Search device, plasma processing device, and search method

Abstract

The purpose of the present invention is to improve the prediction accuracy of a training model in order to reduce the number of times of searching for the value of a parameter to be set in a plasma processing device. A search device (3) searches for the value of a first parameter to be set in a plasma processing device (1) that uses a process gas to machine a machining subject into a target. A control unit (39) selects a plurality of first combinations composed of values that the first parameter can take. The control unit (39) acquires, for each of the first combinations, the value of a second parameter indicating the state of the plasma-converted process gas and the value of a third parameter indicating an evaluation of the machining subject after machining. The control unit (39) generates a training model (82) that takes the first parameter and the second parameter as explanatory variables and the third parameter as an objective variable. The control unit (39) uses the training model (82) to predict the value of the first parameter where the value of each of the third parameters will be a prescribed value.

Description

Searching device, plasma processing apparatus, and searching method

[0001] The present invention relates to a searching device, a plasma processing device, and a searching method.

[0002] As disclosed in Patent Document 1 (JP 2019-165123 A), there is a technology that generates a learning model in which parameters set in a plasma processing apparatus are used as explanatory variables and parameters indicating the evaluation of the processing target after processing are used as target variables, and uses the learning model to search for values ​​of parameters to be set in the plasma processing apparatus in order to process the processing target into a target object.

[0003] In order to reduce the number of searches for parameter values ​​to be set in the plasma processing apparatus, it is desirable to further improve the prediction accuracy of the learning model.

[0004] A searching device according to a first aspect searches for values ​​of one or more first parameters to be set in a plasma processing apparatus that processes a processing object into a target object using a process gas. The searching device includes a control unit and a storage unit. The control unit selects a plurality of first combinations consisting of possible values ​​of the respective first parameters. The control unit sets each of the first combinations in the plasma processing apparatus, and thereby acquires, for each of the first combinations, values ​​of one or more second parameters that indicate the state of the plasma-converted process gas and values ​​of one or more third parameters that indicate an evaluation of the processing object after processing. The control unit stores learning data in the storage unit that associates the first combinations with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters. The control unit uses the learning data to generate a learning model. The learning model uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The control unit uses the learning model to predict values ​​of the first parameters such that the values ​​of the third parameters are predetermined values.

[0005] In the searching device of the first aspect, the learning model uses one or more first parameters set in the plasma processing device as explanatory variables, and also one or more second parameters that indicate the state of the process gas that has been converted into plasma. As a result, the searching device can further improve the prediction accuracy of the learning model.

[0006] A search device of a second aspect is a search device of a first aspect, in which the control unit selects, by experimental design, from a plurality of second combinations consisting of values ​​that each first parameter can take, a first combination that has a smaller number of combinations than the second combination.

[0007] As a result, the searching device can efficiently search for parameter values ​​to be set in the plasma processing device.

[0008] The searching apparatus of a third aspect is the searching apparatus of the first or second aspect, wherein the first parameter includes at least one of plasma generation power, bias power, process gas pressure, process gas flow rate, and process gas type.

[0009] A searching device of a fourth aspect is the searching device of any one of the first aspect to the third aspect, wherein the second parameter includes a parameter that directly or indirectly represents at least one of electron temperature, electron density, radical species, radical density, ion species, ion density, and ion energy.

[0010] A searching device of a fifth aspect is the searching device of the fourth aspect, wherein the parameter indirectly representing the radical species and the parameter indirectly representing the radical density include at least one of the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma, and the ratio of the emission intensities of each of the plurality of emission wavelengths of the plasma, the mass of radicals in the plasma, and the observed intensity of the mass of radicals in the plasma, and the ratio of the observed intensities of each of the plurality of masses of radicals in the plasma, or a value obtained by calculation using at least one of them.

[0011] A searching device of a sixth aspect is the searching device of the fourth aspect, wherein the parameter indirectly representing the ion species and the parameter indirectly representing the ion density include at least one of the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma and the ratio of the emission intensity of each of a plurality of emission wavelengths of the plasma, the mass of the ion, the observed intensity of the mass of the ion and the ratio of the observed intensity of each of a plurality of masses of ions, or a value obtained by calculation using at least one of them.

[0012] A searching device of a seventh aspect is a searching device of any one of the first aspect to the sixth aspect, wherein the third parameter includes at least one of film thickness, processing shape, processing speed, sputtering resistance, and deposited film composition.

[0013] An eighth aspect of the searching device is the searching device of any one of the first to seventh aspects, wherein the control unit acquires values ​​of the second parameter and the third parameter for the predicted values ​​of the first parameter by setting the predicted values ​​of the first parameter in the plasma processing apparatus. The control unit updates the learning data by adding data that associates the predicted values ​​of the first parameter with the acquired values ​​of the second parameter and the acquired values ​​of the third parameter to the learning data. The control unit updates the learning model using the updated learning data.

[0014] With this configuration, the search device of the eighth aspect can further improve the prediction accuracy of the learning model.

