Single crystal silicon manufacturing method

Abstract

To provide a polysilicon rod, etc., capable of improving the poor single crystallization in a FZ method by the ratio of the area of a grain boundary surface corresponding to an interface between particles to a corresponding grain boundary.SOLUTION: In a polysilicon rod, the average of grain boundary characteristics in a region except a seed core in a region of 2 / 3 from the center of the cross section of the polysilicon rod has the following characteristics. A corresponding grain boundary ratio is more than 20%; a grain boundary length is more than 550 mm / mm2; and a random grain boundary length is 800 mm / mm2 or less.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a raw polysilicon for improving the defect rate in single crystal production, and a method for producing the same. [Background technology]

[0002] In the manufacturing of semiconductor devices, the single crystal silicon manufacturing process must maintain productivity by controlling impurities, lattice defects, etc. Currently, the mainstream single crystal manufacturing methods include the FZ (floating zone) method and the CZ (Czochralski) method, but the FZ method is a method in which a polysilicon rod is directly heated by high-frequency heating to obtain a single crystal, and has advantages in terms of impurity control compared to the CZ method, which uses a quartz crucible.

[0003] In the FZ method, defects occur when single crystal growth is hindered, causing dislocations and resulting in crystal defects in the single crystal rod. One of the causes of single crystal growth hinderance is the phenomenon of polysilicon remaining undissolved, causing defects.

[0004] In this FZ method, the crystalline characteristics of the raw polysilicon rod used are closely related to the FZ defects that occur during single crystal production.

[0005] In the single crystal growth process of the FZ method, the occurrence of FZ defects is a major issue because it significantly reduces productivity.

[0006] The Siemens process is the mainstream method for producing polysilicon rods as a raw material for the FZ method, and is a CVD method in which silane gas, the raw material, is deposited in air on a heated silicon rod.

[0007] Patent Documents 1 to 4 disclose polysilicon rods characterized by the area ratio of needle-like crystals and coarse grains and the size of the crystal grains. Patent Documents 5 to 7 disclose polysilicon rods characterized by the Miller indices obtained by X-ray diffraction. <111> or <220> Patent Document 8 discloses a method for selecting a single crystal raw material based on the peak intensity and number of peaks. <222> A polysilicon rod characterized by a diffraction intensity of .gtoreq..gtoreq..times ... [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Japanese Patent Application Laid-Open No. 2008-285403 [Patent Document 2] Japanese Patent Application Laid-Open No. 2013-193902 [Patent Document 3] Japanese Patent Application Laid-Open No. 2014-28747 [Patent Document 4] Japanese Patent Application Publication No. 2017-197431 [Patent Document 5] Japanese Patent Application Laid-Open No. 2013-217653 [Patent Document 6] Japanese Patent Application Laid-Open No. 2015-3844 [Patent Document 7] Japanese Patent Application Laid-Open No. 2016-150885 [Patent Document 8] JP 2019-19010 A DISCLOSURE OF THE INVENTION [Problem to be solved by the invention]

[0009] (1) None of the methods in Patent Documents 1 to 8 provide high quantitativeness and reproducibility. This is because they focus on the coarse particles of polysilicon (size, distribution, crystal orientation, etc.) as the cause of poor single crystallization in the FZ method, which is insufficient.

[0010] The present invention provides polysilicon rods and the like that improve the poor single crystallization caused by the FZ method by utilizing the grain boundary width and the ratio of coincident grain boundaries, which are characteristics of grain boundaries that correspond to the interface between particles.

[0011] For example, the silicon rod with the largest crystal grains is a single crystal silicon rod, and when considering a model of this being made into a single crystal using the FZ method, it can be said that the rate of defects due to the raw materials is zero. When this single crystal is divided, grain boundaries appear. The coincidence grain boundary that is closest to the single crystal bond is Σ3, and grain boundaries that have no coincidence lattice points or no regularity are random grain boundaries, and grain boundaries that contain a lot of Σ3, the bonding plane closest to the single crystal, can be said to be closer to the single crystal.

