Defect inspection device and defect inspection method

Abstract

Provided is a defect inspection device that processes an output signal from a sensor, the output signal being obtained by spirally scanning a sample while performing rotation control of a stage. Said defect inspection device: reads boundary frequencies fc, fl1, and fh1, which are time frequencies of the output signal from the sensor and have been set so as to satisfy the relationship fl1 < fh1 < fc; separates the output signal from the sensor into a first high-frequency components which are time-frequency components equal to or greater than fc, and first low-frequency components which are time-frequency components lower than fc; detects a defect by processing the first high-frequency components; computes, for first low-frequency components that include second low-frequency components lower than fl1, intermediate-frequency components equal to or greater than fl1 and lower than fh1, and second high-frequency components equal to or greater than fh1, a haze signal that is obtained by performing filtering processing for reducing the intermediate-frequency components relative to the second low-frequency components and the second high-frequency components; computes a haze map of the sample on the basis of the haze signal; and outputs the haze map to an output device.

Description

Defect inspection device and defect inspection method

[0001] The present invention relates to a defect inspection apparatus and a defect inspection method for inspecting defects in a sample such as a semiconductor silicon wafer.

[0002] In a defect inspection apparatus that uses scattered light to inspect a semiconductor silicon wafer or the like as a sample for defects, a method is known in which low-frequency variations are removed by two-dimensional polynomial fitting in order to reveal a haze signal that is detected as a low-frequency component compared to the defect (Patent Document 1).Also, a method is known in which image data processing is performed in a direction along the haze value measurement direction to remove noise components (Patent Document 2).

[0003] U.S. Patent No. 4,691,499, U.S. Patent No. 5,463,943

[0004] The haze signal reflects wafer conditions such as roughness and film thickness, making it an important indicator for wafer quality control. However, in addition to signals derived from wafer conditions, changes in signal intensity due to air fluctuations and fluctuations in the beam emission direction can also occur as concentric (annular) noise, which can degrade the sensitivity of the haze signal to be observed. The concentric noise that occurs across the entire wafer due to these various factors is high-frequency compared to the low-frequency components targeted for removal by the method of Patent Document 1, making it difficult to remove using the method of Patent Document 1. Furthermore, the concentric noise occurs at a specific frequency, making it difficult to remove using the method of Patent Document 2, which determines a data processing area with a predetermined number of pixels, in cases such as inspections where the scanning speed varies depending on the distance from the wafer center.

[0005] An object of the present invention is to provide a defect inspection apparatus and a defect inspection method that can reduce noise signals that occur at specific frequency components, such as signals of concentric noise components, and detect a target haze signal with high sensitivity.

[0006] In order to achieve the above object, the present invention provides an illumination optical system that supplies illumination light to a sample, a detection optical system that collects scattered light scattered from an irradiation area on the sample that is irradiated with the illumination light, a sensor that converts the scattered light collected by the detection optical system into an electric signal, a stage on which the sample is placed, a control device that controls the rotation of the stage, and a signal processing device that processes an output signal of the sensor obtained by spirally scanning the sample by controlling the rotation of the stage with the control device, wherein the signal processing device reads boundary frequencies fc, fl1, and fh2 that are time frequencies of the output signal of the sensor and are set to satisfy the relationship fl1<fh1<fc, and converts the output signal of the sensor into an electric signal. and a first high-frequency component which is a time frequency component equal to or higher than a boundary frequency fc, and a first low-frequency component which is a time frequency component less than the boundary frequency fc, processes the first high-frequency component to detect defects, calculates a haze signal for the first low-frequency component including a second low-frequency component less than the boundary frequency fl1, an intermediate-frequency component equal to or higher than the boundary frequency fl1 and less than the boundary frequency fh1, and a second high-frequency component equal to or higher than the boundary frequency fh1, by performing a filter process on the second low-frequency component and the second high-frequency component to reduce the intermediate-frequency component, and calculates a haze map of the sample based on the haze signal and outputs the result to an output device.

[0007] According to the present invention, it is possible to reduce noise signals occurring at specific frequency components, such as signals of concentric noise components, and to detect the haze signal of interest with high sensitivity.

[0008] Schematic diagram of a defect inspection device according to one embodiment of the present invention. Diagram showing an example of a scanning trajectory on a sample surface. Schematic diagram of an example of the configuration of an attenuator. Schematic diagram of an example of the configuration of an oblique incidence illumination optical system of the scattered light inspection illumination unit. Schematic diagram of an example of the configuration of a normal incidence illumination optical system of the scattered light inspection illumination unit. Schematic diagram showing the positional relationship of some sensors that detect scattered light in oblique incidence illumination, viewed from a direction perpendicular to the sample surface. Schematic diagram showing the positional relationship of areas (apertures) where the detection unit detects scattered light, viewed from a direction perpendicular to the sample surface. Diagram showing an example of a Haze map in which concentric noise appears. Diagram explaining a process for reducing noise from a first low-frequency component time waveform in one embodiment of the present invention. Diagram explaining an example of determining boundary frequencies fl1 and fh1 of areas where noise occurs, based on time waveform data. Diagram explaining an example of changing boundary frequencies fl1 and fh1 for each area of ​​the sample. Diagram explaining filtering performed on points sampled in a one-dimensional space in one embodiment of the present invention. Conceptual diagram of a band-pass filter that extracts intermediate frequency components. Diagram showing an example of a two-dimensional filter that removes specific frequency components in an Rθ direction in one embodiment of the present invention. FIG. 1 shows an example of filtering two-dimensional haze data in the R and θ directions using filter coefficient sequences; FIG. 2 shows an example of filtering haze signal intensity for each revolution in one embodiment of the present invention; FIG. 3 shows an example of the frequency characteristics of the average filter for each revolution; FIG. 4 explains an example of noise removal performed for the same position on multiple haze maps obtained by multiple inspections; FIG. 5 shows an example of a GUI for setting filtering to remove noise occurring in a specific frequency range; FIG. 6 explains an example of a flow for performing independent integration processing on a first low-frequency component and a first high-frequency component in one embodiment of the present invention; FIG. 7 explains another example of a flow for performing independent integration processing on a first low-frequency component and a first high-frequency component in one embodiment of the present invention;

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0010] --Defect Inspection Apparatus-- Fig. 1 is a schematic diagram of a defect inspection apparatus according to one embodiment of the present invention. The defect inspection apparatus illustrated in Fig. 1 includes a stage G that moves and carries a sample 1, such as a semiconductor silicon wafer, an inspection system 100 that irradiates illumination light onto the sample 1 and detects light from the sample 1, and a signal processing device 200 that processes an output signal from the inspection system 100. The inspection system 100 included in the defect inspection apparatus of this embodiment is a scattered light inspection system that irradiates illumination light onto the sample 1 and detects scattered light generated by the sample 1. While this embodiment illustrates a defect inspection apparatus equipped with one type of inspection system, the present invention can also be applied to a defect inspection apparatus that also has inspection systems other than scattered light inspection.

[0011] The defect inspection apparatus of this embodiment includes a scattered light inspection and illumination unit A, a detection unit B, a signal processing unit C, a control unit D, an input unit E, a display unit F, and a stage G. The scattered light inspection and illumination unit A and its detection unit B constitute the inspection system 100. The scattered light signal processing unit C constitutes the signal processing unit 200.

[0012] The signal processing device 200 and the control device D are computers, and are configured to include a CPU, FPGA, timer, etc. in addition to ROM, RAM, and other memories. While the signal processing device 200 is assumed to be configured as a single computer that forms a unit with the main body of the defect inspection device (the mechanical parts that perform inspection, such as the inspection system 100 and the stage G), it may also be configured as multiple computers connected via a network. In this case, a server may be used as one of the multiple computers, and the server may also be included as a component of the defect inspection device or the signal processing device. For example, a configuration may be adopted in which a computer attached to the main body of the defect inspection device acquires defect detection signals from the main body, processes the detection signals as necessary, and transmits them to a server, where processing such as defect classification is performed by the server.

[0013] The scattered light signal processing section C of the signal processing device 200 may be realized virtually by software, or may be realized by hardware such as an electronic circuit. Part of these scattered light signal processing sections C, etc. (particularly the upstream processes) can be configured using an FPGA or DSP. Some or all of the functions of these scattered light signal processing sections C, etc. can also be executed by a server. Furthermore, the signal processing device 200 and the control device D can also be configured as a single computer.

[0014] The scattered light inspection illumination unit A is an illumination optical system that supplies illumination light to the sample 1 placed on the stage G. The scattered light inspection illumination unit A includes a laser light source A1, an attenuator A2, an emitted light adjustment unit A3, a beam expander A4, a polarization control unit A5, mirrors A6 and A7, condenser lenses A8 and A9, and a mirror A10. When inspection is performed by a different method using light of a different wavelength, a dichroic mirror may be used as the mirror A10.

