A knowledge graph updating method and device

Abstract

The application provides a knowledge graph updating method, comprising: acquiring a static knowledge graph, the static knowledge graph being constructed based on target data, the target data comprising historical data generated in a semiconductor manufacturing process; constructing a dynamic knowledge graph based on current data, the current data being data generated in the semiconductor manufacturing process in a preset current time period; and updating the static knowledge graph based on the dynamic knowledge graph. The knowledge graph updating method provided by the application uses knowledge generated by analyzing data generated in subsequent semiconductor manufacturing processes, such as defect discovery and root cause, to continuously update and complete the static knowledge graph, thereby ensuring the comprehensiveness of the knowledge graph.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor manufacturing, and in particular to a knowledge graph updating method and device. BACKGROUND

[0002] In the field of semiconductor integrated circuit manufacturing, the yield of chips is crucial. When yield problems occur online, the root cause needs to be investigated as soon as possible and solved in time to protect the normal operation of the production line 7x24 hours.

[0003] Existing yield analysis often requires manual (such as yield analysis engineers) to perform yield analysis to find wafer defects and the root causes corresponding to the defects, which often consumes a lot of manpower and time. Therefore, a knowledge graph fused with engineer experience knowledge can be applied to wafer yield analysis to realize wafer yield analysis based on the knowledge graph, and reasoning on the knowledge graph to realize automatic wafer defect discovery and root cause discovery. The comprehensiveness of the knowledge in the knowledge graph is a key factor in wafer yield analysis based on the knowledge graph, so how to ensure the comprehensiveness of the knowledge in the knowledge graph is a problem to be solved. SUMMARY

[0004] Embodiments of the present application provide a knowledge graph updating method, which uses knowledge generated by analyzing data generated in subsequent semiconductor manufacturing processes, such as defect discovery and root cause, to continuously update and complete the static knowledge graph, ensuring the comprehensiveness of the knowledge in the knowledge graph.

[0005] In a first aspect, the present application provides a knowledge graph updating method, comprising: obtaining a static knowledge graph, the static knowledge graph being constructed based on target data, the target data including historical data generated in a semiconductor manufacturing process; constructing a dynamic knowledge graph based on current data, the current data being data related to manufacturing generated in a current predetermined time period in a semiconductor manufacturing process; and updating the static knowledge graph based on the dynamic knowledge graph.

[0006] The knowledge graph updating method provided by the present application uses knowledge generated by analyzing data generated in subsequent semiconductor manufacturing processes, such as defect discovery and root cause, to continuously update and complete the static knowledge graph, ensuring the comprehensiveness of the knowledge in the knowledge graph.

[0007] In one possible implementation, the static knowledge graph includes a plurality of first nodes and a plurality of first edges connecting the plurality of first nodes, the first nodes representing first entities related to the semiconductor manufacturing in the target data, and the first edges representing association relationships between the plurality of first nodes.

[0008] The dynamic knowledge graph includes a plurality of second nodes and a plurality of second edges connecting the plurality of second nodes, the second nodes representing second entities related to the wafer manufacturing involved in the current data, and the second edges representing association relationships between the plurality of second nodes.

[0009] The static knowledge graph is updated based on the dynamic knowledge graph, including:

[0010] Second nodes different from the first nodes and second edges different from the first edges are supplemented into the static knowledge graph.

[0011] In one possible implementation, the dynamic knowledge graph includes a plurality of

[0012] The static knowledge graph is updated based on the dynamic knowledge graph, including:

[0013] Second nodes different from the first nodes and second edges different from the first edges with a number of occurrences greater than a preset number are supplemented into the static knowledge graph.

[0014] In another possible implementation, the dynamic knowledge graph includes a plurality of, the first edge and the second edge each have a weight;

[0015] The static knowledge graph is updated based on the dynamic knowledge graph, including:

[0016] The weight of the corresponding first edge in the static knowledge graph is adjusted based on the number of occurrences of the second edge in the plurality of dynamic knowledge graphs.

[0017] In another possible implementation, the static knowledge graph is updated based on the dynamic knowledge graph, including:

[0018] The triple information of the static knowledge graph and the triple information of the dynamic knowledge graph are input into a trained graph fusion model to obtain an updated static knowledge graph;

[0019] The triple information of the static knowledge graph represents association relationships between the plurality of first nodes, and the triple information of the dynamic knowledge graph represents association relationships between the plurality of second nodes.

