JP7801010B2 - GaN substrate surface processing method and GaN substrate manufacturing method - Google Patents

Description

The present invention relates to a method for processing the surface of a GaN substrate and a method for manufacturing a GaN substrate.

The market for GaN substrates is expected to expand as substrates for next-generation semiconductor devices. To use GaN substrates in devices, crystals obtained through various bulk crystal growth processes must be cut and formed into the shape of the substrate, and ultimately the substrate surface must be finished to an atomically flat, damage-free, and undisturbed mirror finish. GaN is a highly brittle material that is mechanically hard and chemically stable, making it difficult to polish and known as a difficult-to-process material.

Recent research and development has led to the mirror polishing of GaN substrate surfaces, which is commonly achieved by mechanical polishing such as lapping and polishing followed by chemical mechanical polishing (hereinafter simply referred to as "CMP") using a colloidal silica polishing solution. However, with lapping and polishing, the polishing conditions change as the polishing solution changes over time, and adjusting the gap and parallelism between the wafer and the surface plate takes time. Furthermore, CMP still only achieves a polishing efficiency of a few to tens of nanometers per hour. For this reason, new surface processing processes that can achieve higher efficiency are desired.

For example, Patent Document 1 discloses a grinding device that performs three grinding steps for substrates to be subjected to CMP: rough grinding, medium grinding, and finish grinding, in order to shorten the surface processing time. Specifically, paragraph 0070 of the document describes using grinding wheels with a grit size of #250 to #500 for rough grinding, grinding wheels with a grit size of #1200 to #1800 for medium grinding, and grinding wheels with a grit size of #2500 to #3500 for finish grinding.

JP 2014-65082 A

Patent Document 1 describes an apparatus that can handle a variety of substrates, including sapphire, SiC, and GaN. However, because the polishing conditions for substrates depend heavily on the substrate's material, it is necessary to optimize the surface processing process depending on the material. The grinding apparatus described in Patent Document 1 is said to be able to handle a series of grinding processes for substrates of various materials with a single device in order to shorten processing time. In other words, Patent Document 1 describes highly versatile processing conditions that are not dependent on the material. In recent years, the miniaturization of electronic components has led to demands for higher surface quality for substrates. Meeting this demand requires optimal surface processing processes for each material, but Patent Document 1 has difficulty meeting this demand.

Patent Document 1 also describes the use of CMP polishing after the three-stage grinding described above. However, even in the third stage, finish grinding, a grinding wheel with a grit size of #3500 or less is used, and it is difficult to say that the substrate surface is in a state suitable for CMP. For this reason, if GaN, a known difficult-to-process material, is surface-processed by CMP after grinding with a low-grit wheel, the CMP polishing time will be long, thereby lengthening the total surface processing time. Furthermore, if the CMP polishing time is long, the CMP polishing fluid will seep into the gap between the substrate and the substrate mounting plate, making it more likely to adhere to the backside of the substrate. If the polishing fluid adheres, it will be difficult to remove even in the cleaning process after CMP, raising concerns that a clean substrate may not be obtained.

To shorten the polishing time of CMP, it is generally believed that the surface roughness of the substrate used for CMP polishing must be reduced. One way to achieve this is to use a grinding wheel with a larger grit size. However, to reduce surface roughness using the method described in Patent Document 1, the grit size must be gradually increased. Therefore, to grind with a grit size larger than #3500, the number of grinding steps must be increased to four or more, requiring an enormous amount of time for grinding. Even if the CMP processing time is shortened by increasing the number of grinding steps, the total surface processing time until the CMP polishing is completed increases. GaN, in particular, requires a long surface processing time, and optimizing the grit size and number of steps using the method described in Patent Document 1 requires an enormous amount of time and effort, making it difficult to optimize the surface processing process.

In addition to the above, it is also possible to shorten the polishing time of CMP by optimizing the polishing liquid used in CMP. For example, the material of the free abrasive grains and the chemical components of the polishing liquid can be adjusted taking into account the surface condition of the substrate before CMP. However, since the surface condition of the substrate changes depending on the processing conditions in the pre-CMP process, endless investigations are required to find the optimal conditions. As mentioned above, since the surface processing time for GaN is long, it is difficult to find the optimal conditions, and simply changing the conditions to achieve optimization is not realistic.
As described above, it is difficult to optimize the GaN surface processing process by randomly adjusting the grinding and polishing conditions, and it is not possible to apply the processing processes for substrates other than GaN substrates. For this reason, the optimal final surface processing conditions specifically for GaN have not yet been found.

The present invention aims to provide a GaN substrate surface processing method and a GaN substrate manufacturing method that can process the surface of a GaN substrate in a short time.

The inventors conducted research with the aim of shortening the total surface processing time by reducing the load on CMP, even if this meant a longer grinding time. In order to optimize the processing method before CMP, they investigated the problems with mechanical polishing, which has traditionally been performed in the pre-CMP process. Specifically, the research was conducted using cathodoluminescence (hereinafter referred to as "CL method"), which allows for non-destructive observation of processed substrates. Observing the substrate using CL method revealed that the processing-affected layer is composed of a primary-affected layer and latent scratches. Therefore, the inventors investigated the relationship between the thickness of the primary-affected layer and the depth of latent scratches formed on the surface of a GaN substrate and the processing method.

It is generally known that the thickness of a process-affected layer varies depending on the particle size of the abrasive grains. Therefore, it is thought that the larger the particle size of the abrasive grains, the thicker the process-affected layer. The inventors investigated the behavior of the primary affected layer and latent scratches by performing polishing with loose abrasive grains with an average particle size of 0.5 μm, which is commonly used for mechanical polishing, and grinding with a #3000 grinding wheel with an average particle size of 3 μm, as described in Patent Document 1. Unexpectedly, GaN substrates ground with fixed abrasive grains with an average particle size of 3 μm exhibited a thicker primary affected layer, but shallower latent scratches, and a thinner process-affected layer, compared to GaN substrates ground with free abrasive grains with an average particle size of 0.5 μm. Furthermore, they also found that grinding with fixed abrasive grains caused less damage to the primary affected layer.