[0015] The illumination light emitted from the laser light source A1 is adjusted to a desired beam intensity by the attenuator A2, adjusted to a desired beam position and beam traveling direction by the output light adjustment unit A3, and adjusted to a desired beam diameter by the beam expander A4. The illumination light whose beam diameter has been adjusted is adjusted to a desired polarization state by the polarization control unit A5, and in the case of oblique incidence illumination, is reflected by mirrors A6 and A7 and focused by the condenser lens A8 to illuminate the inspection target area of ​​the sample 1.

[0016] The angle of incidence of the illumination light relative to the surface of the sample 1 (hereinafter referred to as the sample surface) is determined by the position and angle of the reflecting mirrors A3a and A3b of the output light adjusting unit A3, which are arranged in the optical path of the scattered light inspection and illumination unit A. The angle of incidence of the illumination light is set to an angle suitable for detecting minute defects. The larger the illumination incident angle, i.e., the smaller the illumination elevation angle (the angle between the sample surface and the illumination optical axis), the weaker the scattered light from minute irregularities on the sample surface (called haze) and light from patterns on the sample, which become noise compared to the scattered light from minute foreign particles on the sample surface, making it more suitable for detecting minute defects. Therefore, if scattered light from minute irregularities on the sample surface interferes with the detection of minute defects, it is preferable to set the incident angle of the illumination light to 75 degrees or more (an elevation angle of 15 degrees or less). On the other hand, in oblique incidence illumination, the smaller the illumination incidence angle, the greater the absolute amount of scattered light from minute foreign particles. Therefore, if insufficient scattered light from defects hinders the detection of minute defects, it is preferable to set the incident angle of the illumination light to between 60 degrees and 75 degrees (elevation angle between 15 degrees and 30 degrees). Furthermore, for the scattered light inspection illumination, polarization control is performed using a half-wave plate A5a and a quarter-wave plate A5b in the polarization control unit A5 of the scattered light inspection illumination unit A to set the polarization of the illumination to P-polarized light, which increases the scattered light from defects on the sample surface compared to other polarized light. Furthermore, if scattered light from minute irregularities on the sample surface hinders the detection of minute defects, setting the polarization of the illumination to S-polarized light reduces the scattered light from minute irregularities on the sample surface compared to other polarized light.

[0017] 1, if necessary, the illumination light path can be changed by removing mirror A6 from the optical path of the scattered light inspection illumination unit A, and illumination light is incident on condenser lens 9 from polarization control unit A5. In this case, illumination light focused by condenser lens A9 is reflected by mirror A10 and irradiated from a direction substantially perpendicular to the sample surface (perpendicular illumination). Perpendicular illumination, which is incident substantially perpendicular to the sample surface, is suitable for obtaining scattered light from concave defects on the sample surface (polishing scratches or crystal defects in crystalline materials).

[0018] When detecting minute defects near the surface of a sample, the laser light source A1 is one that emits a short-wavelength (355 nm or less) ultraviolet or vacuum ultraviolet laser beam, which has a wavelength that does not easily penetrate into the interior of the sample, and has a high output of 2 W or more. When detecting defects inside the sample, the laser light source A1 is one that emits a visible or infrared laser beam, which has a wavelength that easily penetrates into the interior of the sample.

[0019] As shown in FIG. 3, the attenuator A2 is configured by a combination of a polarizing plate and a wavelength plate, and is used to reduce the intensity of the laser light.

[0020] The output light adjustment unit A3 includes multiple reflecting mirrors. Here, a configuration in which the output light adjustment unit A3 includes two reflecting mirrors, A3a and A3b, is described. However, the configuration of the output light adjustment unit A3 is not limited to this example, and three or more reflecting mirrors may be used as appropriate. Here, a three-dimensional Cartesian coordinate system (XYZ coordinates) is hypothetically defined, and it is assumed that light incident on the reflecting mirrors travels in the +X direction. The first reflecting mirror A3a is positioned to deflect the incident light in the +Y direction (incidence and reflection in the XY plane), and the second reflecting mirror A3b is positioned to deflect the light reflected by the first reflecting mirror in the +Z direction (incidence and reflection in the YZ plane). The reflecting mirrors A3a and A3b adjust the position and propagation direction (angle) of the light emitted from the output adjustment unit A3 by adjusting their respective parallel translations and tilt angles. As described above, by arranging the incident / reflecting surface (XY plane) of the reflecting mirror A3a and the incident / reflecting surface (YZ plane) of the reflecting mirror A3b so that they are perpendicular to each other, it is possible to independently adjust the position and angle of the light emitted from the output light adjustment unit A3 (traveling in the +Z direction) in the XZ plane and in the YZ plane.

[0021] The beam expander A4 has two or more lens groups and functions to expand the diameter of the incident parallel beam. For example, a Galilean beam expander equipped with a combination of a concave lens A4a and a convex lens A4b is used. The beam expander A4 is mounted on a translation stage with two or more axes, allowing for offset adjustment of its own optical axis. The beam expander A4 also has a function for adjusting the tilt angle of the entire beam expander A4 so that the optical axis of the beam expander A4 coincides with the optical axis of a predetermined beam. The expansion rate of the beam diameter can be controlled by adjusting the spacing between the lenses (zoom mechanism). When the light incident on the beam expander A4 is not parallel, adjusting the spacing between the lenses simultaneously expands the beam diameter and collimates it (makes the beam quasi-parallel). The beam can also be collimated by installing a collimating lens upstream of the beam expander A4, independent of the beam expander A4.

[0022] The polarization control unit A5 is composed of a half-wave plate A5a and a quarter-wave plate A5b, and controls the polarization state of the illumination light to an arbitrary polarization state.

[0023] The mirrors A6 and A7 can also be used to adjust the position and angle of obliquely incident light, similarly to the reflecting mirrors A3a and A3b of the emitted light adjusting unit A3.

[0024] The condenser lens A8 is provided with an angle and position adjustment mechanism so that its own optical axis can be aligned with the optical axis of the incident light.

[0025] Scattered light scattered in various directions from an illumination spot, which is an illumination area on the sample surface illuminated with illumination light from the scattered light inspection illumination unit A, is incident on and collected by a plurality of scattered light detection optical systems Bn (n = 1, 2, ...) in the detection unit B. The scattered light collected by the scattered light detection optical systems Bn is detected by scattered light sensors Bn-2 (n = 1, 2, ...) and converted into electrical signals. The detection signals (scattered signal intensities) output from the scattered light sensors Bn-2 are input to the scattered light signal processing unit C. Note that while FIG. 1 shows four scattered light detection optical systems B1-B4 and four scattered light sensors B1-2, B4-2, the defect inspection apparatus may be equipped with many more scattered light detection optical systems and scattered light sensors.

[0026] Digital signals discretized from the output signal of scattered light sensor Bn-2 at a sampling pitch on the sample surface based on a preset rule are input to scattered light signal processing unit C in association with coordinates on the sample surface. Boundary frequencies fc, fl1, and fh2 of the output signal of scattered light sensor Bn-2 are read from the storage device to scattered light signal processing unit C. Boundary frequencies fc, fl1, and fh2 are time frequencies set so as to satisfy the relationship fl1<fh1<fc for the output signal of scattered light sensor Bn-2.

[0027] In the scattered light signal processing unit C, the frequency separation unit C1 separates the input data set of the detection signals of the scattered light sensor Bn-2 into a first high-frequency component, which is a time frequency component equal to or greater than the boundary frequency fc, and a first low-frequency component, which is a time frequency component less than fc.

[0028] Defects such as foreign particles and scratches on the sample surface are detected by processing the first high-frequency component. Specifically, the signal intensity of the first high-frequency component output from the frequency separation unit C1 is compared with a predetermined high-sensitivity threshold in the defect candidate extraction unit C2 to determine whether the signal intensity exceeds the predetermined high-sensitivity value. The high-sensitivity threshold is a value set for the purpose of detecting numerous defect candidates, including false detections, and is set lower than a low-sensitivity threshold set for the purpose of detecting true defects. A dataset of detection signals exceeding the high-sensitivity threshold and an integrated signal of these detection signals constitutes a defect candidate signal. These defect candidate signals are integrated in the defect integration processing unit C4 based on first integration conditions set in advance. The first integration conditions include, for example, a combination of sensors (two or more sensors at different positions) that integrate the signals and an amplification factor for each sensor to integrate the signals. The first integration conditions may also include an amplification factor applied to a signal obtained by integrating two or more signals. Furthermore, the scattered light inspection illumination unit A can select at least one polarization of illumination to irradiate the sample 1, and the first integration conditions can also be set differently for each polarization of the illumination light.

