JP2013019866A - Scanning electron microscope, defect inspection system, and defect inspection apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the ratio of incurring a failure that a defect is not captured in an image when image pickup is attempted in a scan electron microscope to a portion of the defect which occurs on a sample such as a wafer, a display panel, or a magnetic disk.SOLUTION: Defect inspection data having coordinates and a feature amount of a defect are inputted (241), the possibility of detection for a scanning electron microscope is predicted for each defect (242), and on the basis of a prediction result, coordinates to be performed imaging with the scanning electron microscope are sampled (243). An image of a defect portion is then captured by an electronic optical system or an optical system for which an acceleration voltage is set on the basis of the prediction result (244).

Description

  The present invention relates to a scanning electron microscope for observing defects generated on a sample and attached foreign matter, a defect inspection device for detecting coordinates of defects and foreign matter generated on a sample, and a scanning electron microscope and a defect inspection device connected by a network. It relates to a defect inspection system.

  In the manufacturing process of products manufactured using thin film processes such as exposure, development, etching, and film formation, such as integrated circuits, thin film magnetic heads, magnetic disks, solar cells, and liquid crystal displays, the following steps are performed for the purpose of quality control. Defect inspection is performed. (1) A sample (for example, a wafer) to be inspected is inspected by a defect inspection apparatus that makes full use of optical technology and image processing technology, and the coordinates of the generated defect and attached foreign matter on the sample are detected. Hereinafter, defects and foreign substances are collectively referred to as defects. (2) A high-magnification image is taken with a scanning electron microscope with respect to the coordinates of the detected defect, and the type of defect is classified. This is called review work. (3) The classification result is compared with the result of the electrical characteristic test, the cause of the occurrence of the defect is analyzed, and measures are taken.

  When the number of defects detected by the defect inspection apparatus is large, the review work described in (2) above requires a great deal of labor. Therefore, an automatic review (Automatic Review) that automatically captures and collects images of defective parts. Scanning electron microscopes having functions of Defect Review and Automatic Defect Classification for automatically classifying collected images are on the market.

  However, the scanning electron microscope cannot always detect defects detected by the defect inspection apparatus. Defect inspection devices can be broadly classified into bright-field inspection devices that irradiate illumination from directly above and detect the reflected light, and dark-field inspection devices that irradiate illumination obliquely from above and detect scattered light that hits a defect. . Since both devices use a light source such as an Ar laser, a He—Cd laser, or an excimer laser, if the sample is formed of a film that passes the wavelength of the light source, defects in the lower layer are also detected. However, in the electron optical system used by the scanning electron microscope, it is difficult to detect a defect in the lower layer, and it cannot be detected under an acceleration voltage condition of about 1 kV. Patent Document 1 discloses a method of detecting by a try-and-error while switching the detection conditions in order to detect the above-described lower layer defects with a scanning electron microscope.

JP 2005-259396 A JP 2006-145269 A JP 2004-294360 A

Toshiro Tango, et al. “Logistic regression analysis” Asakura Shoten 1996

  The method disclosed in Patent Document 1 first attempts to detect the coordinates of a defect detected by a defect inspection apparatus using normal detection conditions, for example, an acceleration voltage of about 1 kV. If detection is not possible under normal detection conditions, the acceleration voltage is increased and detection is attempted again. However, the conventional method for detecting defects by trial and error has a problem that the time required for the review work becomes long. In addition, detection of a defect by a scanning electron microscope is not a complete nondestructive inspection, but causes some damage to the sample. Therefore, it is not preferable to irradiate the same portion with an electron beam many times.

In order to solve the above problems, for example, the configuration described in the claims is adopted.
The present application includes a plurality of means for solving the above-described problems. To give an example, an inspection image for analyzing the defect is input by inputting information on the defect coordinates detected by the defect inspection apparatus. In a scanning electron microscope having a function of imaging, a function of inputting a feature amount of a defect detection image together with a plurality of defect coordinates detected by the defect inspection apparatus, and an acceleration voltage in the scanning electron microscope using the feature amount A function for predicting for each defect whether or not a defect can be detected in a set condition, an observation image, a function for selecting a plurality of coordinates of an alignment and observation target from a coordinate group input based on a prediction result, and a plurality of selected For each of the coordinates, an acceleration voltage predicted to be detectable is set and an image is captured.

By applying the present invention to a manufacturing process of an integrated circuit, a thin film magnetic head, a solar cell, a magnetic disk, a liquid crystal display, etc., when attempting to take an image of a defective part with a scanning electron microscope, the image shows a defect. Reduce the rate at which failures such as none occur.

It is a figure which shows an example of the outline | summary of a scanning electron microscope. It is a figure which shows an example of the defect inspection system which has a scanning electron microscope and a defect inspection apparatus. It is a figure which shows an example of the outline | summary of a defect inspection apparatus. It is a figure which shows an example of the defect inspection data which has a coordinate and a feature-value. It is a figure which shows an example of the wafer map which showed the coordinate visually. It is a figure explaining an example of the extraction method of the feature-value. It is a figure which shows an example of the result of having predicted the possibility of detection. It is a figure which shows an example of the coefficient for estimating a detection possibility. It is a figure which shows an example of a sampling result. It is a figure which shows an example of a classification result. It is a figure which shows an example of the flowchart which images an image. It is a figure which shows an example of the graphical user interface of a scanning electron microscope. It is a figure which shows an example of the graphical user interface of a scanning electron microscope. It is a figure which shows an example of the graphical user interface of a scanning electron microscope. It is a figure which shows an example of the data which learns whether detection is possible. It is a figure which shows an example of the graphical user interface of a scanning electron microscope. It is a figure which shows the example of the data record format of a defect inspection and analysis data table. It is a figure which shows an example of the flowchart of a process of a detection possibility prediction program. It is a figure which shows an example of the flowchart of a process of a detection possibility learning program.

