Computer programs, information processing devices, and information processing methods

Abstract

This invention provides a computer program, an information processing apparatus, and an information processing method. The computer performs the following processing: acquiring observation data corresponding to multiple observation variables from an observation system; setting prior constraints based on the causal structure of the variable group in a reference system; and deriving the causal structure of the observation variable group in the observation system by performing learning with the prior constraints applied using the acquired observation data.

Description

Technical Field

[0001] This invention relates to computer programs, information processing apparatus, and information processing methods. Background Technology

[0002] In a substrate processing apparatus, substrates are processed based on a process recipe for each step. The process recipe consists of multiple steps, and optimal processing results can be obtained by controlling various parameters such as pressure and temperature at each step. Since the set values ​​of various parameters may differ from step to step, measurement data from multiple sensors installed in the substrate processing apparatus are managed for each substrate.

[0003] Patent document 1 discloses a technique for visualizing data regarding a process with multidimensional independent and dependent variables.

[0004] Patent Document 1: Japanese Patent Publication No. 2022-502806. Summary of the Invention

[0005] The purpose of this disclosure is to provide computer programs, information processing devices, and information processing methods that can derive the causal structure of a set of observed variables in an observation system based on a causal structure that serves as a benchmark.

[0006] The computer program disclosed herein is used to cause a computer to perform the following processes: acquiring observation data corresponding to multiple observation variables from an observation system, setting prior constraints based on the causal structure of the variable group in the reference system, and deriving the causal structure of the observation variable group in the observation system by performing learning with the prior constraints applied using the acquired observation data.

[0007] According to this disclosure, it is possible to derive the causal structure of the observed variable set in the observation system based on the causal structure that serves as a benchmark. Attached Figure Description

[0008] Figure 1 This is an explanatory diagram illustrating the configuration of the information processing system involved in the implementation method.

[0009] Figure 2 It is a block diagram showing the internal structure of an information processing device.

[0010] Figure 3 It is an explanatory diagram that illustrates the internal manifestations of the causal structure.

[0011] Figure 4 This is an explanatory diagram illustrating an example of setting a confidence interval.

[0012] Figure 5 This is a schematic diagram illustrating an example of a causal structure that contains nonlinearity.

[0013] Figure 6 This is an illustrative diagram used to illustrate an example of a causal structure obtained by additional learning based on a general structure.

[0014] Figure 7 This is a flowchart representing the derivation steps of a causal structure using a general structure.

[0015] Figure 8 This is a schematic diagram illustrating a causal structure obtained through supplementary learning.

[0016] Figure 9 This is a schematic diagram illustrating an example of a causal structure under the condition of generating perturbation factors.

[0017] Figure 10 It is a flowchart representing the steps to infer unknown disturbances. Detailed Implementation

[0018] Hereinafter, one embodiment will be described with reference to the accompanying drawings.

[0019] (Implementation Method 1)

[0020] Figure 1 This is an explanatory diagram illustrating the configuration of the information processing system according to the embodiment. The information processing system according to the embodiment includes an information processing device 100 and a substrate processing device 200 that are communicatively connected.

[0021] The substrate processing apparatus 200 is, for example, a conductor manufacturing apparatus that includes at least one of an exposure apparatus, an etching apparatus, a film forming apparatus, an ion implantation apparatus, an ashing apparatus, and a sputtering apparatus. Alternatively, the substrate processing apparatus 200 may also be a display manufacturing apparatus that manufactures FDP (Flat Display Panel) panels such as liquid crystal display panels and organic EL (Electro-Luminescence) panels.

[0022] In the substrate processing apparatus 200, various preset values ​​are set at the start of the process, such as the substrate temperature, the pressure inside the chamber, the gas flow rate, and the voltage applied by the high-frequency power supply. Furthermore, during process execution, the substrate processing apparatus 200 is equipped with multiple sensors that measure the substrate temperature, the pressure inside the chamber, the gas flow rate, and the voltage applied to the upper and lower electrodes. The substrate processing apparatus 200 outputs the preset values ​​set at the start of the process and the measured values ​​measured during process execution as observation data to the information processing apparatus 100.

