Method for detecting minute defects in a substrate
The method enhances the detection of minute defects on semiconductor substrates by repeated measurements and threshold setting to improve sensitivity and precision, addressing issues of haze and false defect identification in existing technologies.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- SHIN ETSU HANDOTAI CO LTD
- Filing Date
- 2023-02-08
- Publication Date
- 2026-06-02
AI Technical Summary
Existing methods for detecting minute defects on semiconductor substrates suffer from issues such as increased haze due to pretreatments, altered defect morphology, and low detection recall rates leading to misidentification of false defects, especially when attempting to detect defects smaller than the detection limit of surface inspection devices.
A method involving repeated measurements at the highest sensitivity to avoid noise-induced LPDs, calculating defect rates, and setting thresholds to ensure accurate detection of minute defects, thereby improving throughput and precision.
Enables high-sensitivity and high-precision detection of minute defects on semiconductor substrates, excluding noise-induced false defects and ensuring accurate identification of actual defects, particularly on mirror-polished silicon single-crystal substrates.
Smart Images
Figure 0007868525000006 
Figure 0007868525000007 
Figure 0007868525000008
Abstract
Description
[Technical Field]
[0001] The present invention relates to a method for detecting minute defects in a substrate. [Background technology]
[0002] In recent years, with the miniaturization and increased integration of process rules, the control of crystal defects in semiconductor crystals, the precision of semiconductor substrate processing, and the high-precision detection and evaluation of minute defects on the surface of semiconductor substrates have become increasingly important. In particular, in advanced processes where EUV (extreme ultraviolet) is used for semiconductor device formation, it is easy to imagine that even defects as small as 10 nm can become device killers.
[0003] In this context, semiconductor substrate surface inspection equipment is becoming capable of detecting even smaller defects, such as those smaller than 20 nm in size, with high sensitivity. However, in state-of-the-art semiconductor manufacturing processes, it is anticipated that defects can occur due to minute flaws that would not be detected even if the detection limit of a surface inspection device were set to 15 nm.
[0004] By making such minute defects apparent through various pretreatments such as heat treatment and etching, it is possible to detect defects smaller than the detection limit of a surface inspection device. For example, Patent Document 1 discloses a technique for detecting defects smaller than the detection limit by applying heat treatment after LPD (Light Point Defect) measurement to form an oxide film, and then performing LPD measurement again.
[0005] Furthermore, Patent Document 2 discloses a technique for surface inspection of an object under inspection in which, when an arbitrary defect candidate is selected from a group of detected defect candidates, the group of defect candidates is classified into a group of defect candidates having feature quantities (amounts dependent on scattered light, defect candidate coordinates, or defect candidate detection recall) less than or equal to the selected defect candidate, and a group of defect candidates having feature quantities greater than the selected defect candidate, and a threshold is calculated based on the classification result to determine whether or not there is a defect on the surface of the object under inspection. [Prior art documents] [Patent Documents]
[0006] [Patent Document 1] Japanese Patent Publication No. 2003-142544 [Patent Document 2] Japanese Patent Publication No. 2009-156573 [Overview of the Initiative] [Problems that the invention aims to solve]
[0007] However, the detection methods requiring pretreatment, including the technology described in Patent Document 1, generally tend to worsen haze, and measurements after heat treatment required a sensitivity equivalent to or lower than that of the As-grown substrate. Furthermore, because the defect morphology changes due to heat treatment, there was a problem in that it was not possible to evaluate the defects themselves that existed in the As-grown substrate.
[0008] Furthermore, Patent Document 2 discloses a method of repeatedly measuring defects on the wafer surface to set a threshold for not detecting false defects. However, if the lower limit of the measurement conditions is reduced to detect very small defects, even such small defects can be detected, the detection recall rate becomes low, leading to misidentification as false defects and uniform exclusion from the list of defect candidates.
[0009] The present invention was made to solve the above problems, and aims to provide a method for detecting minute defects on a substrate surface that can detect minute defects on the substrate surface with high sensitivity and high accuracy. [Means for solving the problem]
[0010] The present invention has been made to achieve the above objective, and provides a method for detecting minute defects in a substrate, comprising: a first step of repeatedly performing measurements on one substrate using a surface inspection device at the highest sensitivity without detecting LPD caused by noise due to the surface roughness of the substrate to detect defects; a second step of calculating the total number of defects relative to the number of measurements and the rate of increase of the total number of defects relative to the number of measurements from the detection results of the first step; a third step of calculating the number of measurements required so that the rate of increase is less than or equal to a predetermined threshold; and a fourth step of repeatedly performing measurements on another substrate manufactured under the same manufacturing conditions using the surface inspection device at the highest sensitivity for the required number of measurements to detect defects.
