JP4844694B2 - Inspection apparatus and defect classification method - Google Patents

Description

The present invention optically detects defects existing in a silicon carbide substrate (SiC substrate) or an epitaxial layer formed on a silicon carbide substrate using a scanning device including a differential interference optical system, and detects the detected defects. The present invention relates to an inspection device to be classified.
The present invention also relates to a defect classification method for capturing a differential interference image on the surface of a silicon carbide substrate or an epitaxial layer and discriminating defects based on the obtained differential interference image.

  A semiconductor device manufacturing method has been developed in which a single crystal layer is formed on a single crystal substrate by an epitaxial growth method, and a device is formed on the formed single crystal layer. In this semiconductor device manufacturing method, a silicon carbide substrate is used as a single crystal substrate, and a silicon carbide single crystal layer is formed on the silicon carbide substrate by an epitaxial growth method. A silicon substrate is used as the single crystal substrate, and a single crystal layer of gallium nitride is formed on the silicon substrate. Since silicon carbide has superior physical and thermal characteristics compared to silicon, a method for manufacturing a device using a semiconductor substrate in which an epitaxial layer of silicon carbide is formed on a silicon carbide substrate is high power. This is extremely useful for manufacturing low-loss semiconductor devices.

  In the semiconductor device manufacturing method described above, in order to improve the manufacturing yield and improve the reliability of the device, the defect existing in the silicon carbide substrate and the epitaxial layer is detected, and the type of the detected defect is determined. Classification is extremely important. Conventionally, defect detection by X-ray topography is known as a method for detecting defects present in a silicon carbide substrate (SiC substrate) (see, for example, Patent Document 1). This defect detection method by X-ray topography has an advantage that crystal defects existing in the SiC substrate or SiC epitaxial layer can be detected nondestructively.

  As a method for optically detecting defects formed on a SiC substrate, a laser scattering type defect inspection apparatus is known (see, for example, Patent Document 2). In this laser scattering type defect inspection apparatus, laser light emitted from a laser diode is incident obliquely on the substrate surface, and scattered light generated on the substrate surface is detected by a photodetector. Then, defect detection is performed based on the detected scattered light, and the detected defect is determined.

Furthermore, it has been reported that the surface of a SiC substrate is imaged using a differential interference microscope, and the type of defect is determined from the differential interference image of the defect. Since the differential interference microscope detects a change in the height of the sample surface as phase difference information, there is an advantage that a minute unevenness appearing on the surface of the SiC substrate can be detected as a luminance image.
JP 2009-44083 A US Pat. No. 7,201,799

  In the defect detection method using X-ray topography described above, crystal defects inside the substrate can be detected, and threading screw dislocations and defects in the basal plane formed in the bulk of the substrate are detected. However, as defects in the substrate and the epitaxial layer, there are not only defects in the bulk but also defects due to scratches or adhesion of foreign substances formed on the substrate surface, and these defects also affect the manufacturing yield. Accordingly, it is also important to be able to detect scratches generated during various processes such as a polishing process and defects due to foreign matter adhesion. However, X-ray topography can detect lattice defects in the bulk, but has a defect that defects formed on the surface of the substrate and the surface of the epitaxial layer, such as stylats and adhesion of foreign matter, cannot be detected. Furthermore, X-ray topography has a drawback that defects having a specific shape such as a carrot defect and a comet defect cannot be detected or distinguished. Therefore, from the viewpoint of defect inspection, the inspection method by X-ray topography has a limit in improving the quality of the epitaxial layer formed on the substrate. In addition, defect detection by the X-ray topography method not only requires a large apparatus for irradiating X-rays, but also has a drawback that it takes a long time to detect defects and the inspection cost is expensive.

  Since the laser scattering type defect inspection apparatus detects scattered light generated on the substrate surface, it is possible to detect defects appearing on the substrate surface. However, since undesired scattered light is incident on the photodetector, the defect detection resolution is low and it is difficult to detect fine defects. In particular, the laser scattering type inspection apparatus has low detection sensitivity in the height direction of the sample surface, so that it is difficult to clearly detect minute unevenness defects of several nm to several tens of nm formed on the surface of the SiC substrate. There are drawbacks. That is, lattice defects formed in the bulk of the SiC substrate appear as irregularities of about several nm to several tens of nm on the substrate surface. Accordingly, in the laser scattered light type defect detection, there is a limit in detecting a fine defect existing in the SiC substrate, and there is a limit in the defect type determination. Furthermore, since the SiC substrate is transparent in the visible region, the irradiated laser light passes through the substrate, is reflected by the back surface of the substrate, and is emitted from the substrate surface. Therefore, it is necessary to remove the influence of the reflected light from the back surface of the substrate. For this reason, it is pointed out that the structure is complicated because a special spatial filter is required.

  The method of detecting a defect using a differential interference microscope has an advantage that a minute uneven shape on the substrate surface can be detected as a luminance image. However, the surface of the substrate has minute irregularities due to step bunching, etc., and undesired scattered light is generated and incident on the light detection means. Therefore, there is a disadvantage that the resolution is lowered, and the reflected light from the back surface of the substrate However, it has been pointed out that the resolution further deteriorates because the light enters the light detection means.

  An important defect that becomes a problem when forming a device in an epitaxial layer formed on a SiC substrate is a micropipe defect. When micropipe defects are formed in the epitaxial layer, there arises a problem that not only the leakage current of the manufactured device increases but also the withstand voltage decreases. Therefore, when manufacturing a device using a SiC substrate, in order to improve the manufacturing yield, it is an urgent issue to detect micropipe defects separately from other defects. However, micropipe defects are only detected as point-like low-brightness images when imaged with a microscope, and cannot be distinguished from edge dislocation defects and spiral dislocation defects that do not become an important problem in device manufacturing. .

  Further, as a defect that becomes a problem when forming a device, there is a defect in the basal plane. If defects in the basal plane are formed in the epitaxial layer, there arises a problem that leakage current of a manufactured device increases.

An object of the present invention is to realize an inspection apparatus that has high detection sensitivity and can detect a minute concave or convex defect of about several nanometers and is not affected by reflected light from the back surface of a substrate.
Another object of the present invention is to realize an inspection apparatus that can detect and detect micropipe defects formed in a SiC substrate and an epitaxial layer separately from other defects.
Another object of the present invention is to provide a defect classification method capable of classifying micropipe defects and defects in the basal plane separately from other defects.

