Process inspection method, apparatus, device, and storage medium

Abstract

This application discloses a process inspection method, apparatus, equipment, and storage medium, belonging to the field of computer technology. The method includes: obtaining the weight of each of multiple influencing factors; acquiring target data corresponding to each of the multiple influencing factors during the manufacturing process; predicting the product yield based on the target data; and, in the event of an abnormal predicted product yield, inspecting the process flow to which each influencing factor belongs, in descending order of its weight, until a problematic process flow is identified. This application prioritizes inspecting the process flow to which the influencing factors with higher weights belong, i.e., prioritizing the inspection of the process flow most likely to have problems. This order of inspection, from highest to lowest probability of problem occurrence, allows for faster identification of problematic process flows, thereby saving manpower, reducing process inspection time, and improving process inspection efficiency.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a process inspection method, apparatus, equipment, and storage medium. Background Technology

[0002] Currently, with rapid technological advancements and the widespread use of integrated circuits, coupled with increasing user demand for electronic products, the importance of integrated circuit production volume is becoming increasingly prominent. One of the most crucial indicators of integrated circuit production volume is chip yield. A high yield saves manufacturing costs, while a low yield wastes resources and increases costs. A chip typically requires numerous process steps to manufacture, and problems in any of these steps can affect chip quality, thus reducing yield. Therefore, the process steps in chip manufacturing are extremely important. Consequently, inspecting the process steps during chip manufacturing is crucial.

[0003] In related technologies, formulas are used To determine the chip yield, let Y represent the chip output, B represent the number of defects per unit area of ​​the chip, A represent the complexity of the manufacturing process, and C represent the cost of the wafer materials used to manufacture the chip. When the chip yield is lower than a preset value, technicians inspect each process step by examining the equipment used in each step of the chip manufacturing process. This allows them to trace back to the process step where the problem occurred and the most serious issue was identified, enabling them to address the problem and improve the chip yield.

[0004] However, in the above method, technicians need to inspect every process step in the chip manufacturing process to trace back to the process where the problem occurred and was most severe. This is a huge waste of time and manpower. Summary of the Invention

[0005] This application provides a process inspection method, apparatus, equipment, and storage medium, which can save process inspection time and improve process inspection efficiency. The technical solution is as follows:

[0006] In a first aspect, a process inspection method is provided, the method comprising:

[0007] Obtain the weight of each of the multiple influencing factors, where each of the multiple influencing factors is a factor that affects the quality of the item during the manufacturing process.

[0008] During the manufacturing process of the item, acquire the target data corresponding to each of the multiple influencing factors;

[0009] Predict the product yield based on the target data;

[0010] If the predicted yield of the product is abnormal, the process flow to which each of the multiple influencing factors belongs is checked in descending order of the weight of each influencing factor until the problematic process flow is found.

[0011] In this application, the weight of each of multiple influencing factors is obtained. Each influencing factor is a factor that affects the quality of the product during the manufacturing process. The larger the weight of an influencing factor, the greater its impact on the quality of the product during the manufacturing process. During the product manufacturing process, target data corresponding to each of the multiple influencing factors is obtained. Then, based on the target data, the product yield is predicted. This predicted yield is the proportion of qualified products in the predicted batch. Subsequently, if the predicted yield is abnormal, the process flow to which each influencing factor belongs is checked in descending order of its weight until the problematic process flow is identified. Therefore, prioritizing the inspection of process flows with higher-weighted influencing factors ensures that process flows with factors that have a greater impact on product quality are inspected first. In other words, process flows most likely to cause product quality abnormalities are inspected first, i.e., process flows most likely to have problems. This process checks each process flow in descending order of the probability of problems, and stops when a problematic process flow is found. In this way, problematic process flows can be found more quickly, thereby saving manpower, saving process inspection time, and thus improving process inspection efficiency.

[0012] Optionally, obtaining the weight of each impact factor among multiple impact factors includes:

[0013] Obtain a directed graph between the multiple influencing factors. The directed path in the directed graph includes a directed edge and the length of the directed edge. The length of the directed edge is the degree of influence of the starting point of the directed edge on the ending point.

[0014] Based on the directed graph, establish the influence relationship matrix of the multiple influencing factors;

[0015] Based on the influence relationship matrix, determine the influence degree and the degree of influence of each of the plurality of influence factors;

[0016] For any one of the multiple influencing factors, the influence degree of the influencing factor is added to the influence degree of the influencing factor to obtain the centrality of the influencing factor;

[0017] The weight of each of the plurality of influence factors is determined based on the centrality of each influence factor.

[0018] Optionally, determining the weight of each impact factor among the plurality of impact factors based on the centrality of each impact factor includes:

[0019] The centrality of each of the multiple impact factors is summed to obtain the target value.

[0020] The weight of an influence factor is obtained by dividing the centrality of the influence factor by the target value.

[0021] Optionally, predicting the yield rate of the item based on the target data includes:

[0022] The target data is input into the prediction model so that the prediction model can output the number of qualified items. The prediction model is used to predict the number of qualified items out of a preset number of items.

[0023] The yield rate of the finished product is obtained by dividing the number of qualified items by the preset quantity.

[0024] Optionally, the prediction model includes an input layer, a feature layer, an enhancement layer, and an output layer. The feature layer performs k linear transformations on the target data to obtain k first feature matrices, and merges the k first feature matrices to obtain a second feature matrix. The enhancement layer performs m nonlinear transformations on the second feature matrix to obtain m third feature matrices, and merges the m third feature matrices to obtain a fourth feature matrix. The output layer merges the second feature matrix and the fourth feature matrix to obtain a fifth feature matrix of size n×1, obtains a parameter matrix of size 1×n for the prediction model, and multiplies the fifth feature matrix with the parameter matrix to obtain the number of qualified items. Here, k, m, and n are all positive integers.

[0025] Optionally, before checking the process flow to which each of the multiple influencing factors belongs in descending order of their weights when the predicted yield of the product is abnormal, the method further includes:

[0026] Obtain a historical dataset, which includes historical inspection data for each item in each batch of multiple batches of items that have been manufactured and have a normal actual yield, and the actual number of qualified items in each batch of multiple batches of items. The historical inspection data is the data corresponding to each of the multiple influencing factors obtained during the manufacturing process of each item.

[0027] Determine the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items;

[0028] The standard yield rate is determined based on the average of the historical test data of the multiple batches of items and the average of the actual qualified quantity of the multiple batches of items.

[0029] If the finished product yield of the item is less than the standard finished product yield, then the finished product yield of the item is determined to be abnormal.

[0030] Optionally, the item is a chip.

[0031] Secondly, a process inspection apparatus is provided, the apparatus comprising:

[0032] The first acquisition module is used to acquire the weight of each of the multiple influencing factors, wherein each of the multiple influencing factors is a factor that will affect the quality of the item during the manufacturing process.

[0033] The second acquisition module is used to acquire target data corresponding to each of the multiple influencing factors during the article manufacturing process;

[0034] The prediction module is used to predict the yield rate of the items based on the target data;

[0035] The inspection module is used to check the process flow to which each of the multiple influencing factors belongs in descending order of the weight of each influencing factor when the predicted yield of the product is abnormal, until the problematic process flow is found.

[0036] Optionally, the first acquisition module is used to:

[0037] Obtain a directed graph between the multiple influencing factors. The directed path in the directed graph includes a directed edge and the length of the directed edge. The length of the directed edge is the degree of influence of the starting point of the directed edge on the ending point.

[0038] Based on the directed graph, establish the influence relationship matrix of the multiple influencing factors;

[0039] Based on the influence relationship matrix, determine the influence degree and the degree of influence of each of the plurality of influence factors;

[0040] For any one of the multiple influencing factors, the influence degree of the influencing factor is added to the influence degree of the influencing factor to obtain the centrality of the influencing factor;

[0041] The weight of each of the plurality of influence factors is determined based on the centrality of each influence factor.

[0042] Optionally, the first acquisition module is used to:

[0043] The centrality of each of the multiple impact factors is summed to obtain the target value.

[0044] The weight of an influence factor is obtained by dividing the centrality of the influence factor by the target value.

[0045] Optionally, the prediction module is used for:

[0046] The target data is input into the prediction model so that the prediction model can output the number of qualified items. The prediction model is used to predict the number of qualified items out of a preset number of items.

[0047] The yield rate of the finished product is obtained by dividing the number of qualified items by the preset quantity.