[0023] The information processing device 100 acquires observation data corresponding to various observation variables from the baseboard processing device 200 of the observation system. The information processing device 100 sets prior constraints based on the causal structure of a reference system, and derives the causal structure of the observation variable group in the observation system by performing learning with imposed prior constraints using the observation data obtained from the baseboard processing device 200 of the observation system. In this embodiment, the observation system and the reference system are the same type of baseboard processing device 200. For example, the causal structure of the reference system can be a causal structure pre-created based on user insights. Alternatively, the causal structure of the reference system can also be a causal structure created based on observation data obtained from a specific baseboard processing device 200.

[0024] Figure 2 This is a block diagram showing the internal structure of the information processing device 100. The information processing device 100 is, for example, a dedicated or general-purpose computer that includes a control unit 101, a storage unit 102, a communication unit 103, an operation unit 10, and a display unit 105.

[0025] The control unit 101 includes a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). The ROM in the control unit 101 stores control programs for the operation of various hardware components of the information processing device 100. The CPU in the control unit 101 reads and executes the control programs stored in the ROM and the computer programs (described later) stored in the storage unit 102, controlling the operation of the various hardware components, thereby enabling the entire device to function as the information processing device of this disclosure. The RAM in the control unit 101 temporarily stores data used during computation.

[0026] In this embodiment, the control unit 101 is configured with a CPU, ROM, and RAM, but its configuration is not limited to the above. The control unit 101 may, for example, include one or more control or arithmetic circuits such as a GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), quantum processor, or volatile or non-volatile memory. Furthermore, the control unit 101 may also include functions such as a clock for outputting time information, a timer for measuring the elapsed time from the start of measurement to the end of measurement, and a counter for counting quantities.

[0027] Storage unit 102 includes storage devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and EEPROM (Electrically Erasable Programmable Read Only Memory). Storage unit 102 stores various computer programs executed by control unit 101 and various data used by control unit 101.

[0028] The computer program (program product) stored in the storage unit 102 includes a causal structure learning program PG1 for causing a computer to perform processing to derive the causal structure of observed variables from the observation data of the substrate processing device 200. The causal structure learning program PG1 can be a single computer program or composed of multiple computer programs. Furthermore, the causal structure learning program PG1 can also be executed collaboratively by multiple computers. Moreover, the causal structure learning program PG1 can also partially utilize existing libraries.

[0029] The computer program containing the causal structure learning program PG1 is provided by a non-transitory recording medium RM that readablely records the computer program. The recording medium RM is a portable memory such as a CD-ROM, USB memory, SD (Secure Digital) card, micro SD card, or compact flash memory (registered trademark). The control unit 101 uses a reading device (not shown) to read various computer programs from the recording medium RM and stores the read computer programs in the storage unit 102. Alternatively, the computer programs stored in the storage unit 102 can also be provided via communication. In this case, the control unit 101 obtains the computer program through communication via the communication unit 103 and stores the obtained computer program in the storage unit 102.

[0030] The communication unit 103 has a communication interface for receiving and transmitting various data with external devices. The communication interface of the communication unit 103 can use a communication interface conforming to communication standards such as LAN (Local Area Network). External devices include the aforementioned board processing device 200, user terminals (not shown), etc. When the control unit 101 inputs data to be transmitted, the communication unit 103 transmits data to the destination external device; when it receives data transmitted from the external device, it outputs the received data to the control unit 101.

[0031] The operation unit 104 is equipped with operating devices such as a touch panel, keyboard, and switches, and accepts various operations and settings from users. The control unit 101 performs appropriate control based on the various operation information provided by the operation unit 104, and stores the setting information in the storage unit 102 as needed.

[0032] The display unit 105 is equipped with display devices such as a liquid crystal monitor and an organic EL (Electro-Luminescence) display, and displays information that should be reported to the user, etc., according to the instructions from the control unit 101.