[0012] (2) Bell-jar reactors are commonly used for CVD reactions using the Siemens process. The reactor's inner walls receive radiation from the heated rod, and if the inner walls are mirror-finished, they have a high reflectivity, effectively returning the radiant energy from the rod to the rod. However, if the inner walls become cloudy, the reflectivity decreases, increasing the absorption of energy by the walls and preventing it from being returned to the rod. The cloudiness is caused by hydrolysis of the raw material chlorosilanes with moisture in the air when the reactor is opened between batches, and the reflectivity tends to decrease with each batch. This makes it difficult to consistently produce polysilicon rods under the same conditions. However, by feeding back the grain boundary characteristics of the previous batch into the reaction conditions for the next batch, it is possible to produce polysilicon with the desired grain boundaries. [Means for solving the problem]

[0013] The inhibiting factor for single crystallization by the FZ method is the characteristics of the grain boundary. By measuring and analyzing this and feeding it back to the manufacturing conditions, it becomes possible to manufacture polysilicon rods suitable for single crystallization by the FZ method.

[0014] Looking at the single crystallization process using the FZ method, the area near the center of the polysilicon rod reaches the single crystal growth surface immediately after melting, so it is easily affected by grain boundaries, while the area near the periphery of the polysilicon rod passes through a heating zone caused by induced current, so it is less affected than the area near the center.

[0015] Specifically, the central area during single crystallization by the FZ method should preferably have a short random grain boundary length and a long grain boundary length, and as it moves away from the center, a shorter grain boundary length is acceptable.

[0016] Therefore, the average of the area excluding the seed core in the 2 / 3 area from the center of the cross section of the polysilicon rod is more than 20% and the grain boundary length is 550 mm / mm 2 of which the random grain boundary length is 800 mm / mm 2 It is also advantageous to use a rod containing polysilicon with a coincidence grain boundary ratio of more than 25% and a grain boundary length of 650 mm / mm. 2 of which the random grain boundary length is 700 mm / mm 2 Polysilicon rods not exceeding 1000 .mu.m are preferred.

[0017] When applied to the entire polysilicon rod, the average of the entire polysilicon rod excluding the seed core is more than 20% and the grain boundary length is 550 mm / mm 2 of which the random grain boundary length is 800 mm / mm 2 It is also advantageous to use a rod containing polysilicon with a coincidence grain boundary ratio of more than 25% and a grain boundary length of 650 mm / mm. 2 of which the random grain boundary length is 700 mm / mm 2 Polysilicon rods not exceeding 1000 .mu.m are preferred.

[0018] The closer the coincidence grain boundary ratio is to 100%, the better, but this manufacturing condition is similar to epitaxial film growth, and current technology has poor cost benefits. Also, when trying to make the grain boundary length long, the random grain boundary length is 700 mm / mm 2 To keep it below this, it is necessary to increase the coincidence grain boundary ratio, and for the reasons mentioned above, the grain boundary length is set to 3000 mm / mm 2 The following is realistic.

[0019] As mentioned above, in the Siemens method for producing polysilicon rods, the environment inside the reactor gradually changes, so it is possible to analyze the polysilicon at regular intervals and feed the results back into the CVD conditions. The coincident grain boundary ratio, which is a characteristic of grain boundaries, the grain boundary length, which is an index of grain boundary width, and the random grain boundary length obtained from these are quantitative values ​​that can be correlated with manufacturing conditions. Furthermore, in product design, it becomes possible to control the grain boundary characteristics from the inner to outer circumference of the polysilicon rod, making it possible to provide polysilicon rods that meet customer requirements. [Effects of the Invention]

[0020] According to one aspect of the present invention, 1. The FZ method reduces the rate of single crystal defects, improves yield and productivity, and 2. Stable production of polysilicon rods becomes possible by feedback of grain boundary characteristics to manufacturing conditions. [Brief explanation of the drawings]

