Method and equipment for predicting vertical surface type of semiconductor product

Abstract

The invention discloses a method and equipment for predicting the vertical surface type of a semiconductor product, and the method comprises the steps: obtaining the front-layer process information and the vertical surface type information of a plurality of semiconductor products, and enabling the front-layer process information to represent a machine table and a cavity which are used for carrying out the front-layer process processing of the corresponding semiconductor product, the vertical surface type information represents the fluctuation of the corresponding semiconductor product in the vertical direction; clustering the plurality of semiconductor products according to the vertical surface type information of the semiconductor products to obtain a defect prediction model comprising a plurality of product groups; for each product group in the defect prediction model, determining a front-layer process label and a defect surface type label of the product group according to vertical surface type information and front-layer process information of the semiconductor product corresponding to the product group; wherein the front-layer process labels and the defect surface type labels of the plurality of product groups in the defect prediction model are at least used for predicting the defect surface type characteristics of the semiconductor products which are not processed by the front-layer process.

Description

Technical Field

[0001] This application relates to the field of semiconductor technology, and in particular to a method and apparatus for predicting the vertical profile of a semiconductor product. Background Technology

[0002] As integrated circuit technology nodes continue to shrink and complex processes are introduced, the requirements for vertical precision control of semiconductor products that carry integrated circuits are becoming increasingly stringent. Taking silicon wafers as an example, during actual processing, interference factors such as the structure or process of the front-layer film, unevenness of the lithography machine stage, and uneven vacuum adsorption on the stage can all cause significant vertical undulations in the produced silicon wafers. This has a profound impact on the imaging quality of the lithographic patterns on the silicon wafers, and consequently affects the yield of the final integrated circuit products.

[0003] To address this issue, relevant technologies typically analyze the surface shape of the processed silicon wafers to screen out defective wafers with excessive vertical undulations. However, this approach fails to reduce defects and improve the yield of silicon wafers. Summary of the Invention

[0004] Therefore, this application discloses the following technical solution:

[0005] The first aspect of this application provides a method for predicting the vertical surface profile of a semiconductor product, comprising:

[0006] Obtain front-end process information and vertical surface profile information for multiple semiconductor products. The front-end process information characterizes the equipment and chambers used to perform front-end process processing on the corresponding semiconductor products, and the vertical surface profile information characterizes the vertical undulations of the corresponding semiconductor products.

[0007] Based on the vertical surface shape information of the semiconductor products, multiple semiconductor products are clustered to obtain a defect prediction model that includes multiple product groups;

[0008] For each product group in the defect prediction model, the front-end process label and defect surface label of the product group are determined based on the vertical surface information and front-end process information of the semiconductor product corresponding to the product group.

[0009] The defect prediction model includes multiple product group front-end process labels and defect surface type labels, which are used at least to predict the defect surface type features of semiconductor products that have not undergone front-end process processing.

[0010] Optionally, the step of clustering multiple semiconductor products based on the vertical surface information of the semiconductor products to obtain a defect prediction model including multiple product groups includes:

[0011] For each semiconductor product, multiple surface profile indicators are determined based on the vertical surface profile information of the semiconductor product to obtain surface profile feature data of the semiconductor product composed of the multiple surface profile indicators;

[0012] Clustering is performed on the surface feature data of the semiconductor products to obtain a defect prediction model that includes multiple product groups.

[0013] Optionally, the step of clustering based on the surface feature data of the semiconductor products to obtain a defect prediction model including multiple product groups includes:

[0014] Identify multiple central feature data;

[0015] Based on the similarity parameters between the central feature data and the surface feature data, the multiple semiconductor products are divided into product groups corresponding to the central feature data;

[0016] Determine whether the similarity parameters between the central feature data and the surface feature data of the semiconductor product in each product group meet the convergence condition;

[0017] If the convergence condition is not met, the center feature data corresponding to the product group is updated according to the surface feature data of the semiconductor products contained in the product group, and the process of dividing the multiple semiconductor products into the product group corresponding to the center feature data according to the similarity parameter of the center feature data and the surface feature data is returned until the convergence condition is met.

[0018] Optionally, determining the front-end process label and defect surface type label of the product group based on the vertical surface type information and front-end process information of the semiconductor product corresponding to the product group includes:

[0019] Among the multiple semiconductor products corresponding to the product group, the first semiconductor product whose similarity parameter between the corresponding surface feature data and the center feature data of the product group satisfies the target condition is selected.

[0020] Based on the vertical surface profile information of each of the first semiconductor products, the defect surface profile features of each of the first semiconductor products are obtained, and the defect surface profile feature that appears most frequently is determined as the defect surface profile label corresponding to the product group.

[0021] Obtain the front-end process information for each of the first semiconductor products, and determine the front-end process information that appears most frequently as the front-end process label corresponding to the product group.

[0022] Optionally, the step of selecting a first semiconductor product from among the multiple semiconductor products corresponding to the product group, wherein the similarity parameter between the corresponding surface feature data and the center feature data of the product group satisfies the target condition, includes:

[0023] For each semiconductor product in the product group, a root mean square similarity parameter is calculated based on a first similarity parameter and a second similarity parameter of the surface feature data corresponding to the semiconductor product and the center feature data of the product group.

[0024] The semiconductor products corresponding to the product group are arranged in descending order according to the root mean square of the similarity parameter, and the first N semiconductor products are determined to be the first semiconductor products that meet the target conditions, where N is a preset integer.

[0025] Optionally, the step of clustering multiple semiconductor products based on the vertical surface information of the semiconductor products to obtain a defect prediction model including multiple product groups includes:

[0026] Obtain the vertical surface profile information of the ultra-flat sheet;

[0027] The vertical surface profile information of each semiconductor product is corrected based on the vertical surface profile information of the ultra-flat wafer to obtain the corrected vertical surface profile information.

[0028] Based on the corrected vertical surface information of the semiconductor products, multiple semiconductor products are clustered to obtain a defect prediction model that includes multiple product groups.

[0029] Optional, also includes:

[0030] Obtain the second defect surface feature corresponding to the second semiconductor product, wherein the second semiconductor product is a semiconductor product processed by the previous layer process;

[0031] Search for a second defect facet label that matches the second defect facet feature among the defect facet labels of the multiple product groups in the defect prediction model;

[0032] Based on the front-end process label of the product group to which the second defect surface label belongs, determine the front-end process information of the second semiconductor product related to the second defect surface feature.