[0020] In another possible implementation, the static knowledge graph is obtained, including:

[0021] A plurality of first entities and relationships between the plurality of first entities are extracted from the target data.

[0022] The static knowledge graph is constructed based on the plurality of first entities and the relationships between the plurality of first entities.

[0023] In another possible implementation,

[0024] The historical data comprises one or more of structured data, semi-structured data and unstructured data;

[0025] The structured data comprises data related to the semiconductor manufacturing equipment and / or data related to the wafer;

[0026] The semi-structured data comprises experience documents generated by engineers based on analysis of the structured data;

[0027] The unstructured data comprises documents related to the semiconductor manufacturing equipment, including one or more of usage instruction information of the semiconductor manufacturing equipment, fault information and repair information corresponding to the fault information.

[0028] In another possible implementation, the current data comprises statistical process control data of several production processes in the current time period;

[0029] The constructing the dynamic knowledge graph based on the current data comprises:

[0030] performing semantic analysis on the statistical process control data to obtain description texts corresponding to the statistical process control data, the description texts describing abnormal information corresponding to the statistical process control data;

[0031] extracting a plurality of second entities and relationships between the second entities from the description texts;

[0032] constructing a dynamic knowledge graph based on the plurality of second entities and the relationships between the second entities.

[0033] In another possible implementation, the semiconductor manufacturing is wafer manufacturing, and the first entities comprise at least two of manufacturing equipment of the wafer, defects of the wafer and root causes corresponding to the defects of the wafer;

[0034] The second entities comprise at least two of the manufacturing equipment of the wafer, the defects of the wafer, the root causes corresponding to the defects of the wafer, batches of the wafer, production steps of the wafer, operations corresponding to the production steps of the wafer, failure categories on the wafer, defect detection layers of the wafer, abnormal events of the wafer and root causes corresponding to the abnormal events of the wafer.

[0035] In a second aspect, the present application provides an updating device of a knowledge graph, comprising:

[0036] An acquisition module is configured to acquire a static knowledge graph, the static knowledge graph being constructed based on target data, the target data including historical data generated in a semiconductor manufacturing process;

[0037] A construction module is configured to construct a dynamic knowledge graph based on current data, the current data being data related to manufacturing generated in the semiconductor manufacturing process in a preset current time period;

[0038] An updating module is configured to update the static knowledge graph based on the dynamic knowledge graph.

[0039] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.

[0040] In a fourth aspect, the present application further provides a computing device comprising a memory and a processor, the memory having instructions stored therein, which, when executed by the processor, cause the method of the first aspect to be implemented.

[0041] In a fifth aspect, the present application provides a computer program or computer program product comprising instructions which, when executed, cause a computer to perform the method of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A flowchart of a knowledge graph updating method provided by an embodiment of the present application;

[0043] Figure 2 A schematic diagram of a static knowledge graph;

[0044] Figure 3 A schematic diagram of a rule template;

[0045] Figure 4 A schematic diagram of a dynamic knowledge graph;

[0046] Figure 5 A schematic diagram of a static knowledge graph after updating;

[0047] Figure 6 A structural schematic diagram of a graph fusion model;

[0048] Figure 7 A structural schematic diagram of a knowledge graph updating device provided by an embodiment of the present application;

[0049] Figure 8 A structural schematic diagram of a computing device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0050] The technical solutions of the present application are described in further detail below with reference to the accompanying drawings and examples.

[0051] Figure 1 A flowchart of a knowledge graph updating method provided by an embodiment of the present application. The method can be executed by any computing device, equipment, platform or device cluster. The specific computing device executing the method is not limited in the present application, and a suitable computing device can be selected for execution as needed. As shown in the figure, the knowledge graph updating method comprises at least steps S101-S103. Figure 1

[0052] In step S101, a static knowledge graph is obtained.

[0053] The static knowledge graph refers to the knowledge accumulated in the past and organized in the form of a graph. The stored knowledge is common sense and general purpose.

[0054] The static knowledge graph can be pre-constructed based on target data, which includes historical data generated in the semiconductor manufacturing process.