[0015] A plasma processing apparatus according to a ninth aspect processes a processing object into a target object using a process gas. The plasma processing apparatus includes a processing chamber, a plasma generation device, and a search device. The processing object is placed in the processing chamber. The plasma generation device generates a plasma-converted process gas in the processing chamber according to the setting of each value of one or more first parameters, and performs plasma processing on the processing object. The search device searches for each value of the first parameter to be set in the plasma generation device. The search device includes a control unit and a storage unit. The control unit selects a plurality of first combinations composed of possible values ​​of each first parameter. By setting each of the first combinations in the plasma generation device, the control unit acquires, for each of the first combinations, each value of one or more second parameters indicating a state of the plasma-converted process gas and each value of one or more third parameters indicating an evaluation of the processing object after processing. The control unit stores learning data in the storage unit, associating the first combinations with each of the acquired values ​​of the second parameters and each of the acquired values ​​of the third parameters. The control unit generates a learning model using the learning data. The learning model uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The control unit uses the learning model to predict values ​​of the first parameter such that the values ​​of the third parameter are predetermined values.

[0016] In the plasma processing apparatus of the ninth aspect, the learning model uses one or more second parameters indicating the state of the plasma-converted process gas as explanatory variables in addition to one or more first parameters set in the plasma generating apparatus, thereby enabling the plasma processing apparatus to further improve the prediction accuracy of the learning model.

[0017] A tenth aspect of the present invention relates to a search method for searching for values ​​of one or more first parameters to be set in a plasma processing apparatus that processes a processing object into a target object using a process gas. The search method includes a selection step, an acquisition step, a generation step, and a prediction step. The selection step selects a plurality of first combinations consisting of possible values ​​of the respective first parameters. The acquisition step sets each of the first combinations in the plasma processing apparatus, and thereby acquires, for each of the first combinations, values ​​of one or more second parameters that indicate the state of the plasma-converted process gas and values ​​of one or more third parameters that indicate an evaluation of the processing object after processing. The generation step stores learning data in a storage unit, associating the first combinations with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters. The generation step generates a learning model using the learning data. The learning model uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The prediction step uses the learning model to predict the values ​​of the first parameters such that the values ​​of the third parameters are predetermined values.

[0018] In the search method of the tenth aspect, the learning model uses one or more second parameters indicating the state of the plasma-converted process gas as explanatory variables in addition to one or more first parameters set in the plasma processing apparatus, thereby further improving the prediction accuracy of the learning model.

[0019] It is a functional block diagram of a plasma processing apparatus. It is a flowchart for explaining an example of a process of the plasma processing apparatus. It is a flowchart for explaining an example of a process of the plasma processing apparatus. It is a diagram showing a comparison of accuracy of learning models.

[0020] (1) Overall Configuration The plasma processing apparatus 1 processes an object to be processed into a target object using a process gas. The plasma processing apparatus 1 is, for example, a film formation apparatus, a pattern processing apparatus, an ion implantation apparatus, etc. The film formation apparatus is, for example, a CVD (Chemical Vapor Deposition) apparatus, a PVD (Physical Vapor Deposition) apparatus, etc. The pattern processing apparatus is, for example, a dry etching apparatus, etc. The ion implantation apparatus is, for example, a plasma doping apparatus, etc.

[0021] Fig. 1 is a functional block diagram of a plasma processing apparatus 1. As shown in Fig. 1, the plasma processing apparatus 1 mainly includes a processing chamber 70, a plasma generation device 2, and a search device 3. The plasma generation device 2 and the search device 3 are communicatively connected via a communication line 90.

[0022] (2) Detailed Configuration (2-1) Processing Chamber The processing chamber 70 holds an object to be processed.

[0023] Various measuring devices for measuring the state of the plasmatized process gas and the state of the processing target after processing are installed in the processing chamber 70. Examples of the measuring devices include a mass spectrometer, an optical emission spectrometer, a film thickness meter, a scanning electron microscope (SEM), an X-ray photoelectron spectroscopy (XPS), a temperature sensor, a pressure sensor, and a flow rate sensor.

[0024] (2-2) Plasma Generation Apparatus The plasma generation apparatus 2 generates plasmatized process gas in the processing chamber 70 according to the set values ​​of one or more first parameters, and performs plasma processing on the object to be processed. Hereinafter, one combination of the values ​​of one or more first parameters may be referred to as a recipe. As shown in FIG. 1 , the plasma generation apparatus 2 mainly includes a storage unit 21, a communication unit 22, and a control unit 29. The control unit 29, the storage unit 21, and the communication unit 22 are communicatively connected via internal wiring.

[0025] (2-2-1) Storage Unit The storage unit 21 is a storage device such as a RAM, a ROM, and an HDD (hard disk drive), etc. The storage unit 21 stores programs executed by the control unit 29, data necessary for executing the programs, etc.

[0026] (2-2-2) Communication Unit The communication unit 22 is a communication interface device for communicating with the searching device 3 via the communication line 90 .

[0027] (2-2-3) Control Unit The control unit 29 is a processor such as a CPU or a GPU. The control unit 29 reads and executes programs stored in the storage unit 21 to realize various functions of the plasma generation device 2. The control unit 19 can write calculation results to the storage unit 21 and read information stored in the storage unit 21 according to the programs.