[0033] Optional, also includes:

[0034] Obtain the third front-end process information of the third semiconductor product, wherein the third semiconductor product is a semiconductor product that has not undergone front-end process processing;

[0035] The third front-end process label that matches the third front-end process information is searched among the front-end process labels of the multiple product groups in the defect prediction model.

[0036] Based on the defect surface type label of the product group to which the third front-layer process label belongs, predict the predicted defect surface type features of the third semiconductor product.

[0037] Optionally, it may also include at least one of the following:

[0038] The processing parameters of the machine and chamber corresponding to the third semiconductor product are adjusted according to the predicted defect surface features to avoid the occurrence of the predicted defect surface features;

[0039] The design parameters of the third semiconductor product are adjusted based on the predicted defect surface features to avoid the occurrence of the predicted defect surface features.

[0040] A second aspect of this application provides an electronic device, including a memory and a processor;

[0041] The memory is used to store computer programs;

[0042] The processor is used to execute the computer program to implement the method for predicting the vertical profile of a semiconductor product provided in any of the first aspects of this application.

[0043] The beneficial effects of this plan are as follows:

[0044] A defect prediction model is constructed based on the front-end process information and vertical surface type information of multiple semiconductor products. The front-end process label and defect surface type label of each product group in the defect prediction model are determined. Therefore, based on the front-end process label and defect surface type label of the product group in the defect prediction model, the defect surface type characteristics of semiconductor products that have not undergone front-end process processing can be predicted. This allows for the implementation of corresponding measures to address the vertical surface type defects that may occur in semiconductor products during front-end process processing, thereby improving the yield of processed semiconductor products. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of a method for predicting the vertical surface profile of a semiconductor product provided in an embodiment of this application;

[0047] Figure 2 This is a schematic diagram illustrating clustering processing based on the surface feature data of a semiconductor product, provided in an embodiment of this application.

[0048] Figure 3 This is a schematic diagram of the structure of a device for predicting the vertical surface profile of a semiconductor product provided in an embodiment of this application;

[0049] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] This application provides a method for predicting the vertical surface profile of a semiconductor product. Please refer to [link to relevant documentation]. Figure 1 The method may include the following steps.

[0052] S101, obtain front-end process information and vertical surface information of multiple semiconductor products. The front-end process information represents the equipment and chamber used to perform front-end process processing on the corresponding semiconductor products, and the vertical surface information represents the vertical undulations of the corresponding semiconductor products.

[0053] The semiconductor product in this embodiment can be a silicon wafer or other wafer-shaped semiconductor product. For ease of explanation, the following description uses a silicon wafer as an example to illustrate the specific implementation of the method provided in this embodiment.

[0054] The silicon wafer in S101 is a semiconductor product that has undergone at least the preceding layer processing. In this embodiment, the preceding layer processing of the silicon wafer refers to all the process steps that the silicon wafer undergoes before the current layer is exposed. Specifically, it may include multiple process steps such as pre-processing, photolithography, etching (ETCH), thin film growth (TF), ion implantation, cleaning, chemical mechanical polishing (CMP), and measurement. Each process step can be realized by a machine and chamber with corresponding processing functions. For example, the photolithography process can be realized by a photolithography machine and a photolithography chamber, and the etching process can be realized by an etching machine and an etching chamber.

[0055] The front-end process information of silicon wafers can also be called front-end context information. The front-end process of silicon wafers is generally implemented through factory automation systems. Therefore, for each silicon wafer that has undergone front-end process processing, the front-end process information of the silicon wafer can be obtained from the factory automation system that processed it.

[0056] The content of the front-end process information for silicon wafers is not limited. For example, in a factory automation system, each process step can correspond to multiple processing machines and chambers used for that step. The front-end process information for a silicon wafer can include, in each process step of the front-end process, the machine identifier of the machine used to process the silicon wafer and the chamber identifier of the chamber used to process the silicon wafer.

[0057] For example, assuming that the front-end processing of silicon wafer A is specifically performed using photolithography equipment numbered 100 and photolithography chamber numbered 201, and etching using etching equipment numbered 305 and etching chamber numbered 402, then the front-end process information of silicon wafer A may include:

[0058] Photolithography process information: Photolithography machine, 100; Photolithography chamber, 201;

[0059] Etching process information: Etching machine, 305; Etching chamber, 402.

[0060] Understandably, the front-end process information of a silicon wafer can include process information for each process step that the silicon wafer undergoes in the front-end process, including but not limited to etching process information for the etching step, thin film process information for the thin film (TF) step, CMP process information for the chemical mechanical polishing (CMP) step, cleaning process information for the cleaning step, etc. The above examples are only examples of the process information for the photolithography and etching steps in the front-end process information.

[0061] The method for obtaining the pre-process information of the silicon wafer in step S101 can be as follows: each time the factory automation system produces a silicon wafer that has undergone pre-processing, it reads the pre-process information of the silicon wafer from the factory automation system and stores the read pre-process information and the identifier of the silicon wafer in the lithography data center in the form of a file. The file format is not limited and can generally be any structured data format, such as an XML file.

[0062] The vertical profile information of a silicon wafer characterizes the height and undulations of the wafer in the vertical direction. The vertical direction refers to the direction perpendicular to the upper surface of the silicon wafer. The form of the vertical profile information is not limited; as an example, it can include the thickness at various locations on the upper surface of the wafer, that is, the distance from each location on the upper surface to the lower surface of the wafer.

[0063] One way to obtain vertical surface profile information is to use a leveling sensor to scan the silicon wafer during the exposure process. This vertical surface profile information can also be stored as a structured XML file in a lithography data center. In other embodiments, other sensors capable of scanning the vertical undulations of the silicon wafer's surface can also be used to obtain the vertical surface profile information; this is not limited to leveling sensors.

[0064] S102, cluster multiple semiconductor products based on the vertical surface information of the semiconductor products to obtain a defect prediction model that includes multiple product groups.

[0065] Clustering multiple semiconductor products can be done in the following ways:

[0066] A1. For each semiconductor product, determine multiple surface features corresponding to the semiconductor product based on the vertical surface features information of the semiconductor product, so as to obtain surface feature data of the semiconductor product composed of multiple surface features.

[0067] A2, based on the surface feature data of semiconductor products, performs clustering to obtain a defect prediction model that includes multiple product groups.

[0068] In step A1, for each silicon wafer, several target areas can be defined on the upper surface of the silicon wafer. For example, a circular area with the center of the silicon wafer as the center, an inner radius of 140 mm, and an outer radius of 150 mm can be defined as a target area. Other target areas can also be set as needed, without limitation.