[0021] [Figure 1] FIG. 1 is a diagram showing the relationship between grain boundary length and coincident grain boundary ratio. [Figure 2] A graph showing the relationship between grain boundary length and the Σ3 coincidence grain boundary ratio. [Figure 3] Figure showing images of the Σ3 coincidence grain boundary, the Σ9 coincidence grain boundary, a random grain boundary, and the Σ3-49 coincidence grain boundary. [Figure 4] 1 is a schematic diagram for explaining an overview of a measurement method 1 in an embodiment of the present invention. [Figure 5] FIG. 2 is a schematic diagram for explaining an overview of a measurement method 2 in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] A horizontal plane perpendicular to the growth direction of the polysilicon rod is cut out, and the crystal orientation of all grains exposed on the measurement surface is measured using EBSD (electron backscatter diffraction) in 1μm increments. The state of the grain boundary is calculated from the difference in orientation and angle between adjacent crystals in the resulting data matrix. A Σ3 coincidence boundary is a grain boundary where one coincidence lattice point appears for every three atoms, and is said to be the grain boundary that is closest to a single crystal among all the coincidence grain boundaries. The large number of coincidence lattice points at a grain boundary means that the thermal and physical properties are closer to those of a single crystal.

[0023] Coincident grain boundary ratio Σ3 to Σ49 detected using EBSD analysis software (TSL Solutions Co., Ltd.) are taken as coincidence grain boundaries. Σ3 and Σ9 account for approximately 80% of all coincidence grain boundaries Σ3 to Σ9, with Σ3 being slightly more abundant than Σ9. As the Σ value increases, the spacing between the coincidence lattice points widens, approaching a random grain boundary. Therefore, in this embodiment, the coincidence grain boundary ratio is calculated using the sum of the Σ3 to Σ9 coincidence grain boundaries and is used as an index. Note that Σ1 is a single crystal.

[0024] Since grain boundaries are the boundaries between grains, when the surface is observed, they are obtained as a surface and are shown as an area, but the information obtained by measuring with an actual device is a line (from the time of surface observation, it becomes the length of the boundary line). Therefore, in this embodiment, the coincidence grain boundary ratio is Coincidence boundary ratio = Observed boundary lines of coincidence grain boundaries / Observed boundary lines of grain boundaries (%) (See Figure 3)

[0025] There are boundary lines exceeding Σ49 in the boundary lines. The "boundary lines of the observed grain boundaries" in the above formula refer to all grain boundaries observed by the above EBSD analysis software. In this embodiment, as described above, "Σ3-49" are called coincidence grain boundaries. The boundary lines of the coincidence grain boundaries are approximately 50-60% of the "boundary lines of the observed grain boundaries."

[0026] In EBSD analysis software, for example, at 150x magnification, the orientation (angle) of the crystals on the observation surface is measured at 1µm intervals. If a difference of more than a certain angle is observed in the obtained continuous data, it is considered to be a grain boundary. The corresponding grain boundary "Σ3~49" can be obtained from the orientation and orientation of the crystals on either side of this grain boundary.

[0027] The order of observed grain boundary boundaries is > boundary lines of "Σ3~Σ9 coincidence grain boundaries" > boundary lines of "Σ3~Σ9 coincidence grain boundaries." The observed grain boundary boundaries include coincidence grain boundaries and grain boundaries that are not coincidence grain boundaries. Therefore, the coincidence grain boundary ratio is the sum of the boundary lines of "Σ3~Σ9 coincidence grain boundaries" divided by the sum of the boundary lines of "Σ3~Σ49 grain boundaries" and boundaries beyond Σ49.

[0028] Grain boundaries with a low density of coincidence lattice points (grain boundaries close to random grain boundaries) have high energy and are unstable. Therefore, if there are many grain boundaries with a low density of coincidence lattice points, this can trigger the shedding of unmelted particles on the FZ melting surface, causing FZ defects. On the other hand, by using a polysilicon rod with physical properties close to those of a single crystal as the raw material in the FZ method, stable melting can be achieved.

[0029] Grain boundary length Currently, it is difficult to accurately measure the grain size of single crystals in polysilicon because it is not possible to distinguish grain boundaries in images such as SEM. By measuring the crystal orientation on a grain-by-grain basis using techniques such as EBSD, the length of the grain boundaries on the measurement surface can be obtained, which indirectly expresses the average grain size. Dividing the grain boundary length on the measurement surface by the measured area gives the grain boundary length per unit area. In this embodiment, this is referred to as the grain boundary length (unit: length / area), which is an index of the width of the grain boundary.

