WO2022102439A1 - モデル管理システム、モデル管理方法及びモデル管理プログラム - Google Patents
モデル管理システム、モデル管理方法及びモデル管理プログラム Download PDFInfo
- Publication number
- WO2022102439A1 WO2022102439A1 PCT/JP2021/040093 JP2021040093W WO2022102439A1 WO 2022102439 A1 WO2022102439 A1 WO 2022102439A1 JP 2021040093 W JP2021040093 W JP 2021040093W WO 2022102439 A1 WO2022102439 A1 WO 2022102439A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- management
- management device
- parameter
- layer
- Prior art date
Links
- 238000007726 management method Methods 0.000 title claims abstract description 473
- 239000000758 substrate Substances 0.000 claims abstract description 42
- 238000004519 manufacturing process Methods 0.000 claims abstract description 19
- 238000004364 calculation method Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims description 113
- 238000000034 method Methods 0.000 claims description 85
- 230000005540 biological transmission Effects 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 7
- 230000006866 deterioration Effects 0.000 claims description 7
- 238000003745 diagnosis Methods 0.000 claims description 7
- 230000005856 abnormality Effects 0.000 claims description 6
- 239000010410 layer Substances 0.000 description 318
- 238000003860 storage Methods 0.000 description 43
- 238000005259 measurement Methods 0.000 description 29
- 238000010586 diagram Methods 0.000 description 27
- 238000013500 data storage Methods 0.000 description 23
- 238000010438 heat treatment Methods 0.000 description 20
- 238000005530 etching Methods 0.000 description 18
- 238000009826 distribution Methods 0.000 description 10
- 230000015654 memory Effects 0.000 description 7
- 238000005457 optimization Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000004088 simulation Methods 0.000 description 6
- 238000004886 process control Methods 0.000 description 5
- 239000002184 metal Substances 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000002068 genetic effect Effects 0.000 description 3
- 239000011229 interlayer Substances 0.000 description 3
- 238000005121 nitriding Methods 0.000 description 3
- 229910021420 polycrystalline silicon Inorganic materials 0.000 description 3
- 238000012369 In process control Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010965 in-process control Methods 0.000 description 2
- 238000012628 principal component regression Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B17/00—Systems involving the use of models or simulators of said systems
- G05B17/02—Systems involving the use of models or simulators of said systems electric
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/04—Programme control other than numerical control, i.e. in sequence controllers or logic controllers
- G05B19/042—Programme control other than numerical control, i.e. in sequence controllers or logic controllers using digital processors
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0208—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the configuration of the monitoring system
- G05B23/0213—Modular or universal configuration of the monitoring system, e.g. monitoring system having modules that may be combined to build monitoring program; monitoring system that can be applied to legacy systems; adaptable monitoring system; using different communication protocols
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0262—Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0283—Predictive maintenance, e.g. involving the monitoring of a system and, based on the monitoring results, taking decisions on the maintenance schedule of the monitored system; Estimating remaining useful life [RUL]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Definitions
- This disclosure relates to a model management system, a model management method, and a model management program.
- the present disclosure provides a model management system, a model management method, and a model management program that efficiently manage the models applied to the substrate manufacturing process.
- the model management system has, for example, the following configuration. That is, It is a model management system that manages the model applied to the board manufacturing process by dividing it into three or more layers.
- the first management unit that manages the model in a predetermined hierarchy, It has one or more second management units that manage the model in a layer one level lower than the first management unit.
- the first management unit is A calculation unit that calculates new model parameters based on each updated model parameter when the model parameters of the model managed by the second management unit are updated.
- the model managed by each of the plurality of lowermost management units belonging to the first management unit has a control unit that controls the setting of the new model parameters.
- FIG. 1 is a diagram showing an example of a system configuration of a model management system in the first phase.
- FIG. 2 is a diagram showing an example of the system configuration of the model management system in the second phase.
- FIG. 3 is a diagram showing an example of the system configuration of the model management system in the third phase.
- FIG. 4 is a diagram showing an example of the system configuration of the model management system in the fourth phase.
- FIG. 5 is a diagram showing an example of the system configuration of the model management system in the fifth phase.
- FIG. 6 is a diagram showing an example of the system configuration of the model management system in the sixth phase.
- FIG. 7 is a diagram showing an example of the system configuration of the model management system in the seventh phase.
- FIG. 8 is a sequence diagram showing the flow of model management processing by the model management system.
- FIG. 9 is a diagram showing an example of the hardware configuration of the x-layer management device.
- FIG. 10 is a diagram showing an example of the functional configuration of the x-layer management device.
- FIG. 11 is a flowchart showing the flow of management processing by the x-layer management device.
- FIG. 12 is a diagram showing an example of the hardware configuration of the execution device.
- FIG. 13 is a diagram showing an example of the functional configuration of the execution device.
- FIG. 14A is a flowchart showing the flow of sequential learning processing by the execution device.
- FIG. 14B is a flowchart showing the flow of batch learning processing by the execution device.
- FIG. 15 is a diagram showing an example of a system configuration of a model management system.
- the model management system according to the first embodiment has a plurality of management devices, and each management device manages a model (model structure and model parameters) in each layer.
- the model managed by each management device may have one type or a plurality of types of model structures (in the first embodiment, a case where each management device manages a model having two types of model structures will be described. do).
- a new model parameter is generated to be a management device of the same layer. Model parameters can be shared between them.
- the types of model parameters that is, the number of models
- the management cost is reduced.
- the management cost cost required for learning
- FIG. 1 is a diagram showing an example of the system configuration of the model management system in the first phase.
- the model management system 100 includes a model management device 110, a learning data storage unit 111, and learning units 112_1 and 112_2.
- model management system 100 is an example of a management unit, that is, the first layer management device 120, the second layer management devices 131 to 133, ...
- model management system 100 includes a board processing device 160, an execution device 170, a board processing device 180, and an execution device 190.
- the first purpose referred to here refers to, for example, an abnormality detection of a board processing device, inference of a product of process processing, deterioration diagnosis of parts, inference of an internal state, and the like.
- the inference of the result of the process processing includes inference of the film formation rate, inference of the etching rate, inference of the shape of the result, and the like.
- the second purpose referred to here refers to, for example, a purpose such as process control.
- model management device 110 and the first layer management device 120 are connected to different networks connected via a firewall (not shown). Further, it is assumed that the first layer management devices 120 to the nth layer management devices 151 and 152 are connected to the same local network in the same board processing factory.
- the second layer management devices 131, 132, 133 to the nth layer management devices 151, 152 a plurality of layered groups of models according to each purpose of each board processing device in the same board processing factory. It is managed separately.
- the second-tier management devices 131, 132, 133 manage the models corresponding to each purpose of each board processing device in the same board processing factory by dividing them into groups of board processing devices having the same "device model”. do. Specifically, when the substrate processing device is a heat treatment device, for example, -The second layer management device 131 manages the model whose device model corresponds to the group of "heat treatment device having an oxidation furnace". -The second layer management device 132 manages a model in which the device model corresponds to the group of "heat treatment devices having a nitriding furnace". The second-level management device 133 manages a model in which the device model corresponds to the group of "heat treatment devices having a metal furnace”.
- the second layer management device 131 manages the model in which the device model corresponds to the group of "etching device that etches the oxide film”.
- the second layer management device 132 manages a model in which the device model corresponds to the group of "etching device for etching Poly-Si film”.
- the third layer management device is a model corresponding to each group grouped by the second layer management devices 131 to 133 among the models corresponding to each purpose of each board processing device in the same board processing factory. Are managed by dividing them into groups of substrate processing devices having the same "chamber type". Specifically, when the substrate processing device is a heat treatment device, for example, -For the third layer management device, among the models whose device model corresponds to the group of "heat treatment devices having a nitriding furnace", the model corresponding to the heat treatment device group whose chamber type is "chamber for processing the interlayer insulating film".