  FIG. 1 is a diagram showing an outline of an example of a scanning electron microscope including an optical microscope according to the present invention. The scanning electron microscope 600 includes an electron source 201 that generates primary electrons 208, an acceleration electrode 202 for accelerating the primary electrons, a focusing lens 203 for converging the primary electrons, and a deflector that deflects the primary electrons in two dimensions. 204 and an objective lens 205 for converging primary electrons on the sample 206. Reference numeral 207 denotes a drive stage on which the sample 206 is mounted. Reference numeral 210 denotes a detector that detects the secondary electron signal 209 generated from the sample 206, and reference numerals 212a and 212b denote reflected electron detectors that detect the reflected electron signal 213, respectively. In the figure, two backscattered electron detectors 212a and 212b are installed facing each other, and each detects a different component of the backscattered electron signal emitted from the sample 206. Reference numeral 211 denotes a digital conversion unit for digitizing the detected signal. Further, the scanning electron microscope 600 has an optical microscope in order to capture an image of a defective portion that cannot be detected by the above-described electron optical system. 221a and 221b are light sources that output light of different wavelengths. For example, an Ar laser, a He-Cd laser, or an excimer laser is used. Light emitted from the light sources 221a and 221b is applied to the sample 206 via the reflection mirrors 222a, 222b, and 222c. Light reflected by the sample 206 or scattered by hitting a defect is detected by the photoelectric conversion units 225a and 225b via the condensing lens 223 and the beam splitter 224 that separates each wavelength. For the photoelectric conversion units 225a and 225b, a photomultiplier, a CCD sensor, a TDI sensor, or the like is used. Signals output from the photoelectric conversion units 225a and 225b are digitized by a digital conversion unit 226. Each part described above is connected to the overall control unit 231 via the bus 230. In addition, the scanning electron microscope 600 includes a central processing unit (CPU) 232, a primary storage device 233, a network interface 234, a keyboard 235, a mouse 236, a display 237, and a secondary storage device 238. The secondary storage device 238 also has a defect inspection data input program 241 for inputting the inspection result of the defect inspection device 601 to the scanning electron microscope via the network interface 234, and for each input coordinate, Detectability prediction program 242 for predicting whether or not detection is possible with scanning electron microscope 600, sampling program 243 for selecting coordinates based on the prediction result of detection availability, and operating the drive stage so that the selected coordinates are reflected in the image An image capturing program 244 for capturing an image, an alignment program 245 for calculating the offset of the coordinate system of the sample with respect to the drive stage so that the defect is surely included in the field of view of the image, and defect classification for classifying the type of defect using the captured image Prediction of program 246 and detectability prediction program 242 Detectability learning program 247 executed to improve accuracy, detection result prediction result, image capturing result, defect classification result, graphical user interface (GUI) program 248 for executing various programs, and defect Defect inspection / analysis data table 500 for recording defect inspection data input from the inspection apparatus and results analyzed based on the data, and prediction equation coefficient data table 520 for recording prediction equation coefficient data used by the detectability prediction program. Etc. are stored. These programs are read from the secondary storage device 238 to the primary storage device 233 and executed by the CPU 232.

  FIG. 2 is an example of a minimum hardware configuration of a defect inspection system having a scanning electron microscope and a defect inspection apparatus according to the present invention. The scanning electron microscope 600 and the defect inspection apparatus 601 are each connected to a network 602, and data is transmitted and received through a network interface included in each apparatus. Of course, a database system or a file server may be further connected to the network 602 to relay data transmission / reception between the scanning electron microscope 600 and the defect inspection apparatus 601.

  FIG. 3 is a diagram showing an outline of an example of the defect inspection apparatus according to the present invention. Similar to the optical microscope included in the scanning electron microscope 600, the defect inspection apparatus 601 includes a light source that outputs different wavelengths 701a and 702b, reflection mirrors 702a, 702b, and 702c, a condensing lens 703, a beam splitter 704 that separates the wavelengths. Photoelectric conversion units 705a and 705b and a drive stage 707 are provided, and signals output from the photoelectric conversion units 705a and 705b are digitized by a digital conversion unit 706. Each of these parts is connected to the overall control unit 721 via a bus 720. In addition, the defect inspection apparatus 601 includes a CPU 722, a primary storage device 723, a network interface 725, a keyboard 726, a mouse 727, a display 728, and a secondary storage device 724. Further, in the secondary storage device 724, a defect coordinate detection program 741 for detecting the coordinates of the defect generated in the sample 206, an image captured at the time of detecting the defect coordinates, the maximum value of luminance, the sum of luminance, and the luminance are saturated. A feature quantity extraction program 742 that extracts feature quantities such as the number of pixels, luminance variance, and volume of a defective part, a graphical user interface (GUI) program 744 that checks the result of defect inspection and executes various programs, etc. Is stored. Further, the same program as the detection possibility prediction program 242 and the detection possibility learning program 247 of the scanning electron microscope 600 may be stored in the secondary storage device 724 of the defect inspection apparatus 601. In that case, it is also necessary to store a detection availability input program 743 for inputting past detection availability data in the scanning electron microscope 600. These programs are read from the secondary storage device 724 to the primary storage device 723 and calculated by the CPU 722.