[0048] Optionally, the prediction model includes an input layer, a feature layer, an enhancement layer, and an output layer. The feature layer performs k linear transformations on the target data to obtain k first feature matrices, and merges the k first feature matrices to obtain a second feature matrix. The enhancement layer performs m nonlinear transformations on the second feature matrix to obtain m third feature matrices, and merges the m third feature matrices to obtain a fourth feature matrix. The output layer merges the second feature matrix and the fourth feature matrix to obtain a fifth feature matrix of size n×1, obtains a parameter matrix of size 1×n for the prediction model, and multiplies the fifth feature matrix with the parameter matrix to obtain the number of qualified items. Here, k, m, and n are all positive integers.

[0049] Optionally, the device further includes:

[0050] The third acquisition module is used to acquire historical datasets, which include historical inspection data of each item in each batch of multiple batches of items that have been manufactured and have normal actual yield, and the actual number of qualified items in each batch of multiple batches of items. The historical inspection data is the data corresponding to each of the multiple influencing factors acquired during the manufacturing process of each item.

[0051] The first determining module is used to determine the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items;

[0052] The second determining module is used to determine the standard yield rate based on the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items.

[0053] The third determining module is used to determine that the finished product rate of the item is abnormal if the finished product rate of the item is less than the standard finished product rate.

[0054] Optionally, the item is a chip.

[0055] Thirdly, a computer device is provided, the computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-described process inspection method.

[0056] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described process inspection method.

[0057] Fifthly, a computer program product containing instructions is provided, which, when run on a computer, causes the computer to perform the steps of the above-described process inspection method.

[0058] It is understood that the beneficial effects of the second, third, fourth, and fifth aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart of a process inspection method provided in an embodiment of this application;

[0061] Figure 2 This is a schematic diagram of a directed graph provided in an embodiment of this application;

[0062] Figure 3 This is a schematic diagram of influence degree and affected degree provided in an embodiment of this application;

[0063] Figure 4 This is a schematic diagram of centrality-causality provided in an embodiment of this application;

[0064] Figure 5 This is a schematic diagram of the structure of a prediction model provided in an embodiment of this application;

[0065] Figure 6 This is a schematic diagram of the structure of a process inspection device provided in an embodiment of this application;

[0066] Figure 7This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0068] It should be understood that "multiple" as mentioned in this application refers to two or more. In the description of this application, unless otherwise stated, " / " indicates "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist, for example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, to facilitate a clear description of the technical solutions of this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first," "second," etc., do not limit the quantity or execution order, and that "first," "second," etc., do not necessarily imply differences.

[0069] Before providing a detailed explanation of the embodiments of this application, the application scenarios of these embodiments will be described first.

[0070] Currently, integrated circuits (ICs) are widely used in electronic products. An integrated circuit is a circuit that uses specific processes to connect transistors, resistors, capacitors, inductors, and other components and wiring required in a circuit. Fabricating an integrated circuit on a small piece of silicon wafer yields a chip; therefore, a chip is a silicon wafer containing integrated circuits, and its size is extremely small. Chips are also called microcircuits or microchips. As the carrier of integrated circuits, a chip is an independent entity that can be used immediately, and its small size and lightweight characteristics have led to its widespread application in various electronic products.

[0071] As user demand for electronic products grows, chip production volume becomes increasingly important. The most crucial metric for measuring chip production volume is the chip yield rate. A higher yield rate indicates that more chips in a batch are of acceptable quality, resulting in higher overall chip production. Manufacturing a chip typically requires numerous process steps, and problems in any of these steps can affect chip quality, thus reducing the yield rate. Therefore, the process steps in chip manufacturing are extremely important. Consequently, process checks are necessary during chip manufacturing.

[0072] In related technologies, formulas are used Determine the chip yield. When the chip yield is lower than the preset value, technicians check each process step in the chip manufacturing process by examining the equipment used in each step. This allows them to trace back to the process step where the problem occurred and the most serious issue was identified. The corresponding problems can then be resolved to improve the chip yield.

[0073] However, the formula used to determine chip yield in the above method is only applicable to scenarios where there are no sensors in the manufacturing process. Modern chip manufacturing processes incorporate numerous sensors that monitor data throughout the process, rendering the formula inapplicable. Furthermore, this method requires technicians to inspect every step of the chip manufacturing process to pinpoint the most problematic step, resulting in significant time and manpower waste.

[0074] Therefore, this application provides a process inspection method that can be applied to scenarios where the process flow during the manufacturing of articles is inspected.

[0075] For example, the process inspection method provided in this application can be applied to scenarios where the process flow in chip manufacturing is inspected. Specifically, the weight of each of the multiple influencing factors that affect chip quality during chip manufacturing is obtained, that is, the degree of influence of each influencing factor on chip quality during chip manufacturing is obtained. Then, based on the data of multiple process flows detected in the chip manufacturing process, the chip yield of this batch of chips is predicted, and then it is determined whether the predicted chip yield is normal. If the chip yield is abnormal, the process flow to which each influencing factor belongs is checked in descending order of the weight of each influencing factor until a problem is found in the process flow. Therefore, prioritizing the inspection of process flows with higher-weighted influencing factors ensures that process flows with factors that have a greater impact on product quality are inspected first. In other words, process flows most likely to cause product quality abnormalities are inspected first, i.e., process flows most likely to have problems. This process checks each process flow in descending order of the probability of problems, and stops when a problematic process flow is found. In this way, problematic process flows can be found more quickly, thereby saving manpower, saving process inspection time, and thus improving process inspection efficiency.

[0076] The process inspection method provided in the embodiments of this application will be explained in detail below.

[0077] Figure 1 This is a flowchart illustrating a process inspection method provided in an embodiment of this application. This method can be applied to computer equipment; see [link / reference]. Figure 1 The method includes the following steps.

[0078] Step 101: The computer device obtains the weight of each of the multiple impact factors.

[0079] Each of these multiple influencing factors represents a factor that affects the quality of the product during the manufacturing process. The weight of each influencing factor indicates the degree of its impact on the product's quality; a higher weight indicates a greater impact on the product's quality.

[0080] Each of these multiple influencing factors can be pre-set by technicians. Optionally, these multiple influencing factors can be multiple process flows in the article manufacturing process, multiple obstacle factors in the article manufacturing process, or a combination of process flows and obstacle factors in the article manufacturing process. For example, the article is a chip. For instance, these multiple influencing factors can be process flows such as photolithography (A1), ion implantation (A2), etching (A3), heat treatment (A4), chemical vapor deposition (A5), physical vapor deposition (A6), molecular beam epitaxy (A7), electroplating (A8), and chemical / mechanical surface treatment (A9). Alternatively, these multiple influencing factors can be multiple obstacle factors such as airborne dust particles (E1), temperature and humidity (E3), pressure (E4), harmful gases generated during manufacturing (E5), the airtightness of the entire cleanroom (E6), static electricity (E7), and electromagnetic interference (E8). For example, these multiple influencing factors can be photolithography (A1), ion implantation (A2), etching (A3), heat treatment (A4), chemical vapor deposition (A5), physical vapor deposition (A6), molecular beam epitaxy (A7), electroplating (A8) and chemical / mechanical surface treatment (A9), airborne dust particles (E1), temperature and humidity (E3), pressure (E4), harmful gases generated during the manufacturing process (E5), the airtightness of the entire cleanroom (E6), static electricity (E7), electromagnetic interference (E8), etc.

[0081] In this case, after the computer equipment obtains the weight of each of the multiple influencing factors, it can determine which of the multiple influencing factors has a greater impact on the quality of the product during the manufacturing process, and which of the multiple influencing factors has a smaller impact on the quality of the product during the manufacturing process.

[0082] Optionally, the computer device can obtain the weight of each of the multiple influencing factors using the decision laboratory method. Specifically, the operation in step 101 can be implemented through the following steps (1)-(5).

[0083] (1) Computer equipment obtains the directed graph between the multiple influencing factors.

[0084] This directed graph is used to represent the degree of influence among multiple influencing factors, that is, the degree of influence of one influencing factor on other influencing factors. The directed graph includes multiple elements and multiple directed paths. The multiple elements are the multiple influencing factors. The directed paths in the graph include directed edges and their lengths. A directed edge is a line connecting one element to another in the graph that has an influence relationship. The length of the directed edge represents the degree of influence of the starting point to the ending point of the directed edge, that is, the degree of influence of one element on another in the directed graph, or the degree of influence of one influencing factor on another among the multiple influencing factors.

[0085] For example, these multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during the manufacturing process (E5), the airtightness of the entire cleanroom (E6), and static electricity (E7). Figure 2 For a directed graph showing the relationships between these multiple influencing factors, see [link to graph]. Figure 2 The directed graph contains 6 elements 201. The line connecting any two elements 201 is a directed edge, and the length of the directed edge represents the degree of influence of one element on the other. For example, if the length of the directed edge between element 201(A1) and element 201(A2) is 2, that is, if the length of the directed edge starting from element 201(A1) and ending at element 201(A2) is 2, then the degree of influence of element 201(A1) on element 201(A2) is 2.

