Prediction method, plasma processing device, model generation method, and information processing device

Abstract

This prediction method includes a step (a) and a step (b). Step (a) involves acquiring the frequency of radio frequency (RF) power being applied to an electrode that generates plasma inside a chamber, and the impedance of the chamber in which plasma is being generated by the RF power from a matching circuit provided in a power feed line through which the RF power flows. Step (b) involves using a predictive model that predicts a change in the impedance of the chamber from at least the frequency of the RF power and time-series data of the impedance of the chamber, to predict a change in the impedance of the chamber from the frequency of the RF power and the time-series data of the impedance of the chamber acquired in step (a).

Description

Prediction method, plasma processing apparatus, model generation method, and information processing apparatus

[0001] The present disclosure relates to a prediction method, a plasma processing apparatus, a model generation method, and an information processing apparatus.

[0002] The following Patent Document 1 describes a specification that "identifies a model corresponding to a processing apparatus for processing a substrate based on a recipe describing conditions for processing the substrate, and a control algorithm for generating a control signal for controlling the processing apparatus so that the state of the processing apparatus becomes a predetermined state based on measured values ​​measured by a measuring instrument provided in the processing apparatus; a simulator for simulating the state of the processing apparatus using the model; a first control signal generation unit for generating the control signal based on the measured values ​​using the control algorithm identified by the identification unit, and inputting the generated control signal to the processing apparatus; and a simulator for simulating the state of the processing apparatus using the control algorithm identified by the identification unit, a second control signal generation unit that generates the control signal based on an output value of a simulator and inputs the generated control signal to the simulator; a first adjustment unit that adjusts values ​​of model parameters included in the model so that a difference between the measurement value and the output value of the simulator becomes small; and a second adjustment unit that adjusts values ​​of control parameters included in the control algorithm used by the second control signal generation unit so that an evaluation value calculated for the output value of the simulator using the model including the adjusted values ​​of the model parameters approaches a target value, wherein the first control signal generation unit generates the control signal using the value of the control parameter adjusted by the second adjustment unit.

[0003] Japanese Patent Application Laid-Open No. 2021-085053

[0004] The present disclosure provides techniques for predicting changes in chamber impedance.

[0005] A prediction method according to one aspect of the present disclosure includes steps (a) and (b). In step (a), the frequency of radio frequency (RF) power applied to an electrode that generates plasma inside the chamber and the impedance of the chamber in which plasma is generated by the RF power are acquired from a matching circuit provided in a feeder line through which the RF power flows. In step (b), a prediction model that predicts changes in the chamber impedance from time-series data of the RF power frequency and the chamber impedance acquired in step (a) are used to predict changes in the chamber impedance from the time-series data of the RF power frequency and the chamber impedance.

[0006] According to the present disclosure, changes in chamber impedance can be predicted.

[0007] FIG. 1 is a diagram schematically illustrating a configuration of a plasma processing apparatus according to an embodiment. FIG. 2 is a diagram schematically illustrating an example of a configuration of an RF power supply unit according to an embodiment. FIG. 3 is a diagram schematically illustrating an example of a configuration of a control device according to an embodiment. FIG. 4 is a diagram illustrating an example of a change in chamber impedance according to an embodiment. FIG. 5 is a diagram illustrating an example of a flow of machine learning according to an embodiment. FIG. 6 is a diagram illustrating an example of a flow of predicting chamber impedance according to an embodiment. FIG. 7 is a diagram illustrating an example of a flow of predicting chamber impedance by repeating calculations of a prediction model according to an embodiment. FIG. 8 is a flowchart illustrating an example of a flow of model generation processing according to an embodiment. FIG. 9 is a flowchart illustrating an example of a flow of matching processing according to an embodiment. FIG. 10 is a timing chart of matching processing according to a comparative example. FIG. 11 is a timing chart of matching processing according to an embodiment. FIG. 12 is a diagram illustrating an example of a schematic configuration of an information processing apparatus according to an embodiment.

[0008] Hereinafter, embodiments of the prediction method, plasma processing apparatus, model generation method, and information processing apparatus disclosed in the present application will be described in detail with reference to the drawings. Note that the disclosed prediction method, plasma processing apparatus, model generation method, and information processing apparatus are not limited to the embodiments.

[0009] In a plasma processing apparatus, a substrate is placed on a support provided within a chamber, and plasma processing is performed by generating plasma within the chamber. When generating plasma within the chamber, the plasma processing apparatus supplies radio frequency (RF) power to the chamber. The RF power is amplified by an amplifier and then supplied to the chamber. A matching circuit such as a variable capacitor is provided in the power supply line connecting the amplifier and the chamber. The RF power is supplied to the chamber by adjusting the capacitance of the matching circuit to match the impedance between the amplifier and the chamber.

[0010] The impedance of a chamber changes from moment to moment. Furthermore, it takes time for the capacitance of a matching circuit to change after the setting value for the matching circuit's capacitance is changed. Conventionally, plasma processing equipment has had to continually adjust the capacitance of the matching circuit in response to changes in the chamber's impedance, making it difficult to maintain a matched state.

[0011] Therefore, a technique for predicting changes in the impedance of the chamber is desired.

[0012] [Embodiment] [Apparatus Configuration] The configuration of a capacitively coupled plasma processing apparatus 1 will be described below as an example of a plasma processing apparatus according to the present disclosure. Fig. 1 is a diagram schematically illustrating the configuration of the plasma processing apparatus 1 according to the embodiment. The plasma processing apparatus 1 is an apparatus that performs plasma etching on a substrate W using plasma. The plasma processing apparatus 1 includes a chamber 10, a gas supply unit 20, an RF (Radio Frequency) power supply unit 30, and an exhaust system 40.

[0013] In this embodiment, the chamber 10 includes a support 11 and an upper electrode showerhead assembly 12. The support 11 is disposed in a lower region of a processing space 10s within the chamber 10. The upper electrode showerhead assembly 12 is disposed above the support 11 and can function as part of a ceiling plate of the chamber 10.

[0014] The support 11 is configured to support the substrate W in the processing space 10s. In this embodiment, the support 11 includes a lower electrode 111, an electrostatic chuck 112, and an edge ring 113. The electrostatic chuck 112 is disposed on the lower electrode 111 and is configured to support the substrate W on the upper surface of the electrostatic chuck 112. A heater (not shown) is provided within the electrostatic chuck 112, and the heater controls the temperature of the substrate W disposed on the electrostatic chuck 112. The edge ring 113 is disposed on the upper surface of the peripheral edge of the lower electrode 111 to surround the substrate W.

[0015] The upper electrode showerhead assembly 12 is configured to supply one or more gases from a gas supply unit 20 into the processing space 10s. In this embodiment, the upper electrode showerhead assembly 12 includes a gas inlet 12a, a gas diffusion chamber 12b, and multiple gas outlets 12c. The gas supply unit 20 and the gas diffusion chamber 12b are in fluid communication via the gas inlet 12a. The gas diffusion chamber 12b and the processing space 10s are in fluid communication via the multiple gas outlets 12c. In this embodiment, the upper electrode showerhead assembly 12 is configured to supply one or more gases from the gas inlet 12a into the processing space 10s via the gas diffusion chamber 12b and the multiple gas outlets 12c.

[0016] The gas supply unit 20 includes one or more gas sources 21 and flow controllers 22. In this embodiment, the gas supply unit 20 is configured to supply one or more process gases from the respective gas sources 21 to the gas inlet 12a via the respective flow controllers 22. The flow controllers 22 may include, for example, mass flow controllers or pressure-controlled flow controllers. Additionally, the gas supply unit 20 may include one or more flow modulation devices that modulate or pulse the flow rates of the one or more process gases.