[0030] Random grain boundary length Grain boundaries other than the Σ3 to Σ9 coincidence boundaries contain a variety of coincidence boundaries, but as the Σ value increases, the spacing between the coincidence lattice points increases and the characteristics of grain boundaries with low Σ values ​​(low grain boundary energy and stability) are lost. For this reason, for convenience, we define the sum of Σ greater than Σ9 as a random grain boundary, and calculate the random grain boundary length from the grain boundary length per unit area.

[0031] To reduce crystal defects and increase yields in the FZ method, it is best to use raw materials with as long a grain boundary length as possible, a low Σ value, a high coincidence boundary ratio, and a short random grain boundary length. However, the coincidence boundary ratio and grain boundary length are in a trade-off relationship in most cases. For example, polysilicon manufactured under conditions that increase the coincidence boundary ratio will have a short grain boundary length. For this reason, it is important to find the best combination of both grain boundary properties.

[0032] The cause of crystal grain shedding, which inhibits single crystal growth, is weak and unstable bonding at grain boundaries. The more random grain boundaries there are, where there are fewer bonds between coincident lattice points, the more likely they are to peel off from the melting surface. A low Σ value and a high proportion of coincident grain boundaries among the grain boundary properties indicate strong and stable bonding at the grain boundaries, making crystal grain shedding less likely. Furthermore, crystal grain shedding due to random grain boundaries occurs at temperatures lower than the melting temperature of the single crystal due to the high energy of the grain boundaries. Therefore, the shed single crystal particles are not sufficiently heated and melted, and unmelted or semi-melted material reaches the single crystal growth surface in clusters, causing crystal defects. The amount of unmelted or semi-melted material depends on the size of the shed crystal particles; the larger they are, the longer they remain, making them more likely to reach the single crystal growth surface.

[0033] The manufacturing conditions for obtaining the desired grain boundary characteristics include the temperature of the rod surface, the reaction pressure, and the concentration of the raw material silane. Regression analysis of these factors yields the coefficient of determination, R 2 A correlation of 0.8 or higher is obtained. The same is true when more parameters are added and machine learning is used. The obtained correlation can be used as feedback to the equipment to track changes in the state inside the reactor and set optimal reaction conditions.

[0034] While the FZ method is moving towards larger diameter equipment, the traditional small diameter equipment is still widely used. The grain boundary characteristics required for each equipment are different. Furthermore, even equipment of the same type has its own quirks, and by performing this analysis, it is possible to manufacture polysilicon rods tailored to the needs.

[0035] As a measurement method, for example, the embodiment shown in FIG. 4 (hereinafter also referred to as "measurement method 1") may be used. The produced silicon rod is cut and sliced ​​at any desired locations (three locations in the embodiment shown in FIG. 4) to cut out samples. The samples thus obtained are measured. Since the properties are basically the same on both sides of the leg of the U-rod, it is also possible to measure only one side of the leg.

[0036] The measurement results showed that in all the cut samples, the average grain boundary characteristics of the area excluding the seed core in the 2 / 3 region from the center of the cross section of the polysilicon rod exceeded 20% and the grain boundary length was 550 mm / mm 2 The random grain boundary length exceeds 800 mm / mm 2 or the average grain boundary characteristics of the entire polysilicon excluding the seed core have a coincidence grain boundary ratio of more than 20% and a grain boundary length of 550 mm / mm 2 The random grain boundary length exceeds 800 mm / mm 2 If it does not exceed this, it indicates that the FZ yield is good under the conditions.

[0037] Therefore, it is expected that subsequent batches of polysilicon rods produced under the same conditions will also produce favorable results. The reactor interior environment gradually changes with each batch, due to factors such as the loss of gloss on the reactor wall and changes in the efficiency of radiant heat, even under the same conditions, but these changes are not dramatic. Therefore, over a certain period of time (e.g., about one month), it is expected that subsequent batches of polysilicon rods produced under the same conditions will also produce favorable results.