[0086] Optionally, for the i-th and j-th impact factors among the multiple impact factors, the computer device can determine the average of the scores of the i-th impact factor on the j-th impact factor set in advance by each of the h technical personnel as the degree of influence of the i-th impact factor on the j-th impact factor, where i and j are both positive integers.

[0087] The score of the i-th impact factor on the j-th impact factor can be pre-set by each of the h technical personnel. For example, the score is a positive integer in the range [0, 4), where "0" indicates that one impact factor has no influence on another, "1" indicates that one impact factor has a slight influence on another, "2" indicates that one impact factor has a moderate influence on another, and "3" indicates that one impact factor has a high influence on another.

[0088] For example, if h is 3, and the multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), and heat treatment (A4), and the scores for the first influencing factor (A1) and the third influencing factor (A3) set in advance by these three technicians are "2", "1", and "3" respectively, then the computer equipment determines the degree of influence of the first influencing factor (A1) on the third influencing factor (A3) as follows:

[0089] (2) The computer equipment establishes the influence relationship matrix of the multiple influencing factors based on the directed graph.

[0090] The influence relationship matrix is ​​used to represent the influence relationship between the multiple influence factors. The influence relationship matrix includes the multiple influence factors and the degree of influence between each influence factor.

[0091] For example, these multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during manufacturing (E5), the overall airtightness of the cleanroom (E6), and static electricity (E7). A directed graph of these multiple influencing factors is shown below. Figure 2 As shown, the computer equipment establishes the influence relationship matrix of these multiple influence factors based on the directed graph of these multiple influence factors, as shown below:

[0092]

[0093] In this influence matrix, the values ​​represent the degree of influence of one influence factor on another. The value in each row of the matrix represents the degree of influence of the influence factor indicated in that row on each of the multiple columns (i.e., each of the multiple influence factors). The value in each column represents the degree to which the influence factor indicated in that column is influenced by each of the multiple rows (i.e., each of the multiple influence factors). For example, C in this influence matrix... 12 This is used to represent the degree of influence of influence factor A1 on influence factor A2, that is, the degree to which influence factor A2 is affected by influence factor A1. For example, C in this influence relationship matrix... 56 It is used to indicate the degree of influence of influence factor E6 on influence factor E7, that is, the degree to which influence factor E7 is affected by influence factor E6.

[0094] Specifically, the operation of the computer device to establish the influence relationship matrix of the multiple influencing factors based on the directed graph is similar to the operation of the computer device to establish a relationship matrix based on a directed graph in related technologies, and will not be described in detail in the embodiments of this application.

[0095] (3) The computer equipment determines the influence degree and the degree of influence of each of the multiple influencing factors based on the influence relationship matrix.

[0096] For any one of these multiple influencing factors, the influence degree of that factor is the total degree of influence that factor has on each of the other multiple influencing factors. The greater the influence degree of this factor, the greater its influence on the other influencing factors.

[0097] The degree of influence of this impact factor is the total influence of each of the multiple impact factors on this impact factor. The greater the degree of influence of this impact factor, the greater the influence of other impact factors on this impact factor, that is, the more susceptible this impact factor is to the influence of other impact factors.

[0098] In this situation, the computer equipment can determine which of the multiple influencing factors has a greater impact on the other influencing factors, and which influencing factor is more susceptible to the influence of the other influencing factors.

[0099] Specifically, step (3) can be performed as follows: the computer device normalizes the influence relationship matrix to obtain the normalized influence matrix; the computer device determines the comprehensive influence matrix based on the normalized influence matrix; and the computer device determines the influence degree and the degree of being influenced of each of the multiple influence factors based on the comprehensive influence matrix.

[0100] This normalization process is used to standardize the influence matrix; that is, the computer device standardizes each element in the influence matrix to a value between [0,1]. The normalized influence matrix is ​​the standardized influence matrix, where each element is a number between [0,1]. Each element in the normalized influence matrix follows the same data distribution; for example, each element in the normalized influence matrix follows a standard normal distribution.

[0101] The comprehensive impact matrix is ​​a matrix used to determine the impact index (influence degree and affected degree) of each of the multiple impact factors.

[0102] The normalization process of the influence relationship matrix by the computer equipment to obtain the standardized influence matrix can be as follows: the computer equipment sums up multiple elements in each row of the influence relationship matrix to obtain multiple first values; the computer equipment sums up multiple elements in each column of the influence relationship matrix to obtain multiple second values; the computer equipment determines the maximum value among the multiple first values ​​and multiple second values; the computer equipment divides each element in the influence relationship matrix by the maximum value to obtain the standardized influence matrix.

[0103] Normalization using the maximum value preserves the original data relationships between elements in the influence matrix, ensuring that the influence relationships between various influencing factors remain unchanged. In this case, using the normalized influence matrix to determine the comprehensive influence matrix results in a more accurate matrix, thereby improving the accuracy of the determined influence degree and affected degree of each influencing factor.

[0104] For example: the influence relationship matrix is The computer device sums up multiple elements in each row of the influence matrix, obtaining multiple first values: 4, 3, 8, 1, 3, 0. Then, the computer device sums up multiple elements in each column of the influence matrix, obtaining multiple second values: 0, 5, 2, 4, 2, 6. The computer device then determines the maximum value to be 8. Finally, the computer device divides each element in the influence matrix by 8 to obtain the normative influence matrix.

[0105] The operation of determining the comprehensive influence matrix based on the standard influence matrix by computer equipment can be as follows: the computer equipment subtracts the standard influence matrix from the identity matrix to obtain the target matrix; the computer equipment multiplies the standard influence matrix with the inverse of the target matrix to obtain the comprehensive influence matrix.

[0106] In this case, the comprehensive influence matrix not only includes the direct influence between each influencing factor, but also the indirect influence of each influencing factor on other influencing factors. Thus, the computer equipment can determine the degree of influence and the degree of being influenced of each influencing factor based on the comprehensive influence matrix.

[0107] The computer equipment determines the influence and affected degree of each of the multiple influence factors based on the comprehensive influence matrix as follows: For any one of the multiple influence factors, the computer equipment sums up the multiple elements in the row corresponding to this influence factor in the comprehensive influence matrix to obtain the influence degree of this influence factor; the computer equipment sums up the multiple elements in the column corresponding to this influence factor in the comprehensive influence matrix to obtain the affected degree of this influence factor.

[0108] For example, the multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during the manufacturing process (E5), the airtightness of the entire cleanroom (E6), and static electricity (E7). The comprehensive influence matrix is ​​as follows: The computer equipment summed the elements in the first row of the comprehensive influence matrix to obtain the influence degree of the first influence factor (A1) as 0.97. Summing the elements in the first column of the matrix yielded an influence degree of 0 for the first influence factor (A1). Summing the elements in the second row of the matrix yielded the influence degree of the second influence factor (A2) as 0.77. Summing the elements in the second column yielded an influence degree of 0.49 for the second influence factor (A2). This process continued until the influence degree of the third influence factor (A3) was 0.77 and its influence degree was 0.62. The influence degree of the fourth influence factor (E5) was 0.13 and its influence degree was 0.45. The influence degree of the fifth influence factor (E6) was 0.11 and its influence degree was 0.53. The influence degree of the sixth influence factor (E7) was 0 and its influence degree was 0.66.

[0109] (4) For any one of the multiple influencing factors, the computer device adds the influence degree of the influencing factor to the influence degree of the influencing factor to obtain the centrality of the influencing factor.

[0110] The centrality of an impact factor indicates the magnitude of its influence on the quality of an item. A higher centrality indicates a greater influence of the factor on the quality of the item, meaning that the factor is more likely to affect the quality of the item.

[0111] In this situation, the computer equipment can determine the extent to which each of the multiple influencing factors affects the quality of the product during the manufacturing process. It can also determine which of the multiple influencing factors are more likely to affect the quality of the product during the manufacturing process, and which have a smaller impact on the quality of the product during the manufacturing process.

[0112] Optionally, after obtaining the centrality of each of the multiple impact factors, the computer device can also determine the causal degree of each of the multiple impact factors.

[0113] The causality degree of an impact factor indicates its influence on other impact factors. The causality degree of an impact factor is the difference between its influence degree and its degree of being influenced. A causality degree greater than 0 indicates that this impact factor has a greater influence on other impact factors among multiple impact factors, meaning this impact factor is a causal factor. A causality degree less than 0 indicates that this impact factor is more influenced by other impact factors among multiple impact factors, meaning this impact factor is an outcome factor. Therefore, the causality degree of an impact factor can indicate whether it is a causal factor or an outcome factor.

