Apparatus for evaluating edge defects in silicon wafers and method thereof
By combining image acquisition and measurement equipment with a learning model, accurate identification and classification of edge defects in silicon wafers were achieved, solving the problem of unclear defect attributes in existing technologies and improving the accuracy and reliability of silicon wafer quality management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- LG SILTRON
- Filing Date
- 2023-03-07
- Publication Date
- 2026-05-29
Smart Images
Figure CN116739970B_ABST
Abstract
Description
Technical Field
[0001] This embodiment relates to an apparatus and method for evaluating edge defects in a silicon wafer. Background Technology
[0002] The Czochralski (CZ) method is commonly used to manufacture silicon wafers. In the CZ method, polycrystalline silicon is placed in a quartz crucible and melted by heating it with a graphite heating element. A seed crystal is immersed in the molten silicon and crystallization occurs at the interface. A single-crystal silicon ingot is grown by pulling the seed crystal while rotating it. Subsequently, the silicon ingot is slicing, etching, and polishing to form wafers.
[0003] As semiconductor process design rules become increasingly miniaturized, defect management in the edge regions of silicon wafers becomes increasingly important. Therefore, techniques capable of assessing edge defects in silicon wafers are crucial.
[0004] However, a technique for assessing edge defects in silicon wafers has not yet been clearly proposed.
[0005] In particular, since the properties of edge defects in silicon wafers are not clearly classified, there is an urgent need for technological development. Summary of the Invention
[0006] Technical issues
[0007] The purpose of these embodiments is to address the aforementioned and other problems.
[0008] Another objective of the embodiments is to provide an apparatus and method for evaluating edge defects in a silicon wafer, which can readily perform edge defect evaluation on the silicon wafer.
[0009] Another objective of the embodiments is to provide an apparatus and method for evaluating edge defects in a silicon wafer, the apparatus and method being able to definitively classify the properties of edge defects in the silicon wafer.
[0010] Another objective of the embodiments is to provide an apparatus and method for evaluating edge defects in a silicon wafer, the apparatus and method being based on learning capable of achieving explicit classification.
[0011] The technical problems of the embodiments are not limited to those described in this section, but include those that can be understood from the description of the invention.
[0012] Technical solution
[0013] According to a first aspect of the embodiments, in order to achieve the above or other objectives, an apparatus for evaluating edge defects in a silicon wafer includes: an image acquisition unit configured to acquire image data of an edge region of the silicon wafer; a data preprocessing unit configured to, when an edge defect is detected in the acquired image data, measure the edge region of the silicon wafer using a measuring device to determine a defect attribute of the detected edge defect; and a processor configured to: perform control to learn the acquired image data to output a defect attribute corresponding to the detected edge defect, and verify the accuracy of the output defect attribute of the edge defect based on the determined defect attribute of the edge defect.
[0014] The measuring equipment includes at least one or more of 3D laser microscopes, SEMs, and TEMs.
[0015] The defect attributes include indentations, scratches, grains, crystals, or ESF.
[0016] The device includes a learning model, and the learning model includes formulas for calculating layer parameters.
[0017] The learning model uses the layer parameter calculation formula to output the defect attribute corresponding to the edge defect detected in the acquired image data.
[0018] When the accuracy of the defect attribute is less than or equal to a threshold, the processor changes the calculation formula for the layer parameters.
[0019] According to a second aspect of this embodiment, a method for evaluating edge defects in a silicon wafer includes: acquiring image data of an edge region of the silicon wafer; when an edge defect is detected in the acquired image data, measuring the edge region of the silicon wafer using a measuring device to determine a defect attribute of the detected edge defect; controlling the acquired image data to learn and output a defect attribute corresponding to the detected edge defect; verifying the accuracy of the output defect attribute of the edge defect based on the determined defect attribute of the edge defect; and verifying the accuracy of the output defect attribute of the edge defect based on the determined defect attribute of the edge defect.
[0020] Beneficial effects
[0021] The effects of an apparatus and method for evaluating edge defects in a silicon wafer according to an embodiment will now be described.
[0022] According to this embodiment, by using a learning model to learn the image data acquired by the image acquisition unit, the defect attributes of edge defects detected in the edge region of a silicon wafer can be accurately and easily identified.
[0023] According to one embodiment, the accuracy of defect attributes can be calculated using the defect attributes output by the learning model each time it is repeatedly trained on a large number of silicon wafers. If the accuracy of the defect attributes is less than or equal to a threshold, the calculation formula for the layer parameters of the learning model can be changed. By periodically changing the calculation formula for the layer parameters of the learning model in this way, the accuracy of the defect attributes output from the learning model can be significantly improved.
