Defect inspection device and defect inspection method

Abstract

Provided is a defect inspection method that uses a plurality of sensors that perform photoelectric conversions on scattered light from a sample surface and output detection signals and a signal processing device that processes the detection signals as inputted from the plurality of sensors. The signal processing device computes filters and gains on the basis of the two-dimensional profiles of the detection signals for defects on the sample surface, uses the filters to perform filter processing with respect to the temporal or spatial change in channel signals obtained from individual sensor channels by scanning the same coordinates on the sample surface multiple times, each sensor channel being the output of the detection signal of a single sensor of the plurality of sensors or the integrated output of the detection signals of a portion of the sensors, outputs a filter-processed signal for each of the sensor channels, multiplies the filter-processed signal for each of the sensor channels by the corresponding gain and then integrates the filter-processed signals, outputs defect evaluation values for a defect type to be detected, and uses a threshold value to detect defects on the basis of the defect evaluation values.

Description

Defect inspection device and defect inspection method

[0001] The present invention relates to a defect inspection device and a defect inspection method for inspecting a sample surface and outputting the position, type, dimensions, etc. of defects (including foreign matter).

[0002] In order to improve the yield of products such as semiconductor substrates and thin film substrates, semiconductor substrates, thin film substrates, etc. manufactured on a production line are used as samples, and defects on the sample surface are inspected. As a defect inspection device used for this inspection, there is known one that simultaneously detects scattered light from the surface of a sample (hereinafter, the sample surface) by a plurality of detection systems at different positions and obtains detailed data on the position, shape, size, etc. of defects (see Patent Document 1, etc.). According to the defect inspection device of Patent Document 1, by simultaneously detecting scattered light from an illumination spot by a plurality of detection systems having different directions with respect to the illumination spot, a lot of data on defects can be obtained.

[0003] Japanese Patent No. 5564807

[0004] For example, various solutions used in the semiconductor manufacturing process may contain extremely small foreign matters of 15 nm or less. In recent years, defect inspection devices have been increasingly required to have the detection performance for defects (hereinafter, micro-defects) that are sufficiently small with respect to the wavelength of such illumination light. Since the scattered light from a defect is proportional to approximately the sixth power of the defect size, the amount of detection light from the defect rapidly decreases due to the miniaturization of the defects to be inspected.

[0005] In the detection of important micro-defects of 15 nm or less, the roughness scattered light generated by minute irregularities on the sample surface can be an obstacle. The roughness scattered light obtained by applying illumination light to a mirror-finished wafer with a polished surface is observed as shot noise due to its statistical variation when detected by a sensor. As a result, the scattered light from the defect (hereinafter, defect detection light) is masked by noise, making high-speed inspection difficult.

[0006] In contrast, the defect inspection apparatus of Patent Document 1 scans the sample by irradiating it with illumination light, detects the light from the illumination spot on the sample and converts it into an electrical signal, and extracts a signal with a certain time width from the high-frequency components obtained by removing the low-frequency components from this electrical signal. Then, from the extracted signals, those with a signal intensity above a certain level are selected and used as defect detection signals to calculate the size of foreign matter or defects present on the sample surface or within the sample. The certain time width is calculated based on the half-width of the defect detection signal, which is determined from the rotation speed of the stage that moves the sample, the coordinate position in the rotation direction of the illumination spot, and the size of the illumination spot. By extracting data only from signals for which the period during which the signal intensity is above a threshold is greater than or equal to a certain time width, low-level noise signals and the like are removed, and noise and defect detection signals are distinguished.

[0007] However, the defect inspection apparatus in Patent Document 1 lacks sufficient consideration for the statistical variability of the signal itself due to a decrease in signal intensity. As defects become smaller, the phenomenon of defect detection light being lower than the amount of scattered light from a sample surface without defects is occurring more frequently. In the defect inspection apparatus in Patent Document 1, signals whose signal intensity is above a threshold for a certain period of time or longer are extracted as defect detection signals. Therefore, if the signal intensity is not above the threshold, it is not extracted as a defect detection signal, and it is difficult to extract defect detection light with a light intensity lower than the amount of scattered light from the sample surface as a defect detection signal.

[0008] The object of the present invention is to provide a defect inspection device and a defect inspection method that can detect various defects, including minute defects, with high sensitivity.

[0009] To achieve the above objective, the present invention provides a defect inspection apparatus comprising: an illumination optical system for irradiating a sample surface with spot-shaped illumination light; a scanning device for scanning the spot-shaped illumination light on the sample surface; a detection system for collecting scattered light from the sample surface; a plurality of sensors for photoelectric conversion of the light collected by the detection system and outputting a detection signal; and a signal processing device for processing the detection signals input from the plurality of sensors. The signal processing device calculates a filter and gain based on a two-dimensional profile of the detection signal of a defect on the sample surface, and uses the output of the detection signal of a single sensor or the integrated output of the detection signals of some of the plurality of sensors as individual sensor channels. It scans the same coordinates on the sample surface multiple times and filters the channel signals obtained from each sensor channel for temporal or spatial changes using the filter, outputs a filtered signal for each sensor channel, multiplies the filtered signals of each sensor channel by a gain corresponding to the defect type to be detected and integrates them to output a defect evaluation value for the defect type to be detected, and detects defects using a threshold value based on the defect evaluation value.

[0010] According to the present invention, various defects, including minute defects, can be detected with high sensitivity.

[0011] This is a schematic diagram of one example configuration of a defect inspection apparatus according to the first embodiment of the present invention. This is a schematic diagram showing an example of the scanning trajectory of a sample by a scanning device provided in the defect inspection apparatus according to the first embodiment of the present invention. This is a schematic diagram showing another example of the scanning trajectory of a sample by a scanning device provided in the defect inspection apparatus according to the first embodiment of the present invention. This is a schematic diagram showing an extracted attenuator provided in the defect inspection apparatus according to the first embodiment of the present invention. This diagram schematically shows the positional relationship between the optical axis of illumination light guided to the sample surface from an oblique direction by an illumination optical system provided in the defect inspection apparatus according to the first embodiment of the present invention and the shape of the illumination intensity distribution, using a cross-section obtained by cutting the sample with a plane perpendicular to the incident plane of illumination light incident on the sample. This diagram schematically shows the positional relationship between the optical axis of illumination light guided to the sample surface from an oblique direction by an illumination optical system provided in the defect inspection apparatus according to the first embodiment of the present invention and the shape of the illumination intensity distribution, using a cross-section obtained by cutting the sample with a plane perpendicular to the incident plane of illumination light incident on the sample and containing the normal to the sample surface. This is a diagram showing the aperture that collects scattered light in the scattered light detection system provided in the defect inspection apparatus according to the first embodiment of the present invention, viewed from above. This is a diagram of the configuration of a scattered light detection system provided in a defect inspection apparatus according to the first embodiment of the present invention, which incidents light onto a sample from the normal direction and collects scattered light from the sample surface. This is a view of Figure 8 from above. This is a top view of the light intensity distribution of scattered light generated by oblique incident illumination on the defect and the sample surface. This is a diagram of the configuration of a Z sensor. This is a processing block diagram of the signal processing according to the first embodiment of the present invention. This is an explanatory diagram of a filter applied in a signal processing device provided in a defect inspection apparatus according to the first embodiment of the present invention. This is an explanatory diagram of a filter applied in the signal processing device according to the first embodiment of the present invention for each positional relationship between the scanning position in the s2 direction and the defect position. This is an explanatory diagram of the input from a plurality of sensors and the processing data output by each processing block in the signal processing device according to the first embodiment of the present invention. This is a diagram showing an example of a defect signal model in the feature space defined by the signals of sensor channels ch1 and chN. This is an explanatory diagram of the similarity between a model of the defect to be detected, represented as a one-dimensional vector, and the detected signal vector. This is an explanatory diagram of the similarity between a model of the defect to be detected, represented as a plane, and the detected signal vector. This is a schematic diagram of the image formation detection system provided in a defect inspection apparatus according to the second embodiment of the present invention.This is an explanatory diagram illustrating the change in image spread of the imaging detection system due to the Z-direction displacement of the sample. This is an explanatory diagram illustrating the change in image spread of the low-angle imaging detection system and the high-angle imaging detection system due to the Z-direction displacement of the sample. This is an explanatory diagram illustrating the spread of the illumination spot due to the Z-direction displacement of the sample. This is an explanatory diagram illustrating the overlap scanning of illumination and the model pattern based thereon according to the second embodiment of the present invention.

[0012] Embodiments of the defect inspection apparatus and defect inspection method of the present invention will be described below with reference to the drawings. The defect inspection apparatus described below as the target of the present invention is used for defect inspection of the surface of a sample (wafer) performed during the manufacturing process of semiconductors, for example. The defect inspection apparatus according to each embodiment is suitable for high-speed execution of the detection of minute defects and the acquisition of data regarding the number, location, dimensions, and type of defects.

[0013] (First Embodiment) - Defect Inspection Apparatus - Figure 1 is a schematic diagram of one example configuration of a defect inspection apparatus according to the first embodiment of the present invention. The defect inspection apparatus 100 according to this embodiment inspects a sample 1 and detects defects such as foreign matter and dents on the surface of the sample 1 (hereinafter referred to as the sample surface), in particular defects of a type appropriate to the inspection purpose. A typical example of the sample 1 is assumed to be a disc-shaped semiconductor silicon wafer having a flat surface on which no pattern is formed. The defect inspection apparatus 100 is composed of a stage ST, an illumination optical system A, a plurality (n) scattered light detection systems B1-Bn, a signal processing device D, a control device E1, a user interface E2, a monitor E3, a Z sensor F, and a secondary storage device DB. The scattered light detection systems B1-Bn each include sensors C1P-CnP and C1S-CnS. When referred to as sensor CiP, it refers to a sensor that detects P-polarized light from the i-th scattered light detection system Bi. Similarly, when referred to as sensor CiS, it refers to a sensor that detects S-polarized light from the i-th scattered light detection system Bi.

[0014] -Stage- The stage ST consists of a sample stage ST1 and a scanning device ST2. The sample stage ST1 is a stage that supports the sample 1. The scanning device ST2 is a device that drives the sample stage ST1 to scan the illumination spot BS over the sample surface, and although not shown in detail in the diagram, it consists of a translation stage, a rotation stage, and a Z stage. The rotation stage is supported by the translation stage via the Z stage, and the sample stage ST1 is supported by the rotation stage. The translation stage moves horizontally together with the rotation stage, and the rotation stage rotates around an axis that extends vertically. The Z stage functions to adjust the height of the sample surface.