[0033] The information processing device 100 in this embodiment can be a single computer or a computer system composed of multiple computers, peripheral devices, etc. Furthermore, the information processing device 100 can also be a virtual machine where the physical entity is virtualized, or it can be a cloud. In this embodiment, the information processing device 100 and the substrate processing device 200 are described as separate structures, but the information processing device 100 can also be disposed inside the substrate processing device 200.

[0034] The causal structure generated by the information processing device 100 will be explained below.

[0035] Figure 3 This is an explanatory diagram used to illustrate the internal representation of a causal structure. A causal structure can be depicted, for example, by a directed acyclic graph (DAG) representing nodes that represent individual observed variables and edges that represent causal relationships between nodes. Figure 3 The upper part represents the causal structure depicted by a directed acyclic graph. Figure 3 The lower part describes its internal performance.

[0036] Causal relationships between observed variables can be modeled using structural equation modeling. In LiNGAM (Linear Non-Gaussian Acyclic Model), one of the structural equation modeling methods, a linear non-cyclic model is used, and it is assumed that the probability distribution of the exogenous variable (error variable) is non-Gaussian.

[0037] As one of the algorithms for exploring structural equation modeling (causal exploration algorithms), the DirectLiNGAM algorithm was proposed (see, for example, S. Shimizu et al. Journal of Machine Learning Research, 12 (Apr): 1225-1248 (2011)). In this algorithm, based on the aforementioned assumptions, the coefficients (edge ​​weights) of the linear regression can be optimized by iteratively performing regression analysis and evaluating the independence of the regression residuals. Edges are then drawn between nodes based on the optimized linear regression coefficients, thus depicting a directed acyclic graph.

[0038] Furthermore, in causal exploration, as long as the edges are not cyclic, there can be cases where edges are drawn from multiple other nodes to a single node. To avoid such overlearning, constraints such as prohibiting the simultaneous drawing of edges from multiple collinear observations to a single observation variable can be added.

[0039] The aforementioned observation variables may also include set values ​​such as the device process formula for the execution of a specified process, measured values ​​from various sensors that accompany the execution of the process, observed or measured values ​​related to the state of the substrate processing apparatus 200 or the consumption of parts, and observed or measured values ​​representing the processing quality of the substrate processing. The information processing apparatus 200 can acquire set values ​​specified by its control device, observed or measured values ​​obtained from various sensors installed in the apparatus or through measuring devices, and observed or measured values ​​obtained from measuring devices installed outside the apparatus, etc., as observation variables. For example, in the actual process of the substrate processing apparatus 200, various set values ​​and observed values ​​such as the coating of the light-transmitting window, OES (Optical Emission Spectrometer), consumption of the lower electrode, consumption of the upper electrode, voltage set value, voltage measurement value of the VI sensor, current measurement value of the VI sensor, and etching amount can be processed as observation variables. Figure 3 In the example, for simplicity, only 6 observed variables are extracted, and only the causal relationships between the extracted 6 observed variables are shown.

[0040] Figure 3 The directed acyclic graph (causal structure) shown in the upper section consists of nodes ND1 to ND6 corresponding to six observed variables (observed variables A to F), and multiple edges EG14, EG24, EG36, EG46, and EG56 representing the causal relationships between the observed variables (between nodes). Figure 3 In the example, nodes ND1 to ND6 are represented by regular octagonal icons, but the shape of the icons is not limited to regular octagons; they can also be circles or other shapes. Figure 3 In the example, the string shown inside the icon represents the variable name of each observed variable.

[0041] exist Figure 3 In the example, edge EG14 connecting node ND1 and node ND4 is drawn. This edge EG14 indicates a causal relationship between the observed variable (observed variable A) corresponding to node ND1 and the observed variable (observed variable D) corresponding to node ND4. Edge EG14 points from node ND1 to node ND4, thus indicating that observed variable A affects observed variable D. The causal relationships between other observed variables are similar.