[0114] Thus, based on the comprehensive influence matrix, the computer equipment can obtain the influence index (including influence degree, degree of influence, centrality and causation degree) of each of the multiple influence factors.

[0115] Optionally, the computer device can also generate an influence-influence diagram and a centrality-causation diagram for each of the multiple influencing factors based on the influence degree, influence level, centrality, and causation degree of each influencing factor.

[0116] For example, the multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during the manufacturing process (E5), the airtightness of the entire cleanroom (E6), and static electricity (E7). The influence, affectedness, centrality, and causation of each of these multiple influencing factors are shown in Table 1 below.

[0117] Table 1

[0118] Impact Influence Centrality Cause degree A1 0.97 0 0.97 0.97 A2 0.77 0.49 1.26 0.28 A3 0.77 0.62 1.39 0.15 E5 0.13 0.45 0.58 -0.32 E6 0.11 0.53 0.64 -0.42 E7 0 0.66 0.66 -0.66

[0119] The embodiments of this application are merely illustrative examples of the influence degree, affected degree, centrality degree and causal degree of each of the multiple influencing factors in Table 1 above. Table 1 above does not constitute a limitation on the embodiments of this application.

[0120] Figure 3 For an impact-affected ratio diagram of these multiple influencing factors, see [link / reference]. Figure 3 , Figure 3This includes coordinate system 301, four quadrants 302, and six influence factors 303. The x-axis of coordinate system 301 represents the degree of influence, and the y-axis represents the degree of being influenced. Quadrant 302 includes two influence factors 303 (A2, A3), indicating that both their influence and degree of being influenced are relatively large. In other words, the second and third influence factors 303 (A2 and A3) have a significant impact on the other six influence factors 303, and these two influence factors 303 (A2 and A3) are also significantly influenced by the other six influence factors 303. Quadrant 302 includes two impact factors 303 (E6 and E7), indicating that these two impact factors 303 (E6 and E7) have relatively small influence but are significantly influenced by other impact factors 303 among the six impact factors 303. Quadrant 302 includes one impact factor 303 (E5), indicating that this impact factor 303 (E5) has both relatively small influence and is relatively easily influenced by other impact factors 303 among the six impact factors 303. The inclusion of one impact factor 303 (A1) in quadrant 302 indicates that this impact factor 303 (A1) has a relatively large influence but is less affected by other impact factors 303 among the six impact factors 303.

[0121] Figure 4 For a centrality-causality plot of these multiple influencing factors, see [link / reference]. Figure 4 , Figure 4This includes coordinate system 401, four quadrants 402, and six influence factors 403. The x-axis of coordinate system 401 represents centrality, and the y-axis represents causality. Quadrant 402 contains three influence factors 403 (A1, A2, A3), indicating that these three influence factors 403 (A1, A2, A3) have relatively high centrality and causality. This means that these three influence factors 403 (A1, A2, A3) are more likely to affect the quality of goods during the manufacturing process, and they are considered causal factors. The third quadrant includes three impact factors 403 (E5, E6, E7), indicating that the centrality and causality of these three impact factors 403 (E5, E6, E7) are relatively small. In other words, these three impact factors 403 (E5, E6, E7) have a relatively small impact on the quality of the product during the manufacturing process, and these three impact factors 403 (E5, E6, E7) are outcome factors.

[0122] In this case, by drawing the influence-affected degree diagram and the centrality-causation degree diagram of these multiple influencing factors, technicians can be provided with a more intuitive understanding of the impact of these multiple influencing factors on the product manufacturing process, thereby enabling technicians to quickly analyze the impact of these multiple influencing factors on the product quality.

[0123] (5) The computer equipment determines the weight of each of the multiple impact factors based on the centrality of each of the multiple impact factors.

[0124] The centrality of one of the multiple influencing factors is used to represent the magnitude of the influence of an influencing factor on the quality of an item during the manufacturing process. In this case, by using the centrality of each of the multiple influencing factors, the weight of each of the multiple influencing factors can be determined more accurately, which means that the magnitude of the influence of each of the multiple influencing factors on the quality of the item can be determined more accurately.

[0125] Specifically, step (5) can be performed as follows: the computer device sums up the centrality of each of the multiple impact factors to obtain the target value; the computer device divides the centrality of an impact factor by the target value to obtain the weight of that impact factor.

[0126] For example, the multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during the manufacturing process (E5), the airtightness of the entire cleanroom (E6), and static electricity (E7). The centrality of each of these influencing factors is shown in Table 1 above. The computer equipment then sums the centralities of each of these influencing factors to obtain a target value of 5.5. Next, the centrality of the first influencing factor (A1), 0.97, is divided by the target value of 5.5, resulting in a weight of 0.177 for the first influencing factor (A1). The centrality of the second influencing factor (A2), 1.26, is divided by the target value of 5.5, resulting in a weight of 0.229 for the second influencing factor (A2). The centrality of the third influencing factor (A3), 1.39, is divided by the target value of 5.5, resulting in a weight of 0.253 for the third influencing factor (A3). Dividing the centerness of the fourth impact factor (E5) from 0.58 by the target value of 5.5 yields a weight of 0.105 for E5. Dividing the centerness of the fifth impact factor (E6) from 0.64 by the target value of 5.5 yields a weight of 0.116 for E6. Dividing the centerness of the sixth impact factor (E7) from 0.66 by the target value of 5.5 yields a weight of 0.12 for E7.

[0127] Step 102: The computer equipment acquires the target data corresponding to each of the multiple influencing factors during the manufacturing process of the item.

[0128] The target data refers to the data collected by sensors in the process flow corresponding to each of the multiple influencing factors in the article manufacturing process. Sensors are present in each process flow of the article manufacturing process to collect data generated by the process equipment in each process flow during article manufacturing. For example, the target data could be data collected by sensors during the etching (A3) process flow, including step time, process chamber pressure, helium back purge, helium pressure back purge, pendulum position, positive pressure, upper electrode loading power, lower electrode loading power, and upper electrode reflection power.

[0129] Optionally, the target data can be the data corresponding to each of the multiple influencing factors directly obtained by computer equipment during the manufacturing process of an item, or it can be the data corresponding to each of the multiple influencing factors obtained by computer equipment during the manufacturing process of each of the multiple items.

[0130] For example, when manufacturing 10 items, a computer device can obtain data corresponding to each of the multiple influencing factors during the manufacturing process of each of the 10 items. Each of the multiple influencing factors corresponds to data for the 10 items. For any one of the multiple influencing factors, the computer device determines the average value of the 10 data points corresponding to that influencing factor. The average value of the 10 data points corresponding to that influencing factor is the target data corresponding to that influencing factor.

[0131] Optionally, after the computer equipment acquires the target data corresponding to each of the multiple influencing factors during the article manufacturing process, it can compare the target data with preset data to determine whether there is a deviation in the target data.

[0132] The preset data can be set in advance. The preset data is the data generated by the process equipment in the process flow to which each of the multiple influencing factors belongs when it is working normally.

[0133] When serious malfunctions occur in the process equipment, the data generated during the manufacturing process will deviate significantly from the data generated during normal operation. As a result, when the target data deviates, the quality of the product is very likely to be substandard. In this case, the target data is meaningless and cannot be processed in subsequent steps.

[0134] Specifically, if the difference between the target data corresponding to at least one of the multiple influencing factors and the preset data corresponding to that influencing factor is greater than a preset threshold, the computer equipment determines that the target data has deviated; if the difference between the target data corresponding to each of the multiple influencing factors and the preset data is less than or equal to the preset threshold, the computer equipment determines that the target data has not deviated.

[0135] The preset threshold can be set in advance, and the preset threshold can be set relatively large. If the difference between the target data corresponding to an impact factor and the preset data corresponding to that impact factor is greater than the preset threshold, it indicates that the difference between the target data and the preset data corresponding to that impact factor is large. In other words, the target data of that impact factor differs significantly from the data generated by the process equipment in the process flow to which that impact factor belongs when operating normally. Therefore, the computer equipment determines that the target data is biased, meaning that the process equipment in the process flow to which this impact factor belongs has a serious abnormality. If the difference between the target data and the preset data is less than or equal to the preset threshold, it indicates that the target data is within the normal range. Therefore, the computer equipment determines that the target data is not biased, meaning that the process equipment in each process flow is operating normally.

[0136] Furthermore, if the computer equipment determines that the target data is deviated, subsequent steps cannot be continued, and the current manufacturing process must be stopped immediately. After replacing or repairing the process equipment, the target data corresponding to each of the multiple influencing factors can be re-acquired during the product manufacturing process. In this way, it can be ensured that the target data acquired by the computer equipment is normal data, that is, data with reference value.