  FIG. 4 is a diagram illustrating an example of data output as an inspection result by the defect inspection apparatus. The defect inspection data 501 has information such as a serial number 401, an x coordinate 402, a y coordinate 403, feature amounts 404, 405, 406, 407, and 408 for each defect detected by the defect inspection apparatus 601. The serial number 401 means the order detected by the defect inspection apparatus 601 by scanning the sample. The x coordinate 402 and the y coordinate 403 mean the position of the defect on the sample. The feature amounts 404 to 408 are feature amounts extracted from an image captured when the defect inspection apparatus 601 scans the sample to detect a defect. For example, the maximum amount of luminance of an image in which a defect site is shown, It has information such as the sum of brightness, the number of pixels with saturated brightness, the variance value of brightness, and the volume of the defective part. The defect inspection data 501 shows an example having five types of feature quantities named F01, F02, F03, F04, and F05. F01 is the maximum luminance value. F02 is the sum of luminance. F03 is the number of pixels with saturated luminance. F04 is a value obtained by multiplying the luminance dispersion value by ten. F05 is the volume of the defective part. This defect inspection data is taken into the defect inspection / analysis data table 500 on the secondary storage device 238 via the network interface 234 of the scanning electron microscope 600 by executing the defect inspection data input program 241.

FIG. 5 is a diagram illustrating an example of a coordinate system definition for a sample. Reference numeral 421 denotes the outer periphery of a wafer which is a sample. Reference numeral 422 denotes a notch for defining the direction of the sample. In this example, a coordinate system is shown in which the notch is positioned on the lower side and the lower leftmost part of the effective area scanned by the defect inspection apparatus is the origin (0, 0). Reference numeral 423 denotes an X axis, and 424 denotes a Y axis. A large number of black dots 425 are examples in which the coordinates of the defect included in the defect inspection data 501 are drawn according to this coordinate system.
FIG. 6 is a diagram for explaining an example of a method for extracting defect feature values. In the defect inspection apparatus 601, an image is obtained from the digital conversion unit 701. The obtained image is two-dimensional matrix data, and each pixel has luminance. FIGS. 4A and 4B are examples of luminance distributions when different defects are reflected in an image. (A) has a peak 801 in the luminance distribution. In (b), a saturated state 802 is seen in the luminance distribution. Since the upper limit and the lower limit of the luminance of the image obtained from the digital conversion unit 706 are determined, when the reflected light or scattered light obtained from the sample is strong as shown in (b), that is, when the defect is large, the saturation state 802 occurs. The feature amount extraction program 742 extracts defect feature amounts from an image having such a luminance distribution. The feature amount F01 is the luminance at the peak 801 in (a) and the luminance in the saturation state 802 in (b). The feature amount F02 is a value obtained by calculating the sum of the luminance of the image regardless of the peak or saturation state in both (a) and (b). The feature amount F03 is zero because there is no saturation in (a), and (b) is the saturation portion, that is, the number of pixels in the saturation portion because there is a saturation state. The feature amount F04 is a value obtained by adding 10 to the value obtained by calculating the variance of the luminance of the image regardless of the peak or saturation state in both (a) and (b). The feature amount F05 is an integrated value of the luminance distribution in (a), but is an estimated value in consideration of the saturation state in (b). Saturation occurs when the reflected or scattered light obtained from the sample is strong. Therefore, when the saturated state 802 is seen, the defect volume is larger than the integrated value of the luminance distribution. Therefore, a Gaussian or quadratic surface is approximated to the luminance distribution excluding the saturated portion, and the integral value of the approximated curved surface is set as the value of the feature amount F05. In this example, the feature amounts F01, F02, F03, F04, and F05 included in the defect inspection data 501 have been described. In addition to this, various feature amounts such as the peripheral length of the defective portion and the aspect ratio of the defective portion are extracted. Also good.

  FIG. 7 is a diagram illustrating an example of a result of executing the detectability prediction program 242. As illustrated in FIG. The detection feasibility prediction data 502 is data obtained by adding values 409 and 411 that predict the probability that each defect cannot be detected by the scanning electron microscope to the defect inspection data 501 and detection prediction results 410 and 412. However, the feature amounts 404 to 408 are omitted here due to space limitations. The detectability prediction data 502 is recorded in the defect inspection / analysis data table 500 shown in FIG.

  Reference numerals 409 and 411 indicate predicted probabilities that each defect cannot be detected by the scanning electron microscope. 409 is a probability that each defect cannot be detected when the acceleration voltage is set to 1 kV, and 411 is a probability that each defect cannot be detected when the acceleration voltage is set to 30 kV.

The prediction process will be described using a logistic regression model in this embodiment.
In order to predict the probability P that a certain defect cannot be detected with a scanning electron microscope, a group of defect features x = (x ₁ , x ₂ ,..., X _N ) included in the defect inspection data 501 is employed. A conditional probability P at which a phenomenon in which a defect cannot be detected occurs under the condition of a defect feature group x = (x ₁ , x ₂ ,..., X _N ),

Modeling is performed using the function F. here,
(1) A linear composite variable that influences the N explanatory variables

(2) Logistic function of Z in function F

(1) and (2) are the models of the following logistic regression model formula.
(3) Logistic regression model formula

The logistic regression model is a model in which a synthetic variable Z of independent variables having a variation range of (−∞ to ∞) and a probability P having a value in the range (0, 1) are linked by a logistic function.
If this equation is transformed,

  It becomes.

