Semiconductor device manufacturing system, server, and foreign matter generation cause determination method

Abstract

The invention provides a technology capable of coping with foreign matters caused by an etching process without processing a coating layer of a member in a container. One of the semiconductor device manufacturing systems according to the present invention is characterized in that: the semiconductor device manufacturing system is provided with a stage on which an application for specifying the cause of generation of foreign matter adhering to a sample processed by a semiconductor manufacturing apparatus is mounted; an application executes the steps of: creating a graph in which a process history is used as a node, and the difference between the number of foreign matters in one connected node and the number of foreign matters in the other connected node is used as an edge; estimating an edge by a neural network using a graph, the neural network having the number of foreign matters and the process history that have been acquired being input; the estimated edge is used to determine a process history that is a cause of generation of the foreign matter, the process history including a manufacturing condition of the semiconductor manufacturing apparatus.

Description

Technical Field

[0001] This invention relates to a semiconductor device manufacturing system, a server, and a method for determining the cause of foreign object generation. Background Technology

[0002] Semiconductor device manufacturing apparatuses, such as plasma processing apparatuses, include substrate processing equipment, vacuum processing chambers, gas supply devices, electrodes for mounting wafers (hereinafter also referred to as "samples"), plasma generation devices, etc. In these substrate processing apparatuses, plasma processing of the wafers, such as etching, is performed by generating plasma gas.

[0003] In recent years, with the miniaturization of semiconductor devices and the reduction in the width of wiring, the impact of foreign objects on the manufacturing process has become greater. In particular, there are concerns about an increased rate of short circuits due to the adhesion of foreign objects that are relatively large than the wiring width, and adverse effects such as manufacturing processes being disrupted by foreign objects. Furthermore, the generation of foreign objects necessitates stopping the semiconductor manufacturing equipment for inspection, during which time semiconductor device production is halted. Therefore, shortening the time from foreign object generation to recovery has become an important challenge.

[0004] As an effective countermeasure to shorten recovery time, one approach is to determine the cause of foreign matter generation early on. For example, as shown in Patent Document 1, a known technique involves differentiating the coatings of components within a processing container to determine which component is the cause of the foreign matter based on the elements present. Furthermore, Patent Document 2 discloses a method that uses machine learning to determine the component causing the foreign matter generation based on the correlation between components within the processing container and foreign matter elements, analyzes cleaning conditions for that component, and performs cleaning until the number of foreign matter items falls below a threshold.

[0005] Specifically, Patent Document 1 addresses the invention of a plasma processing apparatus and a method for determining the replacement of components in the plasma processing apparatus and the surface coating of a component that has deteriorated or corroded within the processing container. The invention discloses that "the coating C1 on the surface of the support member 34 of the microwave transmission plate 31, which is exposed to plasma within the processing container of the plasma processing apparatus and supports the top plate of the processing container, and the coating C2 on the surface of the sidewall 2c connected to the support member 34, are both formed by coating with different materials. If the analysis of particles on the dummy wafer DW shows that their composition originates from the coating C1, the support member 34 is replaced; if the composition originates from the coating C2, it is determined that the sidewall 2c is corroded or deteriorated, and the corresponding component is replaced."

[0006] Furthermore, Patent Document 2 discloses, as an invention of a semiconductor device manufacturing system and a semiconductor device manufacturing method, the following: "A semiconductor manufacturing apparatus system and a semiconductor device manufacturing method are provided for reducing foreign matter that causes adverse effects in the manufacturing process of semiconductor devices. The foreign matter reduction process in the semiconductor device manufacturing system, which includes a semiconductor manufacturing apparatus and a platform for performing foreign matter reduction processing connected to the semiconductor manufacturing apparatus via a network, comprises the following steps: obtaining foreign matter characteristic values ​​using a sample processed by the semiconductor manufacturing apparatus; determining, by machine learning, the components of the semiconductor manufacturing apparatus that are caused by foreign matter based on the obtained foreign matter characteristic values ​​and related data; specifying cleaning conditions for cleaning the semiconductor manufacturing apparatus based on the determined components; and cleaning the semiconductor manufacturing apparatus using the specified cleaning conditions, wherein the related data is the foreign matter characteristic values ​​and related data of the components obtained in advance."

[0007] Existing technical documents

[0008] Patent documents

[0009] Patent Document 1: JP 2016-122772

[0010] Patent Document 2: International Publication No. 2021 / 241242 Summary of the Invention

[0011] The problem that the invention aims to solve

[0012] In the technology of Patent Document 1, different coatings are required on the components within the processing container to identify foreign matter, and multiple coatings of various materials are often needed for a single gas, making it sometimes difficult to achieve a suitable coating. Furthermore, in cases involving materials with high abrasion rates, there is a risk of impacting semiconductor devices. In addition, the technology of Patent Document 2 shows a determination based on the correlation between components and elements, but does not show a determination related to component abrasion. Therefore, even when a deteriorated component is present within the processing container, it is sometimes difficult to determine that the deteriorated component is the cause of the foreign matter. Furthermore, the prior art only considers foreign matter originating from components as the object, and does not include foreign matter originating from the etching process.