[0028] As shown in FIG. 1, the control unit 29 mainly includes an acquisition unit 291 and a plasma processing unit 292 as functional blocks.

[0029] (2-2-3-1) Acquisition Unit The acquisition unit 291 acquires the values ​​of one or more first parameters from the searching device 3. The acquisition unit 291 sets the values ​​of the first parameters acquired from the searching device 3 in the plasma processing unit 292.

[0030] The first parameter includes at least one of a plasma generation power, a bias power, a process gas pressure, a process gas flow rate, and a process gas type. The plasma generation power is, for example, an ICP (Inductively Coupled Plasma) power, a CCP (Capacitively Coupled Plasma) power, an ECR (Electron Cyclotron Resonance) power, etc. The value of the process gas type is, for example, CF ４ , CHF ３ Fluorine-based gases such as CF ４ +H ２ The first parameter may further include a ratio of hydrogen contained in the process gas.

[0031] The acquisition unit 291 acquires the values ​​of one or more second parameters indicating the state of the plasmatized process gas and the values ​​of one or more third parameters indicating an evaluation of the processing target after processing, using various measuring devices installed in the processing chamber 70. The acquisition unit 291 transmits the acquired values ​​of the second parameters and the acquired values ​​of the third parameters to the searching device 3.

[0032] The second parameters include parameters that directly represent at least one of the electron temperature, electron density, radical species, radical density, ion species, ion density, and ion energy. The acquisition unit 291 acquires the electron temperature, electron density, radical species, radical density, ion species, ion density, and ion energy using, for example, a mass analyzer, an emission spectrometer, or the like.

[0033] The second parameter may include a parameter indirectly representing at least one of electron temperature, electron density, radical species, radical density, ion species, ion density, and ion energy. For example, the parameter indirectly representing the radical species and the parameter indirectly representing the radical density include at least one of the following: the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma, and a ratio of the emission intensities of each of the plurality of emission wavelengths of the plasma; the mass of radicals in the plasma, the observed intensity of the mass of radicals in the plasma, and a ratio of the observed intensities of each of the plurality of masses of radicals in the plasma, or a value obtained by calculation using at least one of them. For example, the parameter indirectly representing the ion species and the parameter indirectly representing the ion density include at least one of the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma, and a ratio of the emission intensities of each of the plurality of emission wavelengths of the plasma; the mass of an ion, the observed intensity of the mass of the ion, and a ratio of the observed intensities of each of the plurality of masses of the ion, or a value obtained by calculation using at least one of them. The acquisition unit 291 acquires the emission wavelength of the plasma and the emission intensity of the emission wavelength of the plasma using, for example, an emission spectrometer, etc. The acquisition unit 291 acquires the mass of radicals in the plasma, the observed intensity of the mass of radicals in the plasma, the mass of ions, and the observed intensity of the mass of ions using, for example, a mass spectrometer, etc.

[0034] The third parameter includes at least one of film thickness, processing shape, processing speed, sputtering resistance, and deposited film composition. The acquisition unit 291 acquires the film thickness value using, for example, a film thickness meter, SEM, etc. The processing speed is, for example, an etching rate, a film formation rate, etc. The acquisition unit 291 acquires the processing speed value using, for example, a flow rate sensor, etc. The processing shape is, for example, a processing depth, a selectivity ratio, a maximum diameter, a bottom diameter, and closure, etc. The acquisition unit 291 acquires the processing shape value using, for example, an SEM, etc. The acquisition unit 291 acquires the sputtering resistance and deposited film composition values ​​using, for example, XPS, etc.

[0035] (2-2-3-2) Plasma Processing Unit The plasma processing unit 292 generates plasma-converted process gas in the processing chamber 70 in accordance with the recipe set by the acquisition unit 291, and performs plasma processing on the object to be processed.

[0036] (2-3) Searching Device The searching device 3 searches for a recipe to be set in the plasma generating device 2. As shown in Fig. 1 , the searching device 3 mainly has a storage unit 31, a communication unit 32, an input unit 33, a display unit 34, and a control unit 39. The control unit 39, the storage unit 31, the communication unit 32, the input unit 33, and the display unit 34 are communicatively connected via internal wiring.

[0037] (2-3-1) Storage Unit The storage unit 31 is a storage device such as a RAM, a ROM, and an HDD (hard disk drive). The storage unit 31 stores programs executed by the control unit 39, data required for executing the programs, and the like.

[0038] The storage unit 31 stores learning data 81 and a learning model 82, which will be described later. The storage unit 31 also stores, for example, commercially available software for implementing experimental design and machine learning.

[0039] (2-3-2) Communication Unit The communication unit 32 is a communication interface device for communicating with the plasma generation device 2 via the communication line 90.

[0040] (2-3-3) Input Unit The input unit 33 is, for example, a keyboard and a mouse. A user can use the input unit 33 to input various commands and various pieces of information to the search device 3.

[0041] The user uses the input unit 33 to input possible values ​​of each first parameter (domain of each first parameter).