[0038] If the measurement results satisfy the above conditions and are favorable, a good yield can be produced by performing FZ using the leg that has not been cut out by slicing in Figure 4. In Figure 4, the part that was sliced ​​to create the sample can also be a CZ chunk.

[0039] For example, the following steps can be taken:

[0040] Measure using Measurement Method 1, and grow a single crystal in the FZ using the leg opposite to the leg that meets the above conditions (if the manufacturing equipment is the same type, one will be used as the representative).

[0041] At this time, Measurement method 1 is performed on all silicon rods grown on silicon core wires in the same chamber, and if the above conditions are met and the rod passes, single crystals may be grown in the FZ using the leg opposite to the leg that passed. Measurement method 1 is performed on the representative sample from the inside of the chamber as well as the representative sample from the outside. If the sample satisfies the above conditions and passes, the remaining sample may be grown as a single crystal in the FZ. One representative sample is measured using measurement method 1, and if it satisfies the above conditions and passes, the remaining samples may be grown as single crystals by FZ.

[0042] Furthermore, from the viewpoint of quality, the areas near the electrodes and the bridges are cut and the crack-free central portion becomes the ingot for FZ. Therefore, as another measurement method, for example, the embodiment shown in FIG. 5 may be used.

[0043] As shown in Figure 5, in a mode where a sample is prepared by cutting out only the upper portion near the bridge and the lower portion near the electrode (hereinafter referred to as "measurement method 2"), it is also possible to obtain FZ rods from the leg on which the sample was prepared. Sampling for quality evaluation is performed from a portion outside the effective length of the FZ ingot, mainly from the vicinity of the electrode, and analysis of resistance value, metal composition, etc. is performed.

[0044] In the measurement method 2, for example, the following procedure can be taken.

[0045] Measure using Measurement Method 2, and use both the leg that meets the above conditions and the opposite leg to grow a single crystal in the FZ (if the manufacturing equipment is the same type, one leg will be used as the representative).

[0046] At this time, All inspections may be carried out using Measurement Method 2. Those that meet the above conditions and pass the inspection may be used for single crystal growth by the FZ method. Perform Measurement Method 2 on the outer representative in the same way as the inner representative inside the chamber. If it meets the above conditions and passes the inspection, all of them may be used for single crystal growth by the FZ method. Perform Measurement Method 2 on one of the representatives. If it meets the above conditions and passes the inspection, all of them may be used for single crystal growth by the FZ method.

[0047] If the inspection results are different even when the manufacturing devices are of the same type, or if the same silicon rod cannot be manufactured under the same manufacturing conditions, either Measurement Method 1 or 2 may be performed on each device.

[0048] When using the same device but different lots, and when the characteristics gradually disappear, it is considered that deposits have accumulated inside the bell jar, resulting in a decrease in radiant heat.

[0049] Even in this case, the manufacturing conditions may be continuously reviewed, or the inside of the bell jar may be cleaned to return it to its original state. However, when electrolytic polishing is performed for cleaning the inside of the bell jar, it is a costly process, so continuously reviewing the manufacturing conditions is a practical option.

Examples

[0050] <Relationship between FZ Results and Grain Boundary Characteristics> Creation of Polysilicon Rods Crystal samples were created by the Siemens method using trichlorosilane and hydrogen as raw materials. The grain boundary characteristics were measured by EBSD, and the results of the actual pulling-up experiment by the FZ method are shown below. In the case of the single crystalization experiment by the FZ method, if dislocations occurred in the crystal, it was judged as × (failed). The measurement results are also shown in Figure 1.

Table 1

[0051] Sampling for grain boundary characterization Since it is not practical to measure the grain boundary characteristics of the entire rod, the average grain boundary characteristics were determined by sampling. 1) Wafers with a thickness of 10 mm were cut from both ends of the effective length (electrode side and bridge side removed) of a U-rod removed from a Siemens method CVD apparatus (see Figure 5). 2) At the angle formed by the intersection of the lines connecting the longest and closest parts from the outer periphery of the wafer to the seed core, a line segment a was drawn to the outer periphery, bisecting the angle on the acute angle side. 3) Samples were cut out from the core wire at 20 mm intervals along line segment a, and measurements were taken over a measurement range of 0.5 mm x 0.5 mm or more using a TIM EBSD device with a step of 1.0 microns to determine the average grain boundary characteristics. Calculations were also made taking into account the fact that cylindrical grinding was performed in a later process. 4) In the radial direction of growth, in the sections where the reaction conditions (factors affecting grain boundaries such as rod temperature, reaction pressure, raw material concentration, raw material supply rate, CVD equipment, and radiant heat received by the rod from the outside) were the same throughout the reaction batch, the entire area was determined by measuring representative points.