- Manage and -The other third-tier management equipment corresponds to the group of heat treatment equipment whose chamber type is "chamber for processing the gate insulating film" among the models whose equipment model corresponds to the group of "heat treatment equipment having a nitriding furnace".
- Manage the model and -The other third layer management device corresponds to the group of heat treatment devices whose chamber type is "chamber for processing the gate electrode layer” among the models whose device model corresponds to the group of "heat treatment devices having a metal furnace”.
- Manage the model and -The other third-tier management equipment is a model corresponding to the group of heat treatment equipment whose chamber type is "chamber for processing a metal film” among the models whose equipment model corresponds to the group of "heat treatment equipment having a metal furnace”.
- the substrate processing device is an etching device
- -The third layer management device corresponds to the group of etching devices whose chamber type is "chamber that opens a contact hole" among the models whose device model corresponds to the group of "etching devices that etch Poly-Si film”.
- Manage the model and -The other third-tier management equipment is a group of etching equipment whose chamber type is "chamber for wiring trench processing" among the models whose equipment model corresponds to the group of "etching equipment for etching Poly-Si film”.
- the first layer management device 120 manages the model transmitted from the model management device 110. Further, in the first phase, the first layer management device 120 transmits the model transmitted from the model management device 110 to the second layer management devices 131 to 133.
- the case where the number of the second layer management devices connected to the first layer management device 120 is three is shown, but the case of the second layer management device connected to the first layer management device 120 is shown.
- the number is not limited to three.
- the second-tier management devices 131 to 133 manage the model transmitted from the first-tier management device 120, which is the connection destination.
- the model sent to each may be different.
- the second layer management devices 131 to 133 each transmit the model transmitted from the first layer management device 120 to the third layer management device (not shown) in the first phase.
- the first (n-1) hierarchy management device 141 manages a model transmitted from the connection destination (n-2) hierarchy management device. Further, in the first phase, the first (n-1) layer management device 141 transmits the model transmitted from the (n-2) layer management device to the nth layer management devices 151 and 152.
- the nth layer management device connected to the (n-1) layer management device 141 is two is shown, but it is connected to the (n-1) layer management device 141.
- the number of the nth layer management devices to be performed is not limited to two.
- the nth layer management devices 151 and 152 manage the model transmitted from the connection destination (n-1) layer management device. Further, the nth layer management devices 151 and 152 transmit the model transmitted from the (n-1) layer management device 141 to the execution devices 170 and 190 in the first phase.
- the substrate processing device 160 is a processing device that executes a substrate manufacturing process in a physical space, such as a heat treatment device or an etching device.
- the inference result output from the inference unit 171_2 is stored in the actual measurement data storage unit 172 as actual measurement data in association with the correct answer data, for example.
- the substrate processing apparatus 180 is a processing apparatus that executes a substrate manufacturing process in a physical space, such as a heat treatment apparatus and an etching apparatus.
- FIG. 2 is a diagram showing an example of the system configuration of the model management system in the second phase.
- the execution device 170 in addition to the system configuration of the first phase, the execution device 170 further includes learning units 210_1 and 210_2. Further, the execution device 190 has learning units 220_1 and 220_1.
- FIG. 3 is a diagram showing an example of the system configuration of the model management system in the third phase.
- FIG. 4 is a diagram showing an example of the system configuration of the model management system in the fourth phase.
- FIG. 5 is a diagram showing an example of the system configuration of the model management system in the fifth phase.
- -Model parameters "parameter PA 2”, “parameter P B 2 " transmitted from the nth layer management device 152 in the fourth phase
- new model parameters "parameter P A 3" and "parameter P B 3" are calculated.
- the method of calculating the new model parameter by the first (n-1) layer management device 141 is arbitrary.
- the weight in the case of weighting addition is arbitrary.
- the weight may be set to zero for the parameters that do not meet the predetermined evaluation criteria (may be excluded from the addition target). ..
- FIG. 6 is a diagram showing an example of the system configuration of the model management system in the sixth phase.
- the second (n-2) layer management device will be described as the second layer management device 132.
- the other (n-1) layer management device connected to the second layer management device 132 transmits new model parameters to the second layer management device 132 at a predetermined timing. do.
- the other (n-1) layer management device connected to the second layer management device 131, 133 sets new model parameters at a predetermined timing. It is transmitted to the second layer management devices 131 and 132, respectively.
- FIG. 7 is a diagram showing an example of the system configuration of the model management system in the seventh phase.
- the second layer management device 132 is -In the sixth phase
- the model parameters "parameter PA 3" and "parameter P B 3 " transmitted from the (n-1) hierarchy management device 141
- -In the sixth phase the model parameters transmitted from a plurality of other (n-1) layer management devices connected to the second layer management device 132
- new model parameters "parameter P A 6" and "parameter P B 6" are calculated.
- up to the seventh phase has been described, but further, up to the eighth phase may be executed.
- the models managed by the lowest layer nth layer management devices 151 and 152 belonging to the first layer management device 120 are shared. Further, in the above description, the case where the first to eighth phases are executed once has been described. However, the first to eighth phases may be repeated a plurality of times.
- each phase is not limited to the execution order described above.
- the first to fifth phases may be repeatedly executed a plurality of times, and then the sixth to seventh phases may be executed.
- the eighth phase may be executed after the sixth to seventh phases are repeatedly executed a plurality of times.
- FIG. 8 is a sequence diagram showing the flow of model management processing by the model management system. In executing the model management process shown in FIG. 8, it is assumed that the learning process by the learning units 112_1 and 112_2 has been completed, and the model structure and the model parameters are managed by the model management device 110.
- step S801 the model management device 110 transmits the managed model structure and model parameters to the first layer management device 120.
- the model structure and model parameters transmitted to the first layer management device 120 are sequentially transmitted to the lower layer management device belonging to the first layer management device 120, for example, the first (n-1) layer management. It is assumed that the signal is transmitted to the device 141.
- step S811 the (n-1) hierarchy management device 141 transmits the transmitted model structure and model parameters to the nth hierarchy management device 151.
- step S812 the nth layer management device 151 transmits the transmitted model structure and model parameters to the execution device 170.
- step S813 the execution device 170 sets the transmitted model parameters in the transmitted model structure and executes the model. Further, the execution device 170 associates the inference result output from the model with the correct answer data and stores it as actual measurement data, and performs learning processing using the stored actual measurement data.
- step S814 the execution device 170 transmits the model parameters optimized by performing the learning process using the actually measured data to the nth layer management device 151.
- step S821 the (n-1) layer management device 141 transmits the transmitted model structure and model parameters to the nth layer management device 152.
- step S822 the nth layer management device 152 transmits the transmitted model structure and model parameters to the execution device 190.
- step S823 the execution device 190 sets the transmitted model parameters in the transmitted model structure and executes the model. Further, the execution device 190 stores the inference result output from the model and the correct answer data in association with each other as actual measurement data, and performs learning processing using the stored actual measurement data.
- step S824 the execution device 190 transmits the model parameters optimized by performing the learning process using the actually measured data to the nth layer management device 152.
- step S831 the nth layer management device 151 transmits the optimized model parameters to the (n-1) layer management device 141.
- step S832 the nth layer management device 152 transmits the optimized model parameter to the (n-1) layer management device 141.
- step S833 the (n-1) th-tier management device 141 is a new model based on the model parameters transmitted from the n-th tier management device 151 and the model parameters transmitted from the n-th tier management device 152. Generate parameters.
- step S834 the (n-1) layer management device 141 transmits a new model parameter to the nth layer management device 151. After that, the same processing as in steps S812 to S814 and S831 is performed.
- step S835 the (n-1) layer management device 141 transmits a new model parameter to the nth layer management device 152. After that, the same processing as in steps S822 to S824 and S832 is performed.
