Method for evaluating sic substrate and method for manufacturing sic substrate

Abstract

To provide a method for evaluating a SiC substrate and a method for manufacturing a SiC substrate with which basal surface dislocation density can be easily and quickly estimated.SOLUTION: A method for evaluating a SiC substrate according to the present embodiment includes: an image acquisition step of acquiring an X-ray topographic image of the entire first surface of a SiC substrate; and an estimation step of estimating the basal surface dislocation density of the SiC substrate from the X-ray topographic image of the entire first surface of the SiC substrate on the basis of a learning result of deep learning.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a method for evaluating a SiC substrate and a method for manufacturing a SiC substrate. [Background technology]

[0002] Silicon carbide (SiC) has an electric breakdown field one order of magnitude larger than silicon (Si) and a band gap three times larger. Silicon carbide (SiC) also has properties such as a thermal conductivity about three times higher than silicon (Si). For this reason, silicon carbide (SiC) is expected to be applied to power devices, high-frequency devices, high-temperature operating devices, and the like. For this reason, SiC epitaxial wafers have come to be used in the above-mentioned semiconductor devices in recent years.

[0003] SiC epitaxial wafers are obtained by stacking a SiC epitaxial layer on the surface of a SiC substrate. Hereinafter, the substrate before the SiC epitaxial layer is stacked is referred to as the SiC substrate, and the substrate after the SiC epitaxial layer is stacked is referred to as the SiC epitaxial wafer. SiC substrates are produced by slicing from SiC ingots. SiC devices such as power devices, high frequency devices, and high temperature operating devices are obtained by forming devices in the SiC epitaxial layer of the SiC epitaxial wafer and then chipping the SiC epitaxial wafer.

[0004] In SiC epitaxial wafers, basal plane dislocations (BPDs) are known to be one of the device killer defects that cause fatal defects in SiC devices. For example, when a current is passed in the forward direction through a bipolar device, the recombination energy of the flowing carriers causes partial dislocations of basal plane dislocations inherited from the SiC substrate to the SiC epitaxial layer to move and expand, forming high-resistance stacking faults. These high-resistance areas in the device cause a decrease in the reliability of the device (forward degradation).

[0005] Basal plane dislocations in SiC epitaxial layers are often inherited from those in the SiC substrate, so there is a demand for a method to characterize basal plane dislocations in SiC substrates.

[0006] For example, Patent Document 1 describes the evaluation of the in-plane distribution of basal plane dislocations in a SiC substrate using the correlation between the contrast of an X-ray topographic image and basal plane dislocations.

[0007] Furthermore, for example, Patent Document 1 discloses a method of classifying defects one by one using a neural network. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Patent No. 6197722 [Patent Document 2] US Patent Application Publication No. 2020 / 365685 Summary of the Invention [Problem to be solved by the invention]

[0009] The method described in Patent Document 1 determines dislocation density using only image contrast as a measurement parameter, so the measurement accuracy of dislocation density may be insufficient. Also, for SiC substrates with low dislocation density, sufficient contrast difference cannot be obtained, and the method described in Patent Document 1 may not be able to appropriately evaluate the in-plane distribution of basal plane dislocations.

[0010] The method described in Patent Document 2 requires classification of each defect one by one, and it takes a lot of time to evaluate the overall dislocation density.

[0011] The present invention has been made in consideration of the above problems, and has an object to provide a method for evaluating a SiC substrate and a method for manufacturing a SiC substrate, which are capable of measuring basal plane dislocation density simply and quickly. [Means for solving the problem]

[0012] The inventors focused on the overall image of the X-ray topography image rather than on individual defects. They found that deep learning using the overall image of the X-ray topography image enables analysis that takes into account the strain of the entire crystal, and allows for accurate estimation of the basal plane dislocation density. In addition, by using the overall image of the X-ray topography image rather than focusing on individual defects, the time required for estimating the basal plane dislocation density could be shortened.

