Charged particle beam device and roughness indicator calculation method

Abstract

Roughness measurement with correction for machine tolerances is achieved. Therefore, using first PSD data (representing the power spectral density of a line pattern measured by a reference machine on a pre-formed line pattern on a machine tolerance management wafer, and second PSD data (representing the power spectral density of a line pattern measured by a calibration target machine on a line pattern formed on a machine tolerance management wafer), a correction method is derived to correct the power spectral density of the second PSD data to the power spectral density of the first PSD data. Based on a scanned image of the line pattern formed on the measurement target wafer, the power spectral density of the line pattern formed on the measurement target wafer is measured as third PSD data. The corrected power spectral density is then calculated after correcting the power spectral density of the third PSD data using the derived correction method.

Description

Technical Field

[0001] This disclosure relates to charged particle beam devices and methods for calculating roughness indices, and particularly to the correction of mechanical discrepancies between charged particle beam devices in the measurement of roughness indices appearing at the edges of a pattern that is the object of measurement. Background Technology

[0002] In semiconductor processes, particularly photolithography using extreme ultraviolet (EUV) light, the edge roughness (the unevenness at the pattern ends) of the patterns significantly impacts device yield as patterns become miniaturized. The degree of roughness varies considerably depending on the materials constituting the semiconductor device, the exposure equipment, or the properties and characteristics of the substrate. Especially in mass production, the magnitude of roughness has a significant impact on product performance, thus requiring precise measurement and management of roughness parameters during mass production.

[0003] On the other hand, in the measuring devices used in mass production processes, it is important to minimize the difference in measured values ​​(machine tolerance) between devices. Currently, scanning electron microscopes (SEMs) are mainly used for dimensional and edge roughness measurements in semiconductors. Here, the measuring device used for dimensional measurement of semiconductors is specifically referred to as an SEM-type length measuring device. With the increasing importance of edge roughness measurement in the mass production processes of semiconductor devices using EUV lithography, there is a requirement to reduce machine tolerance in edge roughness measurement for SEM-type length measuring devices used for edge roughness measurement.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2019-39884

[0007] Patent Document 2: Japanese Patent Application Publication No. 2012-151053 Summary of the Invention

[0008] The problem that the invention aims to solve

[0009] As a major cause of machine error in edge roughness measurement using SEM-type length measuring devices, noise mixed into the SEM image is a concern. Patent Document 1 discloses a method for removing noise from an SEM image by differentiating the power spectral density obtained from one edge location from the power spectral density obtained from another edge location for a single SEM image. However, the method described in Patent Document 1 cannot remove inherent device noise components, such as column vibration and power supply noise, that are mixed in equally at both edge locations.

[0010] On the other hand, Patent Document 2 describes a method in which the inherent noise frequency and amplitude of the device are determined from an SEM image obtained by deflecting the electron beam in only one direction, and the noise frequency and amplitude are fed back to a deflector used to scan the electron beam, thereby removing the inherent noise of the device. However, in the method described in Patent Document 2, even if the amplitude of the signal that should be fed back to the deflector is known, the phase is unknown, thus presenting a problem of needing to use trial and error in determining the optimal phase.

[0011] The following presents a charged particle beam device that can correct for mechanical differences between devices when using a charged particle beam device, such as a SEM-type length measuring device, to measure the roughness index of a pattern, as well as a method for calculating the roughness index.

[0012] Methods for solving problems

[0013] A charged particle beam apparatus according to one aspect of the present invention comprises: a charged particle beam optical system that scans a charged particle beam two-dimensionally over a line pattern formed on a sample; a detector that detects electrons emitted from the sample due to irradiation by the charged particle beam; an image processing unit that calculates a roughness index of the line pattern formed on the sample based on a scanned image obtained from a signal detected by the detector; and a power spectral density input unit that inputs first PSD data, which represents the power spectral density of the line pattern measured by a reference charged particle beam apparatus for a line pattern pre-formed on a first wafer, serving as a reference for tolerance management in the roughness index calculation.