[0055] The semiconductor can be a semiconductor in the popular sense such as a wafer or a display screen. The specific solutions of the present application are described below with the wafer as an example.

[0056] In one example, the historical data can include structured data and / or semi-structured data; that is, the historical data used to construct the static knowledge graph can be multi-source heterogeneous data.

[0057] The structured data includes wafer manufacturing equipment-related data and / or wafer-related data. The wafer manufacturing-related data, for example, is numerical data generated by wafer manufacturing equipment, which can include data obtained by sensors during production, including but not limited to temperature, humidity, pressure, voltage, current, equipment utilization, alarm information, etc.; the wafer-related data, for example, is wafer yield data, data obtained by defect detection during the production process, data obtained by electrical testing during the production process, and data of failure types (bins) obtained during final yield testing, such as CPU interval failure, GUP interval failure, and storage interval failure.

[0058] The semi-structured data includes experience documents generated by engineers analyzing the structured data. The experience documents mainly express abnormal data analyzed from the structured data and analyze these abnormal data to infer wafer defects caused by the abnormal data and the root causes corresponding to the wafer defects. The experience documents of engineers are usually semi-structured data.

[0059] ​For example, in the production process, there are some conventional ppt templates to help engineers better summarize and report problems, and the common format is as follows:

[0060] i. Description of the problem

[0061] ii. Means of inspection

[0062] iii. Attributions

[0063] iv. Next step of the correction scheme

[0064] In other words, the target data is multi-source heterogeneous data, including structured data generated by each machine in the manufacturing process of the wafer, and semi-structured experience document data formed by engineers analyzing the structured data.

[0065] In another example, the target data can also include data of a third source, which is a document related to the manufacturing equipment of the wafer, including one or more of the following: usage instruction information of the manufacturing equipment, fault information, and repair information corresponding to the fault.

[0066] The document related to the manufacturing equipment of the wafer is usually unstructured data.

[0067] For example, the document related to the manufacturing of the wafer can be the product manual of the equipment, which contains but is not limited to the following: instructions of the equipment, each small module of the equipment, common failures and solutions.

[0068] From the target data, a plurality of entities and relationships between the plurality of entities are extracted, and a static knowledge graph related to the manufacturing of the wafer is constructed by taking the entities as nodes and the relationships as edges.

[0069] Optionally, the plurality of entities can include event class entities and object class entities, for example, the event class entities can include one or more of the following: equipment (Tool) events, failure type (Bin) events, production step (Step) events, yield events, statistical process control (SPC) events, and root causes of defects; and the object class entities can include one or more of the following: defect detection layer (Layer) objects, equipment (Tool) objects, defect (Defect) objects, failure type (Bin) objects, and production step (Step) objects.

[0070] Optionally, the edges can include multiple types, such as the following: Entity-Event (En-Ev) type edges between object class entities and event class entities, Entity-Entity (En-En) type edges between object class entities, and Event-Event (Ev-Ev) type edges between event class entities.

[0071] Figure 2 A schematic diagram of a static knowledge graph is shown. The square nodes in the figure represent event class entities, the circular nodes represent object class entities, and the edges represent the relationships between the nodes. For example, the "TAD" node and the "Step1" node in the figure, and the "Step-Tool" edge between the nodes, indicate that the device TAD has a relationship with the step Step1, and it can be concluded that the step 1 in the wafer manufacturing process needs the device TAD to participate.

[0072] In step S102, a dynamic knowledge graph is constructed based on current data.

[0073] The current data is data related to manufacturing generated in the current preset time period in the wafer manufacturing process. For example, the current data includes all information related to wafer production, including production steps, devices for each step, related device events, abnormal events, etc.

[0074] The current data can be divided into two categories: one is SPC data generated by wafer manufacturing equipment in the current production cycle, and the other is data obtained by engineers analyzing the SPC data, which can be referred to as business data, including wafer batch, wafer production step, wafer failure category, wafer defect detection layer, wafer abnormal event, wafer defect, and defect corresponding root cause, etc.

[0075] Through knowledge extraction on the current data, a plurality of entities and relationships between the entities are obtained to construct a dynamic knowledge graph. The entities in the dynamic knowledge graph can include at least two of wafer batch, wafer production step, wafer production step corresponding operation, wafer failure category, wafer defect detection layer, wafer abnormal event, and wafer abnormal event corresponding root cause.