[0042] Directed acyclic graphs can be represented by matrices. Figure 3 In the above, the directed acyclic loop represents the causal structure described above. Figure 1The matrix elements are displayed graphically. In this graph, the values ​​of the matrix elements indicate the existence of edges: 0 indicates no edge, and values ​​other than 0 indicate an edge. For example, since the matrix element in row 1, column 4 contains a value of 1.0, it indicates that an edge exists from the node of observation variable A at the index in row 1 to the node of observation variable D at the index in column 4. Conversely, since the matrix elements outside row 1, column 4 contain values ​​of 0, it indicates that no edges exist from the node of observation variable A to observation variables B, C, E, and F. Rows 2 through 6 follow the same pattern.

[0043] The values ​​of the matrix elements represent the edge weights (the strength of the bond between nodes), which are calculated as coefficients in linear regression, for example, in the DirectLiNGAM algorithm. Edge weights can also be represented in directed acyclic graphs. For example, the numerical values ​​representing edge weights can be displayed on a graph, and the thickness (line thickness) and color of the edges can be changed based on their weights.

[0044] In addition, such as Figure 3 As the example shows, edge weights can take negative values. A negative edge weight indicates a negative correlation between nodes. That is, a change in the observed variable at the starting point in the positive direction indicates a change in the observed variable at the ending point in the negative direction. Figure 3 In the example, the matrix element in row 3, column 6 contains a negative value such as -1.2. Therefore, when the observed variable C (the observed variable on the starting side) changes in the positive direction, the observed variable F (the observed variable on the ending side) changes in the negative direction. This matrix element in row 3, column 6 represents the edge from the node of observed variable C to the node of observed variable F, not the edge from the node of observed variable F to the node of observed variable C. In this implementation, it should be noted that the direction of the edge will not be reversed regardless of whether the weight of the edge is positive or negative.

[0045] exist Figure 3 In the example, the method of describing a directed acyclic graph (DAG) using a matrix was illustrated, but a DAG can also be described by an edge list obtained by listing the nodes representing the start and end points of the edges and the weights of the edges. In any case, within the information processing device 100, it is described in either list or array form.

[0046] In addition, confidence intervals are set for the weights of the edges. The confidence intervals are set based on the probability distribution of the exogenous variables (error variables) calculated during causal exploration. Figure 4 This is an explanatory diagram illustrating an example of setting a confidence interval. Assuming the edge weights have a probability distribution defined by their center value and width (e.g., a Gaussian distribution), the confidence interval can be set to a range of ±α (α being a value from 0 to ∞) for the original weights. Figure 4In the example, the weight of the edge between the nodes corresponding to the observed variables A and D is 1.0, but a confidence interval of ±0.1 is set for this weight.

[0047] Furthermore, the same applies when assuming the edge weights have a probability distribution similar to the logistic (Sigmoid) distribution; the confidence interval is set based on the probability distribution of the weights. Figure 4 In the example, the weight of the edge between the nodes corresponding to the observed variables C and F is -1.2, but when the weight is set to be at least less than 0 (i.e., it can be learned to take negative values ​​such as -0.5, but not positive values ​​such as +1.0), the confidence interval is set to be less than 0 (recorded as <0 in the figure).

[0048] Furthermore, while the aforementioned causal exploration assumes a linear relationship between observed variables, it is also possible to explore causal structures that include nonlinearity. Although the function form is unknown, algorithms capable of exploring nonlinearity are known, so these algorithms can be used to explore causal structures containing nonlinearity. When depicting causal structures containing nonlinearity, nodes can be added between nodes that become nonlinear to display the function form.

[0049] Figure 5 This is a schematic diagram illustrating an example of a causal structure containing nonlinearity. Figure 5 In the example, this represents a nonlinear relationship between observed variables A, B, and D, where observed variable D = observed variable A × observed variable B. In this case, a node ND7 is added to display the function form A × B, and nodes ND1, ND2, and ND4 are connected to node ND7 via edges EG17, EG27, and EG74, respectively. To distinguish it from the usual nodes and edges between observed variables, the display method can be changed by altering the color of the node displaying the function form or by using dashed lines to show the edges connecting that node.