[0014] The image processing unit performs the following processing: based on a scanned image of a line pattern formed on a first wafer, it measures the power spectral density of the line pattern formed on the first wafer as second PSD data, and determines a correction method to correct the power spectral density of the second PSD data to the power spectral density of the first PSD data; based on a scanned image of a line pattern formed on a second wafer, it measures the power spectral density of the line pattern formed on the second wafer as third PSD data, and calculates a corrected power spectral density that corrects the power spectral density of the third PSD data using the correction method; and uses the corrected power spectral density to calculate the roughness index of the line pattern formed on the second wafer.

[0015] Effects of the Invention

[0016] This invention enables roughness measurement that corrects for machine error.

[0017] Other issues and new features will become clear from the description and accompanying drawings in this specification. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the SEM-type length measuring device.

[0019] Figure 2 This is a schematic diagram of a scanned image obtained using a SEM-type length measuring device.

[0020] Figure 3 This is a diagram used to illustrate the method for determining the location of edges.

[0021] Figure 4A This is a flowchart of the machine error correction method for LER measurements.

[0022] Figure 4B This is a flowchart of the machine error correction method for LER measurements.

[0023] Figure 5A This is a schematic diagram of the PSD used in LER measurements.

[0024] Figure 5B This is a schematic diagram of the PSD used in LER measurements.

[0025] Figure 6A This is a flowchart of the machine error correction method for LER measurements.

[0026] Figure 6B This is a flowchart of the machine error correction method for LER measurements.

[0027] Figure 7 This is a diagram used to illustrate methods for improving the accuracy of correction functions.

[0028] Figure 8A This is a diagram used to illustrate a roughness correction method that utilizes machine learning metrics.

[0029] Figure 8B This is a diagram used to illustrate a roughness correction method that utilizes machine learning metrics.

[0030] Figure 9 This is a diagram representing an example of a GUI. Detailed Implementation

[0031] The embodiments of the present invention will be described below. Furthermore, the accompanying drawings shown in this embodiment illustrate specific embodiments following the principles of the present invention, but these are for understanding the invention and are not intended to limit its interpretation. In the following embodiments, an SEM-type length measuring device using electrons as a charged particle source is used as an example for description, but the same effect can be obtained when using various ions as charged particle sources.

[0032] Example 1

[0033] Figure 1The structure of the SEM-type length measuring device is shown. As its main components, the SEM-type length measuring device includes: an electron beam optical system for irradiating the sample with an electron beam; a detection system for detecting secondary electrons released from the sample due to electron beam irradiation; a stage mechanism system disposed within a vacuum chamber (not shown); a control system for controlling the constituent elements of the SEM-type length measuring device and processing various information; and an image processing system for measuring the dimensions and edge roughness of the pattern based on the obtained SEM image.

[0034] Specifically, primary electrons 102 generated by electron source 101 are deflected by deflector 104 and focused by objective lens 103, then irradiate the sample 105 mounted on movable stage 106. The movement of objective lens 103 is controlled by objective lens control unit 111, the movement of deflector 104 is controlled by deflector control unit 112, and the movement of movable stage 106 is controlled by stage control unit 107. Furthermore, a negative voltage can also be applied to sample 105 via movable stage 106.

[0035] Secondary electrons 108 generated by the electron beam optical system described above irradiating the sample 105 with primary electrons 102 are detected by the detector 109 constituting the detection system. In the example shown, the detector 109 is positioned closer to the electron source 101 than the deflector 104, but it can also be positioned between the deflector 104 and the objective lens 103, or between the objective lens 103 and the sample 105, as long as the secondary electrons 108 can be detected. Examples of detector 109 structures include an ET (Everhart-Thornley) detector composed of a scintillator, a light guide, or a photomultiplier tube, and a semiconductor detector; however, any detector can be used as long as it can detect electrons. Furthermore, the detector 109 can be mounted in multiple locations. The signal detected by the detector 109 is converted into a digital signal by the A / D converter 110. Signals for each electron beam coordinate are generated in the image processing unit 113, the scanned image is displayed in the display unit 114, and the scanned image is also recorded in the recording unit 115.

[0036] The operation of the worktable control unit 107, A / D converter 110, objective lens control unit 111, deflector control unit 112, image processing unit 113, display unit 114, recording unit 115, and PSD input unit 116 (described later) is controlled by the workstation 117.

