brightness of the image in the scanned area scanned by the deflection signal is displayed on the display device.

Furthermore, regarding the scanning electron microscope 100 shown in FIG1, although there are no particular limitations, it is equipped with a deflector (not shown) that moves the scanning area of the electron beam 103. This deflector is used to display images of patterns of the same shape existing at different positions on a display device. This deflector, also called an image shift deflector, can move the field of view (FOV) position of the scanning electron microscope 100 without moving the sample 108 using a sample stage (e.g., sample stage) 109 that moves the sample 108. It is also possible to make the image shift deflector and the scanning electrode 105 common deflector, and to superimpose the image shift signal and the deflection signal to the deflector.

Detection signals (images, brightness distribution, brightness, etc.) from the scanning electron microscope 100 are supplied to the computer 200 via the control device 120. The computer 200, based on the supplied detection signals, calculates and outputs at least one value relating to changes in the shape of the object being examined or measured. Alternatively, the computer 200 may be integrated with the scanning electron microscope 100.

Control device 120 controls power supply devices 121 and 122 according to instructions from computer 200. Controlling power supply device 122 causes a change in the voltage applied to lead-out electrode 102 and accelerating electrode (not shown). Similarly, controlling power supply device 121 causes a change in the retarding voltage applied to sample 108. Furthermore, control device 120, according to instructions from computer 200, controls the deflection signal supplied to scanning electrode 105 and the signal supplied to objective lens 106. Furthermore, as described above, control device 120 supplies the detection signal output from detector 113 to computer 200.

<Computer Configuration> Next, the computer 200 of the scanning electron microscope 1 with a test measurement model associated with this embodiment will be explained using diagrams. Figure 2 is a functional block diagram of the computer shown in Figure 1. As shown in Figure 2, the computer 200 of the scanning electron microscope 1 with a test measurement model associated with this embodiment includes a suspected SEM waveform calculation unit 201, a feature quantity analysis unit 202, a learning unit 203, a correction unit 204, a measurement unit 205, a memory unit 206, a communication I/O unit 207, an input/output I/O unit 208, a memory medium 223, and a test measurement model 230, which are interconnected via an internal bus 209. Furthermore, the input/output device 221 and the display device 222 are connected via the input/output I/F 208. Here, the processing unit is composed of a suspected SEM waveform calculation unit 201, a feature analysis unit 202, a learning unit 203, a correction unit 204, and a measurement unit 205. It is implemented, for example, by a processor such as a CPU (not shown), a ROM that stores various programs, RAM that temporarily stores data of the calculation process, an external memory device, or other memory devices. At the same time, the processor such as the CPU reads and executes various programs stored in the ROM, and stores the calculation results of the execution results in the cloud via RAM, an external memory device, or a network connection.

Input/output device 221, such as a mouse or keyboard, is used to allow the user to input data and give instructions to the processing unit. Display device 222 is used to display data obtained by the processing unit. In addition, memory unit 206 is used to store data while the processing unit is executing programs. The processing unit reads and executes the program stored in memory medium 223. Figure 3 is a diagram illustrating the program stored in the memory medium shown in Figure 2. Memory medium 223 stores the suspected SEM waveform calculation program 211, feature analysis program 212, learning program 213, calibration program 214, and measurement program 215 in a predetermined area.

The suspected SEM (Scanning Electron Microscope) waveform calculation unit 201 accesses the memory medium 223 via the internal bus 209, reads and executes the suspected SEM waveform calculation program 211. The suspected SEM waveform calculation unit 201, for example, includes information such as pattern shape and material, information on incident electrons such as accelerating voltage, detector configuration, and detection energy filter information. Using a simulation model as input, for example, Monte Carlo calculations, it outputs a suspected SEM image or waveform. The shape of the simulation model, the suspected SEM image, and the waveform are described in detail in Figure 6.

The feature quantity analysis unit 202 accesses the memory medium 223 via the internal bus 209, reads and executes the feature quantity analysis program 212. The feature quantity analysis unit 202 takes the SEM image or waveform captured by the scanning electron microscope 100, or the image waveform generated by the suspected SEM waveform calculation unit 201, as input via the internal bus 209, and outputs the feature quantities fi (i=1 to n) of the image or waveform. The feature quantity fi is, for example, a value capturing the maximum brightness of the image, the peak brightness of the waveform, the width of the peak, and other characteristic values of the waveform.