[0013] In view of the above-mentioned problems, the present invention provides a technology that can deal with foreign matter caused by etching processes without requiring the coating of components inside the container to be treated.

[0014] Methods for solving problems

[0015] To address the aforementioned issues, one representative semiconductor device manufacturing system of the present invention is characterized in that, in a semiconductor device manufacturing system having a platform equipped with an application for determining the cause of foreign matter attached to a sample processed by a semiconductor manufacturing apparatus, the application performs the following steps: creating a graph with process history records as nodes and the difference between the number of foreign matter in one node and the number of foreign matter in the other node as edges; estimating the edges by inputting the number of foreign matter and the process history records and utilizing the graph's neural network; and using the estimated edges to determine the process history records that are the cause of the foreign matter's generation, the process history records including the manufacturing conditions of the semiconductor manufacturing apparatus.

[0016] The effects of the invention

[0017] According to the present invention, a technology is provided that can provide a coating that can handle foreign matter caused by etching processes without requiring treatment of components inside the container.

[0018] Other issues, structures, and effects not mentioned above will become clear from the following description of the manner in which the invention is carried out. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating an example of the hardware structure involved in an embodiment of the present invention.

[0020] Figure 2 This is a diagram illustrating an example of the structure of a device for analyzing the causes of foreign objects.

[0021] Figure 3 This is a diagram illustrating an example of a functional block involved in an embodiment of the present invention.

[0022] Figure 4 This is a diagram illustrating an example of a flowchart related to an embodiment of the present invention.

[0023] Figure 5 This is a diagram illustrating an example of a sequence diagram related to an embodiment of the present invention.

[0024] Figure 6 This is a diagram illustrating an example of a foreign matter generation diagram according to an embodiment of the present invention.

[0025] Figure 7 This is a diagram representing an example of a dataset input into a neural network.

[0026] Figure 8 This is an example of a graphical user interface (GUI) for an application device that illustrates the cause of foreign object generation. Detailed Implementation

[0027] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the present invention is not limited to these embodiments. Furthermore, in the accompanying drawings, the same reference numerals are used to indicate the same parts.

[0028] As will be described later, the semiconductor device manufacturing system 1 of the present invention includes a network capable of accessing the semiconductor manufacturing apparatus 111, the foreign object characteristic value acquisition device 112, and various databases (device database 115, application database 116), and executes a processing flowchart for reducing foreign objects from the point of determining the cause of foreign object generation. Here, "network" refers to, for example, the Internet, a wide area network (WAN), a local area network (LAN), a dedicated line, or an information communication network consisting of a combination of these.

[0029] Furthermore, the semiconductor device manufacturing system 1 according to an embodiment of the present invention includes: a semiconductor manufacturing apparatus 111; and a platform PF implemented via a network for reducing foreign matter analysis from determining the cause of foreign matter generation. Here, "platform" includes a server equipped with applications and a database.

[0030] Furthermore, in this disclosure, the term "PC" is not limited to personal computers; the term "computer (processing device)" also includes servers.

[0031] Furthermore, the term "application" refers to a program executed by a PC.

[0032] In addition, "cloud" refers to computer resources such as application servers, database servers, and shared disks obtained through the network.

[0033] (Hardware structure)

[0034] The following is for reference. Figures 1 to 8 The embodiments of the present invention will be described below.

[0035] Figure 1 This diagram illustrates an example of the hardware structure involved in an embodiment of the present invention. The semiconductor device manufacturing system 1 includes a semiconductor manufacturing apparatus 111, a foreign object characteristic value acquisition device 112, a PC 113, a PC 114, a device database 115, an application database 116, a monitor 117, a cloud 118, a foreign object cause analysis application device 119, a cleaning implementation device 120, a component replacement device 121, and a process formulation discussion device 122. The platform PF includes the device database 115, the application database 116, the monitor 117, the cloud 118, and the foreign object cause analysis application device 119.

[0036] Semiconductor manufacturing apparatus 111, PC 113, foreign object characteristic value acquisition device 112, PC 114, cleaning implementation device 120, component replacement device 121, and process formulation discussion device 122 constitute a closed network CN. Semiconductor manufacturing apparatus 111 is connected to PC 113 in the closed network CN. PC 113 is connected to device database 115 via a local network. Similarly, foreign object characteristic value acquisition device 112 is connected to PC 114 in the closed network CN, and PC 114 is connected to application database 116 via a local network.

[0037] The cloud 118 references the device database 115 and the application database 116, and forwards the referenced information to the foreign object cause analysis application device 119. For security reasons, it is preferable that the permissions from the cloud 118 to the device database 115 and the application database 116 are only for reference.

[0038] Foreign object (FOO) cause analysis application device 119 is, for example, a server that executes the FAO cause analysis application. The FAO cause analysis application device 119 analyzes the cause of the foreign object (hereinafter also referred to as the "foreign object generation cause") from information obtained from the cloud 118, and displays the analysis results on the monitor 117 via the cloud 118. Furthermore, the FAO cause analysis application device 119 may also have a display device, and the display of the analysis results may also be performed by the graphical user interface (GUI) of the FAO cause analysis application device 119 (e.g., a display device), as detailed later. For example, the FAO cause analysis application device 119 can also modify the user-recommended process formula based on the determined foreign object generation cause via the GUI.

