TW202343313A - Program, information processing method, information processing device, and model generation method - Google Patents

Abstract

A program according to one embodiment of the present disclosure causes a computer to execute a process for: acquiring process data including at least two among the type and flow rate of gas supplied to the inside of a semiconductor manufacturing device, chamber internal pressure, and valve opening degree of an automatic pressure control device; inputting the process data to a training model trained to output an estimated pressure value of the inside of the semiconductor manufacturing device when the process data is inputted, and outputting the estimated pressure value; and determining whether a state is an abnormal state on the basis of the estimated pressure value.

Description

Programs, information processing methods, information processing devices and methods for generating models

This technology relates to a processing program, an information processing method, an information processing device, and a model generation method for a semiconductor manufacturing device.

In the past, a monitoring method for detecting abnormalities in semiconductor manufacturing equipment has been proposed. For example, the monitoring method described in Patent Document 1 determines abnormality based on the gas temperature in the vacuum pump part of the semiconductor manufacturing equipment. [Prior technical literature] [Patent Document]

Patent Document 1: Japanese Patent Application Laid-Open No. 5-9759

[Problem to be solved by the invention]

The program described in Patent Document 1 determines the abnormality of the exhaust system based only on the gas temperature, so the accuracy of determining the abnormality of the exhaust system is low.

The present disclosure was developed in view of this situation, and its purpose is to provide a program for determining abnormalities in the exhaust system of semiconductor manufacturing equipment with high accuracy. [Means used to solve problems]

A program according to an embodiment of the present disclosure causes a computer to perform the following processing: acquiring program data including at least two of the type and flow rate of gas supplied to the semiconductor manufacturing equipment, the pressure in the chamber, or the valve opening of the automatic pressure control device. (process data), for a learning model that is learned to output an estimated pressure value in the semiconductor manufacturing device when the aforementioned program data is input, the aforementioned program data is input and the aforementioned estimated pressure value is output, and the determination is made based on the aforementioned estimated pressure value. Whether it is an abnormal state.

A program according to an embodiment of the present disclosure obtains an actual pressure value measured by a vacuum gauge in a semiconductor manufacturing device, and determines whether an abnormal state is present based on the estimated pressure value and the actual measured pressure value.

A program according to an embodiment of the present disclosure acquires the process data in an idle state of the device before starting the manufacturing process of supplying gas to the semiconductor device.

A program according to an embodiment of the present disclosure acquires the process data during a manufacturing process of supplying gas into a semiconductor device.

In the program according to an embodiment of the present disclosure, the program data includes the operating status of the exhaust pump.

A program according to an embodiment of the present disclosure determines an abnormal state when the difference between the estimated pressure value and the measured pressure value is greater than or equal to a predetermined value.

A program according to an embodiment of the present disclosure continuously obtains the difference between the estimated pressure value and the measured pressure value, creates time series data corresponding to the difference with time, and displays the time series data on the display unit.

A program according to an embodiment of the present disclosure predicts the time when the difference will exceed a predetermined value based on the time series data.

An information processing method according to an embodiment of the present disclosure is to obtain program data including at least two of the type and flow rate of gas supplied to the semiconductor manufacturing equipment, the pressure in the chamber, or the valve opening of the automatic pressure control device, and provide the information for the learned The learning model is a learning model that outputs an estimated pressure value in the semiconductor manufacturing device when the program data is input. The program data is input to output the estimated pressure value, and based on the estimated pressure value, it is determined whether an abnormal state is present.

An information processing device according to an embodiment of the present disclosure is provided with: a program data acquisition unit that acquires at least two of the type and flow rate of gas supplied to the semiconductor manufacturing device, the pressure in the chamber, or the valve opening of the automatic pressure control device. and The abnormality determination unit determines whether the abnormality is in an abnormal state based on the estimated pressure value.

A method for generating a model according to an embodiment of the present disclosure is to obtain a pressure value of a vacuum gauge in a semiconductor manufacturing device, obtain the type and flow rate of gas supplied to the semiconductor manufacturing device, the pressure in a chamber, or the valve opening of an automatic pressure control device. The degree includes at least two program data and training data that obtains the aforementioned pressure value. Based on the obtained training data, it is generated when the type and flow rate of the gas supplied to the aforementioned semiconductor manufacturing device, the pressure in the chamber, or When the valve opening of the automatic pressure control device includes at least two program data, a learning model of the estimated pressure value is output. [The effect of invention]

In the program according to one embodiment of the present disclosure, it can be determined with high accuracy whether the semiconductor manufacturing equipment is in an abnormal state.

(Embodiment 1) Specific examples of the program, information processing device, etc. according to the first embodiment will be described below with reference to the drawings.

FIG. 1 is a schematic diagram showing the schematic structure of the information processing device. The semiconductor manufacturing apparatus 1 of this embodiment is an apparatus that performs etching processing, film formation processing, etc. by performing plasma processing on a semiconductor (including MEMS (Micro Electro Mechanical Systems)) substrate. The semiconductor manufacturing apparatus 1 is equipped with an information processing apparatus 30, and the information processing apparatus 30 exchanges and receives the server apparatus 5, data, and a learning model.

The semiconductor manufacturing apparatus 1 includes a chamber 11, a sample stage 13, a gas supply source (gas No. 1 to No. 6), an MFC (Mass Flow Controller) 71, a gas valve (gas valve) 71a, and a vacuum Gauge 72, APC (Automatic Pressure Controller, automatic pressure control device) 73, first pump 74, second pump 75, vacuum gauge 76, exhaust pipe 77, pressure flow meter 78, and thermometer 79.

The gas valve 71a is provided in each flow path from the gas supply source associated with the gas type (gas No. 1 to No. 6 types) to the chamber 11 . MFC71 is provided in each flow path between the gas supply source and the gas valve 71a to control the flow rate of each gas (gas No. 1 to No. 6 flow rate).