[0015] Figure 2 is a schematic diagram showing the scanning trajectory of sample 1 by the scanning device ST2. As will be described later, the illumination light (hereinafter referred to as illumination spot BS) that is irradiated onto the sample surface in a spot shape by the illumination optical system A has an illumination intensity distribution that is long in one direction, as shown in the figure. The direction of the long axis of illumination spot BS is denoted as s2, and the direction intersecting the long axis (for example, the short axis perpendicular to the long axis) is denoted as s1. As the rotating stage rotates, sample 1 rotates and illumination spot BS is scanned in the s1 direction relative to the sample surface. As the translation stage translates, sample 1 moves horizontally and illumination spot BS is scanned in the s2 direction relative to the sample surface. As sample 1 rotates and moves due to the operation of the scanning device ST2, illumination spot BS moves in a spiral trajectory from the center to the outer edge of sample 1, as shown in Figure 2, and the entire surface of sample 1 is scanned. During one rotation of sample 1, illumination spot BS moves in the s2 direction by a distance less than or equal to the length of illumination spot BS in the s2 direction.

[0016] Furthermore, a scanning device can be applied that replaces the rotary stage with another translational stage whose movement axis is extended in a direction intersecting the movement axis of the translational stage in the horizontal plane. In this case, as shown in Figure 3, the illumination spot BS scans the sample surface by overlapping linear trajectories rather than a helical trajectory. Specifically, the first translational stage is driven to translate at a constant speed in the s1 direction, the second translational stage is driven in the s2 direction for a predetermined distance (for example, a distance less than or equal to the length of the illumination spot BS in the s2 direction), and then the first translational stage is again driven to translate in the s1 direction. In this way, the illumination spot BS scans the entire surface of the sample 1 by repeatedly performing linear scanning in the s1 direction and moving in the s2 direction. Compared to this scanning method, the helical scanning method shown in Figure 2 does not involve reciprocating motion, which is advantageous for performing sample inspection in a short time.

[0017] - Illumination Optical System - The illumination optical system A shown in Figure 1 is configured to irradiate the sample 1 placed on the sample stage ST1 with desired illumination light, and includes a group of optical elements. As shown in Figure 1, this illumination optical system A includes a laser light source A1, an attenuator A2, an output light adjustment unit A3, a beam expander A4, a polarization control unit A5, a focusing optical unit A6, reflective mirrors A7-A10, etc.

[0018] The laser light source A1 is a unit that emits a laser beam as illumination light. When detecting minute defects near the sample surface with the defect inspection device 100, the laser light source A1 is used to emit a high-power laser beam of 2W or more in the ultraviolet or vacuum ultraviolet region with a short wavelength (wavelength 355 nm or less) that does not easily penetrate into the interior of the sample 1. The diameter of the laser beam emitted by the laser light source A1 is typically about 1 mm. When detecting defects inside the sample 1 with the defect inspection device 100, the laser light source A1 is used to emit a visible or infrared laser beam with a longer wavelength that easily penetrates into the interior of the sample 1.

[0019] Figure 4 shows a schematic diagram of the attenuator A2. The attenuator A2 is a unit that reduces the light intensity of the illumination light from the laser light source A1. In this embodiment, a configuration combining a first polarizer A2a, a half-wave plate A2b, and a second polarizer 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 to linear polarization by the first polarizer A2a, and then its polarization direction is adjusted to the azimuth angle of the slow axis of the half-wave plate A2b before passing through the second polarizer A2c. By adjusting the azimuth angle of the half-wave plate A2b, the light intensity of the illumination light can be reduced by any ratio. If the degree of linear polarization of the illumination light incident on the attenuator A2 is sufficiently high, the first polarizer A2a can be omitted. Furthermore, the attenuator A2 is not limited to the configuration exemplified in Figure 4; it can also be configured using an ND filter with a gradient density distribution, and the attenuation effect can be adjusted by combining multiple ND filters with different densities.

[0020] - Output Light Adjustment Unit The output light adjustment unit A3 shown in Figure 1 is a unit that adjusts the angle of the optical axis of the illumination light attenuated by the attenuator A2, and in this embodiment it is composed of multiple reflective mirrors A3a and A3b. The configuration is such that the illumination light is reflected sequentially by the reflective mirrors A3a and A3b, but in this embodiment the incident and output surfaces of the illumination light to reflective mirror A3a are configured to be perpendicular to the incident and output surfaces of the illumination light to reflective mirror A3b. The incident and output surfaces are the surfaces that include the optical axis incident on the reflective mirror and the optical axis emitted from the reflective mirror. For example, if a three-dimensional XYZ orthogonal coordinate system is defined and the illumination light is incident on reflective mirror A3a in the +X direction, although it differs from the schematic Figure 1, for example the illumination light is redirected in the +Y direction at reflective mirror A3a and then in the +Z direction at reflective mirror A3b. This is an example where the incident and output surfaces of the illumination light to reflective mirror A3a are the XY plane and the incident and output surfaces to reflective mirror A3b are the YZ plane. Furthermore, although not shown in the diagram, the reflective mirrors A3a and A3b are equipped with mechanisms for translational movement and tilting, respectively. The reflective mirrors A3a and A3b can be translated parallel to themselves, for example, in the incident or outgoing direction of the illumination light, and can also be tilted around the normal to the incident / outgoing surface. This allows for independent adjustment of the offset amount and angle in the XZ plane and the offset amount and angle in the YZ plane of the optical axis of the illumination light emitted from the outgoing light adjustment unit A3 in the +Z direction. In this example, a configuration using two reflective mirrors A3a and A3b is illustrated, but a configuration using three or more reflective mirrors is also possible.

[0021] The beam expander A4 is a unit that expands the diameter of the incident illumination beam and has multiple lenses A4a and A4b. A Galilean type beam expander A4, using a concave lens as lens A4a and a convex lens as lens A4b, can be cited as an example of a beam expander A4. The beam expander A4 is equipped with a mechanism (zoom mechanism) for adjusting the distance between lenses A4a and A4b, and the degree of expansion of the light beam diameter changes by adjusting the distance between lenses A4a and A4b. The degree of expansion of the light beam diameter by the beam expander A4 is, for example, about 5-10 times. In this case, if the beam diameter of the illumination light emitted from the laser light source A1 is 1 mm, the beam system of the illumination light will be expanded to about 5-10 mm. If the illumination light incident on the beam expander A4 is not a parallel beam, collimation (making the light beam quasi-parallel) is also possible in addition to adjusting the beam diameter by adjusting the distance between lenses A4a and A4b. However, the collimation of the light beam may be performed by installing a collimating lens separately from the beam expander A4 upstream of the beam expander A4.

[0022] Furthermore, the beam expander A4 is installed on a translation stage with two or more axes (two or more degrees of freedom) and is configured to allow positional adjustment so that the center coincides with the incident illumination light. In addition, the beam expander A4 is equipped with tilt angle adjustment functions with two or more axes (two or more degrees of freedom) so that the optical axis coincides with the incident illumination light.

[0023] • Polarization Control Unit The polarization control unit A5 is an optical system that controls the polarization state of illumination light and is composed of a half-wave plate A5a and a quarter-wave plate A5b. For example, when oblique incidence illumination is performed by placing a reflective mirror A7 (described later) in the optical path, the amount of scattered light from defects on the sample surface increases compared to polarization other than P-polarization by using the polarization control unit A5 to make the illumination light P-polarized. If scattered light from minute irregularities on the surface of the sample itself (called haze) interferes with the detection of minute defects, the haze can be reduced compared to polarization other than S-polarization by making the illumination light S-polarized. The polarization control unit A5 can also be used to make the illumination circularly polarized or 45-degree polarized, which is intermediate between P-polarization and S-polarization.

[0024] As shown in Figure 1, the reflective mirror A7 moves in parallel in the direction of the arrow by a drive mechanism (not shown) to enter and exit the optical path of the illumination light directed toward the sample 1, and the incident path of the illumination light toward the sample 1 can be switched. By inserting the reflective mirror A7 into the optical path, as described above, the illumination light emitted from the polarization control unit A5 is reflected by the reflective mirror A7 and incident obliquely on the sample 1 via the focusing optical unit A6 and the reflective mirror A8. On the other hand, when the reflective mirror A7 is removed from the optical path, the illumination light emitted from the polarization control unit A5 is incident perpendicularly on the sample 1 via the reflective mirrors A9, A10, polarization control unit B2', reflective mirror B1', and scattered light detection system B3. The polarization control unit B2' includes a half-wave plate Ba' and a quarter-wave plate Bb', similar to the polarization control unit A5.

[0025] Figures 5 and 6 are schematic diagrams showing the positional relationship between the optical axis of the illumination light guided to the sample surface from an oblique direction by the illumination optical system A and the shape of the illumination intensity distribution. Figure 5 schematically shows a cross-section of the sample 1 cut by the incident plane of the illumination light incident on the sample 1. Figure 6 schematically shows a cross-section of the sample 1 cut by a plane perpendicular to the incident plane of the illumination light incident on the sample 1 and containing the normal to the sample surface. The incident plane is the plane containing the optical axis OA of the illumination light incident on the sample 1 and the normal to the sample surface. Note that Figures 5 and 6 show only a part of the illumination optical system A, and for example, the output light adjustment unit A3 and the reflection mirrors A7 and A8 are omitted from the illustration.

[0026] When the reflective mirror A7 is inserted into the optical path, the illumination light emitted from the laser light source A1 is focused by the focusing optical unit A6, reflected by the reflective mirror A8, and incident on the sample 1 at an oblique angle. In this way, the illumination optical system A is configured to allow illumination light to be incident on the sample 1 from a direction inclined with respect to the normal to the sample surface. This oblique incidence illumination is controlled by adjusting the light intensity with the attenuator A2, the beam diameter with the beam expander A4, and the polarization with the polarization control unit A5, thereby homogenizing the illumination intensity distribution within the incident plane. As shown in the illumination profile (illumination intensity distribution) LD1 in Figure 5, the illumination spot formed on the sample 1 has a Gaussian light intensity distribution in the s2 direction, and the length of the beam width l1, defined at 13.5% of the peak, is, for example, about 25 μm to 4 mm.

