Plasma processing apparatus and control parameter estimation method

Abstract

It is required to predict an optimal value of a specific control parameter using a machine learning model. At that time, since an enormous amount of learning data including a plurality of gas types is required for generating a prediction model by machine learning, it is necessary to suppress the number of pieces of learning data using the above-described technique or the like.SOLUTION: In the present invention, when a prediction model is generated using machine learning in order to estimate an optimal operation parameter in a plasma processing apparatus, a physical quantity using a ratio of ions obtained by ionizing each gas in plasma when a plasma processing gas is turned into plasma as an input parameter and an alternative parameter calculated from a physical quantity used to predict an optimal operation parameter when each gas is a single substance are used. Further, it is provided with an estimating part estimating the optimum value of the output parameter corresponding to the input parameter.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a plasma processing apparatus and a method for estimating a control parameter. [Background technology]

[0002] The manufacturing process of semiconductor devices is required to cope with the miniaturization and integration of components included in semiconductor devices, for example, in integrated circuits and nanoelectromechanical systems, structures are being further scaled down to nanoscale. Lithography is a common technique used in the fabrication of semiconductor devices to create fine patterns. This technique involves projecting the device structure onto a resist layer and selectively etching away the exposed substrate. In subsequent processing steps, other materials can be deposited in the etched areas to form integrated circuits.

[0003] For this reason, plasma etching processing equipment has become indispensable in the manufacture of semiconductor devices. In plasma etching, gas is supplied into a processing chamber that has been depressurized to a predetermined vacuum level, and then converted into plasma by an electric field or other means formed inside the vacuum chamber. Highly reactive ions and radicals generated in the plasma then physically and chemically react with the surface of the wafer being processed, thereby etching the wafer.

[0004] In plasma etching processes, applying a radio-frequency voltage to a wafer mounting table is a common technique. When a radio-frequency voltage is applied to a mounting table connected to a radio-frequency power supply via a capacitor, the sheath generated between the plasma and the mounting table has a rectifying effect, causing the mounting table to assume a time-averaged negative voltage due to self-bias. This accelerates positive ions, facilitating etching and increasing the linearity of the positive ions, resulting in anisotropic etching. Adjusting the amplitude of the radio-frequency voltage applied to the mounting table makes it possible to control the etching rate and the perpendicularity of the cross-section formed by etching.

[0005] Generally, a sine wave is used as the high-frequency voltage applied to the wafer mounting table. However, as disclosed in Patent Document 1, a square wave may be used instead. The energy of ions flowing from the plasma into the mounting table is determined by the electric field acting between the plasma and the mounting table. When a sine wave high-frequency voltage is applied, the electric field changes smoothly, causing ions of various energies to flow into the mounting table. However, when a square wave high-frequency voltage is applied, the ion energies are clearly separated into high and low, making it easier to control etching.

[0006] Furthermore, to perform complex and highly accurate etching in a plasma processing system, it is necessary to set optimal control parameters. Therefore, it is necessary to use a machine learning model to predict the optimal values ​​of specific control parameters. To generate a predictive model using machine learning, a huge amount of training data, including multiple gas species, is required.

[0007] As a method for generating such training data, the following technology is disclosed in Patent Document 1. "The model learning unit 108 learns a prediction model using the learning data, the target setting unit 107 sets a target output parameter value by interpolating between a target output parameter value and an output parameter value of the learning data that is closest to the target output parameter value, the processing condition searching unit 109 estimates input parameter values ​​corresponding to the target output parameter value and the target output parameter value using the prediction model, and the model learning unit 108 updates the prediction model using, as additional learning data, pairs of the input parameter values ​​estimated by the processing condition searching unit and output parameter values ​​that are the results of processing by the processing device using the estimated input parameter values." [Prior art documents] [Patent documents]

[0008] [Patent Document 1] JP 2019-159864 A Summary of the Invention [Problem to be solved by the invention]

[0009] However, in Patent Document 1, a large amount of additional training data must still be accumulated in order to obtain desirable input parameters. Therefore, in order to create a prediction model, it is necessary to perform processing in an actual processing device and obtain many pairs of control parameters and output parameters, which makes it difficult to apply to processing devices that are under development and do not have an actual device, or when considering process recipes using new gases. [Means for solving the problem]