[0013] (1) A method for evaluating a SiC substrate according to a first aspect includes an image acquisition step of acquiring an X-ray topographic image of an entire first surface of a SiC substrate, and an estimation step of estimating a basal plane dislocation density of the SiC substrate from the X-ray topographic image of the entire first surface of the SiC substrate based on a learning result of deep learning.

[0014] (2) The method for evaluating a SiC substrate according to the above aspect may further include an image compression step of compressing the X-ray topographic image acquired in the image acquisition step. In the image compression step, the pixel size of the X-ray topographic image is reduced to 0.1 mm 2 The size is set to be equal to or larger than that. The estimation step is performed using the compressed X-ray topographic image.

[0015] (3) The method for evaluating a SiC substrate according to the above aspect may further include a preliminary step of measuring an X-ray topographic image of each of a plurality of SiC substrates and a basal plane dislocation density of each of the plurality of SiC substrates, wherein the preliminary step learns a relationship between the X-ray topographic image and the basal plane dislocation density of each of the SiC substrates by the deep learning.

[0016] (4) In the method for evaluating a SiC substrate according to the above aspect, the number of SiC substrates measured in the preliminary step may be 100 or more.

[0017] (5) In the method for evaluating a SiC substrate according to the above aspect, an image obtained by inverting or rotating the X-ray topographic image acquired in the preliminary step may be used for the learning.

[0018] (6) In the method for evaluating a SiC substrate according to the above aspect, the X-ray topographic image may be a diffraction image including a diffraction vector in a <11-20> direction.

[0019] (7) In the method for evaluating a SiC substrate according to the above aspect, the X-ray topographic image may be a diffraction image including a diffraction vector in a <1-100> direction.

[0020] (8) In the method for evaluating a SiC substrate according to the above aspect, the SiC substrate may have a diameter of 145 mm or more.

[0021] (9) In the method for evaluating a SiC substrate according to the above aspect, the SiC substrate may have a diameter of 195 mm or more.

[0022] (10) In the method for evaluating a SiC substrate according to the above aspect, the diameter of the SiC substrate measured in the pre-processing step may be different from the diameter of the SiC substrate for which an X-ray topographic image is obtained in the image obtaining step.

[0023] (11) In the method for evaluating a SiC substrate according to the above aspect, the basal plane dislocation density of the SiC substrate is 1.0×10 4 / cm 2 The following is also fine.

[0024] (12) A method for producing a SiC substrate according to a second aspect includes the method for evaluating a SiC substrate according to the above aspect. Effect of the Invention

[0025] By using the method for evaluating a SiC substrate according to the above embodiment, the basal plane dislocation density can be measured quickly and non-destructively. [Brief description of the drawings]

[0026] [Figure 1] FIG. 2 is a plan view of the SiC substrate according to the embodiment. [Diagram 2] FIG. 2 is a flow diagram of an evaluation method for a SiC substrate according to the present embodiment. [Diagram 3] The measurement results of Example 1 are shown. [Figure 4] The measurement results of Example 2 are shown. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] The evaluation method and manufacturing method of the SiC substrate according to the present embodiment will be described in detail below with reference to the drawings as appropriate. The drawings used in the following description may show characteristic parts in an enlarged scale for the sake of convenience in order to make the characteristics of the present embodiment easier to understand, and the dimensional ratios of each component may differ from the actual ones. The materials, dimensions, etc. exemplified in the following description are merely examples, and the present invention is not limited thereto, and may be appropriately modified and implemented within the scope of the present invention.

[0028] In this specification, individual orientations are indicated by [] and collective orientations by <>. For negative indices, a "-" (bar) is placed above the number in crystallography, but in this specification, a negative sign is placed before the number.

[0029] In the method for evaluating a SiC substrate according to the present embodiment, the basal plane dislocation density of the SiC substrate is determined.