[0076] Since the SPC data is structured data with pure numerical values, in order to facilitate knowledge extraction, the SPC data needs to be converted into text data, that is, the SPC data is subjected to semantic analysis to obtain a description text corresponding to the SPC, which describes the abnormal information corresponding to the SPC data.

[0077] Then, based on the text data converted from the SPC data, knowledge extraction is performed to extract a plurality of entities and relationships between the entities.

[0078] The semantic recognition of the SPC data can be implemented in various ways, for example, according to a preset rule template, the corresponding description text is matched for the SPC data. For example, according to the rule template, the description text corresponding to the SPC data is matched. Figure 3The rule template shown is matched, for example, the distribution characteristics of the SPC data is "1 point is outside the control limits", that is, "one point falls outside the A zone", and the description text matched by the rule template for the SPC data is "A large shift", that is, "there is a large deviation".

[0079] The semantic recognition of the SPC data can also be realized by inputting the SPC data into the trained semantic recognition model to obtain the description text corresponding to the SPC data. Exemplarily, the framework of the semantic recognition model can be seq2seq, the seq2seq framework includes an encoder and a decoder, the seq2seq framework is trained by a training data set to establish a mapping relationship between a numerical sequence and a text, and finally a Data-To-Text model, that is, a required semantic recognition model, is obtained.

[0080] The current preset time period can be understood as a current update period, for example, the update period of the knowledge graph is one month, and the current preset time period is the time period of the current one month. That is, the information related to the wafer production in the current one month is collected and summarized.

[0081] It can be understood that the meaning of the current production period mentioned above is the wafer production period in the current update period, and the current update period can include a plurality of wafer production periods, for example, the current update period includes five wafer production periods, and the five wafer production periods can all become the current production period.

[0082] In step S103, the static knowledge graph is updated based on the dynamic knowledge graph.

[0083] Unlike the static knowledge graph, the dynamic knowledge graph is a knowledge graph with a time dimension, for example, the nodes thereof include a wafer batch node and an SPC event node of the batch. A dynamic knowledge graph can be generated for each batch of wafers, that is, each production period in the current update period, and a plurality of dynamic knowledge graphs can be generated in one update period, and the plurality of dynamic knowledge graphs form a dynamic knowledge graph that presents events in chronological order.

[0084] The dynamic knowledge graph generally organizes one-time data into one-time graph. When a user analyzes the data result, the data can be converted into one-time graph and fused with the static knowledge graph for visual analysis and reasoning. When the one-time data (e.g., business data or SPC data of one production cycle) is analyzed, the one-time graph is destroyed. If the analyzed data can precipitate new general knowledge (such as a Wafer Bin analysis document), the new general knowledge is fused into the static knowledge graph to supplement the content of the static knowledge graph and complete the update of the static knowledge graph. In other words, the useful knowledge in the dynamic knowledge graph is extracted and fused into the static knowledge graph to complete the knowledge completion of the static knowledge graph.

[0085] In one example, the nodes and edges in the dynamic knowledge graph that are different from the static knowledge graph can be supplemented into the static knowledge graph to form a new static knowledge graph.

[0086] Figure 4 A dynamic knowledge graph in an update cycle is shown. The nodes and connection relationships in the dynamic knowledge graph that are different from the static knowledge graph are extracted and supplemented into the static knowledge graph to update and complete the static knowledge graph. The updated static knowledge graph is shown in FIG. 6B. Figure 1 The nodes and connection relationships in the dynamic knowledge graph that are different from the static knowledge graph are extracted and supplemented into the static knowledge graph to update and complete the static knowledge graph. The updated static knowledge graph is shown in FIG. 6B. Figure 5

[0087] In another example, the current update cycle has a production cycle of multiple wafers, and multiple dynamic knowledge graphs can be generated. The nodes and edges in the dynamic knowledge graph that appear more than a preset number of times (e.g., 5 times) are supplemented into the static knowledge graph to update and complete the static knowledge graph.