[0042] (2-3-4) Display Unit The display unit 34 is, for example, a monitor. The display unit 34 displays a screen for inputting various information, various information stored in the storage unit 31, and the like.

[0043] The display unit 34 displays, for example, the final selected recipe.

[0044] (2-3-5) Control Unit The control unit 39 is a processor such as a CPU or a GPU. The control unit 39 reads and executes programs stored in the storage unit 31 to realize various functions of the search device 3. The control unit 39 can write calculation results to the storage unit 31 and read information stored in the storage unit 31 in accordance with the programs.

[0045] As shown in FIG. 1, the control unit 39 mainly includes, as functional blocks, an acquisition unit 391, a selection unit 392, a generation unit 393, and a prediction unit 394.

[0046] (2-3-5-1) Acquisition Unit The acquisition unit 391 acquires possible values ​​for each first parameter from the user via the input unit 33. The acquisition unit 291 transmits the possible values ​​for each first parameter to the selection unit 392.

[0047] When the acquiring unit 391 receives the first recipe group from the selecting unit 392, it transmits the first recipe group to the plasma generating apparatus 2. The acquiring unit 391 acquires the values ​​of one or more second parameters and the values ​​of one or more third parameters for each recipe of the first recipe group by setting each recipe of the first recipe group in the plasma generating apparatus 2. The acquiring unit 391 transmits the first recipe group received from the selecting unit 392 and the values ​​of the second parameters and the values ​​of the third parameters acquired from the plasma generating apparatus 2 to the selecting unit 392.

[0048] When the acquiring unit 391 receives the first predicted recipe or the second predicted recipe from the predicting unit 394, it transmits the first predicted recipe or the second predicted recipe to the plasma generating apparatus 2. The acquiring unit 391 acquires the values ​​of the second parameters and the values ​​of the third parameters for the first predicted recipe or the second predicted recipe by setting the first predicted recipe or the second predicted recipe in the plasma generating apparatus 2. The acquiring unit 391 transmits the first predicted recipe or the second predicted recipe received from the predicting unit 394 and the values ​​of the second parameters and the third parameters acquired from the plasma generating apparatus 2 to the selecting unit 392.

[0049] (2-3-5-2) Selection Unit When the selection unit 392 receives the possible values ​​of each first parameter from the acquisition unit 391, it selects a group of first recipes (plurality of first combinations) consisting of the possible values ​​of each first parameter.

[0050] Specifically, the selection unit 392 first defines a second recipe group (plurality of second combinations) consisting of values ​​that each of the first parameters can take. For example, the second recipe group is a set of all combinations consisting of values ​​that each of the first parameters can take.

[0051] Next, the selection unit 392 selects, from the second recipe group, a first recipe group that has fewer recipes than the second recipe group using a design of experiments. The design of experiments may be, for example, a D-optimal design, an orthogonal array, or a central composite design. The selection unit 392 executes the design of experiments using, for example, commercially available software for implementing the design of experiments.

[0052] The selection unit 392 transmits the selected first recipe group to the acquisition unit 391 .

[0053] When the selection unit 392 receives the first recipe group, the values ​​of the second parameters, and the values ​​of the third parameters from the acquisition unit 391, it determines whether there is a recipe in the first recipe group in which the values ​​of the third parameters are first predetermined values. The first predetermined values ​​are the values ​​of the third parameters of the target object. If there is a recipe, the selection unit 392 selects the recipe. In other words, the recipe is a recipe for processing the processing object into the target object. If there is no recipe, the selection unit 392 transmits the first recipe group, the values ​​of the second parameters, and the values ​​of the third parameters received from the acquisition unit 391 to the generation unit 393.

[0054] When the selection unit 392 receives the first predicted recipe or the second predicted recipe, the values ​​of the second parameter, and the values ​​of the third parameter from the acquisition unit 391, the selection unit 392 determines whether a first predicted recipe or a second predicted recipe exists in which the values ​​of the third parameter are the first predetermined value. If a first predicted recipe or a second predicted recipe exists, the selection unit 392 selects the recipe. In other words, the recipe is a recipe for processing the processing object into a target object. If a first predicted recipe or a second predicted recipe does not exist, the selection unit 392 transmits the first predicted recipe or the second predicted recipe, the values ​​of the second parameter, and the values ​​of the third parameter received from the acquisition unit 391 to the generation unit 393.

[0055] (2-3-5-3) Generator Upon receiving the first recipe group, the values ​​of the second parameter, and the values ​​of the third parameter from the selector 392, the generator 393 stores training data 81 in the storage unit 31, which associates the recipes of the first recipe group with the values ​​of the second parameter and the values ​​of the third parameter. The generator 393 uses the training data 81 to generate a training model 82. The training model 82 uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The training model 82 is, for example, Gaussian process regression, linear regression, or a regression tree. The generator 393 generates the training model 82 using, for example, commercially available software for implementing machine learning.