The substrate S is placed on the sample stage 13 provided in the chamber 11 . While the substrate S is electrostatically adsorbed to the electrostatic chuck (not shown) provided on the sample stage 13 , gas is supplied from the gas supply source into the chamber, and the substrate S is etched by a plasma source (not shown). Or plasma treatment such as film forming treatment. The vacuum gauge 72 measures the pressure in the chamber 11 (internal chamber pressure).

The semiconductor manufacturing apparatus 1 is heated by a heater (not shown) that heats the wall surface in the chamber 11 . In addition, He (helium) gas is supplied to the back surface of the substrate S placed on the sample stage 13 for cooling. The pressure flow meter 78 performs pressure control and flow measurement of the He gas supplied to the back surface of the substrate S. The thermometer 79 measures the heater temperature.

The APC73 controls the pressure in the chamber 11 (pressure inside the chamber) measured by the vacuum gauge 72 to become the set pressure by adjusting the opening of its own valve (APC opening). The first pump 74 is a pump used to exhaust the reaction product, such as a turbo molecular pump. The second pump 75 is, for example, a dry pump, which is a pump used to assist the first pump 74 . By operating the first pump 74 and the second pump 75 , the reaction product in the chamber 11 is exhausted and the pressure in the chamber 11 is reduced. Hereinafter, the first pump 74 , the second pump 75 and the exhaust pipe 77 are also collectively referred to as an exhaust system. When the processing of the substrate S is completed in the semiconductor manufacturing apparatus 1, the valve of the APC is fully opened to discharge the remaining gas and He gas in the chamber.

In this way, the semiconductor manufacturing apparatus 1 is equipped with various machines and measuring instruments, and the information processing device 30 can obtain various program data regarding the manufacturing process from these machines and measuring instruments. In the example shown in the figure, the information processing device 30 can obtain the types No. 1 to No. 6 from the gas supply source, the flow rate No. 1 to No. 6 from the MFC 71 , the chamber pressure from the vacuum gauge 72 , and the pressure in the chamber from the APC 73 The APC opening is acquired, the motor rotation number of the first pump 74 (first pump rotation number) is acquired from the first pump 74 , the motor rotation number of the second pump 75 (second pump rotation number) is acquired from the second pump 75 , and the motor rotation number of the second pump 75 is acquired from the second pump 75 . The pressure flow meter 78 obtains the He gas pressure and flow rate, and the heater temperature from the thermometer 79, and includes them in the program data. In addition, in this embodiment, the operating states of the first pump 74 and the second pump 75 are the first pump rotation speed and the second pump rotation speed. The first pump rotation number and the second pump rotation number are controlled so as to become a constant rotation number.

The vacuum gauge 76 measures the pressure value (actual measured pressure value) of the exhaust pipe 77 between the first pump 74 and the second pump 75 . The information processing device 30 can obtain the actual measured pressure value from the vacuum gauge 76 .

FIG. 2 is a block diagram showing the structure of the information processing device. The information processing device 30 of this embodiment is composed of a processing unit 31, a memory unit 32, a communication unit 33, a display unit 34, an operation unit 35, a card slot 36, and the like. In addition, in this embodiment, the information processing device 30 is configured as a single device. However, the information processing device 30 may be configured to realize the function of the information processing device 30 by performing distributed processing on a plurality of devices. Alternatively, the information processing device 30 may be embedded in the semiconductor manufacturing device 1 .

The information processing device 30 determines whether the semiconductor manufacturing device 1 is in an abnormal state. The semiconductor manufacturing apparatus 1 supplies gas into the chamber 11 and performs substrate processing. If the semiconductor manufacturing apparatus 1 continues to manufacture semiconductors, the reaction products (deposits) of the gas supplied into the chamber 11 may clog the pipes, or the exhaust capacity may deteriorate due to the deterioration of the first pump 74 or the second pump 75 . Reduced, impairing the reproducibility of substrate processing (plasma processing). The information processing device 30 uses the learning model 32b to output the pressure value (estimated pressure value) of the exhaust system that should be displayed by the vacuum gauge 76 when it is assumed that the exhaust system is normal. When the semiconductor manufacturing apparatus 1 is in an abnormal state, the actual pressure value measured by the vacuum gauge 76 will become higher than the estimated pressure value because the exhaust capacity of the exhaust system is reduced. The information processing device 30 determines whether the semiconductor manufacturing apparatus 1 is in an abnormal state based on the difference between the actual measured pressure value and the estimated pressure value. When it is determined that the semiconductor manufacturing apparatus 1 is in an abnormal state, it notifies the user of the semiconductor manufacturing apparatus 1 . In addition, the information processing device 30 may also determine whether the semiconductor manufacturing apparatus 1 is in an abnormal state based on the first pump rotation number or the second pump rotation number. The reaction product (deposit) of the gas supplied into the chamber 11 may clog the pipe, and the first pump or the second pump may be unable to maintain the rotation speed and decelerate. The information processing device 30 can determine that the semiconductor manufacturing apparatus 1 is in an abnormal state when the first pump rotation number or the second pump rotation number is lower than a predetermined rotation number.

The processing unit 31 is configured using a computing device such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), or a GPU (Graphics Processing Unit). The processing unit 31 reads and executes the program stored in the memory unit 32, thereby performing various processes. In this embodiment, the processing unit 31 reads and executes the program 32a and the learning model 32b stored in the memory unit 32, thereby performing processing of outputting the estimated pressure value in the semiconductor manufacturing device 1 and determining the semiconductor manufacturing device 1 Various processing such as processing whether it is an abnormal state and processing to create time series data.

The memory unit 32 is configured using a large-capacity memory device such as a hard disk (harddisk) or an SSD (Solid State Drive). The memory unit 32 stores various programs executed by the processing unit 31 and various data required for processing by the processing unit 31 . In this embodiment, the memory unit 32 stores the program 32a executed by the processing unit 31, the learning model 32b used for processing by execution of the program 32a, and data that records the data output from the processing unit 31. Library 32c.