[0137] Optionally, after acquiring the target data corresponding to each of the multiple influencing factors during the article manufacturing process, the computer equipment can also process the target data to extract data related to article quality.

[0138] Since computer equipment acquires target data by directly obtaining data collected by sensors in the process flow, and this data is related to the corresponding process flow, some data in the target data is related to the quality of the product, while some data is not related to the quality of the product. Therefore, the computer equipment needs to process the target data to extract the data related to the quality of the product.

[0139] In this case, the target data processed by computer equipment is data related to the quality of the item, thus ensuring the accuracy of the target data.

[0140] For example, the target data corresponding to an impact factor etching (A3) acquired by the computer equipment includes step time, process chamber pressure, back purge of helium gas flow, back purge of helium pressure, pendulum position, positive pressure, upper electrode loading power, lower electrode loading power, upper electrode reflection power, lower electrode reflection power, DC bias, upper RF matching unit 1, upper RF matching unit 2, lower RF matching unit 1, lower RF matching unit 2, gas flow rate, refrigerator temperature, and heating temperatures of the upper middle and bottom sections. The step time, pendulum position, and DC bias are unrelated to the quality of the item, so they are not meaningful. The computer equipment can remove these three data from the target data, leaving only the data related to the quality of the item. That is, the target data after the processing corresponding to the influence factor etching (A3) is the process chamber pressure, helium back purge, helium pressure back purge, positive pressure, upper electrode loading power, lower electrode loading power, upper electrode reflection power, lower electrode reflection power, upper RF matching 1, upper RF matching 2, lower RF matching 1, lower RF matching 2, gas flow rate, refrigerator temperature, and upper and lower heating temperatures.

[0141] Step 103: The computer equipment predicts the yield of the finished product based on the target data.

[0142] The product yield rate is the proportion of qualified products in a predicted batch of products. It indicates the quantity of qualified products in a batch. A higher product yield rate means a greater proportion of qualified products in the predicted batch, and consequently, a larger quantity of qualified products in the batch.

[0143] The primary purpose of the method provided in this application is to promptly identify and resolve problems in the manufacturing process of articles, thereby ensuring the quality of the articles. However, if the yield rate of a batch of articles is directly calculated, and process problems are only checked when the yield rate is low, the process would be sluggish, wasting manufacturing materials and increasing manufacturing costs. Therefore, the method provided in this application predicts the status of the process equipment in the current manufacturing process to obtain the yield rate of the current batch of articles, thereby allowing for early detection of process problems.

[0144] In this situation, computer equipment can predict the yield rate of goods and promptly identify problems in the manufacturing process, thus avoiding waste of manufacturing materials and ensuring the quality of goods.

[0145] Specifically, step 103 can be performed as follows: the computer device inputs the target data into the prediction model to output the number of qualified items through the prediction model; the computer device divides the number of qualified items by a preset quantity to obtain the item yield.

[0146] The qualified quantity refers to the number of qualified items out of a predicted preset quantity. The preset quantity is the total quantity of items in a batch, and it can be set in advance. For example, the preset quantity can be 100.

[0147] This prediction model is used to predict the number of qualified items out of a preset number of items. For example, the prediction model can be a sparse Bayesian width learning model.

[0148] For example, Figure 5This is a schematic diagram of the prediction model, which includes an input layer 501, a feature layer 502, an enhancement layer 503, and an output layer 504. The input layer 501 receives the target data; the feature layer 502 performs k linear transformations on the target data to obtain k first feature matrices, and merges these k first feature matrices to obtain a second feature matrix; the enhancement layer performs m nonlinear transformations on the second feature matrix to obtain m third feature matrices, and merges these m third feature matrices to obtain a fourth feature matrix; the output layer merges the second and fourth feature matrices to obtain a fifth feature matrix of size n×1, obtains the parameter matrix of size 1×n for the prediction model, and multiplies the fifth feature matrix by the parameter matrix to obtain the number of qualified items; where k, m, and n are all positive integers.

[0149] Linear transformation is a linear activation operation. For example, the feature layer can be linearly activated on the target data using the ReLU (Rectified LinearUnit) activation function.

[0150] Nonlinear transformation is a nonlinear activation operation. For example, the enhancement layer can be nonlinearly activated by the Sigmoid (S-shaped growth curve) activation function on the second feature matrix.

[0151] The parameter matrix of this prediction model consists of the model parameters of the prediction model itself, and it represents the optimal model parameters for predicting the number of qualified items. For example, the parameter matrix of this prediction model represents the optimal weights for predicting the number of qualified items.

[0152] In this case, the prediction model contains both linear and nonlinear features, which can improve the model's feature representation ability and thus improve the accuracy of the prediction model in predicting the number of qualified items.

[0153] Optionally, the feature layer 502 can merge the k first feature matrices by accumulating the k first feature matrices, or by horizontally or vertically concatenating the k first feature matrices.

[0154] The following is an illustrative explanation of how the prediction model works.

[0155] For example, input layer 501 receives target data and processes it into a feature matrix X. Then, feature matrix X is input into feature layer 502. Feature layer 502 performs k linear transformations on feature matrix X to obtain k first feature matrices. The i-th linear transformation in the k linear transformations of feature layer 502 on feature matrix X can be implemented using the following formula:

[0156]

[0157] Among them, P i Let i be the first characteristic matrix. For linear activation operations, X is the feature matrix (i.e., the target data), and W... zi β is the weight matrix randomly generated when feature layer 502 performs the i-th linear transformation on the feature matrix X. zi The i-th bias matrix is ​​randomly generated when feature layer 502 performs the i-th linear transformation on the feature matrix X.

[0158] Next, feature layer 502 merges the k first feature matrices to obtain the second feature matrix. The operation is as follows: P k = [P1, P2, ..., P k ], where P k Let P1 be the first first characteristic matrix among k first characteristic matrices, and P2 be the second first characteristic matrix among k first characteristic matrices. k It is the kth first characteristic matrix among the k first characteristic matrices.

[0159] Next, feature layer 502 inputs the second feature matrix into enhancement layer 503. Upon receiving the second feature matrix, enhancement layer 503 performs m nonlinear transformations on it to obtain m third feature matrices. The j-th nonlinear transformation in the m nonlinear transformations performed by enhancement layer 503 on the second feature matrix can be implemented using the following formula:

[0160] B j =τ(P k ·W ej +β ej )

[0161] Among them, B j Let W be the j-th third characteristic matrix, τ(·) be the nonlinear activation operation, and W be the third characteristic matrix. ej β is the weight matrix randomly generated when enhancement layer 503 performs the j-th nonlinear transformation on the second feature matrix. ej The bias matrix is ​​randomly generated when the second feature matrix is ​​subjected to the j-th nonlinear transformation in the enhancement layer 503.

[0162] Next, enhancement layer 503 merges the m third feature matrices to obtain the fourth feature matrix using the following operation: B m = [B1, B2, ..., B m ], where B m B1 is the fourth characteristic matrix, B2 is the first third characteristic matrix among m third characteristic matrices, and B is the second third characteristic matrix among m third characteristic matrices. mLet be the m-th third characteristic matrix among m third characteristic matrices.

[0163] Then, feature layer 502 inputs the second feature matrix into output layer 504, enhancement layer 503 inputs the fourth feature matrix into output layer 504, output layer 504 merges the second feature matrix and the fourth feature matrix into a fifth feature matrix A of size n×1, obtains the parameter matrix O of size 1×n of the prediction model, and then multiplies the fifth feature matrix A with the parameter matrix O to obtain the qualified quantity of the item: y = A·O. Then the output layer outputs the qualified quantity y of the item, which is the qualified quantity y of the item output by the prediction model.

[0164] It is worth noting that before the computer equipment inputs the target data into the prediction model to output the qualified quantity of items, the prediction model needs to be trained.

[0165] Specifically, the computer device can acquire Q training samples and use these Q training samples to train the neural network model to obtain the prediction model.

[0166] The Q training samples can be pre-set. Each of the Q training samples includes sample data and a sample label. The sample data is the data corresponding to each influencing factor in the manufacturing process of the sample item, and the sample label is the number of qualified items in the batch to which the sample item belongs. That is, the input data in each of the Q training samples is the data corresponding to each influencing factor, and the sample label is the number of qualified items corresponding to the input data.

[0167] This neural network model can include multiple network layers, including an input layer, multiple hidden layers, and an output layer. The input layer receives the input data; the output layer outputs the processed data; and the hidden layers, located between the input and output layers, process the data and are invisible to the outside world. This neural network model can be used with... Figure 5 The prediction models shown have the same structure.