[0029] The integrated signal output from the defect integrated processing unit C4 is further compared with a low-sensitivity threshold in a defect detection unit C6, and a signal exceeding the low-sensitivity threshold becomes defect inspection data that is finally output by the signal processing device 200 for defect inspection of foreign matter, scratches, etc. on the sample surface. This defect inspection data can be saved in a database DB or displayed as a defect map on a display device F. Past defect inspection data accumulated in the database DB is also used as teaching data for defect classification.

[0030] The first low-frequency components associated with each sensor output from the frequency separation unit C1 are integrated in the haze integration processing unit C3 based on second integration conditions that are set independently of the first integration conditions. The second integration conditions include, for example, the combination of sensors whose signals are to be integrated and the amplification factor of the signals for each sensor whose signals are to be integrated. The second integration conditions may also include the amplification factor applied to the signal obtained by integrating two or more signals. Furthermore, while the scattered light inspection illumination unit A can select at least one polarization of the illumination to irradiate the sample 1, the second integration conditions can also be set differently for each polarization of the illumination light. The sensor combinations and amplification factors in the second integration conditions may be set separately for haze inspection and different combinations and values ​​from those in the first integration conditions for defect inspection.

[0031] The first low-frequency component signal output from the haze integration processor C3 is separated in the filter processor C5 into a second low-frequency component less than the boundary frequency f11, an intermediate-frequency component equal to or greater than the boundary frequency f11 but less than the boundary frequency fh1, and a second high-frequency component equal to or greater than the boundary frequency fh1. Furthermore, the filter processor C5 performs a filter process to reduce the intermediate-frequency components of the second low-frequency component and the second high-frequency component, and a haze signal is calculated after this filter process. The haze signal thus subjected to the integration process and filtering process in the haze integration processor C3 and the filter processor C5 becomes the haze inspection data finally output by the signal processing device 200 for inspecting the shape of the sample surface, etc. This haze inspection data can be stored in a database DB or output as a haze map to an output device (e.g., the display device F). Past haze inspection data stored in the database DB is also used as training data for haze classification.

[0032] 1 illustrates an example in which the processing by the haze integration processor C3 is performed before the processing by the filter processor C5, but the processing by the haze integration processor C3 may be performed after the processing by the filter processor C5. Also, although an example in which the first high-frequency component and the second low-frequency component are separated by one boundary frequency fc has been illustrated, two boundary frequencies fc1 and fc2 (fc1<fc2) may be set as the boundary frequency fc, and the temporal frequency components equal to or less than fc1 may be defined as the first low-frequency components, and the temporal frequency components equal to or greater than fc2 may be defined as the first high-frequency components.

[0033] The control device D controls the main body of the defect inspection device (inspection system 100, stage G, etc.) during the above inspection operations. The output of the inspection system 100 is associated with coordinates on the sample surface based on control data output by the control device D. Input operations related to the inspection conditions and the display on the display device F are performed by operating the input device E.

[0034] --Example of Scanning-- Figure 2 shows an example of a scanning trajectory on the sample surface. The illuminance distribution shape (illumination spot BS) formed on the sample surface by the scattered light inspection illumination unit A and the sample scanning method will be described using Figure 2. In Figure 2, a circular semiconductor silicon wafer is assumed as the sample 1. The stage G includes a translation stage, a rotation stage, and a Z-stage for adjusting the height of the sample surface (none of which are shown). The illumination spot BS has an illumination intensity distribution that is long in one direction. The longitudinal direction of the illumination spot BS is designated s2, and the direction substantially perpendicular to s2 is designated s1. The illumination spot BS is scanned in the circumferential direction (s1 direction) of a circle centered on the rotation axis of the rotation stage by control of the rotation of the rotation stage by control device D, and simultaneously scanned in the translation direction of the translation stage (s2 direction) by control device D's translation of the translation stage. During one rotation of the sample 1 due to scanning in the scanning direction s1, the sample 1 translates in the direction s2 by a distance equal to or less than the longitudinal length of the illumination spot BS. As a result, the illumination spot BS moves on the sample surface while tracing a spiral orbit T, and the entire surface of the sample 1 is scanned.

[0035] When inspecting the sample surface for defects, first, the sample 1 is placed on the stage G and moved while illuminating light is applied to the sample 1 to scan the sample surface and obtain a detection signal. At this time, the inspection system 100 irradiates the sample 1 with illumination light and detects scattered light from the sample 1 using a scattered light inspection method. As a result, the detection signal from the inspection system 100 is input to the processing device 200. The accuracy of the coordinates of the inspection system 100 is ensured by correction using a correction mechanism provided in the defect inspection device (for example, a mechanism that measures the deviation between the center coordinates of the sample 1 obtained by detecting the outer periphery of the sample 1 and the center of rotation of the sample 1 as a correction amount).

[0036] --Configuration Example of Attenuator-- FIG. 3 is a schematic diagram of an example configuration of the attenuator A2. The attenuator A2 is a unit that attenuates the light intensity of the illumination light from the laser light source A1. In this embodiment, a configuration that combines a first polarizing plate A2a, a half-wave plate A2b, and a second polarizing plate A2c is illustrated. The half-wave plate A2b is configured to be rotatable around the optical axis of the illumination light. The illumination light incident on the attenuator A2 is converted into linearly polarized light by the first polarizing plate A2a, and then the polarization direction is adjusted to the slow axis azimuth angle of the half-wave plate A2b before passing through the second polarizing plate A2c. By adjusting the azimuth angle of the half-wave plate A2b, the light intensity of the illumination light can be attenuated at any desired ratio. If the degree of linear polarization of the illumination light incident on the attenuator A2 is sufficiently high, the first polarizing plate A2a can be omitted. The attenuator A2 is not limited to the configuration illustrated in FIG. 3, but can also be configured using an ND filter with a gradation density distribution, or can be configured such that the attenuation effect can be adjusted by combining multiple ND filters with different densities.

[0037] --Configuration Example of the Scattered Light Inspection Illumination Unit-- Figure 4 is a schematic diagram of an example of the configuration of the oblique incidence illumination optical system of the scattered light inspection illumination unit A, and Figure 5 is a schematic diagram of an example of the configuration of the normal incidence illumination optical system of the scattered light inspection illumination unit A. Figures 4 and 5 show the positional relationship between the optical axis of the illumination light guided to the sample surface by the scattered light inspection illumination unit A and the illumination intensity distribution shape. Figure 4 is a schematic diagram of a cross section of the sample 1 cut along the plane of incidence of the illumination light obliquely incident on the sample 1. Figure 5 is a schematic diagram of a cross section of the sample 1 cut along a plane that is perpendicular to the plane of incidence of the illumination light incident on the sample 1 and includes the normal to the sample surface. The plane of incidence is a plane that includes the optical axis OA of the illumination light incident on the sample 1 from the oblique incidence illumination optical system and the normal to the sample surface at the center of the illumination spot BS. Note that Figures 4 and 5 show only a portion of the scattered light inspection illumination unit A, and for example, the output light adjustment unit A3 and mirrors A6 and A7 are omitted from the illustration.

[0038] When mirror A6 is inserted into the optical path, the illumination light emitted from laser light source A1 is reflected by mirror A7, condensed by condenser lens A8, and obliquely incident on sample 1. In this way, the scattered light inspection illumination unit A is configured to allow illumination light to be incident on sample 1 from a direction oblique to the normal to the sample surface. This oblique incidence illumination has its light intensity adjusted by attenuator A2, its beam diameter adjusted by beam expander A4, and its polarization adjusted by polarization controller A5, thereby homogenizing the illumination intensity distribution within the incident surface. As shown in the illumination intensity distribution (illumination profile) LD1 in Figure 4, the illumination spot formed on sample 1 has a Gaussian light intensity distribution in the s2 direction, and the length of the beam width l1, defined as 13.5% of the peak, is, for example, approximately 25 μm to 4 mm.

[0039] On the other hand, in a plane perpendicular to the incident surface and the sample surface, the illumination spot BS has a light intensity distribution with weak intensity at the periphery relative to the center of the optical axis OA, as shown in the illumination intensity distribution (illumination profile) LD2 in Figure 5. Specifically, the intensity distribution resembles a Gaussian distribution reflecting the intensity distribution of light incident on the condenser lens A9, or a first-order Bessel function or sinc function reflecting the aperture shape of the condenser lens A9. The length l2 of the illumination intensity distribution in the plane perpendicular to the incident surface and the sample surface is set to be shorter than the beam width l1 shown in Figure 4, for example, approximately 1.0 μm to 20 μm, in order to reduce haze generated from the sample surface. The length l2 of this illumination intensity distribution LD2 is the length of the region in the plane perpendicular to the incident surface and the sample surface where the illumination intensity is 13.5% or more of the maximum illumination intensity.