In this embodiment, the program 32a (program product) is provided in the form of being recorded on a recording medium 99 such as a memory card, and the information processing device 30 reads the program 32a from the recording medium 99 and stores it in the memory unit 32 . However, the program 32a may also be written in the memory unit 32 during the manufacturing stage of the information processing device 30, for example. For example, the program 32a may be read from the recording medium 99 by a writing device and written into the memory unit 32 of the information processing device 30 . The program 32a may also be provided through network distribution.

The learning model 32 b is a learning model that outputs an estimated pressure value based on the program data acquired from the semiconductor manufacturing apparatus 1 . In this embodiment, the learning model 32b is inputted with the gas No. 1 to No. 6 types and gas No. 1 to No. 6 flow rates measured by various measuring instruments of the semiconductor manufacturing apparatus 1, the chamber pressure, or the APC. When the opening degree contains at least two program data, the estimated pressure value is output. In addition, the program data may include at least one of the first pump rotation number or the second pump rotation number, and may also include heater temperature or He gas pressure and flow rate. In addition, although only one learning model 32b is shown in FIG. 2, a plurality of learning models 32b may be stored in the memory unit 32. In addition, the learning model 32b can also be stored in a cloud provided by an external server. Details of the learning model 32b will be described later.

The database 32c records the time series data generated by the processing unit 31. The details of the time series data will be explained later. In addition, the data recording department can also record program data, measured pressure values and estimated pressure values in addition to time series data.

The communication unit 33 is connected to the semiconductor manufacturing device 1 through a communication cable or the like, and transmits and receives various data to and from the semiconductor manufacturing device 1 through the communication cable. In this embodiment, the communication unit 33 acquires the program data measured by each measuring device of the semiconductor manufacturing apparatus 1 by receiving communication via a communication cable. In addition, the communication unit 33 may be configured to perform wireless communication with the semiconductor manufacturing apparatus 1 .

The display unit 34 is composed of a liquid crystal display or the like, and displays various images, characters, etc. based on processing by the processing unit 31 . The operation unit 35 accepts the user's operation and notifies the processing unit 31 of the accepted operation. For example, the operation unit 35 is an input device such as a mouse and a keyboard, and these input devices may be detachable from the information processing device 30 . In addition, for example, the operation unit 35 may also accept user operations through an input device such as a touch panel provided on the surface of the display unit 34 .

The card slot 36 can be used to detach and install a recording medium 99 such as a memory card, so as to read and write data to the installed recording medium 99 . The card slot 36 reads data from the recording medium 99 and supplies the data to the processing unit 31 , and writes the data supplied from the processing unit 31 to the recording medium 99 . In this embodiment, the learning model 32b that has undergone mechanical learning in the server device 5 in advance is recorded in the recording medium 99 and provided to the information processing device 30, and the information processing device 30 converts the learning model recorded in the recording medium 99 32b is read into the card slot 36 and stored in the memory unit 32. In addition, in this embodiment, the recording medium 99 is used to exchange data between the information processing device 30 and the server device 5. However, the present invention is not limited to this. For example, the information processing device 30 and the server device 5 may also be used. The server device 5 transmits and receives data through communication such as LAN (Local Area Network) or the Internet.

In addition, the memory unit 32 may also be an external memory device connected to the information processing device 30 . In addition, the information processing device 30 may be a multi computer composed of a plurality of computers, or may be a virtual machine constructed virtually through software. In addition, the information processing device 30 may not have a user interface (user interface) such as the display unit 34 and the operation unit 35. In this case, the information processing device 30 may be operated by an administrator through another device, for example.

In the processing unit 31 of the information processing device 30 of this embodiment, the processing unit 31 reads and executes the program 32a stored in the memory unit 32, thereby causing the program data acquisition unit 31a, the measured pressure value acquisition unit 31b, The estimated pressure value output unit 31c, the abnormality determination unit 31d, the display processing unit 31e, the time series data creation unit 31f, the data recording unit 31g, and the like are implemented as software functional units. In addition, the processing unit 31 has a counter function starting from the time when the semiconductor manufacturing apparatus 1 starts manufacturing the semiconductor. In addition, as an alternative to the counter function, a timer function that measures actual time can also be provided.

The program data acquisition unit 31 a performs a process of acquiring program data from the semiconductor manufacturing apparatus 1 by communicating with the semiconductor manufacturing apparatus 1 through the communication unit 33 . As mentioned above, in this embodiment, the program data includes data measured by various measuring instruments provided in the semiconductor manufacturing apparatus 1 .

The measured pressure value acquisition unit 31 b performs the following processing: communicating with the semiconductor manufacturing apparatus 1 through the communication unit 33 , thereby acquiring the actual measured pressure value measured by the vacuum gauge 76 provided in the semiconductor manufacturing apparatus 1 .

The estimated pressure value output unit 31c reads the learning model 32b from the storage unit 32, inputs the program data acquired by the program data acquisition unit 31a into the learning model 32b, and outputs the estimated pressure value.

The abnormality determination unit 31d determines whether the semiconductor manufacturing apparatus 1 is in an abnormal state based on the actual measured pressure value acquired by the actual measured pressure value acquisition unit 31b and the estimated pressure value output by the estimated pressure value output unit 31c. Specifically, the difference between the actual measured pressure value and the estimated pressure value is calculated, and when the calculated difference is equal to or greater than a preset predetermined value (threshold value), it is determined that the semiconductor manufacturing apparatus 1 is in an abnormal state. In addition, the abnormality determination unit 31d determines that the abnormality is in an abnormal state when the difference continues to be equal to or higher than the threshold value for a certain period of time, or when the difference is calculated to be equal to or higher than the threshold value a certain number of times or more.