[0027] In the plane perpendicular to the incident surface and the sample surface, the illumination spot has a light intensity distribution where the intensity is weaker at the periphery relative to the center of the optical axis OA, as shown in the illumination profile (illumination intensity distribution) LD2 in Figure 6. Specifically, it is an intensity distribution similar to a Gaussian distribution that reflects the intensity distribution of light incident on the focusing optical unit A6, or a first-order Bessel function of the first kind or a sin function that reflects the aperture shape of the focusing optical unit A6. The length l2 of the illumination intensity distribution in the plane perpendicular to the incident surface and the sample surface is shorter than the beam width l1 shown in Figure 5, for example, set to about 1.0 μm to 20 μm, in order to reduce haze generated from the sample surface. This length l2 of the illumination intensity distribution is the length of the region in the plane perpendicular to the incident surface and the sample surface that has an illumination intensity of 13.5% or more of the maximum illumination intensity.

[0028] Furthermore, the angle of incidence of the oblique incidence illumination on the sample 1 (the angle of inclination of the incident optical axis with respect to the normal to the sample surface) is adjusted to an angle suitable for detecting minute defects by adjusting the position and angle of the reflective mirrors A7 and A8. The angle of the reflective mirror A8 is adjusted by the adjustment mechanism A8a. For example, the larger the angle of incidence of the illumination light on the sample 1 (the smaller the illumination elevation angle, which is the angle between the sample surface and the incident optical axis), the weaker the haze that causes noise from scattered light from minute foreign matter on the sample surface, making it suitable for detecting minute defects. From the viewpoint of suppressing the effect of haze on the detection of minute defects, it is preferable to set the angle of incidence of the illumination light to, for example, 75 degrees or more (elevation angle of 15 degrees or less). On the other hand, in oblique incidence illumination, the smaller the angle of incidence of the illumination, the greater the absolute amount of scattered light from minute foreign matter. Therefore, from the viewpoint of aiming to increase the amount of scattered light from defects, it is preferable to set the angle of incidence of the illumination light to, for example, 60 degrees or more and 75 degrees or less (elevation angle of 15 degrees or more and 30 degrees or less).

[0029] -Scattered Light Detection System- The scattered light detection systems B1-Bn are units that collect and detect scattered light from the illumination spot BS on the sample surface, and consist of multiple optical elements including a focusing lens (objective lens). This optical system detects scattered light from the sample surface and performs laser scattering detection. The 'n' in scattered light detection system Bn represents the number of scattered light detection systems, and in this embodiment, the defect inspection apparatus 100 is described as having 13 scattered light detection systems (n=13). The i-th scattered light detection system Bi consists of an objective lens Bi1, a half-wave plate Bi2 that can rotate around the optical axis of the scattered light detection system Bi by a rotation mechanism (not shown), and a polarizing beam splitter Bi3. The polarization direction is controlled by rotating the half-wave plate Bi2, and the polarizing beam splitter Bi3 separates it into two desired beams with orthogonal polarization directions. P-polarized light traveling in a straight line through the polarizing beam splitter Bi3 is detected by the sensor CiP, and S-polarized light reflected by the polarizing beam splitter Bi3 is detected by the sensor CiS.

[0030] Figure 7 shows the apertures of the scattered light detection system B1-B13 that collect scattered light, as viewed from above, and corresponds to the arrangement of each objective lens of the scattered light detection system B1-B13. In the following explanation, with the direction of incidence of oblique incident illumination onto sample 1 as the reference, the direction of propagation of incident light relative to the illumination spot BS on the sample surface (to the right in Figure 7) will be treated as the front, and the opposite direction (to the left in Figure 7) will be treated as the back, as viewed from above. Therefore, the lower side of the figure relative to the illumination spot BS is the right side, and the upper side is the left side.

[0031] Each objective lens of the scattered light detection system B1-B13 is positioned along the upper hemispherical surface of a sphere (celestial sphere) centered on the illumination spot BS relative to the sample 1. This hemispherical surface is divided into 13 regions, apertures L1-L6, H1-H6, and V, and the scattered light detection system B1-B13 collects and focuses the scattered light at its respective corresponding aperture.

[0032] Aperture V is the region that coincides with the zenith and is located directly above the illumination spot BS formed on the sample surface.

[0033] Apertures L1-L6 are regions that equally divide the ring-shaped area surrounding the illumination spot BS at a low angle, and are arranged in the order of apertures L1, L2, L3, L4, L5, and L6 in a counterclockwise direction from the direction of incidence of obliquely incident illumination when viewed from above. Of these apertures L1-L6, apertures L1-L3 are located to the right of the illumination spot BS, with aperture L1 located to the right rear of the illumination spot BS, aperture L2 to the right, and aperture L3 to the right front. Apertures L4-L6 are located to the left of the illumination spot BS, with aperture L4 located to the left front, aperture L5 to the left, and aperture L6 to the left rear.

[0034] The remaining apertures H1-H6 are regions that equally divide the ring-shaped area surrounding the illumination spot BS at a high angle (between apertures L1-L6 and aperture V), and are arranged in the order of apertures H1, H2, H3, H4, H5, and H6 in a counterclockwise direction from the direction of incidence of obliquely incident illumination when viewed from above. The arrangement of the high-angle apertures H1-H6 is shifted by 30 degrees when viewed from above compared to the low-angle apertures L1-L6. Of the apertures H1-H6, aperture H1 is located behind the illumination spot BS, and aperture H4 is located in front of it. Apertures H2 and H3 are located to the right of the illumination spot BS, with aperture H2 located to the right rear of the illumination spot BS and aperture H3 located to the right front. Apertures H5 and H6 are located to the left of the illumination spot BS, with aperture H5 located to the left front of the illumination spot BS and aperture H6 located to the left rear.

[0035] In Figure 1, scattered light incident on the scattered light detection system Bi is focused and guided to the corresponding sensors CiP and CiS. When comparing Figure 1 and Figure 7, for example, the scattered light detection system B1 in Figure 1 can be treated as an example of an optical system that collects scattered light with aperture L4 in Figure 7, the scattered light detection system B2 with aperture L6, and the scattered light detection system B3 with aperture V.

[0036] Figure 8 is a diagram of the configuration of the scattered light detection system B3, into which scattered light emitted from sample 1 in the normal direction is incident, and Figure 9 is a plan view of Figure 8 from above. The scattered light detection system B3 is composed of a focusing lens (objective lens) B3a and an imaging lens B3b, and the scattered light focused by the focusing lens B3a is detected by sensors C3P and C3S via the imaging lens B3b, half-wave plate B32, and polarizing beam splitter B33. In this respect, it is the same as the other scattered light detection systems B1, B2, B4, etc. The scattered light detection system B3 differs from the other scattered light detection systems in that a reflective mirror B1' is positioned at the pupil position between the focusing lens B3a and the imaging lens B3b. As mentioned above, in the case of incident illumination, illumination light is incident on sample 1 from the normal direction via the reflective mirror B1'. Therefore, the focusing lens B3a of the scattered light detection system B3 also serves as a focusing lens that guides incident illumination to sample 1.

[0037] As mentioned earlier, the illumination spot BS has a linear intensity distribution that is long in the s2 direction. As shown in Figure 9, the reflective mirror B1' is longer than the illumination spot BS in the short axis direction (s1 direction) of the linear illumination spot BS when viewed from the side of sensors C3P and C3S, and shorter than the illumination spot BS in the long axis direction (s2 direction). The reflective mirror B1' is positioned at the pupil of the focusing lens B3a, and the directly reflected light incident on the focusing lens B3a from the sample 1 is reflected by the reflective mirror B1', while the scattered light that is not reflected here is guided to the imaging lens B3b.

[0038] -Sensors- Sensors C1P-CnP and C1S-CnS are single-pixel point sensors that convert scattered light focused through corresponding apertures into photoelectric signals and output detection signals. Sensors C1P-CnP and C1S-CnS can use photomultiplier tubes, SiPMs (silicon photomultiplier tubes), etc., which convert weak signals into photoelectric signals with high gain. Typically, SiPMs are more compact and robust against magnetic noise than photomultiplier tubes, while photomultiplier tubes are superior in terms of signal linearity. Therefore, both may be applied simultaneously. A bandpass filter that transmits only the wavelength of the laser light source A1 is provided in front of the light receiving part of the sensor (not shown). In this embodiment, SiPMs are used for sensor C1P-CnP and photomultiplier tubes are used for sensor C1S-CnS. The polarization direction of the light incident on sensors CiP and CiS can be adjusted by setting the rotation angle of the half-wave plate Bi2. For example, although the types of defects requiring highly sensitive detection differ depending on the inspection process, by adjusting the rotation angle of the half-wave plate Bi2, light in the polarization direction requiring highly sensitive detection can be guided to the sensor CiP. The detection signals output from sensors C1P-CnP and C1S-CnS are input to the signal processing device D as needed.

[0039] -Control Device- The control device E1 is a computer that comprehensively controls the defect inspection device 100, and is composed of ROM, RAM, other memory, as well as a CPU, FPGA, timer, etc. The control device E1 is connected to the monitor E3 and the signal processing device D by wire or wireless. The control device E1 is connected to the device E2 to which the user inputs various operations, and various input devices such as a keyboard, mouse, and touch panel are connected to E2 as appropriate. The control device E1 receives input from the encoders of the rotary stage and translation stage, as well as inspection conditions input from the input device E2 according to the operator's operations. Inspection conditions include, for example, the type, size, shape, material, lighting conditions, and defect judgment conditions of sample 1. The control device E1 also outputs command signals that instruct the operation of the stage ST and the illumination optical system A, etc., according to the inspection conditions, and outputs coordinate data of the illumination spot BS synchronized with the defect detection signal to the signal processing device D. The control device E1 also displays the output of the signal processing device D (defect inspection results of sample 1, etc.) on the monitor E3. As shown in Figure 1, the control device E1 is connected to a network, allowing for the input of inspection condition data and the output of inspection results via the network. It is also possible to output inspection results to inspection and measurement devices connected to this network. For example, a DR-SEM (Defect Review-Scanning Electron Microscope), which is an electron microscope for defect inspection, can be connected. In this case, the control device E1 can transmit the inspection results and defect inspection result data from the DR-SEM, the DR-SEM can observe the defects and determine the type of defect, and then the control device E1 can receive these results.