The learning unit 203 accesses the memory medium 223 via the internal bus 209, reads and executes the learning program 213. The learning unit 203 inputs, for example, the pattern shape used for calculation by the suspected SEM waveform calculation unit 201, and the feature quantity fi obtained by analyzing the output of the suspected SEM waveform calculation unit 201 using the feature quantity analysis unit 202, via the internal bus 209. Furthermore, using the feature quantity fi as input, it generates an inspection and measurement model 230 with the size of the pattern shape as output through learning, and stores it, for example, on a recording medium (not shown). Alternatively, the inspection and measurement model 230 can also be configured to be stored in the memory medium 223.

The calibration unit 204 accesses the memory medium 223 via the internal bus 209, reads and executes the calibration program 214. The calibration unit 204 takes the feature quantity fi obtained by the feature quantity analysis unit 202 from analyzing the SEM image and waveform of the calibration sample (with known pattern shape dimensions) captured by the scanning electron microscope 100, and inputs the pattern shape dimensions of the calibration sample and the inspection measurement model 230 via the internal bus 209. Furthermore, it performs calibration to ensure that the output of the inspection measurement model 230 matches the pattern shape dimensions of the calibration sample, thus correcting the inspection measurement model 230. Details of the calibration are described in Figure 9.

The measurement unit 205 accesses the memory medium 223 via the internal bus 209, reads and executes the measurement program 215. The measurement unit 205 examines the measurement model 230 and takes as input the feature quantity fi obtained by analyzing the SEM image and waveform of the sample 108 (FIG. 1) to be examined or measured by the scanning electron microscope 100 and the feature quantity fi obtained by the feature quantity analysis unit 202. The measurement unit 205 outputs the dimensions of the sample 108 (FIG. 1).

Alternatively, the entirety or a portion thereof of the suspected SEM waveform calculation unit 201, feature analysis unit 202, learning unit 203, correction unit 204 and measurement unit 205 may be implemented in a computer different from the computer 200, and the results may be output to the computer 200.

<Flowchart for Creating (Learning and Correcting) the Inspection Measurement Model> Figure 4 is a flowchart illustrating the process of creating (learning and correcting) the inspection measurement model using a scanning electron microscope with an inspection measurement model associated with this embodiment. As shown in Figure 4, in step S301, the simulation model and dimensional parameters are input via the input/output device 221 constituting the computer 200. The input/output device 221 transfers the simulation model and dimensional parameters input via the input/output I/F 208 and the internal bus 209 to the suspected SEM waveform calculation unit 201. In step S302, the suspected SEM waveform calculation unit 201, which constitutes the computer 200, performs a simulation based on the simulation model and dimensional parameters transmitted by the input/output device 221, generating a suspected SEM image. The suspected SEM waveform calculation unit 201 transmits the generated suspected SEM image to the feature quantity analysis unit 202 via the internal bus 209. In step S303, the feature quantity analysis unit 202, which constitutes the computer 200, analyzes the features of the suspected SEM image transmitted by the suspected SEM waveform calculation unit 201. The feature quantity analysis unit 202 transmits the analyzed feature quantity fi and dimensional parameters to the learning unit 203 via the internal bus 209. In step S304, the learning unit 203, which constitutes the computer 200, generates learning data consisting of characteristic quantities fi and dimensional parameters transmitted via the internal busbar 209. In step S305, the learning unit 203, based on the learning data, uses learning to create a test model 230. Steps S302 to S305 are collectively referred to as the learning steps (S311).

Once the actual sample is obtained in step S306, it is imaged using a scanning electron microscope 100 in step S307. Here, if the size of the actual sample is unknown, a ground truth value for the size is obtained using a cross-sectional TEM (Transmission Electron Microscopy) or other device (step S308). If the size of the actual sample is known and design information is used as the ground truth value, step S308 can be omitted. The imaging results obtained using the scanning electron microscope 100 in step S307 are input by the control device 120 via communication I/F 207. The communication I/F207 transmits the imaging results from the scanning electron microscope 100 to the feature quantity analysis unit 202 via the internal bus 209. The feature quantity analysis unit 202 then transmits the feature quantity analysis results to the calibration unit 204 via the internal bus 209. In step S309, the calibration unit 204 inputs the feature quantity analysis results transmitted from the feature quantity analysis unit 202 to the dimensions obtained from the inspection measurement model 230, and compares these dimensions with the baseline true value to correct the inspection measurement model. Afterwards, the calibration unit 204 outputs the corrected inspection measurement model 230 and the calibration range 330. Steps S306 to S309 are collectively referred to as the calibration steps (step S310).