[0024] By installing the improved accuracy learning model described above in an actual production line, the precise defect attributes of edge defects obtained from silicon wafers can be determined without additional equipment. Therefore, defect management of edge regions of silicon wafers becomes more accurate, and reliability can be improved through silicon wafer quality enhancements.
[0025] The further applicability of the embodiments will become apparent from the following detailed description. However, since various changes and modifications within the spirit and scope of the embodiments will be readily understood by those skilled in the art, it should be understood that the detailed description and specific embodiments (such as preferred embodiments) are given by way of example only. Attached Figure Description
[0026] Figure 1 This is a block diagram illustrating an apparatus for evaluating edge defects in a silicon wafer according to one embodiment.
[0027] Figure 2 This is a flowchart illustrating a method for evaluating edge defects in a silicon wafer according to an embodiment.
[0028] Figure 3A and Figure 3B The acquisition of image data of the edge region of a silicon wafer is shown.
[0029] Figure 4 shows that... Figure 1 The defect attributes of edge defects determined in the data preprocessing unit.
[0030] Figure 5 The formulas for calculating layer parameters included in the learning model are shown.
[0031] Figure 6 The input / output table of the learning model is shown. Detailed Implementation
[0032] The embodiments disclosed in this specification will be described in detail below with reference to the accompanying drawings. However, regardless of the reference numerals, the same or similar components are given the same reference numerals, and redundant descriptions will be omitted. For ease of writing, the suffixes "module" and "unit" used for components in the following description are given or used interchangeably, and these suffixes do not inherently have different meanings or functions from each other. Furthermore, the accompanying drawings are provided to facilitate understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the drawings. In addition, when an element such as a layer, region, or substrate is referred to as being "on top of" another element, this includes being directly on top of the other element or where other intermediate elements may exist.
[0033] It should be noted that the artificial intelligence model described in the following embodiments can be composed of various artificial neural network models, such as convolutional neural networks (CNN) and recurrent neural networks (RNN).
[0034] Figure 1 This is a block diagram illustrating an apparatus for evaluating edge defects in a silicon wafer according to one embodiment.
[0035] Reference Figure 1 An apparatus 100 for evaluating edge defects in a silicon wafer according to one embodiment includes a communication interface 110, an input interface 120, a processing processor 130, a sensing circuit 140, an output interface 150, a memory 170, an image acquisition unit 160, a data preprocessing unit 190, and a processor 180. The apparatus 100 according to this embodiment can be an artificial intelligence (AI) device that executes learning algorithms. Therefore, the apparatus 100 and the AI device according to this embodiment can be referred to interchangeably.
[0036] The communication interface 110 can use wired / wireless communication technologies to send data to / receive data from external devices (such as other devices or servers).
[0037] Input interface 120 can receive image data. Furthermore, input interface 120 can acquire various types of data, such as user input interfaces for receiving voice signals from users and acquiring data.
[0038] The learning processor 130 can learn from a model composed of an artificial neural network using training data. Here, the learned artificial neural network can be referred to as the learning model. The learning model can be used to infer the result value of new input data other than the learning data, and the inferred value can be used as the basis for decisions to perform specific operations.
[0039] As will be described later, the learning model of this embodiment may include formulas for calculating layer parameters.
[0040] A learning model can be used to output defect attributes corresponding to edge defects detected in the image data of the image acquisition unit 160. That is, the value of each of the various parameter values can be calculated based on the edge defects detected in the image data. A learning model can be created by layering and modeling each of these different parameter values. Therefore, when the image data of the image acquisition unit 160 is input into the learning model, multiple defect attributes (…) are calculated according to the layer parameter calculation formula of the learning model. Figure 5 One of the defect attributes (from category 1 to category 6) is output.
[0041] Figure 6 The input / output table of the learning model is shown.
[0042] The input / output table of a learning model can consist of multiple codes as inputs and multiple codes as outputs. Here, a code can be a routine for calculating the layer parameters of the learning model. Defect attributes can correspond to each code, but are not limited to this.
[0043] When the maximum number of codes in the input / output table is determined by repeatedly learning the learning model, the defect attribute corresponding to that code can be output from the learning model.