409 and 411 in FIG. 7 are results calculated based on the equation (4). In this embodiment, N = 4, and the coefficients registered in the prediction formula coefficient data table 520 shown in FIG. 8 are substituted for the coefficients a ₁ , a ₂ , a ₃ , a ₄ , and b (calculation of coefficients). The method will be described later). 522 is a coefficient when the acceleration voltage is set to 1 kV, and 523 is a coefficient when the acceleration voltage is set to 30 kV. In this example, the feature value F01 is not used as an explanatory variable because it is not necessary for predicting whether or not it can be detected, and feature values F02, F03, F04, and F05 are used.

  Reference numerals 410 and 412 are prediction results of detection availability. 410 is a prediction result when the acceleration voltage is set to 1 kV, and 412 is a prediction result when the acceleration voltage is set to 30 kV. 410 is the result of predicting “invisible” when 409 is 50% or more, and “visible” when 409 is less than 50%. Similarly, 412 is a result of predicting “invisible” when 411 is 50% or more, and “visible” when 411 is less than 50%. In this example, as described above, the probability is calculated using the logistic regression equation, and “invisible” and “visible” are determined using 50% as a threshold value, but this is not restrictive. For example, a pattern recognition algorithm such as a neural network may be applied instead of the logistic regression equation. Also, instead of determining “invisible” and “visible” with 50% as a threshold value, the threshold value may be optimized each time. In this example, the probability that cannot be detected by the scanning electron microscope is used for 409 and 411, but conversely, the probability that can be detected by the scanning electron microscope may be used. The probability that can be detected with a scanning electron microscope can be calculated by reversing the positive and negative signs in square brackets after the exp in Expression 4.

  FIG. 9 is an example of a result of executing the sampling program 243. The sampling data 503 is data obtained by adding the sampling result 413 to the detectability prediction data 502, and is stored in the defect inspection / analysis data table 500. However, here, due to space limitations, the feature amounts 404 to 408 and the probabilities 409 and 411 that cannot be detected by the scanning electron microscope are omitted. Reference numeral 413 denotes data in which a defect is randomly selected from the coordinates of a large number of defects included in the detection possibility prediction data 502 and having a probability 409, 411 that cannot be detected by the scanning electron microscope having a predetermined value. The blank means that the defect was not selected for the review work, and conversely, a defect written in some way means that the defect was selected for the review work. “1 kV Review” represents that an image is taken with a scanning electron microscope with an acceleration voltage set to 1 kV.

  “1 kV Alignment” represents that an acceleration voltage is set to 1 kV with a scanning electron microscope and an image is captured, and that the captured image is used for alignment of a coordinate system by the method disclosed in Patent Document 2. . The coordinate system alignment is the alignment between the stage coordinate system of the defect inspection apparatus and the scanning electron microscope and the semiconductor wafer coordinate system. The stage coordinate system is an apparatus-specific coordinate system determined according to the axis of movement of the apparatus stage. The semiconductor wafer coordinate system is a coordinate system determined for each semiconductor wafer. In the case of a semiconductor wafer on which a pattern is formed, it is generally determined along a pattern die. In the case of a semiconductor wafer having no pattern, it is determined from the positional relationship between the outer shape of the semiconductor wafer and the V notch or orientation flat. If the alignment between the stage coordinate system of the defect inspection apparatus and the scanning electron microscope and the semiconductor wafer coordinate system is performed at the same position on the semiconductor wafer, the coordinate system in the inspection by the defect inspection apparatus and the review by the scanning electron microscope Match. However, actually, the defect inspection data output from the defect inspection apparatus includes a semiconductor wafer alignment error at the time of inspection and a defect detection position error. For this reason, even if semiconductor wafer alignment is performed in a scanning electron microscope, a desired defect may not always enter the field of view. Therefore, when correcting the error of the defect coordinates, first, the center of the observation field of the scanning electron microscope is moved to the position of the defect coordinates output from the defect inspection apparatus, and the defect existing in the vicinity of the region of the observation field And the position of the detected defect is designated. Such an operation is repeated once or a plurality of times. Then, coordinate conversion is performed so that an error between the designated actual defect position coordinate and the defect coordinate output from the defect inspection apparatus is minimized.

  “30 kV Review” of the sampling result 413 in FIG. 9 represents that an image is captured with the acceleration voltage set to 30 kV with a scanning electron microscope. This indicates that the defect with the serial number of 15 has the prediction result of 410 being “invisible”, but the prediction result of 412 is “visible”, so that the acceleration voltage is set to 30 kV. “Optical Review” represents that an image is picked up with an optical microscope because the scanning electron microscope cannot detect even if the acceleration voltage is set to 30 kV. Here, as the optical microscope, an optical system included in the scanning electron microscope 601 may be used, or any device having an optical system different from the scanning electron microscope 601 may be used.

  FIG. 10 is an example of a result of executing the image capturing program 244 and the defect classification program 246. The classification result data 504 is data obtained by adding the defect classification result 414 and the storage destination 415 of the captured image to the sampling data 503, and is stored in the defect inspection / analysis data table 500. However, here, due to space limitations, the feature amounts 404 to 408 and the probabilities 409 and 411 that cannot be detected by the scanning electron microscope are omitted. The image imaging program 244 images a defect site with a scanning electron microscope or an optical microscope in accordance with the contents described in 413. For the defects marked “1 kV Review” or “1 kV Alignment” in 413, an image is taken with the acceleration voltage set to 1 kV by the electron optical system 201-213, and the image is saved in the storage destination designated by 415. To do. For the defect described as “30 kV Review”, the electron optical system 201 to 213 sets the acceleration voltage to 30 kV, captures an image, and stores the image in the storage destination specified by 415. For the defect described as “Optical Review”, an image is captured by the optical systems 221 to 226 and the image is stored in a storage destination designated by 415. In this example, the storage destination 415 is numbered based on the serial number of the defect 401. Next, the defect classification program 246 classifies the defect for each type of the captured image. Specifically, as disclosed in Patent Document 3, it is known that classification performance is high when defects are classified by combining a rule-based defect classification engine and a learning-type defect classification engine in multiple stages. .