This finding can only be obtained with GaN substrates, which allow for observation of latent scratches using the CL method. Conventionally, the cross section of the substrate surface was observed by cutting, but cutting applies stress to the substrate surface, resulting in a different surface condition than before cutting. Therefore, the conventional processing-affected layers and microcracks are different from what should be observed, and are not the main affected layers or latent scratches identified in this study. In particular, with the conventional method of cutting and then observing GaN substrates, which are difficult-to-process materials, the large stresses applied during processing make it impossible to observe the main affected layers and latent scratches that are observed before cutting for cross-sectional observation.

The inventors investigated the optimal grinding conditions for GaN substrates, focusing on the fact that even if the primary affected layer becomes thick, the thickness of the work-affected layer can be reduced by shallowing latent scratches. They found that while substrates ground using the #3000 grinding wheel described in Patent Document 1 had shallower latent scratches than substrates mechanically polished with free abrasive grains having an average grain size of 0.5 μm, this was not enough to enable CMP polishing to be performed in a short time.

The abrasive grain size (average grain size) of the #3000 grinding wheel used in the above study was 3 μm. Although the above study showed that the thickness of the work-affected layer does not necessarily depend on the abrasive grain size, the inventors deliberately performed grinding using a #6000 grinding wheel with an abrasive grain size (average grain size) of 1.5 μm and observed the surface condition of the substrate. As a result, it was found that with the #6000 grinding wheel, the primary affected layer became thinner and the latent scratches became significantly shallower. It was also found that damage to the primary affected layer was also reduced. It was also found that the CMP processing time was shortened, ultimately resulting in a reduction in the total surface processing time. Furthermore, it was found that depending on the grit size, the difference between the thickness of the primary affected layer and the depth of latent scratches became smaller, further shortening the CMP processing time.
The present invention, which was completed based on these findings, is as follows.

(1) A GaN substrate surface processing method for processing the surface of a GaN substrate by grinding and polishing, comprising: a high-grit grinding step in which the surface of the GaN substrate is ground with a grinding wheel having a grit size of #6000 or greater; and a CMP polishing step in which, after grinding the surface of the GaN substrate in the high-grit grinding step, the surface of the GaN substrate is polished by CMP.

(2) The surface processing method for a GaN substrate according to (1) above, further comprising, before the high-grit grinding step, a rough grinding step of grinding the GaN substrate with a grinding wheel having a grit size of less than #6000, or a mechanical polishing step of polishing the GaN substrate with free abrasive grains having a grain size of more than 0.5 μm.
(3) The method for processing the surface of a GaN substrate according to (1) or (2) above, further comprising a cleaning step of cleaning the GaN substrate after the CMP polishing step.
(4) A surface processing method for a GaN substrate according to any one of (1) to (3) above, wherein the grinding stone used in the high-grit grinding step has a grit size of greater than #8000.
(5) A surface processing method for a GaN substrate according to any one of (1) to (4) above, wherein the grinding wheel is bonded with vitrified bonding.

(6) A method for manufacturing a GaN substrate, comprising the GaN substrate surface processing method described in any one of (1) to (5) above.

1A and 1B are process diagrams showing an example of a manufacturing process for a GaN substrate, where FIG. 1A is a process diagram showing an example of a conventional manufacturing process for a GaN substrate, and FIG. 1B is a process diagram showing an example of a manufacturing process for a GaN substrate to which the GaN substrate surface processing method according to this embodiment is applied. FIG. 2 is a diagram showing the processing method and evaluation items of the substrate surface before CMP, and the slurry conditions for CMP and evaluation items of the substrate surface. FIG. 3 is a schematic diagram showing a device for capturing a cathodoluminescence image (hereinafter referred to as a "CL image" where appropriate). FIG. 4 shows the polishing time of the CMP and the CL images of the substrate surface at each time for a substrate that has been mechanically polished using 0.5 μm diamond abrasive grains and then CMP-polished. FIG. 5 is a graph showing the relationship between the polishing time of CMP and the black line density. FIG. 6 is a diagram showing an example of a means for widening the field of view of the CL image and confirming the presence of black lines after the CMP polishing time of 420 minutes in FIG. FIG. 7 is a table showing the relationship between the grit size and the surface roughness in the high-grit grinding step before CMP. FIG. 8 shows the CMP polishing time and CL images of the substrate surface at each time for a substrate that has been ground using an abrasive stone with a grit size of #8000 and then CMP-processed. FIG. 9 shows the polishing time of the CMP and the CL images of the substrate surface at each time for a substrate that has been ground using a grindstone with a grit size of #30000 and then CMP-processed. Figure 10 is an image diagram showing the degree of alteration of the main altered layer and the depth of latent scratches, where Figure 10(a) is an image diagram showing the damage to the substrate surface after mechanical polishing using 0.5 μm diamond abrasive grains as described in Comparative Example 1, Figure 10(b) is an image diagram showing the damage to the substrate surface after grinding using a grinding stone with a grit size of #30000 as described in Example 3, and Figure 10(c) is an image diagram showing the damage to the substrate surface after grinding using a grinding stone with a grit size of #8000 as described in Example 2.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail with reference to the accompanying drawings, but the present invention is not limited to the following embodiments.
1. Overview of the GaN Substrate Manufacturing Method According to the Present Invention FIG. 1 is a process diagram showing an example of a GaN substrate manufacturing process, with FIG. 1(a) being a process diagram showing an example of a conventional GaN substrate manufacturing process, and FIG. 1(b) being a process diagram showing an example of a GaN substrate manufacturing process to which the GaN substrate surface processing method according to the present embodiment is applied.

As shown in Figure 1(a), a conventional method for manufacturing a GaN substrate (hereinafter sometimes simply referred to as a "substrate") involves first growing a GaN crystal on a GaN seed crystal using, for example, vapor phase epitaxial growth or liquid phase growth (S11). Next, the grown crystal is fixed to a jig and outer diameter grinding is performed to remove protrusions (S12). After outer diameter grinding, the grown crystal is sliced to a thickness of, for example, 100 to 3000 μm (S13), yielding a GaN substrate. Next, chamfering is performed to remove edges created during slicing (S14), and rough grinding is performed to remove surface irregularities on the GaN substrate (S15).