The display processing unit 31e performs processing of displaying various images, characters, etc. on the display unit 34. In this embodiment, the display processing unit 31e displays the determination result of the abnormality determination unit 31d, for example. In addition, the display processing unit 31e may also display the acquired program data, actual measured pressure values, estimated pressure values, or time series data recorded in the database 32c of the memory unit 32 as described below.

The time series data creation unit 31f performs processing of creating time series data. Specifically, the difference between the actual measured pressure value and the estimated pressure value corresponds to the elapsed time from the time when the semiconductor manufacturing apparatus 1 started manufacturing the semiconductor, and is set as time series data. In addition, the measured pressure value and the estimated pressure value can also be included in the time series data. In addition, the difference can also be made to correspond to the actual time.

The data recording unit 31g records the time series data generated by the time series data generating unit in the database 32c of the memory unit 32. The data recording unit 31g registers the time series data in the time series data table of the database 32c every time the time series data creation unit 31f creates the time series data. Details on the time series data form will be explained later.

FIG. 3 is a block diagram showing the structure of the server device 5 of this embodiment. The server device 5 of this embodiment is composed of a processing unit 51, a memory unit 52, a communication unit 53, a card slot 54, and the like. In addition, in this embodiment, although the description is made assuming that one server device 5 performs the processing, a plurality of server devices 5 may perform the processing in a distributed manner. In addition, the processing performed by the server device 5 can also be performed by the information processing device 30 .

The processing unit 51 is configured using a computing processing device such as a CPU, MPU, or GPU. The processing unit 51 reads and executes the server program 52a stored in the memory unit 52, thereby performing various processes such as learning processing of the learning model 32b used by the information processing device 30.

The memory unit 52 is configured using a large-capacity memory device such as a hard disk or an SSD. The memory unit 52 stores various programs executed by the processing unit 51 and various data required for processing by the processing unit 51 . In this embodiment, the storage unit 52 stores the server program 52a executed by the processing unit 51. In this embodiment, the server program 52a is provided in the form of being recorded on a recording medium 99 such as a memory card, and the server device 5 reads the server program 52a from the recording medium 99 and stores it in the storage unit 52 . However, the server program 52a may also be written in the memory unit 52 during the manufacturing stage of the server device 5, for example. For example, the server program 52a may be read from the recording medium 99 by a writing device and written in the memory unit 52 of the server device 5 . The server program 52a may also be provided in the form of distribution over the Internet.

The communication unit 53 communicates with various devices through a network N including an intra-company LAN, a wireless LAN, the Internet, and the like. The communication unit 53 transmits data provided from the processing unit 51 to other devices, and provides data received from other devices to the processing unit 51 .

The card slot 54 can be used to detach and install a recording medium 99 such as a memory card, so as to read and write data to the installed recording medium 99 . The card slot 54 reads data recorded in the recording medium 99 and supplies the data to the processing unit 51 , and writes the data supplied from the processing unit 51 to the recording medium 99 . In this embodiment, the server device 5 writes the learning model 32 b that has been subjected to the learning process in the recording medium 99 through the card slot 56 , and the server device 5 supplies the learning model 32 b to the semiconductor manufacturing device 1 . Information processing device 30. In addition, in this embodiment, a log including a plurality of process data recorded by the information processing device 30 when the semiconductor manufacturing apparatus 1 is in a normal state and a pressure value measured by the vacuum gauge 76 when the exhaust system is in a normal state is used ( The log) data 52b is provided by writing it in the recording medium 99. The server device 5 reads and obtains the log data 52b from the recording medium 99 through the card slot 54 to use it as training data for learning processing.

In addition, the storage unit 52 may be an external storage device connected to the server device 5 . In addition, the server device 5 may be a multi-computer composed of a plurality of computers, or may be a virtual machine constructed virtually through software. In addition, the server device 5 is not limited to the above-mentioned structure, and may include, for example, an operation unit that accepts operation input, a display unit that displays an image, or the like.

In addition, in the processing unit 51 of the server device 5 of this embodiment, the server program 52a stored in the memory unit 52 is read and executed by the processing unit 51, thereby allowing the learning processing unit 51a and the like to function as software. The functional part is implemented. In addition, these functional units are functional units related to the process of generating the learning model 32b, and illustrations and descriptions of other functional units are omitted.

The learning processing unit 51a performs processing to generate the learning model 32b by machine learning the training data. The learning processing unit 51a performs learning processing in the initial learning stage of the learning model 32b, that is, the stage in which the learning model 32b is first generated, using training data prepared in advance by the administrator or developer of the system. The training data used at this time can be created using data accumulated through the operation of semiconductor manufacturing equipment so far.

Figure 4 is an explanatory diagram showing the learning model. The learning model 32b of this embodiment is generated by machine learning using a neural network (Neural Network), for example. However, machine learning can also be performed by methods other than neural networks. For example, various machine learning methods such as LSTM (Long Short Term Memory), Transformer (Transformer), SVM (Support Vector Machine, Support Vector Machine) or k-nearest neighbor method can be used.

The input layer included in the learning model 32b of this embodiment has a plurality of neurons (neurons) that accept input of various program data, and delivers the input program data to the intermediate layer. The intermediate layer has a plurality of neurons that extract feature quantities of the program data, and delivers the extracted feature quantities to the output layer. The output layer has neurons that output estimated pressure values, and outputs estimated pressure values based on feature quantities output from the intermediate layer. In addition, the program data input to the learning model 32b may also be time series data.

In the machine learning of the learning model 32b of this embodiment, normal-time training data can be used. The normal-time training data uses a plurality of normal-time program data as input information, and assumes that the exhaust system is a normal-time vacuum. The pressure value measured by the meter 76 is obtained as the output information. When the normal program data is input, machine learning using a neural network is performed to output the same pressure value as the pressure value measured by the vacuum gauge 76 during the normal time. The program data and pressure value at normal times are, for example, the program data and pressure value at the time of trial operation after installation or maintenance of the semiconductor manufacturing apparatus 1 . The learning model 32b performs learning using normal training data obtained by combining a plurality of program data and pressure values.