[0040] -Configuration example of scattered light detection section- Figure 6 is a schematic diagram showing the positional relationship of some sensors that detect scattered light from oblique incidence illumination, viewed from a direction perpendicular to the sample surface. Figure 7 is a schematic diagram showing the positional relationship of areas (apertures) that detect scattered light and reflected light, viewed from a direction perpendicular to the sample surface.

[0041] The inspection system 100 has multiple detectors B arranged to detect scattered light emitted in multiple directions from the illumination spot BS. The inspection system 100 is equipped with a low-angle detector arranged at a low angle and a high-angle detector arranged at a high angle. Figure 6 shows a plan view of the arrangement of the low-angle detectors BL. The angle between the direction of travel of the oblique incident illumination and the detection direction in a plane parallel to the sample surface is defined as the detection azimuth angle. The low-angle detectors BL in Figure 6 include a low-angle front detector BLf, a low-angle side detector BLs, a low-angle rear detector BLb, and a low-angle front detector BLf', a low-angle side detector BLs', and a low-angle rear detector BLb', which are positioned symmetrically to the low-angle front detector BLf', a low-angle side detector BLs', and a low-angle rear detector BLb', which are positioned symmetrically to the low-angle front detector BLf', a low-angle side detector BLs', and a low-angle rear detector BLb', which are positioned symmetrically to the low-angle front detector BLf', a low-angle side detector BLs', and a low-angle rear detector BLb', which are positioned symmetrically to the low-angle front detector BLf', a low-angle side detector BLs', and a low-angle rear detector BLb', which are positioned symmetrically to the low-angle front detector BLf', a low-angle rear detector BLb ... For example, the low-angle front detection unit BLf is installed at a position where the detection azimuth angle is between 0 degrees and 60 degrees, the low-angle side detection unit BLs is installed at a position where the detection azimuth angle is between 60 degrees and 120 degrees, and the low-angle rear detection unit BLb is installed at a position where the detection azimuth angle is between 120 degrees and 180 degrees.

[0042] 7 is a diagram showing the area (aperture) where the detection unit B collects scattered light as viewed from above, and corresponds to the arrangement of each objective lens of the scattered light detection optical systems B1-B4, etc. In this embodiment, the incident direction of the oblique incidence illumination onto the sample 1 is used as a reference, and the traveling direction of the incident light with respect to the illumination spot BS on the sample surface as viewed from above (to the right in FIG. 7) is defined as the forward direction, and the opposite direction (to the left in FIG. 7) is defined as the backward direction. Therefore, the lower side in FIG. 7 is the right side and the upper side is the left side with respect to the illumination spot BS.

[0043] Each objective lens of the scattered light detection optical system Bn is arranged along the upper half of a hemispherical surface of a sphere (celestial sphere) centered on the illumination spot BS on the sample 1. This hemispherical surface is divided into a total of 12 regions, L1-L6 and H1-H6, and the scattered light detection optical system Bn collects and focuses scattered light in each corresponding region. In this case, there are at least 12 scattered light detection optical systems Bn. (n≧12) Regions L1-L6 are equal divisions of an annular region surrounding the illumination spot BS at a low angle, 360 degrees around it. Looking from above, regions L1, L2, L3, L4, L5, and L6 are arranged counterclockwise from the direction of incidence of the oblique illumination. Of these regions L1-L6, regions L1-L3 are located to the right of the illumination spot BS, region L1 is located to the rear right of the illumination spot BS, region L2 is located to the right, and region L3 is located to the front right. The areas L4-L6 are located on the left side of the illumination spot BS, with the area L4 being located to the left front of the illumination spot BS, the area L5 being located to the left, and the area L6 being located to the left rear.

[0044] The remaining regions H1-H6 are equal divisions of a ring-shaped region surrounding the illumination spot BS 360 degrees at high angles, and are arranged counterclockwise from the direction of incidence of the oblique incidence illumination as viewed from above: regions H1, H2, H3, H4, H5, and H6. Compared to the low-angle regions L1-L6, the high-angle regions H1-H6 are positioned 30 degrees apart as viewed from above. Of the regions H1-H6, region H1 is located behind the illumination spot BS, and region H4 is located in front. Regions H2 and H3 are located to the right of the illumination spot BS, with region H2 located to the rear and right of the illumination spot BS, and region H3 located in front and right of the illumination spot BS. Regions H5 and H6 are located to the left of the illumination spot BS, with region H5 located to the front and region H6 located to the rear and left of the illumination spot BS.

[0045] In Fig. 1, scattered light incident on the scattered light detection optical system Bn is collected and guided to the corresponding sensor Bn-2. Comparing Fig. 1 with Fig. 7, for example, the scattered light detection optical system B1 in Fig. 1 can be considered to exemplify an optical system that collects scattered light in region L6 in Fig. 7, the scattered light detection optical system B2 in region H6 in Fig. 7, the scattered light detection optical system B3 in region H5 in Fig. 7, and the scattered light detection optical system B4 in region L4 in Fig. 7.

[0046] —Haze Map— FIG. 8 is a diagram showing an example of a haze map in which concentric noise obtained from the first low-frequency component appears. In the haze map shown in FIG. 8 , a concentric (annular) noise signal S9 is present across the entire sample surface. Noise occurring in a specific frequency band, such as this noise signal S9, reduces the detection sensitivity of the target haze signal S8 caused by the sample 1. Noise that reduces the detection sensitivity of the haze signal can be generated, for example, by air fluctuations, fluctuations in the laser beam emission direction, brightness changes, and other factors unrelated to the rotation of the sample 1. It is also possible that noise is generated by the chucking of the sample 1 on the stage G, deformation of the sample surface in the Z direction due to distortion of the sample 1, or crystallization / polishing. For example, if the cause of the noise is unrelated to the rotation of the sample 1, differences in the rotation speed of the sample 1 or the scanning mode (constant linear velocity (CLV) / constant angular velocity (CAV)) do not affect the temporal frequency at which noise occurs. Furthermore, noise caused by fluctuations in the height of the sample surface due to the sample 1 being held by a chuck does not occur concentrically but occurs locally (for example, near the chuck).

[0047] -Examples of Filter Processing- There are several possible methods for the filter processing by the filter processing unit C5. In this embodiment, four examples of some methods will be described in order: (1) filter processing defined in one-dimensional time, (2) filter processing for points sampled in one-dimensional space, (3) two-dimensional spatial filter processing, and (4) average value filter processing in units of revolutions.

[0048] (1) Filtering defined in one dimension of time First, the processing of the filter processing unit C5, which reduces the signal intensity of noise occurring in a frequency band equal to or greater than fl1 and equal to or less than fh1 based on the time waveform of the first low-frequency component in this embodiment, relative to the intensity of the haze signal of interest, will be described. Figure 9 is a diagram illustrating the flow of a series of processes for reducing the intermediate frequency component relative to the second low-frequency component and the second high-frequency component, for the second low-frequency component less than fl1, the intermediate frequency component equal to or greater than fl1 and equal to or less than fh1, and the second high-frequency component equal to or greater than fh1, which are included in the first low-frequency component in this embodiment.

[0049] The series of processes described in FIG. 9 are executed by signal processing device 200. Frequency separation unit C1 obtains a digital signal from the output signal of scattered light sensor Bn-2, discretized at a sampling pitch based on a predetermined rule on the sample surface. As shown in FIG. 9, a first low-frequency component signal IN extracted from the output signal of scattered light sensor Bn-2 by frequency separation unit C1 is input to filter processing unit C5. Here, data acquired at regular time intervals is considered. For example, this would be a detection signal synchronized with the laser oscillation frequency. FIG. 9 illustrates a frequency spectrum S91 of first low-frequency component signal IN.

[0050] The second low-frequency component lower than f11 and the second high-frequency component higher than fh1 are Haze signals derived from sample 1. On the other hand, the intermediate-frequency components equal to or higher than f11 and equal to or lower than fh1 are noise components, such as concentric noise signals resulting from fluctuations in the emission direction of the laser light. These intermediate-frequency components are extracted from the first low-frequency component signal IN by a band-pass filter (BPF) C5A. The band-pass filter C5A can be configured using known techniques, such as a process for extracting intermediate-frequency components equal to or higher than f11 and equal to or lower than fh1 from the first low-frequency component signal IN, which has been Fourier-transformed from a time waveform signal to a frequency spectrum S91, or a filter process using a convolution operation.

[0051] The intermediate frequency components extracted by the band-pass filter C5A are subtracted from the first low frequency component signal IN by a subtraction unit C5B. As a result, the intermediate frequency components above fl1 and below fh1 are reduced in the first low frequency component signal IN relative to the second low frequency component and second high frequency component. The first low frequency component signal IN with the reduced intermediate frequency components is inverse Fourier transformed into a time waveform signal. FIG. 9 illustrates a frequency spectrum S92 of the first low frequency component signal IN with the reduced intermediate frequency components relative to the second low frequency component and second high frequency component.