[0037] Figure 2 Indicates by Figure 1 A schematic diagram of the scanned image obtained by the SEM-type length measuring device is shown. Figure 2 The variation in the edge position within a line pattern is called line edge roughness (LER). The measurement method for LER is explained. Figure 2In the middle, the y-coordinate n The line profile in the x-direction at that location is Figure 3 To determine the left edge of the line pattern, set the minimum semaphore on the left side of the line profile to 0%, the maximum semaphore to 100%, and the x-coordinate with a semaphore of 50% to the y-coordinate. n edge position x in n This determination method is called the threshold method. Alternatively, the x-coordinate where the semaphore is outside 50% can also be defined as the edge position x. n Alternatively, methods other than thresholding can be used, such as differentiating the line profile or determining the edge position by matching it to a pre-calculated waveform. This can be achieved by applying multiple y-coordinates... n The edge position x is obtained from the middle n Fourier analysis is performed to obtain the power spectral density (hereinafter referred to as "PSD"). Typically, the LER is expressed as the integral of the PSD with respect to its frequency. The above calculation of edge positions, PSD analysis, and LER calculation can be performed by the image processing unit 113, and the results are displayed on the display unit 114 and recorded on the recording unit 115.

[0038] In addition, the obtained PSD (or LER) will have deviations within the wafer plane. Therefore, for example, hundreds of measurement points are set on the wafer, and the PSD measured for the line pattern at each measurement point is averaged to obtain the PSD of the wafer.

[0039] The obtained PSD contains overlapping variations in the edge positions of the pattern itself (the object of observation) and variations in edge positions caused by the inherent noise of the device. Since the inherent noise of the device introduces machine error, it is necessary to remove the machine error component of the noise from the PSD to reduce machine error in LER measurements. Figure 4A , Figure 4B This describes the specific procedure for obtaining the LER value after correcting the machine error component of the noise from the PSD.

[0040] The process is divided into production Figure 4A The steps and execution of the correction function shown Figure 4B The correction steps are shown. First, for the fabrication... Figure 4A The flow of the correction function shown will be explained. First, a PSD ("PSD") of the pattern formed on the wafer for error management is obtained in a device that serves as the reference for error management (called the "reference machine"). Master (402). Here, PSD Master The PSD can be obtained from a single device, or it can be calculated as the average of PSD values ​​obtained from multiple devices for differential control. Master It is recorded in Records Section 115.

[0041] Next, in the device (called the "correction object machine") that becomes the object of the machine error correction for performing LER measurements, the PSD is calculated. Master The same wafer was used to obtain PSD' (403). Furthermore, the wafer used in the measurement of PSD' is not the same as the one used to obtain PSD. Master The chip is the same as the one used to obtain the PSD, as long as it is confirmed to have the same characteristics. Master If there are multiple LER chips with the same chip, then one of them can be used. Next, the PSD is read in by the calibration machine. Master (404). PSD Master Input can be made from the PSD input section 116. Next, as shown in (Equation 1), PSD' and PSD' are calculated. Master The difference is defined as the correction function PSD. Corr (405).

[0042] PSD Corr =PSD'-PSD Master …(Equation 1)

[0043] Finally, the correction function PSD Corr Recorded in Record Section 115 (406).

[0044] For execution Figure 4B The calibration process is explained below. In the calibration process, the PSD (Plate Size Displacement) of a pattern formed on an arbitrary wafer is measured by a device (called the "calibration target device") that is the object to which the LER measurement value is corrected. Obs (412). Use the correction function PSD recorded in the recording unit 115. Corr As shown in Equation 2, calculate PSD Obs With PSD Corr Differential PSD Obs '(413).

[0045] PSD Obs =PSD Obs -PSD Corr …(Equation 2)

[0046] Next, regarding the frequency of the PSD Obs Integrating, the computer then calculates the difference-corrected LER (LER). Corr (414). Calculated LER Corr The information is displayed on the display unit 114 and recorded in the recording unit 115.