<Measurement Flowchart> Figure 5 is a flowchart illustrating the measurement process performed using a scanning electron microscope with an inspection measurement model associated with this embodiment. The scanning electron microscope 100 performs measurements using the inspection measurement model 230 and calibration range 330 created in the inspection measurement model creation process.

As shown in Figure 5, in step S401, the scanning electron microscope 100 images the sample 108 (Figure 1) to be inspected or measured. In step S402, the feature quantity analysis unit 202 receives the captured SEM image and waveform via the control device 120, communication I/F 207, and internal bus 209, and performs feature quantity analysis. Then, in step S403, the feature quantity analysis results performed by the feature quantity analysis unit 202 are transferred to the measurement unit 205 via the internal bus 209, and the measurement unit 205 outputs the measured value. Next, in step S404, the calibration unit 204 inputs the characteristic quantity analysis results into the measured values or characteristic quantity analysis results obtained by checking the measurement model 230, and compares them with the calibration range 330 to determine whether it has deviated from the calibration range 330. Details regarding the calibration range 330 will be described later. If the determination result indicates that it has deviated from the calibration range 330, an alarm is issued (step S405), prompting adjustment (step S406). Furthermore, the adjustment (step S406) indicates that the calibration steps are repeated using the sample 108 (Fig. 1) that should have been checked or measured when the alarm was issued (step S310). On the other hand, if it is not separated, it is displayed within the correction range (step S407). <Examples of Sample, SEM Image, Distribution, and Feature Quantity> Next, examples of the pattern shape formed on sample 108 and examples of the electron microscope image obtained when the pattern shape is photographed (observed) with a scanning electron microscope 100 will be explained using diagrams. Figure 6 is a diagram illustrating the pattern shape formed on the sample and the electron microscope image. Here, the upper left view of Figure 6 shows an oblique view of the pattern shape, and the upper right view of Figure 6 shows the electron microscope image. The lower left of Figure 6 shows a cross-sectional view of the pattern shape shown in the upper left of Figure 6; the lower right of Figure 6 is a diagram illustrating the brightness distribution (the change in brightness with coordinates) in the electron microscope image shown in the upper right of Figure 6.

Here, regarding sample 108, a semiconductor wafer used for manufacturing semiconductor devices will be used as an example for explanation. Since it is a semiconductor wafer, the material of sample 108 is silicon (Si). Sample 108 is etched to form a pattern shape 500 as shown in the upper left image of Figure 6. Here, regarding the pattern of pattern shape 500, an L/S (Line and Space) with a concave shape will be used as an example for explanation.

In the upper left of Figure 6, UP represents the main surface (first surface) of the line L of pattern shape 500, and SD represents the sidewall (second surface) of line L. Furthermore, DW represents the back surface of sample 108 facing the main surface UP. The electron beam microscope image 510 in the upper right of Figure 6 is an image captured by a scanning electron microscope 100 from directly above sample 108. That is, it is an image captured by irradiating sample 108 with an electron beam 103 from directly above the main surface UP; this is electron microscope image 510.

The lower left view of Figure 6 shows the cross-section 502 of the pattern shape 500 at section position 501 in the pattern shape 500 shown in the upper left view of Figure 6. Furthermore, the lower right view of Figure 6 shows the brightness distribution 512 at section position 511 in the electron microscope image 510 shown in the upper right view of Figure 6. Characteristic quantities fi include, for example, the brightness at the top of the pattern 503 513, the peak brightness near the edge of the pattern 504 514, the width of the distribution, etc.