  FIG. 11 is a diagram illustrating an example of a flowchart for explaining the processing procedure of the review work performed in the scanning electron microscope in which the function of the present invention is introduced. In step 321, the defect inspection data input program 241 is used to read defect inspection data 501 having coordinates and feature amounts from the defect inspection apparatus 601 into the defect inspection / analysis data table 500.

  In step 322, the detection possibility with the scanning electron microscope is predicted for each coordinate and each acceleration voltage using the detection possibility prediction program 242. That is, the defect inspection data 501 is converted into the detection possibility prediction data 502. Processing by the detection possibility prediction program 242 will be described with reference to the flowchart shown in FIG.

  The defect coordinates to be processed first are specified from the defect inspection data 501 and S1011, the defect feature data extracted by the defect inspection apparatus from the image picked up at the specified defect coordinates is passed through the user interface shown in FIG. The data of the feature amount selected by the user is input from the defect inspection / analysis data table 500 (S1012).

For each detection condition (acceleration voltage is 1 kV, 30 kV, etc.), the probability P that the corresponding defect cannot be detected (or the probability 1-P that can be detected) is expressed using the above-described logistic function (Equation 4) and Equation 2. To calculate. Here, the coefficients a ₁ , a ₂ ,..., A _N , b in Equation 2 are corresponding values calculated in advance (described later) by the detectability learning program 247 and stored in the prediction formula coefficient data table 520. And the feature quantity of the defect input in S1012 is substituted into the explanatory variables (x ₁ , x ₂ ,..., X _N ), and the probability P that the identified defect cannot be detected for each detection condition is calculated. Then, S1013 is recorded in the defect inspection / analysis data table 500.

  Subsequently, the probability P that cannot be detected in S1013 can be detected by using a predetermined or optimized threshold value, and a prediction result of detection is obtained and recorded in the defect inspection / analysis data table 500. S1014.

  If there is a next calculation target defect coordinate (S1015), the next calculation target defect coordinate is specified (S1016), and the processing from S1012 is repeated, and if there is no next calculation target defect coordinate. The process of the detectability prediction program 242 is terminated.

  In step 323 of FIG. 11, the sampling program 243 is used to select alignment coordinates and observation target coordinates. That is, the detection possibility prediction data 502 is converted into sampling data 503. Here, the specification of the process for sampling each defect coordinate can vary depending on the scanning electron microscope model, the type of wafer to be processed, the user's request, etc. Will be set.

  Next, in step 324, an image is picked up using the electron optical systems 201 to 213 using the image pickup program 244 by preceding the 413 of the sampling data 503 by the coordinates of the defect described as “1 kV Alignment”.

  In step 325, alignment is performed based on the image captured in step 324. That is, the coordinate system offset between the drive stage 207 and the defect inspection data 501 is calculated so that the defect is surely reflected in the captured image.

  Next, in step 326, an image is captured using the image capturing program 244 with respect to the coordinates of the defect described as “1 kV Review”, “30 kV Review”, and “Optical Review” in 413 of the sampling data 503. As for the coordinates of the defect described as “1 kV Review”, the acceleration voltage is set to 1 kV and an image is picked up by the electron optical systems 201 to 213. As for the coordinates of the defect described as “30 kV Review”, the acceleration voltage is set to 30 kV and an image is taken by the electron optical systems 201 to 213. As for the coordinates of the defect described as “Optical Review”, an image is picked up by the optical systems 221 to 226.

  Next, in step 327, the defect type is classified for each coordinate using the defect classification program 246.

  FIG. 12 is a diagram illustrating an example of a graphical user interface of the scanning electron microscope. A screen 901 drawn by the GUI program 248 displays a detection possibility prediction result 502 by the detection possibility prediction program 242. 915 means the outer periphery of the wafer which is a sample. Reference numeral 916 denotes a notch for defining the direction of the sample. A large number of black dots 917 are defect coordinate groups included in the detection possibility prediction results designated by 911 to 914. Reference numeral 911 designates a serial number of the sample, and reference numeral 912 designates a process name inspecting the sample. In 913, a defect condition to be displayed is selected. In this example, since “Visible (1 kV)” is checked, the coordinates of the defect predicted to be detected by the scanning electron microscope under the condition of the acceleration voltage of 1 kV are displayed. In 914, it is selected whether to draw a wafer map or a die map in the wafer as in this example. In this example, a condition for not displaying the layout of the dies in the wafer is selected. As described above, in the scanning electron microscope according to the present invention, the display of the wafer map can be switched according to the detection possibility prediction result for the set acceleration voltage by having the detection possibility prediction program 242.