[0050] In this embodiment, given the causal structure of the reference system, prior constraints are set based on the causal structure of the reference system, and the causal structure of the observation system is derived by performing learning with prior constraints applied using observation data obtained from the observation system.

[0051] Figure 6 This is an illustrative diagram used to illustrate an example of a causal structure learned by adding to a general structure. Figure 6The upper part represents a general structure. A general structure refers to the causal structure of a reference system created through user insights, or the causal structure of a reference system derived from observation data from a specific substrate processing device 200. Where there is a mechanism that allows for understanding user insights without collecting observation data, a causal structure created based on user insights can be used as a general structure. Users can create a general structure by using the operation unit 104 of the information processing device 100 to describe the nodes corresponding to the observed variables, the edges between the observed variables, and assign weights (the strength of the connection between nodes) to the edges as numerical values.

[0052] Furthermore, in the case of multiple substrate processing devices 200 with similar mechanisms, the general causal structure is common among the devices, so the causal structure created in a specific substrate processing device 200 as a representative can be used as a general structure. In this case, the information processing device 100 can create a general structure by collecting observation data from a specific substrate processing device 200 and using the aforementioned causal exploration algorithm.

[0053] Furthermore, the causal structure of the reference system created based on the user's insights can be corrected based on the observation data obtained from the substrate processing device 200 of the reference system, and the corrected causal structure can be used as a general structure. That is, the causal structure created by the user can also be corrected by learning additionally based on the actual observation data obtained from the reference system, using the weights of the edges set by the user as initial values, and the corrected causal structure can be used as a general structure.

[0054] The aforementioned general structure is pre-created, and the parameters of the general structure, including the weights of the edges and the confidence intervals for the weights (the allowable deviation caused by device differences), are stored in the storage unit 102 of the information processing device 100.

[0055] Figure 6 The lower section shows individual structures. Individual structures refer to the various causal structures obtained through supplementary learning using a general structure. By acquiring observation data from each substrate processing device 200 and performing supplementary learning using the general structure, the causal structures (individual structures) of each device are derived. Compared to creating causal structures from a state without prior knowledge, the amount of observation data required can be reduced, thus improving the efficiency of learning.

[0056] In this implementation, when using a general structure for supplemental learning, the weights in individual structures are adjusted by applying a loss corresponding to the confidence interval to avoid significant deviations from the baseline (original weights). Figure 6In the example of the general structure, the weight of the edge between observed variables A and D is 1.0, and its confidence interval is set to ±0.1. Therefore, to apply a larger loss to weights above 1.1 or below 0.9, such as applying an abs(1.0-w) / 0.1 loss, the weight w of the edge in the individual structure can be adjusted. In the example between observed variables A and D, it is shown that the edge weight is 1.0 in the general structure, which is adjusted to 1.03 in the individual structure.

[0057] In addition, Figure 6 In the example of the general structure, the weight of the edge between observed variables C and F is -1.2, and its confidence interval is set to be less than the baseline value (<0). Therefore, for example, to apply a larger loss to weights that are greater than -1.2, the weight of the edge in the individual structure is adjusted. In the example between observed variables C and F, the weight of the edge is shown to be -1.2 in the general structure, which is adjusted to -3.0 in the individual structure.

[0058] The following describes the steps for deriving a causal structure using a general structure.

[0059] Figure 7 This is a flowchart illustrating the steps of deriving the causal structure using a general structure. The control unit 101 of the information processing device 100 performs the following processing by reading the causal structure learning program PG1 from the storage unit 102 and executing it.

[0060] The control unit 101 acquires observation data corresponding to various observation variables from the substrate processing apparatus 200 of the observation system (step S101). The observation data acquired by the control unit 101 includes data measured by the substrate processing apparatus 200, such as the coating of the light-collecting window, OES, consumption of the lower electrode, consumption of the upper electrode, voltage setting value, voltage measurement value of the VI sensor, current measurement value of the VI sensor, and etching amount, as well as data set by the substrate processing apparatus 200. The control unit 101 acquires this observation data by communicating with the substrate processing apparatus 200 via the communication unit 103.