An inspection apparatus according to the present invention is an inspection apparatus that performs defect inspection on a silicon carbide substrate or an epitaxial layer formed on a silicon carbide substrate,
A light source device that generates an illumination beam, a stage that supports a silicon carbide substrate to be inspected and is movable in a first direction and a second direction orthogonal to the first direction, and the illumination beam on the stage An objective lens that projects toward the silicon carbide substrate disposed; and a light detection means that receives reflected light reflected by the surface of the silicon carbide substrate or the epitaxial layer, and the stage is moved in the first and second directions. A scanning device that scans the surface of the silicon carbide substrate or the epitaxial layer with an illumination beam by moving the surface;
The incident illumination beam disposed in the optical path between the objective lens and the light detection means of the scanning device is converted into first and second sub beams having coherence with each other, and the silicon carbide substrate or the epitaxial layer is converted. A differential interference optical system that synthesizes sub-beams reflected on the surface of the substrate and emits an interference beam including phase difference information related to the surface height of the silicon carbide substrate or the epitaxial layer; and
An image forming means for receiving an output signal from the light detection means and forming a differential interference image on the surface of the silicon carbide substrate or the epitaxial layer, a defect detection means for detecting a defect based on the formed differential interference image, and detection A signal processing device having defect classification means for classifying the defect based on the differential interference image of the generated defect,
The defect classification means includes a first classification means for identifying a defect image having a specific shape based on the shape of the differential interference image of the detected defect, and a point-like low luminance based on the luminance distribution of the differential interference image. A second classification means for identifying a defect image, a point-like high-luminance defect image, and a defect image of brightness and darkness obtained by combining the high-luminance image portion and the low-luminance image portion ;
The first classification means has shape information regarding a specific defect image, determines whether or not the defect image having a specific shape is similar to the specific shape image, and based on the determination result, at least the carrot Classify as defect, triangle defect or comet defect,
The second classifying means classifies a dot-like low luminance defect image as a micropipe defect or a foreign matter adhesion defect, and a light and dark luminance defect image as an edge transition defect, a screw dislocation defect, a basal plane defect or a protrusion. It is classified as a defect .

  In the inspection apparatus according to the present invention, since the confocal differential interference image of the surface of the single crystal substrate is picked up, the height appearing on the surface of the single crystal substrate and the epitaxial layer formed on the single crystal substrate is as small as about several nanometers. It is possible to capture an uneven defect as a luminance distribution image. In addition, since the surface of the single crystal substrate is scanned by the confocal scanning device, a two-dimensional image with higher resolution than that of the differential interference image captured by the differential interference microscope is captured. Lattice defects present in single crystal substrates, especially SiC substrates, and lattice defects present in epitaxial layers appear as slight unevenness changes of several nanometers on the substrate surface or epitaxial layer surface. By doing so, it is possible to detect various crystal defects and defects caused by physical factors as bright and dark luminance distribution images. In addition, since the confocal scanning device can acquire three-dimensional shape information and surface contour shape information on the surface of the substrate and the epitaxial layer, a microscopic image detected as a point-like low-luminance image in the confocal differential interference image. Pipe defects can be detected separately from other defects by acquiring surface contour shape information using the same inspection apparatus. Further, defects in the basal plane detected as a bright and dark image can be detected separately from other defects by comparing coordinates before and after the formation of the epitaxial layer. In this specification, the surface contour shape information is the three-dimensional image information indicating the surface contour of the surface of the single crystal substrate or the epitaxial layer, and the two-dimensional image information (cross-sectional shape information) cut by a plane orthogonal to these surfaces. ).

  As a confocal scanning device, a single illumination beam is scanned by the scanning device, and the reflected light from the surface of the single crystal substrate is received by the light detection means through the pinhole, and the line-shaped illumination beam is scanned. Scanning a single crystal substrate using a multi-beam arranged in a line, and a confocal scanning device that scans by moving the device or stage and receives reflected light from the single crystal substrate by a linear image sensor, and from the substrate surface It is possible to use a confocal scanning device that receives the reflected light of the light by a linear image sensor.

  As the differential interference optical system, various differential interference optical systems such as a Nomarski prism, a lotion prism, and a Walston prism can be used. The differential interference optical system is inserted into the optical path when capturing a confocal differential interference image of a single crystal substrate, and is removed from the optical path when capturing a confocal image. In the present specification, the confocal image means a confocal image taken with the differential interference optical system removed from the optical path.

A defect classification method according to the present invention is a defect classification method for detecting defects present in a silicon carbide substrate or an epitaxial layer formed on a silicon carbide substrate, and classifying the detected defects.
Using a scanning device including a differential interference optical system, capturing a differential interference image of the surface of the silicon carbide substrate or the epitaxial layer; and
A defect detection step of detecting defects based on the imaged differential interference image;
A defect classification step for classifying defects based on differential interference images of detected defects;
The defect classification step is based on a first classification step for identifying a defect image having a specific shape based on the shape of the differential interference image of the detected defect, and on a luminance distribution of the differential interference image of the detected defect, A second classification step for identifying point-like low-luminance defect images, point-like high-luminance defect images, and light-dark luminance defect images,
In the first classification step, for a defect image having a specific shape, it is determined whether or not the detected defect image is similar to the specific shape based on shape information regarding the defect image, and based on the determination result, Classify at least as carrot defect, triangle defect or comet defect,
In the second classification step, the dot-like low-luminance defect image is classified as a micropipe defect or a foreign matter adhesion defect, and the light-dark luminance defect image is classified as an edge transition defect, a screw dislocation defect, a basal plane defect, or a protrusion. It is classified as a defect .

  If there is a defect on the surface of the silicon carbide substrate or the epitaxial layer using a scanning device including a differential interference optical system, irregularities of several nanometers to several tens of nanometers are formed on the surface of the substrate or the epitaxial layer. Accordingly, when the surface of the substrate or the surface of the epitaxial layer is scanned with an illumination beam, a luminance image or luminance distribution image having a specific shape is detected according to a minute unevenness change present on the surface of the silicon carbide substrate or epitaxial layer. Therefore, according to the inspection apparatus of the present invention, it is possible to automatically determine the type of detected defect.

It is a figure which shows an example of the test | inspection apparatus by this invention. It is a figure which shows an example of the signal processing apparatus by this invention. It is a figure which shows the defect classification method by this invention. It is a figure which shows the modification of the test | inspection apparatus by this invention.