[0052] It is also possible to perform a process of multiplying the first low frequency component signal IN, which is input to the subtraction unit C5B without passing through the band pass filter C5A, by an amplification factor of 1 or more, a process of multiplying the first low frequency component signal IN, which is input to the subtraction unit C5B through the band pass filter C5A, by an amplification factor of 1 or less, or both of these processes. The output signal of the subtraction unit C5B becomes the output signal of the filter processing unit C5.

[0053] FIG. 10 shows an example of a method for determining the boundary frequencies f11 and fh1 used in the series of processes shown in FIG. 9 . The boundary frequencies f11 and fh1, which define the frequency range in which noise occurs, i.e., the frequency range in which the sensitivity of the haze signal of the sample 1 is reduced, are calculated from a frequency spectrum S102 obtained by Fourier transforming a time waveform signal S101 of the noise-containing first low-frequency component signal IN by the Fourier transform unit C5C. The boundary frequencies f11 and fh1 can be manually set by an operator based on the frequency spectrum S102 so as to satisfy the condition f11<fh1<fc. However, there are restrictions on the values ​​that can be set for the boundary frequencies f11 and fh1. For example, f11 is set to 5 Hz or more and fh1 is set to 10 kHz or less. Alternatively, the boundary frequencies f11 and fh1 can be automatically calculated based on multiple haze inspection data sets obtained by varying the scanning speed and R-direction pitch using known techniques such as machine learning, and the boundary frequencies f11 and fh1 can be set accordingly. When machine learning is applied, for example, there is a method in which accumulated past haze maps are input, and the frequency band of the noise portion is learned from the noise portion recognized from the image and its frequency spectrum.

[0054] FIG. 11 is a conceptual diagram illustrating an example in which the boundary frequencies f11 and fh1 used in the series of processes shown in FIG. 9 are varied for each region on the sample surface. For example, in the outer peripheral region 11-2 of the sample 1 held by the chuck mechanism, noise with a frequency band different from the frequency of noise resulting from fluctuations in the emission direction of the laser beam may occur at a constant angular interval. In this case, as shown in FIG. 11, it is possible to set different boundary frequencies f11 and fh1 for the inner peripheral region 11-1 and the outer peripheral region 11-2 of the sample 1. For example, the boundary frequencies f11 and fh1 can be set for the inner peripheral region 11-1 to extract and reduce concentric noise, and for the outer peripheral region 11-2, boundary frequencies f11 and fh1 different from those for the inner peripheral region 11-1 can be set to extract and reduce noise resulting from contact with the chuck mechanism.

[0055] (2) Filtering of Points Sampled in One-Dimensional Space Next, we will discuss the case where one-dimensional spatial filtering is performed by the filtering unit C5. For example, the stage G (spindle stage) that rotates the sample 1 is equipped with an encoder that outputs a synchronization pulse signal at regular intervals, sampling a signal synchronized with the encoder pulse. In this case, the sampling pitch p (the distance between adjacent sampling points) in the θ direction varies depending on the position in the R direction on the sample surface. Furthermore, the control device D controls the stage G so that the scanning speed v for spirally scanning the sample 1 varies depending on the radial position (R coordinate) of the sample 1. In this case, to perform filtering on the entire region of the sample 1 for intermediate frequency components between f11 and fh1, the filter coefficient sequence used for filtering is varied depending on the position on the sample surface (R coordinate) in response to at least one of the sampling pitch p and the scanning speed v of the digital signal related to the first low-frequency component obtained by the frequency separation unit C1. Examples of the filter coefficient sequence include the number of data points to be filtered.

[0056] FIG. 12 shows an example of a configuration in which a θ-direction filter coefficient sequence calculation unit 12-4 calculates a θ-direction filter coefficient sequence based on scanning conditions 12-1, such as scanning speed v and sampling pitch p, and filter frequencies 12-2, consisting of fl1 and fh1. A filter processing unit 12-5 filters a first low-frequency component signal 12-3, and outputs a filtered first low-frequency component signal 12-6. If the number of sampling points per rotation synchronized with the encoder pulse is ec, the radial position of the sampling point is r, and the scanning speed is v, the sampling pitch p is 2πr / ec, and the sampling time interval in the θ direction is p / v. Therefore, the time frequency of the filtered data varies in proportion to v / p with respect to the number of data points within the filter processing region. When sampling at fixed angles, in CLV mode, the sampling time interval is not constant but proportional to 1 / v. In CLV mode, the time frequency of the filtered data varies in proportion to v with respect to the number of data points within the filter processing region. The filter coefficient sequence is determined based on this sampling time interval. When a signal synchronized with an encoder pulse is used as the data acquisition timing in spiral scanning, the same time-frequency filtering can be performed by performing filtering with the same number of sampling points over the entire surface of the sample 1.

[0057] As with the first example, the frequency 12-2 where filtering is performed can be set manually using boundary frequencies f11 and fh1 that define the frequency band in which noise occurs. Alternatively, the boundary frequencies f11 and fh1 can be automatically set by calculating the frequency range in which noise occurs using a known technique such as machine learning based on the signal waveform acquired in sampling pitch units.

[0058] FIG. 13 is a conceptual diagram of a band-pass filter that extracts intermediate frequency components. FIG. 13 illustrates an example in which a band-pass filter that extracts a predetermined frequency component in the filter processing unit C5 is generated by combining low-pass filters with different cutoff frequencies. The band-pass filter 13-3 shown in FIG. 13 is applicable to the filter processing of FIGS. 9 and 12 and the filter processing described below. The band-pass filter 13-3 combines a low-pass filter 13-1 with a high cutoff frequency and a low-pass filter 13-2 with a low cutoff frequency. The cutoff frequency of the low-pass filter 13-1 corresponds to the boundary frequency fh1, and the cutoff frequency of the low-pass filter 13-2 corresponds to the boundary frequency fl1. This band-pass filter 13-3 can extract a predetermined frequency component (intermediate frequency component) by taking the difference between the signals obtained by applying the low-pass filters 13-1 and 13-2 to the input signal.

[0059] (3) Two-dimensional spatial filtering Next, we will explain two-dimensional spatial filtering of a two-dimensional Haze map calculated from the sampled data points of the first low-frequency component. In this example, the filtering unit C5 uses a two-dimensional spatial filter that takes into account filter coefficient sequences in the R and θ directions.

[0060] 14 is a diagram showing an example of using a two-dimensional spatial filter as a filter that reduces intermediate frequency components between fl1 and fh1 relative to the second low frequency component and the second high frequency component. The haze map shown in FIG. 14 shows, as an example, two-dimensional spatial filters F14-1 and F14-2 (applied filter regions). Let the translation direction of the spirally scanned sample 1 be the R direction, the rotation direction of the sample 1 that is perpendicular to the R direction be the θ direction, and let fl2 and fh2 be boundary frequencies in the R direction corresponding to the boundary frequencies fl1 and fh1 in the θ direction. In the haze map calculated from the first low frequency component, the boundary frequencies fl2 and fh2 have higher spatial frequencies than the boundary frequencies fl1 and fh1.

[0061] The R-direction and θ-direction filter coefficient sequences of the two-dimensional spatial filter are calculated according to the flow shown in the block diagram of Fig. 15. A filter coefficient sequence is calculated in an R-direction / θ-direction filter coefficient sequence calculation unit 15-4 based on a θ-direction filter frequency 15-1 and an R-direction filter frequency 15-2, and a scanning condition 15-3. Using this filter coefficient sequence, a filter process is performed in the R-direction and θ-direction in a filter processing unit C5 in the same manner as in Fig. 12, and intermediate frequencies are extracted and reduced from a two-dimensional map related to the first low-frequency component signal.

[0062] For the R-direction filter coefficient sequence, if the radius of the sample 1 is R, the scanning speed in the region of interest is v, and the rotation speed is P [rpm], the time interval per revolution at the same θ coordinate is 2πR / v in CLV mode and 60 / P in CAV mode. Taking this into consideration, the R-direction / θ-direction filter coefficient sequence calculation unit 15-4 calculates the R-direction filter coefficient sequence in the same manner as the θ-direction filter coefficient sequence is calculated by one-dimensional spatial sampling, depending on the scanning mode.

[0063] For example, in the R direction, noise signals generated at a specific time frequency occur at the same R-direction pitch at any position in the CAV mode because the scanning time per revolution, 60 / P, is constant regardless of the position on the sample surface. On the other hand, in the CLV mode, the time required for one revolution of the scan becomes shorter the closer to the center of the sample 1, and the R-direction pitch of the noise signal tends to be coarser in the inner peripheral region closer to the center than in the outer peripheral region of the sample 1, and the R-direction filter coefficient sequence is larger in the inner peripheral region than in the outer peripheral region.