[0056] When the generation unit 393 receives the first predicted recipe or the second predicted recipe from the selection unit 392, the generation unit 393 updates the training data 81 by adding data that associates each of the first predicted recipe or the second predicted recipe with each value of the second parameter and each value of the third parameter to the training data 81. The generation unit 393 updates the training model 82 using the updated training data 81.

[0057] (2-3-5-4) Prediction Unit When the generation unit 393 generates or updates the learning model 82, the prediction unit 394 uses the learning model 82 to predict a recipe in which each value of the third parameter becomes the first predetermined value.

[0058] Specifically, the prediction unit 394 defines a third recipe group having a greater number of recipes than the second recipe group and inputs each recipe in the third recipe group into the learning model 82. For example, the third recipe group is a set of all combinations consisting of possible values ​​of each first parameter after increasing the possible values ​​of each first parameter. At this time, each value of the second parameter to be input into the learning model 82 is predicted, for example, from each recipe in the third recipe group. For example, the generation unit 393 uses learning data that associates the recipes in the first recipe group with each value of the second parameter to generate a learning model in which the first parameter is an explanatory variable and the second parameter is an objective variable. The prediction unit 394 inputs each recipe in the third recipe group into the learning model to predict each value of the second parameter to be input into the learning model 82.

[0059] Each value of the second parameter input to the learning model 82 may be, for example, a predefined value. For example, if the second parameter is ion energy, five levels of ion energy, 100, 150, 200, 250, and 300 (unit: eV), are predefined. The prediction unit 394 inputs five ion energy values ​​(five levels) to the learning model 82 for each recipe in the third recipe group.

[0060] The prediction unit 394 determines whether a recipe exists in which the values ​​of the third parameters output from the learning model 82 are first predetermined values. If a recipe exists, the prediction unit 394 transmits to the acquisition unit 391 a recipe (first predicted recipe) in which the values ​​of the third parameters are first predetermined values. If a recipe does not exist, the prediction unit 394 extracts one or more recipes (second predicted recipes) from a set of the third recipe group excluding recipes included in the current learning data 81, and transmits the second predicted recipes to the acquisition unit 391. For example, if the learning model 82 is Gaussian process regression, the prediction unit 394 extracts the second predicted recipe using Bayesian optimization.

[0061] (3) Processing An example of processing performed by the plasma processing apparatus 1 will be described with reference to the flowcharts of FIGS. 2A to 2C.

[0062] As shown in step S1, the user inputs the domain of each first parameter to the search device 3.

[0063] After step S1 is completed, as shown in step S2, the searching device 3 defines a second recipe group and selects a first recipe group from the second recipe group using the experimental design method.

[0064] After completing step S2, the searching device 3 transmits the first recipe group to the plasma generating device 2 as shown in step S3.

[0065] After completing step S3, as shown in step S4, the plasma generating device 2 sets the first recipe group in the plasma processing unit 292 and performs plasma processing. The plasma generating device 2 transmits the values ​​of the second parameters and the third parameters acquired by the plasma processing to the searching device 3.

[0066] After step S4, the searching device 3 determines whether or not there is a recipe in the first recipe group in which the values ​​of the third parameters are the first predetermined values, as shown in step S5. If there is a recipe, the process proceeds to step S14. If there is no recipe, the process proceeds to step S6.

[0067] When proceeding from step S5 to step S6, the searching device 3 stores learning data 81 that associates the recipes of the first recipe group with the values ​​of the second parameters and the values ​​of the third parameters in the storage unit 31. The searching device 3 uses the learning data 81 to generate a learning model 82.

[0068] After completing step S6, the search device 3 defines a third recipe group and inputs each recipe in the third recipe group into the learning model 82, as shown in step S7.

[0069] After step S7, as shown in step S8, the searching device 3 determines whether or not there is a recipe in which the values ​​of the third parameters output from the learning model 82 are the first predetermined values. If there is a recipe, the process proceeds to step S9. If there is no recipe, the process proceeds to step S10.

[0070] When the process proceeds from step S8 to step S9, the searching device 3 transmits the first predicted recipe to the plasma generating device 2.

[0071] When the process proceeds from step S8 to step S10, the searching device 3 transmits the second predicted recipe to the plasma generating device 2.

[0072] After completing step S9 or step S10, as shown in step S11, the plasma generation apparatus 2 sets the first predicted recipe or the second predicted recipe in the plasma processing unit 292 and performs plasma processing. The plasma generation apparatus 2 transmits the values ​​of the second parameters and the third parameters acquired by the plasma processing to the searching device 3.

[0073] After step S11, the searching device 3 determines whether or not there exists a first predicted recipe or a second predicted recipe in which the values ​​of the third parameters are equal to the first predetermined values, as shown in step S12. If there exists a first predicted recipe or a second predicted recipe, the process proceeds to step S14. If there does not exist a first predicted recipe, the process proceeds to step S13.

[0074] When proceeding from step S12 to step S13, the searching device 3 updates the training data 81 by adding data associating each of the first predicted recipe or the second predicted recipe with each value of the second parameter and each value of the third parameter to the training data 81. The generating unit 393 updates the training model 82 using the updated training data 81.