[0039] In addition, the cleaning implementation device 120 cleans the semiconductor manufacturing apparatus 111. The component replacement device 121 determines the conditions for component replacement in the semiconductor manufacturing apparatus 111 and performs the component replacement. The process formulation discussion device 122 discusses changes to the process formulation implemented by the semiconductor manufacturing apparatus 111. Furthermore, these devices 120 to 122 may also be equipped with a GUI, allowing user operations to be accepted via the GUI, reflecting user instructions during cleaning, component replacement, and process formulation discussion, as detailed later. Furthermore, since the cleaning implementation device 120, component replacement device 121, and process formulation discussion device 122 are connected to the PC 113, they can also send information to the device database 115. Moreover, the cleaning implementation device 120, component replacement device 121, and process formulation discussion device 122 can also send information to the foreign matter cause analysis application device 119 and display information on the monitor 117.

[0040] In addition, regarding the foreign matter characteristic value acquisition device 112, one device is shown, but it can also be composed of multiple devices to obtain information such as foreign matter elements, foreign matter number, and foreign matter distribution on the sample.

[0041] Furthermore, the structure of the semiconductor device manufacturing system 1 is not limited to the above-described case. The semiconductor device manufacturing system 1 may also have a structure that includes at least a semiconductor manufacturing apparatus 111, a foreign object characteristic value acquisition device 112, a PC 113, a PC 114, and a platform PF.

[0042] Furthermore, the structure of the platform PF is not limited to the above-described cases. For example, the cloud 118 is not limited to computer resources provided via a network, but can also be constructed from hardware resources. The structure of the platform PF can be appropriately configured as long as an application is installed to determine the cause of foreign matter adhering to samples processed by a semiconductor manufacturing apparatus is installed. For example, as described later... Figures 4 to 6 As explained in the description, the platform PF functions as the foreign object cause analysis unit 215, but the platform PF only needs to be a structure that can execute the installed application.

[0043] (Structure of the foreign object cause analysis application device)

[0044] Figure 2 This diagram illustrates an example of the structure of the foreign object cause analysis application device 119. The foreign object cause analysis application device 119 includes a bus 1190, a processor 1191, a memory 1192, a storage device 1193, an input / output device 1194, and a communication interface 1195. The processor 1191, memory 1192, storage device 1193, input / output device 1194, and communication interface 1195, all connected to the bus 1190, communicate information via the bus 1190. The processor 1191 processes the acquired information. The memory 1192 stores the commands (applications) executed by the processor 1191. The memory 1192 may contain, for example, random access memory (RAM) or other dynamic storage devices, read-only memory (ROM), or other static storage devices used to save temporary variables or other intermediate information during the execution of commands executed by the processor 1191. By executing the commands contained in the memory, the processor 1191 can perform the functions of the foreign object cause analysis unit 215, described later.

[0045] Furthermore, storage device 1193 is used for storing information and commands, and may be composed of a disk or optical disk. Storage device 1193 may be a database storing commands. Input / output device 1194 includes: a display device for displaying information to the user; and an input device for transmitting information and command selections made by the user to processor 1191. For example, the display device is a monitor, and the input device is a mouse and keyboard. Communication interface 1195 enables bidirectional communication via a network. Through communication interface 1195, foreign object cause analysis application device 119 can communicate with cloud 118, cleaning implementation device 120, component replacement device 121, and process formula discussion device 122.

[0046] The structure of the foreign object cause analysis application device 119 has been described, but this disclosure is not limited to this structure. The foreign object cause analysis application device 119 may also be composed of other hardware circuits, or it may be composed of a combination of hardware circuits and software.

[0047] In addition, PC113 and PC114 may also have the same structure as the foreign object cause analysis application device 119.

[0048] Furthermore, the cleaning implementation device 120, the component replacement device 121, and the process formula discussion device 122, in addition to having the same structure as the foreign matter cause analysis application device 119, can also be configured to have a structure corresponding to each function.

[0049] (Structure of functional blocks)

[0050] Figure 3 This is a diagram illustrating an example of the functional blocks involved in an embodiment of the invention. Figure 3 In the diagram, solid lines primarily represent the flow of information, while dashed lines primarily represent the flow of objects. Figure 3 Indicates will Figure 1 The hardware structure is represented by functional blocks. Figure 3 Semiconductor Manufacturing Division 211 and Figure 1 The semiconductor manufacturing device 111 corresponds to this. Figure 3 Foreign object characteristic value acquisition part 212 and Figure 1 The foreign object characteristic value acquisition device 112 corresponds to this. Figure 3 Device database 213 and Figure 1 The device database corresponds to 115. Figure 3 Application Database 214 and Figure 1 The application database 116 corresponds to this. Figure 3 Foreign object cause analysis department 215 and Figure 1 The foreign object cause analysis application device 119 corresponds to this. Figure 3 Process formulation discussion department 216 and Figure 1 The process formulation discussion device 122 corresponds to this. Figure 3 Cleaning Implementation Department 217 and Figure 1 The cleaning implementation device 120 is corresponding. Figure 3 Part replacement part 218 and Figure 1 The component replacement device 121 corresponds to this. The semiconductor manufacturing unit 211, the foreign object characteristic value acquisition unit 212, the process formula research unit 216, the cleaning implementation unit 217, and the component replacement unit 218 are included in the closed network CN, and the device database 213, the application database 214, and the foreign object cause analysis unit 215 are included in the platform PF.