FIG. 1 is a diagram showing an example of an inspection apparatus according to the present invention. In this example, a silicon carbide substrate (SiC substrate) or a silicon carbide substrate on which an epitaxial layer is formed is used as a single crystal substrate to be inspected. Of course, not only a SiC substrate but also a single crystal substrate in which a GaN epitaxial layer (single crystal layer) is formed on a silicon single crystal substrate can be used as the single crystal substrate. A confocal scanning device having a differential interference optical system is used as the inspection device, and the entire surface of the silicon carbide substrate is scanned by the confocal scanning device to capture a confocal differential interference image on the entire surface of the substrate. The obtained confocal differential interference image is subjected to various image processes to detect the defect and its coordinates (address). Moreover, the confocal differential interference image of a defect can be reviewed using the address information of the detected defect, and the detected defect can be classified. Furthermore, the inspection apparatus according to the present invention captures a confocal image of a single crystal substrate and a mode for capturing a confocal differential interference image of the single crystal substrate by selectively inserting a differential interference optical system in the optical path. There are two modes: mode. Since the differential interference optical system detects a change in the height of the sample surface as a change in the phase difference of the scanning light, it is possible to detect a concave and convex defect of several nm to several tens of nm as a luminance change. . Furthermore, since the scanning can be performed in the Z-axis direction (optical axis direction) as a characteristic of the confocal scanning device, the inspection device according to the present invention not only captures the confocal differential interference image of the silicon carbide substrate but also detects it. It is also possible to capture a confocal image of the detected defect, acquire the three-dimensional shape information of the defect from the confocal image, and also acquire the cross-sectional shape information (surface contour information) of the detected defect. Therefore, when determining the defect type, the defect confocal differential interference image, the three-dimensional shape, and the cross-sectional shape can be used, and the defect type can be determined by coordinate comparison of the defect coordinates.

Referring to FIG. 1, a mercury lamp is used as the illumination light source 1. Various illumination light sources other than mercury lamps such as xenon lamps can also be used. The illumination beam emitted from the illumination light source 1 is incident on an optical fiber bundle 2 in which a plurality of optical fibers are stacked in a circle, propagates through the optical fiber, and is emitted as a divergent beam having a substantially circular cross section. Incident. The filter 3 emits green wavelength light (e-line: wavelength 546 nm) from the incident light beam. The light beam emitted from the filter is converted into a parallel beam by the converging lens 4 and enters the slit 5. The slit 5 is disposed at the pupil position of the converging lens 4 and has an elongated opening extending in a first direction (a direction orthogonal to the paper surface). Here, the first direction is referred to as the X direction. The width of the opening of the slit 5 is set to 10 to 20 μm, for example. Therefore, an elongated line-shaped light beam extending in the first direction is emitted from the slit 5. The linear light beam emitted from the slit 5 enters the polarizer 6 and is converted into polarized light having a single vibration surface. This line-shaped polarized beam is reflected by the half mirror 7 that functions as a beam splitter, and enters a vibrating mirror 9 that functions as a scanning device via a relay lens 8.

A drive circuit 10 is connected to the vibration mirror 9, and the drive circuit 10 drives the vibration mirror based on a control signal supplied from the signal processing device 11. The oscillating mirror 9 functions as a beam scanning device that deflects an incident line-shaped light beam in a second direction (Y direction) orthogonal to the first direction. The signal processing device 11 has position information in the Y direction of the light beam based on the angle information of the vibrating mirror. Note that other scanning devices such as a polygon mirror may be used instead of the vibrating mirror. Further, when scanning by moving the stage, the vibrating mirror is not always necessary. The line-shaped light beam emitted from the vibrating mirror 9 enters the differential interference optical system 14 via the relay lenses 12 and 13. In this example, a Nomarski prism is used as the differential interference optical system. The linear polarized beam incident on the Nomarski prism 14 is converted into two sub beams whose vibration planes are orthogonal to each other. A phase difference of (2m + 1) π / 2 is given between these two sub-beams, where m is a natural number. Therefore, it is possible to detect a defect having a height change of several nm formed on the surface of the SiC substrate as a luminance image. Further, the shearing amount of the Nomarski prism is set to 2 μm, for example. The Nomarski prism 14 is detachably arranged in the optical path, and is inserted in the optical path when capturing a confocal differential interference image of the SiC substrate. In other cases, for example, a 3D confocal image of the sample is captured. In the case and when the surface contour image of the sample is taken, it is removed from the optical path.

The two sub beams emitted from the Nomarski prism 14 enter the objective lens 15. The objective lens 15 focuses the incident two line-shaped sub-beams and projects them toward the silicon carbide substrate 17 to be observed arranged on the stage 16. Accordingly, the surface of the SiC substrate 17 is scanned in the second direction (Y direction) perpendicular to each other by two line-shaped sub-beams extending in the first direction (X direction). As a substrate to be inspected, both a SiC substrate on which no epitaxial layer is formed and a SiC substrate on which an epitaxial layer is formed are used. The stage 16 is composed of an XY stage, the position information thereof is detected by a position sensor (not shown), and the position information of the stage is supplied to the signal processing device 11. When a defect is detected by scanning the entire surface of the SiC substrate 17, the entire surface of the SiC substrate can be scanned by moving the stage 16 in a zigzag manner in the Y direction and the X direction while maintaining the oscillating mirror 9 in a stationary state. . Alternatively, the stage 16 can be continuously moved in the first direction, and combined with scanning in the second direction by the oscillating mirror 9, the entire surface of the SiC substrate can be inspected. Also, when reviewing using the address of the detected defect, the stage is moved in the X and Y directions based on the coordinate information of the defect so that the defect is positioned in the field of view, and the vibrating mirror is scanned in the second direction. Thus, the confocal differential interference image in the vicinity of the defect can be taken.

A motor 18 and a motor drive circuit 19 are connected to the objective lens 15, and can be moved along the optical axis direction by a drive control signal supplied from the signal processing device 11. The position of the objective lens in the optical axis direction is detected by the position sensor 20 and supplied to the signal processing device 11. Here, the motor 18 changes the relative distance in the optical axis direction between the objective lens and the SiC substrate on the stage, that is, the relative distance between the focal point of the light beam that scans the substrate surface and the substrate surface. Functions as a means. The objective lens can move in the optical axis direction with a resolution of 10 nm.

Due to the characteristics of the confocal scanning device, the oscillating mirror 9 is driven while moving the objective lens 15 in the direction of the optical axis, and a two-dimensional confocal image of the surface of the SiC substrate is taken a plurality of times. By detecting the generated position in the optical axis direction, it is possible to acquire three-dimensional shape information (three-dimensional shape image) of the SiC substrate surface. Further, based on the acquired three-dimensional shape information, two-dimensional shape information (cross-sectional shape information) indicating the SiC substrate surface as a cross section can be acquired. Therefore, like micropipe defects formed in the epitaxial layer, hollow hole defects can be detected separately from other defects by acquiring surface contour shape information including three-dimensional shape information or cross-sectional shape information. Is possible. When capturing a two-dimensional confocal image, the Nomarski prism 14 is imaged off the optical path.