[0064] The filter coefficient sequence in the θ direction can be determined in the same way as in the case of one-dimensional spatial sampling.

[0065] In the case of CAV mode, the sampling time interval in the θ direction is constant and the scanning speed v increases as one approaches the outer periphery of the sample 1, so that the two-dimensional spatial filter F14-2, which is located closer to the outer periphery than the two-dimensional spatial filter F14-1, becomes spatially longer in the θ direction than the two-dimensional spatial filter F14-1, as shown in Fig. 14. In contrast, in the case of CLV mode, the scanning speed (angular velocity) is slower at the position of the two-dimensional spatial filter F14-2 on the outer periphery than at the position of the two-dimensional spatial filter F14-1, and the length of the two-dimensional spatial filter F14-2 in the θ direction is shorter than that of the two-dimensional spatial filter F14-1.

[0066] - (4) Average value filter processing per rotation - Figure 16 is a diagram showing an example of performing average value filter processing of the Haze signal intensity for each rotation of the spiral scan in this embodiment. In this example, a filter that calculates the average value of the signal for each rotation, such as the i-th rotation 16-1 or the j-th rotation 16-2, is used as the configuration of the filter processing unit C5 for the two-dimensional signal map created from the sampled data points. The average value filter performs convolution processing of the input signal time waveform with a rectangular function corresponding to the filter processing region time interval ΔT. For example, in the case of a rotation speed P [rpm], the average value filter corresponds to a rectangular function filter with a time interval of ΔT = 60 / P.

[0067] FIG. 17 shows an example of the frequency response of an average filter for each revolution. The frequency response of an average filter exhibits frequency characteristics in which the gain is strongest for the dc component and decreases at frequencies near 1 / ΔT. Therefore, this average filter can also be considered a type of time-frequency filter. Because average filters can relatively emphasize signals in a frequency domain sufficiently smaller than 1 / ΔT, they can also be used as an example of the low-pass filters 13-1 and 13-2 shown in FIG. 13. For example, if concentric noise occurs at a predetermined xth revolution, the noise component occurring at the x-1th revolution can be extracted by calculating the difference between the average filter located on the inner side of the xth revolution (e.g., the x-1th revolution) and the average filter located on the outer side of the xth revolution (e.g., the x+1th revolution).

[0068] The number of haze data points in the i-th rotation 16-1 is m, and the k-th haze signal intensity among the m data points is Haze ｉ If (k), the average value of the haze signal intensity of the ith rotation 16-1 can be calculated by the following formula.

[0069]

[0070] This formula allows the calculation of the average value of all frequency components of the haze signal for one scanning revolution. The average value can be calculated in the same way for the jth revolution 16-2. When an average filter is applied to an input signal, signals occurring at frequencies sufficiently lower than 1 / ΔT are relatively emphasized and output. Therefore, the signal obtained by removing the average value of each revolution as an offset for each revolution has suppressed signals occurring as noise at frequencies sufficiently lower than 1 / ΔT. Furthermore, if necessary, the reduction in the intensity of the haze signal due to filtering can be suppressed, for example, by adding the average value (dc component) of the haze signal intensity over the entire surface of the sample 1 to the haze signal intensity obtained by removing the average value for each revolution as an offset.

[0071] --Example of Removal of Noise Occurring at the Same Position-- Figure 18 is a diagram illustrating an example of noise removal performed on the same position on multiple haze maps obtained through multiple inspections. Figure 18 illustrates an example in which a noise signal occurring at the same coordinates in each inspection is calculated as the median value for each measurement point of the multiple inspection results, and this is removed as an offset. Specifically, for each coordinate on the sample surface, a statistical value (e.g., median) of multiple reference haze data 18-1 obtained by previously inspecting the same sample 1 multiple times is calculated as a haze base signal 18-2, and the haze base signal 18-2 is subtracted from the haze data obtained by inspecting the sample 1 again. This makes it possible to remove noise occurring at the same position on the sample surface. This process makes it possible to remove haze signals, etc., that occur when the sample 1 comes into contact with the chuck mechanism.

[0072] In this example, the median of multiple reference Haze data 18-1 is used as the Haze base signal 18-2, but other features such as the average value or the most frequent value may be calculated and used as the Haze base signal 18-2, instead of the median.

[0073] - Example of GUI for two-dimensional spatial filtering - Figure 19 is a diagram showing an example of a GUI for setting up filtering to remove noise occurring in a specific frequency range. Figure 19 shows an example of a GUI for selecting the filtering to be applied, applying the selected processing to raw haze data, and displaying a post-filtering haze map and components removed by the filtering. Specifically, when a data path 19-1 on a database DB containing raw haze data is specified, a raw data haze map 19-2 before filtering is displayed. S8 in the raw data haze map 19-2 is the haze signal of interest, and it is desired to reduce the concentric noise N to make the haze signal S8 more apparent.

[0074] To visualize the haze signal S8, first, the scanning mode used to acquire the raw data haze map 19-2 is selected in the scanning mode 19-3, and the filter processing to be applied is set in the filter processing settings 19-4. In the filter processing settings 19-4, the boundary frequencies fh1 and fl1 of the time frequency filter that removes noise N can be set for the R direction and the θ direction, respectively. By setting these boundary frequencies fh1 and fl1, the corresponding filter coefficient strings are calculated. After setting the boundary frequencies fh1 and fl1, clicking the haze filter execution button 19-5 displays the haze map 19-6 to which the frequency filter has been applied.

[0075] Execution of a filter by setting boundary frequencies f11 and fh1 in the θ direction corresponds to (1) a filter process defined in one-dimensional time, or (2) a filter process performed on points sampled in one-dimensional space. Execution of a filter by setting boundary frequencies f11 and fh1 in the θ direction and R direction corresponds to (3) a two-dimensional spatial filter process.

[0076] In addition, in the filter processing setting 19-4, in addition to setting the boundary frequencies fh1 and fh1, at least one filter to be applied can be selected from "removal of average value per revolution" and "removal of median value from multiple tests." Removal of average value per revolution corresponds to the average value filter processing per revolution (4) described in FIG. 16. Removal of median value from multiple tests corresponds to the filter for removing statistics from multiple tests described in FIG. 18. Two-dimensional spatial filters including these can be executed in addition to or instead of the filter processing executed by setting the boundary frequencies fh1 and fh1.

[0077] - Independence of integration processing of haze inspection and defect inspection - Figure 20 is a diagram illustrating an example of the flow of independent integration processing performed on the first low-frequency component and the first high-frequency component in this embodiment. As can be seen from the explanation so far, in the defect inspection device and defect inspection method of this embodiment, the processing of processing the first high-frequency component to detect defects and the processing of processing the first low-frequency component to calculate a haze signal are independent of each other. This will be explained below using Figure 20.

[0078] The signal processing device 200 integrates first low-frequency components 20-3 obtained from each of the plurality of scattered light sensors Bn-2 based on second integration conditions that are set independently of first integration conditions for integrating first high-frequency components 20-2 obtained from each output signal 20-1 of the plurality of scattered light sensors Bn-2 by a frequency separation unit C1 (not shown in FIG. 20). The first and second integration conditions are set in advance and stored, for example, in a storage device of the signal processing device 200.

[0079] A defect candidate extraction unit C2 (not shown in FIG. 20 ) extracts signals at coordinates of defect candidates that exceed a predetermined high-sensitivity threshold from the first high-frequency components 20-2. The signals associated with the extracted defect candidates are subjected to a first integration process based on first integration conditions in a defect integration processing unit C4, which calculates an integrated signal I1 by integrating the first high-frequency components for each scattered light sensor Bn-2 at the coordinates of the defect candidates. A defect detection unit C6 compares the signal strength of the integrated signal I1 for each scattered light sensor Bn-2 and the signal on which it is based with a predetermined low-sensitivity threshold, and defect candidates that exceed the low-sensitivity threshold are detected as defects. The result is output as a final defect integrated signal 20-4 to, for example, a display device F.

[0080] In parallel with this, the first low-frequency components obtained from the output signals of the multiple scattered light sensors Bn-2 undergo second integration processing in the haze integration processor C3, while including noise signals (e.g., signals related to concentric noise), based on second integration conditions set independently of the first integration conditions. As a result, an integrated signal I2 (e.g., time waveform signal S101 in FIG. 10 ) obtained by integrating the first low-frequency components is calculated for each scattered light sensor Bn-2. As an example of the second integration condition, it is possible to superimpose signals obtained by applying an amplification factor set individually for each scattered light sensor Bn-2 to the first low-frequency components of the output signals of each scattered light sensor Bn-2. As described above with several examples, the noise signals of the integrated signal I2 are reduced in the filter processor C5, and the integrated signal is finally output as a haze integrated signal 20-5 to, for example, a display device F.