[0051] Semiconductor device manufacturing system 1 includes: a semiconductor manufacturing apparatus for processing samples based on process recipes; a foreign matter characteristic value acquisition unit; and a foreign matter cause analysis unit for determining the cause of foreign matter generation in the samples. Here, the case of semiconductor manufacturing is illustrated, and the functional blocks are explained. The case where the semiconductor manufacturing unit 211 performs etching is explained (i.e., the case where the semiconductor manufacturing apparatus 111 is an etching apparatus).

[0052] The sample Sa, after being processed in the semiconductor manufacturing section 211, is transported to the foreign matter characteristic value acquisition section 212. The transport method can be either manual or automatic. Furthermore, the process recipe Re and equipment error information (hereinafter also referred to as "equipment error history record Eh") during processing in the semiconductor manufacturing section 211 are stored in the equipment database 213. The process recipe Re records parameters controlling the operation of the etching apparatus, the type and flow rate of the processing gas, and the temperature and heat of the heater of the sample loading stage. Parameters may also include sensor values ​​installed within the apparatus. In addition, information such as the target of the process recipe, the execution time, and the sample to be processed is also stored in the equipment database 213. The target of the process recipe Re is, for example, the shape of the semiconductor device manufactured according to the process recipe Re. Information associated with these process recipes Re also includes usage history information for multiple process recipes used in the processing of sample Sa, stored in the equipment database 213 as the process history record Rh. The process history record Rh also includes, for example, the manufacturing conditions in the semiconductor manufacturing section 211. The device error history record Eh contains information indicating whether the error was due to a voltage system error or pressing the emergency stop button.

[0053] The foreign object characteristic value acquisition unit 212 uses the sample processed by the semiconductor manufacturing unit 211 to acquire foreign object characteristic values ​​including the number of foreign objects. The foreign object characteristics acquired by the foreign object characteristic value acquisition unit 212, specifically including information such as foreign object elements, the number of foreign objects, and the mapping of foreign objects on the sample (hereinafter also referred to as "foreign object characteristic value Ci"), are stored in the application database 214. Furthermore, if defects can also be detected simultaneously in the foreign object characteristic value acquisition unit 212, it is preferable to also store the defect information as a foreign object characteristic value Ci in the application database 214. In this case, to maintain compatibility between the device database 213 and the application database 214 (e.g., which process and sample were used to produce the product), it is preferable to mark the process recipe Re, process history record Rh, device error history record Eh, and foreign object characteristic value Ci with a unique ID (Identification).

[0054] The foreign object cause analysis unit 215 creates a foreign object generation graph Gr, which represents the process history record Rh as nodes and the increase or decrease in the number of foreign objects as edges. It uses the obtained number of foreign objects and the process history record Rh as input and employs a machine learning model to infer the edges. Specifically, the foreign object cause analysis unit 215 creates the foreign object generation graph Gr based on information stored in the device database 213 and the application database 214. The foreign object generation graph Gr is a directed acyclic graph, where the causes of foreign object generation are represented as nodes and the increase or decrease in the number of foreign objects are represented as edges.

[0055] Furthermore, the foreign object cause analysis unit 215 identifies the edge that is the cause of foreign object generation from the predicted edges. Regarding the description after identification, the machine learning model used is, for example, a neural network. The foreign object cause analysis unit 215 determines whether to perform process recipe Re modification, cleaning implementation in the semiconductor manufacturing unit 211, or component replacement in the semiconductor manufacturing unit 211 to eliminate the identified foreign object generation cause. The means of elimination can be set in accordance with the structure of the semiconductor device manufacturing system 1. For example, the foreign object cause analysis unit 215 sends a foreign object generation chart Gr to at least one of the process recipe analysis unit 216, cleaning implementation unit 217, and component replacement unit 218, causing processing to be performed to eliminate the foreign object generation cause shown in the foreign object generation chart Gr. For example, the process recipe analysis unit 216 improves the portion of the recipe with an increased number of foreign objects based on the obtained foreign object generation chart Gr, and forwards the revised process recipe Rem to the semiconductor manufacturing unit 211. In the cleaning implementation unit 217 and the component replacement unit 218, the conditions for component replacement and cleaning are discussed based on the obtained foreign matter generation chart Gr, and the content of the implementation, the implementation time, and information about the replaced components are also stored in the device database 213. Specifically, the cleaning implementation unit 217 stores information representing the historical implementation history of cleaning performed in the semiconductor manufacturing unit 211, namely the cleaning implementation history Ch, in the device database 213. In addition, the component replacement unit 218 stores information representing the historical replacement history of components included in the semiconductor manufacturing unit 211, namely the component replacement history Ph, in the device database 213.