[0011] This method for detecting minute defects detects defects under conditions that increase sensitivity while avoiding the detection of noise-induced LPDs due to surface roughness by detecting defects at the highest sensitivity. The decrease in the recall rate of minute defect detection due to the increased sensitivity is compensated for by performing multiple measurements. Furthermore, by limiting the number of measurements to a number where the rate of increase in defects when measurements are repeated falls below a threshold, the increase in the number of detected false defects caused by noise that occur by increasing the number of measurements is suppressed, and throughput is improved by limiting the number of measurements.
[0012] Therefore, with this method, even with inspection sensitivities where detection recall is not guaranteed by the surface inspection equipment manufacturer, it is possible to detect and evaluate minute defects with low detection recall as actual defects while excluding noise-induced LPDs, thereby achieving both improved throughput and high-precision measurement.
[0013] In this case, the substrate can be a silicon single crystal substrate that has been mirror-polished.
[0014] In this way, by detecting minute defects as actual defects on the mirror-polished surface of silicon single-crystal substrates, which are widely used as semiconductor crystal substrates, it is possible to detect minute defects that could become device killers, especially in advanced processes where EUV (extreme ultraviolet) is used for semiconductor device formation.
[0015] In this case, the highest sensitivity at which the noise-induced LPD is not detected in the first step can be set by first measuring one substrate with the recommended highest sensitivity, which is the highest sensitivity recommended by the surface inspection apparatus, or with an arbitrary measurement sensitivity where the detection limit is greater than the recommended highest sensitivity, then gradually decreasing the detection limit at arbitrary intervals and performing the measurement to obtain a map of detected defects, ending the measurement when the noise-induced LPD is detected in the obtained map, determining the highest sensitivity at which the noise-induced LPD is not detected from the measurement results, and then decreasing the detection limit to the determined highest sensitivity.
[0016] In this way, by gradually increasing the sensitivity from a sensitivity with high detection accuracy but a large detection limit, and lowering the detection limit, and setting the sensitivity just before LPD is detected as the highest sensitivity, it is possible to easily set the highest sensitivity in which pseudo-defects caused by the surface roughness of the substrate are less likely to be detected.
[0017] In this case, when calculating the total number of defects in the second step, the coordinate data of the defects obtained in each measurement can be compared, and all defects within a specific coordinate range can be considered the same defect.
[0018] In this way, by treating defects within a specific coordinate range obtained in each measurement as the same defect, defects detected at the same location as defects already detected in previous measurements are considered to be the same as previously detected defects.
[0019] Therefore, when calculating the total number of defects, defects that have already been detected are not counted twice, and it is possible to avoid counting more defects than the actual total number of defects.
[0020] At this time, when calculating the increase rate of the total number of defects in the second step, the total number of defects up to the Nth measurement in the measurements repeatedly performed in the first step S N And the total number of defects up to the N+1th measurement, S N+1 Using the aforementioned increase rate PN Let [%] be
Number
[0021] In this way, by setting the growth rate at a certain number of measurements as the change rate of the total number of defects with respect to the number of measurements measured at the previous number of measurements, the growth rate can be easily obtained with high precision.
Advantages of the Invention
[0022] As described above, by using the method for detecting minute defects on a substrate of the present invention, defects on the substrate surface can be detected with higher sensitivity, and smaller defects can be detected with high precision. Therefore, minute defects on the substrate surface can be detected with high sensitivity and high precision.
Brief Description of the Drawings
[0023] [Figure 1] A flowchart as an example of the method for detecting minute defects on a substrate according to an embodiment of the present invention is shown. [Figure 2] This is an example of the experimental results shown in this example, showing the relationship between the number of measurements and the total number of defects. [Figure 3] This is an example of the experimental results shown in this example, showing the relationship between the number of measurements and the growth rate of the total number of defects.
Modes for Carrying Out the Invention
[0024] Hereinafter, the present invention will be described in detail, but the present invention is not limited thereto.
[0025] As described above, there has been a demand for a method capable of detecting minute defects on the substrate surface with high sensitivity and high precision.