[0017] The RF power supply 30 is configured to supply RF power, e.g., one or more RF signals, to one or more electrodes, such as the lower electrode 111, the upper electrode showerhead assembly 12, or both the lower electrode 111 and the upper electrode showerhead assembly 12. The RF power supply 30 corresponds to the supply unit of the plasma processing apparatus of the present disclosure. In this embodiment, the RF power supply 30 is configured to supply a first RF signal for plasma generation to the upper electrode showerhead assembly 12. This results in plasma formation from at least one process gas supplied to the processing space 10s. The frequency of the first RF signal includes a portion of the range of 3 Hz to 3000 GHz. For electronic material processes such as semiconductor processes, the frequency of the RF signal used for plasma generation is preferably within the range of 100 kHz to 3 GHz, more preferably 200 kHz to 150 MHz. For example, the frequency of the first RF signal may be within the range of 27 MHz to 100 MHz.

[0018] Although not shown, other embodiments are considered here. For example, the RF power supply unit 30 may be configured to further supply a second RF signal to the lower electrode 111. The frequency of the second RF signal may be, for example, within a range of 400 kHz to 13.56 MHz. Alternatively, a DC (Direct Current) pulse may be supplied to the lower electrode 111.

[0019] Furthermore, although not shown, other embodiments are contemplated herein. For example, the RF power supply 30 may be configured to supply a first RF signal to the lower electrode 111 and a second RF signal to the lower electrode 111. The RF power supply 30 may also be configured to supply a third RF signal to the upper electrode showerhead assembly 12. In addition, in other alternative embodiments, a DC voltage may be applied to the upper electrode showerhead assembly 12. Furthermore, in various embodiments, the amplitude of one or more RF signals (i.e., the first RF signal, the second RF signal, etc.) may be pulsed or modulated. Amplitude modulation may include pulsing the amplitude of an RF signal between an on state and an off state or between multiple different on states. Furthermore, phase matching of the RF signals may be controlled, and the phase matching of the amplitude modulation of multiple RF signals may be synchronized or asynchronous.

[0020] The exhaust system 40 may be connected to, for example, an exhaust port 10e provided at the bottom of the chamber 10. The exhaust system 40 may include a pressure valve, a vacuum pump such as a turbomolecular pump, a roughing pump, or a combination thereof. The exhaust system 40 evacuates the chamber 10 to a predetermined vacuum level during plasma processing.

[0021] The plasma processing apparatus 1 includes a control device 100. The operation of the plasma processing apparatus 1 is controlled comprehensively by the control device 100. The control device 100 controls each part of the plasma processing apparatus 1.

[0022] [Configuration of RF power supply unit 30] Next, a description will be given of the configuration of the RF power supply unit 30. Fig. 2 is a diagram schematically showing an example of the configuration of the RF power supply unit 30 according to the embodiment. Fig. 2 schematically shows the configuration of a path through which the first RF signal flows.

[0023] The RF power supply unit 30 includes a signal source 31, an input circuit 32, and an amplifier (AMP) 33. The signal source 31 is connected to the input circuit 32. The input circuit 32 is connected to the amplifier 33.

[0024] The signal source 31 generates an RF signal and outputs the generated RF signal to the input circuit 32. The signal source 31 is capable of changing the frequency of the RF signal it generates. The signal source 31 changes the frequency of the RF signal under the control of the control device 100. For example, the signal source 31 changes the frequency of the RF signal by changing the setting value to which the control device 100 sets the frequency.

[0025] The input circuit 32 matches the input impedance on the signal source 31 side with the impedance of the chamber 10. For example, when the impedance of the chamber 10 is 50Ω, the capacitance of the input circuit 32 is adjusted so that the input impedance on the signal source 31 side is 50Ω, thereby matching the input impedance on the signal source 31 side with the impedance of the chamber 10. The RF signal output from the signal source 31 is output to the amplifier 33 via the input circuit 32. In FIG. 2, the RF signal output to the amplifier 33 is ＲＦ It shows:

[0026] The amplifier 33 is connected to the chamber 10 via a power feed line. The amplifier 33 amplifies the power of the RF signal input from the input circuit 32 and outputs the amplified RF signal to the chamber 10. For example, the power of the RF signal output from the signal source 31 is about 10 mW. For example, the amplifier 33 amplifies the power of the RF signal to several kW and outputs it. The RF signal output from the amplifier 33 is P ＡＭＰ It shows:

[0027] The amplifier 33 is configured to be able to measure the power of the RF signal output from the amplifier 33. The amplifier 33 outputs the measured power of the RF signal to the control device 100. For example, the amplifier 33 measures the power of the RF signal P ＡＭＰ Measure the power of the RF signal P ＡＭＰ The measured power value is output to the control device 100.

[0028] The RF power supply unit 30 is provided with a matching circuit 34 on a power feed line connecting the amplifier 33 and the chamber 10. The matching circuit 34 is configured to be able to change its capacitance. For example, the matching circuit 34 has a built-in variable capacitor, and the capacitance can be changed by changing the capacitance of the variable capacitor. The capacitance of the matching circuit 34 changes under the control of the control device 100. For example, the capacitance of the matching circuit 34 changes when the control device 100 changes the setting value for the capacitance of the matching circuit 34.

[0029] The matching circuit 34 is also configured to be able to measure the impedance of the chamber 10. For example, the matching circuit 34 has a built-in sensor that detects the voltage and current of the path through which the RF signal flows, and is able to measure the impedance of the chamber 10 from the voltage and current detected by the sensor. The matching circuit 34 outputs the measured value of the impedance of the chamber 10 to the control device 100.

[0030] The capacitance of the matching circuit 34 is adjusted under the control of the control device 100 so as to match with the impedance of the chamber 10. By matching the impedance between the amplifier 33 and the chamber 10, the RF signal P ＡＭＰ is input to the chamber 10 via the matching circuit 34. The RF signal output from the matching circuit 34 is P ｉｎ Here, if there is an impedance mismatch between the amplifier 33 and the chamber 10, a part of the RF signal output from the matching circuit 34 is reflected from the chamber 10. The RF signal reflected from the chamber 10 is expressed as P ｒｅｆ As the impedance mismatch between the amplifier 33 and the chamber 10 increases, the proportion of the reflected RF signal P ｒｅｆ The power of the RF signal P ｒｅｆ When the power of the RF signal P becomes large, the power loss increases and may cause the plasma to misfire. ｒｅｆ If this is input to the amplifier 33, the amplifier 33 may generate heat, which may cause the amplifier 33 to break down.

[0031] [Device Configuration of Control Device 100] Next, an example of the configuration of the control device 100 according to the embodiment will be described. Fig. 3 is a diagram showing an example of a schematic configuration of the control device 100 according to the embodiment. The control device 100 is, for example, an information processing device such as a computer.

[0032] The control device 100 includes an external I / F (interface) unit 101, a user I / F unit 102, a storage unit 103, and a control unit 104. Note that the control device 100 may include various functional units that are included in known computers in addition to the functional units shown in FIG.

[0033] The external I / F unit 101 is an interface for inputting and outputting information to and from other devices. For example, the external I / F unit 101 is an interface for controlling communications with other devices. One embodiment of the external I / F unit 101 is a network interface card such as a LAN card. For example, the external I / F unit 101 transmits and receives various data to and from the plasma processing apparatus 1 or other devices via a network. The external I / F unit 101 may be an interface such as a USB (Universal Serial Bus) port.

[0034] The user I / F unit 102 includes a reception unit that receives various information input from a user via a keyboard, a touch panel, or the like, and a display unit that displays various information. The control device 100 receives various operations from the reception unit of the user I / F unit 102, and displays various information such as the device status and operating status of the plasma processing device 1 on the display unit.

[0035] The storage unit 103 is a storage device that stores various types of data. For example, the storage unit 103 is a storage device such as a hard disk, a solid state drive (SSD), or an optical disk. Note that the storage unit 103 may also be a data-rewritable semiconductor memory such as a random access memory (RAM), a flash memory, or a non-volatile static random access memory (NVSRAM).