[0040] -Example of Scattered Light Distribution- Figure 10 illustrates the scattered light distribution from the sample surface when obliquely incident illumination is performed from the left side of the figure, viewed from the direction of the sample normal. Light scattered in the direction of direct reflection of the illumination is called forward scattering, and light scattered in the direction of incidence of the illumination is called backscattering, and is indicated as front and back in the figure, respectively. Light scattered to the left and right with respect to the incidence of the illumination is called left lateral scattering and right lateral scattering, respectively, and is indicated as left and right in the figure, respectively. Figure 10(a) shows the scattered light distribution from foreign matter on the sample surface. Scattering from foreign matter on the sample surface is isotropic, and the amount of scattered light at high angles is weaker than the amount of scattered light at low angles. The distribution of scattered light changes depending on the shape of the defect, and the scattering is not always isotropic. Figure 10(b) shows the distribution of roughness scattered light from an arbitrary coordinate on the sample surface when a polished wafer with a polished surface is similarly illuminated by oblique incidence. Roughness scattered light generates shot noise in the sensor, degrading the defect detection sensitivity. Because roughness scattered light exhibits strong backscattering and weak forward scattering, good sensitivity can be obtained with sensors positioned in front or to the side. In epitaxial wafers created by vapor-phase growth of silicon single crystals on the surface of a polished wafer, directional roughness is generated on the surface depending on the orientation of the silicon single crystals on the surface.

[0041] When the sample is a polished wafer, the sensitivity difference depending on the aperture position is almost constant. For example, as shown earlier in Figure 10(a), for foreign matter on the sample surface, the scattered light from defects is almost isotropically the same in intensity at low-angle apertures L1-L6. In the roughness scattered light distribution in Figure 10(b), because the amount of forward scattering is low, foreign matter on the sample surface can always be detected with good sensitivity at the apertures L3 and L4 and the corresponding sensors, even when the wafer is rotated during inspection.

[0042] In contrast, when the roughness of the sample surface is directional, such as with an epitaxial wafer, the intensity of the roughness scattered light changes individually at each aperture position as the wafer rotates, and the sensor that provides good sensitivity changes moment by moment as the angle of the sample changes relative to oblique incidence illumination.

[0043] When the defect on the sample surface is large with respect to the wavelength of the illumination, the scattered light distribution becomes non-isotropic. An example of this is the scratch defect that occurs during wafer polishing. Typically, it has a linear shape on the wafer surface. The longitudinal direction of the line has a low spatial frequency, and the scattered light has a high forward scattering ratio. The transverse direction has a high spatial frequency, and the ratio of backscattering is high. FIG. 10(c) shows the scattered light distribution when the longitudinal direction of the scratch defect is rotated by +45° with respect to the front-rear direction, and FIG. 10(d) shows the distribution when rotated by -45°. Based on this scattered light distribution, it is possible to improve the defect detection sensitivity or identify the type of defect.

[0044] - Z Sensor - The defect inspection apparatus 100 includes a Z sensor F that measures the height of the sample surface, that is, the distance between the illumination optical system A or the scattered light detection systems B1 - Bn and the sample surface. FIG. 11 is a configuration diagram of the Z sensor F. F1A is a light source, and in this embodiment, a laser is used. The wavelength irradiated by F1A is different from the wavelength of the laser light source A1 to prevent stray light in the sensors C1P - C3P, C1S - C3S. The light emitted from F1A is condensed into a beam spot by the condenser lens of F1B and obliquely incident on the sample 1. The irradiation position of the beam spot changes due to the movement of the sample 1 in the Z direction. This directly reflected light is condensed by F2A and measured by the two-dimensional sensor F2B. The image obtained by F2B is transferred to the height calculation unit F2C, and the height of the sample surface is measured based on the principle of triangulation from the position of the reflected light. The measured height is transferred to the control device E1.

[0045] - Overview of the Signal Processing Device - The signal processing device D is a computer that processes detection signals input from sensors C1P-CnP and C1S-CnS. Like the control device E1, it is composed of ROM, RAM and other memory, as well as a CPU, FPGA, timer, etc. One example of the signal processing device D is a single computer that forms a unit with the main body of the defect inspection device 100 (stage, illumination optical system, scattered light detection system, etc.), which is the controlled part of the defect inspection device 100. However, it may also be composed of multiple computers. In this case, a server can be used as one of the multiple computers, and the server may also be included as a component of the defect inspection device 100. For example, a computer attached to the main body of the device can acquire the defect detection signal from the main body of the device, process the detection data as needed, and send it to the server, where the server can perform processing such as defect detection and classification.

[0046] In this embodiment, the signal processing device D is configured to include a single sensor channel processing unit D1, an integrated channel processing unit D2, a defect determination unit D3, and a filter setting unit D4. The single sensor channel processing unit D1, the integrated channel processing unit D2, the defect determination unit D3, and the filter setting unit D4 may be virtually implemented by software or by hardware such as electronic circuits. Some of these single sensor channel processing units D1, etc. (especially the upstream processes) can be configured using FPGAs or DSPs. Furthermore, some or all of the functions of these single sensor channel processing units D1, etc. can be executed on a server.

[0047] The single-sensor channel processing unit D1 uses the detection signal output of a single sensor or the integrated output of the detection signals of some of the plurality of sensors C1P - CnP, C1S - CnS as individual sensor channels, scans the same coordinates on the sample surface multiple times, filters the channel signals obtained from each sensor channel for temporal or spatial changes, and outputs a filtered signal for each sensor channel (Figs. 12 - 15). In this embodiment, the single-sensor channel processing unit D1 performs the above filtering process using a filter (filter coefficient array) set by the filter setting unit D4. For example, the single-sensor channel processing unit D1 assumes multiple relative positional relationships between the scanning position of the illumination spot BS on the sample surface and an unknown defect, sets multiple filters, and outputs a filtered signal with a virtually narrowed scanning interval.

[0048] The integrated channel processing unit D2 integrates the filtered signals of each sensor channel and outputs a defect evaluation value for the type of defect to be detected (Figs. 12, 15, 16 - 18). At that time, the integrated channel processing unit D2 multiplies the filtered signal for each sensor channel by the corresponding gain (weight coefficient array) and then sums them up to calculate the defect evaluation value. Also, the integrated channel processing unit D2 can calculate the similarity between the detection signal vector based on the channel signals obtained from each sensor channel by scanning the same coordinates on the sample surface multiple times and the model of the type of defect to be detected, and change the defect evaluation value based on this similarity. The similarity can be calculated, for example, based on the L2 norm or the square of the L2 norm of the processing area of the filtering process. Also, the integrated channel processing unit D2 detects defect candidates using a threshold value based on the defect evaluation value for each type of defect. The threshold value used here is set, for example, to a curved surface shape that is convex on the opposite side of the origin in the feature amount space defined by the signals of each sensor channel when the variation of the scattered light from the defect is larger than the variation of the scattered light from the normal part that does not contain defects, thereby improving the noise separation performance.

[0049] The defect determination unit D3 integrates the defect evaluation values related to the defect candidates detected for each type of defect by the integrated channel processing unit D2 to determine the defect.

[0050] The filter setting unit D4 calculates the filter used for filtering in the single sensor channel processing unit D1 and the gain multiplied by the filtered signal integrated in the integrated channel processing unit D2, based on the two-dimensional profile of the detection signal of defects on the sample surface (Figures 13 and 14). The filter setting unit D4 calculates the noise characteristics (e.g., frequency characteristics) and variability related to scattered light from normal areas on the sample surface, and these are used in the calculation of gain and filter. For example, the filter setting unit D4 calculates the variability corresponding to the dispersion of the detection signal from the channel signal, and sets the gain used in the integrated channel processing unit D2 based on the expected detection signal for each of the one or more defect types targeted for detection for each sensor channel and the above variability. The gain is calculated, for example, for each sensor channel by normalizing the expected detection signal for each defect type by the above variability. The expected detection signal is calculated, for example, based on the focusing characteristics of the illumination spot BS of the illumination optical system A or the point image intensity distribution of the scattered light detection system B1-Bn. Furthermore, the filter setting unit D4 normalizes the expected detection signal according to the noise characteristics and sets the filter to be used in the single sensor channel processing unit D1. The filter setting unit D4 can also change the filter and gain based on the distance measured by the Z sensor F.

[0051] The single sensor channel processing unit D1, integrated channel processing unit D2, defect determination unit D3, and filter setting unit D4 will be described in order below.

[0052] -Single Sensor Channel Processing Unit D1- Figure 12 is a processing block diagram of the signal processing unit D. As shown in Figure 12, the single sensor channel processing unit D1 includes a plurality (N) of filter units D1i (filter units D11-D1N). Each filter unit D1i filters the channel signal input from one of the plurality (N) of sensor channels chi (sensor channels ch1-chN) corresponding to that sensor channel. The channel signal is obtained by converting the time-series detection signal (sensor output) of each sensor channel into a digital data sequence using an AD converter. All N filter units D1i have the same configuration. N is the total number of sensor channels and the total number of filter units D1i. Sensor channel chi is the i-th sensor channel, and filter unit D1i is the i-th filter unit. In Figure 12, sensor channel ch1 and filter unit D11 are the 1st sensor channel and filter unit, and sensor channel chN and filter unit D1N are the N-th sensor channel and filter unit. Figure 12 shows two filter sections D11 and D1N, but in reality, there are many more filter sections D11-D1N. In this embodiment, the number of sensor channels ch1-chN (N) is the same as the number of sensors C1P-CnP and C1S-Cn (2n), and each sensor channel ch1-chN is assumed to be an individual output channel of sensors C1P-CnP and C1S-Cn. However, for example, the output obtained by electrically coupling the sensor outputs of multiple sensors can be made into one sensor channel, in which case the number of sensor channels will be less than the number of sensors implemented.

[0053] -Filter section D1i- Each of the N filter sections D1i includes a low-pass filter D1ia, a difference calculation section D1ib, an s1-direction filter D1ic, a buffer memory D1id, an s2-direction filter D1ie, and a buffer memory D1if.