[0026] As a result of diligent study on the above-mentioned problems, the present inventors have found that a method for detecting minute defects on a substrate can detect minute defects on the substrate surface with high sensitivity and accuracy, and have completed the present invention. This method is characterized by comprising: a first step of repeatedly performing measurements on one substrate using a surface inspection device at the highest sensitivity without detecting LPD caused by noise due to the surface roughness of the substrate; a second step of calculating the total number of defects relative to the number of measurements and the rate of increase of the total number of defects relative to the number of measurements from the detection results of the first step; a third step of calculating the number of measurements required so that the rate of increase is below a predetermined threshold; and a fourth step of repeatedly performing measurements on another substrate manufactured under the same manufacturing conditions using the surface inspection device at the highest sensitivity for the required number of measurements.
[0027] The following describes a method for detecting minute defects in a substrate according to an embodiment of the present invention, with reference to the drawings.
[0028] Figure 1 shows a flowchart illustrating an example of a method for detecting minute defects in a substrate according to an embodiment of the present invention. First, prepare the substrate to be used for defect detection. The substrate is not particularly limited as long as surface defects need to be detected, but a silicon single crystal substrate can be used as an example. In this case, the surface of the silicon single crystal substrate is polished to a mirror finish.
[0029] Thus, by using a mirror-polished silicon single-crystal substrate, which is widely used as a semiconductor crystal substrate, as the target for defect detection, and by detecting minute defects on the mirror-polished surface as actual defects, it is possible to detect minute defects that could become device killers, especially in advanced processes where EUV (extreme ultraviolet) is used for semiconductor device formation. Unless otherwise specified, the following explanation will use the case where a mirror-polished silicon single-crystal substrate is prepared as the substrate.
[0030] In this context, "defects" refer to crystal defects that exist as-grown (grown-in defects), and examples include LEP (Large Etching Pit), COP (Crystal Originated Particle), and OSF (Oxidation Induced Stacking Fault).
[0031] The method for manufacturing a mirror-polished silicon single crystal substrate is not particularly limited, but examples of processing methods include manufacturing a silicon single crystal ingot using the Czochralski process, cutting the manufactured ingot with a wire saw or the like, and then polishing both sides.
[0032] The method of mirror polishing is not particularly limited as long as a mirror surface with the desired surface roughness can be obtained, but for a silicon single crystal substrate, one example is a method in which the substrate is sandwiched between upper and lower polishing plates to which polishing cloth is attached, and the upper and lower polishing plates are rotated while supplying polishing compound to polish both sides of the substrate.
[0033] (1st step) In the first step shown in Figure 1, a single substrate is repeatedly measured using a surface inspection device at the highest sensitivity, without detecting LPD caused by noise due to the surface roughness of the substrate, to obtain the defect coordinates. The surface inspection device used for defect detection is not particularly limited as long as it is capable of detecting defects of the desired size. However, when detecting defects in a silicon single-crystal substrate, a wafer surface defect inspection device, such as a particle counter, can be given as an example. An example of such a wafer surface defect inspection device is the KLA-Tencor Surfscan SP7XP.
[0034] LPDs caused by noise due to substrate surface roughness are false defects resulting from the misdetection of noise caused by the substrate surface roughness. These LPDs are detected uniformly across the entire measurement range, or in a characteristic shape (e.g., a fan shape) at a specific angle.
[0035] If the sensitivity is increased to the point where noise-induced LPDs (Low-Pressure Disorders) are generated, depending on the shape of the LPD, the noise-induced LPDs may cover the measurement range, hindering defect detection. Therefore, measurements are performed at the highest sensitivity setting, avoiding the detection of noise-induced LPDs due to the surface roughness of the substrate. This allows for the detection of minute defects under conditions that increase sensitivity while avoiding the detection of noise-induced LPDs.
[0036] There are no particular limitations on how to set the highest sensitivity so as not to detect noise-induced LPDs, but the following methods can be used as examples. First, a single substrate is measured using a surface inspection device with a high detection reproducibility, using the recommended maximum sensitivity (the highest sensitivity recommended for the surface inspection device) or any measurement sensitivity with a lower detection limit greater than the recommended maximum sensitivity, i.e., a measurement sensitivity within the recommended range, and the defect coordinates are obtained.