Then, mechanical lapping is performed using loose abrasive grains (S16), followed by precision polishing using loose abrasive grains such as diamond (S17). CMP is then performed to remove any altered layers formed on the surface of the substrate (S18), and finally the substrate is cleaned (S19).

In the lapping (S16) and precision polishing (S17) steps shown in Figure 1(a), the polishing liquid can wear out the surface plate, and the polishing liquid can change over time. Increasing the polishing pressure or rotation speed of the surface plate shortens the polishing time, but this also increases damage to the surface plate, requiring more frequent surface plate maintenance and reducing productivity. Meanwhile, to prevent the formation of a processing-affected layer, it is necessary to reduce the abrasive grain size used in lapping (S16) and precision polishing (S17). However, the smaller the abrasive grain size, the longer the polishing time, and it becomes more difficult to control grain size variation. As such, it has been difficult to shorten the total surface processing time with conventional surface processing methods.

In contrast, as shown in Figure 1(b), a GaN substrate manufacturing method incorporating the GaN substrate surface processing method of the present invention performs high-grit grinding (S27) using a grinding wheel with abrasive grains of #6000 or greater (average grain size of the abrasive grains of 1.5 μm or less) instead of the lapping (S16) and precision polishing (S17) of Figure 1(a). This reduces the total surface processing time. Note that in Figure 1(b), the processes other than the high-grit grinding (S27) and CMP polishing process (S28) can generally be performed under the same conventional GaN substrate manufacturing conditions as in Figure 1(a). The GaN substrate surface processing method of the present invention is described in detail below.

2. GaN Substrate Surface Processing Method According to the Present Invention The surface processing method according to the present invention is a method specialized for GaN substrates, which are difficult-to-process materials, and includes a high-grit grinding step (S27) of grinding the surface of the GaN substrate with a grinding stone having a grit size of #6000 or more (average grain size of abrasive grains of 1.5 μm or less), and a CMP polishing step (S28) of polishing the surface of the GaN substrate by CMP after grinding the surface of the GaN substrate in the high-grit grinding step (S27). Each step will be described in detail.

2-1 High-grit Grinding Step In the high-grit grinding step (S27) of the present invention, the surface of the GaN substrate is ground using a grinding stone of #6000 or larger (average grain size of abrasive grains of 1.5 μm or smaller).
The high-grit grinding step (S27) of the present invention is performed using a conventional grinding device. The substrate fixed to a surface plate is ground with a grinding wheel attached to a motor. Grinding conditions such as the rotation speed of the grinding wheel and the pressure applied to the substrate are not particularly limited, but as long as the grinding wheel grit falls within the ranges described below, the thickness of the work-affected layer and the depth of latent scratches can be reduced regardless of the rotation speed, pressure, or feed rate. In this embodiment, the grinding time for the high-grit grinding step is preferably 1 to 20 minutes, and more preferably 1 to 5 minutes. The grinding rate may be 5 to 60 μm/min, and may be 5 to 20 μm/min.

The grit size of the abrasive grains is #6000 or higher (average grain size of the abrasive grains is 1.5 μm or less). A grit size of #6000 or higher causes less damage to the substrate, thinner primary altered layers, and less deep latent scratches, shortening the CMP polishing time and reducing the total surface processing time. Furthermore, replacing mechanical polishing with mechanical grinding facilitates automated manufacturing of high-grit grinding processes and shortens the total surface processing time. Furthermore, in the present invention, the grain size of the abrasive grains can be measured using a particle size distribution analyzer using laser diffraction, dynamic light scattering, image analysis, gravitational sedimentation, or other methods.

There are no particular limitations on the material of the abrasive grains. Various abrasive grains can be used, including alumina, alumina zirconia, silicon carbide, CBN, and diamond.

If the grinding wheel used in the high-grit grinding process (S27) has a small grit size, the grinding is completed quickly due to the large abrasive grain size, but the damage to the substrate is significant and the latent scratches are deep. Therefore, the CMP polishing process (S28) must be extended to remove the latent scratches. On the other hand, if the grit size is large, the grinding time is slightly longer, but the damage to the primary altered layer is small and the latent scratches are shallow, thereby shortening the CMP polishing process (S28). Since the polishing time by CMP is significantly longer than the time required for mechanical grinding, shortening the CMP polishing time can shorten the total surface processing time. The grit size of the grinding wheel is preferably greater than #8000 (average abrasive grain size less than 1.0 μm), more preferably #10000 or greater (average abrasive grain size 0.7 μm or less), and most preferably #13000 or greater (abrasive grain size 0.5 μm or less). In particular, when using a high grit size of #13000 or higher, the relationship between the grain size of the abrasive grains and the hardness of GaN results in long polishing times, which has traditionally been avoided. However, in this invention, the knowledge that the polishing time of the CMP polishing process described below can be shortened was first obtained through observations of GaN using the CL method. As a result, the total processing time was shortened.

While there is no particular upper limit, increasing the grit size makes it possible to reduce the thickness of the primary altered layer and the depth of latent scratches during the grinding process, further shortening the time for the CMP polishing process. On the other hand, if the grit size is increased too much, the increased grinding time will be longer than the shortened CMP polishing time. For this reason, simply increasing the grit size does not necessarily shorten the total surface processing time from grinding to polishing. From the perspective of shortening this total time, #50000 or less (average grain size of abrasive grains of 0.1 μm or more) is preferable, and #30000 or less (average grain size of abrasive grains of 0.2 μm or more) is even more preferable.

The binder for the grinding wheel can be any commonly used material. It must be able to withstand grinding resistance, and when the abrasive grains wear down, they must be able to break down due to increased grinding resistance, allowing the abrasive grains to self-regenerate appropriately. Examples include resinoid, rubber, metal, shellac, and vitrified, but vitrified is preferred as it changes little over time and has excellent durability.

Stress during processing causes a primary alteration layer and latent scratches to form on the surface of a surface-processed substrate. Substrates ground using the high-grit grinding step (S27) of the present invention are conventionally ground using a high-grit grinding wheel to reduce the degree of alteration in the primary alteration layer (hereinafter referred to as "damage" as appropriate). For GaN substrates, using a large-grit grinding wheel, as in the high-grit grinding step (S27), results in a primary alteration layer thickness of approximately 700 to 1000 nm. However, grinding with a high-grit grinding wheel results in minimal damage to the primary alteration layer, resulting in less alteration and less deep latent scratches. The depth of the latent scratches is approximately 1.0 to 2.0 μm from the surface of the substrate.