<Details of Learning and Correction> Next, details of learning and correction will be explained. Figure 7 is a diagram illustrating the dimensional parameters used to create learning data. A set of dimensional parameters 602 is prepared for the region 601 larger than the assumed range of dimensional variation during mass production, representing the ratio of dimension A to dimension B. Here, an example of preparing 6 × 5 = 30 dimensional parameter sets 602 is shown. Although only dimensions A and B are shown for illustration purposes, sometimes other dimensions are varied to prepare more dimensional parameter sets 602. Furthermore, sometimes dimensional parameter sets 602 are prepared at non-equidistant intervals in the parameter space. Here, the range of dimensional parameters in the learning data is referred to as the learning range 603. Additionally, in Figure 7, dimensions A and B represent, for example, the height of the pattern and the width of the top of the pattern. However, this is not the only limitation.

An example of generating a suspected SEM image using the size parameter set 602 of Figure 7 and performing feature analysis is shown in Figure 8. As shown in Figure 8, a feature dataset 702 is obtained corresponding to the size parameter set 602. At this time, the inner region connected at the outermost point of the feature space is called the learning range 703. Learning is performed, for example, using linear or nonlinear regression of the feature fi. At this time, the measurement model 230 is examined, for example, as shown in the following equation (1), represented by a weighted nonlinear combination of features.

Here, n is the number of eigenvalues, a _₀ and a _{_ij} are coefficients (constant values), f _^ij are eigenvalues, and N is the number of nonlinear combinations considered. Examples of n=3 and N=1 are shown in equation (2) below, and examples of n=2 and N=2 are shown in equation (3) below. Alternatively, equation (1) can be replaced with a more complex polynomial, or methods employing non-numerical representations, such as deep learning models, can be used.

For the example of correction, Figure 9 will be used for explanation. For simplicity, an example of the linear combination of equation (2) is shown. Assume that the learning unit 203 (Figure 2) constituting the computer 200 has learned the learning data 801 and obtained the inspection measurement model 801. The plot of the estimated value p and the reference truth value of size A when the calibration sample was measured is plot 803. In the left figure of Figure 9, the tilt of the line is kept and only the offset is corrected to obtain the corrected inspection measurement model 804. In the right figure of Figure 9, the tilt and offset of the line are corrected to obtain the corrected inspection measurement model 805. Here, although a simplified case of linear combination is shown, and although this is generally referred to as correcting the inspection measurement model based on the measured data of the calibration sample and the benchmark truth value (sometimes also called correction), in addition to correcting the inspection measurement model, it is also possible to perform correction by transforming the characteristic quantity fi into the reporting quantity fi' using any function. Generally speaking, to perform higher-order corrections, more measurement data of the calibration sample is required.

The correction range, which is simultaneously established during calibration, will be explained. Figure 10 is a diagram illustrating an example of the correction range expressed in feature space. The left diagram of Figure 10 shows an example of a correction range 901 defined by the range of each feature fi at the time of measurement of the calibration sample. The right diagram of Figure 10 shows an example of a correction range 902 defined by the relationship between the multiple features based on the existence range between the multiple features at the time of measurement of the calibration sample. Furthermore, although the right diagram of Figure 10 is presented as a 2D graph for illustration purposes, the correction range 902 can also be defined in a higher-order multidimensional space. Moreover, the correction range can be defined not only in feature space but also by dimensional values. The left image of Figure 11 shows an example of defining the correction range 1001 of the measured value p using the maximum and minimum values of the reference truth size of the correction sample. Furthermore, as shown in the right image of Figure 11, the correction range 1002 can also be defined using multiple dimensions (pi and pj). Additionally, although the right image of Figure 11 is presented as a 2D graph for illustration purposes, the correction range can also be defined in a higher-order multidimensional space.

<Regarding Deviation from the Calibration Range of Measurements> Figure 12 is an explanatory diagram regarding deviation from the calibration range. Here, the calibration ranges 902 and 1002 from the right-hand images of Figure 10 and Figure 11 are used for explanation. Deviation from the calibration range means that the characteristic quantity f and the estimated value p obtained from the image captured from the corresponding sample 108 being inspected or measured are outside the calibration ranges 902 and 1002. In Figure 12, measured data 1101 and measured data 1111 are outside the calibration range (902, 1002), while measured data 1102 and measured data 1112 are not outside the calibration range (902, 1002).

So far, for the sake of simplicity, the explanation of deviation from the calibration range has been provided. However, sometimes the data obtained from measuring sample 108 deviates from both the calibration range and the learning range. In this case, the possibility of measurement errors larger than those caused by deviation from the calibration range is high. Below, the deviation from the learning range will be explained, and a flowchart showing the deviation from the learning range will be shown at the end.