[0030] FIG. 1 is a plan view of a SiC substrate 10 according to this embodiment. <0001> The figure is a view seen from the direction. The SiC substrate 10 is made of, for example, n-type SiC. The polytype of the SiC substrate 10 is not particularly limited and may be any of 2H, 3C, 4H, and 6H. The SiC substrate 10 is, for example, 4H—SiC.

[0031] The SiC substrate 10 has a substantially circular shape in plan view. The SiC substrate 10 may have a notch N for grasping the direction of the crystal axis. Instead of the notch N, the SiC substrate 10 may have an orientation flat.

[0032] The diameter of the SiC substrate 10 is, for example, 145 mm or more, preferably 149 mm or more, more preferably 155 mm or less, and further preferably 151 mm or less. The diameter of the SiC substrate 10 is, for example, 195 mm or more, preferably 199 mm or more, more preferably 205 mm or less, and further preferably 201 mm or less. The SiC substrate 10 may be, for example, a 6-inch substrate or an 8-inch substrate.

[0033] Basal plane dislocations are dislocations that exist in the (0001) plane (=c-plane), which is the basal plane of SiC substrate 10. The basal plane dislocation density is calculated by dividing the number of basal plane dislocations present in SiC substrate 10 by the area of ​​SiC substrate 10.

[0034] 2 is a flow diagram of the method for evaluating a SiC substrate according to the present embodiment. The method for evaluating a SiC substrate according to the present embodiment includes, for example, a preparatory step S1, an image acquiring step S2, an image compressing step S3, and an estimating step S4.

[0035] In the pre-process S1, the X-ray topographic image of each of the multiple SiC substrates and the basal plane dislocation density of each of the multiple SiC substrates are measured. These measured data are stored in the memory of the evaluation device. The memory may be stored in a server located outside the evaluation device. The learning program of the evaluation device learns using these stored data. The learning program learns the relationship between the X-ray topographic image and the basal plane dislocation density of each SiC substrate by deep learning. The data collected in the pre-process S1 corresponds to the teacher data for deep learning. In the pre-process S1, a learning process for deep learning is performed. When the actual evaluation of the SiC substrate is performed using a trained evaluation device, it is not necessary to perform the pre-process S1 at the time of evaluation.

[0036] In the preliminary step S1, the relationship between the X-ray topographic image of the SiC substrate and the basal plane dislocation density of the SiC substrate in the X-ray topographic image is obtained as training data. The X-ray topographic image of the entire SiC substrate contains information on not only dislocations but also the overall strain.

[0037] The X-ray topographic image is a topographic image of the entire first surface of the SiC substrate. By performing learning using the entire image of the SiC substrate, rather than a local image of the defective portion, classification of local images is not required for the estimation process in estimation step S4, and the time required to estimate the basal plane dislocation density can be shortened. In addition, the captured X-ray topographic image is 2 The image may be compressed to a size of 250 x 250 pixels, and the compressed image may be used for learning. The pixel size indicates the size of one pixel. For example, learning may be performed using an image obtained by compressing a captured X-ray topography image to 250 x 250 pixels.

[0038] The X-ray topographic image of the SiC substrate measured in the preliminary step S1 may be a diffraction image including a diffraction vector in the <11-20> direction, or may be a diffraction image including a diffraction vector in the <1-100> direction.

[0039] The number of basal plane dislocations contained in the SiC substrate measured in the preliminary step S1 may be determined from an X-ray topographic image, a photoluminescence method, a mirror electron microscope, or by measuring etch pits after KOH etching. In order to obtain highly reliable data as training data, in the preliminary step S1, it is preferable to evaluate the etch pits after KOH etching and classify the basal plane dislocations.