[0010] In order to solve the above problems, one representative plasma processing apparatus of the present invention is a plasma processing apparatus including a processing chamber in which a sample is plasma-processed, a high-frequency power supply that supplies high-frequency power for generating plasma, an automatic matching box that controls reflection of the high-frequency power, and a gas supply device that supplies a gas for plasma processing to the processing chamber, an estimation unit that estimates output parameters from input parameters; The estimation unit includes a prediction model generated by machine learning from learning data that is a set of input parameters and output parameters, including ionization energy when the process gas is assumed to be a single gas. [Effects of the Invention]

[0011] According to the present invention, it is possible to generate a prediction model that can be applied to a processing apparatus for which an actual apparatus does not exist or to a process recipe using a new gas. Problems, configurations, and effects other than those described above will become apparent from the following description of the preferred embodiments of the invention. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a block diagram showing an example of the configuration of a plasma processing apparatus according to an embodiment. [Figure 2] FIG. 2 is a block diagram showing an example of the configuration of an automatic matching box in the plasma processing apparatus according to the embodiment. [Figure 3] FIG. 3 is a diagram for explaining an outline of generation of a prediction model by machine learning in the plasma processing apparatus according to the embodiment. [Figure 4] FIG. 4 is a diagram illustrating the movement of the matching element and the change in the reflected wave when the prediction model is used in the plasma processing apparatus according to the embodiment. [Figure 5] FIG. 5 is a block diagram showing an example of a hardware configuration. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In describing the embodiment, the same reference numerals will be used in the drawings to designate parts having the same functions, and repeated description will be omitted as a general rule. However, the present invention should not be construed as being limited to the description of the following embodiments. Those skilled in the art will readily understand that the specific configurations can be modified without departing from the spirit or intent of the present invention.

[0014] (Terminology explanation) The "first physical quantity" means, for each gas, the cross section of a collision between a gas molecule and an electron, the probability that a gas molecule will be ionized when colliding with an electron, and the probability that the gas will collide with an electron again after being ionized and return to a neutral gas. The "second physical quantity" refers to the ionization energy required to convert each gas into plasma. The "surrogate parameter" refers to the average ionization energy of each gas. Specifically, it is a value obtained by weighting the second physical quantities (ionization energies of each gas) of the gases by the first physical quantities and partial pressures (flow rate ratios) of each gas. [Example]

[0015] (Outline of plasma processing equipment) First, a plasma processing apparatus according to an embodiment will be described with reference to Fig. 1. Fig. 1 shows an example of a schematic configuration of a plasma processing apparatus according to an embodiment. 1 shows a microwave ECR plasma etching apparatus as an example of a plasma processing apparatus 100. The diagram shows a schematic diagram of a vacuum processing chamber 101 provided in the plasma processing apparatus 100, including electrodes disposed therein and electric and magnetic field generators disposed thereoutside. The vacuum processing chamber 101 has a structure in which the upper part of a vacuum vessel 102 is hermetically sealed with a dielectric window 103 and the lower part is connected to an exhaust system consisting of a turbo molecular pump 111 and a dry pump 112 via a variable conductance valve 110 . A shower plate 121 having a plurality of pores 120 is installed directly below the dielectric window 103, and the space 122 between the dielectric window 103 and the shower plate 121 is connected to a gas supply mechanism 124 via a gas pipe 123, and the space 122 and the vacuum processing chamber 101 are in communication with each other via the pores 120.

[0016] The pressure inside the vacuum processing chamber 101 is maintained at a desired value by adjusting the opening of the variable conductance valve 110 through feedback control based on the value measured by the pressure gauge 104, and by adjusting the amount of air exhausted from the vacuum processing chamber 101 by the turbomolecular pump 111.

[0017] A microwave supply mechanism consisting of a microwave power supply 130, a waveguide 131, an output monitor 132, an automatic matching box 133, and a cavity resonator 134 is installed above the vacuum processing chamber 101. The microwaves output from the microwave power supply 130 propagate through the waveguide 131 to the cavity resonator 134, where they are adjusted to a microwave distribution suitable for plasma generation, and are introduced into the vacuum processing chamber 101 through the dielectric window 103 and the shower plate 121. At this time, the typical frequency of the microwaves is 2.45 GHz.