- steps S831 to S835 is repeated in the layer higher than the first (n-1) layer management device 141.
- each of the plurality of second-tier management devices transmits the optimized model parameters to the first-tier management device 120, which is the connection destination.
- step S842 the first-tier management device 120 generates new model parameters using the model parameters transmitted from a plurality of second-tier management devices one-tier lower than the first-tier management device 120.
- the new model parameters generated by the first layer management device 120 are transmitted to a plurality of second layer management devices one level lower connected to the first layer management device 120.
- FIG. 9 is a diagram showing an example of the hardware configuration of the x-layer management device.
- the x-layer management device 900 has a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, and a RAM (Random Access Memory) 903. Further, the xth layer management device 900 has a GPU (Graphics Processing Unit) 904. A processor (processing circuit, Processing Circuit, Processing Circuitry) such as CPU 901 and GPU 904 and a memory such as ROM 902 and RAM 903 form a so-called computer.
- CPU Central Processing Unit
- ROM Read Only Memory
- RAM Random Access Memory
- the xth layer management device 900 has a GPU (Graphics Processing Unit) 904.
- a processor processing circuit, Processing Circuit, Processing Circuitry
- CPU 901 and GPU 904 and a memory such as ROM 902 and RAM 903 form a so-called computer.
- the x-layer management device 900 has an auxiliary storage device 905, a display device 906, an operation device 907, an I / F (Interface) device 908, and a drive device 909.
- the hardware of the x-layer management device 900 is connected to each other via the bus 910.
- the CPU 901 is an arithmetic device that executes various programs (for example, a model management program described later) installed in the auxiliary storage device 905.
- ROM 902 is a non-volatile memory and functions as a main storage device.
- the ROM 902 stores various programs, data, and the like necessary for the CPU 901 to execute various programs installed in the auxiliary storage device 905.
- the ROM 902 stores boot programs such as BIOS (Basic Input / Output System) and EFI (Extensible Firmware Interface).
- the RAM 903 is a volatile memory such as a DRAM (Dynamic Random Access Memory) or a SRAM (Static Random Access Memory), and functions as a main storage device.
- the RAM 903 provides a work area that is expanded when various programs installed in the auxiliary storage device 905 are executed by the CPU 901.
- the GPU 904 is an arithmetic device for image processing, and in the present embodiment, when the model management program is executed by the CPU 901, high-speed arithmetic is performed by parallel processing on the two-dimensional data.
- the GPU 904 is equipped with an internal memory (GPU memory), and temporarily holds information necessary for performing parallel processing on two-dimensional data.
- the auxiliary storage device 905 stores various programs, various data used when various programs are executed by the CPU 901, and the like.
- the distribution information storage unit 1011, the update information storage unit 1012, and the lower layer information storage unit 1013, which will be described later, are realized in the auxiliary storage device 905.
- the display device 906 is, for example, a display device that displays the internal state of the x-layer management device 900.
- the operation device 907 is an input device used by the user of the x-layer management device 900 to input various instructions to the x-layer management device 900.
- the I / F device 908 is a connection device for transmitting / receiving data to / from another management device.
- the drive device 909 is a device for setting the recording medium 920.
- the recording medium 920 referred to here includes a medium such as a CD-ROM, a flexible disk, a magneto-optical disk, or the like, which records information optically, electrically, or magnetically. Further, the recording medium 920 may include a semiconductor memory or the like for electrically recording information such as a ROM or a flash memory.
- the various programs installed in the auxiliary storage device 905 are installed, for example, by setting the distributed recording medium 920 in the drive device 909 and reading the various programs recorded in the recording medium 920 by the drive device 909. Will be done.
- various programs installed in the auxiliary storage device 905 may be installed by being downloaded via a network (not shown).
- FIG. 10 is a diagram showing an example of the functional configuration of the x-layer management device.
- the model management program is installed in the x-layer management device 900, and when the program is executed, the x-layer management device 900 is the upper layer information acquisition unit 1001 and the sharing unit. Functions as 1002. Further, the xth layer management device 900 functions as a lower layer information acquisition unit 1003, an optimization unit 1004, and an upper layer information transmission unit 1005.
- the upper layer information acquisition unit 1001 is a management device of the upper layer one layer, and when both the model structure and the model parameters or the model parameters are transmitted from the management device connected to the xth layer management device 900. To get this. Further, the upper layer information acquisition unit 1001 stores the acquired model structure and model parameters in the distribution information storage unit 1011. Alternatively, the upper layer information acquisition unit 1001 updates the model parameters already stored in the distribution information storage unit 1011 by using the acquired model parameters.
- the sharing unit 1002 is an example of a control unit, and when a model and model parameters are newly stored in the distribution information storage unit 1011, the sharing unit 1002 is read out and is a management device in the lower layer of the first layer. It is transmitted to the management device connected to the hierarchy management device 900. Further, the sharing unit 1002 reads out the model parameters stored in the distribution information storage unit 1011 when the model parameters are updated, and is a management device of a layer one level lower than the layer x layer management device 900. Send to the management device to be connected.
- the lower layer information acquisition unit 1003 is a management device in the lower layer of the first layer, and acquires the model parameters when the model parameters are transmitted from the management device connected to the xth layer management device 900. Further, the lower layer information acquisition unit 1003 stores the acquired model parameters in the lower layer information storage unit 1013.
- the optimization unit 1004 is an example of a calculation unit, and is notified by the lower layer information acquisition unit 1003, reads out the model parameters stored in the lower layer information storage unit 1013, and generates new model parameters.
- the method of calculating the new model parameter is arbitrary, and for example, a new model parameter may be generated by weighting and adding a plurality of model parameters read from the lower layer information storage unit 1013.
- the conditions for the optimization unit 1004 to generate new model parameters are arbitrary, and new model parameters may be generated when any conditions are satisfied. For example, it may be generated on the condition that a predetermined period has elapsed, or may be generated on the condition that a plurality of model parameters of a predetermined number or more are stored.
- the plurality of model parameters refer to the model parameters transmitted from each of the plurality of management devices in the lower layer of the first layer and the plurality of management devices connected to the xth layer management device 900.
- the optimization unit 1004 updates the model parameters already stored in the distribution information storage unit 1011 by using the generated new model parameters. Further, the optimization unit 1004 stores the generated new model parameter in the update information storage unit 1012.
- the upper layer information transmission unit 1005 is an example of a transmission unit, and when a new model parameter is stored in the update information storage unit 1012, it is read out and is a management device of the upper layer one layer at a predetermined timing. Then, it is transmitted to the management device to which the x-layer management device 900 is connected.
- FIG. 11 is a flowchart showing the flow of management processing by the x-layer management device.
- step S1101 the upper layer information acquisition unit 1001 of the xth layer management device 900 acquires the model structure and the model parameters from the management device of the upper layer one layer to which the xth layer management device 900 is connected, and distributes the information. It is stored in the storage unit 1011.
- step S1102 the sharing unit 1002 of the x-layer management device 900 reads out the model structure and the model parameters from the distribution information storage unit 1011 and is a management device of a layer one level lower than the distribution information storage unit 1011. Send to the management device to be connected.
- the lower layer information acquisition unit 1003 of the xth layer management device 900 is a management device of the lower layer of the first layer, and the model parameter is different from any of the management devices connected to the xth layer management device 900. Determine if it was sent.
- step S1103 If it is determined in step S1103 that the model parameter has been transmitted (YES in step S1103), the process proceeds to step S1104.
- step S1104 the lower layer information acquisition unit 1003 of the xth layer management device 900 stores the transmitted model parameter in the lower layer information storage unit 1013.
- the optimization unit 1004 reads the model parameters from the lower layer information storage unit 1013, generates new model parameters, and stores the model parameters already stored in the distribution information storage unit 1011 and the update information storage unit 1012, respectively. Update.