[0044] According to one embodiment, the accuracy of defect attributes can be calculated using the defect attributes output by the learning model each time it is repeatedly trained on a large number of silicon wafers. If the accuracy of the defect attributes is less than or equal to a threshold, the calculation formula for the layer parameters of the learning model can be changed. By periodically changing the calculation formula for the layer parameters of the learning model in this way, the accuracy of the defect attributes output from the learning model can be significantly improved.
[0045] By implementing a learning model with improved accuracy into an actual production line, the precise defect attributes of edge defects obtained from silicon wafers can be determined without additional equipment. Therefore, defect management of edge regions of silicon wafers becomes more accurate, and reliability can be improved through silicon wafer quality enhancements.
[0046] The artificial intelligence model to be loaded into the device 100 according to this embodiment can be learned in the learning processor 130 and can also be sent from an external device or server.
[0047] The sensing circuit 140 can use various sensors to acquire at least one of the following: internal information of the AI device 100, surrounding environmental information of the AI device 100, and user information.
[0048] In addition, the sensing circuit 140 may include a camera, and the camera may be used to receive images for determining defects in the silicon wafer.
[0049] The image acquisition unit 160, which will be described later, may be included in the sensing circuit 140, but is not limited thereto.
[0050] Output interface 150 can produce outputs related to vision, hearing, or touch.
[0051] The memory 170 can store data that supports various functions of the AI device 100. For example, the memory 170 can store learning data, learning models, learning history, and input data obtained from the input interface 120.
[0052] The image acquisition unit 160 can acquire image data of the edge region of the silicon wafer. The image acquisition unit 160 may be referred to as a camera, image scanner, etc.
[0053] like Figure 3A and Figure 3B As shown, after the optical lens 320 of the image acquisition unit 160 focuses on each of the sub-region partitions 1 to 5 of the edge region 312 of the silicon wafer 310, image data can be acquired in each sub-region partition 1 to 5. Sub-region partitions 1 to 5 can be, for example, the lower edge region, the upper edge region, and the side region of the edge region 312 of the silicon wafer 310.
[0054] As an example, the main body (not shown) of the image acquisition unit 160 can move along the X-axis, Y-axis, and Z-axis, and can also rotate. When the image acquisition unit 160 moves and rotates along the X-axis, Y-axis, Z-axis, etc., image data can be acquired in each of the sub-regions 1 to 5 of the edge region 312 of the silicon wafer 310 through the optical lens 320 of the image acquisition unit 160.
[0055] As another example, the main body of the image acquisition unit 160 can be fixed. In this case, multiple optical lenses 320 simultaneously focus on each of the multiple sub-regions 1 to 5 of the silicon wafer 310, enabling image data to be acquired simultaneously at the same time.
[0056] In this embodiment, the image acquisition unit 160 can analyze the image data acquired in each of the multiple sub-region partitions 1 to 5 of the edge region 312 of the silicon wafer 310 to detect edge defects from the corresponding image data.
[0057] Meanwhile, edge defect detection can be performed by the processor 180. In this case, image data from each of the multiple sub-region partitions 1 to 5 of the edge region 312 of the silicon wafer 310 acquired in the image acquisition unit 160 can be sent to the processor 180. The processor 180 can analyze the image data acquired from each of the multiple sub-region partitions 1 to 5 of the edge region 312 of the silicon wafer 310 to detect edge defects from the corresponding image data.
[0058] When an edge defect is detected in the image data acquired by the image acquisition unit 160, the data preprocessing unit 190 can use a measuring device to measure the edge region 312 of the silicon wafer 310 and determine the defect properties of the detected edge defect. The measuring device may include at least one of a 3D laser microscope, SEM, and TEM.
[0059] When no edge defects are detected in the image data acquired by the image acquisition unit 160, the operation of the data preprocessing unit 190 may not be performed. Conversely, the operation of the data preprocessing unit 190 may be performed even if no edge defects are detected in the image data acquired by the image acquisition unit 160. In this case, edge defects not detected by the image acquisition unit 160 can be detected by the measuring device.
[0060] like Figures 4A to 4E As shown, various defect attributes can be determined through measurements using measuring equipment. Defect attributes can represent edge types.
[0061] Defect attributes include, for example, indentations ( Figure 4A ), scratches ( Figure 4B ), particles ( Figure 4C ), ESF ( Figure 4D ) and crystals ( Figure 4E ).
[0062] For example, defects can be identified by their properties (such as indentation). Figure 4A ), scratches ( Figure 4B ) and particles ( Figure 4C To determine the silicon wafer to be polished.