[0168] When a computer device trains a neural network model using Q training samples, for each of the Q training samples, the input data from that training sample is input into the neural network model to obtain output data. A loss function is used to determine the loss value between the output data and the sample labels in that training sample. The parameters in the neural network model are then adjusted based on this loss value. After adjusting the parameters of the neural network model based on each of the Q training samples, the adjusted neural network model is the prediction model.

[0169] Optionally, the computer device can iteratively optimize the parameters and hyperparameters of the neural network model based on Bayesian theory to obtain the optimal parameters and hyperparameters of the neural network model. The hyperparameter is the learning rate of the neural network model, and a suitable learning rate needs to be obtained during the training process. The optimal parameters of the neural network model obtained by the computer device then constitute the parameter matrix of the prediction model.

[0170] Optionally, the posterior probability of the parameters of the neural network model is p(ε|λ, α), where α is a hyperparameter of the neural network model. The optimal parameters of the neural network model (i.e., the parameter matrix of the prediction model) can be obtained by calculating the maximum value of p(ε|λ, α). The specific steps are as follows:

[0171] First, calculate the logarithm of p(ε|λ, α). Since p(ε|λ, α) ≈ p(λ|ε)p(ε|α), the logarithm of p(ε|λ, α) can be expressed by the following formula:

[0172]

[0173] Where p(λ|ε) is the product yield rate, λ is the number of qualified samples in the Q training samples, and λ a Let ε be the number of qualified samples corresponding to the a-th training sample among the Q training samples. Let ε be the parameters of the neural network model, and p(ε|α) be the probability that the parameters of the neural network model are ε under the hyperparameter α. ζ a Let ζ be the probability that the item in the a-th training sample out of the Q training samples is of acceptable quality. a =σ{f(Γ) a ;ε)},σ{·} represents probability,Γ a Let f(Γ) be the sample data of the a-th training sample in the Q training samples. a ;ε) is the likelihood function between the sample data of the a-th training sample in the Q training samples and the parameters of the neural network model, 1-ζ a Let ε be the probability that the item in the a-th training sample out of the Q training samples is substandard. T Let A be the transpose of the parameters of this neural network model, A be the fifth feature matrix, and ζ be a constant. Where α a Let be the value of the hyperparameter corresponding to the a-th training sample among the Q training samples.

[0174] Secondly, the computer device calculates the first-order partial derivative of the parameter ε in the logarithmic formula of p(ε|λ, α) above, and obtains... in, It is the first-order partial derivative of the parameter ε in the logarithmic formula. For representing the first-order partial derivative operation on the parameter ε, Ψ is a positive definite matrix, which is obtained after performing the first-order partial derivative operation on the parameter ε. ζ is the transpose of the matrix composed of the qualified probabilities of the items in each of the Q training samples. Here, ζ = [ζ1, ζ2, …, ζ Q T .

[0175] After that, the computer device calculates the second-order partial derivative of the parameter ε in the logarithmic formula of p(ε|λ, α) above, and obtains where is the second-order partial derivative of the parameter ε in the logarithmic formula, Ψ T is the transpose of this positive definite matrix, Υ is a diagonal matrix, Υ = diag(u1, u2, …, υ a , … υ Q ), where diag() is used to construct a diagonal matrix, and υ a is the product of the qualified probability and the unqualified probability of the item in the a-th training sample, that is, u a = ζ a (1 - ζ a ).

[0176] After the computer device calculates the first-order partial derivative and the second-order partial derivative of the parameter ε in the logarithmic formula of p(ε|λ, α) above, the parameter ε of this neural network can be iteratively optimized multiple times through the following formula.

[0177]

[0178] where ε new is the parameter value of this neural network model after the current iterative optimization, ε old is the parameter value of this neural network model before the current iterative optimization, is 's inverse matrix. In this way, through multiple iterative optimizations, the optimal parameter value of this neural network can be obtained, that is, the parameter matrix of this prediction model can be obtained.

[0179] After that, assuming that the parameter ε of this neural network model follows a Gaussian distribution, the two statistical quantities commonly used to adjust the parameters of the neural network model in the Gaussian distribution are the mean and the covariance. Then, the computer device can determine the statistical quantities of the Gaussian distribution through the following formula after each iterative optimization.

[0180] ∑ ε =(Ψ T ΥΨ + A) -1

[0181]

[0182] where, ∑​ε Let μ be the covariance of the parameters of the neural network model, and let μ be a matrix. ε Let be the mean of the parameters of this neural network. The number of qualified items to be used as a transition in the process of finding a suitable hyperparameter α.

[0183] After obtaining the mean and covariance of the parameters of the neural network, the computer device can use the mean and covariance to iteratively optimize the hyperparameter α.

[0184] When the hyperparameter α is optimal, the log-marginal likelihood expansion of the probability p(λ|α) that the number of qualified items is λ is:

[0185]

[0186] Where J = Υ + ΨA -1 Ψ T A -1 Let be the inverse of the fifth characteristic matrix. Assume... If the derivative of the hyperparameter α is 0, then the hyperparameter α of the neural network model can be iteratively optimized using the following formula:

[0187]

[0188] Where e represents the number of iterations, that is, the e-th iteration optimizes the hyperparameter α. Let α be the hyperparameter of the neural network model after the e-th iteration optimization. e Let (∑) be the hyperparameters of the neural network model before the e-th iteration optimization. ε ) e Let (μ) be the e-th diagonal element in the covariance. ε ) e Let represent the mean obtained after the e-th iteration of optimization, where a and b are constants. Through multiple iterations of optimization, the optimal hyperparameters of the neural network can be obtained.

[0189] In this case, the computer device can find the optimal parameters and hyperparameters of the neural network model through multiple iterations of optimization. Finding the optimal parameters and hyperparameter values ​​of the neural network means that the neural network model has been trained into the prediction model.

[0190] It is worth noting that after predicting the yield of an item, the computer equipment must also determine whether the predicted yield is abnormal.

[0191] Specifically, the operation by which the computer equipment determines whether the predicted yield rate of an item is abnormal can be as follows: The computer equipment acquires a historical dataset, which includes historical inspection data for each item in each batch of multiple batches of items that have been manufactured and whose actual yield rate is normal, as well as the actual number of qualified items in each batch; the computer equipment determines the average value of the historical inspection data of the multiple batches of items and the average value of the actual number of qualified items in the multiple batches of items; the computer equipment determines the standard yield rate based on the average value of the historical inspection data of the multiple batches of items and the average value of the actual number of qualified items in the multiple batches of items; if the yield rate of the item is less than the standard yield rate, the computer equipment determines that the yield rate of the item is abnormal; if the yield rate of the item is greater than or equal to the standard yield rate, the computer equipment determines that the yield rate of the item is normal.

[0192] Historical datasets are collections of data related to the manufacturing process of completed items. Optionally, the historical dataset may also include multiple actual model parameters, which are the optimal model parameters generated by the predictive model when predicting the number of qualified items in each batch of items across multiple batches. In this case, the computer device can also determine the average value of these multiple actual model parameters.

[0193] Historical testing data consists of data corresponding to each of the multiple influencing factors obtained during the manufacturing process of each item.

[0194] The actual yield rate is the proportion of qualified items out of a batch of manufactured goods. A normal actual yield rate indicates that a large proportion of qualified items are in a batch of manufactured goods. The actual qualified quantity is the number of qualified items in a batch of manufactured goods.

[0195] The standard yield rate is used to determine whether the yield rate of an item is abnormal. If the yield rate is greater than or equal to the standard yield rate, it means that the yield rate is not lower than this standard, and the yield rate is considered normal. If the yield rate is less than the standard yield rate, it means that the yield rate has not met this standard, and the yield rate is considered abnormal.

[0196] In this case, the data in the historical dataset is the historical test data of each item in each batch of multiple batches of items when the actual yield is normal. This means that the data in the historical dataset is of reference value for determining the standard yield. Therefore, the standard yield determined by the computer equipment based on the data in the historical dataset is relatively accurate.

[0197] The operation of determining the standard yield rate by computer equipment based on the average of historical test data of multiple batches of items and the average of the actual qualified quantity of multiple batches of items can be as follows: The computer equipment determines the standard yield rate by the following formula based on the average of historical test data of multiple batches of items, the average of the actual qualified quantity of multiple batches of items, and the average of multiple actual model parameters.