The two reflected beams reflected from the surface of the SiC substrate are collected by the objective lens 15 and enter the differential interference optical system 14. The two reflected sub-beams are synthesized by the differential interference optical system 14 to form an interference beam including the height change of the SiC substrate surface as phase difference information. For example, when a concave or convex defect of about several nanometers exists on the surface of the SiC substrate, one of the two sub beams incident on the SiC substrate surface scans on the defect, and the other sub beam is normal. Since the surface portion is scanned, a phase difference corresponding to the height of the defect is introduced between the two sub beams. As a result, the interference beam emitted from the differential interference optical system 14 includes, as phase difference information, an unevenness change of about several nm that appears on the surface of the SiC substrate due to crystal defects.

The interference beam emitted from the Nomarski prism 14 propagates in the opposite direction along the original optical path, enters the vibrating mirror 9 through the relay lenses 13 and 12, and is descanned by the vibrating mirror. The interference beam emitted from the vibration mirror 9 passes through the lens 8 acting as an imaging lens, passes through the half mirror 7, and enters the analyzer 21. The analyzer 21 is arranged in a crossed Nicols relationship with respect to the polarizer 6. Accordingly, light other than the polarized light synthesized in the Nomarski prism 14 is blocked, and only light constituting the differential interference image is transmitted through the analyzer 21.

The line-shaped interference beam transmitted through the analyzer 21 enters the linear image sensor 23 through the positioner 22. The linear image sensor 23 has a plurality of light receiving elements arranged in a direction corresponding to the first direction, and receives the incident line-shaped interference beam. Each light receiving element of the linear image sensor converts phase difference information included in the interference beam into luminance information. Therefore, the unevenness of about several nm formed on the surface of the SiC substrate or the surface of the epitaxial layer is displayed as a luminance image. The light receiving element array arranged in the linear shape of the linear image sensor is almost equivalent to a light receiving element array in which a pin hole is disposed on the front surface of each light receiving element because the entrance opening is limited by the frame. Therefore, the confocal optical system having the differential interference optical system is configured by receiving the reflected light from the SiC substrate surface by the linear image sensor.

The electric charge accumulated in each light receiving element of the linear image sensor is sequentially read out by a read control signal supplied from the signal processing device 11 and output as a one-dimensional image signal on the surface of the SiC substrate. The one-dimensional image signal output from the linear image sensor is amplified by the amplifier 24 and supplied to the signal processing device 11 through the camera link. The signal processing device 11 includes an image processing board, and generates a two-dimensional image of the SiC substrate surface using the received one-dimensional image signal, position information of the vibrating mirror, position information of the stage, and the like. In addition, the generated two-dimensional image is subjected to various image processes including a filtering process, a binarization process, and a threshold comparison process to detect a defect and acquire its coordinates.

The SiC substrate is transparent in the visible light wavelength range. For this reason, when the surface of the SiC substrate is scanned with a light beam, the incident light beam is transmitted through the inside of the SiC substrate, and the reflected light reflected from the back surface of the SiC substrate is incident on the detector, resulting in a decrease in resolution. . For this reason, both the case where the SiC substrate is imaged by the differential interference microscope and the case where the defect is detected by the laser scattering method have a drawback that the resolution is low and the accuracy of defect detection is lowered. In contrast, the confocal inspection apparatus according to the present invention has a configuration that is substantially equivalent to a pinhole disposed on the front surface of the linear image sensor. Out of the optical path, it does not enter the light receiving element of the linear image sensor, but only the reflected light reflected by the surface of the SiC substrate enters the linear image sensor. As a result, by using the inspection apparatus according to the present invention, it is possible to capture a confocal differential interference image having a higher resolution than the differential interference image obtained by the differential interference microscope, and to perform defect detection with higher detection accuracy. Is possible.

The differential interference optical system detects minute irregularities of about several nanometers formed on the sample surface as a phase difference, so that SiC substrates are caused by crystal defects existing on the substrate such as edge dislocation defects and dislocations in the basal plane. When unevenness of about several nm is formed on the surface, these unevenness can be detected as a luminance image. Furthermore, with respect to protrusion defects and concave defects, upward and downward slopes are detected as low-intensity images or high-intensity images on the differential interference image, so the brightness change between the brightness and darkness displayed in the captured differential interference image It is possible to determine whether the defect is a concave defect or a convex defect. Therefore, it is possible to easily determine whether the defect is a pit defect or a bump defect based on the captured bright and dark differential interference image.

Since the inspection apparatus shown in FIG. 1 constitutes a confocal scanning apparatus having a differential interference optical system, a confocal differential interference image on the surface of the SiC substrate can be taken, and further, three-dimensional contour shape information on the surface of the SiC substrate and It is also possible to acquire surface contour shape information indicating the contour of the SiC substrate surface as a cross-sectional image. That is, as a characteristic of the confocal optical system, when the focal point of the scanning beam is located on the sample surface, reflected light having the maximum luminance is incident on the linear image sensor, and linearly as the focal point of the scanning beam is displaced from the sample surface. The amount of reflected beam incident on the image sensor is reduced. Therefore, while moving the objective lens along the optical axis direction, that is, moving the focusing point of the line-shaped scanning beam along the optical axis direction, scanning with the vibrating mirror captures a plurality of two-dimensional confocal images. Then, by detecting the position in the optical axis direction of the objective lens that generates the maximum luminance signal of the linear image sensor for each pixel, three-dimensional shape information indicating the contour of the SiC substrate surface is acquired. Further, from the acquired three-dimensional shape information, it is also possible to acquire surface contour information that indicates a surface contour that shows a SiC substrate cut by an arbitrary surface as a cross section. Therefore, by capturing a three-dimensional image of the detected defect, the shape of the defect can be displayed in three dimensions, and information useful for determining the type of defect can be obtained. It is also possible to determine micropipe defects based on the two-dimensional contour shape information and three-dimensional surface contour shape information on the SiC substrate surface.

Next, it will be described in what form the defect formed in the SiC substrate and the epitaxial layer is captured in the confocal differential interference image captured by the inspection apparatus according to the present invention.
[Micropipe defects]
Micropipes are defects in the form of hollow holes. Therefore, when the scanning beam scans over the micropipe defect, the reflected light from the bottom surface of the hole does not enter the linear image sensor or only a small amount of reflected light, so that it is detected as a dot-like low luminance image. .
[Foreign matter adhesion]
When a foreign object adheres to the surface, a foreign object with high reflectivity such as metal is detected as a dot-like high-intensity image, and when a foreign object with low reflectivity adheres, it is detected as a dot-like dark low-intensity image .
[Downfall]
The downfall is formed by a lump being deposited on the epitaxial layer during the growth of the epitaxial layer, and has a form having two inclined surfaces, upward and downward. Therefore, when it is captured as a differential interference image, it is detected as a bright and dark luminance image in which a bright image portion and a dark image portion are combined.
[Edge dislocation defects]
Edge dislocation defects appear as pit structures on the surface of the SiC substrate or epitaxial layer. Therefore, it has a form having two downward and upward inclined surfaces, and in the differential interference image, it is detected as a bright and dark luminance image in which the low luminance image portion and the high luminance image portion are combined.
[Spiral dislocation defects]
Similar to the edge dislocation defect, the screw dislocation appears as a pit structure on the surface of the SiC substrate and the epitaxial layer, and is detected as a bright and dark luminance image.
[Flaws in the basal plane]
The defects in the basal plane appear as pit structures on the surface of the SiC substrate and the surface of the epitaxial layer. Therefore, the differential interference image is detected as a bright and dark luminance image.
[bump]
A bump is a protrusion-like defect, and has a downward slope and an upward slope, and is thus detected as a bright and dark luminance image.
[scratch]
Scratches may be formed during the polishing process of the substrate. Since this scratch has a linear concave structure, it is detected as a linear bright and dark luminance image on the confocal differential interference image.