[0069] It should be noted that the target area is the same for each silicon wafer. Referring to the example above, the target area for each silicon wafer is a ring-shaped region with an inner radius of 140 mm and an outer radius of 150 mm.

[0070] After determining the target area, statistical calculations can be performed on the vertical surface features of silicon wafers located within that target area to obtain n surface feature parameters for the silicon wafer. n is an arbitrary preset integer. For example, n can be set to 20 or 30.

[0071] As an example, for a silicon wafer A, the thickness of silicon wafer A at various locations within the aforementioned target area can be obtained from the vertical surface profile information of silicon wafer A. Then, the root mean square value of multiple thickness values ​​is calculated to obtain a surface profile index of silicon wafer A, denoted as KPI1. A Calculate the average of multiple thickness values ​​to obtain another surface profile index for silicon wafer A, denoted as KPI2. A Calculate the 3sigma value (i.e., 3 times the standard deviation) of multiple thickness values ​​to obtain another aspect ratio of silicon wafer A, denoted as KPI3. A ; Calculate the upper and lower limits of the range of multiple thickness values ​​to obtain another surface profile index for silicon wafer A, denoted as KPI4. A This process continues until the n surface features of silicon wafer A are calculated.

[0072] After obtaining n surface feature parameters of a silicon wafer, the vector formed by these n surface feature parameters is the surface feature data corresponding to that silicon wafer. Referring to the example above, the surface feature data of silicon wafer A could be (KPI1...). A KPI2 A KPI3 A KPI4 A ...KPIn A ).

[0073] In step A2, a grouping number K can be set. This value determines the number of product groups included in the constructed defect prediction model, that is, how many product groups are divided into multiple silicon wafers. The number of groups can be set by the relevant user based on experience, for example, it can be set to any value between 2 and 5.

[0074] Once the number of groups is determined, multiple silicon wafers can be clustered based on the surface feature data corresponding to each silicon wafer to obtain a defect prediction model consisting of K groups.

[0075] Specifically, clustering can be performed based on whether the corresponding surface feature data are similar. If the surface feature data of two silicon wafers are similar, the two silicon wafers are grouped into the same group. If the surface feature data of two silicon wafers are not similar, the two silicon wafers are grouped into different groups.

[0076] S103, for each product group in the defect prediction model, determine the front-end process label and defect surface label of the product group based on the vertical surface information and front-end process information of the semiconductor product corresponding to the product group; the front-end process labels and defect surface labels of multiple product groups in the defect prediction model are used to predict the defect surface features of semiconductor products that have not undergone front-end process processing.

[0077] In step S103, the defect surface features of each silicon wafer can be obtained based on the vertical surface features of that silicon wafer.

[0078] For each silicon wafer, the defect surface features can characterize both the presence of vertical surface defects (i.e., vertical surface defects) and the characteristics of the existing vertical surface defects. For example, the defect surface features of a silicon wafer could be: tornado-like undulations at the 12 o'clock position.

[0079] For methods to obtain the defect surface features of silicon wafers, please refer to the relevant technologies in the field of silicon wafer defect detection, which will not be elaborated in this embodiment.

[0080] After obtaining the defect surface features of each silicon wafer, for each product group, the defect surface features of each silicon wafer in the product group can be statistically analyzed. If one or more defect surface features appear frequently and account for a large proportion in a certain product group, then one or more defect surface features can be identified as the defect surface feature corresponding to that product group. This label indicates that if a silicon wafer belongs to that product group, then the silicon wafer is likely to have the defect surface feature indicated by the defect surface feature label.

[0081] For example, suppose a product group X contains 20 silicon wafers, and 15 of the 20 silicon wafers have a tornado-like undulation defect surface feature in the 12 o'clock direction. The number of silicon wafers with other defect surface features is less than 10. Then the defect surface feature label corresponding to this product group can be determined as: tornado-like undulation in the 12 o'clock direction.

[0082] The methods for identifying the front-end process label can be:

[0083] For each product group, the front-end process information of each silicon wafer in the product group is compared to determine which silicon wafers in the product group have the same front-end process information, and specifically which front-end process information is the same. Then, the front-end process information that appears more frequently and accounts for a larger proportion in the product group is determined as the front-end process label of the product group. The front-end process label of the product group can characterize that most of the silicon wafers belonging to the product group are processed by the specific machine and / or chamber indicated by the front-end process label.

[0084] As an example, suppose a product group X contains 20 silicon wafers, and the front-end process information of 18 of these 20 wafers includes the following:

[0085] Photolithography chamber, 201; Etching equipment, 305; CMP equipment, 508;

[0086] In other words, these 18 silicon wafers all went through the photolithography chamber 201, the etching machine 305 and the CMP machine 508 during the front-end processing.

[0087] In this case, the corresponding front-end process label for the product group can be determined as: lithography chamber, 201; etching machine, 305; CMP machine, 508.

[0088] As can be seen, after determining the front-end process label and defect surface type label for each group in the above manner, the correspondence between various front-end process information and defect surface type features can be obtained based on the defect prediction model. For example, based on product group X in the previous example, the following correspondence between front-end process information and defect surface type features can be obtained:

[0089] There are tornado-like undulations at the 12 o'clock position - photolithography chamber, 201; etching machine, 305; CMP machine, 508.

[0090] Therefore, when a silicon wafer that has not yet undergone pre-processing needs to be pre-processed, the machine and chamber expected to be used to process the silicon wafer can be used to predict the likely defect surface features of the silicon wafer based on the correspondence indicated by the above defect prediction model. In this way, countermeasures can be taken to eliminate or weaken the possible defect surface features, thereby improving the yield of the processed silicon wafer.

[0091] The beneficial effects of this embodiment are as follows:

[0092] A defect prediction model is constructed based on the front-end process information and vertical surface type information of multiple semiconductor products. The front-end process label and defect surface type label of each product group in the defect prediction model are determined. Therefore, based on the front-end process label and defect surface type label of the product group in the defect prediction model, the defect surface type characteristics of semiconductor products that have not undergone front-end process processing can be predicted. This allows for the implementation of corresponding measures to address the vertical surface type defects that may occur in semiconductor products during front-end process processing, thereby improving the yield of processed semiconductor products.