The machine learning of the learning model 32b is not limited to the first time. The learning model 32b can also be re-learned using new training data if necessary, or new training data can be added to the current training data to re-learn. At this time, the server device 50 can also regularly transmit the re-learned learning model 32b to the information processing device 30, and the processing unit 31 of the information processing device 30 can also read the re-learned learning model 32b.

In this embodiment, the learning process of the learning model 32b is performed in the server device 5 (see FIG. 3). The server device 5 uses the above-mentioned training data prepared in advance and performs mechanical learning through a neural network, thereby generating the learning model 32b. The learning model 32b generated by the server device 5 is provided to the information processing device 30 via the recording medium 99. In addition, the learning model 32b can also be learned in the information processing device 30.

Figure 5 is an explanatory diagram showing a time series data form. The processing unit 31 of the information processing device 30 acquires the program data and the actual measured pressure value every 10 seconds, for example, and outputs the estimated pressure value. The processing unit 31 calculates the difference between the acquired actual measured pressure value and the estimated pressure value, and registers the difference in the memory unit 32 in association with the elapsed time from the start of semiconductor manufacturing to the acquisition of the program data and the actual measured pressure value. The database 32c. This processing is performed one by one, and a time series data form is created in the database 32c. In addition, as shown in FIG. 5 , in the time series data sheet, the measured pressure value and the estimated pressure value can also be registered in association with the elapsed time.

FIG. 6 is a schematic diagram showing an example of a screen displayed by the information processing device. The display unit 34 of the information processing device 30 displays a time series data form. In addition, as shown in FIG. 6 , the information processing device 30 can also display charts based on the time series data form. The graph displayed on the display unit 34 displays the actual measured pressure value, the estimated pressure value, and the difference, with the vertical axis as the pressure value and the horizontal axis as the elapsed time. In addition, the graph shows a threshold value (for example, 100 Pa) as a criterion for determining whether or not the difference is abnormal. It can be confirmed that the information processing device 30 determines that the semiconductor manufacturing device is a semiconductor manufacturing device because the difference exceeds the threshold value. 1 abnormal status. In this embodiment, the actual measured pressure value is shown as a dotted line, the estimated pressure value is shown as a dotted line, and the difference is shown as a solid line. Furthermore, when it is determined that the processing unit 31 is in an abnormal state, the display unit 34 displays a warning message notifying the user of the abnormal state. In addition, when the information processing device 30 determines that the processing unit 31 is in an abnormal state, it may also notify the user of the abnormal state by changing the background color of the screen, lighting a lamp, or outputting a sound.

Figure 7 is a flow chart showing the procedure for generating a learning model. The information processing device 30 obtains program data from the semiconductor manufacturing apparatus 1 in normal condition (S1), and further obtains the pressure value measured by the vacuum gauge 76 of the semiconductor manufacturing apparatus 1 (S2). The information processing device 30 associates the program data with the pressure value to create training data (S3). The information processing device 30 determines whether the generated training data is more than a certain number sufficient to generate the learning model 32b (S4). When the number of training data does not reach a certain number (S4: NO), the changes flow into the chamber. The type of gas or gas flow rate and other conditions within 11 are determined (S5), and the program data is obtained again (S1). When the amount of training data is more than a certain number (S4: YES), the information processing device 30 transmits the training data to the server device 5 (S6), and causes the server device 5 to generate the learning model 32b (S7) .

FIG. 8 is a flowchart showing a procedure of processing performed by the information processing device. The information processing apparatus 30 causes the semiconductor manufacturing apparatus 1 to start semiconductor manufacturing (S11). The information processing device 30 obtains the program data from the semiconductor manufacturing apparatus 1 (S12), and further obtains the actual measured pressure value from the vacuum gauge 76 of the semiconductor manufacturing apparatus 1 (S13). The information processing device 30 inputs the acquired program data into the learning model 32b (S14), and outputs the estimated pressure value (S15). The information processing device 30 calculates the difference between the obtained actual measured pressure value and the output estimated pressure value (S16), and associates the calculated difference with time to create time series data (S17). The information processing device 30 determines whether the difference in the time series data is greater than a threshold value (S18). When the difference does not reach the threshold value (S18: No), the information processing device 30 returns the process to S12. When the difference is greater than the threshold value (S18: Yes), the information processing device 30 displays a warning message on the display unit 34 (S19), and determines whether to terminate the process (S20). The information processing device 30 ends the processing, for example, when an instruction to end the processing is input through the operation unit 35 (S20: Yes). When the process is not ended (S20: No), the information processing device 30 returns the process to S12 and obtains the program data again.

According to the program of this embodiment and the processing performed by the information processing device 30, an abnormality in the semiconductor manufacturing device 1 can be determined with high accuracy. Thereby, the user of the semiconductor manufacturing apparatus 1 can cope with the abnormal state. In addition, by visualizing the transition of the difference between the actual measured pressure value and the estimated pressure value, the user can also speculate on the cause of the abnormality.

(Embodiment 2) The information processing device 30 may also predict the time when the difference becomes the predetermined value or more when the difference between the actual measured pressure value and the estimated pressure value is not more than a predetermined value (threshold value). The program, information processing device, etc. of Embodiment 2 will be described below with reference to the drawings. Among the structures of Embodiment 2, the same components as those of Embodiment 1 are designated by the same reference numerals, and detailed description thereof is omitted.