Next, a method for detecting a micropipe defect and a basal plane defect that strongly influences the device manufacturing yield will be described. First, a micropipe defect detection method will be described. Micropipe defects formed on the SiC substrate and the epitaxial layer are observed as point-like low-luminance images, and are difficult to distinguish from other defects such as defects due to foreign matter adhesion. Therefore, in the present invention, the surface contour information of the SiC substrate surface or the epitaxial layer surface is acquired using the characteristics of the confocal optical system to determine whether or not there is a micropipe defect. When a dot-like low-luminance image is detected, the stage is moved to the address position, and the detected dot-like defect is positioned in the field of view. Subsequently, a plurality of two-dimensional confocal images are picked up while moving the objective lens in the optical axis direction to form a three-dimensional image of the defect, and surface contour information is acquired. Since the micropipe is a hollow hole having a diameter of about several μm, when the surface contour shape is acquired, it is displayed as an image of a deep hole in the bottom surface or a hollow hole image in which the bottom surface is not detected. Therefore, it is possible to determine whether or not the detected defect is a micropipe defect by acquiring the surface contour shape information for the defect detected as a point-like low-luminance image.

Next, a method for detecting defects in the basal plane will be described. Defects in the basal plane formed on the SiC substrate appear as pit structures on the substrate surface. Therefore, when imaging is performed using the differential interference optical system, a bright and dark luminance image is detected, and defects in the basal plane formed in the epitaxial layer are also detected as a bright and dark luminance image that appears on the epitaxial layer surface. On the other hand, defects in the basal plane formed in the epitaxial layer are formed by propagation of defects in the basal plane formed in the SiC substrate into the epitaxial layer. Therefore, the defects in the basal plane formed in the epitaxial layer appear at a position separated from the defect position on the SiC substrate surface as a starting point by a distance defined by the thickness of the epitaxial layer and the off angle along the step flow direction. Therefore, it is determined whether or not the defects in the basal plane formed in the epitaxial layer are defects in the basal plane by comparing the defect coordinates before and after the formation of the epitaxial layer. That is, the coordinate information of the defect detected as a bright and dark image by the defect inspection of the SiC substrate before the epitaxial layer formation is stored in the defect memory. Further, the coordinates of the defect detected as a bright and dark image after the epitaxial layer is formed are detected. Then, the coordinates are compared, and when the two defect coordinates are separated by a distance defined by the thickness of the epitaxial layer and the off-angle, the pit defects detected in the defect inspection of the epitaxial layer are defects in the basal plane. Judge that there is. As described above, by comparing the coordinates of defects detected before and after the formation of the epitaxial layer, it is possible to detect the defects in the basal plane formed in the epitaxial layer separately from other defects.

Next, the defect inspection method will be described. As a characteristic of the SiC substrate, when a dislocation defect exists in the SiC substrate, the dislocation defect existing on the surface of the SiC substrate propagates in the epitaxial layer when an epitaxial layer is formed thereafter. For example, if basal plane dislocations are formed on the SiC substrate before the epitaxial layer is formed, the basal plane dislocations become edge dislocation defects in the epitaxial layer when the epitaxial layer is formed on the SiC substrate. Or, it becomes a dislocation defect in the basal plane. In addition, defects having specific shapes such as carrot defects and half-moon defects formed in the epitaxial layer are often formed due to screw dislocations in the SiC substrate. In view of the characteristics of these SiC substrates, the following two methods can be used as the defect inspection method according to the present invention. The first inspection method scans the entire surface of the SiC substrate before forming the epitaxial layer, acquires a confocal differential interference image over the entire surface of the SiC substrate, and detects a defect and its coordinates from the confocal differential interference image. Thereafter, an epitaxial layer is formed on the SiC substrate. Then, scanning is performed on the entire surface of the epitaxial layer after the epitaxial layer is formed, and a confocal differential interference image is captured on the entire surface of the epitaxial layer. Then, defect detection is performed on the SiC substrate after the epitaxial layer is formed. In this method, since confocal differential interference images before and after the formation of the epitaxial layer are obtained, it is possible to compare defect images before and after the formation of the epitaxial layer. It is also possible to detect epi defects formed during the growth of the epitaxial layer.

The second inspection method uses the inspection apparatus according to the present invention shown in FIG. 1 to capture a confocal differential interference image over the entire surface of the SiC substrate before the epitaxial layer is formed, and to obtain the acquired confocal differential interference image. Image processing is performed for the defect and its address (coordinate information) is acquired. Next, an epitaxial layer is formed on the SiC substrate after inspection. Subsequently, the SiC substrate on which the epitaxial layer is formed is mounted again on the inspection apparatus, and the portion designated by the defect address detected by the previous defect inspection is positioned in the visual field of the inspection apparatus on the surface of the epitaxial layer. Then, a confocal differential interference image of the surface of the epitaxial layer is taken. Then, for the confocal differential interference image of the defect on the surface of the epitaxial layer, the defect is classified from the viewpoint of the shape, size, and luminance distribution of the defect. According to this inspection method, since the confocal differential interference image is not captured on the entire surface of the substrate after the epitaxial layer is formed, but only a part is captured on the substrate, there is an advantage that the inspection time is shortened.

FIG. 2 is a diagram illustrating an example of the signal processing device 11 that performs defect detection and defect classification. In this example, the entire surface of the SiC substrate before the epitaxial layer is formed is scanned to detect defects, and the entire SiC substrate after the epitaxial layer is formed is detected to detect defects. Perform defect classification. The one-dimensional image signal output from the linear image sensor 23 and amplified by the amplifier 24 is converted into a digital signal by the A / D converter 30 and supplied to the signal processing device 11. A stage position signal (digital signal) indicating the position of the stage 16 that supports the substrate is also supplied to the signal processing device 11. Further, an objective lens position signal (digital signal) output from the position sensor 20 indicating the position of the objective lens in the optical axis direction is also supplied to the signal processing device 11. In this example, the signal processing apparatus 11 is configured by software executed by a computer, and various means operate under the control of the control means 31. The signal lines from the control means 31 are not shown in the drawing because the drawings are crossed.