The thickness of the primary altered layer can be determined by multiplying the polishing time and polishing rate at which the brightness of the CL image, in areas other than the latent scratches, becomes equivalent to that of the as-grown crystal during CMP polishing, using a CL image (described below). The as-grown crystal in this invention is obtained by crystal growth (S11) shown in Figure 1(b). Therefore, in this invention, a CL image of the obtained GaN surface is captured using an SEM (described below), and the average brightness data (average pixel value) of the captured CL image is pre-obtained as the brightness of the as-grown crystal using image analysis software (SM-300 Series) provided with the SEM. The thickness of the primary altered layer can be determined by multiplying the polishing time and polishing rate at which the brightness becomes equivalent to that of the as-grown crystal during CMP polishing. The depth of the latent scratches can be determined by multiplying the polishing time and polishing rate of the CMP polishing. The polishing rate of CMP is described below.

Furthermore, the surface roughness of the substrate ground in the high-grit grinding step (S27) is preferably 2.0 nm or less, more preferably 1.0 nm or less, even more preferably 0.5 nm, particularly preferably 0.1 nm or less, and most preferably 0.05 nm or less. When performing surface processing of GaN substrates as in the present invention, if the surface roughness after the high-grit grinding step (S27) is preferably 2.0 nm or less, damage is minimal and latent scratches do not become deep. Therefore, it is easy to achieve a surface roughness of approximately 0.1 nm or less after surface processing is completed with a short CMP polishing time (S28).

A GaN substrate processed by the GaN substrate surface processing method of the present invention has a depth of latent scratches formed on the surface that is reduced by 30% or more, preferably 40% or more, and more preferably 60% or more, compared to a GaN substrate processed by the conventional processing method shown in Figure 1(a). Since there is no significant difference in the polishing ability of the CMP polishing step (S28) described below, the polishing time for the CMP polishing step (S28) becomes shorter as the latent scratches become shallower. Because the majority of the surface polishing time is spent in the CMP polishing step (S28), shortening this polishing time also shortens the total surface processing time.

Figure 2 shows the processing method and evaluation items for the substrate surface before CMP, as well as the CMP slurry conditions and evaluation items for the substrate surface. As shown in Figure 2, the surface roughness of a substrate polished using conventional mechanical polishing can be observed using a scanning electron microscope (SEM), atomic force microscope (AFM), transmission electron microscope (TEM), or cathodoluminescence (CL). The thickness of the primary altered layer and the depth of latent scratches can be observed using the CL method.

2-2. CMP Polishing Step The CMP polishing step (S28) of the present invention is carried out after the surface of the substrate is ground in the high grit grinding step (S27).
A typical method for the CMP polishing step (S28) is to attach the substrate to a carrier, press the substrate against a polishing pad, and rotate both the substrate and the polishing pad while supplying a polishing liquid between the substrate and the polishing pad.

The polishing solution used in the CMP polishing process (S28) generally contains abrasive grains such as alumina or silica, and is acidified with an oxidizing agent such as hydrogen peroxide or persulfuric acid. The oxidizing agent oxidizes the surface of the substrate, and the resulting oxide film is then removed with the abrasive grains to achieve polishing. The abrasive grain size is 100 nm or less, and the abrasive grain concentration should be 20 to 60 mass% of the total mass of the polishing solution. The abrasive grain size can be measured using the same method as for the abrasive grains in grinding wheels.

The conditions for the CMP polishing step (S28) are not particularly limited, but the rotation speed of the carrier is 20 to 1000 rpm, and the supply rate of the polishing liquid is 50 to 1000 ml/h. Under these conditions, the polishing rate falls within a predetermined range. For example, the polishing rate may be 50 to 3000 nm/h. Latent scratches remain on the obtained substrate, but as shown in FIG. 2, the presence or absence of latent scratches can be confirmed using a cathode luminescence image. The polishing time may be a time at which the density of the black lines visible in the cathode luminescence image described below is expected to be, for example, 1 cm ^- 2 or less.

Figure 3 is a schematic diagram showing a device for capturing a cathodoluminescence image (hereinafter referred to as a "CL image") . The CL device 10 comprises an electron beam irradiation device (electron beam generator) 20, a sample stage 30, a (CL light) detector 40, and a computing device (controller) 50. The electron beam irradiation device 20 (electron beam generator) irradiates an electron beam 21 onto a substrate 60 placed on the sample stage 30. The (CL light) detector 40 detects CL light 22 emitted from the substrate 60. The computing device (controller) 50 performs various processes based on the data detected by the (CL light) detector 40. Furthermore, since the electron beam irradiation device (electron beam generator) 20, sample stage 30, and computing device (controller) 50 may be attached to an SEM, the CL device 10 may be built into the SEM. The calculation device (control device) 50 includes a CPU (Central Processing Unit) that performs various processes, memory, non-volatile storage, input devices such as a keyboard and microphone, a monitor, and an input/output interface. This hardware can perform the following calculations using various functions (not shown) that are controlled by programs stored in the storage device.

In the cathodoluminescence (CL) method, when an electron beam is irradiated onto the surface of a GaN crystal material, CL light is observed, which is due to the radiative recombination process near the GaN crystal surface. Evaluation using the CL method mainly uses the electron beam irradiation function installed in an SEM, and a CL light intensity mapping image (hereinafter simply referred to as a "CL image") can be obtained, similar to an SEM image. Using this CL image, latent scratches present below the substrate surface can be visualized as black lines, rather than processing scratches present on the substrate surface. Therefore, this method is extremely effective as a method for determining whether or not processing damage exists within the surface of a GaN substrate, as reported, for example, in the following document (Hideo Aida, Hidetoshi Takeda, Koji Koyama, Haruji Katakura, Kazuhiko Sunakawa, and Toshiro Doi, "Chemical Mechanical Polishing of Gallium Nitride with Colloidal Silica", Journal of the Electrochemical Society, 158 (12) H1206-H1212 (2011)).
An example of calculating the polishing time of the CMP polishing step (S28) from the change over time of the CL image will be described in detail.