- the sharing unit 1002 transmits the updated new model parameter to the management device of the layer one level lower and connected to the xth layer management device 900.
- the upper layer information transmission unit 1005 transmits the updated new model parameter to the management device of the upper layer of the first layer and to the management device to which the xth layer management device 900 is connected. Then, the process proceeds to step S1105.
- step S1103 determines whether the model parameter has been transmitted (NO in step S1103). If it is determined in step S1103 that the model parameter has not been transmitted (NO in step S1103), the process directly proceeds to step S1105.
- step S1105 the upper layer information acquisition unit 1001 of the xth layer management device 900 is a management device of the upper layer of the first layer, and the model parameter is transmitted from the management device connected to the xth layer management device 900. Judge whether or not.
- step S1105 If it is determined in step S1105 that the model parameter has been transmitted (YES in step S1105), the process proceeds to step S1106.
- step S1106 the upper layer information acquisition unit 1001 of the xth layer management device 900 acquires the model parameters transmitted from the management device of the upper layer one layer to which the xth layer management device 900 is connected. Further, the upper layer information acquisition unit 1001 of the xth layer management device 900 updates the model parameters already stored in the distribution information storage unit 1011 by using the acquired model parameters. Further, the sharing unit 1002 transmits the updated new model parameter to the management device connected to the xth layer management device 900, which is the management device of the next lower layer, and proceeds to step S1107.
- step S1105 determines whether the model parameter has been transmitted (NO in step S1105). If it is determined in step S1105 that the model parameter has not been transmitted (NO in step S1105), the process directly proceeds to step S1107.
- step S1107 the x-layer management device 900 determines whether or not to end the management process. If it is determined in step S1107 that the management process is not completed (NO in step S1107), the process returns to step S1103.
- step S1107 determines whether the management process is to be completed (YES in step S1107). If it is determined in step S1107 that the management process is to be completed (YES in step S1107), the management process is terminated.
- FIG. 12 is a diagram showing an example of the hardware configuration of the execution device. Since the hardware configurations of the execution devices 170 and 190 are substantially the same as the hardware configurations of the x-layer management device 900, only the differences will be described here.
- the CPU 1201 is an arithmetic device that executes various programs (for example, an execution program described later) installed in the auxiliary storage device 1205.
- the auxiliary storage device 1205 stores various programs and various data used when various programs are executed by the CPU 1201.
- the model information storage units 1321 and 1323 and the actual measurement data storage units 172-1 and 172-2, which will be described later, are realized in the auxiliary storage device 1205.
- the hardware configuration shown in FIG. 12 is an example, and the execution devices 170 and 190 may be configured by, for example, an FPGA (Field-programmable gate array). Alternatively, the execution devices 170 and 190 may be configured by, for example, an AI chip (semiconductor specialized in AI (Artificial Intelligence) processing).
- FIG. 13 is a diagram showing an example of the functional configuration of the execution device.
- the execution device 170 has a virtual space (for real-time control, simulation) that processes data collected from the physical space and executes or updates a model. Further, the execution device 170 has a storage group for storing data collected from the physical space, model parameters updated in the virtual space, and the like.
- ⁇ Board processing device 160 In the physical space where the execution device 170 collects data, ⁇ Board processing device 160 and A measuring device 1301 for measuring data used by the inference unit 171_1 to execute the model or the sequential learning unit 1311 to perform the sequential learning of the model during the substrate processing by the substrate processing apparatus 160.
- the virtual space (for real-time control) of the execution device 170 includes a sequential learning unit 1311 and 1313, an instruction unit 1312, a notification unit 1314, and an inference unit 171_1 and 171_2.
- the sequential learning unit 1311 performs sequential learning (online learning) based on the data collected from the measuring device 1301 and adjusts the model parameters of the model arranged in the inference unit 171_1.
- the model parameters adjusted by the sequential learning unit 1311 may be stored in the model information storage unit 1321.
- the inference unit 171_1 executes a model based on the data collected from the measuring device 1301, and inference results (for example, anomaly detection inference result, process processing result inference result, component). Outputs the inference result of deterioration diagnosis, inference result of internal state, etc.).
- the inference result output from the inference unit 171_1 is notified to the instruction unit 1312, and is stored in the actual measurement data storage unit 172_1 in association with the correct answer data, for example.
- the instruction unit 1312 gives an instruction according to the inference result notified from the inference unit 171-1 (for example, the inference result of abnormality detection, the inference result of the result of process processing, the inference result of component deterioration diagnosis, the inference result of the internal state, etc.). Generate (for example, the target value of the process). Further, the instruction unit 1312 transmits the generated instruction to the substrate processing device 160. In this case, the substrate processing apparatus 160 calculates the control value based on the transmitted instruction (for example, the target value of the process) and controls the control.
- the instruction unit 1312 may be realized by a model having a learning function.
- the sequential learning unit 1313 performs sequential learning (online learning) based on the data collected from the measuring device 1302, and adjusts the model parameters of the model arranged in the inference unit 171-2.
- the model parameters adjusted by the sequential learning unit 1313 may be stored in the model information storage unit 1323.
- the inference unit 171_2 executes the model based on the data collected from the measuring device 1302, and outputs the inference result (for example, the inference result (control value) in the process control).
- the inference result output from the inference unit 171_2 is notified to the notification unit 1314, and is stored in the actual measurement data storage unit 172_2 in association with the correct answer data, for example.
- the notification unit 1314 transmits the inference result (for example, the inference result (control value) in the process control) notified by the inference unit 171-2 to the board processing device 160.
- the board processing apparatus 160 performs control according to the transmitted inference result (for example, the inference result (control value) in the process control).
- the instruction unit 1312 is arranged in the virtual space (for real-time control), but the instruction unit 1312 may be configured to be realized in the substrate processing device 160.
- the virtual space (for simulation) of the execution device 170 includes learning units 210_1 and 210_2.
- FIG. 14A is a flowchart showing the flow of sequential learning process by the execution device.
- step S1403 the inference units 171-1 and 171-2 of the execution device 170 execute the model based on the data collected from the board processing device 160 and output the inference result.
- step S1404 the sequential learning units 1311 and 1313 of the execution device 170 perform sequential learning based on the data collected from the substrate processing device 160, and adjust the model parameters, respectively.
- step S1405 the execution device 170 stores the inference result output from the inference units 171_1 and 171_2 in the actual measurement data storage unit 172_1 or 172_2 in association with the correct answer data.
- step S1406 the execution device 170 determines whether or not to end the sequential learning process. If it is determined in step S1406 that the sequential learning process is not completed (NO in step S1406), the process proceeds to step S1407.
- step S1407 the execution device 170 determines whether or not a new model parameter has been transmitted from the nth layer management device 151. If it is determined in step S1407 that the new model parameter has not been transmitted (NO in step S1407), the process returns to step S1403.
- step S1407 determines whether a new model parameter has been transmitted (YES in step S1407). If it is determined in step S1407 that a new model parameter has been transmitted (YES in step S1407), the process proceeds to step S1408.
- step S1408 the execution device 170 verifies the suitability of the model parameters by performing a simulation using the model parameters transmitted from the nth layer management device 151.
- step S1408 If it is not verified in step S1408 that the model parameters are appropriate (NO in step S1408), the process returns to step S1403.
- step S1408 determines whether the model parameters are appropriate. If it is verified in step S1408 that the model parameters are appropriate (YES in step S1408), the process returns to step S1402. As a result, the model in which the new model parameters are set is distributed to the corresponding inference unit.
- step S1406 determines whether the sequential learning process is to be completed (YES in step S1406). If it is determined in step S1406 that the sequential learning process is to be completed (YES in step S1406), the sequential learning process is terminated.
- FIG. 14B is a flowchart showing the flow of batch learning process by the execution device.