[0088] In another example, the current update cycle has a production cycle of multiple wafers, and multiple dynamic knowledge graphs can be generated. The edges in the static knowledge graph also have weight information. Based on the number of times the edges in the multiple dynamic knowledge graphs appear, the weight of the corresponding edges in the static knowledge graph is adjusted.

[0089] For example, when the number of times the edges in the dynamic knowledge graph appear is more, it indicates that the relationship often occurs and the knowledge is more important. If the edge does not exist in the static knowledge graph, the edge can be supplemented into the static graph and set to a larger weight. If the edge exists in the static knowledge graph, the weight of the edge can be increased so that the weight calculation of the edge is increased in the reasoning process.

[0090] ​In another example, the triple information of the static knowledge graph and the triple information of the dynamic knowledge graph can be input into the trained graph fusion model to obtain an updated static knowledge graph; wherein the triple information of the static knowledge graph represents the association relationship between a plurality of first nodes; and the triple information of the dynamic knowledge graph represents the association relationship between a plurality of second nodes. For example, Figure 5 The triple information in the illustrated knowledge graph can include (Step 1, Step-Tool, TAD), (Step 1, Layer-Step, Metal 1), etc., that is, the triple information represents node 1 in the knowledge graph, the relationship between node 1 and node 2, and node 2. That is, the triple relationship of a knowledge graph is obtained, that is, the expression information of the knowledge graph is obtained, that is, the triple information of the knowledge graph can be obtained to construct the knowledge graph.

[0091] The structure of the graph fusion model is as shown in Figure 6 The model is a relationship path model based on a recurrent skip network, which can effectively explore the semantic information of long-tail entities in the knowledge graph, and to some extent, solve the problem of insufficient local expression of existing knowledge graphs. The relationship path can be understood as a chain structure formed by the interlacing of entities and relationships in the knowledge graph, that is, the relationship path of the knowledge graph is mastered, the composition information of the knowledge graph is mastered, and the updated static knowledge graph can be constructed, thereby solving the multi-node path information problem.

[0092] The input of the graph fusion model is the triple information of the static knowledge graph and the triple information of the dynamic knowledge graph, and the output is the updated static knowledge graph. The training process of the graph fusion model is as follows:

[0093] Input: Knowledge graph triple sequence G

[0094] Output: Trained graph fusion model

[0095] 1: Initialize all parameters by the Xavier method

[0096] 2: while There is no convergence do

[0097] 3: Calculate h t ,ω

[0098] 4: for i←1to T do

[0099] 5: Update the coefficient of the weight matrix ω

[0100] 6: Update the parameters in RSN

[0101] 7: Calculate h ′ t

[0102] 8: end for

[0103] 9: Calculate the objective function using the formula

[0104] 10: Update all parameters using the gradient descent method

[0105] 11: end while

[0106] where, represents the loss function of the graph fusion model, and the meaning of step 2 is to continue training if the loss function of the model does not converge, that is, the model is trained until the loss function of the model converges.h t represents the output of a neuron in the model, and represents the weight of a neuron in the model.

[0107] In the production process of the wafer, the triple information of the dynamic knowledge graph containing correct knowledge and the triple information of the static knowledge graph to be updated are input into the trained graph fusion model for fusion to obtain the updated static knowledge graph, that is, the new knowledge generated in the subsequent production process of the wafer is supplemented to the static knowledge graph to ensure the comprehensiveness and invalidity of the static knowledge graph.

[0108] In summary, the updating method of the knowledge graph provided in the embodiment of the present application constructs a static knowledge graph from the prior knowledge of engineers, forms some common and general knowledge, and then fuses some new general knowledge in the dynamic knowledge graph generated in the subsequent wafer manufacturing process into the static knowledge graph to realize the updating and supplementing of the static knowledge graph and ensure the comprehensiveness and timeliness of the knowledge in the static knowledge graph.