[0093] In this embodiment, multiple silicon wafers obtained can be clustered based on the surface feature data of semiconductor products using various methods. One possible method is:

[0094] B1, identifying multiple central characteristic data;

[0095] B2, based on the similarity parameters of the central feature data and the surface feature data, divides multiple semiconductor products into product groups corresponding to the central feature data;

[0096] B3, determine whether the similarity parameters between the central feature data and the surface feature data of semiconductor products in each product group meet the convergence condition;

[0097] B4. If the convergence condition is not met, update the central feature data corresponding to the product group based on the surface feature data of the semiconductor products contained in the product group, and return to execute the step of classifying multiple semiconductor products into the product group corresponding to the central feature data based on the similarity parameters of the central feature data and the surface feature data, until the convergence condition is met.

[0098] In step B1, K central feature data can be randomly determined based on the predetermined number of groups K. The central feature data is also a vector composed of n indicators, where the value of each indicator is randomly determined.

[0099] Taking K equal to 2 as an example, two central feature data can be randomly determined in B1. Please refer to [link / reference]. Figure 2 (1) and (2), Figure 2 In (1), each green dot represents the surface feature data corresponding to a silicon wafer. Figure 2 In (2), the red cross points represent one randomly determined central feature data, and the blue cross points represent another randomly determined central feature data.

[0100] In step B2, for each surface feature data, the similarity parameter between the surface feature data and each central feature data can be calculated. Based on the similarity parameter, the central feature data with the highest similarity to the surface feature data is determined. The surface feature data and its corresponding silicon wafer are divided into the group of the central feature data with the highest similarity. After dividing each surface feature data in this way, K product groups can be obtained.

[0101] The similarity parameter can include the Euclidean distance dist(X, Y) between the face feature data and the center feature data.

[0102] This parameter can be calculated using the following formula (1).

[0103]

[0104] Where X represents the surface feature data, x i Let Y represent the i-th index of the face feature data, and let Y represent the center feature data. i This represents the i-th index of the central feature data. The smaller the Euclidean distance between the surface feature data and the central feature data, the higher the similarity between them. Therefore, when grouping, the surface feature data and its corresponding silicon wafer can be assigned to the group of central feature data with the smallest Euclidean distance.

[0105] The similarity parameter can also include the normalized Euclidean distance sim1 between the face feature data and the center feature data. This parameter can be calculated using the following formula (2), where the Euclidean distance dist(X, Y) is the Euclidean distance between the face feature data and the center feature data calculated according to formula (1).

[0106] sim1=1÷(1+dist(X,Y)) (2)

[0107] The larger the normalized Euclidean distance between the surface feature data and the central feature data, the higher the similarity between them. Therefore, when grouping, the surface feature data and its corresponding silicon wafer can be assigned to the group of the central feature data with the largest normalized Euclidean distance.

[0108] The similarity parameter can include the cosine similarity cos(X, Y) between the face feature data and the center feature data. This parameter can be calculated using the following formula (3).

[0109]

[0110] The greater the cosine similarity between surface feature data and central feature data, the higher the degree of similarity between the two. Therefore, when grouping, the surface feature data and its corresponding silicon wafer can be assigned to the group of central feature data with the highest cosine similarity.

[0111] The similarity parameter can include the normalized cosine similarity sim2 between the face feature data and the center feature data. This parameter can be calculated using the following formula (4).

[0112] sim2=0.5+0.5cos(X,Y) (4)

[0113] The greater the normalized cosine similarity between surface feature data and central feature data, the higher the degree of similarity between the two. Therefore, when grouping, the surface feature data and its corresponding silicon wafer can be assigned to the group of central feature data with the highest normalized cosine similarity.

[0114] The similarity parameter can also include the root mean square value of the Euclidean distance and cosine similarity between the face feature data and the central feature data, denoted as RMS(dist(X,Y), cos(X,Y)), or it can include the root mean square value of the normalized Euclidean distance and normalized cosine similarity between the face feature data and the central feature data, denoted as RMS(sim1, sim2). RMS indicates that the root mean square value of the data within the parentheses is calculated. For the specific calculation method of the root mean square value, please refer to relevant technologies, which will not be elaborated further here.

[0115] The larger the root mean square value of the surface feature data and the center feature data, the higher the similarity between them. Therefore, when grouping, the surface feature data and its corresponding silicon wafer can be assigned to the group of the center feature data with the largest root mean square value.

[0116] The similarity parameter can also include any combination of parameters from the previous examples, such as normalized Euclidean distance and normalized cosine similarity, or it can include Euclidean distance and cosine similarity.

[0117] When multiple parameters are involved, grouping can be prioritized based on one of them. For example, when Euclidean distance and cosine similarity are involved, grouping should be prioritized based on cosine similarity.

[0118] After grouping according to step B2, multiple facet feature data can be assigned to groups corresponding to two central feature data. Please refer to [link / reference]. Figure 2 (3), where the red dots are assigned to the group corresponding to the red cross, and the blue dots are assigned to the group corresponding to the blue cross.

[0119] In step B3, for each product group, the similarity parameters between the feature data of each face shape in the product group and the central feature data of the group can be obtained. Then, the average of multiple similarity parameters is taken as the group similarity parameter of the product group. Then, it is determined whether the group similarity parameter of each product group meets the convergence condition.

[0120] If the group similarity parameter of at least one product group does not meet the convergence condition, it is determined that the similarity parameter of the center feature data and the surface feature data of the semiconductor product in each product group does not meet the convergence condition. In this case, step B4 can be executed.

[0121] If the similarity parameters of each product group meet the convergence condition, then the similarity parameters of the central feature data and the surface feature data of the semiconductor product in each product group meet the convergence condition. In this case, the multiple product groups at this time can be used as the output of the defect prediction model.

[0122] In this embodiment, if the similarity parameter used in B2 is negatively correlated with the degree of similarity (e.g., the similarity parameter is Euclidean distance), then the convergence condition can be that the group similarity parameter is less than or equal to a preset convergence threshold. If the similarity parameter used in B2 is positively correlated with the degree of similarity (e.g., the similarity parameter is normalized Euclidean distance, cosine similarity, or normalized cosine similarity), then the convergence condition can be that the group similarity parameter is greater than or equal to a preset convergence threshold. Different convergence thresholds can be set for different similarity parameters based on relevant experience; this embodiment does not impose any limitations.

[0123] If the similarity parameters include the aforementioned multiple parameters, and at least one similarity parameter does not meet the convergence condition, then step B4 can be executed. If each parameter meets the convergence condition, then the defect prediction model is output.