[0198] p(r|ω)=σ{f(ψ;ω)} r [1-σ{f(ψ;ω)}] 1-r

[0199] Where p(r|ω) is the standard yield, r is the average of the actual qualified quantity of the multiple batches of items, ω is the average of the multiple actual model parameters, σ{·} represents the probability, σ{f(ψ;ω)} is the probability that each batch of items in the multiple batches is qualified, f(ψ;ω) is the function between the average of the historical test data and the average of the multiple actual model parameters, ψ is the average of the historical test data of the multiple batches of items, and 1-σ(f(ψ;ω)} is the probability that each batch of items in the multiple batches is unqualified.

[0200] The above explains whether the finished product rate of the item is abnormal as determined by the computer equipment. When the computer equipment determines that the finished product rate of the item is abnormal, it means that the number of qualified items among the preset quantity of items may be relatively small, which means that there may be a problem with the process equipment in the current process flow. At this time, the following step 104 should be performed to check the process flow.

[0201] Step 104: If the predicted yield of the product is abnormal, the computer equipment checks the process flow to which each of the multiple influencing factors belongs in descending order of their weights until the problematic process flow is identified.

[0202] A manufacturing process refers to the steps involved in manufacturing an item. For example, if the item is a chip, the manufacturing processes include photolithography, ion implantation, etching, heat treatment, chemical vapor deposition, physical vapor deposition, molecular beam epitaxy, electroplating, and chemical / mechanical surface treatment. A manufacturing process includes both the equipment used to manufacture the item and the execution environment. The equipment is used to manufacture the item, and the execution environment is the working environment of the equipment within that process.

[0203] These multiple influencing factors correspond to multiple technological processes in the manufacturing of an item; that is, one influencing factor is related to one technological process in the manufacturing process. For example, the influencing factor lithography (A1) refers to the lithography process in the manufacturing process. In this case, the technological process to which the influencing factor lithography (A1) belongs is lithography. As another example, the influencing factors temperature and humidity (E3) refer to the temperature and humidity in the execution environment of the lithography process. Therefore, the technological process to which the influencing factor temperature and humidity (E3) belongs is lithography.

[0204] The higher the weight of an influencing factor, the greater the impact of the process to which this influencing factor belongs on the quality of the product. In the case of abnormal product yield, the probability that the abnormal product yield is caused by the process to which this influencing factor belongs is higher. In other words, the probability that there is a problem with the process to which this influencing factor belongs is higher when the product yield is abnormal.

[0205] If the predicted yield of the product is low, it indicates a potential problem with the equipment in the current process flow. In this case, the processes associated with each of the multiple influencing factors can be checked in descending order of their weight. Prioritizing the processes associated with factors with higher weights ensures that processes with the greatest impact on product quality are checked first—that is, processes most likely to cause quality abnormalities. This process checks processes in descending order of probability of problem until a problematic process is identified. This allows for faster identification of problematic processes, saving manpower and time, and ultimately improving process inspection efficiency.

[0206] For example, the multiple influencing factors are photolithography (A1), ion implantation (A2), etching (A3), harmful gases generated during the manufacturing process (E5), airtightness of the entire cleanroom (E6), and static electricity (E7). The weight of each of these multiple influencing factors is shown in Table 2 below.

[0207] Table 2

[0208] Impact Factor A1 A2 A3 E5 E6 E7 Weight 0.177 0.229 0.253 0.105 0.116 0.12

[0209] The embodiments of this application are merely illustrative examples of the weights of the multiple influencing factors as shown in Table 2 above, and Table 2 above does not constitute a limitation on the embodiments of this application.

[0210] The computer equipment can check the process flow to which each influencing factor belongs during the manufacturing process according to the order of weight from high to low as shown in Table 2. Specifically, since the weight of the third influencing factor (A3) is greater than the weight of the other influencing factors, meaning that the third influencing factor (A3) has the greatest impact on the quality of the product among the multiple influencing factors, it is very likely that there is a problem in the process flow to which the third influencing factor (A3) belongs. Therefore, the computer equipment first checks the process flow (etching) to which the third influencing factor (A3) belongs, that is, it first checks whether there is a problem with the process equipment in the etching process. If there is no problem with the process equipment in the etching process, it continues to check the process flow to which the next influencing factor belongs. Since the weight of the second influencing factor (A2) is less than the weight of the third influencing factor (A3) but greater than the weight of the other influencing factors, the computer equipment then checks the process flow (ion implantation) to which the second influencing factor (A2) belongs, that is, it checks whether there is a problem with the process equipment in the ion implantation process. If there is a problem with the process equipment in the ion implantation process, it means that the problem affecting the quality of the product during the manufacturing process has been found, and the remaining process flows are no longer checked. In this way, the inspection begins with the process flow to which the factor with the greatest impact on the quality of the product belongs. That is, we can start by checking the process flow to which the factor with the greatest impact on the quality of the product belongs. This process flow is most likely to be the key reason for the low yield of the product. Once the problem is found, there is no need to continue to check the process flow to which the remaining factors belong. In this way, it is not necessary to check all the process flows, thus saving time.

[0211] Furthermore, after identifying problematic processes, technicians can repair the equipment involved to ensure timely detection and resolution of issues, thereby guaranteeing product quality.

[0212] In this embodiment, the computer device acquires the weight of each of a plurality of influencing factors, where each influencing factor is a factor that affects the quality of the article during the manufacturing process. The larger the weight of an influencing factor, the greater its impact on the quality of the article during manufacturing. During the manufacturing process, target data corresponding to each of the multiple influencing factors is acquired. Then, based on the target data, the yield rate is predicted; this predicted yield rate represents the proportion of qualified articles in the predicted batch. Subsequently, if the predicted yield rate is abnormal, the process flow corresponding to each influencing factor is checked in descending order of weight until a problematic process flow is identified. Therefore, prioritizing the inspection of process flows with higher-weighted influencing factors ensures that process flows with factors that have a greater impact on product quality are inspected first. In other words, process flows most likely to cause product quality abnormalities are inspected first, i.e., process flows most likely to have problems. This process checks each process flow in descending order of the probability of problems, and stops when a problematic process flow is found. In this way, problematic process flows can be found more quickly, thereby saving manpower, saving process inspection time, and thus improving process inspection efficiency.

[0213] Figure 6 This is a schematic diagram of a process inspection device provided in an embodiment of this application. The process inspection device can be implemented as part or all of a computer device by software, hardware, or a combination of both. This computer device can be described below. Figure 7 The computer equipment shown. See also Figure 6 The device includes: a first acquisition module 601, a second acquisition module 602, a prediction module 603, and an inspection module 604.

[0214] The first acquisition module 601 is used to acquire the weight of each of the multiple influencing factors, where each of the multiple influencing factors is a factor that will affect the quality of the item during the manufacturing process.

[0215] The second acquisition module 602 is used to acquire the target data corresponding to each of the multiple influencing factors during the article manufacturing process;

[0216] Prediction module 603 is used to predict the yield of finished products based on target data;

[0217] The inspection module 604 is used to check the process flow to which each of the multiple influencing factors belongs in descending order of the weight of each influencing factor when the predicted yield of the product is abnormal, until the problematic process flow is found.

[0218] Optionally, the first acquisition module 601 is used for:

[0219] Obtain the directed graph between the multiple influencing factors. The directed path in the graph includes the directed edge and the length of the directed edge. The length of the directed edge is the degree of influence of the starting point of the directed edge on the ending point.

[0220] Based on the directed graph, establish the influence relationship matrix of the multiple influencing factors;

[0221] Based on the influence relationship matrix, determine the influence degree and the degree of influence of each of the multiple influence factors;

[0222] For any one of these multiple influencing factors, the centrality of the influencing factor is obtained by adding its influence degree to its affected degree.

[0223] The weight of each of the multiple impact factors is determined based on the centrality of each impact factor.

[0224] Optionally, the first acquisition module 601 is used for:

[0225] The centrality of each of the multiple impact factors is summed to obtain the target value.

[0226] The weight of an impact factor is obtained by dividing its centrality by the target value.

[0227] Optionally, the prediction module 603 is used for:

[0228] The target data is input into the prediction model so that the prediction model can output the number of qualified items. The prediction model is used to predict the number of qualified items out of a preset number of items.

[0229] Divide the number of qualified items by the preset quantity to obtain the yield of the finished product.

[0230] Optionally, the prediction model includes an input layer, a feature layer, an enhancement layer, and an output layer. The feature layer performs k linear transformations on the target data to obtain k first feature matrices, and merges these k first feature matrices to obtain a second feature matrix. The enhancement layer performs m nonlinear transformations on the second feature matrix to obtain m third feature matrices, and merges these m third feature matrices to obtain a fourth feature matrix. The output layer merges the second and fourth feature matrices to obtain a fifth feature matrix of size n×1, obtains the parameter matrix of size 1×n of the prediction model, and multiplies the fifth feature matrix with the parameter matrix to obtain the number of qualified items. Here, k, m, and n are all positive integers.