[0018] Solenoid coils 140, 141, and 142 are installed to surround vacuum vessel 102 and cavity resonator 134, and generate a magnetic field within vacuum processing chamber 101 by applying current from coil power supply 143. Electron cyclotron resonance (ECR) occurs in a region where the strength of this magnetic field and the frequency of the microwaves introduced from the microwave supply mechanism satisfy a specific relationship. For example, for a 2.45 GHz microwave, ECR occurs in a region where the magnetic field strength is 0.0875 T. In this region, electrons efficiently receive energy from the microwaves, are accelerated, and collide with gas supplied by gas supply mechanism 124, promoting dissociation and ionization. This generates plasma 144, which diffuses within vacuum processing chamber 101.

[0019] The region where plasma is generated can be controlled by the magnetic field configuration inside vacuum processing chamber 101, and the magnetic field configuration can be controlled by the value of the current flowing through solenoid coils 140, 141, and 142. In addition, since plasma moves along magnetic field lines, it can be controlled by the magnetic field configuration.

[0020] A sample stage fixed by beams (not shown) is installed inside the vacuum processing chamber 101. The sample stage, vacuum vessel 102, turbomolecular pump 111, and cavity resonator 134 are cylindrical and share the same central axis, so that the generation of plasma 144 and the gas flow are uniform about the central axis. A wafer 150 to be etched is transferred to the top of the sample stage by a transfer device such as a robot arm provided in the plasma processing apparatus 100.

[0021] The top and side surfaces of the electrode 151 are covered with a dielectric film 152, and electrostatic chucking electrodes 153 are installed inside the dielectric film 152 on both the center and outer periphery of the sample stage. An electrostatic chucking voltage generator 154 is connected to the electrostatic chucking electrodes 153 from the outside of the vacuum chamber 102. Different voltages are applied from the electrostatic chucking voltage generator 154 to the center and outer periphery of the electrostatic chucking electrodes 153, generating an attractive force between the wafer 150 and the electrostatic chucking electrode 153, thereby fixing the wafer 150 to the top of the sample stage. In addition, a temperature control film 155 is installed inside the dielectric film 152 and is connected to a temperature control mechanism 156 outside the vacuum chamber 102. This allows the temperature of the sample stage to be controlled, thereby controlling the temperature of the wafer 150 and thus controlling the plasma processing.

[0022] When the plasma 144 is generated, a sheath 145 is generated between the plasma 144 and the wafer 150 , and a sheath 146 is generated between the plasma 144 and the ground 157 , respectively, and particles in the plasma 144 pass through the sheath 145 and reach the wafer 150 .

[0023] When a radio frequency voltage is output from the radio frequency bias power supply 158 connected to the electrode 151, an electric circuit is formed through the automatic matching box 159, the electrode 151, the dielectric film 152, the wafer 150, the sheath 145, the plasma 144, and the sheath 146 to the ground 157, generating a radio frequency voltage in the wafer 150, causing the wafer 150 to have a self-bias voltage. This accelerates ions incident on the wafer 150 from the plasma 144, enabling anisotropic etching. This radio frequency voltage has a frequency lower than the output frequency of the microwave power supply 130 but high enough to apply the voltage to the wafer 150 via the dielectric film 152; typically, a frequency in the range of several hundred kHz to several MHz is used. The automatic matching box 159 also performs impedance matching so that the power output from the radio frequency bias power supply 158 is efficiently transmitted to the sheath 145.

[0024] All of the above components are connected to a control computer in the control unit 160 and controlled in accordance with a preset operation sequence called a recipe. The control computer in the control unit 160 can be realized by a general computer system 500 shown in Fig. 5. This can also include an estimation unit and a prediction model. The hardware configuration of the control computer can include a CPU 520, an input / output interface 530, a memory 510 including ROM and RAM, a storage device 501, and a network interface 540 connecting these. This makes it possible to store external input data in a storage device or the like and output the results of information processing.