[0124] If the convergence condition is not met, step B4 can be executed to redetermine new center feature data for each product group based on the face feature data of that group.

[0125] Please see Figure 2 (4) In cases where the convergence condition is not met, it can be determined that... Figure 2 In (3), the surface feature data corresponding to the red dots are used to redetermine a new central feature data, such as... Figure 2 As shown by the red intersection in (4), according to Figure 2 In (3), the surface feature data corresponding to the blue dots are used to redetermine a new central feature data, such as... Figure 2 The blue intersection point is shown in (4).

[0126] After determining the new central feature data in this way, you can return to step B2 to regroup based on the new central feature data. For example, based on... Figure 2 After the new central feature data determined in (4) is regrouped, the silicon wafer and the corresponding surface feature data can be reclassified into Figure 2 The two groups shown in (5) can be repeatedly executed from step B2 to B4 until the convergence condition is met after a certain grouping.

[0127] For example, when the convergence condition is met, it can be obtained Figure 2 As shown in (6), the defect prediction model includes two product groups, namely Figure 2 In (6), the red and blue groups, each of which can include a central feature data, i.e. Figure 2 The central feature data corresponding to the red intersection points in (6) and the central feature data corresponding to the blue intersection points.

[0128] When constructing a defect prediction model based on the above method, the way to determine the front-end process label and defect surface type label for product grouping can be:

[0129] Among multiple semiconductor products corresponding to a product group, select the first semiconductor product whose similarity parameters between the corresponding surface feature data and the center feature data of the product group meet the target conditions;

[0130] Based on the vertical surface profile information of each first semiconductor product, the defect surface profile features of each first semiconductor product are obtained, and the defect surface profile feature that appears most frequently is determined as the defect surface profile label corresponding to the product group.

[0131] Obtain the front-end process information for each first semiconductor product, and determine the front-end process information that appears most frequently as the front-end process label corresponding to the product group.

[0132] In other words, unlike the aforementioned method of determining labels, this embodiment first needs to filter the silicon wafers within the product group based on the surface feature data and the similarity parameters of the center of the product group when determining the label for each product group. A portion of the silicon wafers whose similarity parameters meet the target conditions are selected. Then, the defect surface feature that appears most frequently in this portion of silicon wafers is determined as the defect surface feature label, and the front-end process information that appears most frequently is determined as the front-end process label for the group.

[0133] For example, suppose a product group X has 10 silicon wafers whose surface feature data similarity parameters meet the target conditions. The most frequent defect surface feature among these 10 silicon wafers is a tornado-like undulation at the 12 o'clock position. This defect surface feature appears in 8 silicon wafers that meet the target conditions. The most frequent front-end process information includes: lithography chamber, 201; etching machine, 305. This front-end process information appears in 9 silicon wafers that meet the target conditions. Therefore, it can be determined that the defect surface feature label for product group X is: a tornado-like undulation at the 12 o'clock position, and the corresponding front-end process label is: lithography chamber, 201; etching machine, 305.

[0134] In this embodiment, if the similarity parameter meets the target condition, the similarity between the surface feature data corresponding to the silicon wafer and the central feature data of the product group is high; if the target condition is not met, the similarity between the surface feature data corresponding to the silicon wafer and the central feature data of the product group is low.

[0135] One method for selecting the first semiconductor product that meets the target conditions based on similarity parameters is as follows:

[0136] For each semiconductor product in the product group, the root mean square of the similarity parameter is calculated based on the first similarity parameter and the second similarity parameter of the surface feature data corresponding to the semiconductor product and the center feature data of the product group.

[0137] The semiconductor products corresponding to the product group are sorted in descending order according to the root mean square of the similarity parameter, and the first N semiconductor products are determined as the first semiconductor product that meets the target condition, where N is a preset integer.

[0138] The first similarity parameter here can be the aforementioned normalized Euclidean distance, and the second similarity parameter can be the aforementioned normalized cosine similarity. In this case, the root mean square of the similarity parameter can be RMS(sim1, sim2). Alternatively, the first similarity parameter can be the aforementioned Euclidean distance, and the second similarity parameter can be the aforementioned cosine similarity. In this case, the root mean square of the similarity parameter can be RMS(dist(X, Y), cos(X, Y)). That is to say, for any silicon wafer, the root mean square of the similarity parameter corresponding to the silicon wafer can be the root mean square value of the first and second similarity parameters of the surface feature data of the silicon wafer and the central feature data of the group to which it belongs. The specific calculation method of the root mean square value can be found in the relevant technology and will not be elaborated here.

[0139] Since a larger root mean square of the similarity parameter corresponding to the surface feature data indicates a higher degree of similarity between the surface feature data and the central feature data, all silicon wafers in a product group can be sorted in descending order according to the root mean square of the corresponding similarity parameter. The larger the root mean square of the similarity parameter, the earlier the wafer is sorted, and the smaller the root mean square of the similarity parameter, the later the wafer is sorted. Then, the first N silicon wafers after sorting are selected as the first silicon wafer in the product group that meets the target conditions (i.e., the first semiconductor product mentioned above).

[0140] N can be set according to the number of silicon wafers contained in the product group. For example, when a product group contains 20 silicon wafers, N can be set to 10, 15, or other values ​​without limitation.

[0141] In some alternative embodiments, to eliminate the interference of uneven workpiece stage height, the defect prediction model can also be obtained as follows:

[0142] C1, obtain the vertical surface profile information of the ultra-flat sheet;

[0143] C2, based on the vertical surface profile information of the ultra-flat wafer, correct the vertical surface profile information of each semiconductor product to obtain the corrected vertical surface profile information;

[0144] C3 clusters multiple semiconductor products based on their corrected vertical surface information to obtain a defect prediction model that includes multiple product groups.

[0145] The vertical surface profile information of the ultra-flat wafer can be obtained by scanning the wafer using a leveling sensor or other sensors with similar functions. This surface profile information reflects the vertical undulations of the workpiece stage. The scanned vertical surface profile information of the ultra-flat wafer can also be stored in a lithography data center using an XML file format.

[0146] The correction method in step C2 can be to subtract the vertical surface profile information of the ultra-flat wafer from the vertical surface profile information of the silicon wafer to obtain the corrected vertical surface profile information.

[0147] The method of clustering based on the corrected vertical surface information in step C3 can be the same as the method of clustering based on vertical surface information in the previous embodiment, and will not be described again.

[0148] In some optional embodiments, after the defect prediction model is constructed, it can be updated periodically based on the front-end process information and vertical surface information of the subsequently produced new silicon wafers to improve the accuracy of the prediction results obtained based on the defect prediction model.