[0109] Based on the same concept as the foregoing method embodiment, the embodiment of the present application further provides a knowledge graph updating device 700, which comprises units or modules for realizing each step in the method shown in the figure. Figure 1

[0110] Figure 7 A structural diagram of a knowledge graph updating device provided in the embodiment of the present application is shown in the figure. Figure 7 As shown in the figure, the knowledge graph updating device 700 at least comprises:

[0111] The acquisition module 701 is configured to acquire a static knowledge graph, wherein the static knowledge graph is constructed based on target data, and the target data includes historical data related to manufacturing generated in the wafer manufacturing process and experience documents formed by engineers analyzing wafer data; ​

[0112] The construction module 702 is configured to construct a dynamic knowledge graph based on current data, the current data being manufacturing-related data generated in the wafer manufacturing process in a current preset time period;

[0113] The updating module 703 is configured to update the static knowledge graph based on the dynamic knowledge graph.

[0114] The updating device 700 of the knowledge graph according to the embodiments of the present application can correspond to the method described in the embodiments of the present application, and the above and other operations and / or functions of each module in the updating device 700 of the knowledge graph are respectively used to realize the corresponding processes of each method in the embodiments of the present application, and the specific implementation can be referred to the description above, and for the sake of brevity, will not be repeated here. Figure 1

[0115] The embodiments of the present application further provide a computing device, comprising at least one processor, a memory and a communication interface, the processor being used to execute the method. Figure 1 The computing device can be a server or a terminal device.

[0116] Figure 8 The computing device provided by the embodiments of the present application is shown in a structural schematic diagram.

[0117] As shown in Figure 8 , the computing device 800 comprises at least one processor 801, a memory 802, a communication interface 803. Wherein, the processor 801 and the memory 802 and the communication interface 803 are in communication connection, and can realize communication through wireless or wired and the like. The communication interface 803 is used to receive user instruction or information sent by acquisition device; the memory 802 stores computer instruction, and the processor 801 executes the computer instruction, and executes the method in the foregoing method embodiment.

[0118] It should be understood that, in the embodiments of the present application, the processor 801 can be a central processing unit CPU, and the processor 801 can also be other general-purpose processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), field programmable gate arrays (field programmable gate array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0119] ​The memory 802 can include read-only memory and random access memory, and provide instructions and data to the processor 801. The memory 802 can also include non-volatile random access memory.

[0120] The memory 802 can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically EPROM (EEPROM) or flash memory. The volatile memory can be random access memory (RAM) used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchlink dynamic random access memory (SLDRAM) and direct rambus random access memory (DR RAM).

[0121] It should be understood that the computing device 800 according to the embodiments of the present application can perform the method shown in the embodiments of the present application, and the detailed description of the method implemented is described above. For the sake of brevity, it will not be repeated here. Figure 1

[0122] The embodiments of the present application provide a computer readable storage medium, which stores a computer program, when the computer program is executed by a processor, the knowledge graph updating method mentioned above is implemented.

[0123] The embodiments of the present application provide a computer program or computer program product, which includes instructions, when the instructions are executed, the computer executes the knowledge graph updating method mentioned above.

[0124] ​Those skilled in the art should further understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are realized 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 realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0125] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented in hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0126] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for updating a knowledge graph, characterized in that, include: A static knowledge graph is acquired, constructed based on target data, which includes historical data generated during semiconductor manufacturing. The historical data includes one or more of structured data, semi-structured data, and unstructured data. The structured data includes data related to semiconductor manufacturing equipment and / or wafer-related data. The semi-structured data includes experience documents generated by engineers analyzing the structured data. The unstructured data consists of documents related to the semiconductor manufacturing equipment, including one or more of the following: user manuals, fault information, and repair information corresponding to the fault information. The static knowledge graph includes multiple first nodes and multiple first edges connecting the multiple first nodes. Each first node represents a first entity related to semiconductor manufacturing in the target data, and each first edge represents the association between the multiple first nodes. Semiconductor manufacturing is wafer manufacturing. The first entity includes at least two of the following: wafer manufacturing equipment, wafer defects, and root causes corresponding to the wafer defects. A dynamic knowledge graph is constructed based on current data, which refers to data generated during the semiconductor manufacturing process within the current preset time period. The current data includes descriptive text obtained from Statistical Process Control (SPC) data generated by the wafer manufacturing equipment in the current production cycle, and business data derived from engineer analysis of the SPC data. This business data includes wafer batches, wafer production steps, wafer failure categories, wafer defect detection layers, wafer anomalies, wafer defects, and the root causes of those defects. The dynamic knowledge graph includes multiple second nodes and multiple second edges connecting these nodes. A second node represents a second entity related to semiconductor manufacturing in the current data, and a second edge represents the association between the multiple second nodes; the dynamic knowledge graph includes multiple entities, with one dynamic knowledge graph corresponding to each wafer production cycle; the second entity includes at least two of the following: wafer manufacturing equipment, wafer defects and their root causes, wafer batch, wafer production steps, operations corresponding to the wafer production steps, failure categories on the wafer, wafer defect detection layer, wafer anomalies and their root causes; The static knowledge graph is updated based on the dynamic knowledge graph; The step of updating the static knowledge graph based on the dynamic knowledge graph includes: Second nodes that appear more than a preset number of times, are different from the first node, and second edges that are different from the first edge are added to the static knowledge graph.