[0056] (One example of what was handled)

[0057] Figure 4 This is a flowchart illustrating an example of an embodiment of the present invention. In step S1, the sample Sa processed in the semiconductor manufacturing unit 211 is transported to the foreign matter characteristic value acquisition unit 212. Next, in step S2, the process formula Re of the semiconductor manufacturing unit 211 is stored in the device database 213. Next, in step S3, the number of foreign matter, foreign matter elements, etc., are acquired in the foreign matter characteristic value acquisition unit 212, and the acquired information is stored in the application database 214.

[0058] Next, the foreign object cause analysis unit 215 performs the following steps: It creates a graph with process history records as nodes and the difference between the number of foreign objects in one node and the number of foreign objects in the other node as edges; it estimates the edges using a neural network based on the obtained number of foreign objects and the process history records; and it uses the estimated edges to determine the process history records that are the cause of the foreign object's formation. Furthermore, in the step of estimating the edges, it further inputs information representing the component replacement history of the semiconductor manufacturing apparatus (i.e., component replacement history records) and information representing the cleaning implementation history of the semiconductor manufacturing apparatus (i.e., cleaning implementation history records) to estimate the edges. Furthermore, based on the determined cause of the foreign object's formation, the application recommends modifications to the manufacturing conditions.

[0059] Specifically, in step S4, the foreign matter cause analysis unit 215 represents the process recipe Re and the process recipe history (process history Rh) stored in the device database 213 as nodes. Additionally, during the period up to step S4, the cleaning implementation unit 217 stores the cleaning implementation history Ch in the device database 213. Furthermore, the component replacement unit 218 stores the component replacement history Ph in the device database 213.

[0060] In S5, the foreign object cause analysis unit 215 represents the increase or decrease of the number of foreign objects as edges, and uses the number of foreign objects and the historical process records as inputs to infer the edges using a machine learning model (e.g., a neural network). When using a neural network to infer the edges, the maximum likelihood method is used. The maximum likelihood method is a method of estimating the most likely value based on known observation data. In the embodiments of the present invention, it is used to infer the edges of the foreign object generation chart Gr, i.e., the increase or decrease of the number of foreign objects, based on the relationship between input information such as past process formula Re and the number of foreign objects. Mathematical formula (1) refers to the objective function of the maximum likelihood estimation.

[0061]

Mathematical Formula 1

[0062]

[0063] In the mathematical formula (1) here, the parameter Θ represents the input information, i.e., the number of nodes and outliers, x (x1, x2, ..., xn (n is a positive integer). When n is uncertain but a general case is shown, the suffix n is omitted and it is shown as x to represent the edge. Learning is performed by mimicking a neural network to maximize the likelihood of edge x. Furthermore, in cases where the learning accuracy is low, it is preferable to change the basis function used in the maximum likelihood method. In this case, it is preferable to create an outlier generation chart for each outlier element. That is, by estimating the increase or decrease of the number of outliers for each outlier element, such as the large increase of outliers in edge A for outlier element A, the learning accuracy can be improved.

[0064] Next, in step S6, the foreign object cause analysis unit 215 determines whether the number of foreign objects obtained in step S3 is below a threshold. If the number of generated foreign objects exceeds the threshold (step S6 "No"), the process proceeds to step S7. Alternatively, if the number of foreign objects is less than the threshold (step S6 "Yes"), the flowchart processing ends.

[0065] Next, in step S7, the foreign object cause analysis unit 215 determines the node that is the cause of foreign object generation based on the predicted edge. Step S7 is the step of determining the node that is the cause of foreign object generation based on the predicted edge in step S5, the foreign object element, the component replacement history record Ph, etc. Regarding step S7, since multiple types of foreign object generation causes are predicted, the probability of the predicted certainty is displayed according to the label of each foreign object generation cause. For example, it is expected that the probability of gas ejection pipe is 45%, insufficient cleaning time in the process formula is 35%, and other causes are 20%, etc.

[0066] Next, in step S8, the foreign matter cause analysis unit 215 determines the cause of the foreign matter generation. Step S8 is a conditional branching step based on whether the cause of foreign matter generation determined in S7 is due to insufficient cleaning, process formulation, or damage to the component. Through conditional branching, the foreign matter cause analysis unit 215 performs the component replacement discussion in step S9, the cleaning condition discussion in step S10, and the process formulation discussion in step S11.

[0067] After any of steps S9 to S11 is implemented, based on the results of the discussion, as step S12, countermeasures are implemented in at least one of the process formulation discussion unit 216, cleaning implementation unit 217, and component replacement unit 218. After implementing the countermeasures in step S12, step S1 is executed again, and in step S6, it is verified whether the number of foreign objects is below the threshold.

[0068] Figure 5 This is a diagram illustrating an example of a sequence diagram related to an embodiment of the present invention. Figure 5 Steps S1 to S12 in the middle Figure 4 Steps S1 to S12 in the text correspond to each other respectively. Figure 5 In indicating and Figure 4 Based on common phenomena, and Figure 4 In comparison, additional detailed information was recorded. The following explanation will focus on... Figures 1 to 4 The common explanations are sometimes simplified or omitted.