- step S1411 the execution device 170 determines whether or not a certain amount of actual measurement data has been accumulated in the actual measurement data storage units 172_1 and 172_2. If it is determined in step S1411 that a certain amount of actually measured data is not accumulated (NO in step S1411), the process proceeds to step S1416.
- step S1411 determines whether a certain amount of actually measured data has been accumulated (YES in step S1411), the process proceeds to step S1412.
- step S1412 the learning units 210_1 and 210_2 of the execution device 170 read the actual measurement data from the actual measurement data storage units 172_1 and 172_2.
- step S1413 the learning units 210_1 and 210_2 of the execution device 170 perform additional learning processing (batch learning processing) using the read actual measurement data.
- step S1415 the execution device 170 transmits the optimized model parameters to the nth layer management device 151.
- step S1416 the execution device 170 determines whether or not to end the batch learning process. If it is determined in step S1409 that the batch learning process is not completed (NO in step S1416), the process returns to step S1411.
- step S1416 determines whether the batch learning process is to be completed (YES in step S1416). If it is determined in step S1416 that the batch learning process is to be completed (YES in step S1416), the batch learning process is terminated.
- the model management system 100 is -The model applied to the board manufacturing process is managed by dividing it into three or more layers.
- It has a management device (second management unit) of x + 1) hierarchy.
- -The management device that manages the model in the xth layer is a new model based on each updated model parameter when the model parameters of the model managed in one or more (x + 1) layers are updated. Calculate the parameters.
- -The management device that manages the model in the xth layer controls so that new model parameters are set in the model managed by each of the plurality of management devices in the lowest layer belonging to the management device that manages the model in the xth layer. do.
- the model is divided into three or more layers, and the model parameters are shared among the management devices of the same layer, so that the number of models is aggregated and the management cost is reduced.
- the model applied to the substrate processing process can be efficiently managed.
- the model is divided into three or more layers as a whole. You may manage it.
- FIG. 15 is a second diagram showing an example of the system configuration of the model management system.
- the model management device 1510 manages the model in the substrate processing factory ⁇ and the model in the substrate processing factory ⁇ .
- the substrate processing factory ⁇ has the first layer management device 120 and the second layer management devices 131 and 132, and is a management device of the third layer or higher. Does not have.
- the second layer management devices 131 and 132 are the lowest layer management devices, and are connected to the execution devices 170 and 190, respectively.
- the model management device 1510 is, for example, in the seventh phase. -In the sixth phase, the model parameters transmitted from the first layer management device 120 of the substrate processing factory ⁇ and -In the sixth phase, the model parameters transmitted from the first layer management device 120 of the substrate processing factory ⁇ and Calculate new model parameters based on. Further, the model management device 1510 manages new model parameters in the seventh phase, and uses the new model parameters in the first layer management device 120 of the board processing factory ⁇ and the first layer of the board processing factory ⁇ . Send to the management device.
- model parameters of the model managed by each of the plurality of management devices are updated using the new model parameters. It will be.
- the model is divided into two layers in the same board processing factory, and a common management device for managing the models of a plurality of board processing factories is arranged outside the factory. It is managed by dividing it into three or more layers. Thereby, according to the second embodiment, the same effect as that of the first embodiment can be enjoyed.
- the first layer management device to the nth layer management device are separate management devices, but the first layer management device to the nth layer management device are integrated. It may be formed as a device.
- the management device when the model parameter newly generated by the x-layer management device is transmitted, the management device is one layer lower than the x-layer management device and is the xth layer. It has been described as transmitting to all the management devices connected to the hierarchy management device. However, the transmission destination of the model parameter may be a management device one level lower than the xth layer management device, and may be configured to be transmitted to some management devices connected to the xth layer management device. ..
- the management process when the model management system manages the model according to each purpose of the existing board processing apparatus has been described.
- the management device that manages the model according to each purpose of the new board processing device is -Connected to a management device that manages models corresponding to a new group of board processing devices -Obtain the model structure and model parameters (latest model parameters at the time of addition) from the management device on the next higher level to which the management device is connected. -Manage the acquired model and model parameters and send them to the corresponding execution device.
- the latest model can be arranged in the inference section of the corresponding execution device as a model according to each purpose of the newly added board processing device. become. That is, the management device that manages the model according to each purpose of the new board processing device controls so that the latest model parameters are set in the model according to each purpose of the new board processing device.
- the latest model parameters are transmitted to the management device that manages the model according to each purpose of the new board processing device, the latest model parameters are also transmitted to other management devices in the same layer. .. However, the latest model parameters may not be transmitted to other management devices in the same layer, and the set model parameters may be continuously managed. Further, when transmitting the latest model parameters, a simulation may be performed in advance and the transmission may be performed after the verification is completed.
- the model used in the first and second embodiments is, for example, a machine learning model including deep learning.
- ⁇ RNN Recurrent Neural Network
- RSTM Long Short-Term Memory
- CNN Convolutional Neural Network
- R-CNN Regular Convolutional Neural Network
- YOLO You Only Look Once
- SSD Single Shot MultiBox Detector
- GAN Geneative Adversarial Network
- SVM Simple Vector Machine
- Decision tree ⁇ Random Forest And so on.
- a model using a genetic algorithm such as GA (Genetic Algorithm) or GP (Genetic Programming), or a model learned by reinforcement learning may be used.
- GA Genetic Algorithm
- GP Genetic Programming
- the models used in the first and second embodiments are PCR (Principal Component Regression), PLS (Partial Least Square), LASTO, ridge regression, linear polypoly, autoregressive model, moving average model, and autoregressive migration. It may be a model obtained by general statistical analysis other than deep learning, such as an average model or an ARX model.
- the model management program is executed independently by each of the management devices in each layer.
- the management device of each layer is composed of, for example, a plurality of computers and the model management program is installed on the plurality of computers, it is executed in the form of distributed computing. May be good.
- the management device has been described as being composed of separate devices for each layer, but if the management is divided for each layer, the management devices of different layers are integrated. It may be configured by a specific device. Alternatively, some functions may be shared and executed among the management devices of different layers.
- the download source may be, for example, a server device that stores the model management program in an accessible manner.
- the server device may be, for example, a device that accepts access from a management device via a network (not shown) and downloads a model management program on condition of billing. That is, the server device may be a device that provides a service for providing a model management program.