The one-dimensional image signal input to the signal processing device 11 is sent to the two-dimensional image generation means 32 to form a two-dimensional image, that is, a two-dimensional confocal differential interference image. The two-dimensional image signal is supplied to the first image memory 33, and the confocal differential interference image on the surface of the SiC substrate before the epitaxial layer is formed is accumulated in the first image memory 33. The two-dimensional image signal formed by the two-dimensional image generation unit 32 is supplied to the defect detection unit 34. The defect detection means 34 is also supplied with a stage position signal and position information of each light receiving element of the linear image sensor. The defect detection means 34 includes filtering means, binarization means, and threshold value comparison means, and performs image processing on the input two-dimensional image to detect defects. At the same time, the coordinates of the detected defect are also acquired using the stage position signal and the position information of each light receiving element of the linear image sensor. The coordinates of the detected defect are stored in the first defect memory 35.

When observing a defect formed on the SiC substrate, the first image memory 33 is accessed under the control of the control means 31 using the defect coordinate information stored in the first defect memory 35 and designated. An image of a predetermined size including the generated defect can be taken out, output as a defect image of the SiC substrate, and displayed on the monitor. Further, when classifying defects on the SiC substrate, the defect image is extracted from the first image memory using the defect coordinate information stored in the defect memory under the control of the control means 31. It supplies to the classification means 36. Then, the defect classification means 36 determines the defect type and outputs the defect type.

  The two-dimensional image formed by the two-dimensional image generation unit 32 and the position signal of the objective lens are supplied to the three-dimensional shape information acquisition unit 37. The three-dimensional shape information acquisition means 37 detects the position in the optical axis direction that generates the maximum luminance value for each pixel of a plurality of two-dimensional confocal images captured while moving the objective lens in the optical axis direction. Three-dimensional shape information representing a three-dimensional image of the surface is acquired. The obtained three-dimensional shape information is supplied to the defect classification means 36 and used for defect classification. It is also possible to output as a three-dimensional image output and display a three-dimensional image of the defect on the monitor.

The three-dimensional shape information acquired by the three-dimensional shape information acquisition unit 37 is also supplied to the cross-sectional shape information acquisition unit 38. The cross-sectional shape information acquisition means 38 uses, as a cross-sectional image, a two-dimensional contour shape obtained by cutting the SiC substrate surface along a specified line by a specific surface based on the three-dimensional shape information, that is, the contour shape of the SiC substrate surface. The cross-sectional shape information shown is generated. The acquired cross-sectional shape information is supplied to the defect classification means 36 and used for detection of micropipe defects. It is also possible to output the cross-sectional shape as an image output, display it on the monitor, and display on the monitor the cross-sectional shape shown by cutting the portion including the micropipe defect by a plane orthogonal to the substrate surface. When detecting a micropipe defect, it is also possible to determine whether or not it is a micropipe defect based on the three-dimensional shape information.

  After the defect inspection of the SiC substrate is completed, the SiC substrate is sent to an epitaxial layer growth apparatus, and an epitaxial layer (single crystal layer) is formed on the SiC substrate.

  The substrate on which the epitaxial layer is formed is mounted again on the inspection apparatus that has performed the defect inspection, and the above-described defect inspection is performed, and defect detection and defect coordinate acquisition are performed on the surface of the epitaxial layer. That is, a two-dimensional confocal differential interference image on the surface of the epitaxial layer is generated by the two-dimensional image generating means 32, and the generated two-dimensional image is stored in the second image memory 39. Further, the generated two-dimensional confocal differential interference image is sent to the defect detection means 34, and defects formed on the surface of the epitaxial layer are detected, and the defect coordinates are stored in the second defect memory 40.

  After the defect inspection is completed for the epitaxial layer, defect classification is performed on the detected defects. FIG. 3 is a diagram showing an example of a defect classification method according to the present invention. In step 1, the defects are classified based on the shape of the defect image. As defects formed in the epitaxial layer, defects such as a carrot defect, a triangle defect, a comet defect, a half moon defect, and a scratch are detected as a luminance image having a specific shape when captured as a differential interference image. Therefore, in this example, first, a luminance image having a specific shape is extracted based on the shape of the defect image. Then, carrot defects, triangle defects, comet defects and the like are classified based on the shape of the luminance image (step 1).

  Subsequently, the luminance distribution image having no specific shape is classified based on the luminance distribution of the defect image (step 2). As the classification based on the luminance distribution of the defect image, a point-like low luminance image, a point-like high luminance image, and a light / dark luminance distribution image obtained by combining the low luminance image portion and the high luminance image portion are extracted. For example, a micropipe defect and a defect due to adhesion of foreign matter having a low reflectance are detected as a dot-like low-brightness image when captured as a differential interference image. In this case, a micropipe defect is extracted using the cross-sectional shape information. Moreover, the defect detected as a high-intensity image is determined to be a defect caused by adhesion of a foreign object having a high reflectance such as metal. Further, the defects detected as the brightness distribution image of brightness and darkness include pit defects and bump defects. For bright and dark image defects, it is determined whether the detected defect is a convex defect or a concave defect based on the generation order of the bright image portion and the dark image portion in the scanning direction. Therefore, projection defects such as bumps and concave defects such as pits are classified based on the order in which bright and dark image portions occur. Further, when it is determined that the defect is a concave defect (pit defect) based on the order of occurrence of light and dark, an edge dislocation defect, a spiral dislocation defect, and a basal plane defect are included. In this case, the defect in the basal plane is extracted by coordinate comparison described later.

  As a result of classification based on the luminance distribution, for defects detected as point-like low-luminance images, the three-dimensional shape information of the defective part is acquired, the cross-sectional shape is acquired based on the three-dimensional shape information, and the cross-sectional shape of the defective part is obtained. Based on the above, it is determined whether the defect is a micropipe defect or a defect due to foreign matter adhesion (step 3). That is, when the cross-sectional shape of the defect portion is an image including a hollow hole having a bottom surface with a depth about the thickness of the epitaxial layer or a hollow hole without a bottom surface, the low luminance image is a defect image due to a micropipe defect. Is determined.

  As a result of classification based on the luminance distribution, pit defects detected as a bright and dark image are classified by comparing the coordinates of defect coordinates before and after the formation of the epitaxial layer, and defects in the basal plane are extracted (step 4). That is, when a pit defect is detected in the defect detection on the surface of the epitaxial layer, the pit defect detected in the defect inspection of the SiC substrate before the epitaxial layer formation is searched based on the coordinates of the bright and dark image, and the thickness of the epitaxial layer is determined. It is searched whether or not a pit defect exists at a position separated in a direction opposite to the step flow direction by a distance defined by the off angle. When pit defects exist at positions separated by a distance defined by the thickness and off-angle of the epitaxial layer on the SiC substrate surface, the bright and dark image detected in the defect detection of the epitaxial layer is an image due to defects in the basal plane. Judge that there is.