[0069] In step S1, the processed sample (processed sample) from the semiconductor manufacturing unit 211 is transported to the foreign object characteristic value acquisition unit 212. Alternatively, as a control experiment before step S1, steps S2 to S12 can be performed using a sample not processed in the semiconductor manufacturing unit 211 (pre-processed sample). The pre-processed sample can be a sample that has not undergone all processing of the process formulation, or a sample that has undergone processing of a predetermined process formulation in stages. Therefore, whenever the process formulation is implemented, the foreign object characteristic value Ci can be obtained by the foreign object characteristic value acquisition unit 212 (described later), thus obtaining information related to the increase or decrease in the number of foreign objects corresponding to the edge.

[0070] Next, in step S2, the process recipe Re of the semiconductor manufacturing unit 211 is saved in the device database 213. Furthermore, the information saved in the device database 213 is not limited to the process recipe Re. For example, information related to the historical record of device errors generated in the semiconductor manufacturing unit 211 when the process recipe Re is executed may also be saved in the device database 213.

[0071] Next, in step S3, the foreign matter characteristic value acquisition unit 212 acquires the foreign matter characteristic value Ci for the processed sample (sample Sa) and stores it in the application database 214. Additionally, as an experimental study, the foreign matter characteristic value Ci is also acquired for the sample before processing and stored in the application database 214.

[0072] Next, in step S4, the foreign object cause analysis unit 215 represents the process recipe Re and the historical records of process recipe Re (process history record Rh) stored in the device database 213 as nodes. In addition to process recipe Re and device error history record Eh, the device database 213 also includes component replacement history record Ph and cleaning implementation history record Ch. The foreign object cause analysis unit 215 can also represent nodes based on this information.

[0073] Next, in S5, the foreign object cause analysis unit 215 represents the increase or decrease in the number of foreign objects as edges, and uses a machine learning model to infer the edges using the number of foreign objects and the process history as input. In addition to using the number of foreign objects and the process history Rh, the input to the machine learning model can also include the component replacement history Ph and the cleaning implementation history Ch.

[0074] Next, in steps S6 to S11, the foreign object cause analysis unit 215 determines whether the number of foreign objects obtained in step S3 is below a threshold. If the number of foreign objects is greater than the threshold, the foreign object cause analysis unit 215 determines the side that is the cause of the foreign object generation based on the predicted side, such as discussions on implementing component replacement, discussions on cleaning conditions, and discussions on process formulations.

[0075] Next, as step S12, countermeasures are implemented. For example, when performing countermeasures to clean the conditions, the foreign object cause analysis unit 215 instructs the cleaning implementation unit 217 to change the cleaning conditions. Here, in addition to changing and implementing the cleaning conditions, the cleaning implementation unit 217 also determines whether the components in the semiconductor manufacturing unit 211 need to be replaced. If the replacement of the device component is required, the cleaning implementation unit 217 instructs the device component replacement unit 218 to replace the device component. Furthermore, if cleaning has been performed, the cleaning implementation history Ch is updated; and if the device component has been replaced, the component replacement history Ph is updated.

[0076] Furthermore, if the process formulation has been modified, the foreign matter cause analysis unit 215 instructs the process formulation discussion unit 216 to modify the process formulation. The process formulation discussion unit 216 sends the modified process formulation Rem to the semiconductor manufacturing unit 211. The semiconductor manufacturing unit 211 executes the process based on the modified process formulation Rem.

[0077] (Chart showing foreign object generation)

[0078] Figure 6 This is a diagram illustrating an example of a foreign matter generation graph Gr according to an embodiment of the present invention. Figure 6The graph shows eight nodes connected by directional (arrow) edges. Furthermore, no closed loops are formed in the connected paths, resulting in a directed acyclic graph (DAG) for the foreign matter generation graph Gr. Each node in the DAG graph Gr contains information indicating the cause of foreign matter generation. Examples include information on the type and flow rate of gases recorded in the process recipe Re, and information on the degree of plasma exposure since the replacement of components within the processing vessel, as shown in the component replacement history Ph. All nodes containing this information are connected, and edges are predicted to arrive at the final foreign matter count. Edge prediction uses the maximum likelihood method, learning from the overall distribution of past DAG graph Gr to determine the most probable edge values. As a simple example, in a graph where the past foreign matter count X is represented by nodes A, B, C, and the end time of node C, the increase or decrease in the foreign matter count Y when nodes A, B, and D are considered to be due to the change from node C to node D. This example is very simple, so a neural network is not needed. However, in reality, the number of foreign object generation graphs Gr is proportional to the number of times the etching and foreign object characteristic values ​​are obtained, and the number of nodes to be learned and the number of foreign objects become enormous. Therefore, in this disclosure, the following method is adopted: by using a neural network that utilizes the same method as Bayesian inference, the distribution of each edge is inferred, representing the increase or decrease of the number of edges, i.e., the number of foreign objects, in response to the input.