[0047] Figure 5A It shows Figure 4A PSD in Corr, PSD and PSD Master Relationship, Figure 5B It shows Figure 4B PSD in Obs '、PSD Obs and PSD Corr The relationship between PSD and PSD Master They are obtained from the same chip (or a chip with the same LER size), and the PSDs of the patterns themselves are identical. Therefore, as PSD' and PSD Master The difference PSD Corr This represents the noise difference between the reference machine and the calibration target machine. Therefore, it is obtained from the PSD obtained in the calibration target machine with an arbitrary pattern. Obs Subtract PSD Corr A PSD with noise removed and differential components were obtained. Obs '.

[0048] Example 2

[0049] The PSD obtained by the SEM-type length measuring device contains both the roughness component inherent in the pattern itself and the noise component. Furthermore, the noise component in the PSD includes a noise component (random noise component) with a certain intensity at any frequency, and methods for removing this random noise component from the PSD are well known (e.g., Patent Document 1). In Embodiment 2, a correction function PSD for machine error correction is disclosed based on the PSD after random noise removal. Corr The method.

[0050] Figure 6A , Figure 6B This describes the flow of this embodiment. The basic flow and... Figure 4A , Figure 4B The same. The PSD (Plate Deposit Surface Mount) is used to obtain the pattern formed on the wafer in the reference machine for differential management. Master (602) After that, according to PSD Master Calculate the PSD after removing random noise components. Master (603). Similarly, the PSD is also used in the calibration object machine to obtain it. Master After measuring PSD' (604) with the same chip, the PSD (PSD”) (605) after removing the random noise component from PSD' is obtained. Then, PSD is read into the calibration target machine. Master '(606), based on PSD Master The correction function PSD, defined by Equation 3, is obtained from 'and PSD'. Corr (607)

[0051] PSD Corr =PSD"-PSDMaster '…(Formula 3)'

[0052] Finally, the correction function PSD Corr Recorded in Record Section 115 (608).

[0053] For execution Figure 6B The calibration process is explained below. The PSD (PSD) of any chip is obtained from the calibration target machine. Obs After (611), calculate the PSD after removing the random noise component. Obs (612). Then, calculate the value from PSD. Obs "Subtract PSD" Corr The PSD after (PSD) Obs "')(613), by integrating the frequency PSD Obs ', calculate the LER after machine error correction (LER) Corr (614)

[0054] Thus, in Example 2, a PSD with the random noise removed can be obtained with the random noise component corrected.

[0055] The choice between using the method of Example 1 or the method of Example 2 depends on the application of the mass production process. When process management is performed using the LER calculated from the PSD after removing frequency-independent random noise, the method of Example 2 is preferred; when process management is performed using the LER calculated from the PSD without removing random noise, the method of Example 1 is preferred. This allows for mass production process management by reducing machine error management values ​​while maintaining the continuity of management values.

[0056] The following describes a variation of the correction method for the machine error component described in Example 1 or Example 2. First, a correction function PSD is obtained with higher precision. Corr The method. The correction function PSD obtained in Examples 1 and 2. Corr Depending on the number of edges used for PSD parsing, a limited amount of noise overlaps due to measurement bias. This noise affects the accuracy of PSD error correction. Therefore, as... Figure 7 As shown, the correction function PSD calculated by (Equation 1) or (Equation 3) is... Corr Implementing smoothing processing allows for the precise determination of the correction function PSD. Corr As a smoothing method, moving averages or any function can be used for approximation.

[0057] Furthermore, in Embodiments 1 and 2, the correction function PSD is defined by difference as in Equations (1) and (3).Corr However, the correction function PSD can also be defined using other methods. Corr For example, (Equation 4-1) in Example 1 uses PSD' and PSD Master The ratio defines the correction function PSD. Corr .

[0058] PSD Corr =PSD Master / PSD'…(Formula 4-1)

[0059] In this case, the PSD after machine error correction for any wafer Obs It can be calculated using (Equation 5-1).

[0060] PSD Obs =PSD Corr ×PSD Obs …(Equation 5-1)

[0061] In the case of Example 2, calculations were performed using (Equation 4-2) and (Equation 5-2) respectively.