<Regarding the separation from the learning range during measurement> Figure 13 is a diagram illustrating the relationship between the calibration range 902 and the learning range 703. If the data obtained during calibration exceeds the learning range 703, it is necessary to review the dimensional parameters of step S301 (Figure 4) and repeat the learning step S311 (Figure 4). Therefore, during measurement, the calibration range 902 is contained within the learning range 703.

Figure 14 illustrates the separation from the learning range 703. Separation from the learning range 703 means that the characteristic quantity f and the estimated value p derived from the image captured of the corresponding sample 108 being inspected or measured are outside the learning range 703 and/or the learning range 603 (Figure 7). In Figure 14, measurement data 1301 is separated from the learning range 703, measurement data 1302 is separated from the correction range 902, and the learning range 703 is not separated. Measurement data 1303 is not separated from either the correction range 902 or the learning range 703.

<Flowchart for Creating an Inspection Measurement Model Considering the Detachment from the Learning Range (Learning and Correction)> Figure 15 is a flowchart illustrating the creation of an inspection measurement model considering the detachment from the learning range for a scanning electron microscope with an inspection measurement model associated with this embodiment. Here, only the parts different from Figure 4 will be explained. The learning unit 203 (Figure 2) constituting the computer 200, after step S304 during learning data creation, performs learning range creation (step S1401) in parallel with step S305, creating learning range 1401. It includes steps S302 to S305 and step S1401, and is called learning step S1402.

<Measurement Flowchart Considering the Detachment from the Learning Scope> Figure 16 is a measurement flowchart considering the detachment from the learning scope, using a scanning electron microscope associated with the measurement model described in this embodiment. Only the parts different from Figure 5 are explained here. The correction unit 204 determines whether the device has deviated from the correction range 330 (step S404). If it has, it further determines whether the device has deviated from the learning range 1401 (step S1501). If the device has not deviated, it issues an alarm indicating that it has deviated from the correction range 330 (step S405), prompting adjustment (step S406). On the other hand, if the device has deviated from the learning range 1401, it issues an alarm indicating that it has deviated from the learning range 1401 (S1502), prompting relearning (step S1503). Here, "learn again" means to review the simulation model and dimensional parameters of step S301 (Figure 15) and then execute the learning step S1402 (Figure 15) again.

Figure 17 illustrates an example where the measured data has fallen outside the calibration range. As shown in the left image of Figure 17, an alarm is issued when the measured data group 1601 has fallen outside the calibration range 902. At this point, as shown in the right image of Figure 17, an analysis is performed on the distribution of the calibration range 902 and the measured data group 1601. Additional calibration points are selected so that all measured data points 1601 are included. Here, measured data 1602, measured data 1603, and measured data 1604 are used as additional calibration data to redefine the new calibration range 1605.

When calibration is required, measurements need to be performed using cross-sectional TEM, AFM (Atomic Force Microscope), or similar instruments at the point where the measurement data is obtained. In this embodiment, the calibration range 902 and the measurement data group 1601 are analyzed, and adjustments are made with fewer calibration points, thereby reducing the effort required to obtain the baseline truth data.

Figure 18 is a diagram illustrating the graphical user interface (GUI) of the display device shown in Figure 2. As shown in Figure 18, the GUI consists of execution button group 1710 for each program, simulation model and parameter setting area 1720 for the suspected SEM waveform calculation program 211, setting area 1730 for the feature analysis program 212, setting area 1740 for the learning program 213, setting area 1750 for the calibration program 214, setting area 1760 for the measurement program 215, and display area 1770.

In the simulation model and parameter setting area 1720, input the shape of the model and the parameter settings. In the eigenvalue analysis program setting area 1730, specify which part of the waveform's eigenvalues to extract, either displayed or defined by eigenvalue calculation formulas, even if not graphically shown on the GUI. In the learning program setting area 1740, select learning methods such as regression or deep learning, and choose from drop-down menus for the model used, such as linear or nonlinear, or define it numerically. In the calibration program setting area 1750, select from drop-down menus the calibration method for the model selected in the learning program setting area 1740, such as offset (other methods are explained in Figure 9). In the settings area 1760 of the measurement program, the model to be used is selected from a drop-down menu from multiple models defined by pattern shape (Line and space, Hole, etc.) and representative dimensions of the pattern.