[0081] Figure 21 is a diagram illustrating another example of the flow of performing independent integration processing on the first low-frequency component and the first high-frequency component in this embodiment. In this example, the processing of the first high-frequency component is the same as in the example of Figure 20, and therefore is not shown in Figure 21. Unlike the example of Figure 20, Figure 21 illustrates a configuration in which filter processing is performed on the first low-frequency component of each scattered light sensor Bn-2 before the haze integration processor C3.

[0082] The output signal 21-1 of each scattered light sensor Bn-2 is separated into a first high-frequency component 21-2 and a first low-frequency component 21-3 by a frequency separation unit C1 (not shown in FIG. 21). The low-frequency component 21-3 of each scattered light sensor Bn-2 is subjected to filtering in a corresponding filter processing unit C5-n (n = 1, 2, ...) to reduce, for example, concentric circular noise, and then subjected to a second integration process in a haze integration processing unit C3. The first low-frequency component I2 integrated by the haze integration processing unit C3 is output to, for example, a display device F as a final haze integrated signal 21-4.

[0083] - Example of Simultaneous Detection of Two Polarizations - Figure 22 is a diagram illustrating an example of integration processing of haze signals of two polarizations in which scattered light is simultaneously detected in this embodiment. In this example, an individual amplification factor is set for each of the two polarizations detected simultaneously. The amplification factor applied to the output signal is set individually for each sensor and polarization within the dotted-line frame G22 in the figure for each scattered light sensor Bn-2. The configuration within the dotted-line frame G22 is shown only for sensor 1; the other sensors are similar and therefore not shown. Sensor 1 independently detects the signal intensities of polarization 1 signal 22-2 and polarization 2 signal 22-3. Examples of polarization 1 and polarization 2 include a combination of S-polarized light and P-polarized light. The polarization 1 signal 22-2 and polarization 2 signal 22-3 are subjected to individually set polarization 1 individual channel amplification factor 22-4 and polarization 2 individual channel amplification factor 22-5, respectively, before being separated into a defect signal and a haze signal (including noise) in the frequency separation unit C1. The haze signals of polarization 1 and 2 of each sensor output from the frequency separation unit C1 are multiplied by individually set haze amplification factors 22-6 and 22-7 for polarization 1 and polarization 2, respectively, and then undergo a second integration process in a haze integration processor C3 to become an integrated first low-frequency component I2. Although not shown in Fig. 22, in this example as well, filtering is performed after or before the second integration process, as in the examples of Figs. 20 and 21.

[0084] In the case of simultaneous detection of two polarizations, not only signal intensity but also phase information is obtained as a haze integrated signal. Therefore, the haze map display GUI 22-7 displayed on the display device F can output not only a haze intensity map but also a haze phase map. These intensity maps and phase maps can be compared with intensity map and phase map models for each haze type stored in a database DB, allowing the type of observed haze to be automatically determined. Furthermore, based on the obtained intensity map and phase map, the values ​​of the individual channel amplification factor and haze amplification factor within the dotted frame G22 can be manually or automatically adjusted so as to improve the sensitivity of the haze signal of interest.

[0085] --Setting Example of Two-Dimensional Spatial Filtering for Each Type of Haze Signal of Interest-- Figure 23 shows an example of a GUI for setting filtering and integration processes suitable for observing a type of haze signal of interest. In Figure 23, the address where the raw haze data is located is specified in the data path 23-1, the type of haze signal to be observed is specified in the haze type 23-2, and the sensor for displaying the detected haze map is set in the sensor setting section 23-3. A two-dimensional haze map consisting of raw data of the first low-frequency component of the sensor set in the sensor setting section 23-3 is then displayed in the haze map display section 23-4. Clicking the processing execution button 23-5 in this state displays an integrated two-dimensional haze map obtained by performing filtering and integration processes on the signals from each sensor under the current settings in the haze map display section 23-7. At the same time, a recipe setting display section 23-6 is displayed on the screen for setting filtering and integration processes that will reveal the haze signal corresponding to the selected haze type 23-2.

[0086] In the recipe setting display unit 23-6, the conditions for the filtering and integrating processes can be set according to the type of haze signal of interest. Furthermore, the conditions can be manually changed by viewing the haze map displayed on the haze map display unit 23-7 after the integrating process, or by switching to and displaying a haze map based on raw haze data for each sensor on the haze map display unit 23-4. Furthermore, when inspecting for defects in the shape of the sample surface, such as roughness, based on the haze signal, as in the example of FIG. 24 described below, the results of the defect inspection can be displayed on the screen of FIG. 23, and the conditions can be changed to more clearly show the haze signals of defective areas detected in the inspection results, and the two-dimensional haze map can be redisplayed.

[0087] Specifically, in the recipe setting display unit 23-6, it is possible to check the scanning mode (CLV mode / CAV mode), the filter frequencies f11 and fh1 set for the R direction and the θ direction, respectively, as well as the amplification factor set for each polarization of each sensor, a two-dimensional spatial filter, etc. Based on the haze map after application of the filter and integration processes displayed in the haze map display unit 23-7, by manually changing the recipe settings for the filter processing and integration processes in the recipe setting display unit 23-6 and clicking the recipe update button 23-8, a haze map processed with the modified settings is displayed in the haze map display unit 23-7. The signal processing device 200 can also be configured to perform machine learning to determine settings for suppressing signals such as noise N relative to the haze signal intensity of interest based on the settings for the filter processing and integration processes when displaying the haze map and the haze map after processing, and to automatically update the recipe settings.

[0088] Figure 24 shows an example of automatic defect detection based on a haze signal. Figure 24 illustrates a case where defects are detected based on an integrated haze signal obtained by integrating the filtered first low-frequency components of each scattered light sensor Bn-2. Unlike defect inspection for foreign matter, scratches, etc., based on the first high-frequency component, defect detection based on the first low-frequency component is an inspection for defects in the shape of the sample surface, such as roughness. In addition to the integrated haze signal, various other haze-related signals obtained by the scattered signal processing unit C, such as an individual haze signal from any filtered scattered light sensor Bn-2 (e.g., scattered light sensor Bn-1), can also be used as the basis for defect detection.

[0089] In the example of FIG. 24 , the scattered light signal processing unit C includes a defect detection unit 24-2, a feature extraction unit 24-3, and a defect extraction unit 24-4. When an integrated haze signal 24-1 is input to the defect detection unit 24-2, the defect detection unit 24-2 compares the integrated haze signal 24-1 with a predetermined threshold, and extracts defect candidates having a signal intensity exceeding the threshold. The feature extraction unit 24-3 calculates feature information such as signal intensity, contrast, size, aspect ratio, and scattering distribution for the extracted defect candidates. The defect extraction unit 24-4 determines whether the defect candidate is a shape defect or other defect based on the feature information. If the defect candidate is determined to be a defect, the defect is displayed in a haze map display unit 24-5 on the display screen of the display device F. The defect extraction unit 24-4 can perform defect determination based on a preset determination algorithm, or it can automatically perform defect detection based on known techniques such as machine learning based on defect information stored in a database DB. The haze map display unit 24-5 can display only the haze signal S8 determined to be defective, or can output a background signal such as concentric noise N together with the signal and highlight the position of the haze signal S8 of interest by coloring it or surrounding it with a dotted line. The signal determined to be defective is displayed on the haze map display unit 24-5, and the defect data is output from a defect data output unit 24-6 (for example, a communication device that outputs data to a network, a recording device that records data on a storage medium, etc.). Note that, although an example has been described above in which a haze map is displayed and data is output for signals determined to be defective on the sample surface, a configuration in which a haze map is displayed and data is output for defect candidates detected by the defect detection unit 24-2 can also be used.

[0090] -Effects- (1) In this embodiment, the frequency separation unit C1 removes high-frequency components derived from minute defects and the like from the scattered light intensity signals obtained by each sensor to obtain first low-frequency components. This makes it possible to extract components that generate scattered signals over a wide spatial range, such as the roughness of the sample 1. Because the first low-frequency components contain not only signals derived from the roughness of the sample 1 but also noise (such as concentric noise), further filtering is performed to reduce intermediate-frequency components derived from noise and the like relative to the second low-frequency and second high-frequency components derived from the sample 1. By relatively reducing the signal intensity of the noise components in this way, noise signals generated at specific frequency components, such as the signal of the concentric noise component, can be reduced, enabling the haze signal of interest to be detected with high sensitivity.