[0036] The storage unit 103 stores an operating system (OS) and various programs executed by the control unit 104. For example, the storage unit 103 stores programs for executing a model generation process and a matching process, which will be described later. Furthermore, the storage unit 103 stores various data used by the programs executed by the control unit 104. For example, the storage unit 103 stores recipe data 103a, learning data 103b, and model data 103c. The storage unit 103 can also store other data in addition to the data exemplified above. The various programs and data may be stored in a computer-readable computer recording medium (e.g., a hard disk, an optical disk such as a DVD, a flexible disk, a semiconductor memory, etc.). The various programs and data may also be transmitted from other devices as needed and used online.

[0037] The recipe data 103a is data that stores process conditions for plasma processing to be performed on the substrate W. For example, the process conditions include the pressure in the chamber 10 that is set when performing the plasma processing. The process conditions also include the conditions for the gas supplied to the chamber 10 during the plasma processing. For example, the process conditions include, as gas conditions, the type of gas used in the plasma processing and the gas flow rate. The process conditions also include the conditions for the power supplied to the chamber 10 during the plasma processing. For example, the process conditions include, as power conditions, the frequency and power of the RF power applied to electrodes such as the first RF signal. The process conditions may also include various other conditions necessary for performing the plasma processing, such as the set temperature in the chamber 10 and the distance between the electrodes.

[0038] The learning data 103b is data for learning used to generate a prediction model. Details of the learning data 103b will be described later. The model data 103c is data of a prediction model that predicts a change in the impedance of the chamber 10.

[0039] The control unit 104 is a device that controls the control device 100. The control unit 104 may be an electronic circuit such as a central processing unit (CPU) or a microprocessing unit (MPU), or an integrated circuit such as an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA). The control unit 104 has an internal memory for storing programs and data. The control unit 104 reads various programs stored in the storage unit 103 and executes the processes of the read programs. The control unit 104 functions as various processing units when the programs run. For example, the control unit 104 includes a plasma processing control unit 104a, a matching control unit 104b, a generation unit 104c, an acquisition unit 104d, and a prediction unit 104e. In this embodiment, the control unit 104 has the functions of the plasma processing control unit 104a, the matching control unit 104b, the generation unit 104c, the acquisition unit 104d, and the prediction unit 104e. However, the functions of the plasma processing control unit 104a, the matching control unit 104b, the generating unit 104c, the acquiring unit 104d, and the predicting unit 104e may be distributed among a plurality of control units.

[0040] The plasma processing control unit 104a controls each part of the plasma processing apparatus 1 and controls the plasma processing.

[0041] Here, a brief description will be given of a flow of the plasma processing apparatus 1 according to the embodiment, under the control of the plasma processing control unit 104a, performing plasma processing on the substrate W. When performing plasma processing in the plasma processing apparatus 1, the substrate W is transported by a transport mechanism such as a transport arm (not shown) and placed on the support part 11.

[0042] The plasma processing controller 104a controls each component of the plasma processing apparatus 1 to perform plasma processing according to the process conditions of the recipe data 103a stored in the memory unit 103. For example, the plasma processing controller 104a controls the exhaust system 40 to evacuate the chamber 10 so that the pressure inside the chamber 10 matches the pressure inside the chamber 10 specified by the process conditions. The plasma processing controller 104a also controls the gas supply unit 20 to supply various gases from the gas supply unit 20 and introduce process gases into the chamber 10 through the upper electrode showerhead assembly 12 according to the gas conditions specified by the process conditions. For example, the plasma processing controller 104a controls the gas supply unit 20 to introduce various gases used in the plasma processing into the chamber 10 at flow rates specified by the gas conditions. The plasma processing controller 104a also controls the RF power supply unit 30 to supply RF power from the RF power supply unit 30 under the power conditions specified by the process conditions to generate plasma in the chamber 10 and perform plasma processing on the substrate W. For example, the plasma processing control unit 104a controls the RF power supply unit 30 to supply RF power such as a first RF signal from the RF power supply unit 30 at a frequency and power that satisfy the power conditions.

[0043] The matching control unit 104b controls at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the impedance of the chamber 10 measured by the matching circuit 34. For example, the matching control unit 104b controls the signal source 31 of the RF power supply unit 30 to adjust the RF signal P reflected from the chamber 10. ｒｅｆ For example, the matching control unit 104b controls the signal source 31 to sweep the frequency of the RF power, and the RF signal P reflected from the chamber 10 ｒｅｆ After changing the frequency of the RF power, the matching control unit 104b controls the matching circuit 34 to minimize the RF signal P reflected from the chamber 10. ｒｅｆ The matching control unit 104b may control the matching circuit 34 to match the impedance of the chamber 10 by changing only either the frequency of the RF power or the capacitance of the matching circuit 34.

[0044] The impedance of the chamber 10 changes from moment to moment. FIG. 4 is a diagram showing an example of the change in impedance of the chamber 10 according to the embodiment. FIG. 4 schematically shows a Smith chart, in which the impedance of the chamber 10 at times t-Δt, t, and t+Δt during plasma processing is plotted. Δt is, for example, 10 μsec. In this way, the impedance of the chamber 10 changes from moment to moment.

[0045] It takes time for the capacitance of the matching circuit 34 to change after the setting value for the variable capacitor capacitance is changed. For example, it takes about 100 msec for the capacitance of the matching circuit 34 to change after the setting value for the variable capacitor capacitance is changed. For this reason, in the plasma processing apparatus 1, it has been difficult to maintain a matched state even if the RF power frequency and the capacitance of the matching circuit 34 are adjusted in response to changes in the impedance of the chamber 10.

[0046] Therefore, the plasma processing apparatus 1 according to the embodiment uses a prediction model to predict the impedance of the chamber 10 during plasma processing. Then, the plasma processing apparatus 1 according to the embodiment controls at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the predicted impedance.

[0047] Here, the prediction model will be described.

[0048] The prediction model is generated by performing machine learning on the training data 103b. The plasma processing apparatus 1 first generates the training data 103b and stores it in the storage unit 103.

[0049] When generating the learning data 103b, the plasma processing apparatus 1 performs plasma processing and measures the impedance of the chamber 10, the frequency of the RF power, and the power of the RF power during the plasma processing. For example, during the plasma processing, the plasma processing apparatus 1 measures the impedance of the chamber 10, the frequency of the RF power, and the power of the RF power by adjusting the frequency of the RF power and the capacitance of the matching circuit 34 in a tracking manner so as to match the impedance of the chamber 10. The impedance of the chamber 10 is measured by the matching circuit 34. The frequency of the RF power may be measured from the set value of the frequency of the signal source 31, or may be measured by providing a sensor for measuring the frequency of the RF power in the RF power supply unit 30. The power of the RF power may be calculated using the amplification factor of the amplifier 33, and may be calculated based on the RF signal P measured by the amplifier 33. ＡＭＰ The plasma processing apparatus 1 then generates time-series data in which the impedance of the chamber 10 during plasma processing, the frequency of the RF power, and the power of the RF power are arranged in chronological order.

[0050] The plasma processing apparatus 1 performs plasma processing under various process conditions and generates time-series data for each process condition. For example, the plasma processing apparatus 1 performs plasma processing on the substrate W under process conditions and at least partially changed process conditions, and generates time-series data for each process condition. When there are multiple process conditions under which the plasma processing is performed on the substrate W, the plasma processing apparatus 1 performs plasma processing under various process conditions including the multiple process conditions under which the plasma processing is performed on the substrate W, and generates time-series data for each process condition.

[0051] The plasma processing apparatus 1 generates a data set including the process condition and time-series data for each process condition.

[0052] An example of a data set will be described. For example, if the process number k is the process number identifying the process conditions under which the plasma processing was performed, and the time elapsed since the start of the plasma processing is time t, the plasma processing apparatus 1 generates a data set including the following time series data for each process condition. Time t may be, for example, the time elapsed since the gas supply unit 20 started to supply gas, or the time elapsed since the plasma was ignited in the chamber 10.