[0054] The filter unit D1i receives the channel signal (time-series data) of the corresponding sensor channel. When the channel signal is input to the filter unit D1i, the low-pass filter D11a sequentially generates a low-frequency signal from the channel signal, and the difference calculation unit D11b subtracts this low-frequency signal from the channel signal to obtain a high-frequency signal. The low-frequency signal output from the low-pass filter D11a is sent not only to the difference calculation unit D1ib but also to the filter setting unit D4 via the output port D11_o3. The output of the low-pass filter D11a is used as a measured value of the amount of scattered light from the normal part of the sample surface in the calculation of filters (filter coefficient sequence) such as filter 1300b in Figure 13. The filters will be explained using Figure 13.

[0055] The illumination profile 1300a shown in Figure 13(a) is the expected time response of the scattered light intensity obtained from minute defects when illumination light is scanned across the sample surface in the s1 direction. This illumination profile 1300a corresponds to the illumination profile LD2 in the s1 direction of the illumination spot BS shown in Figure 6, and can be calculated based on the illumination profile LD2 and the speed at which the defects cross the illumination spot BS (scanning speed). The noise contained in the detection signal obtained by detecting scattered light with a point sensor is mainly shot noise. However, if the roughness scattered light on the sample surface contains coherent noise, or if the sensor itself contains noise, the noise will not appear white but will have frequency characteristics.

[0056] The expected detection signal 1301a shown in Figure 13(b) is the frequency characteristic obtained by the Fourier transform of the illumination profile 1300a. The expected detection signal 1301a for each channel signal can be calculated based on the focusing ability of the illumination optical system A (focusing characteristics of the illumination spot BS) or the point image intensity distribution of the corresponding scattered light detection system B1-Bn. The noise characteristic 1301b shown in Figure 13(b) is the frequency characteristic of the noise obtained by the Fourier transform of the roughness scattered light signal detected in the normal, defect-free area of ​​the sample surface. In the s1-direction filter D1ic of each filter unit D1i, the noise component is separated from each channel signal using a filter 1300b as exemplified in Figure 13(a). The filter 1300b is calculated in the filter setting unit D4 for each filter unit D1i by normalizing the complex conjugate number (expected detection signal 1301a) obtained by the Fourier transform of the illumination profile 1300a using the noise characteristic 1301b, and then performing an inverse Fourier transform on it. The size of the filter applied by the s1 direction filter D1ic is set to roughly match the beam width of the illumination spot BS in the s1 direction.

[0057] In the example shown in Figure 13(b), since the noise is low frequency relative to the defect signal, the filter 1300b has a steep profile relative to the illumination profile 1300a. However, the profile of filter 1300b changes depending on the noise characteristic 1301b. For example, the higher the frequency of the noise characteristic 1301b, the lower the frequency profile of the calculated filter 1300b, and the higher the frequency response of the noise relative to the defect signal can be. Also, noise does not necessarily originate only from roughness scattered light on the sample surface. Since photons from defects are probabilistically detected by the sensor relative to the expected illumination profile intensity, shot noise is generated. This noise is stronger at the peak position of the illumination profile, so lowering the frequency of filter 1300b improves the noise separation performance.

[0058] Figure 13(c) is a graph illustrating the expected detection signal obtained from the same defect on the sample surface using multiple illumination spots BS with the same coordinate in the s1 direction but different coordinates in the s2 direction during helical scanning. The illumination positions of the illumination spots BS on the sample surface differ in the s2 direction for each rotation of the scan, and these illumination positions are shifted by approximately half the length of the illumination spot BS in the s2 direction (for example, the beam width l1 in Figure 5). Because the movement pitch of the illumination spot BS in the s2 direction per rotation of the helical scan is narrower than the length of the illumination spot BS in the s2 direction, the same defect is detected in multiple rotations of scanning. In this case, the expected light intensities BS-a1d, BS-a2d, and BS-a3d obtained from a certain minute defect in each rotation of scanning change from the illumination profiles BS-a1, BS-a2, and BS-a3 of the illumination spot BS for each rotation of scanning, as illustrated in Figure 13(c).

[0059] Figure 13(d) illustrates the s1-direction profile of the expected detection signal obtained for each scan of the same microdefect. The expected signal profile BS-b1 shown in Figure 13(d) represents the s1-direction profile of the detection signal for the expected light intensity BS-a1d exemplified in Figure 13(c). Similarly, the expected signal profiles BS-b2 and BS-b3 represent the s1-direction profiles of the detection signals for the expected light intensity BS-a2d and BS-a3d, respectively.

[0060] Figure 14 is an explanatory diagram of the filters applied for each positional relationship between the scanning position in the s2 direction and the defect position. Figure 14(a) is identical to Figure 13(d). In contrast, Figures 14(b) and 14(c) illustrate the s1-direction profile of the expected detection signal obtained in each scan when the scanning position in the s2 direction changes for the same minute defect as in Figure 14(a). The illumination position of the illumination spot BS in the s2 direction for each scan is known, but the positional relationship between the illumination spot BS and the defect for each scan is unknown. The expected detection light amount obtained in each scan changes as shown in Figures 14(a) to 14(c) depending on this positional relationship between the illumination spot BS and the defect. Therefore, in the filter setting unit D4, multiple filters are set for each filter unit D1i, as explained in Figure 13, assuming multiple relative positional relationships between the scanning position of the illumination spot BS on the sample surface and the unknown defect.

[0061] Figures 14(d) to 14(f) illustrate the expected detection signal profile in the s1 direction obtained in each scan, assuming the positional relationship between the illumination spot BS and the defect is the same as in Figures 14(a) to 14(c), respectively, and the beam width l2 (Figure 6) of the illumination spot BS widens in the s1 direction. The position of the illumination spot BS is known under certain conditions, but it changes depending on the height of the sample surface. In addition, the atmospheric pressure and temperature around the main body of the defect inspection device 100 can cause thermal expansion of the structure holding the illumination optical system A and the scattered light detection system B1-Bn, or a change in the refractive index of the air, which can also change the position of the illumination spot BS. Therefore, in the filter setting unit D4, the assumed illumination profile 1300a of the illumination spot BS is corrected based on the height of the sample surface measured by the Z sensor F (distance between the illumination optical system A or the scattered light detection system B1-Bn and the sample surface), and the filter is changed according to the height of the sample surface. Although not shown in the diagram, the defect inspection device 100 is also equipped with a barometer and a thermometer, and the filter setting unit D4 corrects the illumination profile 1300a according to the ambient pressure and temperature measured by the barometer and thermometer, and the filter is changed according to the pressure and temperature.

[0062] Furthermore, the s1-direction filter D1ic calculates the square of the high-frequency signal output by the difference calculation unit D11b, and filters this squared value as well. The filter applied to this filtering is also set by the filter setting unit D4. The length of this filter also matches the length of the lighting spot BS in the s1 direction, and the filter coefficient is a uniform value corresponding to the length of the filter. The output of the s1-direction filter D1ic for this squared value is used in a later stage to calculate the L2 norm.

[0063] The output of these s1-direction filters D1ic is stored in buffer memory D1id every few turns of the helical scan.

[0064] Furthermore, as explained in Figure 14, the positional relationship in the s2 direction between the detected defect and the illumination spot BS is unknown. Therefore, in the s2-direction filter D1ie of each filter unit D1i, the filtered signal stored in the buffer memory D1id is further filtered in the s2 direction using multiple filters set in the filter setting unit D4 by changing the assumed positional relationship in the s2 direction between the illumination spot BS and the defect. A filtered signal with a virtually narrowed scanning interval in the s2 direction is output. Since the illumination light profile is Gaussian and can be separated into components in the s1 and s2 directions, the filtered signal that has been filtered in the s1 direction by the s1-direction filter D1ic is further filtered in the s2 direction by the s2-direction filter D1ie, resulting in a result similar to that of a two-dimensional filter. Of the multiple filtered signals calculated by the s2-direction filter D1ie, the one with the highest evaluation value is selected and output. As for the calculation method for the evaluation value of the filtered signal, one can choose to select the one with the maximum output from multiple filtered signals calculated by the s2-directional filter D1ie, or to select the one with the highest correlation value between the filter and the acquired signal.

[0065] Furthermore, the s2-direction filter D1ie also applies filtering in the s2 direction to the output of the s1-direction filter D1ic, which relates to the squared value of the output of the difference calculation unit D1ib.

[0066] The outputs of these s2-directional filters D1ie are stored in buffer memory D1if and output to integrated channel processing unit D2.

[0067] The two filter processing results stored in buffer memory D1if will be explained using Figure 15.

[0068] Figure 15 is an explanatory diagram of the inputs from multiple sensors and the processing data output by each processing block in the signal processing device according to the first embodiment of the present invention. The channel signals BSi (channel signals BS1-BSN) shown in Figure 15 represent the waveforms of the channel signals input from sensor channels ch1-chN, respectively.

[0069] As shown in Figure 10, the amount of light from the defect signal and noise included in the channel signal changes depending on the scattered light from the normal part of the sample surface, the type of defect, the angle of the defect relative to the illumination spot BS, etc. In the example in Figure 15, even for detection signals obtained simultaneously for the same defect, the defect signal of channel signal BS1 of sensor channel ch1 is stronger than the defect signal of channel signal BSN of sensor channel chN. Waveforms BS1-c1, BS1-c2, and BS1-c3 each show the amount of light detected by sensor channel ch1 when scanned in the s1 direction at different positions in the s2 direction. Waveforms BSN-c1, BSN-c2, and BSN-c3 each show the amount of light detected by sensor channel N when scanned in the s1 direction at different positions in the s2 direction.

[0070] The channel signals BS1-BSN are subjected to a two-dimensional filtering process in each filter section D1i (filter sections D11-D1N), defined by two directions, s1 and s2. While Figure 12 illustrates an example where two one-dimensional filters are combined, such as the s1-direction filter D1ic and the s2-direction filter D1ie, a two-dimensional filter based on a two-dimensional profile of the defect detection signal on the sample surface may also be applied.

[0071] When detecting scattered light by focusing it using a focusing detection system, the amount of defect signal light changes depending on the positional relationship between the defect and the illumination spot BS. However, in the case of a bare wafer with no patterns on the sample surface, the change in noise intensity due to scanning position is small. As a result, the filter in the s2 direction is almost proportional to the intensity profile of the illumination spot BS.