[0037] The recommended maximum sensitivity is, for example, the maximum sensitivity recommended by the manufacturer of the surface inspection equipment. The manufacturer's recommended maximum sensitivity refers to the manufacturer's recommended maximum sensitivity, which is adjusted so that the detection recall rate of the smallest particle size defect among the defects detected by measuring a reference wafer multiple times is higher than a threshold (e.g., 95%).
[0038] In terms of specific sensitivity, when using the KLA-Tencor Surfscan SP7XP as a surface inspection device, measurements can be taken at 12.5 nm Up in Oblique mode. Subsequently, measurements are taken by gradually decreasing the detection limit at arbitrary intervals, and a map of the detected defects is obtained.
[0039] The measurement is terminated when noise-induced LPDs are detected in the resulting map. Specifically, as the detection limit is gradually reduced, depending on the surface roughness of the substrate, LPDs (Low-Profile Disorders) with a characteristic shape (e.g., a fan shape) may be detected uniformly across the entire map or at a specific angle. Since these LPDs are clearly false defects caused by misdetection of noise due to the substrate's surface roughness, the measurement is terminated as soon as these LPDs are detected.
[0040] After the measurement is complete, the highest sensitivity at which no noise-induced LPDs are detected is determined from the measurement results. Specifically, the highest sensitivity should be one level lower than the sensitivity at which noise-induced LPDs were detected.
[0041] Finally, the maximum sensitivity is set by reducing the detection limit to the highest sensitivity level, that is, the highest sensitivity level at which no LPDs are detected uniformly in the plane or at a specific angle in the resulting map. By reducing the detection limit to the highest sensitivity level at which no noise-induced LPDs are detected in this way, it is possible to set a maximum sensitivity level at which false defects are less likely to be detected.
[0042] Furthermore, the narrower the interval for reducing the detection limit, the higher the maximum sensitivity can be achieved, while the wider the interval, the shorter the measurement time can be. Therefore, it is appropriate to select the interval appropriately based on the balance between the required maximum sensitivity and measurement time. The above explains how to set the highest sensitivity.
[0043] Once the highest sensitivity is set, the coordinate data of the defects is obtained by repeatedly performing measurements to detect the defects. At this time, the number of repetitions is not particularly limited, but it is at least 2 times, and from an operational standpoint, it is preferable to be 30 times or less. More preferably, it can be 15 to 30 times.
[0044] Here, the set maximum sensitivity is higher than the manufacturer's recommended maximum sensitivity, within the range where noise-induced LPD is not detected. In principle, increasing the measurement sensitivity beyond the manufacturer's recommended maximum sensitivity makes it possible to detect defects smaller than the minimum particle size at the manufacturer's recommended maximum sensitivity, but this results in a lower detection reproducibility.
[0045] In other words, if the sensitivity is increased beyond the manufacturer's recommended maximum sensitivity, there is a higher probability that actual minute defects will not be detected in a single measurement. Therefore, to compensate for the detection recall rate, it is effective to repeatedly perform measurements to detect defects.
[0046] Furthermore, increasing measurement sensitivity means that even items with a low signal-to-noise ratio will be identified as defects. In other words, the risk of misidentifying noise as a defect (known as a false defect) increases.
[0047] This pseudo-defect is, (1) By making the detection limit too small, the threshold for the signal-to-noise ratio that is judged as a defect becomes small, and scattering caused by the surface roughness of the substrate is detected as a defect. (2) Those caused by noise that occurs accidentally, They can be broadly divided into two types.
[0048] It is known that false defects (1) can be detected as LPDs uniformly across the entire map or as characteristic shapes (e.g., fan-shaped) at a specific angle, depending on the surface roughness of the substrate, and can be easily identified. Furthermore, in the first step, the highest sensitivity is set so that false defects (1), i.e., noise-induced LPDs, are not detected, so even if the measurement for detecting defects is repeated, the possibility of detecting false defects (1) is extremely low.
[0049] Regarding pseudo-defects (2), it is presumed that they occur randomly with a certain probability during measurement, making them difficult to identify from the map, and the detection recall rate is also estimated to be low. Therefore, by setting a separate threshold for detection recall, it is possible to efficiently extract defects that are highly likely to be false defects. For example, a defect that is detected only once out of 30 measurements can be labeled as having a high probability of being a false defect.
[0050] (2nd process) Next, in the second step, the total number of defects relative to the number of measurements, and the rate of increase of the total number of defects relative to the number of measurements, are calculated based on the detection results from the first step, specifically the defect coordinates obtained in the first step. The total number of defects referred to here is the sum of the number of defects detected in the measurements up to that point.