[0079] Figure 7 This is a diagram representing an example of a dataset input to a neural network. In Figure 7 In the dataset, the item "Time Elapsed Since Component Replacement" indicates the elapsed time since the component replacement point. Additionally, the item "Process Recipe" indicates the type and flow rate of the gas shown in Process Recipe Re. The item "Discharge Time" indicates the time spent discharging, and "Equipment Errors" indicates the number of equipment errors that occurred. Figure 7 The dataset illustrates, for example, three processing steps (data No. 1 to data No. 3) for a single sample. Figure 7 The diagram shows the process recipe used for each data No., with the process recipes contained in data No. 1 to data No. 3 corresponding to the process history record Rh. Similarly, the information on "Elapsed time since component replacement" contained in data No. 1 to data No. 3 corresponds to the component replacement history record Ph. Furthermore, the information on "Equipment error" contained in data No. 1 to data No. 3 corresponds to the equipment error history record Eh.

[0080] in addition, Figure 7The data represents information configured on the nodes, but separately from this data, when creating a graph Gr to represent the relationships between nodes, a matrix representing the relationships between nodes needs to be defined. This matrix representing the relationships between nodes can be controlled using determinants.

[0081] Figure 8 This is a diagram illustrating an example of the graphical user interface (GUI) of the foreign object cause analysis application device 119. Figure 8 The GUI is, for example, the display screen shown on the display device of the foreign object cause analysis application device 119. The GUI includes a display divided into four parts: the upper left ul, the upper right ur, the lower left ll, and the lower right lr.

[0082] The foreign object generation chart Gr is displayed in the upper left corner of the screen (ul). Alternatively, in addition to displaying the overall image of the foreign object generation chart Gr, detailed information about each node can be displayed when the mouse hovers over it, and detailed information about each edge can be displayed when the mouse hovers over it. Furthermore, the detailed information about each node can be edited on the screen, and the predicted changes to the edges based on the edits are also displayed.

[0083] The ur display is shown in the upper right corner of the screen. Figure 3 Step S7 identifies the edge that is the cause of the foreign object generation. As the cause of the foreign object generation, the specific component name, gas name, gas flow rate, etc., are displayed.

[0084] The revised process recipe Re is displayed in the lower left corner of the screen. The revised recipe is displayed by referencing the parameters of the process recipe Re with the lowest foreign object count from the past foreign object generation chart Gr. Here, when actually manufacturing a semiconductor device with parameters changed with an eye toward foreign object count, there is a possibility of shape mismatch. Therefore, it is preferable to also refer to parameters that are close to the target shape in the referenced past foreign object generation chart Gr. The determination of the target shape can be based on the target of the process recipe Re stored in the device database 115.

[0085] The cleaning conditions and recommended replacement parts are displayed in the lower right corner of the screen. When performing dry cleaning methods such as metal cleaning, it is possible that the coating inside the container may peel off during cleaning. To prevent this, the minimum required cleaning formula is displayed. Furthermore, regarding component replacement, since the semiconductor manufacturing apparatus needs to be stopped during component replacement, it is preferable to display and notify the user of the minimum required component replacement.

[0086] Furthermore, while the description describes a scenario where the display screen of the foreign object cause analysis application device 119 is divided into four parts, this disclosure is not limited to this scenario. It is also possible to divide the screen into more than four parts. Additionally, it is possible to display specific information in a magnified manner without dividing the screen. Furthermore, it is also possible to display information contained in the device database 213 and the application database 214, in addition to the aforementioned foreign object generation chart Gr and other information.

[0087] (Function, effect)

[0088] As described above, according to this disclosure, a technology can be provided that can handle foreign matter caused by etching processes without requiring the coating to be processed on components inside the container.

[0089] In semiconductor device manufacturing equipment, when foreign objects are generated, it is easy to determine the phenomenon that is not limited to the component as the cause of the foreign object generation. Therefore, the foreign object can be disposed of in a shorter time and without incurring unnecessary costs than in the past.

[0090] Furthermore, according to this disclosure, all information that could be the cause of foreign object generation is represented as nodes, and the relationships between the edges of these nodes, i.e., the interrelationships between the causes of foreign object generation, can be taken into account. For example, information such as the time since the replacement of a device component and the extent of plasma treatment performed since the component replacement can also be represented as nodes, and the relationship between these nodes and the etching formula for detecting foreign objects can be represented as a graph, enabling high-precision prediction of the relationship between each edge.

[0091] Furthermore, according to this disclosure, it is possible to recommend appropriate information regarding etching processes and other factors that could be the cause of foreign matter generation. For example, it is possible to predict which processes within the etching formulation will generate a large amount of foreign matter and recommend improvements to those predicted processes. Additionally, if corrosion of a component is the cause of foreign matter generation, it is recommended to replace or clean the component.

[0092] The above describes the embodiments of the present invention, but the present invention is not limited to the above embodiments, and various modifications can be made without departing from the spirit of the present invention.

[0093] The following description is intended to be part of the content of this invention, but is not limited thereto.

[0094] (Method 1)

[0095] A semiconductor device manufacturing system includes a platform equipped with an application for determining the cause of foreign matter attached to a sample processed by a semiconductor manufacturing apparatus. In the semiconductor device manufacturing system, the application performs the following steps: creating a graph with process history records as nodes and the difference between the number of foreign matter in one node and the number of foreign matter in the other node as edges; estimating the edges using a neural network that takes the obtained number of foreign matter and the process history records as input; and using the estimated edges to determine the process history records that are the cause of the foreign matter, the process history records including the manufacturing conditions of the semiconductor manufacturing apparatus.