[0149] One application of the defect prediction model constructed in this embodiment is to quickly identify the machine or chamber that may cause the vertical surface defect in a newly produced silicon wafer that has undergone previous processing when it is determined that the wafer has a vertical surface defect.

[0150] Specifically, if a silicon wafer processed by a previous layer is found to have a vertical surface defect, the machine or chamber that may have caused the vertical surface defect can be identified as follows:

[0151] D1, obtain the second defect surface feature corresponding to the second semiconductor product, the second semiconductor product being a semiconductor product processed by the previous layer process;

[0152] D2, find the second defect facet label that matches the second defect facet feature in the defect facet labels of multiple product groups in the defect prediction model;

[0153] D3. Based on the front-end process label of the product group to which the second defect surface type label belongs, determine the front-end process information of the second semiconductor product related to the second defect surface type feature.

[0154] The second semiconductor product in D1 can refer to a silicon wafer that has undergone previous processing and has been confirmed to have vertical surface defects, or it can be simply referred to as the second silicon wafer. In D1, the second silicon wafer can be scanned to obtain its vertical surface information, and then the defect surface features of the second silicon wafer can be obtained based on the vertical surface information, i.e., the second defect surface features of D1.

[0155] In D2, the defect facet label of each product group in the defect prediction model can be compared with the second defect facet feature. If the defect facet label of a certain product group contains the second defect facet feature, then the defect facet label can be determined to be the second defect facet label.

[0156] For example, the second defect surface feature of the second silicon wafer may be a tornado-like undulation in the 6 o'clock direction. If the defect surface label of a certain product group Y also includes a tornado-like undulation in the 6 o'clock direction, then the defect surface label of product group Y can be determined to be a matching second defect surface label.

[0157] In D3, the preceding process label of the product group to which the second defect surface label belongs can be obtained, and the preceding process label can be identified as the preceding process information that may lead to the second defect surface feature.

[0158] Based on the aforementioned example, assuming the front-end process label corresponding to product group Y is: lithography chamber, 207; etching machine, 306, then it can be determined that the front-end process information of the second semiconductor product related to the second defect surface feature is lithography chamber, 207; etching machine, 306. That is, it can be determined that the tornado-like undulations in the 6 o'clock direction of the second silicon wafer may be caused by processing of at least one of the lithography chamber 207 and the etching machine 306.

[0159] Another application of the defect prediction model constructed in this embodiment is to predict the possible vertical surface defects of a silicon wafer before actually performing front-end processing on the silicon wafer, based on the machine and chamber expected to be used to process the silicon wafer.

[0160] Prediction methods can include:

[0161] E1, obtain the third front-end process information of the third semiconductor product, which is a semiconductor product that has not undergone front-end process processing;

[0162] E2, search for the third front-end process label that matches the third front-end process information in the front-end process labels of multiple product groups in the defect prediction model.

[0163] E3 predicts the predicted defect surface features of the third semiconductor product based on the defect surface label of the product group to which the third front-end process label belongs.

[0164] Third semiconductor products can refer to silicon wafers that have not yet undergone previous processing steps, and can also be called third silicon wafers.

[0165] The third front-end process information of the third silicon wafer can characterize the equipment and chambers that are expected to be used for front-end process processing of the third silicon wafer. Specifically, it can include the equipment identifiers of the equipment and the chamber identifiers of the chambers that are expected to be used for front-end process processing of the third silicon wafer at each process stage.

[0166] The third front-end process information can be automatically configured by the factory automation system. Correspondingly, the third front-end process information can be obtained from the factory automation system in step E1.

[0167] In step E2, the front-layer process label and the third front-layer process information of each product group in the defect prediction model can be compared. If a certain front-layer process label appears in the third front-layer process information, it can be determined that the front-layer process label is the third front-layer process label that matches the third front-layer process information.

[0168] For example, suppose the front-end process label corresponding to a certain product group Z is: lithography chamber, 208; etching machine, 304, and the third front-end process information also includes lithography chamber, 208; etching machine, 304, then it can be determined that the front-end process label is a matching third front-end process label.

[0169] In step E3, the defect surface type label of the product group to which the third front-end process label belongs can be obtained, and the defect surface type features contained in the defect surface type label can be identified as the predicted defect surface type features of the third semiconductor product.

[0170] For example, assuming that the defect surface type label corresponding to the aforementioned product group Z is that there are tornado-like undulations in the 6 o'clock direction, it can be determined that the predicted defect surface type feature of the third semiconductor product is that there are tornado-like undulations in the 6 o'clock direction. That is, it is determined that after the third silicon wafer is processed according to the third front-layer process information, the possible vertical surface type defect is: there are tornado-like undulations in the 6 o'clock direction.

[0171] Optionally, after predicting potential vertical surface defects in the silicon wafer to be processed, countermeasures can be taken in at least one of the following ways to avoid or mitigate the potential vertical surface defects:

[0172] Countermeasure 1: Adjust the processing parameters of the machine and chamber corresponding to the third semiconductor product according to the predicted defect surface features to avoid the occurrence of the predicted defect surface features;

[0173] The second countermeasure is to adjust the design parameters of the third semiconductor product based on the predicted defect surface features in order to avoid the occurrence of predicted defect surface features.

[0174] In the first countermeasure, relevant personnel can adjust the processing parameters of the machine or chamber used to process the third silicon wafer based on the instructions from the third front-end process information. Furthermore, the processing parameters of the machine or chamber indicated in the matched third front-end process label can be adjusted to avoid the occurrence of predicted defect surface features.

[0175] In the second countermeasure, the design parameters of the third silicon wafer can be adjusted based on the predicted defect surface features. This ensures that the third silicon wafer used for processing has sufficient processing allowance in areas where corresponding vertical surface defects may exist, thereby avoiding the occurrence of predicted defect surface features in those areas. The adjustment of design parameters can be made by relevant personnel based on the predicted defect surface features and their experience.

[0176] In some optional embodiments, indicator monitoring can also be performed based on the obtained defect prediction model. Specifically, for each product group, the surface feature data of those silicon wafers in that product group that have the defect surface feature label of that group can be obtained, and the values ​​of these surface feature data on various indicators can be statistically analyzed to obtain the threshold corresponding to each indicator.