[0075] After completing step S13, the process returns to step S7, and the search device 3 inputs each recipe in the third recipe group into the learning model 82.

[0076] When the process proceeds from step S5 or step S12 to step S14, the searching device 3 selects a recipe in which the value of the third parameter becomes the first predetermined value. In other words, the searching device 3 selects a recipe for processing the processing object into a target object.

[0077] (4) Verification (4-1) Premise In this verification, we verified how many times the plasma processing apparatus 1 of this embodiment performs plasma processing (the number of recipes to be set in the plasma generation apparatus 2) before selecting a recipe for processing the processing object into a target object.

[0078] In this verification, the plasma processing apparatus 1 is a dry etching apparatus. In other words, the plasma processing apparatus 1 processes an object to be processed into a target object by plasma processing using an etching gas (process gas).

[0079] The source power and bottom power supplied to the processing chamber 70 are high frequency power supplies of 13.56 MHz.

[0080] The processing target is SiO ２ (TEOS).

[0081] The first parameters are the ICP power, the bias power, the pressure of the etching gas, the flow rate of the etching gas, the proportion of hydrogen contained in the etching gas, and the type of the etching gas.

[0082] As the domain of the first parameter, the ICP power has four levels of values, namely (400, 600, 800, 1000) (unit: W). The bias power has three levels of values, namely (100, 200, 300) (unit: W). The etching gas pressure has four levels of values, namely (10, 20, 30, 40) (unit: mTorr). The etching gas flow rate has three levels of values, namely (50, 75, 100) (unit: sccm). The hydrogen ratio contained in the etching gas has four levels of values, namely (0, 0.25, 0.5, 0.75). The type of etching gas is selected from four levels (CF) corresponding to the respective values ​​of the hydrogen ratio contained in the etching gas. ４ , C.F. ４ +H ２ , C.F. ４ +H ２ , C.F. ４ +H ２ ) has four levels of values.

[0083] The second parameter is H ２ This is the observed intensity of F+ ions (mass m / z=21).

[0084] The third parameter is the etching rate.

[0085] The second recipe group is a set of all combinations of possible values ​​for each first parameter. Since the ICP power, bias power, etching gas pressure, etching gas flow rate, and hydrogen content in the etching gas each have four, three, four, three, and four levels, respectively, the second recipe group is made up of 576 (= 4 x 3 x 4 x 3 x 4) recipes. Note that the value of the hydrogen content in the etching gas and the value of the type of etching gas correspond one-to-one, so the type of etching gas is omitted.

[0086] The third recipe group includes all possible combinations of the first parameters, where the ICP power has seven levels of values ​​(400, 500, 600, 700, 800, 900, 1000) (unit: W), the bias power has five levels of values ​​(100, 150, 200, 250, 300) (unit: W), the etching gas pressure has seven levels of values ​​(10, 15, 20, 25, 30, 35, 40) (unit: mTorr), the etching gas flow rate has six levels of values ​​(50, 60, 70, 80, 90, 100) (unit: sccm), and the hydrogen ratio in the etching gas has four levels of values ​​(0, 0.25, 0.5, 0.75). The third recipe group includes 5,880 (= 7 x 5 x 7 x 6 x 4) recipes.

[0087] The experimental design is a D-optimal design.

[0088] The training model 82 is a Gaussian process regression.

[0089] Since the learning model 82 is Gaussian process regression, the prediction unit 394 extracts the second predicted recipe from the set obtained by excluding the recipes included in the current learning data 81 from the third recipe group, in order from the recipes whose surface or curve indicating the expected value is closest to the first predetermined value range.

[0090] (4-2) Results As a result, recipes for processing the processing object into the target object were selected simply by setting 45 recipes out of a total of 576 recipes in the plasma generating apparatus 2. Of the 45 recipes, 2 recipes were selected for processing the processing object into the target object, and 43 recipes were recipes used in the learning data 81.

[0091] The accuracy of the final learning model 82 (first learning model 821) was also compared with that of a second learning model 822 generated using learning data obtained by removing the second parameter from the final learning data 81. FIG. 3 shows a comparison of the accuracy of the learning models. The right diagram of FIG. 3 shows the accuracy of the first learning model 821, and the left diagram of FIG. 3 shows the accuracy of the second learning model 822. The plot points correspond to the 43 recipes included in the final learning data 81. The horizontal axis represents the etching rate obtained by setting the recipe in the plasma generation apparatus 2. The vertical axis represents the etching rate predicted by inputting the recipe into the learning model (in the first learning model 821 shown on the right, in addition to the recipe, the value of the second parameter is also input into the learning model). Therefore, it is desirable for the plot points to be aligned diagonally.

[0092] As shown in Figure 3, the plot points from the first learning model 821 are arranged more diagonally than the plot points from the second learning model 822. The coefficient of determination for the first learning model 821 was 0.9347, and for the second learning model 822 was 0.8885.

[0093] (5) Features (5-1) Conventionally, there is a technology that generates a learning model in which parameters set in a plasma processing device are used as explanatory variables and parameters indicating the evaluation of a processing target after processing are used as objective variables, and uses the learning model to search for values ​​of parameters to be set in the plasma processing device in order to process the processing target into a target object.