- Model management system 110 Model management device 120: First layer management device 131 to 133: Second layer management device 141: First (n-1) layer management device 151, 152: nth layer management device 160, 180: Board processing equipment 170, 190: Execution device 171_1, 171-2: Inference unit 191_1, 191_2: Inference unit 210_1, 210_2: Learning unit 220_1, 220_2: Learning unit 1001: Upper layer information acquisition unit 1002: Sharing unit 1003: Lower layer information Acquisition unit 1004: Optimization unit 1005: Upper layer information transmission unit
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Manufacturing & Machinery (AREA)
- Economics (AREA)
- Primary Health Care (AREA)
- Theoretical Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- General Factory Administration (AREA)
Abstract
Description
基板製造プロセスに適用されるモデルを3階層以上の階層に分けて管理するモデル管理システムであって、
所定の階層においてモデルを管理する第1の管理部と、
前記第1の管理部の1階層下位の階層においてモデルを管理する1つ以上の第2の管理部と、を有し、
前記第1の管理部は、
前記第2の管理部が管理するモデルのモデルパラメータが更新された場合に、更新された各モデルパラメータに基づいて、新たなモデルパラメータを算出する算出部と、
前記第1の管理部に属する最下層の複数の管理部それぞれにより管理されるモデルに、前記新たなモデルパラメータが設定されるよう制御する制御部とを有する。
<モデル管理システムのシステム構成>
はじめに、第1の実施形態に係るモデル管理システムのシステム構成について説明する。なお、第1の実施形態に係るモデル管理システムは、複数の管理装置を有しており、各管理装置は、各階層において、モデル(モデル構造及びモデルパラメータ)を管理する。各管理装置が管理するモデルはモデル構造が1種類であっても、複数種類であってもよい(第1の実施形態では、各管理装置が2種類のモデル構造のモデルを管理する場合について説明する)。また、第1の実施形態に係るモデル管理システムでは、モデルに対して学習処理が行われることでモデルパラメータが最適化された場合に、新たなモデルパラメータを生成することで、同一階層の管理装置間でモデルパラメータを共有化することができる。
図1は、第1のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図1に示すように、モデル管理システム100は、モデル管理装置110、学習用データ格納部111、学習部112_1、112_2を有する。
・第2階層管理装置131は、装置機種が“酸化炉を有する熱処理装置”のグループに対応するモデルを管理し、
・第2階層管理装置132は、装置機種が“窒化炉を有する熱処理装置”のグループに対応するモデルを管理し、
・第2階層管理装置133は、装置機種が“メタル炉を有する熱処理装置”のグループに対応するモデルを管理する。
・第2階層管理装置131は、装置機種が“酸化膜をエッチングするエッチング装置”のグループに対応するモデルを管理し、
・第2階層管理装置132は、装置機種が“Poly-Si膜をエッチングするエッチング装置”のグループに対応するモデルを管理する。
・第3階層管理装置は、装置機種が“窒化炉を有する熱処理装置”のグループに対応するモデルのうち、チャンバ種が“層間絶縁膜を処理するチャンバ”の熱処理装置のグループに対応するモデルを管理し、
・他の第3階層管理装置は、装置機種が“窒化炉を有する熱処理装置”のグループに対応するモデルのうち、チャンバ種が“ゲート絶縁膜を処理するチャンバ”の熱処理装置のグループに対応するモデルを管理し、
・他の第3階層管理装置は、装置機種が“メタル炉を有する熱処理装置”のグループに対応するモデルのうち、チャンバ種が“ゲート電極層を処理するチャンバ”の熱処理装置のグループに対応するモデルを管理し、
・他の第3階層管理装置は、装置機種が“メタル炉を有する熱処理装置”のグループに対応するモデルのうち、チャンバ種が“メタル膜を処理するチャンバ”の熱処理装置のグループに対応するモデルを管理する。
・第3階層管理装置は、装置機種が“Poly-Si膜をエッチングするエッチング装置”のグループに対応するモデルのうち、チャンバ種が“コンタクトホール開口を行うチャンバ”のエッチング装置のグループに対応するモデルを管理し、
・他の第3階層管理装置は、装置機種が“Poly-Si膜をエッチングするエッチング装置”のグループに対応するモデルのうち、チャンバ種が“配線用トレンチ加工を行うチャンバ”のエッチング装置のグループに対応するモデルを管理する。
・第2階層管理装置によりグループ分けされ、かつ、
・第3階層管理装置によりグループ分けされた、
各グループに対応するモデルを、更に、「プロセスグループ」が同じ基板処理装置のグループに分けて管理する。具体的には、基板処理装置が、熱処理装置である場合、例えば、
・第4階層管理装置は、チャンバ種が“層間絶縁膜を処理するチャンバ”の熱処理装置のグループに対応するモデルのうち、プロセスグループ=“第1のプロセス群”を担当する熱処理装置のグループに対応するモデルを管理し、
・他の第4階層管理装置は、チャンバ種が“層間絶縁膜を処理するチャンバ”の熱処理装置のグループに対応するモデルのうち、プロセスグループ=“第2のプロセス群”を担当する熱処理装置のグループに対応するモデルを管理する。
・第4階層管理装置は、チャンバ種が“コンタクトホール開口を行うチャンバ”のエッチング装置のグループに対応するモデルのうち、プロセスグループ=“第1のプロセス群”を担当するエッチング装置のグループに対応するモデルを管理し、
・他の第4階層管理装置は、チャンバ種が“コンタクトホール開口を行うチャンバ”のエッチング装置のグループに対応するモデルのうち、プロセスグループ=“第2のプロセス群”を担当するエッチング装置のグループに対応するモデルを管理する。
・第n階層管理装置151から送信されたモデル(モデル構造=“モデルA”、モデルパラメータ=“パラメータPA0”)が配された推論部171_1と、
・第n階層管理装置151から送信されたモデル(モデル構造=“モデルB”、モデルパラメータ=“パラメータPB0”)が配された推論部171_2と、
を有する。
・第n階層管理装置152から送信されたモデル(モデル構造=“モデルA”、モデルパラメータ=“パラメータPA0”)が配された推論部191_1と、
・第n階層管理装置152から送信されたモデル(モデル構造=“モデルB”、モデルパラメータ=“パラメータPB0”)が配された推論部191_2と、
を有する。
図2は、第2のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図2に示すように、第2のフェーズにおいて、モデル管理システム100は、上記第1のフェーズのシステム構成に加えて、更に、実行装置170が、学習部210_1、210_2を有する。また、実行装置190が、学習部220_1、220_2を有する。
図3は、第3のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図3に示すように、第3のフェーズにおいて、第n階層管理装置151は、
・モデル構造=“モデルA”、“モデルB”と、
・第2のフェーズにおいて学習部210_1、210_2から送信されたモデルパラメータ=“パラメータPA1”、“パラメータPB1”と、
を管理する。
・モデル構造=“モデルA”、“モデルB”と、
・第2のフェーズにおいて学習部220_1、220_2から送信されたモデルパラメータ=“パラメータPA2”、“パラメータPB2”と、
を管理する。
図4は、第4のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図4に示すように、第4のフェーズにおいて、第n階層管理装置151は、所定のタイミングで、モデルパラメータ=“パラメータPA1”、“パラメータPB1”を、第(n-1)階層管理装置141に送信する。
図5は、第5のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図5に示すように、第5のフェーズにおいて、第(n-1)階層管理装置141は、
・第4のフェーズにおいて第n階層管理装置151から送信された、モデルパラメータ=“パラメータPA1”、“パラメータPB1”と、
・第4のフェーズにおいて第n階層管理装置152から送信された、モデルパラメータ=“パラメータPA2”、“パラメータPB2”と、
に基づいて、新たなモデルパラメータ=“パラメータPA3”、“パラメータPB3”を算出する。なお、第(n-1)階層管理装置141による新たなモデルパラメータの算出方法は任意であり、例えば、“パラメータPA1”と“パラメータPA2”とを重み付け加算して、“パラメータPA3”を算出してもよい。同様に、“パラメータPB1”と“パラメータPB2”とを重み付け加算して、“パラメータPB3”を算出してもよい。
・モデル構造=“モデルA”、“モデルB”と、
・接続先である第(n-1)階層管理装置141から送信された、新たなモデルパラメータ=“パラメータPA3”、“パラメータPB3”と、
を管理する。
図6は、第6のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図6に示すように、第6のフェーズにおいて、第(n-1)階層管理装置141は、所定のタイミングで、新たなモデルパラメータ=“パラメータPA3”、“パラメータPB3”を、第(n-2)階層管理装置に送信する。なお、ここでは、第(n-2)階層管理装置が、第2階層管理装置132であるとして説明する。
図7は、第7のフェーズにおけるモデル管理システムのシステム構成の一例を示す図である。図7に示すように、第7のフェーズにおいて、第2階層管理装置132は、
・第6のフェーズにおいて、第(n-1)階層管理装置141から送信された、モデルパラメータ=“パラメータPA3”、“パラメータPB3”と、