  FIG. 13 is a diagram illustrating an example of a graphical user interface of the scanning electron microscope. A screen 902 drawn by the GUI program 248 displays information on the defect with the serial number 15 in the classification result data 504 obtained by the image capturing program 244 and the defect classification program 246. A captured image is displayed in a square frame 924. In this example, it can be seen that the defect 926 is detected in the lower layer of the circuit pattern 925. Reference numeral 921 denotes the coordinates of the displayed defect. Reference numeral 922 denotes information written in 413 of the classification result data 504. Reference numeral 923 denotes a class described in 414 of the classification result data 504. Thus, in the scanning electron microscope according to the present invention, by having the detection possibility prediction program 242, different detection conditions, that is, different acceleration voltages can be automatically set, and an image of the scanning electron microscope can be taken. The conditions can also be displayed together with the image.

  FIG. 14 is a diagram illustrating an example of a graphical user interface of the scanning electron microscope. A screen 903 drawn by the GUI program 248 displays the result of totalizing the classification result data 504 as a mosaic diagram 931. The horizontal axis of the mosaic diagram 931 indicates detection conditions generated from the information 413, that is, “1 kV Review” and “1 kV Alignment” are “1 kV”, “30 kV Review” is “30 kV”, and “Optical Review” is “Optical”. It is defined as follows. The vertical axis represents the class of the result 414 of defect classification. In this example, it can be seen that the defect captured by the electron optical system with the acceleration voltage set to 1 kV is “Particle”, followed by “Dimple” and “Pattern”. Further, the number of defects taken by the electron optical system with the acceleration voltage set to 30 kV is “Under”, followed by “Pattern”. In addition, it can be seen that a “False” painted black, that is, an image in which a defect could not be detected is also captured. Thus, in the scanning electron microscope according to the present invention, by having the detection possibility prediction program 242, different detection conditions, that is, different acceleration voltages can be automatically set, and an image of the scanning electron microscope can be taken. Defect classification results can be tabulated for each detection condition.

  FIG. 15 is an example of the detectability result data input to the detectability learning program 247. The detection result result data 505 is selected by randomly selecting coordinates from the defect inspection data without using the detection possibility prediction program 242, and setting the acceleration voltage 1 kV or the acceleration voltage 30 kV for the selected coordinates. This is a result of taking an image and evaluating whether or not a defect can be detected. The detection result data 505 is preferably not only the inspection result of one wafer but also the result of inspection of a plurality of wafers. The detectability result data 505 includes feature amounts 404, 405, 406, 407, and 408 included in the defect inspection data, a detection condition 416 of the scanning electron microscope, and a detectability evaluation result 417 of the scanning electron microscope.

  FIG. 16 is a diagram illustrating an example of a graphical user interface of the scanning electron microscope. On the screen 904 drawn by the GUI program 248, the detection possibility learning program 247 is called to calculate the coefficient of the prediction formula. In 941, the name of the feature quantity included in the detection result result data 505 is displayed, and the name of the feature quantity used for the prediction formula can be selected from the name. In this example, a state in which the feature amounts F02, F03, F04, and F05 are selected is displayed. After selecting the name of the feature amount, clicking the button 942 with the mouse causes the detection possibility learning program 247 to be called, and multiple logistic regression analysis is performed for each of the detection conditions 416.

FIG. 19 shows a flowchart of the processing of the detection feasibility learning program 247. Detectable / unusable result data 505 serving as an input of this program is a defect inspection / analysis data table obtained by collecting feature amount data newly subjected to defect inspection from the defect inspection apparatus via the network interface 234 upon activation of this program. The defect inspection data registered in 500 or already registered in the defect inspection / analysis data table 500 is read and used. In this case, the data in each column of the detection condition 416 and the detection result evaluation result 417 is to create a defect corresponding to the detection result data by inputting the result of the user's observation with the scanning electron microscope. Become. (S1021)
The feature quantity selected by the user is received from the graphical user interface of FIG. (S1022)
Subsequently, the logistic regression coefficient is calculated.
The logistic regression model equation shown in Equation 4 is a group of N explanatory variables x = (x ₁ , x ₂ ,..., X explaining the probability P of occurrence of an event in which a certain defect on the wafer cannot be detected by a scanning electron microscope. _N ), which is applied to a binary random variable Y where Y = 1 when an event in which a certain defect cannot be detected occurs and Y = 0 when a certain defect can be detected.

The coefficients of the linear combination of Equation ₂ are shown in a vector expression as a = (a ₁ , a ₂ ,..., A _N , b). Assume that n independent samples have been collected in order to estimate a.
(1) Vector of binary random variable Y: Y = (Y ₁ , Y ₂ ,..., Y _n ) ^t n × 1
(2) Observation value vector of random variable Y: y = (y ₁ , y ₂ ,..., Y _n ) ^t n × 1
(3) Matrix of explanatory variables: X = (x ₁ , x ₂ ,..., X _n ) ^t n × N
(4) Description variable vector of the i-th _{sample: x i = (x i1,} x i2, ......, x iN) t N × 1
The probability that the observed value of the random variable Y is y is

  It becomes. This probability as a function of parameter a under the condition given observation value y,

  Is called likelihood.

  When the number of samples n is sufficiently larger than the number of parameters N included in the model (n is 4 to 5 times N or more), the maximum likelihood method that maximizes this likelihood is used to obtain a true parameter value. Asymptotically close values are obtained.

For example, using the maximum likelihood method described in Non-Patent Document 1, N feature quantities selected by the user in S1022 are used as explanatory variables, and the detection result data created in S1021 is used as a sample. = (A ₁ , a ₂ ,..., A _N , b) is estimated (S1023).

  Chi-square test is performed for each feature amount data used for calculating the coefficient of the logistic regression equation. It is assumed that normal distribution data of past performance values is obtained for each feature amount data. For each feature amount, the range of values is divided into m regions, and the detection result data 505 for the number of samples n is distributed as an observation frequency to one of the respective range regions. Further, an expected frequency by which n samples are distributed to each range area is determined according to data of a normal distribution of past actual values.