[0063] EPI (epitaxial) silicon wafers are used for direct product manufacturing and require more precise edge defect management than polished wafers. Therefore, epitaxial silicon wafers need to be classified using more specific defect attributes. Thus, EPI silicon wafers can be categorized by defect attributes such as indentation (…). Figure 4A ), scratches ( Figure 4B ), particles ( Figure 4C ), ESF ( Figure 4D ), crystal ( Figure 4ETo determine.
[0064] The processor 180 can determine at least one executable operation of the AI device 100 based on information determined or generated using data analysis algorithms or machine learning algorithms. The processor 180 can execute the determined operation by controlling the components of the AI device 100.
[0065] According to this embodiment, the processor 180 can control the overall operation of the device 100.
[0066] The processor 180 can control the image data to be learned so as to output defect attributes corresponding to edge defects detected in the edge region 312 of the silicon wafer 310.
[0067] Artificial intelligence models, such as learning models, can be stored in memory 170.
[0068] As an example, processor 180 can load a learning model from memory 170 and control the learning model to be learned.
[0069] As another example, learning processor 130 can load a learning model from memory 170 and control the learning model. Learning processor 130 may be included in processor 180.
[0070] like Figure 5 As shown, the learning model can include layer parameter calculation formulas. Here, the parameters are: minimum / maximum gray value, average gray value, gray standard deviation, area, X / Y center coordinates, X / Y centroid coordinates, area of each X / Y axis, roundness, compactness, contour length, convexity, rectangularity, fitted major / minor axis / angle, ratio of the major semi-axis to the minor semi-axis of the ellipse (anisometry), volume, structure factor, radius of the smallest circumcircle, radius of the largest incircle, average contour distance, contour distance standard deviation, roundness, polygon approximation parameters, hole area, diameter, Euler number, minimum rectangle length / short width, average gray value / gray standard deviation of at least two neighboring pixels, etc.
[0071] exist Figure 5 In this context, 'a' through 'u' can be threshold values for each parameter. Each parameter can be a parameter used for edge defects detected in the edge region 312 of the silicon wafer 310 detected by the image acquisition unit 160. Formulas comparing these parameters with thresholds can be layered. Therefore, a final output value can be output in a layered calculation manner using these layered calculation formulas. The final output value can be one of defect attribute categories 1 through 6. For example, category 1 can be an indentation, category 2 can be a scratch, and category 3 can be a particle. For example, category 4 can be ESF, and category 5 can be crystallization. For example, category 6 can be another defect attribute not defined in this embodiment.
[0072] Despite Figure 5 The document shows six defect attribute categories, 1 through 6, but more defect attributes are possible.
[0073] The learning model can use layer parameter calculation formulas to output defect attributes corresponding to edge defects detected in the image data of image acquisition unit 160.
[0074] Meanwhile, artificial intelligence models (such as learning models) have been described as being learned by processor 180, but the apparatus 100 according to the embodiment may also receive artificial intelligence models learned from external devices or servers.
[0075] Continue to refer to Figure 1 The processor 180 can verify the accuracy of the defect attributes of the edge defects output from the learning model based on the combined attributes of the edge defects determined by the data preprocessing unit 190.
[0076] When an edge defect is detected in the image data of the edge region 312 of the silicon wafer 310 acquired by the image acquisition unit 160, the data preprocessing unit 190 can measure the edge region 312 of the silicon wafer 310, detect and analyze the edge defect corresponding to the edge defect detected in the image data of the image acquisition unit 160 in the measured edge region 312, and determine the defect attribute of the edge defect.
[0077] Image data acquired by the image acquisition unit 160 can be input into the learning model. Specifically, various parameter values obtained from the image data acquired by the image acquisition unit 160 can be input into the learning model. Each of these parameter values can be obtained by analyzing the image data acquired by the image acquisition unit 160. In this case, the parameters obtained from the image data can be related to... Figure 5 The parameters shown are the same as those for the layers.
[0078] The learning model can calculate the layer parameter calculation formula based on each parameter value obtained from the image data to output a defect attribute from multiple defect attribute categories 1 to 6.
[0079] In this case, for the same edge defect detected in silicon wafer 310, processor 180 can determine whether the defect attributes determined by data preprocessing unit 190 are the same as the defect attributes output from output model.
[0080] If the defect attributes determined by the data preprocessing unit 190 are the same as the defect attributes output by the output model, then the layer parameter calculation formula is well set in the learning model, which means that the defect attribute classification of the output model is accurate.
[0081] If the defect attributes determined by the data preprocessing unit 190 are different from the defect attributes output from the output model, then the layer parameter calculation formula is incorrectly set in the learning model, which means that the defect attribute classification of the output model is inaccurate.