[0072] As explained in Figure 13, for example, the filter unit D11 normalizes the expected detection signal expected for sensor channel ch1 by the frequency intensity of the noise in the frequency domain to obtain a filter, and outputs a filtered signal B-11-a that has been filtered by this filter. At the same time, the filter unit D11 outputs the L2 norm B-11-b of the processing region for which filtering is performed. The L2 norm B11-b is calculated, for example, as the square root of the filtered value obtained by squaring the output of the difference calculation unit D11b. The value obtained by dividing the filtered signal B-11-a by the L2 norm B-11-b becomes the correlation coefficient, which represents the similarity to the assumed defect model. The other filter units D11 are similar; for example, the filter unit D1N outputs a filtered signal B-1N-a and an L2 norm B-1N-b for sensor channel cnN. Note that the value output from the filter unit D1i along with the filtered signal B-1i-a does not necessarily have to be the L2 norm B-1i-b; it may be the square of the L2 norm B-1i-b. For each sensor channel chi, the filtered signal B-1i-a and the L2 norm B-1i-b are output from the output ports D1i_o1 and D1i_o2 (Figure 12) of the filter unit D1i to the integrated channel processing unit D2.

[0073] - Integrated Channel Processing Unit D2 - The integrated channel processing unit D2 includes multiple (M) channel data integration processing units D2i (channel data integration processing units D21-D2M) corresponding to typical defect types. Each channel data integration processing unit D2i receives and integrates the outputs of multiple (N) filter units D1i (filter units D11-D1N) of the single sensor channel processing unit D1. Each of the M channel data integration processing units D2i has the same configuration and includes a signal integration unit D2iA, an L2 norm integration unit D2iB, and a defect candidate detection unit D2iC. Each of the channel data integration processing units D21-D2M targets a different defect type, and the gain (weighting coefficient sequence) multiplied by the output of the filter units D11-D1N during integration differs according to the defect type.

[0074] -Signal Integration Unit D2iA- The signal integration unit D2iA of each channel data integration processing unit D2i calculates a linear evaluation value B-2i-a (for example, the linear evaluation value B-21-a in Figure 15) for each type of defect to be detected and outputs it to the defect candidate detection unit D2iC.

[0075] The channel signal of each sensor channel chi differs depending on the type of defect, as the amount of light detected from the defect and the noise. The signal integration unit D2iA first normalizes the filtered signal B-1i-a input from the filter unit D1i of each sensor channel chi by the standard deviation of the noise contained in the channel signal of each sensor channel chi, thereby making the amount of noise in the filtered signal B-1i-a of each sensor channel chi constant. This is called whitening. For example, if we let the filtered signal B-1i-a input from the filter unit D1i at a certain timing t be d(i,t), then the signal integration unit D2iA divides this input data d(i,t) by the standard deviation std_n(i) and calculates the normalized signal (= d(i,t) / std_n(i)).

[0076] Next, each signal integration unit D2iA linearly sums the filtered signals B-1i-a of each whitened sensor channel chi. The sum of these filtered signals B-1i-a of each sensor channel chi is the linear evaluation value. At this time, by multiplying the filtered signals B-1i-a of each whitened sensor channel chi by the corresponding gain g1, and then summing the filtered signals B-1i-a multiplied by the gain g1 to calculate the linear evaluation value, the degree of separation between the expected detection signal and the noise can be maximized.

[0077] The gain g1 is calculated in the filter setting unit D4 according to the variation in the channel signal of each sensor channel chi. The gain g1 is the value obtained by dividing the expected detection signal related to defects in the channel signal of each sensor channel chi by the standard deviation of the noise light intensity. For example, in the filter setting unit D4, the standard deviation of the noise light intensity std_n(i) is successively calculated for each sensor channel chi based on the output of the low-pass filter D1ia of each sensor channel chi. In addition, the filter setting unit D4 calculates the expected detection signal s(i) for each type of defect that is expected to be detected for each sensor channel chi. The gain g1 is calculated for each type of defect (for each signal integration unit D2iA) and for each sensor channel chi (g1 = s(i) / std_n(i)). The gain g1 can also be calculated based on the profile of the detection signal of defects on the sample surface, similar to the filter, and can be calculated based on, for example, two-dimensional profiles in the s1 and s2 directions.

[0078] In addition to calculating the standard deviation of noise light intensity std_n(i) in the filter setting unit D4 based on the output of the low-pass filter D11a, the channel data integration processing unit D2i may also calculate it based on the average of the squares of the L2 norm B-1i-b output from the output port D11_o2.

[0079] Furthermore, considering the input data d(i,t) including normalization, the expected detection signal s(i) can be considered as a value normalized by the variance contained in the signal. The variation in the normal portion contained in the input data d(i,t) is generally understood to be shot noise. The calculation of the gain g1 described above is equivalent to normalizing the expected detection signal s(i) of each sensor channel chi, expressed in photon units, by the output of each low-pass filter D1ia, which is also expressed in photon units. Therefore, it is also possible to pre-normalize the input data d(i,t) with the square root of the output of the low-pass filter D1ia expressed in photon units, and then multiply the normalized input data d(i,t) by g1 (= s(i) / std_n(i)).

[0080] Here, if there is variability in the expected detection signal s(i) of the defect, and there is a discrepancy between this variability and the variability of the background noise, the linear evaluation value cannot obtain optimal signal-to-noise ratio (SN) separation. Figure 16 shows an example of a model M3 of the defect signal for the defect type to be detected in the feature space defined by the signals of sensor channels ch1 and chN. The linear threshold (boundary surface) 1601 shown in the figure is the optimal threshold where the linear evaluation value for a given point on the sample surface has the same variability whether it is a detected defect or noise. Here, we assume that the ratio of the probability that the point is a defect to the probability that it is a normal part without defects is constant. Although Figure 16 is shown in two dimensions as an example, in reality, model M3 can be represented in a multidimensional space, and the threshold 1601 can be defined on a plane.

[0081] On the other hand, when the linear evaluation value for a point on the sample surface varies depending on whether it is a defect or noise, the threshold at which the ratio of the probability that the point is a defect to the probability that it is normal is constant cannot be represented by a straight line or a plane. If the variation when the point is a defect is greater than the variation when it is normal, the optimal threshold is a curve or surface that is convex in the opposite direction from the origin, as shown in thresholds 1602 and 1603. The curvature of these thresholds increases as the divergence between the variation when the point is a defect and the variability when it is normal increases. Conversely, if the variation when the point is a defect is smaller than the variability when it is normal, the optimal threshold is a curve or surface that is convex towards the origin, as shown in thresholds 1604 and 1605. The curvature of these thresholds also increases as the divergence between the variation when the point is a defect and the variability when it is normal increases.

[0082] Based on the above, the signal integration unit D2iA calculates the squared value of the filtered signal B-1i-a of each whitened sensor channel chi, in addition to calculating the linear evaluation value, and also performs a linear sum of these squared values. The sum of these squared values ​​is the squared evaluation value. When calculating the squared evaluation value, the squared values ​​of the filtered signal B-1i-a of each sensor channel chi are multiplied by the corresponding gain g2 and then added. The gain g2 used here, like the filter and gain g1, is based on the profile of the defect detection signal on the sample surface, and can be calculated based on, for example, two-dimensional profiles in the s1 and s2 directions. The gain g2 is calculated based on the variability of the detection signal. For example, the gain g2 is calculated by subtracting 1 from the variance of the variability of the expected detection signal of defects in the filtered signal B-1i-a of each whitened sensor channel chi, and then normalizing this value by twice the variance of the variability of the expected detection signal. Then, in the signal integration unit D2iA, the squared evaluation value is added to the linear evaluation value for each defect type (for each signal integration unit D2iA), and the linear evaluation value B-2i-a, which has the squared evaluation value added, is output as one evaluation value of the detection signal for each defect type. This makes it possible to realize a threshold represented by a curve (surface) as shown in Figure 16 as a fixed threshold. The linear evaluation value B-21-a shown in Figure 15 is the linear evaluation value output from the signal integration unit D21A, and the defect separation from noise is improved compared to the filtered signal B-1i-a input from each filter unit D1i.

[0083] -L2 Norm Integration Unit D2iB- The L2 Norm Integration Unit D2iB of each channel data integration processing unit D2i calculates a similarity based on the composite L2 norm B-2i-b described later for each defect type to be detected and outputs it to the defect candidate detection unit D2iC. The composite L2 norm B-21-b shown in Figure 15 is an example of a composite L2 norm calculated by the L2 Norm Integration Unit D21B. For example, in each L2 Norm Integration Unit D2iB, in the same manner as the signal integration unit D2iA, the L2 norm B-1i-b input from the filter unit D1i of each sensor channel chi is normalized by the standard deviation of the noise and whitened. Then, each L2 norm integration unit D2iB calculates a composite L2 norm B-2i-b by calculating the square root of the sum of the L2 norms B-1i-b of each whitened sensor channel chi, and transmits the calculated composite L2 norm B-2i-b to the defect candidate detection unit D2iC.

[0084] Each L2 norm integration unit D2iB further calculates a similarity by dividing the linear evaluation value B-2i-a calculated by the corresponding signal integration unit D2iA by the combined L2 norm B-2i-b, and outputs it to the defect candidate detection unit D2iC. The similarity calculated here makes it possible to determine the signal profile of the defect model, and furthermore, the scattering intensity distribution between sensor channels, while excluding the influence of the detected signal intensity.

[0085] Figure 17 is an explanatory diagram illustrating the similarity between the model of the defect to be detected, represented as a one-dimensional vector, and the detected signal vector. Although the feature space in Figure 17 is represented in two dimensions, in reality, the defect model is represented in a multi-dimensional space with a dimension represented by the product of the number of sensor channels and the number of elements of the filter applied to each sensor channel. Figure 17 shows the processing of the channel data integration processing unit D21, but the processing of other channel data integration processing units is similar.