FIG. 9 is a block diagram showing the structure of the information processing device according to Embodiment 2. The information processing device 30 of this embodiment further includes a prediction unit 31h in addition to the functional units of the first embodiment. The prediction unit 31h predicts the time when the difference between the actual measured pressure value and the estimated pressure value becomes more than a threshold value based on the time series data recorded in the database 32c of the memory unit 32. Specifically, an approximate function representing the correlation between time and differences in time series data is obtained by, for example, the least squares method or the maximum approximation method, and based on the approximate function, it is predicted how many seconds until the difference becomes a critical value or more, the semiconductor manufacturing apparatus 1 becomes abnormal. In addition, the information processing device 30 can also use learning models such as seq2seq (sequence to sequence), LSTM, or Transformer to predict the time when the abnormal state becomes abnormal. At this time, time series data of multiple periods are used in the training data used to generate the learning model.

FIG. 10 is a schematic diagram showing an example of a screen displayed by the information processing device according to Embodiment 2. When the information processing device 30 of this embodiment determines that the semiconductor manufacturing device 1 is not in an abnormal state, it displays the predicted time until the difference between the actual measured pressure value and the estimated pressure value becomes more than a threshold value and the semiconductor manufacturing device 1 becomes an abnormal state. . In addition, as shown in FIG. 10 , a one-dot chain line may be used to display the prediction of the transition of the graph showing the difference. In addition, when the information processing device 30 determines that the semiconductor manufacturing apparatus 1 is in an abnormal state, similarly to Embodiment 1, the information processing device 30 displays a warning message notifying the user of the abnormal state (see FIG. 6 ). In addition, when the change amount of the difference and difference transition is small and it is impossible to predict the time when it will exceed the threshold value, the predicted time is not displayed.

FIG. 11 is a flowchart showing the processing procedure performed by the information processing device according to the second embodiment. S21 to S30 are the same processes as S11 to S20 of Embodiment 1. When the difference between the actual measured pressure value and the estimated pressure value does not reach the threshold value (S28: No), the information processing device 30 determines whether it is possible to predict the time until the semiconductor manufacturing apparatus 1 becomes an abnormal state (S31). When time prediction is possible (S31: Yes), the time until the difference between the actual measured pressure value and the estimated pressure value becomes more than the threshold value and the semiconductor manufacturing apparatus 1 becomes an abnormal state is predicted (S32), and the predicted time is displayed. to the display unit 34 (S33), and the process returns to S22. When time prediction cannot be performed (S31: No), the information processing device 30 returns the process to S22.

According to the program of this embodiment and the processing performed by the information processing device 30, when there is a sign that the semiconductor manufacturing apparatus 1 becomes an abnormal state, it is possible to predict the time when the abnormal state will occur. Thereby, the user of the semiconductor manufacturing apparatus 1 can respond to abnormal conditions or avoid abnormal conditions.

(Embodiment 3) The server device 5 of Embodiment 3 generates a second learning model by fine-tuning a learning model 32b (first learning model) by including the data from the semiconductor manufacturing device 1 (first semiconductor manufacturing device). ) obtained by learning the first training data of the pressure value (first pressure value) obtained by the vacuum gauge 76. When a semiconductor manufacturing apparatus is installed at a place of use, the type of pump or the layout or length of the exhaust pipe is changed depending on the place of use. The information processing device of this embodiment outputs the estimated pressure value of the semiconductor manufacturing device after changing the type of pump or the layout or length of the exhaust pipe by using the second learning model generated by finely adjusting the first learning model. .

FIG. 12 is a schematic diagram showing the schematic configuration of the second semiconductor manufacturing apparatus. The second learning model is used to output an estimated pressure value of the second semiconductor manufacturing apparatus 1B that is different from the semiconductor manufacturing apparatus 1 (first semiconductor manufacturing apparatus). The second semiconductor manufacturing apparatus 1B is different from the first semiconductor manufacturing apparatus 1 because, for example, the types of the first pump 74 and the second pump 75 or the layout or length of the exhaust pipe 77 as shown in FIG. 12 are different from the first semiconductor manufacturing apparatus 1B. There may be cases where the correct estimated pressure value cannot be output when using the learning model 32b. The server device 5 is based on the pressure value measured by the vacuum gauge 76 of the second semiconductor manufacturing device 1B when the exhaust system is normal using a plurality of program data of the second semiconductor manufacturing device 1B as input information. The second pressure value) is used as the second training data obtained from the output information to fine-tune the learning model 32b and generate the second learning model. Although the amount of the second training data used when generating the second learning model is smaller than the amount of training data (first training data) used when generating the learning model 32b, more than the amount of the first training data may be used. the second training material.

FIG. 13 is a flow chart showing the procedure for generating the second learning model. The information processing device 30B (see FIG. 12 ) of the second semiconductor manufacturing device 1B acquires the program data from the second semiconductor manufacturing device 1B (S41), and further acquires the pressure value from the vacuum gauge 76 of the second semiconductor manufacturing device 1B (S42). ). The information processing device 30B associates the program data with the pressure value to create second training data (S43). The information processing device 30B determines whether the number of second training data is sufficient to perform fine adjustment or exceeds a certain number (S44). When the number of second training data does not reach a certain number (S44: No), the change flows into the second semiconductor manufacturing process. Conditions such as the type of gas or the flow rate of the gas in the chamber 11 of the device 1B (S45), and the program data is obtained again (S41). When the number of second training data is more than a certain number (S44: Yes), the information processing device 30B transmits the second training data to the server device 5 (S46), and causes the server device 5 to read the first learning model (S47), and generates a second learning model through fine adjustment (S48).

According to the model generation method of this embodiment, it is possible to quickly provide a learning model for estimated pressure values of semiconductor manufacturing equipment that differs in the type of output pump or the layout or length of the exhaust pipe.

(Modification) In each of the above-described embodiments, the processing unit 31 of the information processing device 30 determines whether the exhaust system is in an abnormal state by comparing the estimated pressure value and the actual measured pressure value. However, it is not limited to this. In the memory unit 32 of the information processing device 30, the range of the pressure value of the exhaust system that is expected when the exhaust system is normal is stored. The processing unit 31 may also assume that the estimated pressure value is the range stored in the memory unit 32. When the pressure value is outside the range, it is determined that the exhaust system is in an abnormal state.