[0082] However, verifying the accuracy of a learning model for an edge defect is difficult. Therefore, in this embodiment, accuracy can be improved by iteratively verifying the accuracy of the learning model for each of various silicon wafers, including polished silicon wafers and EPI silicon wafers.
[0083] Despite this iterative accuracy verification, the calculation formulas for layer parameters in the learning model can be modified when the accuracy of the defect attributes output from the learning model is less than or equal to a threshold. For example, the threshold for each layer parameter calculation formula can be changed, or the position of each layer parameter calculation formula can be altered. The accuracy of the defect attributes output from the learning model can vary depending on the defect attributes, but is not limited to this.
[0084] In the following text, reference will be made to Figure 2 A method for evaluating edge defects in a silicon wafer is described according to one embodiment.
[0085] Figure 2 This is a flowchart illustrating a method for evaluating edge defects in a silicon wafer according to an embodiment.
[0086] Reference Figure 1 and Figure 2 The image acquisition unit 160 can acquire image data of the edge region 312 of the silicon wafer (S210).
[0087] like Figure 3A and Figure 3B As shown, the optical lens 320 of the image acquisition unit 160 can acquire image data of the edge region 312 of the silicon wafer 310 by focusing on the periphery of the edge region 312 of the silicon wafer 310. The image data may or may not have edge defects.
[0088] Image data can be analyzed to detect edge defects. Edge defect candidates can be detected based on various parameter values from the image data, and edge defects can be detected through precise analysis of these edge defect candidates.
[0089] At the same time, it is usually difficult to detect edge defects in the edge region 312 of the silicon wafer 310, and it is impossible to classify or determine the specific defect attributes of the edge defects.
[0090] In this embodiment, image data acquired by the image acquisition unit 160 can be used as a learning model to classify or determine defect attributes. That is, when various parameter values of the image data are input into the learning model, the defect attributes of edge defects detected in the image data can be output. In this case, the accuracy of the defect attributes classified or determined by the learning model is certainly high.
[0091] S220, S230 and S240, which will be described below, relate to methods for improving the accuracy of defective attributes classified or identified in a learning model.
[0092] In other words, the data preprocessing unit 190 can provide reference defect attributes for verifying the accuracy of the defect attributes output from the learning model.
[0093] Therefore, the data preprocessing unit 190 can determine the defect attributes of the edge defects detected in the image data (S220). That is, the data preprocessing unit 190 can determine the specific defect attributes of the edge defects detected in the image data in the image acquisition unit 160.
[0094] First, a measuring device can be used to measure the edge region 312 of the silicon wafer 310. For example, the measuring device may include one or more of a 3D laser microscope, SEM, and TEM.
[0095] For example, the number of defect attributes can vary for polished silicon wafers and EPI silicon wafers. For instance, in the case of polished silicon wafers, there can be three defect attributes, such as indentation (…). Figure 4A ), scratches ( Figure 4B ) and particles ( Figure 4C However, it is not limited to this. In the case of EPI silicon wafers, there can be five defect properties, such as indentation ( Figure 4A ), scratches ( Figure 4B ), particles ( Figure 4C ), ESF ( Figure 4D ) and crystals ( Figure 4E (but not limited to this).
[0096] The data preprocessing unit 190 can analyze data measured using a measuring device related to the edge region 312 of the silicon wafer 310, thereby determining the defect attributes of edge defects present in the edge region 312 of the silicon wafer 310.
[0097] Meanwhile, the processor 180 can control the image data to be learned so as to output defect attributes corresponding to edge defects (S230).
[0098] For example, processor 180 can load the learning model into memory 170, and then input image data acquired by image acquisition unit 160 into the learning model. That is, various parameter values of edge defects detected in the image data can be input into the learning model.
[0099] The learning model can include layer parameter calculation formulas. Therefore, the learning model can receive the parameter values for each edge defect detected in image data and output the defect attributes of the edge defects by calculating the layer parameter calculation formulas. That is, as... Figure 5 As shown, the calculation formula for the calculation layer parameters can output one of the multiple defect attribute categories 1 to 6.
[0100] The processor 180 can verify the accuracy of the defect attributes of the edge defects output from the learning model (S240).
[0101] In other words, the processor 180 can compare the defect attributes of the edge defects determined by the data preprocessing unit 190 with the defect attributes of the edge defects output from the learning model. The defect attributes of the edge defects determined by the data preprocessing unit 190 and the defect attributes of the edge defects output from the learning model can be the same edge defects detected in the image data acquired by the image acquisition unit 160.