[0061] The control unit 101 reads the parameters of the general structure from the storage unit 102 (step S102). In this embodiment, a causal structure pre-created with respect to the substrate processing apparatus 200 of the same type as the observation system's substrate processing apparatus 200 is used as the general structure. In step S102, the control unit 101 reads the parameters, including the weights of the edges and the confidence intervals for the weights, from the storage unit 102 regarding the general structure.

[0062] The control unit 101 sets prior constraints based on the parameters of the general structure read from the storage unit 102 (step S103). The parameters of the general structure include the weights of the edges and the confidence intervals for the weights, so the control unit 101 can set prior constraints based on these parameters. For example, if the weights of the edges in the general structure are set to w0, the confidence intervals are set to ±α, and the weights of the individual structures to be calculated are set to w, the loss expressed by abs(w0-w) / α can be set as a prior constraint. Furthermore, the loss is not limited to the above formula; any significant difference from the prior distribution of the weights can be set as the loss.

[0063] The control unit 101 uses the observation data acquired in step S101 to perform learning with the prior constraints set in step S103 applied, to derive the causal structure of the observed variable group in the observation system (step S104). The control unit 101 uses a causal exploration algorithm to apply prior constraints to explore the causal relationships between observed variables and derive the overall causal structure of the observed variables. Alternatively, as a prior constraint, the generation of new edges can be allowed, provided that the causal structure maintains directed acyclicity. By performing learning that allows the addition of edges, new insights are obtained under new conditions, or unexpected structural changes are learned and visualized.

[0064] For example, when using the DirectLiNGAM algorithm, the control unit 101 optimizes the coefficients of the linear regression by assuming linearity between observed variables, acyclicity of the causal structure, non-Gaussian probability distribution of exogenous variables, and independence between different exogenous variables, through iterative regression analysis and evaluation of the independence of regression residuals. The control unit 101 generates a directed acyclic graph by drawing edges between nodes based on the optimized coefficients. This yields a causal structure among observed variables that may contain edges originating from multiple nodes with collinearity.

[0065] Alternatively, edge pruning can be performed after deriving the causal structure in step S104. That is, based on the causal structure derived in step S104, the control unit 101 detects edges drawn from multiple other nodes with collinearity to a single node. If such edges exist, to impose a constraint prohibiting the simultaneous drawing of edges from multiple collinear observations to a single observation, the causal exploration learning can continue by designating edges other than the edge with the best accuracy as prohibited edges. By implementing edge pruning, edge misidentification can be prevented, and a more reliable causal structure can be learned.

[0066] The control unit 101 outputs the derived cause-effect structure (step S105). The control unit 101 may also display the derived cause-effect structure on the display unit 105, or notify the user terminal (not shown) through the communication unit 103.

[0067] Figure 8 This is a schematic diagram illustrating an example of a causal structure obtained through appended learning. When appended learning is allowed, there are cases where new edges are added from the general structure. Figure 8 The example shows the use of Figure 3 The causal structure shown is the result of additional learning as a general structure, with the state of edge EG13 added. Additionally, in Figure 8 In the example, edge EG56, where the weight change error exceeds the threshold, is shown by a thick line. Without changes to the equipment or process, it can be assumed that there are no additions, disappearances, or weight changes exceeding the threshold. However, if these changes are observed through supplementary learning, they can be identified as equipment anomalies or changes in the mechanism itself, which can be used for investigating the causes.

[0068] As described above, in Embodiment 1, the causal structure of the observation system can be derived from the observation data obtained by the observation system using the general structure obtained from the substrate processing apparatus 200 of the reference system. In Embodiment 1, since additional learning based on the general structure is performed, the causal structure can be derived from a small amount of observation data obtained from the observation system, thereby improving the efficiency of the learning process.