・第6のフェーズにおいて、第2階層管理装置132に接続する他の複数の第(n-1)階層管理装置から送信されたモデルパラメータと、
に基づいて、新たなモデルパラメータ=“パラメータPA6”、“パラメータPB6”を算出する。
・モデル構造=“モデルA”、“モデルB”と、
・接続先である第2階層管理装置132から送信された、新たなモデルパラメータ=“パラメータPA6”、“パラメータPB6”と、
を管理する。
・モデル構造=“モデルA”、“モデルB”と、
・接続先である第(n-1)階層管理装置141から送信された、新たなモデルパラメータ=“パラメータPA6”、“パラメータPB6”と、
を管理する。
次に、モデル管理システム100によるモデル管理処理の流れについてシーケンス図を用いて説明する。図8は、モデル管理システムによるモデル管理処理の流れを示すシーケンス図である。なお、図8に示すモデル管理処理を実行するにあたり、学習部112_1、112_2による学習処理は完了しており、モデル管理装置110により、モデル構造とモデルパラメータとが管理されているものとする。
次に、第1階層管理装置120から第n階層管理装置151、152までの各管理装置のハードウェア構成について説明する。なお、第1階層管理装置120から第n階層管理装置151、152までの各管理装置は、いずれも同様のハードウェア構成を有するため、ここでは、第x階層(xは1~nの任意の整数)の管理装置である第x階層管理装置のハードウェア構成について説明する。図9は、第x階層管理装置のハードウェア構成の一例を示す図である。
次に、第1階層管理装置120から第n階層管理装置151、152の各管理装置の機能構成について説明する。なお、第1階層管理装置120から第n階層管理装置151、152の各管理装置は、いずれも同様の機能構成を有するため、ここでも、第x階層(xは1~nの任意の整数)の管理装置である第x階層管理装置の機能構成について説明する。
次に、第x階層管理装置900による管理処理の流れについて説明する。図11は、第x階層管理装置による管理処理の流れを示すフローチャートである。
次に、実行装置170、190のハードウェア構成について説明する。図12は、実行装置のハードウェア構成の一例を示す図である。なお、実行装置170、190のハードウェア構成は、第x階層管理装置900のハードウェア構成と概ね同じであるため、ここでは、相違点のみを説明する。
次に、実行装置170の機能構成について説明する。図13は、実行装置の機能構成の一例を示す図である。図13に示すように、実行装置170は、物理空間より収集したデータを処理し、モデルを実行または更新する仮想空間(リアルタイム制御用、シミュレーション用)を有する。また、実行装置170は、物理空間より収集したデータ及び仮想空間において更新されたモデルパラメータ等を格納するストレージ群を有する。
・基板処理装置160と、
・基板処理装置160による基板処理の際に、推論部171_1がモデルを実行するのに、あるいは、逐次学習部1311がモデルの逐次学習を行うのに用いられるデータを測定する測定装置1301と、
・基板処理装置160による基板処理の際に、推論部171_2がモデルを実行するのに、あるいは、逐次学習部1313がモデルの逐次学習を行うのに用いられるデータを測定する測定装置1302と、
が含まれる。
次に、実行装置170による逐次学習処理及びバッチ学習処理の流れについて説明する。
図14Aは、実行装置による逐次学習処理の流れを示すフローチャートである。
図14Bは、実行装置によるバッチ学習処理の流れを示すフローチャートである。
以上の説明から明らかなように、第1の実施形態に係るモデル管理システム100は、
・基板製造プロセスに適用されるモデルを3階層以上の階層に分けて管理する。
・第x階層においてモデルを管理する管理装置(第1の管理部)と、第x階層においてモデルを管理する管理装置に接続し、1階層下位の階層においてモデルを管理する1つ以上の第(x+1)階層の管理装置(第2の管理部)とを有する。
・第x階層においてモデルを管理する管理装置は、1つ以上の第(x+1)階層において管理されるモデルのモデルパラメータが更新された場合に、更新された各モデルパラメータに基づいて、新たなモデルパラメータを算出する。
・第x階層においてモデルを管理する管理装置は、第x階層においてモデルを管理する管理装置に属する最下層の複数の管理装置それぞれにより管理されるモデルに、新たなモデルパラメータが設定されるよう制御する。
上記第1の実施形態では、同一の基板処理工場内において、モデルを3階層以上の階層に分けて管理する場合について説明した。しかしながら、モデルを3階層以上の階層に分ける方法はこれに限定されない。
・第6のフェーズにおいて、基板処理工場αの第1階層管理装置120から送信されたモデルパラメータと、
・第6のフェーズにおいて、基板処理工場βの第1階層管理装置120から送信されたモデルパラメータと、
に基づいて、新たなモデルパラメータを算出する。また、モデル管理装置1510は、第7のフェーズにおいて、新たなモデルパラメータを管理するとともに、該新たなモデルパラメータを、基板処理工場αの第1階層管理装置120及び基板処理工場βの第1階層管理装置に送信する。
上記第1及び第2の実施形態では、第1階層管理装置から第n階層管理装置を、それぞれ、別体の管理装置としたが、第1階層管理装置から第n階層管理装置は、一体の装置として形成してもよい。
・新たな基板処理装置のグループに対応するモデルを管理する管理装置に接続され、
・該管理装置が接続された1階層上位の階層の管理装置より、モデル構造とモデルパラメータ(追加時の最新のモデルパラメータ)とを取得し、
・取得したモデルとモデルパラメータを管理するとともに、対応する実行装置に送信する。
・RNN(Recurrent Neural Network)、
・LSTM(Long Short-Term Memory)、
・CNN(Convolutional Neural Network)、
・R-CNN(Region based Convolutional Neural Network)、
・YOLO(You Only Look Once)、
・SSD(Single Shot MultiBox Detector)、
・GAN(Generative Adversarial Network)、
・SVM(Support Vector Machine)、
・決定木、
・Random Forest
等のいずれかであってもよい。
110 :モデル管理装置
120 :第1階層管理装置
131~133 :第2階層管理装置
141 :第(n-1)階層管理装置
151、152 :第n階層管理装置
160、180 :基板処理装置
170、190 :実行装置
171_1、171_2 :推論部
191_1、191_2 :推論部
210_1、210_2 :学習部
220_1、220_2 :学習部
1001 :上位層情報取得部
1002 :共有化部
1003 :下位層情報取得部
1004 :最適化部
1005 :上位層情報送信部
Claims (9)
- 基板製造プロセスに適用されるモデルを3階層以上の階層に分けて管理するモデル管理システムであって、
所定の階層においてモデルを管理する第1の管理部と、
前記所定の階層より1階層下位の階層においてモデルを管理する1つ以上の第2の管理部と、を有し、
前記第1の管理部は、
前記第2の管理部が管理するモデルのモデルパラメータが更新された場合に、更新された各モデルパラメータに基づいて、新たなモデルパラメータを算出する算出部と、
前記第1の管理部に属する最下層の複数の管理部それぞれにより管理されるモデルに、前記新たなモデルパラメータが設定されるよう制御する制御部と
を有するモデル管理システム。 - 前記第1の管理部が管理するモデルよりも1階層上位の階層のモデルを管理する管理部に、前記新たなモデルパラメータを送信する送信部を更に有する、請求項1に記載のモデル管理システム。
- 前記階層は、前記基板製造プロセスを実行する装置の装置機種、前記基板製造プロセスを実行する装置のチャンバ種、前記基板製造プロセスを実行する装置のプロセスグループにより分けられる、請求項1に記載のモデル管理システム。
- 前記モデルは、前記基板製造プロセスを実行する装置の異常検知の推論結果、プロセス処理の結果物の推論結果、部品劣化診断の推論結果、内部状態の推論結果のいずれかを出力するモデル、または、前記基板製造プロセスの制御における推論結果を出力するモデルのいずれかである、請求項1に記載のモデル管理システム。
- 前記算出部は、予め定められた条件を満たす場合に、前記新たなモデルパラメータを算出し、前記制御部は、予め定められたタイミングで、前記最下層の複数のモデルに、新たなモデルパラメータが設定されるよう制御する、請求項1に記載のモデル管理システム。
- 前記制御部は、前記最下層の複数の管理部それぞれにより管理されるモデルの一部に、前記新たなモデルパラメータが設定されるよう制御する、請求項1に記載のモデル管理システム。
- 前記制御部は、
前記基板製造プロセスを実行する新たな装置が追加された場合、該追加された装置に対応するモデルに、前記新たなモデルパラメータが設定されるよう制御する、請求項1に記載のモデル管理システム。 - 基板製造プロセスに適用されるモデルを3階層以上の階層に分けて管理するモデル管理システムのモデル管理方法であって、
所定の階層において第1の管理部がモデルを管理する第1の管理工程と、
前記所定の階層より1階層下位の階層において、1つ以上の第2の管理部がモデルを管理する第2の管理工程と、を有し、
前記第1の管理工程は、
前記第2の管理工程において管理されるモデルのモデルパラメータが更新された場合に、更新された各モデルパラメータに基づいて、新たなモデルパラメータを算出する算出工程と、
前記第1の管理部に属する最下層の複数の管理部それぞれにより管理されるモデルに、前記新たなモデルパラメータが設定されるよう制御する制御工程と
を有するモデル管理方法。 - 基板製造プロセスに適用されるモデルを3階層以上の階層に分けて管理するモデル管理システムのコンピュータに、
所定の階層において第1の管理部がモデルを管理する第1の管理工程と、
前記所定の階層より1階層下位の階層において、1つ以上の第2の管理部がモデルを管理する第2の管理工程と、を実行させるためのモデル管理プログラムであって、
前記第1の管理工程は、
前記第2の管理工程において管理されるモデルのモデルパラメータが更新された場合に、更新された各モデルパラメータに基づいて、新たなモデルパラメータを算出する算出工程と、
前記第1の管理部に属する最下層の複数の管理部それぞれにより管理されるモデルに、前記新たなモデルパラメータが設定されるよう制御する制御工程と
を有するモデル管理プログラム。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020237018580A KR20230101855A (ko) | 2020-11-10 | 2021-10-29 | 모델 관리 시스템, 모델 관리 방법 및 모델 관리 프로그램 |