The CL method makes it possible to visually evaluate the presence or absence of latent scratches. Because the observation area of a CL image is generally several tens of microns square, it is difficult to guarantee the absence of latent scratches over a wide area, such as the entire surface of a substrate. Furthermore, latent scratches are observed in CL images as extremely small dotted lines less than 0.5 microns wide.

Thus, it is necessary to make it possible to observe with a realistic field of view that can be achieved with conventional devices, and to make it possible to accurately observe even extremely slight altered layers.
For example, a typical CL image size that allows efficient image acquisition by temporarily pausing processing during processing is assumed to be approximately 35 μm x 50 μm. Immediately after the start of CMP, black lines cannot be recognized, but as processing approaches the middle, the number of black lines observed in the CL observation area decreases to a countable level. When processing has progressed to the point where not a single black line can be observed in this observation area, the black line density becomes approximately 10 ⁴ cm ^-2 .

In subsequent observations, the black line density becomes approximately 10 ⁴ cm ⁻² or less, and the CL observation area must be expanded, so the observation area can be expanded by moving the measurement stage on which the substrate is placed vertically and horizontally to determine the black line density. The black line density is the value obtained by dividing the number of black lines observed in the observation area by the area of the observation area.

In the present invention, the black line density at each CMP polishing time is plotted from the middle to the end of processing, and an approximate line is drawn based on the black line density information within the range where the black line density is 10 ⁴ cm ⁻² or less. The polishing time is estimated as the time when the black line density reaches 1 cm ⁻² as the CMP polishing time, and the depth of the latent scratches can be calculated by multiplying the CMP polishing rate by the polishing time.

In the present invention, the black line density observable in a general observation area is measured from the polishing time when the black line density becomes 10 ⁶ cm ⁻² or less, and the decrease in the black line density with the passage of polishing time is grasped within the polishing time when the black line density becomes 10 ⁴ cm ⁻² to 1 cm ⁻² . In this way, the CMP polishing time can be estimated with high accuracy.

If the CMP polishing time is short, the CMP polishing fluid will not get between the substrate and the substrate mounting plate, preventing the polishing fluid from adhering to the back surface of the substrate. In this embodiment, a high-grit grinding process suitable for processing GaN substrates is performed before the CMP polishing process, thereby shortening the CMP polishing time. In this embodiment, the CMP polishing time is preferably 1 to 30 hours. This is a significantly shorter processing time than conventional processing methods.

In this way, by understanding the decreasing trend of the black line density, it is possible to estimate the polishing time and polishing amount required to achieve a black line density of 1 cm ^-2 , and this polishing amount corresponds to the depth of the latent scratches. By performing the high-grit grinding (S27) shown in Figure 1(b) using various grits in a similar manner, it is possible to accurately predict the grit size and the depth of the latent scratches in the GaN substrate.

The thickness of the primary altered layer can be measured by multiplying the polishing rate of CMP polishing by the polishing time required to obtain a CL image of the area other than the latent scratches observed during CMP polishing, where the brightness of the CL image is equivalent to the brightness of the as-grown crystals obtained as described above. However, when the grinding wheel used in high-grit grinding (S27) is #50000 or smaller (average abrasive grain size of 0.1 μm or larger), the depth of the latent scratches will be equal to or greater than the thickness of the primary altered layer. Therefore, when the depth of the latent scratches can be estimated using the CL method as described above, the primary altered layer can be removed by removing the latent scratches, without measuring the thickness of the primary altered layer.

2-3. Rough Grinding Step or Mechanical Polishing Step The GaN substrate surface processing method according to the present invention may include a rough grinding step (S25) in which the surface of the GaN substrate is ground using a grinding stone having a grit size of less than #6000 (average grain size of abrasive grains exceeding 1.5 μm) before the high-grit grinding step (S27). The rough grinding step (S25) can be performed using a grinding stone having a grit size smaller than that used in the high-grit grinding step (S27). Alternatively, instead of the rough grinding step (S25), a mechanical polishing step (S25) in which the GaN substrate is polished using free abrasive grains having an average grain size exceeding 0.5 μm may be included.

The rough grinding step (S25) is carried out using a standard grinding device, just like the high-grit grinding step (S27). The thickness of the primary altered layer and the depth of latent scratches do not need to be considered in the rough grinding step (S25). There are no particular restrictions on the grinding conditions, but the rotation speed of the surface plate should be 100-500 rpm, the rotation speed of the grinding wheel should be 200-1000 rpm, and the feed rate should be 5-50 μm/min. The grinding time should be 1-5 minutes for a substrate with a diameter of 2 inches, for example, and can be determined appropriately depending on the size of the substrate.

The upper limit of the grit size of the abrasive grains used in the grinding stone in the rough grinding step (S25) is preferably less than #6000 (average grain size of the abrasive grains exceeds 1.5 μm), more preferably #4000 or less (average grain size of the abrasive grains is 2.0 μm or more), and even more preferably #3000 or less (average grain size of the abrasive grains is 3.0 μm or more). The grain size of the abrasive grains is the same as in the high-grit grinding step (S27). The lower limit of the grit size of the abrasive grains is not particularly limited, but is preferably #240 or more (average grain size of the abrasive grains is 127 μm or less), more preferably #400 or more (average grain size of the abrasive grains is 75 μm or less), and even more preferably #600 or more (average grain size of the abrasive grains is 30 μm or less).
When the rough grinding step (S25) is performed, the grinding time of the high-grit grinding step (S27) is shortened, and ultimately the total surface processing time may be shortened.
The material of the abrasive grains, the binder of the grinding wheel, and the concentration of the abrasive grains in the grinding wheel are the same as those in the high-grit grinding step (S27).

The mechanical polishing step (S25) is performed in the same manner as in the conventional method. The substrate is attached to a carrier, and the substrate is pressed against a polishing pad while a polishing liquid is supplied between the substrate and the polishing pad, while both the substrate and the polishing pad are rotated. The polishing liquid used in the mechanical polishing step (S25) contains free abrasive grains with a particle size exceeding 0.5 μm. The abrasive grains may be made of, for example, diamond, alumina, or silica. The abrasive grain concentration may be 1 to 20 mass% of the total mass of the polishing liquid. The particle size of the abrasive grains is preferably 1.0 μm or more, and more preferably 1.5 μm or more. There is no particular upper limit, but it is sufficient if it is 5 μm or less. The particle size of the abrasive grains is defined in the same way as that of abrasive grains in a grinding wheel, with the sieve mesh corresponding to the maximum particle size. The components of the polishing liquid other than the abrasive grains may be the same as in the conventional method.