[0025] (Automatic matching box) Next, the structure of automatic matching box 133 will be described with reference to FIG. FIG. 2 shows the structure of automatic matching box 133 that suppresses the reflection of microwaves introduced into vacuum processing chamber 101 through dielectric window 103 and shower plate 121. As shown in FIG. The automatic matching device 133 suppresses the reflected wave by moving the positions of the matching elements 201, 202, and 203 from their initial positions and adjusting the impedance based on the signal output from the control device 200 through feedback control based on the output value of the reflected wave measured by the output monitor 132.

[0026] In automatic matching to suppress microwave reflection, when the plasma processing conditions are changed, the plasma characteristics such as density change, and the plasma impedance changes accordingly. Automatic matching box 133 operates to suppress microwave reflection in response to this change in plasma impedance. However, because the matching element is mechanically driven, the matching operation is slow and it may not be able to keep up with changes in the plasma processing conditions, which may result in the reflected waves not being suppressed when the plasma processing conditions are changed.

[0027] Furthermore, plasma is not stable immediately after it is generated, and even after it stabilizes, its characteristics, such as plasma density, may change, and the plasma impedance may also change. As a result, even if the position of the matching element is changed based on the plasma impedance immediately after the plasma is generated, the plasma impedance may have changed by the time the matching element position is changed, and the reflected wave of the plasma may not be suppressed. Furthermore, if the reflected wave of the plasma cannot be suppressed, the power absorbed by the plasma decreases, which may result in a decrease in plasma density, making it difficult to maintain the plasma, or a decrease in etching rate.

[0028] For this reason, a prediction model is required that outputs the position of the matching element of the automatic matching box for controlling the reflected waves of the plasma when the process recipe, including the setting values ​​of the processing equipment and the gases used for processing, is used as input data. The following describes how to create a prediction model that outputs the positions of the matching elements of such an automatic matching box.

[0029] (dielectric constant of plasma) The reflected wave from the plasma varies depending on the gas species, and is known to be expressed by the dielectric constant and dielectric loss tangent of the plasma. The dielectric constant of the plasma can be calculated using the following formula:

number

[0030] (Electron density prediction) Next, the prediction of electron density will be described. Electron density prediction is usually performed by predicting the electron density in the plasma using the ionization energy and other operating parameters for each gas, and then predicting the density of the entire plasma from the sum of the electron densities of the individual gases. However, instead of calculating the electron density for each gas, it is possible to estimate the electron density for the entire plasma by averaging the ionization energies and treating the gas as a single gas. In other words, the electron density can be predicted based on the sum of the partial pressures of a plurality of gases.

[0031] Therefore, in the present disclosure, a prediction model can be created by combining the partial pressures of each gas into a single parameter. This prevents the number of training data from becoming too large. Furthermore, by using the shape of the vacuum vessel of the plasma processing apparatus as an input parameter, the prediction model can be applied to processing apparatuses and processing conditions that were not used when acquiring the training data for creating the prediction model.

[0032] (Matching element position prediction model) Next, in order to address the above-mentioned problem, a method for generating a model for predicting the position of a matching element will be described with reference to FIG. FIG. 3 is a diagram for explaining an outline of generation of a prediction model by machine learning.

[0033] (Training data) In Fig. 3, the input parameters P1, P2, P X are parameters input into the prediction model, such as the partial pressure of each gas, microwave output, solenoid coil current value, and ionization energy when the process gas is assumed to be a single gas (hereinafter sometimes referred to as "alternative parameters"). Furthermore, output parameters M1, M2, and M3 represent parameters to be predicted by the prediction model, specifically, the position of the matching element of the microwave automatic matching box. wherein the alternative parameter is calculated by weighting the second physical quantity of the process gas by the first physical quantity and partial pressure of the process gas and averaging the second physical quantity of the process gas; the first physical quantity is a physical quantity for predicting the rate at which each of the plurality of process gases is ionized; The second physical quantity is the energy required to ionize each element of the process gas.

[0034] The first physical quantity includes a collision cross section between gas molecules and electrons, a probability that gas molecules will be ionized when colliding with electrons, and a probability that the gas will collide with electrons again after being ionized and return to a neutral gas.