  As a measure to measure the degree of discrepancy between observed frequency and expected frequency,

Is used. If the normal distribution of past actual values is correct and the number of samples n is sufficiently large, Z has a χ ² distribution with ν degrees of freedom. Using the χ ² distribution data with the degree of freedom ν, an upper probability P is obtained from the calculated Z (S1024).

The calculated values of the coefficients a ₁ , a ₂ ,..., A _N , b of the logistic regression equation and the result of performing the chi-square test for each feature value data are displayed on the graphical user interface shown in FIG. S1025).

  In 943 of FIG. 16, whether or not each feature amount selected in 941 is necessary for prediction of whether or not detection is possible is displayed for each detection condition. In this example, the p-value, that is, the significance probability based on the chi-square test is 0.05 or less for all of F02, F03, F04, and F05. Therefore, accurate prediction is realized by using these selected feature quantities. It means you can do it. Reference numeral 520 denotes a coefficient of the prediction formula calculated by the detectability learning program 247.

  The user verifies the display of FIG. 16 described above, determines whether the selection of the feature amount is appropriate and determines whether the coefficient of the prediction formula is valid, and inputs the following instruction (S1026).

When reviewing the feature amount of the prediction formula, the input from the user reselecting the feature amount is accepted from 941 (S1028), and the processing from S1023 to S1025 is repeated.
The selection of the feature quantity is appropriate and the user judges that the coefficient of the prediction formula is valid, receives an input instructed to click the button 944 with the mouse, and calculates the coefficient a ₁ , a of the calculated logistic regression formula ₂ ,..., A _N , b are stored for each detection condition and stored in the prediction formula coefficient data table 520 (S 1027).

  In the first embodiment, the example of the alternative of 1 kV and 30 kV has been described as the acceleration voltage set in the scanning electron microscope. In the first embodiment, different prediction formulas are used with different coefficients, as in the prediction formula coefficient data table 520 of FIG. 8, depending on whether the acceleration voltage is set to 1 kV or the acceleration voltage is set to 30 kV. .

  In the second embodiment, a method for predicting an appropriate acceleration voltage instead of the alternative will be described. In the second embodiment, an acceleration voltage is added to the explanatory variable of Equation 2 to generate a coefficient of the mathematical formula. That is, in the first embodiment, the feature quantity of the defect inspection data 501 is substituted for x1 to xN in Formula 2, but in the second embodiment, the feature quantity is substituted for x1 to xN-1, where xN is a term of acceleration voltage. And

  Expression 9 is an expression converted to an expression for calculating the acceleration voltage term of Expression 5, that is, xN.

  Each coefficient in Equation 9 is learned in substantially the same manner as each coefficient in Equation 5, but the detection result data 505 used for learning is not limited to the two types of 1 kV and 30 kV shown in 416. It is desirable to evaluate and input the possibility of detection even in the case of more acceleration voltages.

In order to predict an appropriate acceleration voltage using Equation 9, a feature amount obtained from defect inspection data, a coefficient of a prediction formula, and an arbitrary predictability probability are input to a variable P. For example, the probability that it cannot be detected by the scanning electron microscope is set to 10%, and 10 is substituted into the variable P in Equation 9. As a result, x _N , that is, an appropriate acceleration voltage is calculated, and the calculated acceleration voltage is set in the scanning electron microscope to capture an image.

  In the third embodiment, an example in which the detectability prediction program 242 is used in the defect inspection apparatus 601 will be described. As already described with reference to FIG. 3, the detectability prediction program 242 can be used by storing it in the secondary storage device 724 included in the defect inspection apparatus 601. In that case, it is necessary to capture the detection result data 505 from the scanning electron microscope 600. Therefore, the secondary storage device 724 of the defect inspection apparatus 601 requires the detection availability input program 743.

  As a method of utilizing the detection possibility prediction program 242 in the defect inspection apparatus 601, the defect inspection apparatus 601 may be used for image capturing conditions as in the first embodiment, but for filtering the defect coordinate data 501 output from the defect inspection apparatus 601. It can also be used. Specifically, in the defect coordinate data 501, the coordinates of the defect that cannot be detected by the scanning electron microscope are not output. In that case, the defect inspection apparatus 601 calculates the prediction result 502 so as not to output the coordinates of a defect having a high probability that it cannot be detected by the scanning electron microscope, for example, the coordinate of a defect having a probability that the detection of 409 cannot be detected is 70% or more. To.