[0053] (Data set) Impedance (t, k) RF power frequency (t, k) RF power power (t, k) Matching condition (t, k) Process condition (k)

[0054] The process number k is a number assigned to each process condition. The impedance (t, k) is the impedance of the chamber 10 at time t during plasma processing under the process conditions with the process number k. The RF power frequency (t, k) is the frequency of the RF power at time t during plasma processing under the process conditions with the process number k. The RF power power (t, k) is the power of the RF power at time t during plasma processing under the process conditions with the process number k. The matching condition is the matching condition of the matching circuit 34 at time t during plasma processing under the process conditions with the process number k. The capacitance of the matching circuit 34 is adjusted to match the impedance of the chamber 10 during plasma processing. The matching condition may be, for example, the capacitance of the matching circuit 34 or a setting value for setting the capacitance of the matching circuit 34. In this embodiment, the matching condition is the capacitance of the matching circuit 34. The process condition (k) is the process condition with the process number k.

[0055] The impedance (t, k), the frequency (t, k), the power (t, k), and the matching condition (t, k) included in the data set are time-series data. The plasma processing apparatus 1 stores the data set of each process condition in the storage unit 103 as learning data 103b.

[0056] The generation unit 104c performs machine learning using the training data 103b as training data to generate a prediction model that predicts changes in the impedance of the chamber 10. For example, the generation unit 104c reads out the process conditions and time-series data of each data set included in the training data 103b. The generation unit 104c performs machine learning using the process conditions and time-series data of each read data set to generate a prediction model that predicts changes in the impedance of the chamber 10. The model used for machine learning may be any model capable of machine learning time-series data. The model used for machine learning may be a neural network model capable of machine learning time-series data. Examples of models used for machine learning include a recurrent neural network (RNN), a gated recurrent unit (GRU), a long short-term memory (LSTM), and a transformer. The impedance of the chamber 10 changes due to changes in by-products generated during plasma processing, and deposits in the chamber 10 also change. Therefore, the impedance of the chamber 10 also changes depending on the history of the conditions in the chamber 10 since the start of plasma processing. For example, the impedance of the chamber 10 also changes depending on the frequency of the RF power supplied to the chamber 10 from the start of plasma processing and the history of the impedance of the chamber 10. For this reason, the plasma processing apparatus 1 according to this embodiment performs machine learning using time-series data on process conditions and during plasma processing. The generation unit 104c performs machine learning using the training data 103b on a neural network model capable of machine learning time-series data, and generates a prediction model that predicts changes in the impedance of the chamber 10.

[0057] FIG. 5 is a diagram illustrating an example of a machine learning flow according to an embodiment. FIG. 5 schematically illustrates an example of a model structure of a learning model 120 that performs machine learning on learning data 103b. The learning model 120 receives process conditions and time-series data T from a certain time to time t as input, and outputs impedance at time t+Δt, which is Δt ahead of time t. For example, the learning model 120 shown in FIG. 5 includes fully connected layers (Affine) 121 and 124, a model 122, and a flattening layer (Flatten) 123. The model 122 is a neural network model capable of machine learning time-series data, such as an RNN, GRU, LSTM, or Transformer. Here, for time-series data (t, k) of process conditions (k), the time-series data (t, k) from the start of plasma processing to time t is defined as time-series data (T, k). In the learning model 120 shown in FIG. 5 , the impedance (T, k), the RF power frequency (T, k), the RF power power (T, k), and the matching condition (T, k) from the start of plasma processing to time t are similarly expressed, and these are input to the model 122. Furthermore, the process condition (k) is input to the fully connected layer 121. The fully connected layer 121 performs an operation on the input process condition (k) and outputs the operation result data to the fully connected layer 124. The model 122 performs an operation on the input time-series data and outputs the operation result data to the flattening layer 123. The flattening layer 123 performs an operation on the data input from the model 122 and outputs the operation result data to the fully connected layer 124. The fully connected layer 124 performs an operation on the data input from the fully connected layer 121 and the flattening layer 123, and outputs a predicted value of the impedance at time t + Δt as the operation result. The learning model 120 is not limited to the model structure shown in FIG. 5, and may have any model structure as long as it is capable of predicting the change in impedance of the chamber 10.

[0058] In machine learning, learning is performed using the learning data 103b as training data, and parameters included in the learning model 120 are optimized so that the predicted impedance value approaches the measured impedance value. For example, in machine learning, for each process condition in the learning data 103b, the process condition (k) and time t are sequentially changed, and time series data (T, k) up to time t is input to the learning model 120. In machine learning, parameters included in the learning model 120 are optimized so that the predicted impedance value at time t + Δt output by the learning model 120 for each process condition approaches the measured impedance value at time t + Δt of the learning data 103b. The learning model 120 thus subjected to machine learning can predict changes in the impedance of the chamber 10. Note that the machine learning method is not limited to this, and any method may be used.

[0059] The generation unit 104c stores the learning model that has undergone machine learning as a prediction model that predicts changes in the impedance of the chamber 10 in the storage unit 103 as model data 103c.

[0060] Next, a flow of controlling at least one of the frequency of RF power and the capacitance of the matching circuit 34 using a prediction model during plasma processing by the plasma processing apparatus 1 according to the embodiment will be described.

[0061] During plasma processing, the acquisition unit 104d acquires the frequency of the RF power supplied to the chamber 10, the power of the RF power, and the impedance of the chamber 10 in which plasma is generated by the RF power. For example, during plasma processing, the amplifier 33 measures the power of the output RF signal. The matching circuit 34 measures the impedance of the chamber 10. During plasma processing, the acquisition unit 104d acquires the power of the RF signal from the amplifier 33 and acquires the impedance of the chamber 10 from the matching circuit 34. The frequency of the RF power may be acquired from the set value of the frequency of the signal source 31, or may be acquired by measuring the frequency of the RF power using a sensor provided in the RF power supply unit 30. During plasma processing, the acquisition unit 104d acquires the frequency of the RF power, the power of the RF signal, and the impedance of the chamber 10 at a predetermined period.

[0062] Furthermore, during plasma processing, the acquiring unit 104d acquires a matching condition of the matching circuit 34. The matching condition may be the capacitance of the matching circuit 34, or may be a set value that sets the capacitance of the matching circuit 34. For example, during plasma processing, the acquiring unit 104d acquires the capacitance of the matching circuit 34 at a predetermined period.

[0063] The acquiring unit 104d also acquires process conditions for the plasma processing currently being performed from the recipe data 103a.

[0064] The prediction unit 104e predicts the impedance of the chamber 10 using the prediction model stored in the model data 103c.

[0065] The prediction unit 104e generates time-series data from the start of plasma processing to the present regarding the impedance of the chamber 10, the frequency of the RF power, the power of the RF signal, and the matching conditions acquired by the acquisition unit 104d. For example, when the process condition (k) of the plasma processing currently being performed is set as k, and the current elapsed time of the plasma processing is set as t, the prediction unit 104e generates the impedance (T, k), the frequency (T, k) of the RF power, the power (T, k), and the matching conditions (T, k).

[0066] The prediction unit 104e predicts the impedance of the chamber 10 from the impedance (T, k), the frequency of the RF power (T, k), the power of the RF power (T, k), the matching conditions (T, k), and the process conditions (k) using a prediction model stored in the model data 103c.

[0067] 6 is a diagram illustrating an example of a flow for predicting the impedance of the chamber 10 according to the embodiment. Fig. 6 schematically illustrates an example of the model structure of the prediction model 130. The prediction model 130 is a learning model that has undergone machine learning, and therefore has the same model structure as the learning model 120 in Fig. 5 .

[0068] The prediction unit 104e inputs the impedance (T, k), the frequency (T, k) of the RF power, the power (T, k) of the RF power, and the matching condition (T, k) to the model 122, and inputs the process condition (k) to the fully connected layer 121, to perform calculations of the prediction model 130. The prediction model 130 outputs the measured value of the impedance at time t+Δt as the calculation result.

[0069] The prediction unit 104e can also predict the impedance of the chamber 10 at times further in the future than time t+Δt by repeating calculations of the prediction model 130 based on the measured impedance value at time t+Δt.