[0062] PSD Corr =PSD Master ' / PSD'…(Formula 4-2)

[0063] PSD Obs "' = PSD Corr ×PSD Obs "...(Equation 5-2)"

[0064] Furthermore, the error correction method is not limited to correction methods using functions as described above; it can also be a machine learning-based correction method. Figure 8A , Figure 8B This describes how machine learning is used to calculate the PSD after correction for machine error in any chip. Obs '、PSD Obs The method of "". Figure 8A Indicates the steps of learning. Figure 8B This indicates the step of error correction using a learned model. Supervised learning is a type of machine learning, such as... Figure 8A As shown, the PSD' or PSD" of the pattern formed on the wafer used in the machine tolerance management, obtained in the calibration target machine, is used as input for learning so that the PSD of the pattern of the same wafer obtained in the reference machine is... Master or PSD Master'This becomes the output.' Here, as an example of a learning algorithm, Deep Neural Networks, Convolutional Neural Networks, Generative Adversarial Networks, etc., can be used. Furthermore, any algorithm that can infer PSD from 'PSD' or 'PSD' can be used. Master or PSD Master The algorithm can be applied. In the step of machine error correction, such as... Figure 8B As shown, the PSD obtained from an arbitrary chip in the calibration object machine is input into the learned model. Obs or PSD Obs "PSD after output differential correction" Obs 'or PSD Obs ”'.

[0065] The error correction method in LER measurement described above is preferably set according to the optical conditions used to obtain the SEM image in the LER measurement, specifically the irradiation energy of the primary electron 102 in the sample 105, the current of the primary electron 102, the type of detector 109 used to obtain the SEM image, and the scanning speed of the primary electron 102 scanning the sample 105. This is because if these optical conditions change, the amount of noise superimposed on the SEM image will change.

[0066] Figure 9 This illustrates an example of a GUI displayed on the display unit 114 via the PSD input unit 116. The upper part of the GUI screen has a display area 901 for a SEM image used for measurement, where the wafer ID 903 (representing the wafer to be measured), the measurement coordinates 904 (representing the measurement location), the irradiation energy 905 (optical condition), the probe current 906, the magnification 907, the scanning method 908, the detector type 909, and the measurement item 910 can be specified. In the example shown, the area corresponding to the specified measurement coordinates is displayed on the SEM image (902). Here, when LER is selected as the measurement item 910, the machine error correction checkbox 911 is activated. When the machine error correction checkbox 911 is selected, PSD can be input. Master or PSD Master In this example, the PSD corresponding to the optical conditions is specified in box 912. Master or PSD Master 'Data'.

[0067] In this example, it is shown that the operator specifies the PSD via the GUI displayed on the display unit 114 through the PSD input unit 116. Master or PSD MasterExamples of data can be obtained, but a PSD corresponding to predetermined optical conditions can also be transmitted via network between a reference machine and a calibration target machine. Master or PSD Master The data is input into the calibration target machine.

[0068] Furthermore, the invention is not limited to the LER measurement described so far; it can also be applied to the measurement of other roughness parameters of line patterns, specifically, to the measurement of line width roughness (LWR). In this case, Figure 2 In the middle, measuring the y-coordinate y n Line Width (CD) n This will be for multiple y n The calculated line width CD n The result of Fourier analysis is set as PSD. If the method of the above-described embodiments or variations is applied, the LWR after machine error correction can be obtained.

[0069] Symbol Explanation

[0070] 101: Electronic source; 102: Primary electron; 103: Objective lens; 104: Deflector; 105: Sample; 106: Movable stage; 107: Stage control unit; 108: Secondary electron; 109: Detector; 110: A / D converter; 111: Objective lens control unit; 112: Deflector control unit; 113: Image processing unit; 114: Display unit; 115: Recording unit; 116: PSD input unit; 117: Workstation.

Claims

1. A charged particle beam device, characterized in that, have: A charged particle beam optical system that scans a charged particle beam in two dimensions onto a line pattern formed on a sample; A detector that detects electrons released from the sample due to irradiation by the charged particle beam; The image processing unit calculates the roughness index of the line pattern formed on the sample based on the scanned image obtained from the signal detected by the detector; and The power spectral density input unit receives first PSD data, which represents the power spectral density of a line pattern pre-formed on a first wafer, measured by a reference charged particle beam device as a reference for error management in the roughness index calculation. The image processing unit performs the following processing: Based on the scanned image of the line pattern formed on the first wafer, the power spectral density of the line pattern formed on the first wafer is measured as the second PSD data, and a correction method is derived to correct the power spectral density of the second PSD data to the power spectral density of the first PSD data. Based on a scanned image of a line pattern formed on a second wafer, the power spectral density of the line pattern formed on the second wafer is measured as third PSD data. A corrected power spectral density, obtained by correcting the power spectral density of the third PSD data using the aforementioned correction method, is then calculated. The corrected power spectral density is used to calculate the roughness index of the line pattern formed on the second wafer.