The conditions for the mechanical polishing step (S25) are not particularly limited, but a polishing rate of 50 to 300 nm/h is sufficient. The carrier rotation speed is 20 to 2000 rpm, and the polishing liquid supply rate is 1 to 20 ml/h. The polishing rate is 1 to 30 μm/h.

2-4. Cleaning Step The GaN substrate surface processing method according to the present invention may include a cleaning step (S29) of cleaning the GaN substrate after the CMP polishing step (S28).
In the cleaning after the CMP polishing step (S28), it is preferably adopted in order to prevent the polishing liquid used in CMP from remaining and contaminating the substrate. As the detergent used in the cleaning step (S29), an alkaline cleaning liquid that electrostatically repels the substrate and abrasive grains is generally considered to be effective, but it is not limited to this.

The substrate can be cleaned using a conventional method. For example, after the CMP polishing step (S28), the substrate is placed on a spinner and scrubbed for 20 to 60 seconds while a cleaning solution is supplied to the substrate at a flow rate of approximately 50 to 300 ml/min. This cleaning method can be performed using a commercially available cleaning machine. In this embodiment, the CMP polishing time is short, so the CMP polishing solution does not penetrate between the substrate and the substrate mounting plate, and the polishing solution is less likely to adhere to the backside of the substrate. This allows the polishing solution to be easily removed in this cleaning step, resulting in a clean substrate. If the polishing time in the CMP polishing step is long, the polishing solution will begin to adhere, making it impossible to clean within the above-mentioned cleaning time. In this embodiment, as mentioned above, the high-grit grinding step that precedes the CMP polishing step is suitable for GaN substrates, so the CMP polishing step time can be shortened, thereby enabling the cleaning time to be shortened as well.

An example of a surface polishing method specifically for GaN substrates will be described.
As an example, the total surface processing time required for the surface processing methods of the example and comparative example was investigated based on FIGS.

1. Comparative Example 1
1) Preparation of GaN Substrates As shown in Figure 1(a), GaN crystals were grown by vapor-phase epitaxial growth (S11). CL images of the grown GaN crystals were taken using a scanning electron microscope (SEM, Topcon Corporation, Model SM-300) equipped with a CL photodetector at an acceleration voltage of 10 kV, a probe current of 90, a working distance (WD) of 22.5 mm, and a magnification of 2000x. From the CL images, the average brightness data (average pixel values) of the CL images was obtained as the brightness of the as-grown crystal using the image analysis software (SM-300 Series) provided with the SEM. Then, after outer diameter grinding (S12), the substrate was sliced (S13), and the edges were chamfered (S14) to prepare circular GaN substrates with a thickness of 400 μm and a diameter of 2 inches.

2) Rough Grinding The GaN substrate was fixed to a surface plate and ground for 5 minutes using a grinding wheel with a grit size of #600 (average particle size: 30 μm) at a feed rate of 20 μm/min (S15). The average particle size of the abrasive grains used in the grinding wheel was measured using a dynamic light scattering particle size distribution analyzer (LB-500 manufactured by Horiba, Ltd.). The average particle size was calculated from the obtained particle size distribution.

3) Lapping, precision polishing (mechanical polishing)
Next, use the polishing liquid that the concentration of diamond abrasive grains with an average particle size of 3 μm is 10 mass% relative to the total mass of the polishing liquid, the supply amount of the polishing liquid is 10 ml/h, and the polishing rate is 20 μm/h, carry out lapping processing for 120 minutes (S16).Then, use the polishing liquid that the concentration of diamond abrasive grains with an average particle size of 0.5 μm is 10 mass% relative to the total mass of the polishing liquid, and carry out precision polishing for 180 minutes, and the polishing rate is 1 μm/h (S17).It should be noted that the average particle size of the abrasive grains used in grinding wheel is the average particle size measured using above-mentioned device under the same conditions.

4) Observation of the substrate surface The substrate surface was observed before CMP polishing. To observe the substrate surface, an atomic force microscope (AFM) was used to measure the surface roughness (Ra), and the surface irregularities were expressed as shades of gray. Ra was 0.4 nm, indicating that the surface irregularities were minimal.

CL images of the substrate surface were also taken. CL images were taken using a scanning electron microscope (SEM, Topcon Corporation, Model SM-300) equipped with a CL photodetector. The images were taken at an accelerating voltage of 10 kV, a probe current of 90, a working distance (WD) of 22.5 mm, and a magnification of 2000x. The results are shown in Figure 10(a). The short arrow indicates the thickness of the primary altered layer 110, and the long arrow indicates the depth of the latent scratches 120. The thickness of the primary altered layer was approximately 200 nm, while the depth of the latent scratches reached a maximum of 2.4 μm. The depth of the latent scratches corresponds to the amount of polishing by CMP, as described below, and is obtained by multiplying the CMP polishing time and polishing rate, as described below. The thickness of the primary altered layer was obtained by multiplying the polishing time at which the brightness of the CL image in the area other than the latent scratches became equivalent to the brightness of the as-grown crystal by the CMP polishing rate.

5) CMP Polishing: Using a polishing solution containing 35-45% by mass of silica abrasive grains with an average particle size of 60 nm and acidified with hydrogen peroxide, the polishing was performed at a carrier rotation speed of 30 rpm, a polishing solution supply rate of 50 ml/h, and a polishing rate of 180 nm/h (S18). Subsequently, as shown in Figure 4, CL images of the substrate surface were taken every predetermined polishing time. CL images were observed using a scanning electron microscope (SEM, Topcon Corporation, Model SM-300) equipped with a CL photodetector at an accelerating voltage of 10 kV, a probe current of 90, a working distance (WD) of 22.5 mm, and a magnification of 2000x. The particle size of the silica abrasive grains was the average particle size measured under the same conditions using the above-mentioned equipment.