[0035] (Generating training data) Specifically, as shown in Figure 3, the learning data is created by taking the partial pressure of each gas for each recipe as input parameters for the learning data, and the positions M1, M2, and M3 of the matching elements when the reflected waves are suppressed as output parameters. In this case, one set of learning data is (P1-A, P2-A, P3-A, ... Px-A, M1-A, M2-A, M3-A).

[0036] The learning data may be data obtained by actual measurement in a plasma processing apparatus, or may be obtained by estimation using a plasma simulation, or may be obtained by estimation using a machine learning model. In the embodiment of the present disclosure, plasma processing is first performed under each predetermined plasma processing condition, and the position of the matching element when reflected waves are successfully suppressed is recorded. In other words, a set of data on the conditions for each recipe and the matching element positions M1, M2, and M3 when reflected waves are successfully suppressed is accumulated as learning data. Estimation by plasma simulation is a method of simulating plasma behavior without actually performing plasma processing by using numerical calculations such as the CIP (Constrained Interpolation Profile Scheme) method. Estimation by machine learning model is a method of estimating by creating a predictive model using machine learning, with the operating conditions of actual plasma processing as input parameters and the average ionization energy as output parameters.

[0037] In addition, when data obtained from a plasma processing apparatus with a vacuum container shape different from that of the plasma processing apparatus to which the invention is applied is used as input parameters for the learning data, by including parameters related to the vacuum container shape as input parameters, it becomes possible to apply the estimation results even between plasma processing apparatuses with different vacuum container shapes.

[0038] (Creating a predictive model) Next, a model is created by machine learning using the input parameters and output parameters as training data. For this, multiple regression analysis, support vector machine (SVM), or deep learning may be used.

[0039] By performing machine learning using a set of the energy required to generate this plasma, the total process gas flow rate, the microwave output, the current value flowing through the solenoid coil, and the corresponding position of the matching element as learning data, a predictive model can be generated in which the plasma processing conditions are used as input parameters and the position of the matching element is used as an output parameter. In this case, the energy required to generate plasma when the process gas is assumed to be a single gas may be a value obtained by plasma simulation, or a value estimated by another machine learning model.

[0040] First, to create a prediction model, a set of specific operational parameters as output parameters and operational parameters other than the output parameters as input parameters is obtained. These data may be data obtained in a plasma processing apparatus that performs processing using optimal values ​​estimated using the prediction model, or may be data obtained in another plasma processing apparatus.

[0041] In order to create a predictive model using data obtained from other plasma processing devices, the shape of the vacuum chamber of the plasma processing device is also used as an input parameter, so data must be obtained from multiple plasma processing devices having vacuum chambers of different shapes. In order to obtain the set of output parameters and input parameters, data obtained in an actual plasma processing apparatus may be used, or a set of data obtained by a simulation that models the plasma processing apparatus may be used.

[0042] Using the obtained set of output parameters and input parameters, a predictive model is created using multiple regression analysis, support vector machine (SVM), and deep learning. By performing processing in the plasma processing apparatus using the operation parameters estimated by the prediction model created as described above, processing with higher accuracy becomes possible.

[0043] (effect) By using this prediction model to change the position of the matching element to a pre-predicted position before plasma is generated when the plasma processing conditions are changed, the time required to search for the position of the matching element and the time during which the reflected wave cannot be suppressed and the plasma becomes unstable can be shortened. The example shown in Figure 4 is experimental data from a plasma processing system using process gases containing Ar, Cl2, and CHF3. Without prediction (left side of Figure 4), it takes time for the matching element to move based on the feedback signal from the control device, during which time more than 10% of the microwaves are reflected, causing plasma instability. However, with prediction (right side), control is possible without generating microwave reflections. In other words, it can be confirmed that the reflected waves that occurred when the prediction model was not used are sufficiently suppressed by using the prediction model.

[0044] In addition, the partial pressures of each gas required for the machine learning model are combined into a single alternative parameter to create the machine learning model, which prevents the amount of training data from becoming enormous, and also makes it possible to apply the machine learning model even to processing devices and processing conditions that were not used in the training data used to create the machine learning model.