[0051] However, when calculating the total number of defects in the second step, the coordinate data of the defects obtained from the repeated measurements in the first step can be compared, and all defects within a specific coordinate range can be considered the same defect. As a result, defects detected at the same location as defects already detected in previous measurements are treated as the same as the already detected defects and are not added together.
[0052] Therefore, when calculating the total number of defects, defects that have already been detected are not counted twice. Thus, it is possible to avoid counting more defects than the actual total number of defects.
[0053] The specific coordinate range is not particularly limited, but it is preferably 50 to 100 μm, taking into consideration the defect density of the substrate and the coordinate accuracy of the surface inspection device.
[0054] Furthermore, when calculating the rate of increase in the total number of defects in the second process, S is the sum of defects up to the Nth measurement in the measurements repeatedly performed in the first step. N And the total number of defects up to the N+1th measurement, S N+1 Using the growth rate P N [%]of,
number
[0055] In this way, by defining the rate of increase in a given number of measurements as the rate of change in the total number of defects compared to the number of measurements taken in the previous measurement, the rate of increase can be easily and accurately determined.
[0056] (3rd step) Next, in the third step, the number of measurements required to ensure that the rate of increase in the total number of defects calculated in the second step falls below a predetermined threshold is set. In the repeated measurements with low sensitivity and detection recall performed in the first step, the total number of defects approaches a certain value (the number of actual defects) as the number of measurements increases. However, assuming that pseudo-defects (2) (those caused by noise that occur by chance) occur with a certain probability, it is presumed that the number of pseudo-defects (2) will increase as the number of repeated measurements increases.
[0057] In other words, the sum of the pseudo-defects (2) that can be detected to a small extent in each measurement increases monotonically. Therefore, it can be estimated that, after a certain number of measurements, the rate of increase in pseudo-defects will become greater than the rate of increase in actual defects.
[0058] Therefore, by setting a threshold for the rate of increase of the total number of defects and limiting the number of measurements required to the number of measurements at which the rate of increase falls below the threshold, it is possible to prevent an increase in unnecessary false defects caused by increasing the number of measurements, thereby achieving both improved throughput and high-precision measurement, making it efficient. This threshold is not particularly limited, but a larger threshold will result in fewer false defects being detected, while a smaller threshold is expected to detect the majority of actual defects. Therefore, the threshold should be appropriately selected based on the balance between the number of false defects detected and the number of actual defects, but it is preferably set to less than 1%.
[0059] (4th step) Next, in the fourth step, another silicon single crystal substrate manufactured under the same manufacturing conditions is subjected to a surface inspection device, and the number of measurements required in the third step, at the highest sensitivity set in the first step, is repeated to detect defects and obtain the coordinate data of the defects. This makes it possible to detect defects with low detection reproducibility with high sensitivity, efficiency, and high accuracy. The reason for targeting substrates manufactured under the same manufacturing conditions is that if the manufacturing conditions differ, the measurement conditions, such as the highest sensitivity that does not detect LPD caused by noise due to surface roughness, which was calculated in the first step, will change. [Examples]
[0060] The present invention will be described in detail below with reference to examples, but this is not intended to limit the present invention. Defects in a mirror-polished silicon single-crystal substrate were detected using the substrate micro-defect detection method of the present invention. The number of defects detected was compared when defects were detected at the manufacturer's recommended measurement sensitivity, and when the detection method and sensitivity were the same as the present invention, but the number of measurements was fewer. The specific procedure is as follows.
[0061] (Example 1) First, as samples for evaluation, we manufactured silicon single crystal ingots using the Czochralski method, and then processed the manufactured ingots into wafers and polished them to a mirror finish to prepare three 300mm diameter mirror-polished silicon single crystal substrates.
[0062] Next, using one of the three silicon single-crystal substrates prepared, the first step involved using a surface inspection device (KLA-Tencor Surfscan SP7XP, Oblique mode) to detect defects on the silicon single-crystal substrate surface at a depth of 12.5 nm Up, which is within the recommended measurement sensitivity range, in order to determine the highest sensitivity.
[0063] Next, using the same surface inspection device, defects present on the silicon single-crystal substrate surface were detected in Oblique mode by gradually increasing the sensitivity. From the results obtained, the highest sensitivity at which LPDs caused by noise due to the substrate surface roughness were not detected was determined to be 10.6 nm Up. Subsequently, the defect detection measurement was repeated 15 times at 10.6 nm Up.