[0091] (2) If filtering is performed in units of a predetermined number of pixels in the scanning direction, the time required to scan a predetermined number of pixels varies depending on the distance from the center of the sample 1, so the frequency of the noise to be removed varies depending on the position of the sample 1, and it is not possible to remove a specific frequency across the entire surface of the sample. When scanning the sample 1 in a constant angular velocity scanning mode (CAV), the scanning speed becomes faster toward the periphery of the sample 1, and the frequencies removed by filtering in units of a predetermined number of pixels also become higher. When scanning in a constant linear velocity scanning mode (CLV), the scanning speed becomes slower toward the outer regions of the sample 1 within the CLV region, and the frequencies removed by filtering in units of a predetermined number of pixels also become lower. Therefore, when the filtering range is specified by the number of pixels, it is necessary to adjust the number of pixels used in filtering depending on the position of the sample 1 in order to remove a specific frequency.

[0092] In contrast, in this embodiment, since filtering is performed on specific time frequency components, it is possible to reduce the specific frequency components over the entire surface of the sample 1 .

[0093] (3) As with defect detection, in order to improve the sensitivity of haze detection, haze and noise are distinguished using a signal obtained by integrating signals from multiple sensors. Conventionally, integration conditions that maximize the sensitivity of detecting minute defects have been uniformly applied to haze signals as well. Integration conditions include superimposing signals from each sensor using a predetermined amplification factor, or superimposing signals from a predetermined combination of sensors. However, because the scattering distribution of minute defects and that of haze are different, it is not appropriate to use the same integration conditions, such as amplification factors, for haze as for minute defects.

[0094] In contrast, in this embodiment, the first integration condition for the output signals of each sensor for integrating defect signals and the second integration condition for the output signals of each sensor for integrating haze signals are set independently, which allows desirable settings of integration conditions such as amplification factors so that the haze signal of interest can be detected with high sensitivity.

[0095] (4) Since the Haze map to which the filtering and integration processes have been applied is displayed, an operator such as a user can visually confirm information such as the state of the sample surface.

[0096] Furthermore, the settings used in the filtering and integration processes can be adjusted manually or automatically according to the output haze map. For example, if strong concentric noise occurs in a particular sensor, and the concentric noise also appears strongly in the results after integration, the integration conditions can be adjusted manually or automatically, such as by lowering the amplification factor of the sensor that is producing strong concentric noise, thereby improving the sensitivity of the haze signal of interest.

[0097] Furthermore, the signal processing device 200 can be configured to automatically adjust recipe settings by, for example, performing machine learning on the output haze map and learning using the intensity of concentric noise as an index. By viewing the haze map output with the adjusted recipe settings, an operator such as a user can confirm the validity of the adjusted recipe settings and whether adjustment is necessary. This allows for automatic recipe settings for optimal haze detection, taking into account differences in optical conditions and other machine differences.

[0098] - Additional remarks - As in the embodiment, the defect inspection method can be realized by computer processing. Therefore, the defect inspection device according to the present invention can be configured by adding a processing device (corresponding to signal processing device 200) in which the defect inspection method is programmed to an existing defect inspection device equipped with inspection system 100. The defect inspection device according to the present invention can also be configured by programming the defect inspection method in signal processing device 200 of an existing defect processing device equipped with inspection system 100.

[0099] Furthermore, although the defect inspection apparatus including the inspection system 100 has been described, the present invention can also be applied to a defect inspection apparatus including two or more types of inspection systems. In this case, the signal obtained for each inspection system is separated into a first low-frequency component and a first high-frequency component by a frequency separation unit, and a haze signal is obtained for each inspection system by performing filtering and integration processing on the first low-frequency component. For example, if there is a difference in the haze signals, such as one type of haze signal being easily detected by the inspection system 100 and another type of haze signal being easily detected by an inspection system provided separately from the inspection system 100, it is possible to use haze signals from different inspection systems for each haze signal of interest.

[0100] 1...sample, 200...signal processing device, A...scattered light inspection illumination unit (illumination optical system), Bn...scattered light detection optical system (detection optical system), Bn-2...scattered light sensor (sensor), C...signal processing unit, C1...frequency separation unit, C2...defect candidate extraction unit, C3...haze integration processing unit, C4...defect integration processing unit, C5...filter processing unit, C6...defect detection unit, D...control device, DB...database, F...display device (output device), G...stage, p...sampling pitch, v...scanning speed

Claims

1. An optical system for supplying illumination light to a sample; an optical system for detecting scattered light scattered from an irradiation area on the sample illuminated with the illumination light; a sensor for converting the scattered light detected by the optical system into an electrical signal; a stage for placing the sample on the stage; a control device for controlling the rotation of the stage; and a signal processing device for processing an output signal from the sensor obtained by spirally scanning the sample using the control device to control the rotation of the stage, wherein the signal processing device reads boundary frequencies fc, fl1, and fh2 which are time frequencies of the output signal from the sensor and which are set to satisfy the relationship fl1<fh1<fc, separates the output signal from the sensor into a first high-frequency component which is a time frequency component equal to or greater than the boundary frequency fc, and a first low-frequency component which is a time frequency component less than the boundary frequency fc, and processes the first high-frequency component to detect defects, a haze signal obtained by filtering the second low-frequency component and the second high-frequency component, the haze signal being a first low-frequency component including a second low-frequency component that is less than the boundary frequency fl1, an intermediate-frequency component that is equal to or greater than the boundary frequency fl1 and less than the boundary frequency fh1, and a second high-frequency component that is equal to or greater than the boundary frequency fh1; and a haze map of the sample based on the haze signal, the haze map being output to an output device.

2. A defect inspection device according to claim 1, characterized in that the process of detecting defects by processing the first high frequency component and the process of calculating the haze signal by processing the first low frequency component are independent of each other.

3. A defect inspection device according to claim 1, wherein the boundary frequency fl1 is 5 Hz or more, and the boundary frequency fh1 is 10 kHz or less.

4. A defect inspection device according to claim 1, comprising a plurality of the sensors, and wherein the signal processing device integrates the second low frequency components obtained from the plurality of sensors based on second integration conditions set independently of first integration conditions for integrating the first high frequency components obtained from the plurality of sensors.

5. A defect inspection device according to claim 4, wherein the first integration condition and the second integration condition include a combination of sensors for integrating signals and an amplification factor of the signal for each sensor.

6. The defect inspection device according to claim 5, characterized in that the first integration condition and the second integration condition include an amplification factor for a signal obtained by integrating output signals from two or more sensors at different positions.

7. A defect inspection device according to claim 5, wherein the illumination optical system is capable of selecting at least one polarization of illumination to irradiate the sample, and the first integration condition and the second integration condition are different for each polarization.

8. A defect inspection device according to claim 4, wherein conditions for the filtering process and the integrating process can be set according to the type of the haze signal.

9. A defect inspection device according to claim 8, wherein after the filtering process and the integration process based on the conditions are performed and a two-dimensional Haze map is displayed, the settings of the conditions can be changed.

10. A defect inspection apparatus according to claim 1, wherein the boundary frequencies fl1 and fh1 differ depending on the position on the surface of the sample.

11. A defect inspection device according to claim 1, wherein the control device controls the stage so that the scanning speed at which the sample is spirally scanned varies depending on the radial position of the sample, and the signal processing device inputs a digital signal discretized from the output signal of the sensor at a sampling pitch on the sample, and varies the coefficient sequence of the filter used in the filtering process depending on the position on the sample in accordance with at least one of the sampling pitch of the digital signal and the scanning speed of the stage.

12. A defect inspection device as described in claim 1, wherein the translation direction of the spirally scanned sample is defined as the R direction, and the rotation direction of the sample, which is perpendicular to the R direction, is defined as the θ direction, and the boundary frequencies fl2, fh2 in the R direction corresponding to the boundary frequencies fl1, fh1 in the θ direction have higher spatial frequencies than the boundary frequencies fl1, fh1 in the Haze map calculated from the first low frequency component.

13. A defect inspection method using a defect inspection device comprising: an illumination optical system that supplies illumination light to a sample; a detection optical system that collects scattered light scattered from an irradiation area on the sample that is irradiated with the illumination light; a sensor that converts the scattered light collected by the detection optical system into an electrical signal; and a stage on which the sample is placed, comprising: setting boundary frequencies fc, fl1, and fh2 so that the time frequencies of the output signal of the sensor obtained by spirally scanning the sample satisfy the relationship fl1<fh1<fc; separating the output signal of the sensor into a first high-frequency component that is a time frequency component equal to or greater than the boundary frequency fc and a first low-frequency component that is a time frequency component less than the boundary frequency fc; and processing the first high-frequency component to detect defects. a haze signal obtained by filtering the second low-frequency component and the second high-frequency component, the first low-frequency component including a second low-frequency component that is less than the boundary frequency fl1, an intermediate-frequency component that is equal to or greater than the boundary frequency fl1 and less than the boundary frequency fh1, and a second high-frequency component that is equal to or greater than the boundary frequency fh1; and a haze map of the sample based on the haze signal.

Images