[0052] <Example of reaction condition analysis and feedback> In a polysilicon manufacturing device using the Siemens method, an electric current is passed through a silicon seed core connected to an electrode, generating heat and maintaining a constant temperature. The gas phase is made up of hydrogen and chlorosilane, and a polysilicon deposition layer is formed on the surface of the heated silicon seed core, forming a polysilicon rod. This reaction is taken as an example.

[0053] The reaction conditions were the chlorosilane concentration and the surface temperature of the polysilicon rod during the CVD reaction, and the relationship between these and the grain boundary characteristics was analyzed. The results are shown in a diagram in Figure 2. Assuming that point A is the current condition, if only the chlorosilane concentration is increased from point A, the grain boundary characteristics will change toward point B, and if only the rod temperature is decreased from point A, the grain boundary characteristics will change toward point C. Furthermore, if the chlorosilane concentration is set to point B and the rod temperature to point C, polysilicon with the grain boundary characteristics of point D will be obtained.

[0054] By applying actual measurement results to Figure 2 and keeping it up to date, it is possible to constantly adjust the optimal reaction conditions to obtain polysilicon rods with the desired grain boundary characteristics.

[0055] Feedback to manufacturing conditions method 1: As a method for designing the grain boundary characteristics from the center to the periphery of the polysilicon rod, when the diameter is small, the rod surface temperature is relatively increased to increase the area with a high corresponding grain boundary ratio, and as the diameter increases, the surface temperature of the silicon rod is lowered and the chlorosilane concentration is increased (to prevent the heat inside the silicon rod from rising), thereby increasing the grain boundary length and manufacturing a polysilicon rod with an "optimal area."

[0056] Feedback to manufacturing conditions method 2: As the diameter increases, high frequency waves are applied, the surface temperature is increased (this is possible by preventing the heat inside the silicon rod from rising), and the chlorosilane concentration in the chamber is increased to produce a polysilicon rod with an "appropriate range."

Claims

1. In the average grain boundary characteristics of the entire polysilicon excluding the seed core, the coincidence grain boundary ratio exceeds 20% and the grain boundary length is 550 mm / mm 2 The random grain boundary length exceeds 800 mm / mm 2 A method for producing single crystal silicon, in which a polysilicon rod not exceeding 10 ...

2. The polysilicon rod has a coincidence grain boundary ratio of more than 25% and a grain boundary length of 650 mm / mm 2 The random grain boundary length exceeds 700 mm / mm 2 The method for producing single crystal silicon according to claim 1, wherein the total amount of silicon dioxide does not exceed 1000 ppm.

3. The polysilicon rod has a coincidence grain boundary ratio of not more than 90% and a grain boundary length, which is an index of the width of the grain boundary, of 3000 mm / mm 2 The method for producing single crystal silicon according to claim 1 or 2, wherein the total amount of silicon dioxide does not exceed 1000 ppm.

4. 4. The method for producing single crystal silicon according to claim 1, wherein the polysilicon rod is cut and sliced ​​at arbitrary locations to obtain a plurality of samples, and the polysilicon rod is selected using the coincidence grain boundary ratio, the grain boundary length, and the random grain boundary length in the average of the grain boundary properties of the samples.

5. 4. The method for producing single crystal silicon according to claim 1, wherein a sample is prepared by cutting out an upper portion of the polysilicon rod near the bridge and a lower portion of the polysilicon rod near the electrode, and the polysilicon rod is selected using the coincidence grain boundary ratio, the grain boundary length, and the random grain boundary length in the average of the grain boundary characteristics of the sample.

Images