[0064] Next, in the second step, the total number of defects relative to the number of measurements, and the rate of increase of the total number of defects relative to the number of measurements were calculated. Specifically, as shown in Figure 2, the sum of defects was calculated by summing the number of defects detected for each measurement result over 15 measurements. Furthermore, the rate of increase P per measurement was calculated from the total number of defects. N This was calculated as the rate of change in the total number of defects compared to the number of measurements taken in the previous measurement, as shown in Figure 3. Furthermore, as a third step, when the threshold for the rate of increase of the total number of defects was set to 1%, the minimum number of measurements required for the threshold to be less than 1% was calculated. As shown in Figure 3, the rate of increase of the total number of defects became less than 1% after 11 measurements, so the number of measurements was determined to be 11.
[0065] Next, in the fourth step, one of the two silicon single crystal substrates prepared (excluding the substrate used in the first step) was measured 11 times using the same inspection equipment as in the first step, in the same Oblique mode, with a sensitivity of 10.6 nm Up, to detect defects.
[0066] As a result, 234 defects were detected. When the defects were observed using a scanning electron microscope based on the obtained coordinate data, even defects with low detection recall were observed as defects, confirming that they were actual defects.
[0067] (Comparative Example 1) Of the three silicon single-crystal substrates prepared in Example 1, the remaining substrate that was not used for measurement was measured once using the same surface inspection device as in Example 1, in the same Oblique mode, at 12.5 nm Up, which is within the manufacturer's recommended measurement sensitivity range. As a result, 31 defects were detected. Although this silicon single-crystal substrate is of the same quality as in Example 1, the number of defects is low, indicating that even smaller defects could not be detected.
[0068] (Comparative Example 2) The silicon single-crystal substrate evaluated in Comparative Example 1 was measured once using the same surface inspection device as in Example 1, in the same Oblique mode, with a detection limit of 10.6 nm Up, which is lower than the manufacturer's recommended measurement sensitivity. As a result, 147 defects were detected. Compared to Example 1, the number of defects was lower, indicating that defects could not be detected with high accuracy.
[0069] (Comparative Example 3) The silicon single-crystal substrates evaluated in Comparative Examples 1 and 2 were measured three times at the same detection limit as in Example 1, specifically at 10.6 nm Up, which is lower than the manufacturer's recommended measurement sensitivity. As a result, 193 defects were detected. Compared to Example 1, the number of defects was lower, indicating that defects could not be detected with high accuracy.
[0070] The results for Example 1 and Comparative Examples 1-3 are shown in Table 1.
[0071] [Table 1]
[0072] As described above, according to the embodiment of the present invention, even with an inspection sensitivity for which the detection recall rate is not guaranteed by the surface inspection equipment manufacturer, the sensitivity is set to exclude noise-induced LPDs, and the number of measurements is set to a threshold of the increase rate that allows for the detection of minute defects with low detection recall rates as actual defects and avoids the counting of unnecessary false defects, thereby enabling the detection and evaluation of minute defects as actual defects, and enabling the detection of minute defects on the substrate surface with high sensitivity and high accuracy.
[0073] On the other hand, in Comparative Example 1, measurements were taken using a detection sensitivity guaranteed by the surface inspection equipment manufacturer, and therefore, minute defects could not be adequately detected as actual defects.
[0074] Furthermore, Comparative Example 2 used the same sensitivity as Example 1 to exclude noise-induced LPDs, resulting in a higher number of counted defects than Comparative Example 1. However, because the measurement was performed only once and the defect detection measurement was not repeated, it was unable to adequately detect minute defects with a low detection recall rate as actual defects compared to Example 1.
[0075] Furthermore, in Comparative Example 3, the sensitivity was set to the same level as in Example 1, capable of excluding noise-induced LPDs, and the defect detection measurements were repeated. As a result, a larger number of defects were counted compared to Comparative Examples 1 and 2. However, because the number of measurements was set below the threshold for the rate of increase, minute defects with a low detection recall rate could not be sufficiently detected as actual defects compared to Example 1.