In each of the above-described embodiments, the processing unit 31 of the information processing device 30 determines whether the exhaust system is in an abnormal state in the manufacturing process state (during the manufacturing process) when gas is supplied into the chamber, but the invention is not limited to this. The processing unit 31 may determine whether the exhaust system is in an abnormal state in the idle state of the device before starting the manufacturing process. In addition, in this case, the program data includes data or measured values obtained from various machines or measuring devices when the device is idle.

(Additional explanation) The following is a supplementary explanation of each status of the device. Figure 14 is a table comparing the process data and machine status in the device startup status, device idle status, and manufacturing process status. The starting state of the device is a state when the pressure in the chamber is vacuum pumped to below a predetermined reference pressure. The first pump 74 and the second pump 75 are started to vacuum pump the pressure in the chamber from atmospheric pressure to below the reference pressure. . The heater is started and the temperature rises from normal temperature to the preset temperature. A chiller (lower electrode cooling coal) that cools the back side of the substrate S is started, and the temperature rises and falls from normal temperature to a preset temperature. The air valve 71a is closed. The APC opening changes from the closed state of 0% opening to the maximum opening of 100% opening. The pressure in the chamber fluctuates due to vacuum suction from atmospheric pressure to below the reference pressure. The plasma source system is stopped.

In the idle state of the device, the first pump 74 and the second pump 75 are running, and the heater and the cooling device are controlled to be within a preset temperature range. The air valve 71a is closed, and the APC opening is fixed at the maximum opening (100%). The pressure in the chamber is a high vacuum state below the reference pressure, and the plasma source is stopped.

In the manufacturing process state, the first pump 74 and the second pump 75 are running, and the heater and the cooling device are controlled to be within a preset temperature range. The air valve 71a is opened. In addition, the MFC controls the gas flow rate in the chamber to be the flow rate set in the program setting indicating the substrate processing conditions. The APC opening is controlled so that the pressure in the chamber changes to the pressure set in the program setting. The pressure in the chamber becomes the pressure set in the program setting. The plasma source applies high-frequency power set in the program settings.

As shown in Figure 14, in the idle state of the device, program data changes less. Therefore, the processing unit 31 of the information processing device 30 can also determine with high accuracy whether the exhaust system is in an abnormal state by determining whether the estimated pressure value is within the range of the pressure value stored in the memory unit 32 . In addition, the processing unit 31 may determine that the exhaust system is in an abnormal state when the estimated pressure value is outside the range of the pressure value stored in the memory unit 32 during the manufacturing process state.

The following is a supplementary explanation of the APC opening. FIG. 15 is a graph showing an example of changes in chamber pressure and APC opening. The semiconductor manufacturing apparatus 1 enters the apparatus idle state when the pressure in the chamber becomes less than a predetermined reference pressure value after performing vacuum suction to reduce the pressure in the chamber while the apparatus is being started. Thereafter, an instruction to start the semiconductor manufacturing process is input to the semiconductor manufacturing apparatus 1, and the APC opening is changed to increase the pressure in the chamber to the pressure (set pressure) set in the program setting, thereby entering the manufacturing process state.

The horizontal axis of the graph shown in FIG. 15 shows the elapsed time (seconds) after the instruction to start the semiconductor manufacturing process is input, and the vertical axis shows the APC opening (%) or the chamber pressure (Pa). In the graph of FIG. 15 , the transition of the pressure in the chamber is shown by a solid line, and the transition of the APC opening is shown by a dotted line.

When the device is starting up, the APC opening is controlled from 0% to 100%. Thereby, the pressure in the chamber is reduced from atmospheric pressure to below the reference pressure, and the device becomes an idle state. When moving from the device idle state to the manufacturing process state, the semiconductor manufacturing apparatus 1 reduces the APC opening to about 5%, for example, as shown in FIG. 15 , and maintains the pressure in the chamber at the set pressure. In the example shown in Fig. 15, the set pressure is 4Pa. In addition, as shown near 10 seconds to 11 seconds in the graph of Fig. 15, the pressure in the chamber may change from the set pressure. The pressure in the chamber changes, for example, when the plasma source starts applying high-frequency power. When the pressure in the chamber fluctuates, the semiconductor manufacturing apparatus 1 controls the APC so that the pressure in the chamber becomes a designated pressure and changes the opening of the APC. Therefore, the APC opening degree variable data included in the program data of the learning model is input.

All views of the embodiments disclosed this time are illustrative and should not be construed as limiting the present invention. The technical features described in each embodiment can be combined with each other, and the scope of the present invention is intended to include all changes within the scope of the patent application and a range equal to the scope of the patent application. In addition, the independent claims and dependent claims described in the patent application scope may be combined with each other in various combinations regardless of the form of reference. Furthermore, although the form in which a claim citing two or more other claims is described in the scope of the patent application (multiple appendix form) is used, the invention is not limited to this. It may also be recorded in the form of multiple appendices (multiple attachments, multiple requests) that cite at least one appendage.