[0040] In the preliminary step S1, it is preferable to measure the relationship between the X-ray topographic images and the basal plane dislocation density of 100 or more SiC substrates as training data. The more training data there is, the higher the accuracy of estimating the basal plane dislocation density in the estimation step S4. When a sufficient number of SiC substrates cannot be prepared, an image obtained by inverting or rotating an X-ray topographic image of one SiC substrate may be used as part of the training data. For example, a combination of the X-ray topographic image of the first substrate and the basal plane dislocation density of the first substrate may be used as first data, a combination of an image obtained by inverting the X-ray topographic image of the first substrate and the basal plane dislocation density of the first substrate may be used as second data, and a combination of an image obtained by rotating the X-ray topographic image of the first substrate and the basal plane dislocation density of the first substrate may be used as third data.

[0041] The diameters of the multiple SiC substrates measured in the pre-process S1 do not necessarily have to be the same. For example, the multiple SiC substrates measured in the pre-process S1 may include both 6-inch substrates and 8-inch substrates.

[0042] The position of the SiC substrate in the SiC ingot may be added to the training data. The SiC substrate is obtained by cutting out the SiC ingot. The tendency of the basal plane dislocation density to occur may differ between the tail side at the beginning of crystal growth and the head side at the later stage of crystal growth in the SiC ingot. By adding data on the position of the SiC ingot from which the SiC substrate was cut out to the training data, the accuracy of the estimation of the basal plane dislocation density in the estimation step S4 can be improved.

[0043] The SiC substrate evaluation device performs a pre-process S1 to learn the correlation between the X-ray topographic image of the entire SiC substrate and the basal plane dislocation density.

[0044] Next, a SiC substrate to be actually evaluated is prepared. The SiC substrate to be actually evaluated has a basal plane dislocation density of 1.0×10 4 / cm 2 It is preferable that the diameter of the SiC substrate to be actually evaluated is equal to or different from the diameter of the SiC substrate measured in the preliminary step S1. For example, the diameter of the SiC substrate to be actually evaluated may be larger than the diameter of the SiC substrate measured in the preliminary step S1. When the diameter of the SiC substrate to be actually evaluated is equal to the diameter of the SiC substrate measured in the preliminary step S1, the evaluation accuracy is improved.

[0045] In the image acquisition step S2, an X-ray topographic image of the entire first surface of the SiC substrate to be actually evaluated is acquired. The first surface of the SiC substrate from which the X-ray topographic image is acquired may be a chemical mechanical polished (CMP) surface or a ground surface prior to CMP polishing.

[0046] The X-ray topographic image of the SiC substrate measured in the image acquisition step S2 may be a diffraction image including a diffraction vector in the <11-20> direction, or may be a diffraction image including a diffraction vector in the <1-100> direction. The X-ray topographic image of the SiC substrate measured in the image acquisition step S2 is a diffraction image including a diffraction vector in the same direction as the X-ray topographic image of the SiC substrate measured in the preliminary step S1.

[0047] Next, in the image compression step S3, the X-ray topographic image acquired in the image acquisition step is compressed. In the image compression step S3, the pixel size of the X-ray topographic image is compressed to 0.1 mm. 2 If the diameter of the SiC substrate measured in the preliminary step S1 differs from the diameter of the SiC substrate to be actually evaluated, the X-ray topographic images obtained by measuring each SiC substrate are compressed to the same size (e.g., 250 × 250 pixels).

[0048] The resolution of each X-ray topographic image is reduced by performing the image compression step S3. By performing the image compression step S3, the calculation in the estimation step S4 can be performed based on a low-resolution image, and the calculation load of the evaluation device is reduced. If the calculation processing capacity of the evaluation device is sufficiently high, the image compression step S3 does not need to be performed.

[0049] Next, an estimation step S4 is performed. In the estimation step S4, the basal plane dislocation density of the SiC substrate to be evaluated is estimated from an X-ray topographic image of the entire first surface of the SiC substrate to be evaluated based on the learning result of the deep learning.

[0050] The X-ray topograph image used for the image analysis in the estimation step S4 is the X-ray topograph image acquired in the image acquisition step S2. When the image compression step S3 is performed, the X-ray topograph image compressed in the image compression step S3 is used for the image analysis in the estimation step S4.