[0096] (Method 2)

[0097] In the semiconductor device manufacturing system described in Method 1, the graph is characterized as a directed acyclic graph.

[0098] (Method 3)

[0099] In the semiconductor device manufacturing system described in Method 1 or Method 2, in the step of estimating the edge, information representing the replacement history of components of the semiconductor manufacturing apparatus, i.e., component replacement history, and information representing the cleaning implementation history performed by the semiconductor manufacturing apparatus, i.e., cleaning implementation history, are further input to estimate the edge.

[0100] (Method 4)

[0101] In any of the semiconductor device manufacturing systems described in Method 1 to Method 3, the manufacturing conditions are modified by the application based on the determined cause of the foreign object's generation.

[0102] (Method 5)

[0103] A server for determining the cause of foreign matter attached to a sample processed by a semiconductor manufacturing apparatus, characterized in that the server performs the following steps: creating a graph with process history as nodes and the difference between the number of foreign matter in the nodes of one connected party and the number of foreign matter in the nodes of the other connected party as edges; estimating the edges by inputting the number of foreign matter and the process history and utilizing the graph's neural network; and using the estimated edges to determine the process history that is the cause of the foreign matter, the process history including the manufacturing conditions of the semiconductor manufacturing apparatus.

[0104] (Method 6)

[0105] A method for determining the cause of foreign matter generation, comprising the steps of: creating a graph with process history records as nodes and the difference between the number of foreign matter in one node and the number of foreign matter in the other node as edges; estimating the edges by inputting the number of foreign matter and the process history records and utilizing the graph's neural network; and using the estimated edges to determine the process history records that constitute the cause of the foreign matter generation, wherein the process history records include the manufacturing conditions of the semiconductor manufacturing apparatus.

[0106] Explanation of reference numerals in the attached figures

[0107] 1: Semiconductor device manufacturing system; 111: Semiconductor manufacturing apparatus; 112: Foreign object characteristic value acquisition device; 113: Personal computer for connecting semiconductor manufacturing apparatus; 114: Personal computer for connecting foreign object characteristic value acquisition device; 115: Apparatus database; 116: Application database; 117: Monitor; 118: Cloud; 119: Foreign object cause analysis application device; 120: Cleaning implementation device; 121: Component replacement device; 122: Process formulation discussion device; 211: Semiconductor manufacturing department; 212: Foreign object characteristic value acquisition department; 213: Apparatus database; 214: Application database; 215: Foreign object cause analysis department; 216: Process formulation discussion department; 217: Cleaning implementation department; 218: Component replacement department; 1190: Bus; 1191: Processor; 1192: Memory; 1193: Storage device; 1194: Input / output device; 1195: Communication interface.

Claims

1. A semiconductor device manufacturing system comprising a platform equipped with an application for determining the cause of foreign matter adhering to a sample processed by a semiconductor manufacturing apparatus. The semiconductor device manufacturing system is characterized in that the application performs the following steps: Create a graph with process history records as nodes and the difference between the number of foreign objects in the nodes of one connected party and the number of foreign objects in the nodes of the other connected party as edges. The edges are inferred by using the number of foreign objects obtained from the input and the process history, and by utilizing a neural network of the graph; and The process history used to determine the cause of the foreign object's formation is based on the presumed edge. The process history includes the manufacturing conditions of the semiconductor manufacturing apparatus.

2. The semiconductor device manufacturing system according to claim 1, wherein, The chart is a directed acyclic chart.

3. The semiconductor device manufacturing system according to claim 1, wherein, In the step of estimating the edge, information representing the replacement history of components of the semiconductor manufacturing apparatus, namely the component replacement history, and information representing the cleaning implementation history performed by the semiconductor manufacturing apparatus, namely the cleaning implementation history, are further input to estimate the edge.

4. The semiconductor device manufacturing system according to claim 1, wherein, Based on the identified causes of foreign matter generation, the application recommends modifications to the manufacturing conditions.

5. A server for performing processing to determine the cause of foreign matter adhering to a sample processed by a semiconductor manufacturing apparatus. The server is characterized by performing the following steps: Create a graph with process history records as nodes and the difference between the number of foreign objects in the nodes of one connected party and the number of foreign objects in the nodes of the other connected party as edges. The edges are inferred by using the number of foreign objects obtained from the input and the process history, and by utilizing a neural network of the graph; and The process history used to determine the cause of the foreign object's formation is based on the presumed edge. The process history includes the manufacturing conditions of the semiconductor manufacturing apparatus.

6. A method for determining the cause of foreign matter generation, comprising determining the cause of foreign matter attached to a sample processed by a semiconductor manufacturing apparatus. The method for determining the cause of foreign object generation is characterized by the following steps: Create a graph with process history records as nodes and the difference between the number of foreign objects in the nodes of one connected party and the number of foreign objects in the nodes of the other connected party as edges. The edges are inferred by using the number of foreign objects obtained from the input and the process history, and by utilizing a neural network of the graph; and The process history used to determine the cause of the foreign object's formation is based on the presumed edge. The process history includes the manufacturing conditions of the semiconductor manufacturing apparatus.

Images