[0070] 7 is a diagram illustrating an example of a flow for predicting the impedance of the chamber 10 by repeating the calculations of the prediction model 130 according to the embodiment. Fig. 7 shows a case where the calculations of the prediction model 130 are repeated twice to predict the impedance of the chamber 10 at time t+2Δt. In Fig. 7, the upper part shows a flow for predicting the impedance of the chamber 10 at time t+Δt shown in Fig. 6.

[0071] When predicting the impedance of the chamber 10 at time t+2Δt, the measured value of the impedance at time t+Δt is used for the impedance of the chamber 10 at time t+Δt. The prediction unit 104e generates the impedance (T+Δt,k) by adding the measured value of the impedance at time t+Δt to the impedance (T,k).

[0072] It is assumed that the frequency, power, and matching condition of the RF power are constant between time t and time t+Δt. The prediction unit 104e generates the RF power frequency (T+Δt,k) by adding the RF power at time t to the RF power frequency (T,k) as the RF power frequency at time t+Δt. The prediction unit 104e also generates the RF power power (T+Δt,k) by adding the RF power power at time t to the RF power power (T,k) as the RF power power at time t+Δt. The prediction unit 104e also generates the matching condition (T+Δt,k) by adding the matching condition at time t to the matching condition (T,k) as the matching condition at time t+Δt.

[0073] The prediction unit 104e inputs the impedance (T+Δt,k), the frequency of the RF power (T+Δt,k), the power of the RF power (T+Δt,k), and the matching condition (T+Δt,k) to the model 122, and inputs the process condition (k) to the fully connected layer 121, to perform calculations of the prediction model 130. The prediction model 130 outputs the measured value of the impedance at time t+2Δt as the calculation result.

[0074] The prediction unit 104e repeats calculations using the prediction model, thereby being able to predict the impedance of the chamber 10 at a time further in the future than time t+Δt.

[0075] The prediction unit 104e uses the prediction model to predict the impedance of the chamber 10 for a time beyond the time required for matching. For example, the prediction model predicts the impedance of the chamber 10 for a time 10 μsec into the future, assuming Δt to be 10 μsec. The matching circuit 34 takes 100 msec from the time the set value is changed until the capacitance changes. In this case, the prediction unit 104e uses the prediction model to repeat calculations to predict the impedance of the chamber 10 for a time 100 msec or more into the future.

[0076] The matching control unit 104b adjusts at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the impedance of the chamber 10 predicted by the prediction unit 104e. The matching control unit 104b adjusts at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the impedance of the chamber 10 at the time predicted by the prediction unit 104e. For example, if it takes 100 msec for the capacitance of the matching circuit 34 to change, the matching control unit 104b changes the setting value of the matching circuit 34 100 msec before the predicted time. This allows the matching control unit 104b to match the matching circuit 34 to the impedance of the chamber 10 at the predicted time.

[0077] [flowchart]

[0078] First, an example of the flow of a model generation process including a model generation method according to an embodiment will be described. Fig. 8 is a flowchart showing an example of the flow of a model generation process according to an embodiment.

[0079] The generation unit 104c periodically measures the frequency of the RF power and the impedance of the chamber 10 in which plasma is generated by the RF power to obtain time-series data (step S10). For example, the generation unit 104c obtains the time-series data by reading out the process conditions and each time-series data of each data set included in the training data 103b. Step S11 corresponds to step (a) of the model generation method of the present disclosure.

[0080] The generation unit 104c performs machine learning using the acquired time-series data to generate a prediction model that predicts a change in impedance of the chamber 10 (step S11). For example, the generation unit 104c performs machine learning using the process conditions and each time-series data of each read data set to generate the prediction model. Step S11 corresponds to step (d) of the prediction method disclosed herein and step (b) of the model generation method disclosed herein.

[0081] The generating unit 104c stores the generated prediction model data in the storage unit 103 as model data 103c (step S12), and ends the process.

[0082] Next, an example of the flow of the matching process including the prediction method according to the embodiment will be described. Fig. 9 is a flowchart showing an example of the flow of the matching process according to the embodiment. The matching process shown in Fig. 9 is executed when plasma processing is started.

[0083] The acquisition unit 104d acquires the frequency of the RF power supplied to the chamber 10, the power of the RF power, the impedance of the chamber 10, the matching conditions of the matching circuit 34, and the process conditions of the plasma processing being performed (step S20). Step S20 corresponds to step (a) of the prediction method of the present disclosure.

[0084] The prediction unit 104e uses the prediction model of the model data 103c to predict the impedance of the chamber 10 from the RF power frequency, RF power power, impedance of the chamber 10, matching conditions of the matching circuit 34, and process conditions (step S21). For example, the prediction unit 104e generates time-series data from the start of plasma processing to the present for the impedance of the chamber 10, RF power frequency, RF signal power, and matching conditions acquired by the acquisition unit 104d. For example, when the process condition of the ongoing plasma processing is (k) and the elapsed time of the plasma processing is time t, the prediction unit 104e generates the impedance (T, k), RF power frequency (T, k), RF power power (T, k), and matching conditions (T, k). Using the prediction model, the prediction unit 104e predicts the impedance of the chamber 10 from the impedance (T, k), RF power frequency (T, k), RF power power (T, k), matching conditions (T, k), and process conditions (k). Step S21 corresponds to step (b) of the prediction method of the present disclosure.

[0085] The matching control unit 104b determines whether matching is necessary for the predicted impedance of the chamber 10 (step S22). For example, the matching control unit 104b determines that matching is necessary when the predicted impedance of the chamber 10 has changed by a predetermined value or more from the current impedance of the chamber 10.

[0086] If matching is not required (step S22: No), the process proceeds to step S24, which will be described later.

[0087] On the other hand, if matching is necessary (step S22: Yes), the matching control unit 104b adjusts at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the impedance of the chamber 10 predicted by the prediction unit 104e (step S23). Step S23 corresponds to step (c) of the prediction method of the present disclosure.

[0088] The matching control unit 104b determines whether the plasma processing has been completed (step S24). If the plasma processing has not been completed (step S24: No), the matching control unit 104b proceeds to step 20. On the other hand, if the plasma processing has been completed (step S24: Yes), the matching control unit 104b ends the processing.

[0089] In this way, the plasma processing apparatus 1 according to the embodiment can predict changes in the impedance of the chamber 10. Furthermore, the plasma processing apparatus 1 can maintain a matched state by adjusting at least one of the frequency of the RF power and the capacitance of the matching circuit 34 so as to match the predicted impedance of the chamber 10.

[0090] Next, a comparison between the embodiment and a comparative example will be described. First, as a comparative example, an example of a case where a matching process is performed to adjust the frequency of the RF power and the capacitance of the matching circuit 34 in a follow-up manner will be described. FIG. 10 is a timing chart of the matching process according to the comparative example. In the matching process according to the comparative example, when the impedance of the chamber 10 is measured, at least one of the frequency of the RF power and the capacitance of the matching circuit 34 is adjusted so as to match the measured impedance of the chamber 10. For example, in the matching process according to the comparative example, the frequency of the RF power is swept, and the RF signal P reflected from the chamber 10 is adjusted. ｒｅｆ Then, in the matching process according to the comparative example, the RF signal P reflected from the chamber 10 is changed to a frequency at which the RF power is minimized. ｒｅｆ The capacitance of the matching circuit 34 is changed so that the impedance of the chamber 10 is reduced. For example, in the matching process according to the comparative example, when the impedance of the chamber 10 is measured at timing T1, matching is started for the impedance of the chamber 10 measured at timing T1. Then, in the matching process according to the comparative example, at timing T2, the change in the capacitance of the matching circuit 34 to match the impedance of the chamber 10 measured at timing T1 is completed. However, there are cases where the impedance of the chamber 10 changes, and the impedance of the chamber 10 differs between timing T1 and timing T2. For this reason, it is difficult to maintain a matched state in the matching process according to the comparative example.