[0069] (Implementation Method 2)

[0070] In Implementation 2, it is explained that the unknown perturbation factors are inferred based on the learning of new observation data after acquiring the learned causal structure (general structure).

[0071] Furthermore, the overall structure of the system and the internal structure of the information processing device 100 are the same as in Embodiment 1, so their description is omitted.

[0072] Figure 9 This is a schematic diagram illustrating an example of a causal structure under conditions where perturbation factors have occurred. Through causal exploration, we obtain... Figure 9 In the case of the causal structure consisting of nodes ND1 to ND6 as shown, node ND4 is only connected to node ND1 upstream, so the observed variable D of node ND4 should be determined only by the observed variable A of node ND1. Similarly, node ND3 is only connected to node ND2 upstream, so the observed variable C of node ND3 should be determined only by the observed variable B of node ND2.

[0073] However, since the changes in the observed variables of nodes ND4 and ND3, which should be determined solely by the observed variables A and B of nodes ND1 and ND2, are clearly not explainable by them alone, it is inferred that there are variable factors. In cases where variable factors are inferred, the unexplained elements are extracted and defined as perturbation nodes to supplement the causal structure. Figure 9In the example, the state of node NDC is shown as a supplement to the perturbation nodes for nodes ND3 and ND4.

[0074] Furthermore, it can be argued that the causes of collinearity are common, and by combining them, it can be determined that the differences from observed variables A and B to observed variables D and C are due to the same cause. In this way, by visualizing the nodes representing the causes, common effects can be identified, providing insights into tracing the causes.

[0075] Figure 10 This is a flowchart illustrating the steps for inferring unknown disturbances. The control unit 101 of the information processing device 100 derives a causal structure through the same steps as in Embodiment 1, and determines whether there are any variable factors that cannot be explained using the derived causal structure (step S201). For example, after obtaining... Figure 9 In the case of the causal structure shown, if the observed variables A and B are changed, it is understood that the observed variables D and C should change. However, in actual experiments, even if the observed variables A and B are changed, the observed variables D and C do not change, which indicates that there is some variable factor. If it is determined that there is no variable factor (S201: No), the control unit 101 ends the processing of this flowchart.

[0076] If the control unit 101 determines that there is a variable factor that cannot be explained by the derived causal structure (S201: Yes), the variable factor is defined as a disturbance node (step S202).

[0077] The control unit 101 determines whether there are multiple defined disturbance nodes (step S203). If it is determined that there is only one defined disturbance node (S203: No), the control unit 101 supplements the learning with a single disturbance node and generates a causal structure containing a single disturbance node (step S204).

[0078] If it is determined that there are multiple defined disturbance nodes (S203: Yes), the control unit 101 determines whether there is collinearity among the disturbance nodes (step S205). If it is determined that there is collinearity (S205: Yes), the control unit 101 merges the collinear disturbance nodes into one and defines it as a common factor (step S206). The control unit 101 supplements the learning with one or more disturbance nodes to generate a causal structure containing one or more disturbance nodes (step S207).

[0079] If it is determined in step S205 that there is no collinearity among the disturbance nodes (S205: No), the control unit 101 supplements multiple disturbance nodes for learning and generates a causal structure containing multiple disturbance nodes (step S208).

[0080] As described above, in Implementation 2, by visualizing the perturbation nodes, common effects can be identified, and users can be provided with insights into investigating the causes.

[0081] The matters described in each embodiment can be combined with each other. Furthermore, the independent and dependent claims described in the claims can be combined with each other in all possible combinations, regardless of their referencing form. Moreover, the claims do not use the form of referring to two or more other claims (multiple reference form), but are not limited to this. Multiple reference form or multiple claims referring to at least one multiple claim (multiple reference to multiple) may also be used.

[0082] The embodiments disclosed herein should be considered illustrative rather than limiting in all respects. The scope of the invention is set forth in the claims, not in the foregoing sense, and is intended to include all modifications within the meaning and scope of the claims.