  DESCRIPTION OF SYMBOLS 201 ... Electron source, 202 ... Accelerating electrode, 203 ... Condensing lens, 204 ... Deflector, 205 ... Objective lens, 206 ... Sample (for example, wafer), 207 ... Driving stage, 208 ... Primary electron, 209 ... Secondary electron signal, 210 ... Secondary electron detector, 211 ... Digital converter, 212a, 212b ... Reflected electron detector, 213 ... Reflected electron signal, 221a, 221b ... Light source, 222a, 222b, 222c ... Reflecting mirror, 223 ... Condensing lens, 224 ... beam splitter, 225a, 225b ... photoelectric conversion unit, 226 ... digital conversion unit, 230 ... bus, 231 ... overall control unit, 232 ... CPU, 233 ... primary storage device, 234 ... network interface, 235 ... keyboard, 236 ... Mouse, 237 ... display, 238 ... secondary storage device, 241 ... defect inspection data 242 ... Sampling program, 244 ... Image capturing program, 245 ... Alignment program, 246 ... Defect classification program, 247 ... Detection feasibility learning program, 248 ... GUI program, 321 ... Defect inspection data Reading step, 322 ... detectability prediction step, 323 ... coordinate sampling step, 324 ... alignment imaging step, 325 ... alignment step, 326 ... review imaging step, 327 ... defect classification step, 401 ... defect detection step Serial number 402 ... X coordinate 403 ... Y coordinate 404 ... feature amount F01, 405 ... feature amount F02, 406 ... feature amount F03, 407 ... feature amount F04, 408 ... feature amount F05, 409, 411 ... cannot be detected 410, 412 ... Predicted detection result, 413 ... Detection condition, 414 ... Defect classification class, 415 ... Image storage location, 416 ... Detection condition, 417 ... Detectability result, 421 ... Outer frame of sample, 422 ... notch, 423 ... X coordinate axis, 424 ... Y coordinate axis, 425 ... defect point, 500 defect inspection / analysis data table, 501 ... defect inspection data, 502 ... detectability prediction result, 503 ... sampling data, 504 ... classification result data 505 ... detectability result data, 520 ... prediction formula coefficient data table, 522,523 ... prediction formula coefficients, 600 ... scanning electron microscope, 601 ... defect inspection apparatus, 602 ... network, 701a, 701b ... light source, 702a, 702b , 702c: Reflection mirror, 703: Condensing lens, 704: Beam splitter, 705a, 7 5b: photoelectric conversion unit, 706: digital conversion unit, 707 ... drive stage, 720 ... bus, 721 ... general control unit, 722 ... CPU, 723 ... primary storage device, 724 ... secondary storage device, 725 ... network interface, 726 ... Keyboard, 727 ... Mouse, 728 ... Display, 741 ... Defect coordinate detection program, 742 ... Feature extraction program, 743 ... Detectability input program, 744 ... GUI program, 801 ... Peak of luminance distribution, 802 ... Saturation of luminance distribution State, 901, 902, 903, 904 ... screen, 911 ... sample serial number, 912 ... inspection process name, 913, 914 ... display condition selection, 915 ... sample outer frame, 916 ... notch, 917 ... defect point, 921 ... Defect coordinates, 922 ... Detection condition, 923 ... Defect classification result class 924 ... image, 925 ... circuit pattern, 926 ... lower layer defect, 931 ... mosaic view, 941 ... feature amount of the selected conditions, 942, 944 ... execution button, 943 ... significant probability for each feature quantity.

Claims (7)

In the scanning electron microscope having the function of inputting the information that the defect inspection apparatus has detected the defect coordinates on the sample and taking an observation image for analyzing the defect,
A function of inputting the feature amount of the defect detection image together with the coordinates of the plurality of defects detected by the defect inspection device,
A setting condition of acceleration voltage in the scanning electron microscope using the feature amount, a function for predicting whether or not a defect can be detected in the observation image, and
A function for selecting a plurality of coordinates for alignment and observation from a group of coordinates input based on the prediction result;
A scanning electron microscope having a function of setting an acceleration voltage that is predicted to be detectable and picking up an image for each of a plurality of selected coordinates. Collect sample data that includes imaging results that confirm the acceleration voltage that can be detected against the defect coordinates on the measurement sample in advance.
2. The function of calculating a coefficient of a logistic regression equation created based on a probability of an event in which a defect cannot be detected and a feature amount obtained from an image based on the sample data. Scanning electron microscope. The scanning electron microscope is provided with an optical microscope,
2. The function of capturing an image according to claim 1, wherein the image is captured by selecting either an electron optical system or an optical system for each defect based on the prediction result of the detection possibility. Scanning electron microscope. Present a graphical user interface to the user,
Accepting that the user selects a feature quantity to be used in the prediction formula from the list of feature quantities, creating a logistic regression model using the selected feature quantity, and calculating the logistic regression formula coefficients. The scanning electron microscope according to claim 2, further comprising a function of presenting a test result of whether or not the sample data of each selected feature amount is effective in predicting whether or not detection is possible. In the function of predicting whether or not the defect can be detected, an optimum acceleration voltage is calculated from the feature amount instead of selecting an acceleration voltage setting condition from a plurality of prescribed values.
In the function of capturing an image, the image is captured by setting the calculated optimum acceleration voltage instead of the acceleration voltage predicted to be detectable from a plurality of predetermined prescribed values. The scanning electron microscope according to claim 1. A defect inspection apparatus that detects the coordinates of the defect on the sample and a scanning electron microscope that captures an observation image of the coordinates are connected via a network,
A function for predicting the setting condition of the acceleration voltage in the scanning electron microscope using the feature amount of the image obtained when the defect inspection apparatus detects the coordinates of the defect, the ability to detect the defect in the observation image for each defect,
Based on the prediction result, a function of selecting a plurality of defects to be aligned and observed from a group of coordinates detected by the defect inspection apparatus;
A defect inspection system having a function of setting an acceleration voltage predicted to be detectable by a scanning electron microscope for each of a plurality of selected coordinates and capturing an image. A function to detect the coordinates of defects on the specimen;
The feature amount is calculated from the defect image obtained when detecting the coordinates of the defect, and the condition value of acceleration voltage when detecting each defect from the observation image with the scanning electron microscope and the detectability are used. And the ability to predict for each defect,
A defect inspection apparatus having a function of outputting coordinates of a defect having a high probability of being detected by a scanning electron microscope based on a prediction result.

Images