[0086] The vector BS1v in Figure 17 is the whitened vector representation of the light intensity of the channel signal BS1 obtained by the multiple scans in Figure 15. Similarly, the vector BSNv in Figure 17 is the whitened vector representation of the light intensity of the channel signal BSN obtained by the multiple scans in Figure 15. The detected signal vector B-V is a composite vector of these two vectors BS1v and BSNv. Model M1 is the vector representation of the defect model intended for detection by the channel data integration processing unit D21. The vector B-21-av is a vector representation of the linear evaluation value B-21-a in Figure 15, and corresponds to the projection of the detected signal vector B-V onto Model M1. There is an angle difference of θ1 between the vector of Model M1 and the detected signal vector B-V. In this case, cos(θ1) corresponds to the similarity. The variability of the defect detection signal changes depending on the interference state between the defect and the sample surface, and is therefore larger than the variability of the sample surface by the amount of change due to the interference state. In this case, the threshold Th21 (decision boundary surface) for detecting defect candidates is a curved surface (for example, thresholds 1602, 1603 in Figure 16) that is located on the origin side of a linear threshold (for example, threshold 1601 in Figure 16) in the feature space defined by the channel signals of each sensor channel, and is convex on the opposite side of the origin.

[0087] Figure 17 illustrates an example where the defect model M1 is represented by a vector. However, in some cases, it is preferable to represent the defect model not only as a vector but also as a surface or space. For example, as shown in Figures 10(c) and 10(d), the scattering distribution from the sample surface of a scratch defect changes depending on the orientation of the defect, making it difficult to represent the scattering distribution with a single vector. In this case, it is preferable to increase the dimensions and represent the model as a surface or space.

[0088] Figure 18 is an explanatory diagram of the similarity between a model of the defect intended for detection, represented in a plane, and the detected signal vector. The feature space in Figure 18 is represented in three dimensions, but in reality it is even more multidimensional. BS1-V and BSN-V in Figure 18 are the same as in Figure 17. Vector BS2-V is a vector representation of the detection signal detected by sensor channel ch2. The composite vector of vectors BS1-V, BS2-V, and BSN-V is the detected signal vector B-V. The defect model M3 is a plane containing vectors whose orientation changes when a linear scratch defect is shifted by ±Δθ around the illumination spot BS. Vector B22 is a vector representation of the linear evaluation value and corresponds to the projection of the detected signal vector B-V onto model M3. There is an angle difference of θ2 between the plane of model M3 and the detected signal vector B-V. cos(θ2) corresponds to the similarity.

[0089] -Defect Candidate Detection Unit D2iC- In each channel data integration processing unit D2i, three signals are transmitted to the defect candidate detection unit D2iC: the linear evaluation value and the squared evaluation value calculated by the signal integration unit D2iA, and the similarity value calculated by the L2 norm integration unit D2iB. The defect candidate detection unit D2iC detects defect candidates based on the input linear evaluation value, squared evaluation value, and similarity value. In this embodiment, the defect candidate detection unit D2iC calculates a similarity evaluation value by first obtaining an evaluation value that takes a value between -1 and 1 by subtracting 1 from twice the similarity value, and then normalizing it by the variance of the evaluation value. Then, each defect candidate detection unit D2iC calculates a defect evaluation value by multiplying each of the linear evaluation value, squared evaluation value, and similarity evaluation value by the gain given by the parameter and performing a linear sum, and detects defect candidates by comparing this defect evaluation value with a predetermined threshold. The defect evaluation values ​​of the extracted defect candidates are output from each defect candidate detection unit D2iC to the defect determination unit D3. In the channel data integration processing units D21-D2M, the expected detection signal differs depending on the type of defect being detected, but the processing itself is the same.

[0090] Furthermore, defect candidates can also be detected by, for example, setting a parameter k for similarity, calculating the k-th power of similarity, multiplying the linear sum of the linear evaluation value and the squared evaluation value by the k-th power of similarity to calculate a defect evaluation value, and comparing this defect evaluation value with a predetermined threshold. When k is set to 1, the evaluation does not use similarity. When k is set to a value greater than 1, the defect evaluation value decreases as the deviation of the detected signal vector B-V from the model increases and the similarity decreases, thus increasing the noise reduction effect. On the other hand, when k is set to a value less than 1, the defect evaluation value increases as the deviation of the detected signal vector B-V from the model increases. k can be set according to the detection purpose.

[0091] -Defect Detection Unit D3- The defect detection unit D3 further integrates the defect evaluation values ​​related to the defect candidates input from each defect candidate detection unit D2iC of the integrated channel processing unit D2 to determine defects. For example, the defect detection unit D3 performs defect determination by performing a logical OR operation on the defect evaluation values ​​of the defect candidates input from each defect candidate detection unit D2iC. At this time, in order to classify the type of defect, it is desirable that the defect detection unit D3 performs a process to add a flag to each defect evaluation value that indicates which defect candidate was input from which defect candidate detection unit D2iC. However, if the types of defects that need to be detected in the inspection and the types of defects that do not need to be detected are clearly defined in advance, it is also possible to input only the defect candidates of the types of defects that need to be detected into the defect detection unit D3, and exclude the types of defects that do not need to be detected from the defect determination process. Defects detected by the defect detection unit D3 are transmitted to the control device E1 and displayed, for example, on the inspection result screen of the monitor E3.

[0092] -Filter setting unit D4- As described above, the filter setting unit D4 sequentially calculates and sets the filters used in the single sensor channel processing unit D1 and the gains used in the integrated channel processing unit D2.

[0093] -Effect- In this embodiment, noise is removed from the channel signal by a filter based on the signal expected to be obtained from the defect, thereby making visible the defect detection signal related to minute defects buried in the noise. Furthermore, by integrating the detection signals after filtering for each sensor channel chi using a gain normalized by the variation in the detection signal, a defect evaluation value corresponding to the defect and its characteristics can be obtained, and defect determination can be made. Therefore, a variety of defects, including minute defects, can be detected with high sensitivity.

[0094] Furthermore, by assuming multiple relative positional relationships between the scanning position of the lighting spot BS and unknown defects, and setting multiple filters, and outputting a filtered signal with a virtually narrowed scanning interval, the accuracy of defect detection, such as the detection accuracy of defect features and 2D position, can be further improved.

[0095] Furthermore, the variability of the defect detection signal changes depending on the interference state between the defect and the sample surface, and therefore tends to be greater than the variability of the sample surface due to the variation caused by the interference state. In this case, the accuracy of defect detection can be improved by setting the threshold for defect determination in the shape of a curved surface that is convex to the opposite side of the origin in the feature space defined by the signals of each sensor channel.

[0096] Furthermore, while the profile of the expected detection signal may change due to fluctuations in the height of the sample surface during inspection, the accuracy of defect detection can be maintained in response to these height fluctuations by changing the filter based on the distance measured by the Z sensor F.

[0097] Furthermore, by reflecting the similarity of the defect to the defect model in the defect evaluation value, it is possible to accurately detect defects according to their characteristics and accurately classify the defect type.

[0098] (Second Embodiment) In the first embodiment, a focusing detection system was used for the scattered light detection system B1-Bn, and point sensors were used for sensors C1P-CnP and C1S-CnS, but this configuration can be modified in design. In the second embodiment, an example of applying the invention to a defect inspection device that employs an imaging detection system for the scattered light detection system B1-Bn and line sensors for sensors C1P-CnP and C1S-CnS will be described. Except for the scattered light detection system B1-Bn and sensors C1P-CnP and C1S-CnS, the hardware configuration of this embodiment is the same as that of the first embodiment.

[0099] - Oblique Imaging Detection System - The oblique imaging detection system B'' shown in Figure 19 is one example of the configuration of the scattered light detection system B1-Bn in this embodiment, and comprises an objective lens B1'', a half-wave plate B2'', a polarizing beam splitter B3'', half-wave plates B4P'', B4S'', and imaging lenses B5P'', B5S''.

[0100] The half-wave plate B2" is rotatable by a rotating mechanism (not shown), allowing the polarization direction of the light detected by the objective lens B1" to be rotated to a desired direction. The polarizing beam splitter B3" splits the detected light into two optical paths: P-polarized and S-polarized. The optical path for P-polarization includes the half-wave plate B4P", imaging lens B5P", and line sensor CP". Similarly, the optical path for S-polarization includes the half-wave plate B4S", imaging lens B5S", and line sensor CS". The half-wave plates B4P" and B4S" are set on rotating stages (not shown), respectively, and their polarization directions are rotated to optimize the detection efficiency of the line sensors CP" and CS".

[0101] The line sensors CP'' and CS'' are tilted with respect to the optical axis of the oblique imaging detection system B'' so as to be conjugate to the illumination spot BS on the sample surface. The optical axis of the oblique imaging detection system B'' is tilted with respect to the sample surface. Therefore, the working distance from the objective lens of the oblique imaging detection system B'' differs between regions at the position in the s2 direction of the illumination spot BS. Consequently, if the light-receiving surface of the line sensors CP'' and CS'' is perpendicular to the optical axis of the oblique imaging detection system B'', the entire image of the illumination spot BS will not be formed on the light-receiving surface. Therefore, as described above, the light-receiving surface of the sensor is tilted with respect to the optical axis so as to be conjugate to the sample surface.

[0102] - Depth of focus of the detection system - Since the imaging detection system has a shallower depth of focus than the light-gathering detection system, the control device E1 controls the working distance between the oblique imaging detection system B'' and the sample surface to be constant on the order of micrometers, based on the output value of the Z sensor F.

[0103] However, the helical scanning speed of stage ST can reach tens of meters per second in high-speed inspection modes, making it difficult to keep the working distance constant according to the ideal design value. As a result, the working distance fluctuates, which can cause focus blurring in the oblique imaging detection system B'' and shifts in the image formation position.

[0104] This will be explained using Figure 20. Figure 20 shows how light transmitted through apertures L1, L3, L4, L6, H2, H3, H5, and H6 in Figure 7 spreads in the s2 direction on the line sensor in each oblique imaging detection system B'' when the working distance is shifted by ΔZ from the design position. When the illumination spot BS is shifted from the focal position, the image on each sensor surface expands and shifts due to the focus shift. The image tilt angle and shift amount are small for the high-angle H2P, H3P, H5P, and H6P, while the tilt angle is large for the low-angle L1P, L3P, L4P, and L6P.