[0083] For example, in this embodiment, the substrate processing apparatus 200 is described as both a reference system and an observation system. The observation system for the monitored object is not limited to the substrate processing apparatus 200, but can also be a manufacturing apparatus that performs any manufacturing process for electrical equipment, chemical products, pharmaceuticals, food, or other chemical products. Furthermore, the reference system and observation system are not limited to apparatuses or systems that perform a specific manufacturing process, but can also be any system that appropriately combines human living environments, economic activities, meteorological environments, etc.

[0084] Explanation of reference numerals in the attached figures

[0085] 100 Information processing device; 101 Control unit; 102 Storage unit; 103 Communication unit; 104 Operation unit; 105 Display unit; PG1 Causal structure learning program; RM Recording medium; 200 Board processing device.

Claims

1. A computer program for causing a computer to perform the following processes: Obtain observation data corresponding to multiple observation variables from the observation system; Prior constraints are set based on the causal structure of the variable group in the reference system; By performing learning with the aforementioned prior constraints using the acquired observation data, the causal structure of the observation variable set in the observation system is derived.

2. The computer program according to claim 1, wherein, The causal structure of the reference system is a pre-created causal structure based on the user's insights. The computer program is used to cause the computer to perform the following processes: Obtain the parameters of the causal structure; The prior constraints are set based on the obtained parameters.

3. The computer program according to claim 1, wherein, The causal structure of the reference system is a causal structure derived in advance based on the observation data obtained from observing the reference system. The computer program is used to cause the computer to perform the following processes: Obtain the parameters of the causal structure; The prior constraints are set based on the obtained parameters.

4. The computer program according to claim 1, wherein, The causal structure of the reference system is a causal structure created based on the user's insights. The computer program is used to cause the computer to perform the following processes: Obtain information about the causal structure and observe the data obtained from the reference system; The causal structure is corrected based on the acquired observation data; The prior constraints are set based on the parameters of the modified causal structure.

5. The computer program according to claim 1, wherein, The computer program is used to cause the computer to perform the following processes: Using nodes representing each observed variable and edges representing causal relationships between nodes, a directed acyclic graph representing the causal structure of the group of observed variables is generated. Output the generated directed acyclic graph.

6. The computer program according to claim 5, wherein, The computer program is used to cause the computer to perform the following processes: In cases where the causal structure of the observation system contains nonlinearity, nodes for displaying functional forms are added between the nodes that become nonlinear.

7. The computer program according to claim 5, wherein, The computer program is used to cause the computer to perform the following processes: The loss corresponding to the confidence interval for the weights between the nodes is set as the prior constraint.

8. The computer program according to claim 7, wherein, The confidence interval is set as a numerical range that reflects the probability distribution of the weights.

9. The computer program according to claim 5, wherein, The prior constraints, conditional on maintaining the directed acyclicity of the causal structure, allow the generation of new edges.

10. The computer program according to claim 5, wherein, The computer program is used to cause the computer to perform the following processes: Compare the weights of the edges in the causal structure of the reference system with the weights of the edges in the causal structure of the observation system derived through the learning process; If the weights of the edges change by more than the error through the learning process, it is determined that an anomaly has been detected in the observation system.

11. The computer program according to claim 1, wherein, The computer program is used to cause the computer to perform the following processes: Based on the observation data of the observation system, infer the perturbation factors not included in the causal structure of the reference system.

12. The computer program according to claim 1, wherein, The reference system is a first substrate processing device. The observation system is a second substrate processing device of the same type as the first substrate processing device.

13. An information processing apparatus, wherein, Equipped with at least one processor The processor acquires observation data corresponding to multiple observation variables from the observation system, sets prior constraints based on the causal structure of the variable group in the reference system, and derives the causal structure of the observation variable group in the observation system by performing learning with the prior constraints applied using the acquired observation data.

14. An information processing method, wherein a computer performs the following processing: Obtain observation data corresponding to multiple observation variables from the observation system; Prior constraints are set based on the causal structure of the variable group in the reference system; By performing learning with the aforementioned prior constraints using the acquired observation data, the causal structure of the observation variable set in the observation system is derived.

Images