[0051] The evaluation device, which has been trained by deep learning in the preliminary step S1, inputs an X-ray topographic image of the entire first surface of the SiC substrate, and outputs the basal plane dislocation density of the SiC substrate. Because the correlation between the X-ray topographic image and the basal plane dislocation density has been learned by deep learning, the evaluation device can estimate with high accuracy the basal plane dislocation density of the SiC substrate being evaluated.

[0052] As described above, the method for evaluating a SiC substrate according to this embodiment estimates the basal plane dislocation density of the SiC substrate using a topographic image of the entire surface of the first surface of the SiC substrate. If the basal plane dislocation density is obtained by classifying the defects one by one, it is necessary to classify each defect one by one, and it takes time to measure the basal plane dislocation density. In contrast, the method for evaluating a SiC substrate according to this embodiment can obtain the basal plane dislocation density of the SiC substrate from the overall image rather than classifying each defect one by one, thereby shortening the time required for evaluating the basal plane dislocation density.

[0053] In addition, the topographic image of the entire first surface of the SiC substrate includes information on defects and distortion of the SiC substrate. The method for evaluating a SiC substrate according to this embodiment performs learning based on a topographic image that includes both defect information, which is microscopic information on the SiC substrate, and distortion information, which is macroscopic information on the SiC substrate, and therefore has high estimation accuracy for the basal plane dislocation density.

[0054] In addition, the SiC substrate evaluation method according to the present embodiment can reduce the computational load of the evaluation device that estimates the basal plane dislocation density by performing the image compression step S3. In order to accurately estimate the basal plane dislocation density, it is common to increase the image resolution, but in the present embodiment, it has been found that the basal plane dislocation density can be estimated with high accuracy even if the image resolution is reduced.

[0055] Furthermore, by specifying the pixel size of the image in the image compression step S3, the basal plane dislocation density of the SiC substrate to be evaluated can be estimated even if the diameter of the SiC substrate used in the measurement in the preliminary step S1 differs from the diameter of the SiC substrate to be evaluated. For example, it is difficult to fabricate an 8-inch substrate, and it is difficult to prepare many samples in the preliminary step S1. Even in this case, the preliminary step S1 can be performed with a 6-inch substrate, and the basal plane dislocation density of the 8-inch substrate can be estimated based on the learning results for the 6-inch substrate.

[0056] Furthermore, the SiC substrate evaluation method according to this embodiment can be incorporated into one step of the manufacturing method for SiC substrates and SiC epitaxial wafers. The SiC substrate evaluation method according to this embodiment can easily and quickly evaluate the basal plane dislocation density of a SiC substrate, so that defective products with high basal plane dislocation density can be removed during the manufacturing process, thereby increasing the yield rate of ... 4 / cm 2 It can also be applied when the temperature is as low as below.

[0057] Although the preferred embodiment of the present invention has been described in detail above, the present invention is not limited to the specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims. EXAMPLES

[0058] First, 800 X-ray topographic images of a SiC substrate with a diameter of 150 mm were prepared. Some of these images were then inverted and rotated to prepare 4,000 X-ray topographic images. Each of the 4,000 X-ray topographic images was compressed to 250 x 250 pixels.

[0059] Next, the evaluation device was trained through deep learning to learn the correlation between 4,000 X-ray topograph images and the number of basal plane dislocations in the SiC substrate for each X-ray topograph image. The programming language used for the learning program was Python-based. The program was created using tkinter for the GUI application and Tensorflow-Keras for the deep learning module. Information on the position in the SiC ingot from which the SiC substrate was cut was also added to the training data.