[0102] When the defect attributes of the edge defect determined by the data preprocessing unit 190 are the same as the defect attributes of the edge defect output from the learning model, the processor 180 can determine that the layer parameter calculation formula set in the learning model is likely valid. For example, if the defect attribute of the edge defect determined in the data preprocessing unit 190 is indentation, and the defect attribute of the edge defect output from the learning model is also indentation, then it is determined that the layer parameter calculation formula set in the learning model is valid.
[0103] When the defect attributes of the edge defect determined by the data preprocessing unit 190 differ from the defect attributes of the edge defect output from the learning model, the processor 180 determines that the layer parameter calculation formula set in the learning model may be invalid. For example, if the defect attribute of the edge defect determined by the data preprocessing unit 190 is an indentation, but the defect attribute of the edge defect output from the learning model is a scratch, then the layer parameter calculation formula set in the learning model is determined to be invalid.
[0104] Therefore, when it is determined that the layer parameter calculation formula set in the learning model is invalid, the processor 180 can change the layer parameter calculation formula.
[0105] In this way, after changing the calculation formula of the layer parameters of the learning model, the accuracy of the defect attributes output from the learning model can be verified again by executing S210, S220, S230 and S240.
[0106] By repeatedly performing such operations, the accuracy of the defect attributes output from the learning model can be improved.
[0107] As another embodiment, even without changing the calculation formula of the layer parameters of the learning model, the number of learning operations of the learning model can be increased by repeatedly executing S210, S220, S230 and S240, thereby improving the accuracy of the defect attributes output from the learning model.
[0108] The above detailed description should not be construed as limiting in all respects, but rather as illustrative. The scope of the embodiments should be determined by a reasonable interpretation of the appended claims, and all variations within the equivalent scope of the embodiments are included within the scope of the embodiments.
Claims
1. An apparatus for evaluating edge defects in a silicon wafer, comprising: A memory, the memory including a learning model, wherein the learning model includes layer parameter calculation formulas; An image acquisition unit, configured to acquire image data of the edge region of the silicon wafer; A data preprocessing unit, configured to, when an edge defect is detected in the acquired image data, use a measuring device to measure the edge region of the silicon wafer to determine a second defect attribute of the detected edge defect; and processor, The processor is configured as follows: Detect the edge defects from the acquired image data; The detected edge defects are analyzed to obtain parameters of the detected edge defects. By inputting the acquired parameters of the detected edge defects into the learning model and comparing the acquired parameters of the detected edge defects with parameter thresholds in a hierarchical manner using the layer parameter calculation formula, control is exercised to learn the acquired image data, so as to output a first defect attribute corresponding to the detected edge defects. The accuracy of the output of the first defect attribute of the detected edge defect is verified based on the determined second defect attribute of the detected edge defect, and When the accuracy of the first defect attribute output is less than or equal to the threshold, change the parameter threshold or the position of each layer parameter calculation formula in the layer parameter calculation formula.
2. The apparatus according to claim 1, wherein, The measuring equipment includes at least one or more of 3D laser microscopes, SEMs, and TEMs.
3. The apparatus according to claim 1, wherein, The first defect attribute and the second defect attribute respectively include one of indentation, scratch, grain, crystal or ESF.
4. A method for evaluating edge defects in a silicon wafer, comprising: Acquire image data of the edge region of the silicon wafer; When an edge defect is detected in the acquired image data, a measuring device is used to measure the edge region of the silicon wafer to determine a second defect attribute of the detected edge defect; Detect the edge defects from the acquired image data; The detected edge defects are analyzed to obtain parameters of the detected edge defects. The acquired image data is learned by inputting the parameters of the detected edge defects into a learning model stored in a memory and comparing the parameters of the acquired edge defects with parameter thresholds in a hierarchical manner using the layer parameter calculation formula included in the learning model, so as to output a first defect attribute corresponding to the detected edge defects. The accuracy of the first defect attribute of the detected edge defect is verified based on the determined second defect attribute of the detected edge defect. as well as When the accuracy of the first defect attribute output is less than or equal to the threshold, change the parameter threshold or the position of each layer parameter calculation formula in the layer parameter calculation formula.
5. The method according to claim 4, wherein, The measuring equipment includes at least one of a 3D laser microscope, a SEM, and a TEM.
6. The method according to claim 4, wherein, The first defect attribute and the second defect attribute respectively include one of indentation, scratch, grain, crystal or ESF.