[0045] The invention made by the inventor has been specifically described above based on the embodiments, but the present invention is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments are detailed to clearly explain the present invention, but the present invention is not necessarily limited to those having all of the described configurations.

[0046] Furthermore, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add a configuration of another embodiment to a configuration of one embodiment. Furthermore, it is also possible to add other configurations to the configuration of each embodiment, or to delete or replace part of the configuration. Note that the components and relative sizes shown in the drawings are simplified and idealized to make the present invention easier to understand, and the actual shape may be more complex.

[0047] The structures and methods described in the above-described embodiments are not limited to those of the above-described embodiments, but include various application examples. Furthermore, the present disclosure also includes the following aspects.

[0048] (Aspect 1) a processing chamber in which the sample is plasma processed; a high frequency power supply that supplies high frequency power for generating plasma; an automatic matching box for controlling high frequency reflection; a sample stage on which the sample is placed; a gas supply device for supplying a gas for plasma processing to the processing chamber; In a plasma processing apparatus having an exhaust device for exhausting the processing chamber, an estimation unit that estimates output parameters from input parameters; The plasma processing apparatus is characterized in that the estimation unit is equipped with a prediction model generated by machine learning from learning data which is a set of input parameters and output parameters including ionization energy when the process gas is assumed to be a single gas.

[0049] (Aspect 2) In the plasma processing apparatus according to aspect 1, the ionization energy of the process gas, when assumed to be a single gas, is calculated by weighting a second physical quantity of the process gas by a first physical quantity and a partial pressure of the process gas and averaging the second physical quantity of the process gas; the first physical quantity is a physical quantity for predicting the ionization rate of each of a plurality of process gases when the process gases are ionized, The second physical quantity is the energy required to ionize each element of the process gas. A plasma processing apparatus characterized by:

[0050] (Aspect 3) In the plasma processing apparatus according to aspect 2, The first physical quantity includes a collision cross section between a gas molecule and an electron, a probability that a gas molecule is ionized when colliding with an electron, and a probability that the gas is ionized and then collides with an electron again to return to a neutral gas. A plasma processing apparatus characterized by:

[0051] (Aspect 4) In the plasma processing apparatus according to any one of aspects 1 to 3, The input parameters include the sum of the pressure or flow rate of the process gas, the microwave output, and the current value flowing through the solenoid coil, and the output operation parameter is the position of the matching element of the automatic matching box. A plasma processing apparatus characterized by:

[0052] (Aspect 5) In the plasma processing apparatus according to any one of aspects 1 to 4, When the estimation unit includes a prediction model using learning data obtained from a plasma processing apparatus having a different vacuum vessel shape, the input data includes input parameters relating to the vacuum vessel shape. A plasma processing apparatus characterized by:

[0053] (Aspect 6) In the plasma processing apparatus according to any one of aspects 1 to 5, The learning data was obtained using a prediction model in which the operating conditions during plasma processing were used as input parameters and the average ionization energy was used as an output parameter. A plasma processing apparatus characterized by:

[0054] (Aspect 7) In the method for estimating a control parameter using the plasma processing apparatus according to the first aspect, A prediction model is generated by machine learning from learning data that is a set of input parameters and output parameters including ionization energy when the process gas is assumed to be a single gas; A control parameter estimation method for obtaining an estimated value of a control parameter by inputting the input parameter into the prediction model.

[0055] (Aspect 8) In the control parameter estimation method according to aspect 7, the ionization energy of the process gas, when assumed to be a single gas, is calculated by weighting a second physical quantity of the process gas by a first physical quantity and a partial pressure of the process gas and averaging the second physical quantity of the process gas; the first physical quantity is a physical quantity for predicting the ionization rate of each of a plurality of process gases when the process gases are ionized, The second physical quantity is the energy required to ionize each element of the process gas. A method for estimating a control parameter, comprising:

[0056] (Aspect 9) In the control parameter estimation method according to aspect 8, The first physical quantity includes a collision cross section between a gas molecule and an electron, a probability that a gas molecule is ionized when colliding with an electron, and a probability that the gas is ionized and then collides with an electron again to return to a neutral gas. A method for estimating a control parameter, comprising:

[0057] (Aspect 10) In the method for estimating a control parameter according to any one of aspects 7 to 9, The input parameters include the sum of the pressure or flow rate of the process gas, the microwave output, and the current value flowing through the solenoid coil, and the output parameter is the position of the matching element of the automatic matching box. A method for estimating a control parameter, comprising: [Explanation of symbols]

[0058] 100 Plasma processing device 101 Vacuum Processing Chamber 130 Microwave Power Supply 133 Automatic matching box 151 Electrode 152 Dielectric film 153 Electrostatic Adsorption Electrode 154 Electrostatic chucking voltage generator 158 High frequency bias power supply 160 control section

Claims

1. a processing chamber in which the sample is plasma processed; a high frequency power supply that supplies high frequency power for generating plasma; an automatic matching box for controlling high frequency reflection; a sample stage on which the sample is placed; a gas supply device for supplying a gas for plasma processing to the processing chamber; In a plasma processing apparatus having an exhaust device for exhausting the processing chamber, an estimation unit that estimates output parameters from input parameters; The plasma processing apparatus is characterized in that the estimation unit is equipped with a prediction model generated by machine learning from learning data which is a set of input parameters and output parameters including ionization energy when the process gas is assumed to be a single gas.

2. 2. The plasma processing apparatus according to claim 1, the ionization energy of the process gas, when assumed to be a single gas, is calculated by weighting a second physical quantity of the process gas by a first physical quantity and a partial pressure of the process gas and averaging the second physical quantity of the process gas; the first physical quantity is a physical quantity for predicting the ionization rate of each of a plurality of process gases when the process gases are ionized, The second physical quantity is the energy required to ionize each element of the process gas. A plasma processing apparatus characterized by:

3. 3. The plasma processing apparatus according to claim 2, The first physical quantity includes a collision cross section between a gas molecule and an electron, a probability that a gas molecule is ionized when colliding with an electron, and a probability that the gas is ionized and then collides with an electron again to return to a neutral gas. A plasma processing apparatus characterized by:

4. 2. The plasma processing apparatus according to claim 1, The input parameters include the sum of the pressure or flow rate of the process gas, the microwave output, and the current value flowing through the solenoid coil, and the output operation parameter is the position of the matching element of the automatic matching box. A plasma processing apparatus characterized by:

5. 5. The plasma processing apparatus according to claim 1, When the estimation unit includes a prediction model using learning data obtained from a plasma processing apparatus having a different vacuum vessel shape, the input data includes input parameters relating to the vacuum vessel shape. A plasma processing apparatus characterized by:

6. 5. The plasma processing apparatus according to claim 1, The learning data was obtained using a prediction model in which the operating conditions during plasma processing were used as input parameters and the average ionization energy was used as an output parameter. A plasma processing apparatus characterized by:

7. 2. The method for estimating a control parameter using the plasma processing apparatus according to claim 1, A prediction model is generated by machine learning from learning data that is a set of input parameters and output parameters including ionization energy when the process gas is assumed to be a single gas; A control parameter estimation method for obtaining an estimated value of a control parameter by inputting the input parameter into the prediction model.

8. 8. The control parameter estimation method according to claim 7, the ionization energy of the process gas, when assumed to be a single gas, is calculated by weighting a second physical quantity of the process gas by a first physical quantity and a partial pressure of the process gas and averaging the second physical quantity of the process gas; the first physical quantity is a physical quantity for predicting the ionization rate of each of a plurality of process gases when the process gases are ionized, The second physical quantity is the energy required to ionize each element of the process gas. A method for estimating a control parameter, comprising:

9. 9. The control parameter estimation method according to claim 8, The first physical quantity includes a collision cross section between a gas molecule and an electron, a probability that a gas molecule is ionized when colliding with an electron, and a probability that the gas is ionized and then collides with an electron again to return to a neutral gas. A method for estimating a control parameter, comprising:

10. 10. The control parameter estimation method according to claim 9, The input parameters include the sum of the pressure or flow rate of the process gas, the microwave output, and the current value flowing through the solenoid coil, and the output parameter is the position of the matching element of the automatic matching box. A method for estimating a control parameter, comprising:

Images