  The defect classification method described above can also be applied to the defect classification on the SiC substrate surface before the epitaxial layer is formed. In the SiC substrate, there are no carrot defects, triangle defects, and comet defects, but defects such as scratches are determined based on shape information. Further, since the coordinates in the basal plane defect cannot be compared, the classification process up to step 3 is executed.

  With reference to FIG. 2, the processing content of the defect classification means 36 of the inspection apparatus according to the present invention will be described. After the defect detection process is completed, defect classification is performed. First, under the control of the control means 31, the second image memory 39 is accessed, and a predetermined size of a part determined to be a defect using the defect coordinates stored in the second defect memory 40 Are extracted and supplied to the defect classification means 36. The input defect image is supplied to the first defect classification means 50. The first defect classification means 50 classifies the input image based on the shape. That is, the first defect classification means 50 has shape information regarding a specific image such as a carrot defect or a comet defect, and determines whether or not the received image is similar to the specific shape. By this shape determination, defect classification such as carrot defect and comet defect is performed on the defect image having a specific shape.

  A defect image of a predetermined size at a part determined to be a defect is also supplied to the second defect classification means 51. The second classifying unit 52 classifies the received defect image into a point-like high-intensity image, a point-like low-intensity image, or a light-dark image based on the luminance distribution. Further, for bright and dark images, it is determined whether the defect is a convex defect or a concave defect. In the case of a bright / dark image, it is determined whether a protrusion defect such as a bump or a pit defect is based on the order of occurrence of a bright image portion and a dark image portion.

  When it is determined by the second classifying means that the image is a dot-like low-intensity image, the third defect portion means 52 is activated to perform a micropipe defect detection process. In this case, the stage moves based on the coordinate information of the defect image, and the three-dimensional shape information and the cross-sectional shape information are acquired by positioning the defect coordinates in the field of view. Then, the cross-sectional shape information formed by the cross-sectional shape information acquisition 38 is supplied to the third defect classification means 52. The second defect classification means 52 determines whether or not a hollow hole is formed on the surface of the epitaxial layer based on the received cross-sectional shape information, and if it is determined that a hollow hole is formed, the defect The image is determined to be due to a micropipe defect, and the micropipe defect is output. On the other hand, when it is determined that the hollow hole is not formed, it is determined that the defect image is a low-reflectance foreign matter adhesion.

  If the second defect classification means 51 determines that the defect is a pit defect, the fourth defect classification means 53 operates. The fourth defect classification means 53 has epitaxial layer thickness information and off-angle information. Then, the coordinate information of the defect image received by accessing the second defect memory is acquired. Based on the obtained coordinate information, the first defect memory is searched and it is determined whether or not pit defects exist at positions on the surface of the SiC substrate separated by a distance defined by the thickness of the epitaxial layer and the off angle. To do. When it is determined that a pit defect exists on the surface of the SiC substrate before the formation of the epitaxial layer, it is determined that the received defect image is an image caused by a defect in the basal plane, and the fact is output. On the other hand, if it is determined that there is no pit defect at a position separated by a corresponding distance, it is determined that the defect image is a defect other than a defect in the basal plane, for example, a blade dislocation defect or a scratch image.

  As described above, according to the inspection apparatus of the present invention, it is possible to detect the defect formed in the epitaxial layer and to automatically determine the type of the detected defect. In particular, it is possible to automatically determine and output micropipe defects and basal plane defects that have a strong influence on the performance of devices formed in the epitaxial layer.

  FIG. 4 is a view showing a modification of the inspection apparatus according to the present invention. In this example, an inspection apparatus that uses the plurality of scanning beams arranged along the first direction to form the SiC substrate surface will be described. In addition, the same code | symbol is attached | subjected and demonstrated to the component same as the component used in FIG. A laser light source 60 is used as an illumination light source. The laser beam emitted from the laser light source 60 enters the diffraction grating 61 and is converted into a plurality of light beams (multi-beams) arranged along the first direction. The multi-beam enters the polarization beam splitter 64 through the first and second relay lenses 62 and 63, passes through the polarization beam splitter 64, and enters the oscillating mirror 9. The vibrating mirror 9 deflects a plurality of incident light beams along a second direction orthogonal to the first direction. The plurality of light beams reflected by the oscillating mirror 9 enter the Nomarski prism 14 via the third and fourth relay lenses 65 and 66 and the quarter wavelength plate 67. Each light beam incident on the Nomarski prism 14 is converted into two sub beams whose vibration surfaces are orthogonal to each other. A phase difference of (2m + 1) π / 2 is given between these two sub-beams, where m is a natural number.

  The two sub beams emitted from the Nomarski prism 14 enter the objective lens 15. The objective lens 15 focuses the incident sub-beam and projects it toward the silicon carbide substrate 17 to be observed arranged on the stage 16. Accordingly, the surface of the SiC substrate 17 is scanned in the second direction orthogonal to the arrangement direction of the light beams by the two rows of sub-beams arranged along the first direction.

  The sub beam reflected by the surface of the SiC substrate 17 is condensed by the objective lens 15 and enters the Nomarski prism 14. In the Nomarski prism, the reflected beams related to each other are synthesized to generate a plurality of interference beams including a change in the height of the SiC substrate surface as a phase difference. The plurality of interference beams are incident on the oscillating mirror 9 through the quarter-wave plate 67 and the fourth and third relay lenses 66 and 65. Then, it is descanned by the vibrating mirror and enters the polarization beam splitter 64. Since a plurality of incident interference beams pass through the quarter wavelength plate twice, they are reflected by the polarization plane of the polarization beam splitter and enter the linear image sensor 69 through the imaging lens 68. The linear image sensor 69 includes a one-dimensional line sensor in which a plurality of photodiodes are arranged in a line in a direction corresponding to the first direction. Each interference beam then enters the corresponding photodiode. The electric charge accumulated in each photodiode is sequentially read out by the drive signal supplied from the signal processing device 11 and supplied to the signal processing device 11. In the signal processing apparatus, defect detection and classification of the detected defect are performed based on FIGS.

  The present invention is not limited to the above-described embodiments, and various modifications and changes can be made. For example, in the above-described embodiment, the relative distance between the focal point of the scanning beam and the substrate surface is changed by moving the objective lens in the optical axis direction, but the objective lens is fixed and the SiC substrate is supported. It is also possible to change the relative distance between the focal point of the scanning beam and the substrate surface by moving the stage in the optical axis direction.