[0091] FIG. 11 is a timing chart of the matching process according to the embodiment. In the matching process according to the embodiment, when the impedance of the chamber 10 is acquired (measured), the prediction unit 104e predicts the impedance of the chamber 10. For example, when the impedance of the chamber 10 is measured at time T1, the prediction unit 104e predicts the impedance of the chamber 10 at time T2. Then, in the matching process according to the embodiment, the frequency of the RF power and / or the capacitance of the matching circuit 34 are adjusted to match the impedance of the chamber 10 at the predicted time T2. For example, the matching control unit 104b adjusts the frequency of the RF power and / or the capacitance of the matching circuit 34 to match the impedance of the chamber 10 at the predicted time T2. For example, in the matching process according to the embodiment, at time T2, the change in the capacitance of the matching circuit 34 to match the impedance of the chamber 10 at the predicted time T2 is completed. As a result, the matching process according to the embodiment can match the frequency of the RF power and the capacitance of the matching circuit 34 to the impedance of the chamber 10 at the predicted time T2 at time T2, thereby maintaining the matched state.

[0092] In this way, the plasma processing apparatus 1 according to the embodiment performs control to match the predicted value, and therefore can maintain a matched state more stably than the comparative example that does not use the predicted value.

[0093] Furthermore, since the plasma processing apparatus 1 according to the embodiment can stably maintain a matched state, the change in the plasma state is small, and accordingly the change in impedance is also small. Therefore, impedance matching is easier to achieve.

[0094] Furthermore, the plasma processing apparatus 1 according to the embodiment reduces the reflected RF power (RF signal P ｒｅｆ ) is smaller, power loss is reduced, and energy savings can be expected.

[0095] Furthermore, since the plasma processing apparatus 1 according to the embodiment performs matching based on a predicted value, it can follow changes in impedance of the chamber 10. This makes it possible to prevent plasma misfires and breakdowns in the amplifier 33.

[0096] In addition, the plasma processing apparatus 1 according to the embodiment can increase the time available for matching control by extending the timing T2 at which the impedance is predicted by the prediction unit 104e, thereby improving the impedance control range.

[0097] Furthermore, the plasma processing apparatus 1 according to the embodiment can predict changes in apparatus characteristics over time by predicting the impedance of the chamber 10 using a prediction model. Although the results of plasma processing on substrates W may vary from lot to lot due to changes over time, predicting these changes over time makes it possible to control the plasma processing apparatus 1 so that the processing results of the substrates W remain constant. By applying time-series data and a model, the plasma processing apparatus 1 can predict future changes over time and perform control using the predictions.

[0098] In the above embodiment, the process condition is defined as process condition (k), which is a condition that does not change during plasma processing. However, this is not limiting. If the process condition changes depending on time t during plasma processing, the process condition in the data set may be time-series data. For example, if the process condition for process number k changes depending on time t, the process condition in the data set may be process condition (t, k), which is set as the process condition at time t. In this case, a prediction model may be generated using process condition (t, k) instead of process condition (k), and the impedance of chamber 10 may be predicted using the prediction model.

[0099] In the above embodiment, the impedance (t, k), the RF power frequency (t, k), the RF power power (t, k), the matching condition (t, k), and the process condition (k) are used to generate a prediction model and perform prediction using the prediction model. However, this is not limited to this. The RF power power (t, k), the matching condition (t, k), and the process condition (k) are not required as data used to generate the prediction model and perform prediction using the prediction model. For example, if the plasma processing apparatus 1 repeats plasma processing under the same process conditions, the process condition (k) may not be included in the data used to generate the prediction model and perform prediction using the prediction model. If the RF power power and the matching condition change little during plasma processing, the RF power power (t, k) and the matching condition (t, k) may not be included in the data used to generate the prediction model and perform prediction using the prediction model. In this embodiment, the RF power frequency and the impedance of the chamber 10 are at least the data required to generate the prediction model and perform prediction using the prediction model. In the plasma processing apparatus 1 according to the embodiment, it is sufficient that the data set used for generating the prediction model and for making predictions using the prediction model includes at least the frequency of the RF power and the impedance of the chamber 10 .

[0100] In the above embodiment, Δt is set to 10 μsec, and the impedance of the chamber 10 at a time 10 μsec into the future is predicted using the prediction model. In the above embodiment, the impedance of the chamber 10 at a time more than 10 μsec into the future is predicted by repeating calculations using the prediction model. However, this is not limited to this. The future time (Δt) predicted by the prediction model may be any value. The prediction model may predict the impedance of the chamber 10 at a time more than 10 μsec into the future in a single calculation. For example, the prediction model may predict the impedance of the chamber 10 at a time longer than the time required for matching in a single calculation. For example, the prediction model may predict the impedance of the chamber 10 at a time longer than 100 msec into the future in a single calculation.

[0101] In the above embodiment, the control device 100 of the plasma processing apparatus 1 performs machine learning using the training data 103b to generate a prediction model. However, this is not limited to this. The generation of the prediction model may be performed by an information processing device, such as a computer, separate from the plasma processing apparatus 1. FIG. 12 is a diagram showing an example of a schematic configuration of an information processing device 200 according to an embodiment. Since the information processing device 200 is partially identical to the control device 100, the same components are denoted by the same reference numerals and their descriptions are omitted below, and differences will be mainly described. The information processing device 200 stores recipe data 103a, training data 103b, and model data 103c in the storage unit 103. The generation unit 104c performs machine learning using the training data 103b to generate a prediction model that predicts changes in the impedance of the chamber 10. For example, the generation unit 104c reads the process conditions and time-series data of each data set included in the training data 103b. The generation unit 104c performs machine learning using the process conditions and time-series data of each data set that has been read out, to generate a prediction model that predicts a change in the impedance of the chamber 10. The generation unit 104c uses the learned model obtained by the machine learning as a prediction model, and stores the prediction model data as model data 103c in the storage unit 103. The plasma processing apparatus 1 may store the model data 103c generated in this manner in the storage unit 103, and predict the impedance of the chamber 10 using the prediction model of the model data 103c.

[0102] As described above, the prediction method according to the embodiment includes steps (a) and (b). In step (a), the frequency of RF power applied to an electrode that generates plasma inside the chamber 10 and the impedance of the chamber 10 in which plasma is generated by the RF power are acquired from the matching circuit 34 provided in the power supply line through which the RF power flows. In step (b), a prediction model that predicts changes in the impedance of the chamber 10 from time-series data of the RF power frequency and the impedance of the chamber 10 acquired in step (a) is used to predict changes in the impedance of the chamber 10. In this way, the prediction method according to the embodiment can predict changes in the impedance of the chamber 10.

[0103] The prediction method according to the embodiment further includes step (c), in which the impedance of the supply side that supplies RF power is adjusted so as to match the impedance of the chamber 10 predicted in step (b). In this way, the prediction method according to the embodiment can match the chamber 10 side and the supply side that supplies RF power.

[0104] Furthermore, in step (c), at least one of the frequency of the RF power and the capacitance of the matching circuit 34 provided in the power supply line through which the RF power supplied to the chamber 10 flows is adjusted so as to match the impedance of the chamber 10 predicted in step (b). In this way, the prediction method according to the embodiment can match the chamber 10 side with the supply side that supplies the RF power.

[0105] Furthermore, the matching circuit 34 is configured to be able to measure the impedance of the chamber 10. In the step (a), the impedance of the chamber 10 is acquired from the matching circuit 34. In this way, the prediction method according to the embodiment can acquire the impedance of the chamber 10 during plasma processing.

[0106] Furthermore, the prediction model predicts the impedance of the chamber 10 from time-series data of the RF power frequency and the impedance of the chamber 10. In step (a), the RF power frequency and the impedance of the chamber 10 are periodically acquired. In step (b), using the prediction model, a change in the impedance of the chamber 10 is predicted from the time-series data of the RF power frequency and the impedance of the chamber 10 periodically acquired in step (a). As a result, the prediction method according to the embodiment can predict a change in the impedance of the chamber 10 due to the influence of the frequency of the RF power supplied to the chamber 10 and the impedance of the chamber 10 from the start of plasma processing.