In display area 1770, the execution results of various programs are displayed. Figure 18 shows an example of the execution of measurement program 215, displaying the measured value 1771 and simultaneously displaying an alarm 1772 indicating that the calibration range has deviated. Additional calibration data used for adjustment, such as re-measuring a specific chip on wafer map 1773, displays the current measurement coordinate information 1774. Furthermore, chips 1775–1777 on wafer map 1773 correspond to the additional calibration data 1602–1604 (right view of Figure 17). The user performs adjustments based on these displays. Although the explanation uses a feature space, it is also possible to compare the calibration range with the distribution of the measurement data set using the space of the estimated values (measured values) shown in the right view of Figure 11.

When calibration is required, measurements need to be performed using cross-sectional TEM, AFM (Atomic Force Microscope), etc., at the locations where measurement data has been obtained. In this embodiment, the calibration range 902 and measurement data group 1601 (left panel of Figure 17) are analyzed, and adjustments are made with fewer calibration points, thereby reducing the effort required to obtain the reference truth data.

In this embodiment, although learning and calibration are described separately, regarding the learning data, several calibration samples are prepared, SEM images of the calibration samples are obtained, and a feature set 702 (Fig. 8) is created. Measurements are performed on the calibration samples using, for example, cross-sectional TEM, AFM, etc., and a set of dimensional parameters 602 (Fig. 7) is obtained. Equations (1) to (3) are executed using the set of dimensional parameters 602 (Fig. 7) and the set of feature data 702 (Fig. 8). Deep learning model learning can also be performed. In this case, the learning range and the calibration range are equal.

Furthermore, although this embodiment uses dimensional measurement as an example for explanation, the monitoring (alarm) of the learning range and correction range separation, and the timing of the observation point for efficient adjustment, are widely used and not limited in dimensional measurement due to the use of formulas and models for processing. It can also be applied to defect inspection, defect classification, pattern recognition, and image conversion (image quality improvement, high-magnification estimation).

As described above, this embodiment provides a scanning electron microscope with an inspection measurement model and a method for correcting the inspection measurement model. This allows users to recognize the decrease in the reliability of the inspection or measurement results and to make adjustments efficiently. [Embodiment 2]

This embodiment illustrates a computer 200 that not only issues a warning when a user goes out of the learning scope but also suggests a suitable learning scope for relearning.

The flowcharts for checking the measurement model considering the case of the learning range being disengaged in Figure 15 above, and the measurement flowcharts considering the case of the learning range being disengaged in Figure 16, are the same as in Embodiment 1. The difference lies in the action taken when issuing an alarm in step S1502 (Figure 16).

Figure 19 is a diagram illustrating an example where measurement data in a scanning electron microscope with a test measurement model associated with this embodiment has deviated from the learning range. As shown in the upper left of Figure 19, in the case of an alarm related to deviating from the learning range (step S1502 (Figure 16)), that is, when the measurement data group 1801 is located outside the learning range 703, an analysis is performed on the position of the size group 1811 in the learning range 603 in the space of the size parameters, which is inferred from the measurement data group 1801 shown in the upper right of Figure 19, in order to define a new learning range 1812 in a manner that includes the size group 1811. Furthermore, as shown in the lower part of Figure 19, a set of dimensional parameters 1813 is proposed for use in learning the new learning range 1812.

This means that even if the learning model is outside the learning range, there are still situations where the size can be estimated (i.e., extrapolation). In such cases, the new learning range can be roughly grasped based on the estimated size outside the learning range.

In accordance with this embodiment, in addition to the effects of Embodiment 1, it is easy to know what level of simulation data needs to be added when resimulating.

Furthermore, the present invention is not limited to the embodiments described above, but includes various variations. For example, the embodiments described above are detailed for the purpose of explaining the present invention in an easily understandable manner, and are not necessarily limited to having all the described components. In addition, a part of the components of one embodiment may be replaced with the components of other embodiments, and the components of other embodiments may be added to the components of a certain embodiment.