2. The method according to claim 1, characterized in that, Both the first edge and the second edge have weights; The step of updating the static knowledge graph based on the dynamic knowledge graph includes: The weight of the corresponding first edge in the static knowledge graph is adjusted based on the number of times the second edge appears in multiple dynamic knowledge graphs.

3. The method according to claim 1 or 2, characterized in that, The step of updating the static knowledge graph based on the dynamic knowledge graph includes: The triple information of the static knowledge graph and the triple information of the dynamic knowledge graph are input into the trained graph fusion model to obtain the updated static knowledge graph. The triple information of the static knowledge graph represents the association relationship between the multiple first nodes; the triple information of the dynamic knowledge graph represents the association relationship between the multiple second nodes.

4. The method according to claim 1 or 2, characterized in that, The current data includes statistical process control data for several production processes within the current time period; The construction of a dynamic knowledge graph based on current data includes: Semantic analysis is performed on the statistical process control data to obtain the descriptive text corresponding to the statistical process control data, and the descriptive text describes the abnormal information corresponding to the statistical process control data. Extract multiple second entities from the description text, as well as the relationships between the multiple second entities; A dynamic knowledge graph is constructed based on the plurality of second entities and the relationships between the plurality of second entities.

5. A knowledge graph updating device, characterized in that, include: The acquisition module is configured to acquire a static knowledge graph, which is constructed based on target data. The target data includes historical data generated during semiconductor manufacturing. The historical data includes one or more of structured data, semi-structured data, and unstructured data. The structured data includes data related to semiconductor manufacturing equipment and / or wafer-related data. The semi-structured data includes experience documents generated by engineers analyzing the structured data. The unstructured data consists of documents related to the semiconductor manufacturing equipment, including one or more of the following: user manual information, fault information, and repair information corresponding to the fault information. The static knowledge graph includes multiple first nodes and multiple first edges connecting the multiple first nodes. Each first node represents a first entity related to semiconductor manufacturing in the target data, and each first edge represents the association between the multiple first nodes. Semiconductor manufacturing is wafer manufacturing. The first entity includes at least two of the following: wafer manufacturing equipment, wafer defects, and root causes corresponding to the wafer defects. The construction module is configured to build a dynamic knowledge graph based on current data, which is manufacturing-related data generated during the semiconductor manufacturing process within a current preset time period. The current data includes descriptive text obtained from Statistical Process Control (SPC) data generated by the wafer manufacturing equipment in the current production cycle, and business data derived from engineer analysis of the SPC data. The business data includes wafer batches, wafer production steps, wafer failure categories, wafer defect detection layers, wafer anomalies, wafer defects, and the root causes of the defects. The dynamic knowledge graph includes multiple second nodes and multiple second edges connecting the multiple second nodes. Each second node represents a second entity related to semiconductor manufacturing in the current data, and each second edge represents the association between the multiple second nodes. Multiple dynamic knowledge graphs are included, with one dynamic knowledge graph corresponding to each wafer production cycle. The second entity includes at least two of the following: the wafer manufacturing equipment, the wafer defect and the root cause of the wafer defect, the wafer batch, the wafer production step, the operation corresponding to the wafer production step, the failure category on the wafer, the wafer defect detection layer, the wafer abnormal event and the root cause of the wafer abnormal event. The update module is configured to add second nodes that appear more than a preset number of times, are different from the first node, and second edges that are different from the first edge to the static knowledge graph.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed in a computer, it causes the computer to perform the method described in any one of claims 1-4.

Images