2. The charged particle beam device according to claim 1, characterized in that, The first wafer is a wafer with a line pattern formed for tolerance management, and the second wafer is a wafer with a line pattern formed for the object being measured. The first wafer is one of a plurality of wafers, and the power spectral density of the line patterns formed on the plurality of wafers is the same.

3. The charged particle beam device according to claim 1, characterized in that, The first PSD data and the second PSD data are average values ​​of the power spectral density of the line patterns measured on the first wafer.

4. The charged particle beam device according to claim 1, characterized in that, The first PSD data is the average power spectral density of the line pattern measured by the plurality of reference charged particle beam devices on the line pattern formed on the first wafer.

5. The charged particle beam device according to claim 1, characterized in that, The edge position of the line pattern formed on the wafer is measured, and the line edge roughness is calculated as a roughness index of the line pattern formed on the wafer, or... The linewidth of the line pattern formed on the wafer is measured, and the linewidth roughness is calculated as a roughness index of the line pattern formed on the wafer.

6. The charged particle beam device according to claim 1, characterized in that, As a correction method, the difference between the power spectral density of the second PSD data and the power spectral density of the first PSD data is calculated, and the power spectral density of the third PSD data is corrected based on the difference.

7. The charged particle beam device according to claim 6, characterized in that, The difference is smoothed, and the power spectral density of the third PSD data is corrected based on the smoothed difference.

8. The charged particle beam device according to claim 1, characterized in that, As a correction method, the ratio of the power spectral density of the second PSD data to the power spectral density of the first PSD data is calculated, and the power spectral density of the third PSD data is corrected based on the ratio.

9. The charged particle beam device according to claim 8, characterized in that, The ratio is smoothed, and the power spectral density of the third PSD data is corrected based on the smoothed ratio.

10. The charged particle beam device according to claim 1, characterized in that, As the correction method, a model is learned that takes the power spectral density of the second PSD data as input and the power spectral density of the first PSD data as output, and the power spectral density of the third PSD data is corrected by inputting the power spectral density of the third PSD data into the learned model.

11. The charged particle beam device according to claim 1, characterized in that, The power spectral density of the first to third PSD data is the power spectral density after removing frequency-independent random noise components.

12. The charged particle beam device according to claim 1, characterized in that, The power spectral density input unit inputs the first PSD data according to optical conditions.

13. The charged particle beam device according to claim 12, characterized in that, The optical conditions include at least one of the following: the irradiation energy of the charged particle beam in the sample, the current of the charged particle beam, the type of detector, and the scanning speed at which the charged particle beam scans the sample.

14. A method for calculating a roughness index, wherein a charged particle beam device uses first PSD data and second PSD data to calculate the roughness index of a line pattern formed on a second wafer, wherein, The first PSD data represents the power spectral density of a line pattern pre-formed on a first wafer, measured by a reference charged particle beam device as a benchmark for tolerance management in roughness index calculation; the second PSD data represents the power spectral density of a line pattern formed on the first wafer, measured by the charged particle beam device as the object of tolerance management; the roughness index calculation method comprises the following steps: Find a correction method to correct the power spectral density of the second PSD data to the power spectral density of the first PSD data; Based on a scanned image of a line pattern formed on the second wafer, the power spectral density of the line pattern formed on the second wafer is measured as third PSD data. A corrected power spectral density is then calculated after correcting the power spectral density of the third PSD data using the correction method. The corrected power spectral density is used to calculate the roughness index of the line pattern formed on the second wafer.

15. The roughness index calculation method according to claim 14, characterized in that, The power spectral density of the first to third PSD data is the power spectral density after removing frequency-independent random noise components.

Images