Furthermore, the aforementioned components can also be implemented in hardware, either partially or entirely, for example, by designing integrated circuits. Additionally, the aforementioned components can be implemented in software by a processor interpreting and executing programs that implement individual functions. The programs, tables, files, and other information implementing each function can be stored on recording devices such as memory, hard drives, and SSDs (Solid State Drives), or on recording media such as IC cards, SD cards, and DVDs.

Furthermore, control lines, information lines, etc., are only shown as required in the description; it is not necessarily true that all control lines, information lines, etc., are shown on the product. They can also be considered as almost all components actually interconnected.

1: Scanning electron microscope with inspection and measurement model 100: Scanning electron microscope 101: Electron source 102: Lead-out electrode 103: Electron beam 104: Focusing lens 105: Scanning electrode 106: Objective lens 108: Sample 109: Sample stage 110: Electron 111: Secondary electron 112: Conversion electrode 113: Detector 120: Control device 121, 122: Power supply device 200: Computer 201: Suspected SEM waveform calculation unit 202: Feature quantity analysis unit 203: Learning unit 204: Calibration unit 205: Measurement unit 206: Memory unit 207: Communication I/F 208: Input/Output I/F 209: Internal Bus 211: Suspected SEM Waveform Calculation Program 212: Feature Analysis Program 213: Learning Program 214: Calibration Program 215: Measurement Program 221: Input/Output Device 222: Display Device 223: Memory Media 230: Check Measurement Model 330, 902, 1002: Calibration Range 500: Pattern Shape 501: Cross-Section Position 502: Cross-Section 510: Electron Beam Microscope Image 511: Cross-Section Position 512: Brightness Distribution 601: Desired Region 602: Size Parameter Set 603, 703: Learning Range 702: Feature Data Set 901: Range of Each Feature fi (Calibration Range) 902: Correction range defined by the relational nature of complex eigenvalues 1001: Correction range for measured value p 1002: Correction range defined by a complex number of dimensions (pi and pj) 1101, 1102, 1111, 1112: Measured data 1801: Measured data set 1811: Estimated dimension set 1812: New learning range 1813: Dimensional parameter set

[Figure 1] is an overall schematic diagram of the scanning electron microscope with an inspection measurement model associated with Embodiment 1 of the present invention. [Figure 2] is a function block diagram of the computer shown in Figure 1. [Figure 3] is a diagram illustrating the program stored in the memory medium shown in Figure 2. [Figure 4] is a flowchart illustrating the actions of creating (learning and correcting) the inspection measurement model using the scanning electron microscope with the inspection measurement model associated with Embodiment 1. [Figure 5] is a flowchart illustrating the measurement actions performed using the scanning electron microscope with the inspection measurement model associated with Embodiment 1. [Figure 6] is a diagram illustrating the pattern shape formed on the sample and the electron microscope image. [Figure 7] is a diagram illustrating the dimensional parameters used to create learning data. [Figure 8] is a diagram illustrating an example of the results of eigenvalue analysis. [Figure 9] is a diagram illustrating an example of correction. [Figure 10] is a diagram illustrating an example of the correction range expressed in eigenvalue space. [Figure 11] is a diagram illustrating an example of the correction range defined by dimensional values. [Figure 12] is an explanatory diagram regarding the separation of the correction range. [Figure 13] is a diagram illustrating the relationship between the correction range and the learning range. [Figure 14] is an explanatory diagram regarding the separation of the learning range. [Figure 15] is a flowchart illustrating the inspection measurement model using a scanning electron microscope with an inspection measurement model associated with Embodiment 1, taking into account the separation of the learning range. [Figure 16] is a measurement flowchart using a scanning electron microscope with an inspection measurement model associated with Embodiment 1, taking into account the separation of the learning range. [Figure 17] is a diagram illustrating an example where the measurement data has separated from the calibration range. [Figure 18] is a diagram illustrating a screen display example (GUI) of the display device shown in Figure 2. [Figure 19] is a diagram illustrating an example in which measurement data in a scanning electron microscope with a measurement model associated with Embodiment 2 of the present invention has deviated from the learning scope.