[0076] This specification includes the following embodiments: [1]: A method for detecting minute defects in a substrate, The first step involves repeatedly performing a surface inspection on a single substrate using a surface inspection device to detect defects at the highest sensitivity without detecting LPD caused by noise due to the surface roughness of the substrate. A second step involves calculating, from the detection results of the first step, the total number of defects relative to the number of measurements, and the rate of increase of the total number of defects relative to the number of measurements. A third step involves calculating the number of measurements required so that the aforementioned increase rate falls below a predetermined threshold, A fourth step involves repeatedly performing a surface inspection on another substrate manufactured under the same manufacturing conditions using the surface inspection device to detect defects at the highest sensitivity for the required number of measurements, A method for detecting minute defects in a substrate, characterized by including [a specific component]. [2]: A method for detecting minute defects in the substrate according to [1], characterized in that the substrate is a silicon single crystal substrate that has been mirror-polished. [3]: The highest sensitivity in which the noise-induced LPD is not detected in the first step is First, one of the substrates is measured using the surface inspection apparatus at the recommended maximum sensitivity, which is the highest sensitivity recommended by the apparatus, or at any measurement sensitivity where the detection limit is greater than the recommended maximum sensitivity. Subsequently, the measurement is performed by gradually decreasing the detection limit at arbitrary intervals to obtain a map of the detected defects. A method for detecting minute defects in a substrate according to [1] or [2] above, characterized in that the measurement is terminated when the noise-induced LPD is detected in the obtained map, the highest sensitivity at which the noise-induced LPD is not detected is determined from the measurement results, and the detection lower limit is set by reducing it to the determined highest sensitivity. [4]: When calculating the total number of defects in the second step, A method for detecting minute defects in a substrate according to any of the above [1] to [3], characterized by comparing the coordinate data of defects obtained in each measurement and considering all defects within a specific coordinate range as the same defect. [5]: When calculating the increase rate of the total number of defects in the second step, the total number of defects S up to the Nth measurement in the measurements repeatedly performed in the first step N and the total number of defects S up to the (N + 1)th measurement N+1 are used to calculate the increase rate P N [%] as [Equation] The method for detecting minute defects on a substrate according to any one of [1] to [4] above, characterized in that it is set as such.
[0077] Note that the present invention is not limited to the above embodiment. The above embodiment is an example, and any configuration that has substantially the same configuration as the technical idea described in the claims of the present invention and exhibits the same operational effects is included in the technical scope of the present invention.
Claims
1. A method for detecting minute defects in a substrate, The first step involves repeatedly performing a surface inspection on a single substrate using a surface inspection device to detect defects at the highest sensitivity without detecting LPDs caused by noise due to the surface roughness of the substrate. A second step involves calculating, from the detection results of the first step, the total number of defects relative to the number of measurements, and the rate of increase of the total number of defects relative to the number of measurements. A third step involves calculating the number of measurements required so that the aforementioned increase rate falls below a predetermined threshold, A fourth step involves repeatedly performing a surface inspection on another substrate manufactured under the same manufacturing conditions using the surface inspection device to detect defects at the highest sensitivity for the required number of measurements, Includes, In the second step, when calculating the rate of increase of the total number of defects, Using the total number of defects S N up to the Nth measurement in the measurements repeatedly performed in the first step, and the total number of defects S N+1 up to the N+1th measurement, the increase rate P N [%] is calculated as follows: [Math 1] A method for detecting minute defects in a substrate, characterized by the following:
2. The method for detecting minute defects in a substrate according to claim 1, characterized in that the substrate is a silicon single crystal substrate that has been mirror-polished.
3. The highest sensitivity in which the noise-induced LPD is not detected in the first step is First, one of the substrates is measured using the surface inspection apparatus at the recommended maximum sensitivity, which is the highest sensitivity recommended by the apparatus, or at any measurement sensitivity where the detection limit is greater than the recommended maximum sensitivity. Subsequently, the measurement is performed by gradually decreasing the detection limit at arbitrary intervals to obtain a map of the detected defects. A method for detecting minute defects in a substrate according to claim 1 or 2, characterized in that the measurement is terminated when the noise-induced LPD is detected in the obtained map, the highest sensitivity at which the noise-induced LPD is not detected is determined from the measurement results, and the detection lower limit is set by reducing it to the determined highest sensitivity.
4. In the second step described above, when calculating the total number of defects, A method for detecting minute defects in a substrate according to claim 1 or 2, characterized in that the coordinate data of defects obtained from each measurement are compared and all defects within a specific coordinate range are considered to be the same defect.