US18/250,636 US20230393540A1 (en) | 2020-11-10 | 2021-10-29 | Model management system, model management method, and model management program |
JP2022561826A JPWO2022102439A1 (ja) | 2020-11-10 | 2021-10-29 | |
CN202180073627.8A CN116507981A (zh) | 2020-11-10 | 2021-10-29 | 模型管理系统、模型管理方法和模型管理程序 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2020-187177 | 2020-11-10 | ||
JP2020187177 | 2020-11-10 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022102439A1 true WO2022102439A1 (ja) | 2022-05-19 |
Family
ID=81602235
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2021/040093 WO2022102439A1 (ja) | 2020-11-10 | 2021-10-29 | モデル管理システム、モデル管理方法及びモデル管理プログラム |
Country Status (6)
Country | Link |
---|---|
US (1) | US20230393540A1 (ja) |
JP (1) | JPWO2022102439A1 (ja) |
KR (1) | KR20230101855A (ja) |
CN (1) | CN116507981A (ja) |
TW (1) | TW202226078A (ja) |
WO (1) | WO2022102439A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180253646A1 (en) * | 2017-03-05 | 2018-09-06 | International Business Machines Corporation | Hybrid aggregation for deep learning neural networks |
JP2019093969A (ja) * | 2017-11-24 | 2019-06-20 | 株式会社東芝 | 車両空調制御装置、車両の空調機の制御方法及びプログラム |
WO2020094325A1 (en) * | 2018-11-07 | 2020-05-14 | Asml Netherlands B.V. | Determining a correction to a process |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3605405A4 (en) | 2017-03-21 | 2020-12-02 | Preferred Networks, Inc. | SERVER DEVICE, TRAINED MODEL DELIVERY PROGRAM, TRAINED MODEL DELIVERY PROCESS, AND DRIVEN MODEL DELIVERY SYSTEM |
WO2019163823A1 (ja) | 2018-02-20 | 2019-08-29 | 株式会社Abeja | 情報処理システム |
-
2021
- 2021-10-27 TW TW110139823A patent/TW202226078A/zh unknown
- 2021-10-29 WO PCT/JP2021/040093 patent/WO2022102439A1/ja active Application Filing
- 2021-10-29 KR KR1020237018580A patent/KR20230101855A/ko unknown
- 2021-10-29 US US18/250,636 patent/US20230393540A1/en active Pending
- 2021-10-29 CN CN202180073627.8A patent/CN116507981A/zh active Pending
- 2021-10-29 JP JP2022561826A patent/JPWO2022102439A1/ja active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180253646A1 (en) * | 2017-03-05 | 2018-09-06 | International Business Machines Corporation | Hybrid aggregation for deep learning neural networks |
JP2019093969A (ja) * | 2017-11-24 | 2019-06-20 | 株式会社東芝 | 車両空調制御装置、車両の空調機の制御方法及びプログラム |
WO2020094325A1 (en) * | 2018-11-07 | 2020-05-14 | Asml Netherlands B.V. | Determining a correction to a process |
Also Published As
Publication number | Publication date |
---|---|
KR20230101855A (ko) | 2023-07-06 |
CN116507981A (zh) | 2023-07-28 |
TW202226078A (zh) | 2022-07-01 |
US20230393540A1 (en) | 2023-12-07 |
JPWO2022102439A1 (ja) | 2022-05-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110022230A (zh) | 基于深度强化学习的服务链并行部署方法及装置 | |
CN108768716A (zh) | 一种微服务路径选择方法及装置 | |
EP4100888A1 (en) | Computational graph optimization | |
CN111917642B (zh) | 分布式深度强化学习的sdn网络智慧路由数据传输方法 | |
CN108684046A (zh) | 一种基于随机学习的接入网服务功能链部署方法 | |
US10908675B2 (en) | Electronic device for signal processing with embedded optimization of electrical energy consumption and a corresponding method | |
EP4012563A1 (en) | Profiling and application monitoring for edge devices based on headroom | |
CN110247795A (zh) | 一种基于意图的云网资源服务链编排方法及系统 | |
CN109202895A (zh) | 设计支持装置、设计支持方法及存储设计支持程序的介质 | |
Wang et al. | A reinforcement learning approach for online service tree placement in edge computing | |
WO2021025075A1 (ja) | 訓練装置、推定装置、訓練方法、推定方法、プログラム及びコンピュータ読み取り可能な非一時的記憶媒体 | |
Hashemi et al. | A multi‐objective method for virtual machines allocation in cloud data centres using an improved grey wolf optimization algorithm | |
CN115168027A (zh) | 一种基于深度强化学习的算力资源度量方法 | |
Srivastava et al. | Implementation of ant colony optimization in economic load dispatch problem | |
WO2022102439A1 (ja) | モデル管理システム、モデル管理方法及びモデル管理プログラム | |
CN116956744A (zh) | 基于改进粒子群算法的多回路沟槽电缆稳态温升预测方法 | |
CN115016947A (zh) | 负载分配方法、装置、设备及介质 | |
CN110135725A (zh) | 一种线缆装配序列规划方法、装置及设备 | |
CN116089083A (zh) | 一种多目标数据中心资源调度方法 | |
CN115220818A (zh) | 基于深度强化学习的实时依赖型任务卸载方法 | |
JP2887969B2 (ja) | ニューラルネットワークによる部品配置最適化方法 | |
CN111027709B (zh) | 信息推荐方法、装置、服务器及存储介质 | |
Nugraheni et al. | Hybrid Metaheuristics for Job Shop Scheduling Problems. | |
CN113390414A (zh) | 一种军用油料投送路径规划方法 | |
Divya | Solving travelling salesman problem using ant systems: a programmer’s approach |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21891686 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2022561826 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202180073627.8 Country of ref document: CN |
|
ENP | Entry into the national phase |
Ref document number: 20237018580 Country of ref document: KR Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21891686 Country of ref document: EP Kind code of ref document: A1 |