120: Control Device

200: Computer

201: Suspected SEM waveform calculation department

202: Eigenvalue Analysis Section

203: Study Department

204:Correction Department

205: Measurement Department

206: Memory Department

207: Communications I/F

208: Input/Output I/F

209: Internal Busbar

221: Input/output device

222: Display Device

223: Memory Media

230: Inspect the measurement model

Claims (13)

A scanning electron microscope with an inspection and measurement model includes: a feature quantity analysis unit that takes an electron microscope image or waveform of the sample to be inspected or measured as input and calculates a first feature quantity, and simultaneously takes an electron microscope image or waveform of a calibration sample as input and calculates a second feature quantity; a learning unit that generates the inspection and measurement model based on the second feature quantity; and a calibration unit that compares the first feature quantity with a predetermined calibration range based on the first feature quantity and the inspection and measurement model, and issues an alarm when the first feature quantity deviates from the calibration range. As in claim 1, a scanning electron microscope with an inspection measurement model, wherein the aforementioned correction unit corrects the aforementioned inspection measurement model by ensuring that the output of the aforementioned inspection measurement model is consistent with the aforementioned second characteristic quantity. As in claim 1, a scanning electron microscope with an inspection measurement model, wherein the aforementioned calibration unit issues an alarm when the aforementioned first characteristic quantity deviates from the aforementioned calibration range, and deviates from a learning range that includes the aforementioned predetermined calibration range or is wider than the aforementioned predetermined calibration range. As in claim 1, a scanning electron microscope with an inspection measurement model, wherein the aforementioned correction unit corrects the aforementioned correction range by including the aforementioned first characteristic quantity when the aforementioned first characteristic quantity deviates from the aforementioned correction range, thereby correcting the aforementioned inspection measurement model. As in claim 3, a scanning electron microscope with an inspection measurement model, wherein the aforementioned correction unit corrects the aforementioned learning range in a manner that includes the aforementioned first feature when the aforementioned first feature deviates from the aforementioned correction range and deviates from a learning range that includes the aforementioned predetermined correction range or is wider than the aforementioned predetermined correction range. As in claim 4, a scanning electron microscope with an inspection measurement model, wherein the aforementioned correction unit corrects the aforementioned learning range in a manner that includes the aforementioned first feature when the aforementioned first feature deviates from the aforementioned correction range and deviates from a learning range that includes the aforementioned predetermined correction range or is wider than the aforementioned predetermined correction range. For example, in claim 5, a scanning electron microscope with an inspection measurement model, wherein the comparison between the aforementioned first characteristic and the aforementioned correction range is used to analyze the spatial distance of the characteristic. As in claim 7, a scanning electron microscope with an inspection measurement model, wherein, it has a display device, and on the screen of the aforementioned display device, there is at least an area for displaying alarms, an area for visually suggesting recalibration of the location within the aforementioned sample to be inspected or measured, and an area for selecting learning methods and models. A method for correcting the aforementioned inspection and measurement model of a scanning electron microscope having an inspection and measurement model includes the following steps: A feature quantity analysis unit inputs an electron microscope image or waveform of the sample to be inspected or measured and calculates a first feature quantity; simultaneously inputs an electron microscope image or waveform of a calibration sample and calculates a second feature quantity; A learning unit generates the aforementioned inspection and measurement model based on the aforementioned second feature quantity; and a correction unit corrects the aforementioned inspection and measurement model by ensuring that the output of the aforementioned inspection and measurement model is consistent with the aforementioned second feature quantity. The method for correcting the measurement model in claim 9 includes the following steps: the aforementioned correction unit issues an alarm when the aforementioned first feature quantity deviates from the predetermined correction range and deviates from the learning range that includes the aforementioned predetermined correction range or is wider than the aforementioned predetermined correction range. The method for correcting the inspection measurement model as described in claim 9 includes the following steps: the aforementioned correction unit corrects the aforementioned correction range in a manner that includes the aforementioned first feature when the aforementioned first feature is outside the aforementioned correction range, thereby correcting the aforementioned inspection measurement model. The method for correcting the measurement model in claim 10 includes the following steps: the correction unit corrects the learning range in a manner that includes the first feature when the first feature deviates from the correction range and deviates from the learning range that includes the predetermined correction range or is wider than the predetermined correction range. The method for correcting the measurement model in claim 11 includes the following steps: the correction unit corrects the learning range in a manner that includes the first feature when the first feature deviates from the correction range and deviates from the learning range that includes the predetermined correction range or is wider than the predetermined correction range.