[0107] Furthermore, the prediction model predicts the impedance of the chamber 10 from time-series data of the RF power frequency, the RF power power, and the impedance of the chamber 10. In step (a), the RF power frequency, the RF power power, and the impedance of the chamber 10 are periodically acquired. In step (b), the prediction model is used to predict a change in the impedance of the chamber 10 from the time-series data of the RF power frequency, the RF power power, and the impedance of the chamber 10 periodically acquired in step (a). As a result, the prediction method according to the embodiment can predict a change in the impedance of the chamber 10 due to the influence of the RF power frequency, the RF power power, and the impedance of the chamber 10 supplied to the chamber 10 from the start of plasma processing.

[0108] The prediction model also predicts changes in the impedance of the chamber 10 based on time-series data of the RF power frequency and the impedance of the chamber 10, and the conditions of the gas supplied to the chamber 10. In step (a), the RF power frequency and the impedance of the chamber 10 are periodically acquired, and the gas conditions are further acquired from a plasma processing recipe that stores process conditions including the conditions of the gas supplied to the chamber 10 during plasma processing. In step (b), the prediction model is used to predict changes in the impedance of the chamber 10 based on the time-series data of the RF power frequency and the impedance of the chamber 10 periodically acquired in step (a) and the gas conditions acquired from the recipe. As a result, the prediction method according to the embodiment can predict changes in the impedance of the chamber 10 due to the influence of the frequency of the RF power supplied to the chamber 10 and the impedance of the chamber 10 and the influence of the gas supplied to the chamber 10 from the start of plasma processing.

[0109] Furthermore, the prediction model predicts changes in the impedance of the chamber 10 from time-series data of the RF power frequency, the impedance of the chamber 10, and the capacitance of the matching circuit 34. In step (a), the RF power frequency, the impedance of the chamber 10, and the capacitance of the matching circuit 34 are periodically acquired. In step (b), the prediction model is used to predict changes in the impedance of the chamber 10 from the time-series data of the RF power frequency, the impedance of the chamber 10, and the capacitance of the matching circuit 34 acquired in step (a). As a result, the prediction method according to the embodiment can predict changes in the impedance of the chamber 10 due to the influence of the frequency of the RF power supplied to the chamber 10, the impedance of the chamber 10, and the capacitance of the matching circuit 34 from the start of plasma processing.

[0110] Furthermore, the prediction model predicts a change in the impedance of the chamber 10 from at least the frequency of the RF power, the impedance of the chamber 10, and the previous prediction result of the impedance of the chamber 10. In this way, the prediction method according to the embodiment can predict a change in the impedance of the chamber 10.

[0111] Furthermore, the prediction method according to the embodiment further includes step (d) before steps (a) and (b). In step (d), machine learning is performed using the periodically measured frequency of the RF power and time-series data of the chamber 10 to generate a prediction model. In step (b), a change in the impedance of the chamber 10 is predicted from the frequency of the RF power and the impedance of the chamber 10 acquired in step (a) using the prediction model generated in step (d). In this way, the prediction method according to the embodiment can predict a change in the impedance of the chamber 10.

[0112] The plasma processing apparatus 1 according to the embodiment includes a chamber 10, an RF power supply unit 30 (supply unit), an acquisition unit 104d, and a prediction unit 104e. The RF power supply unit 30 supplies RF power to generate plasma inside the chamber 10. The acquisition unit 104d acquires the frequency of the RF power supplied from the RF power supply unit 30 and the impedance of the chamber 10 in which plasma is generated by the RF power. The prediction unit 104e predicts a change in the impedance of the chamber 10 from the frequency of the RF power and the impedance of the chamber 10 acquired by the acquisition unit 104d using a prediction model that predicts a change in the impedance of the chamber 10 from at least the frequency of the RF power and the impedance of the chamber 10. This allows the plasma processing apparatus 1 according to the embodiment to predict a change in the impedance of the chamber 10.

[0113] Moreover, the model generation method according to the embodiment includes steps (a) and (b). In step (a), time-series data is acquired by periodically measuring the frequency of the RF power that generates plasma inside the chamber 10 and the impedance of the chamber 10 in which the plasma is generated by the RF power. In step (b), machine learning is performed using the time-series data acquired in step (a) to generate a prediction model that predicts changes in the impedance of the chamber 10. In this way, the model generation method according to the embodiment can generate a prediction model that can predict changes in the impedance of the chamber 10.

[0114] The information processing apparatus 200 according to the embodiment also includes a storage unit 103 and a generation unit 104c. The storage unit 103 stores time-series data that periodically measures the frequency of the RF power that generates plasma inside the chamber 10 and the impedance of the chamber 10 in which plasma is generated by the RF power. The generation unit 104c performs machine learning using the time-series data stored in the storage unit 103 to generate a prediction model that predicts changes in the impedance of the chamber 10. This allows the information processing apparatus 200 according to the embodiment to generate a prediction model that can predict changes in the impedance of the chamber 10.

[0115] It should be noted that the disclosed embodiments are illustrative in all respects and should not be considered limiting. Indeed, the above-described embodiments may be embodied in various forms. Furthermore, the above-described embodiments may be omitted, substituted, or modified in various forms without departing from the scope and spirit of the appended claims.

[0116] In the above embodiment, the plasma processing is performed on a semiconductor wafer as the substrate W, but the present invention is not limited to this. The substrate W may be a glass substrate or the like.

[0117] In the above embodiment, the plasma processing performed on the substrate W is a plasma processing such as plasma etching, but the present invention is not limited to this. The plasma processing may be any processing using plasma. For example, the plasma processing may be a film formation process, a modification process, or a heat treatment such as ashing.

[0118] In addition, the following supplementary notes are disclosed regarding the above-described embodiment.

[0119] (Supplementary Note 1) A prediction method comprising: (a) a step of acquiring the frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a power feeder through which the RF power flows; and (b) a step of predicting a change in the impedance of the chamber from the time series data of the frequency of the RF power and the impedance of the chamber acquired in the step (a) using a prediction model that predicts a change in the impedance of the chamber from time series data of at least the frequency of the RF power and the impedance of the chamber.

[0120] (Supplementary Note 2) The prediction method according to Supplementary Note 1, further comprising the step of: (c) adjusting the impedance of the supply side that supplies the RF power so as to match the impedance of the chamber predicted by step (b).

[0121] (Supplementary Note 3) The prediction method according to Supplementary Note 2, wherein the step (c) adjusts at least one of a frequency of the RF power and a capacitance of a matching circuit provided in a power supply line through which the RF power supplied to the chamber flows, so as to match the impedance of the chamber predicted in the step (b).

[0122] (Supplementary Note 4) The prediction method according to Supplementary Note 3, wherein the matching circuit is configured to be able to measure the impedance of the chamber, and the step (a) acquires the impedance of the chamber from the matching circuit.

[0123] (Supplementary Note 5) The prediction method according to any one of Supplementary Notes 1 to 4, wherein the step (a) periodically acquires the frequency of the RF power and the impedance of the chamber, and the step (b) predicts a change in the impedance of the chamber using the prediction model from the time-series data of the frequency of the RF power and the impedance of the chamber periodically acquired in the step (a).

[0124] (Supplementary Note 6) The prediction method according to any one of Supplementary Notes 1 to 5, wherein the prediction model predicts the impedance of the chamber from time-series data of the frequency of the RF power, the power of the RF power, and the impedance of the chamber; the step (a) periodically acquires the frequency of the RF power, the power of the RF power, and the impedance of the chamber; and the step (b) uses the prediction model to predict a change in the impedance of the chamber from the time-series data of the frequency of the RF power, the power of the RF power, and the impedance of the chamber periodically acquired in the step (a).