  Furthermore, in the above-described embodiments, the confocal scanning device using a line-shaped scanning beam and the confocal scanning device that scans the sample surface with multiple beams have been described as the confocal scanning device. It is also possible to use a confocal scanning device that scans the substrate surface.

  Further, in the above-described embodiment, the confocal scanning device including the differential interference optical system that detects the defect of the epitaxial layer is used as the means for detecting the defect of the SiC substrate. However, the means for detecting the defect of the SiC substrate is different. It is also possible to inspect the epitaxial layer for defects using the detected defect coordinates.

  Further, in the above-described embodiments, the Nomarski prism is used as the differential interference optical system, but other differential interference optical systems such as a lotion prism and a Walston prism can be used.

DESCRIPTION OF SYMBOLS 1 Illumination light source 2 Optical fiber 3 Filter 4 Focusing lens 5 Slit 6 Polarizer 7 Half mirror 8, 12, 13 Relay lens
DESCRIPTION OF SYMBOLS 9 Vibration mirror 10 Drive circuit 11 Signal processing apparatus 14 Nomarski prism 15 Objective lens 16 Stage 17 SiC substrate 18 Motor 19 Drive circuit 20 Position detection sensor 21 Analyzer 22 Positioner 23 Linear image sensor 24 Amplifier 30 A / D converter 31 Control means 32 Two-dimensional image generation means 33 First image memory 34 Defect detection means 35 First defect memory 36 Defect classification means 37 Three-dimensional shape information acquisition means 38 Cross-sectional shape information acquisition means

Claims (8)

An inspection apparatus for performing a defect inspection on a silicon carbide substrate or an epitaxial layer formed on a silicon carbide substrate,
A light source device that generates an illumination beam, a stage that supports a silicon carbide substrate to be inspected and is movable in a first direction and a second direction orthogonal to the first direction, and the illumination beam on the stage An objective lens that projects toward the silicon carbide substrate disposed; and a light detection means that receives reflected light reflected by the surface of the silicon carbide substrate or the epitaxial layer, and the stage is moved in the first and second directions. A scanning device that scans the surface of the silicon carbide substrate or the epitaxial layer with an illumination beam by moving the surface;
The incident illumination beam disposed in the optical path between the objective lens and the light detection means of the scanning device is converted into first and second sub beams having coherence with each other, and the silicon carbide substrate or the epitaxial layer is converted. A differential interference optical system that synthesizes sub-beams reflected on the surface of the substrate and emits an interference beam including phase difference information related to the surface height of the silicon carbide substrate or the epitaxial layer; and
An image forming means for receiving an output signal from the light detection means and forming a differential interference image on the surface of the silicon carbide substrate or the epitaxial layer, a defect detection means for detecting a defect based on the formed differential interference image, and detection A signal processing device having defect classification means for classifying the defect based on the differential interference image of the generated defect,
The defect classification means includes a first classification means for identifying a defect image having a specific shape based on the shape of the differential interference image of the detected defect, and a point-like low luminance based on the luminance distribution of the differential interference image. A second classification means for identifying a defect image, a point-like high-luminance defect image, and a defect image of brightness and darkness obtained by combining the high-luminance image portion and the low-luminance image portion ;
The first classification means has shape information regarding a specific defect image, determines whether or not the defect image having a specific shape is similar to the specific shape image, and based on the determination result, at least the carrot Classify as defect, triangle defect or comet defect,
The second classifying means classifies a dot-like low luminance defect image as a micropipe defect or a foreign matter adhesion defect, and a light and dark luminance defect image as an edge transition defect, a screw dislocation defect, a basal plane defect or a protrusion. An inspection apparatus characterized by being classified as a defect . 2. The inspection apparatus according to claim 1, wherein when the defect image is identified as a bright and dark luminance defect image by the second classification means, it is determined whether the defect is in the basal plane by comparing the address of the defect before and after the formation of the epitaxial layer. inspection and wherein the determining.   3. The inspection apparatus according to claim 1, wherein the signal processing device further stores an image memory that stores a differential interference image formed by the image forming unit, and an address memory that stores an address of a detected defect. And an inspection apparatus for outputting a differential interference image of a detected defect as defect image information.   4. The inspection apparatus according to claim 3, wherein the signal processing device is the same as a first image memory that stores a differential interference image of a defect imaged in an inspection of a silicon carbide substrate before an epitaxial layer is formed. A second image memory for storing a differential interference image of a defect imaged in the inspection after the epitaxial layer is formed on the silicon carbide substrate; and a defect detected in the inspection of the silicon carbide substrate before the epitaxial layer is formed A first address memory for storing the address of the first address memory, and a second address memory for storing the address of the defect detected in the inspection after the epitaxial layer is formed for the same single crystal. An inspection apparatus that outputs differential interference images of defects before and after being performed as defect image information.   5. The inspection apparatus according to claim 1, wherein the light source device generates a linear illumination beam extending in one direction or a plurality of illumination beams arranged in a line, and The light detection means is constituted by a linear image sensor having a plurality of light receiving elements arranged in a line, and the scanning device is constituted as a confocal scanning device. A defect classification method for detecting defects existing in a silicon carbide substrate or an epitaxial layer formed on a silicon carbide substrate and classifying the detected defects,
Using a scanning device including a differential interference optical system, capturing a differential interference image of the surface of the silicon carbide substrate or the epitaxial layer; and
A defect detection step of detecting defects based on the imaged differential interference image;
A defect classification step for classifying defects based on differential interference images of detected defects;
The defect classification step is based on a first classification step for identifying a defect image having a specific shape based on the shape of the differential interference image of the detected defect, and on a luminance distribution of the differential interference image of the detected defect, A second classification step for identifying point-like low-luminance defect images, point-like high-luminance defect images, and light-dark luminance defect images,
In the first classification step, for a defect image having a specific shape, it is determined whether or not the detected defect image is similar to the specific shape based on shape information regarding the defect image, and based on the determination result, Classify at least as carrot defect, triangle defect or comet defect,
In the second classification step, the dot-like low-luminance defect image is classified as a micropipe defect or a foreign matter adhesion defect, and the light-dark luminance defect image is classified as an edge transition defect, a screw dislocation defect, a basal plane defect, or a protrusion. defect classification method characterized by classifying as a defect. 7. The inspection apparatus according to claim 6, wherein when the defect image is identified as a bright / dark luminance defect image in the second classification step, whether or not the defect is in the basal plane by comparing addresses of defects before and after the formation of the epitaxial layer. A defect classification method characterized by determining . The defect classification method according to claim 6 or 7, wherein the defect detection step includes a first detection step of acquiring a defect existing in the silicon carbide substrate before the epitaxial layer is formed and an address thereof;
A defect classification method comprising: forming a epitaxial layer on the silicon carbide substrate; and detecting a defect present in the formed epitaxial layer.

Images