[0177] For example, suppose the defect surface type label of a certain group is that there are tornado-like undulations in the 12 o'clock direction. We can obtain the silicon wafers in the product group that have this vertical surface type defect, calculate the mean of the indicators related to the vertical surface type information in the 12 o'clock direction in the surface type feature data of these silicon wafers, such as calculating the mean of KPI4 in the 12 o'clock direction, and determine the calculated mean as the threshold corresponding to the vertical surface type defect of tornado-like undulations in the 12 o'clock direction.

[0178] For example, suppose there are 10 silicon wafers in the group with a tornado-like undulation defect in the 12 o'clock direction. We can obtain the KPI4 of these 10 silicon wafers, calculate the average of the 10 KPI4s, and determine the result as the threshold corresponding to the vertical surface defect of the tornado-like undulation in the 12 o'clock direction.

[0179] Therefore, after a silicon wafer is processed, it is possible to directly test whether the corresponding index of the silicon wafer is greater than the corresponding threshold to determine whether the silicon wafer has a corresponding vertical surface defect, which can improve the efficiency of detecting vertical surface defects.

[0180] Based on the aforementioned example, after obtaining the threshold corresponding to the vertical surface defect of tornado-like undulations at the 12 o'clock position, if the KPI4 of a newly processed silicon wafer is greater than the threshold, it can be determined that the silicon wafer has the defect of tornado-like undulations at the 12 o'clock position; if the KPI4 of the silicon wafer is not greater than the threshold, it can be determined that the silicon wafer does not have the defect of tornado-like undulations at the 12 o'clock position.

[0181] This application also provides an apparatus for predicting the vertical surface profile of a semiconductor product. Please refer to [link to relevant documentation]. Figure 3 This is a schematic diagram of the structure of the device, which may include the following units.

[0182] The acquisition unit 301 is used to acquire front-end process information and vertical surface information of multiple semiconductor products. The front-end process information represents the equipment and chamber used to perform front-end process processing on the corresponding semiconductor products, and the vertical surface information represents the height and undulation of the corresponding semiconductor products in the vertical direction.

[0183] Clustering unit 302 is used to cluster multiple semiconductor products based on the vertical surface information of the semiconductor products to obtain a defect prediction model that includes multiple product groups.

[0184] The determination unit 303 is used to determine the front-end process label and defect surface type label of each product group in the defect prediction model based on the vertical surface type information and front-end process information of the semiconductor product corresponding to the product group; the front-end process label and defect surface type label of multiple product groups in the defect prediction model are used to predict the defect surface type features of semiconductor products that have not undergone front-end process processing.

[0185] Optionally, when clustering unit 302 performs clustering processing on multiple semiconductor products based on the vertical surface shape information of the semiconductor products to obtain a defect prediction model including multiple product groups, it is specifically used for:

[0186] For each semiconductor product, multiple surface profile indicators are determined based on the vertical surface profile information of the semiconductor product to obtain surface profile feature data of the semiconductor product composed of multiple surface profile indicators.

[0187] Clustering is performed on the surface feature data of semiconductor products to obtain a defect prediction model that includes multiple product groups.

[0188] Optionally, when clustering unit 302 performs clustering processing based on the surface feature data of semiconductor products to obtain a defect prediction model including multiple product groups, it is specifically used for:

[0189] Identify multiple central feature data;

[0190] Based on the similarity parameters between the central feature data and the surface feature data, multiple semiconductor products are divided into product groups corresponding to the central feature data.

[0191] Determine whether the similarity parameters between the central feature data and the surface feature data of semiconductor products in each product group meet the convergence condition;

[0192] If the convergence condition is not met, update the central feature data corresponding to the product group based on the surface feature data of the semiconductor products contained in the product group, and return to execute the step of classifying multiple semiconductor products into the product group corresponding to the central feature data based on the similarity parameters of the central feature data and the surface feature data, until the convergence condition is met.

[0193] Optionally, when determining the front-end process label and defect surface type label of the product group based on the vertical surface type information and front-end process information of the semiconductor product corresponding to the product group, the determining unit 303 is specifically used for:

[0194] Among multiple semiconductor products corresponding to a product group, select the first semiconductor product whose similarity parameters between the corresponding surface feature data and the center feature data of the product group meet the target conditions;

[0195] Based on the vertical surface profile information of each first semiconductor product, the defect surface profile features of each first semiconductor product are obtained, and the defect surface profile feature that appears most frequently is determined as the defect surface profile label corresponding to the product group.

[0196] Obtain the front-end process information for each first semiconductor product, and determine the front-end process information that appears most frequently as the front-end process label corresponding to the product group.

[0197] Optionally, when determining unit 303 selects the first semiconductor product from among multiple semiconductor products corresponding to a product group whose similarity parameters between the corresponding surface feature data and the center feature data of the product group satisfy the target condition, it is specifically used for:

[0198] For each semiconductor product in the product group, the root mean square of the similarity parameter is calculated based on the first similarity parameter and the second similarity parameter of the surface feature data corresponding to the semiconductor product and the center feature data of the product group.

[0199] The semiconductor products corresponding to the product group are sorted in descending order according to the root mean square of the similarity parameter, and the first N semiconductor products are determined as the first semiconductor product that meets the target condition, where N is a preset integer.

[0200] Optionally, when clustering unit 302 performs clustering processing on multiple semiconductor products based on the vertical surface shape information of the semiconductor products to obtain a defect prediction model including multiple product groups, it is specifically used for:

[0201] Obtain the vertical surface profile information of the ultra-flat sheet;

[0202] The vertical surface profile information of each semiconductor product is corrected based on the vertical surface profile information of the ultra-flat wafer to obtain the corrected vertical surface profile information.

[0203] Based on the corrected vertical surface information of semiconductor products, multiple semiconductor products are clustered to obtain a defect prediction model that includes multiple product groups.

[0204] Optionally, the device further includes an application unit 304 for:

[0205] Obtain the second defect surface feature corresponding to the second semiconductor product, wherein the second semiconductor product is a semiconductor product processed by the previous layer process;

[0206] Find the second defect facet label that matches the second defect facet feature among the defect facet labels of multiple product groups in the defect prediction model;

[0207] Based on the front-layer process label of the product group to which the second defect surface feature label belongs, determine the front-layer process information of the second semiconductor product related to the second defect surface feature.

[0208] Optionally, application unit 304 can also be used for:

[0209] Obtain the third front-end process information of the third semiconductor product, which is a semiconductor product that has not undergone front-end process processing;

[0210] Search for the third front-end process label that matches the third front-end process information in the front-end process labels of multiple product groups in the defect prediction model.