[0060] In Example 1, the basal plane dislocation density of the SiC substrate on the tail side (initial stage of crystal growth) of the SiC ingot was determined. The basal plane dislocation density was determined as both an estimated value using the SiC substrate evaluation method according to this embodiment and an actual measured value. The actual measured value was a value obtained by classifying etch pits after KOH etching and determining the number of basal plane dislocations. The estimated value was determined using an image obtained by acquiring an X-ray topographic image of the SiC substrate to be measured and compressing it to 250 x 250 pixels. FIG. 3 shows the evaluation results of Example 1. The correlation coefficient R of Example 1 2 was 0.6886, confirming a high correlation between the measured and estimated values.

[0061] In Example 2, the basal plane dislocation density of the SiC substrate on the head side (later crystal growth) of the SiC ingot was determined. The basal plane dislocation density was determined both as an estimated value using the SiC substrate evaluation method according to this embodiment and as an actual measured value. The actual measured value was a value obtained by classifying etch pits after KOH etching and determining the number of basal plane dislocations. The estimated value was determined using an image obtained by acquiring an X-ray topographic image of the SiC substrate to be measured and compressing it to 250 x 250 pixels. FIG. 4 shows the evaluation results of Example 2. The correlation coefficient R of Example 2 2 was 0.7254, confirming a high correlation between the measured and estimated values.

[0062] As shown in Examples 1 and 2, the basal plane dislocation density of a SiC substrate can be easily estimated by using the SiC substrate evaluation method according to this embodiment. Even when a compressed image was used, the basal plane dislocation density could be obtained with sufficient accuracy. [Explanation of symbols]

[0063] 10...SiC substrate, N...notch, S1...preliminary process, S2...image acquisition process, S3...image compression process, S4...estimation process

Claims

1. an image acquiring step of acquiring an X-ray topographic image of the entire first surface of the SiC substrate; and estimating a basal plane dislocation density of the SiC substrate from an X-ray topographic image of the entire first surface of the SiC substrate based on a learning result of deep learning.

2. The method further comprises an image compression step of compressing the X-ray topographic image acquired in the image acquisition step, In the image compression step, the pixel size of the X-ray topographic image is reduced to 0.1 mm ２ The above size, The method for evaluating a SiC substrate according to claim 1 , wherein the estimation step is performed using a compressed X-ray topographic image.

3. The method further includes a preliminary step of measuring an X-ray topographic image of each of the plurality of SiC substrates and a basal plane dislocation density of each of the plurality of SiC substrates; The method for evaluating a SiC substrate according to claim 1 , wherein in the preliminary step, a relationship between an X-ray topographic image and a basal plane dislocation density of each SiC substrate is learned by the deep learning.

4. The method for evaluating a SiC substrate according to claim 3 , wherein the number of SiC substrates measured in the preliminary step is 100 or more.

5. The method for evaluating a SiC substrate according to claim 3 , wherein an image obtained by inverting or rotating the X-ray topographic image acquired in the preliminary step is used for the learning.

6. 2. The method for evaluating a SiC substrate according to claim 1, wherein the X-ray topographic image is a diffraction image including a diffraction vector in a <11-20> direction.

7. 2. The method for evaluating a SiC substrate according to claim 1, wherein the X-ray topographic image is a diffraction image including a diffraction vector in a <1-100> direction.

8. The method for evaluating a SiC substrate according to claim 1 , wherein the SiC substrate has a diameter of 145 mm or more.

9. The method for evaluating a SiC substrate according to claim 1 , wherein the SiC substrate has a diameter of 195 mm or more.

10. The method for evaluating a SiC substrate according to claim 3 , wherein a diameter of the SiC substrate measured in the preliminary step is different from a diameter of the SiC substrate for which an X-ray topographic image is obtained in the image obtaining step.

11. The basal plane dislocation density of the SiC substrate is 1.0×10 ４ / cm ２ The method for evaluating a SiC substrate according to claim 1 , wherein:

12. A method for manufacturing a SiC substrate, comprising the method for evaluating a SiC substrate according to any one of claims 1 to 11.

Images