The black line density was calculated from the number of black lines in each CL image taken as shown in Figures 4(a) to 4(e), and the data was plotted as shown in Figure 5, with the black line density on the vertical axis and the polishing time on the horizontal axis. When the CMP polishing time was less than 90 minutes and the black line density was 2 x 10 ⁶ cm ^-2 or more, it was difficult to measure the black line density in the 2000x CL image. When the CMP polishing time was 90 minutes or more but less than 420 minutes, it was possible to measure the black line density. When the polishing time was 420 minutes or more, the number of black lines in the observation area was one. When the CMP polishing time exceeded 420 minutes, almost no black lines were visible in the 2000x CL image, so the black line density was measured while enlarging the field of view.

Observing black lines in a CL image requires a magnification of approximately 2000x. Therefore, it is difficult to accurately calculate the black line density by observing a wide area with CL. Therefore, while maintaining the 2000x magnification of the CL image, the observation area was moved and still images of the CL image were captured at any desired position. Based on the obtained images, the black line density over a wide area was confirmed and the polishing time was estimated. All CL images shown on the right side of Figure 6 were taken at 2000x magnification, and the letters in each CL image indicate the capture position of the CL image within the field of view for each CMP polishing time shown on the left side of Figure 6. Similarly, the black line density for each polishing time was plotted as shown in Figure 5. A linear fit was performed using the plot for black line densities of 10 ⁴ cm ⁻² or less. It was found that the polishing time for the CMP polishing step (S18), where the black line density was 1 cm ⁻² or less, was approximately 800 minutes (796 minutes). Therefore, it was determined that 800 minutes × 180 nm/h ≈ 2400 nm of thickness was sufficient for CMP polishing. Therefore, the depth of the latent scratch can be estimated to be about 2400 nm.

The thickness of the primary altered layer was calculated by taking CL images using the same scanning electron microscope as in S11 at each CMP polishing time. For areas other than latent scratches in the CL images, the polishing time at which the brightness of the CL image became equivalent to the brightness of the as-grown crystals obtained after S11 was multiplied by the polishing rate of 180 nm/h for CMP polishing. As a result, the thickness of the primary altered layer in Comparative Example 1 was 200 nm.

6) Cleaning After the CMP polishing was completed, the substrate was cleaned using an alkaline cleaning solution (S19).

2. Example 2
1) Preparation of GaN Substrate and Rough Grinding A roughly ground substrate was obtained through the same steps as in Comparative Example 1 (S21 to S25).

2) High-grit Grinding Next, the substrate after rough grinding was subjected to grinding for 2 minutes at a grinding rate of 10 μm/min using a grinding wheel with a grit size of #8000 (average grain size: 1.0 μm) (S27). The average grain size of the abrasive grains used in the grinding wheel was the average grain size measured under the same conditions using the above-mentioned device.

3) Observation of Substrate Surface The substrate surface was observed before CMP polishing. A non-contact surface profiler NewView 7300 manufactured by Zygo was used to observe the substrate surface, and the surface roughness (Ra) was measured. The results are shown in FIG. 7. The surface roughness Ra was 1.3 nm, which was found to be greater than the surface roughness of Comparative Example 1.

Furthermore, as in Comparative Example 1, the thickness of the primary altered layer on the substrate surface and the depth of the latent scratches were measured. The results are shown in Figure 10(c). The short arrow indicates the thickness of the primary altered layer 130, and the long arrow indicates the depth of the latent scratches 140. The thickness of the primary altered layer was found to be approximately 900 nm, which is thicker than Comparative Example 1. Meanwhile, the maximum depth of the latent scratches was 1500 nm, which is approximately 40% less than Comparative Example 1. This is thought to result in a 40% reduction in CMP polishing time.

4) CMP Polishing CMP polishing was performed using the same process as in Comparative Example 1. As shown in Figures 8(a) to 8(e), in each CL image taken at 2000x magnification as in Comparative Example 1, the black line density was calculated from the number of black lines, and the vertical axis was plotted with black line density and the horizontal axis was plotted with polishing time. When the polishing time was 390 minutes, the number of black lines in the observation area was one. To estimate the polishing time, CL images were taken with a wider observation field as in Comparative Example 1, and the black line density for each polishing time was plotted. A linear fit was performed using the plots where the black line density was 10 ⁴ cm ^-2 or less. As a result, it was found that the polishing time required to achieve a black line density of 1 cm ^-2 or less was approximately 500 minutes, demonstrating a significant reduction in the total surface processing time compared to Comparative Example 1.

5) Cleaning After the CMP polishing was completed, the substrate was cleaned in the same manner as in Comparative Example 1 (S29).

3. Example 3
In Example 2, the surface of the substrate was processed in the same manner as in Example 1, except that the high-grit grinding stone was changed from #8000 to #30000 (average grain size: 0.2 μm). The average grain size of the abrasive grains used in the grinding stone was measured under the same conditions using the above-mentioned device.
The substrate surface was observed and the surface roughness (Ra) was measured before CMP polishing in the same manner as in Example 2. The results are shown in Figure 7. The surface roughness Ra was 1.8 nm, which was found to be greater than the surface roughness in Example 2.

Furthermore, as in Comparative Example 1, the thickness of the primary alteration layer on the substrate surface and the depth of the latent scratches were measured. The results are shown in Figure 10(b). The short arrow indicates the thickness of the primary alteration layer (150), and the long arrow indicates the depth of the latent scratches (160). The thickness of the primary alteration layer was found to be approximately 700 nm, which is thicker than Comparative Example 1. Meanwhile, the depth of the latent scratches was a maximum of approximately 1000 nm, which is approximately 60% less than Comparative Example 1. This also shortens the CMP polishing time. Therefore, it is believed that the CMP polishing time will be significantly shortened.

Polishing by CMP was performed using the same process as in Example 2. As shown in Figures 9(a) to 9(d), the black line density was calculated from the number of black lines in each CL image taken at 2000x magnification as in Comparative Example 1, and the vertical axis was the black line density, and the horizontal axis was the polishing time. At a polishing time of 300 minutes, the number of black lines in the observation area was one. To estimate the polishing time, CL images were taken with a wider observation field as in Comparative Example 1, and the black line density for each polishing time was plotted. A linear fit was performed using the plots where the black line density was 10 ⁴ cm ^-2 or less. The polishing time required for the black line density to reach 1 cm ^-2 or less was 330 minutes, demonstrating a significant reduction in the total surface processing time compared to Comparative Example 1.
After the CMP polishing was completed, the substrate was cleaned in the same manner as in Comparative Example 1.