[0105] Figure 21 illustrates the changes in this profile in detail. Figure 21(a) shows L3P1-L3P4, which represent the s2-direction profiles of the P-polarized signals detected in the oblique imaging detection system B" related to aperture L3, when the amount of Z-direction displacement of the sample surface is different for each case. Figure 21(b) shows H3P1-H3P4, which represent the s2-direction profiles of the P-polarized signals detected in the oblique imaging detection system B" related to aperture H3, when the amount of Z-direction displacement of the sample surface is different for each case. The Z-direction positional displacement amounts of L3P1 and H3P1 are the same. Similarly, the Z-direction positional displacement amounts of L3P2 and H3P2, L3P3 and H3P3, and L3P4 and H3P4 are the same for each case. It can be seen that the asymmetry of focus blur occurring in the s2 direction increases in the oblique imaging detection system B" related to low-angle aperture L3 compared to the oblique imaging detection system B" related to high-angle aperture H3. Furthermore, compared to the oblique imaging detection system B'' with a high-angle aperture H3, the oblique imaging detection system B'' with a low-angle aperture L3 exhibits a larger amount of image shift in the s2 direction.

[0106] In this embodiment, the signal processing device D holds profile data of the detection signal corresponding to the Z-direction position of the sample surface for each sensor channel. Note that focus blur also differs depending on the field of view. Therefore, profile data of the detection signal corresponding to the Z-direction position of the sample surface for each sensor channel is prepared for each field of view.

[0107] - Depth of focus of the illumination system - In the second embodiment, the beam width of the illumination spot BS in the s1 direction is reduced compared to the first embodiment. Figure 22 shows the relationship between the amount of displacement ΔZ of the sample surface in the Z direction relative to the focal point position of the illumination optical system A and the beam width of the illumination spot BS in the s1 direction. Unlike the scattered light detection system B1-Bn, even if the sample surface is displaced in the Z direction, the illumination spot BS does not shift in the s1 direction, but if the sample surface is displaced from the focal point position, it becomes out of focus and the beam width of the illumination spot BS expands according to the amount of displacement ΔZ. Also, when the sample surface moves in the Z direction, the illumination spot BS shifts in the s2 direction, and the expected defect signal intensity and noise intensity at a specific pixel of the line sensor change.

[0108] -Filter Section- In the s1-direction filter D1ic of each sensor channel chi, in the first embodiment 1, filtering was performed according to the illumination profile in the s1 direction. In contrast, in this embodiment, in the s1-direction filter D1ic of each sensor channel chi, in addition to filtering according to the illumination profile in the s1 direction, filtering that takes into account focus blur in the s2 direction is combined to perform two-dimensional filtering. As explained in Figure 21, the signal processing device D holds the illumination profile in the s2 direction according to the position in the Z direction of the sample surface. The filter used for filtering in the s2 direction is calculated by the filter setting unit D4 in the same manner as the filter used for filtering in the s1 direction. That is, the illumination profile in the s2 direction (Figure 21) according to the position in the Z direction of the sample surface is frequency-converted, and the noise characteristics are also frequency-converted. The filter is calculated by normalizing the complex conjugate of the illumination profile with the power of the noise and performing an inverse conversion. The filter is changed according to the position in the Z direction of the sample surface. In cases where shot noise is dominant and the noise has a flat frequency response, the optimal approach is to perform a convolution with a profile that inverts the profile showing focus blur in the s2 direction.

[0109] The s2-direction filter applied in the s2-direction filter D1ie will be explained using Figure 23. Figure 23(a) shows the illumination light quantity expected to be obtained with two scans. Even when an imaging detection system is used, as in the first embodiment, the movement pitch of the illumination spot BS in the s2 direction per scan is set to be narrower than the beam width of the illumination spot BS in the s2 direction, so that the illumination spot BS overlaps and scans the same coordinates multiple times. Profile p2301 is the illumination light quantity expected in the i-th scan, and profile p2302 is the illumination light quantity expected in the i+1-th scan. For the same defect, the illumination light quantity p2301-a is irradiated in the i-th scan, and the illumination light quantity p2301-b is irradiated in the i+1-th scan. The intensity of the incident light to the line sensor in the imaging detection system differs from the intensity of the incident light to the point sensor in the focusing detection system, and can be determined by the position of the pixel of the line sensor.

[0110] In this embodiment, the filter normalizes the expected signal intensity multiple times by the square of the expected noise amplitude. When the main component of the noise is shot noise, the noise dispersion matches the number of photons in the roughness scattered light and is proportional to the signal intensity. As a result, while a filter proportional to the detected signal value is set for the light-gathering detection system, for the imaging detection system, adding the detected signal values ​​with the same coefficient for pixels at any position yields the optimal result. On the other hand, when electrical noise cannot be ignored and shot noise is not the main component, the noise can be estimated as the sum of the dispersion of shot noise and the dispersion of electrical noise, so a gain is set for each pixel of the line sensor and linear addition is performed.

[0111] Figure 23(b) shows the vector representation of the expected value of the detected signal. Vector V23-a is the vector representing the expected amount of light obtained in the i-th scan, and vector V23-b is the vector representing the expected amount of light obtained in the i+1-th scan. The defect model is contained in the plane spanned by vectors V23-a and V23-b. The ratio of vectors V23-a and V23-b is expected to be equal to the ratio of the irradiated light quantities p2301-a and P2301-b shown in Figure 23(a). In this embodiment, the similarity (= cosθ) is calculated based on the angle θ between the vector V23-c assumed from this expected value and the detected signal vector V23-d.

[0112] In reality, the detected signal vector V23-d has dimensions equal to the product of the number of samples relative to the beam width in the s1 direction of the illumination spot BS, the sum of the number of scans performed multiple times on the same coordinates with illumination spots s2 at different positions in the s2 direction, and the number of sensor channels. Therefore, the similarity is not determined solely by the ratio of illumination during the i-th and i+1th scans. However, it is possible to intentionally reduce the number of dimensions used to calculate the similarity; for example, it is possible to calculate the similarity using only the ratio of illumination light intensity.

[0113] (Addendum) The present invention is not limited to the embodiments described above, and various modifications are included. For example, each of the embodiments described above is described in detail in order to explain the present invention in an easy-to-understand manner, and not all of the described configurations are necessarily required. It is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations.

[0114] In each embodiment, the control lines and information lines shown are those deemed necessary for explanatory purposes, and not all control lines and information lines are necessarily shown in the actual product. In practice, it can be assumed that almost all components are interconnected.

[0115] 1...Sample, 100...Defect inspection device, A...Illumination optical system, BSi...Channel signal, B-1i-a...Filtered signal, B1-Bn...Scattered light detection system (detection system), BS...Illumination spot (spot-shaped illumination light), CiP, CiS...Sensor, chi...Sensor channel, D...Signal processing device, g1, g2...Gain, s(i)...Expected detection signal, Th21...Threshold

Claims

1. A defect inspection apparatus comprising: an illumination optical system that irradiates a sample surface with spot-shaped illumination light; a scanning device that scans the sample surface with the spot-shaped illumination light; a detection system that collects scattered light from the sample surface; a plurality of sensors that convert the light collected by the detection system into photoelectric signals and output detection signals; and a signal processing device that processes the detection signals input from the plurality of sensors, wherein the signal processing device calculates a filter and a gain based on a two-dimensional profile of the detection signal of a defect on the sample surface; scans the same coordinates on the sample surface multiple times using the output of the detection signal of a single sensor or the integrated output of the detection signals of some of the plurality of sensors as individual sensor channels and filters the channel signals obtained from each sensor channel for temporal or spatial changes using the filter, outputs a filtered signal for each sensor channel; multiplies the filtered signals of each sensor channel by the corresponding gain and integrates them to output a defect evaluation value for the defect type to be detected; and detects a defect using a threshold value based on the defect evaluation value.

2. A defect inspection apparatus according to claim 1, wherein the signal processing device sets the filter by normalizing the expected detection signal based on the focusing characteristics of the spot-shaped illumination light of the illumination optical system or the point image intensity distribution of the detection system with the noise characteristics of the normal portion on the sample surface.

3. A defect inspection apparatus according to claim 1, wherein the signal processing device sets up a plurality of filters by assuming a plurality of relative positional relationships between the scanning position of a spot-shaped illumination light on the sample surface and an unknown defect, and outputs a filtered signal with a virtually narrowed scanning interval.

4. A defect inspection apparatus according to claim 1, wherein the signal processing device sets the threshold in a curved shape that is convex to the opposite side of the origin in the feature space defined by the signals of each sensor channel.

5. A defect inspection apparatus according to claim 1, comprising a Z sensor for measuring the distance between the illumination optical system or the detection system and the sample surface, wherein the signal processing device changes the filter based on the distance measured by the Z sensor.

6. A defect inspection apparatus according to claim 1, wherein the signal processing device calculates the similarity between a detection signal vector obtained from each sensor channel by scanning the same coordinates on the sample surface multiple times and a model of the defect type to be detected, and changes the defect evaluation value based on the similarity.

7. A defect inspection apparatus according to claim 6, wherein the signal processing device calculates the L2 norm of the processing area of ​​the filter processing and calculates the similarity based on the L2 norm.

8. A defect inspection apparatus according to claim 1, wherein the signal processing device calculates the variation of the detection signal from the channel signal; calculates the gain for each of the one or more defect types to be detected from the expected detection signal and the variation for each of the sensor channels; calculates the defect evaluation value for each of the defect types to be detected by multiplying the filtered signal for each sensor channel by the gain corresponding to the defect type and integrating them; and determines a defect by integrating the defect evaluation values ​​for each defect type.

9. A defect inspection method using a defect inspection device comprising: an illumination optical system for irradiating a sample surface with spot-shaped illumination light; a scanning device for scanning the sample surface with the spot-shaped illumination light; a detection system for collecting scattered light from the sample surface; and a plurality of sensors that convert the light collected by the detection system into photoelectric signals and output detection signals, wherein the output of a detection signal from one of the plurality of sensors or the combined output of detection signals from some of the sensors are used as individual sensor channels, the same coordinates on the sample surface are scanned multiple times, and the channel signals obtained from each sensor channel are filtered for temporal or spatial changes, and a filtered signal is output for each sensor channel, the filtered signals from each sensor channel are integrated, and a defect evaluation value for the defect type to be detected is output, and a defect is detected based on the defect evaluation value, the defect inspection method being characterized in that: the variation of the detection signal is calculated from the channel signals; the expected detection signal is normalized by the variation for each sensor channel to obtain a gain; and the filtered signals for each sensor channel are multiplied by the corresponding gain and then summed to calculate the defect evaluation value.

Images