[0211] Based on the defect surface type label of the product group to which the third front-layer process label belongs, predict the predicted defect surface type characteristics of the third semiconductor product.

[0212] Optionally, the device may also include an adjustment unit 305 for at least one of the following:

[0213] The processing parameters of the machine and chamber corresponding to the third semiconductor product are adjusted according to the predicted defect surface features in order to avoid the occurrence of the predicted defect surface features.

[0214] The design parameters of the third semiconductor product are adjusted based on the predicted defect surface features to avoid the occurrence of predicted defect surface features.

[0215] The working principle of the apparatus for predicting the vertical surface profile of a semiconductor product provided in this embodiment can be found in the relevant steps of the method for predicting the vertical surface profile of a semiconductor product provided in any embodiment of this application, and will not be repeated here.

[0216] This application also provides an electronic device; please refer to [link to relevant documentation]. Figure 4 The device may include a memory 401 and a processor 402.

[0217] Memory 401 is used to store computer programs.

[0218] The processor 402 is used to execute a computer program to implement the method for predicting the vertical profile of a semiconductor product provided in any embodiment of this application.

[0219] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0220] For ease of description, the above systems or devices are described separately by function, divided into various modules or units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0221] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0222] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0223] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting the vertical surface profile of a semiconductor product, characterized in that, include: Obtain front-end process information and vertical surface profile information for multiple semiconductor products. The front-end process information characterizes the equipment and chambers used to perform front-end process processing on the corresponding semiconductor products, and the vertical surface profile information characterizes the vertical undulations of the corresponding semiconductor products. Based on the vertical surface shape information of the semiconductor products, multiple semiconductor products are clustered to obtain a defect prediction model that includes multiple product groups; For each product group in the defect prediction model, the front-end process label and defect surface label of the product group are determined based on the vertical surface information and front-end process information of the semiconductor product corresponding to the product group. The defect prediction model includes multiple product group front-end process labels and defect surface type labels, which are used at least to predict the defect surface type features of semiconductor products that have not undergone front-end process processing.

2. The method according to claim 1, characterized in that, The step of clustering multiple semiconductor products based on their vertical surface profile information to obtain a defect prediction model comprising multiple product groups includes: For each semiconductor product, multiple surface profile indicators are determined based on the vertical surface profile information of the semiconductor product to obtain surface profile feature data of the semiconductor product composed of the multiple surface profile indicators; Clustering is performed on the surface feature data of the semiconductor products to obtain a defect prediction model that includes multiple product groups.

3. The method according to claim 2, characterized in that, The step of clustering based on the surface feature data of the semiconductor products to obtain a defect prediction model including multiple product groups includes: Identify multiple central feature data; Based on the similarity parameters between the central feature data and the surface feature data, the multiple semiconductor products are divided into product groups corresponding to the central feature data; Determine whether the similarity parameters between the central feature data and the surface feature data of the semiconductor product in each product group meet the convergence condition; If the convergence condition is not met, the center feature data corresponding to the product group is updated according to the surface feature data of the semiconductor products contained in the product group, and the process of dividing the multiple semiconductor products into the product group corresponding to the center feature data according to the similarity parameter of the center feature data and the surface feature data is returned until the convergence condition is met.

4. The method according to claim 1, characterized in that, The step of determining the front-end process label and defect surface type label of the product group based on the vertical surface type information and front-end process information of the semiconductor product corresponding to the product group includes: Among the multiple semiconductor products corresponding to the product group, the first semiconductor product whose similarity parameter between the corresponding surface feature data and the center feature data of the product group satisfies the target condition is selected. Based on the vertical surface profile information of each of the first semiconductor products, the defect surface profile features of each of the first semiconductor products are obtained, and the defect surface profile feature that appears most frequently is determined as the defect surface profile label corresponding to the product group. Obtain the front-end process information for each of the first semiconductor products, and determine the front-end process information that appears most frequently as the front-end process label corresponding to the product group.

5. The method according to claim 4, characterized in that, The step of selecting a first semiconductor product from among the multiple semiconductor products corresponding to the product group, whose similarity parameter between the corresponding surface feature data and the center feature data of the product group satisfies the target condition, includes: For each semiconductor product in the product group, a root mean square similarity parameter is calculated based on a first similarity parameter and a second similarity parameter of the surface feature data corresponding to the semiconductor product and the center feature data of the product group. The semiconductor products corresponding to the product group are arranged in descending order according to the root mean square of the similarity parameter, and the first N semiconductor products are determined to be the first semiconductor products that meet the target conditions, where N is a preset integer.

6. The method according to claim 1, characterized in that, The step of clustering multiple semiconductor products based on their vertical surface profile information to obtain a defect prediction model comprising multiple product groups includes: Obtain the vertical surface profile information of the ultra-flat sheet; The vertical surface profile information of each semiconductor product is corrected based on the vertical surface profile information of the ultra-flat wafer to obtain the corrected vertical surface profile information. Based on the corrected vertical surface information of the semiconductor products, multiple semiconductor products are clustered to obtain a defect prediction model that includes multiple product groups.

7. The method according to claim 1, characterized in that, Also includes: Obtain the second defect surface feature corresponding to the second semiconductor product, wherein the second semiconductor product is a semiconductor product processed by the previous layer process; Search for a second defect facet label that matches the second defect facet feature among the defect facet labels of the multiple product groups in the defect prediction model; Based on the front-end process label of the product group to which the second defect surface label belongs, determine the front-end process information of the second semiconductor product related to the second defect surface feature.

8. The method according to any one of claims 1 to 7, characterized in that, Also includes: Obtain the third front-end process information of the third semiconductor product, wherein the third semiconductor product is a semiconductor product that has not undergone front-end process processing; The third front-end process label that matches the third front-end process information is searched among the front-end process labels of the multiple product groups in the defect prediction model. Based on the defect surface type label of the product group to which the third front-layer process label belongs, predict the predicted defect surface type features of the third semiconductor product.

9. The method according to claim 8, characterized in that, It also includes at least one of the following: The processing parameters of the machine and chamber corresponding to the third semiconductor product are adjusted according to the predicted defect surface features to avoid the occurrence of the predicted defect surface features; The design parameters of the third semiconductor product are adjusted based on the predicted defect surface features to avoid the occurrence of the predicted defect surface features.

10. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is used to execute the computer program to implement the method for predicting the vertical profile of a semiconductor product as described in any one of claims 1 to 9.

Images