4. Example 1
In Example 2, the surface of the substrate was processed in the same manner as in Example 2, except that the grinding stone for high grinding was changed from #8000 to #6000 (average particle size: 1.5 μm). The average particle size of the abrasive grains used in the grinding stone was measured under the same conditions using the above-mentioned device.
As in Example 2, the substrate surface was observed before CMP polishing, and the surface roughness (Ra) was measured. The results are shown in Figure 7. The surface roughness Ra was 1.0 nm, which was found to be smaller than the surface roughness of Example 2. The thickness of the main altered layer was 1000 nm, and the depth of the latent scratches was 2000 nm. Compared to Comparative Example 1, the depth of the latent scratches was significantly shallower.
After the CMP polishing was completed, the substrate was cleaned in the same manner as in Comparative Example 1.

5. Comparative Example 2
In Example 2, the surface of the substrate was processed in the same manner as in Example 2, except that the high-grit grinding stone was changed from #8000 to #3000 (average grain size: 3.0 μm). The average grain size of the abrasive grains used in the grinding stone was measured under the same conditions using the above-mentioned device.
The substrate surface was observed before CMP polishing, and the surface roughness (Ra) was measured in the same manner as in Example 2. The surface roughness Ra was 5.0 nm, which was found to be greater than the surface roughness of Example 2. The thickness of the primary altered layer was 2500 nm, and the depth of the latent scratches was 3000 nm. It was found that the primary altered layer was thicker and the latent scratches were deeper than in any of the other Examples.
After the CMP polishing was completed, the substrate was cleaned in the same manner as in Comparative Example 1.

Figure 10 is an image showing the degree of deterioration of the primary alteration layer and the depth of latent scratches. Figure 10(a) is an image showing damage to the substrate surface 100 after mechanical polishing using diamond abrasive grains with an average particle size of 0.5 μm as described in Comparative Example 1. Figure 10(b) is an image showing damage to the substrate surface 100 after grinding using a grinding stone with a grit size of #30000 as described in Example 3. Figure 10(c) is an image showing damage to the substrate surface 100 after grinding using a grinding stone with a grit size of #8000 as described in Example 2. In these figures, the primary alteration layer and latent scratches are schematically shown in the near-surface region of the cross section of the substrate, and the magnitude of damage to the primary alteration layer is expressed as a shade based on the brightness of the CL image. To make the degree of damage easier to recognize, the damage is quantified using a relative ratio to Figure 10(a).

As shown in Figure 10(a), the substrate whose surface was processed using the conventional process had a thin primary altered layer, but the brightness of the CL image was low, indicating significant damage. Furthermore, for latent scratches, the brightness of the CL image was low and black lines were observed deep from the substrate surface 100, indicating that significant damage had reached the deepest depths.

In contrast, as shown in Figure 10(c), the substrate whose surface was processed using the process of Example 2 had a primary alteration layer that was approximately three times thicker than that of Comparative Example 1, but the brightness of the CL image was lower than that of Comparative Example 1, and the damage to the primary alteration layer was small. If we assume that the damage shown in Figure 10(a) is 1, the damage in Figure 10(c) was 0.4. It was also found that the depth of latent scratches was significantly reduced.

Furthermore, as shown in Figure 10(b), although the thickness of the primary altered layer of the substrate subjected to surface processing using the steps of Example 3 was more than twice as thick as that of Comparative Example 1, the brightness of the CL image was even lower than that of Example 2, and damage was significantly reduced. Assuming that the damage shown in Figure 10(b) is 1, the damage in Figure 10(b) was 0.1. It was also found that the depth of latent scratches was further reduced.

10 Cathodoluminescence (CL) device 20 Electron beam irradiation device (electron beam generator)
21 electron beam 22 CL light 30 sample stage 40 (CL light) detector 50 arithmetic unit (control unit)
60 Substrate 100 Substrate surface 110, 130 150 Thickness of main affected layer 120, 140 160 Depth of latent scratch

Claims (6)

A method for processing the surface of a GaN substrate by grinding and polishing, comprising:
a high-grit grinding step of grinding the surface of the GaN substrate with a grinding wheel having a grit size of #6000 or more;
a CMP polishing step of polishing the surface of the GaN substrate by CMP after grinding the surface of the GaN substrate by the high-grit grinding step,
In the high-grit grinding step, a process-affected layer composed of a primary affected layer and latent scratches is formed on the surface of the GaN substrate,
The process-affected layer has a thickness of the main affected layer of 700 nm or more and 1000 nm or less, and a depth of the latent scratches of 1000 nm or more and 2000 nm or less, and
a surface roughness Ra of the GaN substrate after the high-grit grinding step and before the CMP polishing step is 1.0 nm or more and 1.8 nm or less ;
the primary affected layer is a layer having a thickness determined by multiplying a polishing time and a polishing rate at which the brightness of the cathode luminescence image of an area other than the latent scratches becomes equal to the brightness of an as-grown crystal measured in advance, using a cathode luminescence image, in the CMP polishing step;
the latent scratches have a depth determined by multiplying the polishing time of the CMP polishing step by the polishing rate . Before the high-grit grinding step,
a rough grinding step of grinding the GaN substrate with a grinding wheel having a grit size of less than #6000; or
2. The method for processing the surface of a GaN substrate according to claim 1, further comprising a mechanical polishing step of polishing the GaN substrate with free abrasive grains having an average grain size exceeding 0.5 μm. The GaN substrate surface processing method according to claim 1 or 2, further comprising: a cleaning step of cleaning the GaN substrate after the CMP polishing step. The GaN substrate surface processing method according to any one of claims 1 to 3, wherein the grinding wheel used in the high-grit grinding step has a grit size greater than #8000. The GaN substrate surface processing method according to any one of claims 1 to 4, wherein the grinding wheel is bonded with vitrified cement. A GaN substrate manufacturing method comprising the GaN substrate surface processing method described in any one of claims 1 to 5.