[0231] Optionally, the device further includes:

[0232] The third acquisition module is used to acquire historical datasets, which include historical inspection data of each item in each batch of multiple batches of items that have been manufactured and have normal actual yield, and the actual number of qualified items in each batch of multiple batches of items. The historical inspection data is the data corresponding to each of the multiple influencing factors acquired during the manufacturing process of each item.

[0233] The first determining module is used to determine the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items;

[0234] The second determining module is used to determine the standard finished product rate based on the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items.

[0235] The third determination module is used to determine that the product yield is abnormal if the product yield is less than the standard yield.

[0236] Alternatively, the item is a chip.

[0237] In this embodiment, the weight of each of a plurality of influencing factors is obtained. Each influencing factor is a factor that affects the quality of the product during the manufacturing process. The larger the weight of an influencing factor, the greater its impact on the quality of the product during the manufacturing process among the plurality of influencing factors. During the product manufacturing process, target data corresponding to each of the plurality of influencing factors is obtained. Then, based on the target data, the product yield is predicted. This predicted product yield is the proportion of qualified products in the predicted batch. Subsequently, if the predicted product yield is abnormal, the process flow to which each influencing factor belongs is checked in descending order of the weight of each of the plurality of influencing factors until a problematic process flow is found. Therefore, prioritizing the inspection of process flows with higher-weighted influencing factors ensures that process flows with factors that have a greater impact on product quality are inspected first. In other words, process flows most likely to cause product quality abnormalities are inspected first, i.e., process flows most likely to have problems. This process checks each process flow in descending order of the probability of problems, and stops when a problematic process flow is found. In this way, problematic process flows can be found more quickly, thereby saving manpower, saving process inspection time, and thus improving process inspection efficiency.

[0238] It should be noted that the process inspection device provided in the above embodiments is only illustrated by the division of the above functional modules when inspecting the process flow in the manufacturing process of articles. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0239] The functional units and modules in the above embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of this application.

[0240] The process inspection apparatus and process inspection method embodiments provided in the above embodiments belong to the same concept. The specific working process and technical effects of the units and modules in the above embodiments can be found in the method embodiment section, and will not be repeated here.

[0241] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 7 As shown, the computer device 7 includes a processor 70, a memory 71, and a computer program 72 stored in the memory 71 and executable on the processor 70. When the processor 70 executes the computer program 72, it implements the steps in the process inspection method in the above embodiments.

[0242] Computer device 7 can be a general-purpose computer device or a special-purpose computer device. In specific implementations, computer device 7 can be a desktop computer, portable computer, handheld computer, tablet computer, or other terminal, or computer device 7 can be a network server. This application embodiment does not limit the type of computer device 7. Those skilled in the art will understand that... Figure 7 The computer device 7 is merely an example and does not constitute a limitation on the computer device 7. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0243] Processor 70 can be a Central Processing Unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0244] In some embodiments, memory 71 may be an internal storage unit of the computer device 7, such as a hard disk or RAM of the computer device 7. In other embodiments, memory 71 may be an external storage device of the computer device 7, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, etc., provided on the computer device 7. Furthermore, memory 71 may include both internal storage units and external storage devices of the computer device 7. Memory 71 is used to store the operating system, applications, boot loader, data, and other programs. Memory 71 may also be used to temporarily store data that has been output or will be output.

[0245] This application also provides a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0246] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps in the various method embodiments described above.

[0247] This application provides a computer program product that, when run on a computer, causes the computer to perform the steps described in the various method embodiments above.

[0248] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above method embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, ROM (Read-Only Memory), RAM (Random Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage devices. The computer-readable storage medium mentioned in this application can be a non-volatile storage medium; in other words, it can be a non-transient storage medium.

[0249] It should be understood that all or part of the steps of the above embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented in whole or in part as a computer program product. The computer program product includes one or more computer instructions. The computer instructions can be stored in the above-described computer-readable storage medium.

[0250] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0251] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0252] In the embodiments provided in this application, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0253] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0254] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A process inspection method characterized by, The method includes: Obtain the weight of each of the multiple influencing factors, where each of the multiple influencing factors is a factor that affects the quality of the item during the manufacturing process. During the manufacturing process of the item, acquire the target data corresponding to each of the multiple influencing factors; The target data is input into the prediction model so that the prediction model can output the number of qualified items. The prediction model is used to predict the number of qualified items out of a preset number of items. Divide the number of qualified items by the preset quantity to obtain the finished product rate of the items; If the predicted yield of the product is abnormal, the process flow to which each of the multiple influencing factors belongs is checked in descending order of the weight of each influencing factor until the problematic process flow is found. The prediction model comprises an input layer, a feature layer, an enhancement layer and an output layer, the feature layer is configured to perform k times of linear transformation on the target data respectively to obtain k first feature matrices, and perform merging on the k first feature matrices to obtain a second feature matrix; the enhancement layer is configured to perform m times of nonlinear transformation on the second feature matrix respectively to obtain m third feature matrices, and perform merging on the m third feature matrices to obtain a fourth feature matrix; and the output layer is configured to perform merging on the second feature matrix and the fourth feature matrix to obtain a fifth feature matrix with a size of n 1, obtain a parameter matrix with a size of 1 n of the prediction model, multiply the fifth feature matrix and the parameter matrix to obtain the number of qualified articles; wherein the k, the m and the n are positive integers.

2. The method as described in claim 1, characterized in that, The step of obtaining the weight of each impact factor among multiple impact factors includes: Obtain a directed graph between the multiple influencing factors. The directed path in the directed graph includes a directed edge and the length of the directed edge. The length of the directed edge is the degree of influence of the starting point of the directed edge on the ending point. Based on the directed graph, establish the influence relationship matrix of the multiple influencing factors; Based on the influence relationship matrix, determine the influence degree and the degree of influence of each of the plurality of influence factors; For any one of the multiple influencing factors, the influence degree of the influencing factor is added to the influence degree of the influencing factor to obtain the centrality of the influencing factor; The weight of each of the plurality of influence factors is determined based on the centrality of each influence factor.

3. The method as described in claim 2, characterized in that, The step of determining the weight of each of the plurality of impact factors based on the centrality of each of the plurality of impact factors includes: The centrality of each of the multiple impact factors is summed to obtain the target value. The weight of an influence factor is obtained by dividing the centrality of the influence factor by the target value.

4. The method as described in claim 1, characterized in that, Before checking the process flow to which each of the multiple influencing factors belongs, in order of decreasing weight among the predicted abnormal yield of the product, the process further includes: Obtain a historical dataset, which includes historical inspection data for each item in each batch of multiple batches of items that have been manufactured and have a normal actual yield, and the actual number of qualified items in each batch of multiple batches of items. The historical inspection data is the data corresponding to each of the multiple influencing factors obtained during the manufacturing process of each item. Determine the average value of the historical test data of the multiple batches of items and the average value of the actual qualified quantity of the multiple batches of items; The standard yield rate is determined based on the average of the historical test data of the multiple batches of items and the average of the actual qualified quantity of the multiple batches of items. If the finished product yield of the item is less than the standard finished product yield, then the finished product yield of the item is determined to be abnormal.

5. The method according to any one of claims 1-4, characterized in that, The item in question is a chip.

6. A process inspection device, characterized in that, The device includes: The first acquisition module is used to acquire the weight of each of the multiple influencing factors, wherein each of the multiple influencing factors is a factor that will affect the quality of the item during the manufacturing process. The second acquisition module is used to acquire target data corresponding to each of the multiple influencing factors during the article manufacturing process; The prediction module is used to input the target data into the prediction model so as to output the number of qualified items through the prediction model. The prediction model is used to predict the number of qualified items out of a preset number of items. The number of qualified items is divided by the preset number to obtain the product yield. The inspection module is used to check the process flow to which each of the multiple influencing factors belongs in the order of the weight of each influencing factor from high to low when the predicted yield of the product is abnormal, until the problematic process flow is found. The prediction model includes an input layer, a feature layer, an enhancement layer, and an output layer. The feature layer performs k linear transformations on the target data to obtain k first feature matrices, and then merges these k first feature matrices to obtain a second feature matrix. The enhancement layer performs m nonlinear transformations on the second feature matrix to obtain m third feature matrices, and then merges these m third feature matrices to obtain a fourth feature matrix. The output layer merges the second feature matrix and the fourth feature matrix to obtain a matrix of size n. The fifth feature matrix of 1 is used to obtain the size of the prediction model as 1. The parameter matrix of n is multiplied by the fifth feature matrix to obtain the number of qualified items; wherein, k, m and n are all positive integers.

7. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images