[0125] (Supplementary Note 7) The prediction method according to any one of Supplementary Notes 1 to 6, wherein the prediction model predicts a change in the impedance of the chamber from time-series data of the frequency of the RF power and the impedance of the chamber and conditions of a gas supplied to the chamber; the step (a) periodically acquires the frequency of the RF power and the impedance of the chamber, and further acquires the gas conditions from a plasma processing recipe that stores process conditions including conditions of a gas supplied to the chamber in plasma processing; and the step (b) predicts a change in the impedance of the chamber using the prediction model from the time-series data of the frequency of the RF power and the impedance of the chamber that are periodically acquired in the step (a) and the gas conditions acquired from the recipe.

[0126] (Supplementary Note 8) The prediction method according to Supplementary Note 3 or 4, wherein the prediction model predicts a change in the impedance of the chamber from time-series data of the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit; the step (a) periodically acquires the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit; and the step (b) uses the prediction model to predict a change in the impedance of the chamber from the time-series data of the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit acquired in the step (a).

[0127] (Supplementary Note 9) The prediction method according to any one of Supplementary Notes 1 to 8, wherein the prediction model predicts a change in the impedance of the chamber from at least the frequency of the RF power, the impedance of the chamber, and a prediction result of the impedance of the chamber immediately before.

[0128] (Supplementary Note 10) The prediction method according to any one of Supplementary Notes 1 to 9, further comprising, before the steps (a) and (b), a step (d) of performing machine learning using the periodically measured frequency of the RF power and time-series data of the chamber to generate the prediction model, wherein the step (b) predicts a change in the impedance of the chamber from the frequency of the RF power and the impedance of the chamber acquired in the step (a) using the prediction model generated in the step (d).

[0129] (Supplementary Note 11) A plasma processing apparatus comprising: a chamber; a supply unit configured to supply RF (Radio Frequency) power that generates plasma inside the chamber; an acquisition unit configured to acquire the frequency of the RF power supplied from the supply unit and the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a feeder line through which the RF power flows; and a prediction unit configured to predict a change in impedance of the chamber from the frequency of the RF power and time series data of the impedance of the chamber acquired by the acquisition unit, using a prediction model that predicts a change in impedance of the chamber from time series data of at least the frequency of the RF power and the impedance of the chamber.

[0130] (Supplementary Note 12) A model generation method including: (a) a step of acquiring time series data of a frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and of an impedance of the chamber in which plasma is generated by the RF power, measured periodically from a matching circuit provided in a feeder line through which the RF power flows; and (b) a step of performing machine learning using the time series data acquired in the step (a) to generate a prediction model that predicts a change in the impedance of the chamber.

[0131] (Supplementary Note 13) An information processing device having: a memory unit configured to store time series data of a frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and an impedance of the chamber in which plasma is generated by the RF power, measured periodically from a matching circuit provided in a feeder line through which the RF power flows; and a generation unit configured to perform machine learning using the time series data stored in the memory unit to generate a prediction model that predicts a change in the impedance of the chamber.

[0132] REFERENCE SIGNS LIST 1 Plasma processing apparatus 10 Chamber 11 Support 12 Upper electrode showerhead assembly 20 Gas supply unit 21 Gas source 22 Flow rate controller 30 RF power supply unit 31 Signal source 32 Input circuit 33 Amplifier 34 Matching circuit 40 Exhaust system 100 Control device 101 External I / F unit 102 User I / F unit 103 Memory unit 103a Recipe data 103b Learning data 103c Model data 104 Control unit 104a Plasma processing control unit 104b Matching control unit 104c Generation unit 104d Acquisition unit 104e Prediction unit 120 Learning model 121, 124 Bonding layer 122 Model 123 Flattening layer 130 Prediction model 200 Information processing apparatus W Substrate

Claims

1. A prediction method comprising: (a) a step of acquiring the frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a power supply line through which the RF power flows; and (b) a step of predicting changes in the impedance of the chamber from the time series data of the frequency of the RF power and the impedance of the chamber acquired in step (a) using a prediction model that predicts changes in the impedance of the chamber from time series data of at least the frequency of the RF power and the impedance of the chamber.

2. The prediction method according to claim 1, further comprising the step of: (c) adjusting the impedance of the supply side that supplies the RF power so as to match the impedance of the chamber predicted by step (b).

3. The prediction method according to claim 2, wherein step (c) adjusts at least one of the frequency of the RF power and the capacitance of a matching circuit provided in a power supply line through which the RF power supplied to the chamber flows, so as to match the impedance of the chamber predicted by step (b).

4. The prediction method according to claim 3, wherein the matching circuit is configured to be able to measure the impedance of the chamber, and step (a) acquires the impedance of the chamber from the matching circuit.

5. The prediction method according to claim 1, wherein step (a) periodically acquires the frequency of the RF power and the impedance of the chamber, and step (b) uses the prediction model to predict changes in the impedance of the chamber from the time-series data of the frequency of the RF power and the impedance of the chamber periodically acquired in step (a).

6. The prediction method according to claim 1, wherein the prediction model predicts the impedance of the chamber from time-series data of the frequency of the RF power, the power of the RF power, and the impedance of the chamber; step (a) periodically acquires the frequency of the RF power, the power of the RF power, and the impedance of the chamber; and step (b) uses the prediction model to predict changes in the impedance of the chamber from the time-series data of the frequency of the RF power, the power of the RF power, and the impedance of the chamber periodically acquired in step (a).

7. The prediction method according to claim 1, wherein the prediction model predicts a change in the impedance of the chamber from time-series data of the frequency of the RF power and the impedance of the chamber and conditions of a gas supplied to the chamber, and wherein step (a) periodically acquires the frequency of the RF power and the impedance of the chamber, and further acquires the gas conditions from a plasma processing recipe that stores process conditions including conditions of a gas supplied to the chamber in plasma processing, and step (b) uses the prediction model to predict a change in the impedance of the chamber from the time-series data of the frequency of the RF power and the impedance of the chamber periodically acquired in step (a) and the gas conditions acquired from the recipe.

8. A prediction method according to claim 3 or 4, wherein the prediction model predicts changes in the impedance of the chamber from time-series data of the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit; step (a) periodically acquires the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit; and step (b) uses the prediction model to predict changes in the impedance of the chamber from the time-series data of the frequency of the RF power, the impedance of the chamber, and the capacitance of the matching circuit acquired in step (a).

9. The prediction method according to claim 1, wherein the prediction model predicts a change in the chamber impedance from at least the frequency of the RF power, the chamber impedance, and the immediately preceding prediction result of the chamber impedance.

10. The prediction method according to claim 1, further comprising, prior to steps (a) and (b), (d) performing machine learning using the periodically measured frequency of the RF power and time-series data of the chamber to generate the prediction model, wherein step (b) predicts a change in the impedance of the chamber from the frequency of the RF power and the impedance of the chamber acquired in step (a) using the prediction model generated in step (d).

11. A plasma processing apparatus comprising: a chamber; a supply unit configured to supply RF (Radio Frequency) power to generate plasma inside the chamber; an acquisition unit configured to acquire the frequency of the RF power supplied from the supply unit and the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a feeder line through which the RF power flows; and a prediction unit configured to predict changes in the impedance of the chamber from the frequency of the RF power and the time series data of the impedance of the chamber acquired by the acquisition unit, using a prediction model that predicts changes in the impedance of the chamber from time series data of at least the frequency of the RF power and the impedance of the chamber.

12. A model generation method comprising: (a) a step of acquiring time series data of the frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and of the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a power supply line through which the RF power flows; and (b) a step of performing machine learning using the time series data acquired in step (a) to generate a predictive model that predicts changes in the impedance of the chamber.

13. An information processing device having: a memory unit configured to store time series data periodically measuring the frequency of RF (Radio Frequency) power applied to an electrode that generates plasma inside a chamber, and the impedance of the chamber in which plasma is generated by the RF power from a matching circuit provided in a power supply line through which the RF power flows; and a generation unit configured to perform machine learning using the time series data stored in the memory unit to generate a prediction model that predicts changes in the impedance of the chamber.

Images