[0094] In order to reduce the number of searches for parameter values ​​to be set in the plasma processing apparatus, it is desirable to further improve the prediction accuracy of the learning model.

[0095] The searching device 3 of this embodiment searches for values ​​of one or more first parameters to be set in the plasma processing device 1, which processes a processing object into a target object using a process gas. The searching device 3 includes a control unit 39 and a storage unit 31. The control unit 39 selects multiple first combinations consisting of possible values ​​of each first parameter. By setting each first combination in the plasma processing device 1, the control unit 39 acquires, for each first combination, values ​​of one or more second parameters indicating the state of the plasma-converted process gas and values ​​of one or more third parameters indicating an evaluation of the processing object after processing. The control unit 39 stores learning data 81 in the storage unit 31, associating the first combination with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters. The control unit 39 uses the learning data 81 to generate a learning model 82. The learning model 82 uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The control unit 39 uses the learning model 82 to predict the values ​​of the first parameters such that the values ​​of the third parameters are predetermined values.

[0096] In the searching device 3 of this embodiment, the learning model 82 uses, as explanatory variables, one or more second parameters that indicate the state of the plasmatized process gas in addition to one or more first parameters that are set in the plasma processing device 1. As a result, the searching device 3 can further improve the prediction accuracy of the learning model 82.

[0097] (5-2) In the searching device 3 of this embodiment, the control unit 39 selects, by experimental design, from among a plurality of second combinations constituted by values ​​that each of the first parameters can take, a first combination that has a smaller number of combinations than the second combination.

[0098] As a result, the searching device 3 can efficiently search for parameter values ​​to be set in the plasma processing device.

[0099] (5-3) In the searching device 3 of this embodiment, the first parameters include at least one of the plasma generation power, bias power, process gas pressure, process gas flow rate, and process gas type.

[0100] (5-4) In the searching device 3 of this embodiment, the second parameters include a parameter that directly or indirectly represents at least one of the electron temperature, electron density, radical species, radical density, ion species, ion density, and ion energy.

[0101] (5-5) In the searching device 3 of this embodiment, the parameters indirectly representing the radical species and the parameters indirectly representing the radical density include at least one of the following: the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma, the ratio of the emission intensities of each of the multiple emission wavelengths of the plasma, the mass of radicals in the plasma, the observed intensity of the mass of radicals in the plasma, and the ratio of the observed intensities of each of the multiple masses of radicals in the plasma, or a value obtained by calculation using at least one of them.

[0102] (5-6) In the searching device 3 of this embodiment, the parameter indirectly representing the ion species and the parameter indirectly representing the ion density include at least one of the following: the emission wavelength of the plasma, the emission intensity of the emission wavelength of the plasma, the ratio of the emission intensity of each of the plurality of emission wavelengths of the plasma, the mass of the ion, the observed intensity of the mass of the ion, and the ratio of the observed intensity of each of the plurality of masses of the ions, or a value obtained by calculation using at least one of them.

[0103] (5-7) In the searching device 3 of this embodiment, the third parameters include at least one of the film thickness, the processed shape, the processing speed, the sputtering resistance, and the composition of the deposited film.

[0104] (5-8) In the searching device 3 of this embodiment, the control unit 39 acquires the values ​​of the second parameters and the values ​​of the third parameters for the predicted values ​​of the first parameters by setting the predicted values ​​of the first parameters in the plasma processing device 1. The control unit 39 updates the learning data 81 by adding data that associates the predicted values ​​of the first parameters with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters to the learning data 81. The control unit 39 updates the learning model 82 using the updated learning data 81.

[0105] As a result, the search device 3 can further improve the prediction accuracy of the learning model 82.

[0106] (5-9) The plasma processing apparatus 1 of this embodiment processes a processing object into a target object using a process gas. The plasma processing apparatus 1 includes a processing chamber 70, a plasma generation device 2, and a search device 3. The processing object is placed in the processing chamber 70. The plasma generation device 2 generates plasma-converted process gas in the processing chamber 70 according to the setting of each value of one or more first parameters, and performs plasma processing on the processing object. The search device 3 searches for each value of the first parameter to be set in the plasma generation device 2. The search device 3 includes a control unit 39 and a storage unit 31. The control unit 39 selects multiple first combinations composed of possible values ​​for each first parameter. By setting each first combination in the plasma generation device 2, the control unit 39 acquires, for each first combination, each value of one or more second parameters indicating the state of the plasma-converted process gas and each value of one or more third parameters indicating an evaluation of the processing object after processing. The control unit 39 stores learning data 81 in the storage unit 31, which associates the first combination with each of the acquired values ​​of the second parameter and each of the acquired values ​​of the third parameter. The control unit 39 uses the learning data 81 to generate a learning model 82. The learning model 82 uses the first parameter and the second parameter as explanatory variables and the third parameter as a target variable. The control unit 39 uses the learning model 82 to predict each value of the first parameter such that each value of the third parameter becomes a predetermined value.