1,1B:Semiconductor manufacturing equipment 5,50:Server device 11: Chamber 13: Sample table 30,30B:Information processing device 31:Processing Department 31a: Procedure data acquisition department 31b: Measured pressure value acquisition part 31c: Estimated pressure value output part 31d: Abnormality determination department 31e: Display processing department 31f: Time series data creation department 31g:Data Recording Department 31h: Forecasting Department 32,52:Memory Department 32a: Program 32b: Learning model 32c:Database 33,53: Ministry of Communications 34:Display part 35:Operation Department 36,54,56: Card slot 51:Processing Department 51a: Learning Processing Department 52a:Server program 52b: Log data 71:MFC 71a:Air valve 72: Vacuum gauge 73:APC 74:First pump 75:Second pump 76: Vacuum gauge 77:Exhaust pipe 78: Pressure flow meter 79: Thermometer 99:Recording media

FIG. 1 is a schematic diagram showing the schematic structure of a semiconductor manufacturing apparatus. FIG. 2 is a block diagram showing the structure of the information processing device. Figure 3 is a block diagram showing the structure of the server device. Figure 4 is an explanatory diagram showing the learning model. Figure 5 is an explanatory diagram showing a time series data form. FIG. 6 is a schematic diagram showing an example of a screen displayed by the information processing device. Figure 7 is a flow chart showing the procedure for generating a learning model. FIG. 8 is a flowchart showing a procedure of processing performed by the information processing device. FIG. 9 is a block diagram showing the structure of the information processing device according to Embodiment 2. FIG. 10 is a schematic diagram showing an example of a screen displayed by the information processing device according to Embodiment 2. FIG. 11 is a flowchart showing the processing procedure performed by the information processing device according to the second embodiment. FIG. 12 is a schematic diagram showing the schematic configuration of the second semiconductor manufacturing apparatus. FIG. 13 is a flow chart showing the procedure for generating the second learning model. Figure 14 is a table comparing the process data and machine status in the device startup status, device idle status, and manufacturing process status. FIG. 15 is a graph showing an example of changes in chamber pressure and APC opening.

1:Semiconductor manufacturing equipment

5:Server device

11: Chamber

13: Sample table

30:Information processing device

71:MFC

71a:Air valve

72: Vacuum gauge

73:APC

74:First pump

75:Second pump

76: Vacuum gauge

77:Exhaust pipe

78: Pressure flow meter

79: Thermometer

S:Substrate

Claims (12)

A program that causes a computer to: Obtain process data including at least two of the type and flow rate of gas supplied to the semiconductor manufacturing equipment, the pressure in the chamber, or the valve opening of the automatic pressure control device, For a learning model that is learned to output the estimated pressure value in the semiconductor manufacturing device when the aforementioned program data is input, the aforementioned program data is input and the aforementioned estimated pressure value is output, And it is determined whether it is an abnormal state based on the aforementioned estimated pressure value. The program of claim 1 obtains the actual pressure value measured by the vacuum gauge in the semiconductor manufacturing device, and determines whether the abnormal state is based on the estimated pressure value and the actual measured pressure value. For example, the program of claim 1 or 2 obtains the process data in the idle state of the device before starting the manufacturing process of supplying gas to the semiconductor device. For example, the program of claim 1 or 2 obtains the aforementioned process data during the manufacturing process of supplying gas to the semiconductor device. For example, claim the program of any one of items 1 to 4, wherein the aforementioned program data includes the operating status of the exhaust pump. For example, according to the program of claim 2, when the difference between the actual measured pressure value and the estimated pressure value is more than a predetermined value, an abnormal state is determined. For example, the program of claim 2 or 6 continuously obtains the difference between the aforementioned estimated pressure value and the aforementioned measured pressure value, Create time series data in which the aforementioned difference corresponds to time, And the aforementioned time series data is displayed on the display unit. For example, the program of claim 7 predicts the time when the difference will exceed a predetermined value based on the time series data. An information processing method that obtains program data including at least two of the type and flow rate of gas supplied to a semiconductor manufacturing device, the pressure in a chamber, or the valve opening of an automatic pressure control device, For a learning model that is learned to output the estimated pressure value in the semiconductor manufacturing device when the aforementioned program data is input, the aforementioned program data is input and the aforementioned estimated pressure value is output, And it is determined whether it is an abnormal state based on the aforementioned estimated pressure value. An information processing device having: The process data acquisition unit acquires process data including at least two of the type and flow rate of gas supplied to the semiconductor manufacturing equipment, the pressure in the chamber, or the valve opening of the automatic pressure control device; The estimated pressure value output unit inputs the program data and outputs the estimated pressure value to the learning model that is learned to output the estimated pressure value in the semiconductor manufacturing device when the program data is input; and The abnormality determination unit determines whether the abnormality is in an abnormal state based on the estimated pressure value. A method for generating a model is to obtain the pressure value of a vacuum gauge in a semiconductor manufacturing device, Obtain program data including at least two of the type and flow rate of the gas supplied to the aforementioned semiconductor manufacturing equipment, the pressure in the chamber or the valve opening of the automatic pressure control device, and the training data obtained as the aforementioned pressure value, Based on the acquired training data, when program data including at least two of the type and flow rate of the gas supplied to the semiconductor manufacturing equipment, the pressure in the chamber, or the valve opening of the automatic pressure control device is input, an estimated pressure is output value learning model. A method for generating a model is to obtain a first pressure value of a vacuum gauge in a first semiconductor manufacturing device, Obtaining program data including at least two of the type and flow rate of the gas supplied to the first semiconductor manufacturing device, the pressure in the chamber, or the valve opening of the automatic pressure control device, and the obtained third pressure value a training material, Based on the obtained first training data, program data including at least two of the type and flow rate of the gas supplied to the first semiconductor manufacturing device, the pressure in the chamber, or the valve opening of the automatic pressure control device is generated. , the first learning model that outputs the estimated pressure value, Obtain the second pressure value of the vacuum gauge in the second semiconductor manufacturing device, Obtaining program data including at least two of the type and flow rate of the gas supplied to the second semiconductor manufacturing device, the pressure in the chamber or the valve opening of the automatic pressure control device, and the number of the obtained second pressure values a second training material that is less than the aforementioned first training material, Based on the obtained second training data, the type and flow rate of the gas supplied to the second semiconductor manufacturing device, the pressure in the chamber, or the valve opening of the automatic pressure control device are generated by finely adjusting the first learning model. When at least two of the program data are included, a second learning model of the estimated pressure value is output.

Images