[0107] In the plasma processing apparatus 1 of this embodiment, the learning model 82 uses, as explanatory variables, one or more second parameters that indicate the state of the plasmatized process gas, in addition to one or more first parameters that are set in the plasma generating apparatus 2. As a result, the plasma processing apparatus 1 can further improve the prediction accuracy of the learning model 82.

[0108] (5-10) The search method of this embodiment searches for values ​​of one or more first parameters to be set in a plasma processing apparatus 1 that processes a processing object into a target object using a process gas. The search method includes a selection step S2, an acquisition step S4, a generation step S6, and prediction steps S7 and S8. The selection step S2 selects multiple first combinations composed of possible values ​​of each first parameter. The acquisition step S4 acquires, for each first combination, values ​​of one or more second parameters indicating the state of the plasma-converted process gas and values ​​of one or more third parameters indicating an evaluation of the processing object after processing by setting each first combination in the plasma processing apparatus 1. The generation step S6 stores training data 81 in the memory unit 31, associating the first combination with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters. The generation step S6 generates a training model 82 using the training data 81. The learning model 82 uses the first parameter and the second parameter as explanatory variables and the third parameter as a response variable. Prediction steps S7 and S8 use the learning model 82 to predict the values ​​of the first parameter such that the values ​​of the third parameter are predetermined values.

[0109] In the search method of this embodiment, the learning model 82 uses, as explanatory variables, one or more second parameters that indicate the state of the plasmatized process gas, in addition to one or more first parameters that are set in the plasma processing apparatus 1. As a result, the search method can further improve the prediction accuracy of the learning model 82.

[0110] (6) Modifications (6-1) Modification 1A In this embodiment, the plasma processing apparatus 1 includes the searching device 3. However, the searching device 3 may be separated from the plasma processing apparatus 1. In this case, the communication line 90 is, for example, the Internet.

[0111] (6-2) Although the embodiments of the present disclosure have been described above, it will be understood that various changes in form and details can be made without departing from the spirit and scope of the present disclosure as defined in the claims.

[0112] REFERENCE SIGNS LIST 1 Plasma processing apparatus 2 Plasma generation apparatus 3 Searching apparatus 31 Storage unit 39 Control unit 70 Processing chamber 81 Learning data 82 Learning model S2 Selection step S4 Acquisition step S6 Generation step S7, S8 Prediction step

[0113] Japanese Patent Application Laid-Open No. 2019-165123

Claims

1. A searching device (3) for searching for values ​​of one or more first parameters to be set in a plasma processing device (1) for processing an object to be processed into a target object using a process gas, the searching device comprising: a control unit (39) and a memory unit (31), wherein the control unit: selects a plurality of first combinations consisting of possible values ​​of each of the first parameters; by setting each of the first combinations in the plasma processing device, obtains, for each of the first combinations, values ​​of one or more second parameters indicating a state of the process gas that has been plasmatized and values ​​of one or more third parameters indicating an evaluation of the object to be processed after processing; stores in the memory unit learning data (81) that associates the first combinations with the obtained values ​​of the second parameters and the obtained values ​​of the third parameters; and uses the learning data to generate a learning model (82) in which the first parameter and the second parameter are explanatory variables and the third parameter is a target variable; and uses the learning model to predict the values ​​of each of the first parameters such that the values ​​of the third parameters are predetermined values. Search device (3).

2. The search device (3) of claim 1, wherein the control unit selects the first combination having a smaller number of combinations than the second combination from among a plurality of second combinations constituted by values ​​that each of the first parameters can take, by an experimental design method.

3. The searching device (3) according to claim 1 or 2, wherein the first parameter includes at least one of a plasma generation power, a bias power, a pressure of the process gas, a flow rate of the process gas, and a type of the process gas.

4. The searching device (3) according to any one of claims 1 to 3, wherein the second parameter includes a parameter that directly or indirectly represents at least one of an electron temperature, an electron density, a radical species, a radical density, an ion species, an ion density, and an ion energy.

5. The searching device described in claim 4, wherein the parameter indirectly representing the radical species and the parameter indirectly representing the radical density include at least one of the following, or a value obtained by calculation using at least one of: the emission wavelength of the plasma, and the emission intensity of the emission wavelength of the plasma, a ratio of the emission intensities of each of a plurality of emission wavelengths of the plasma, the mass of radicals in the plasma, and the observed intensity of the mass of radicals in the plasma, and a ratio of the observed intensities of each of a plurality of masses of radicals in the plasma.

6. The searching device according to claim 4, wherein the parameter indirectly representing the ion species and the parameter indirectly representing the ion density include at least one of the following, or a value obtained by calculation using at least one of: an emission wavelength of plasma, and an emission intensity of the emission wavelength of plasma, a ratio of the emission intensities of each of a plurality of emission wavelengths of plasma, an ion mass, and an observed intensity of the ion mass, and a ratio of the observed intensities of each of a plurality of ion masses.

7. The searching device (3) according to any one of claims 1 to 6, wherein the third parameter includes at least one of a film thickness, a processing shape, a processing speed, a sputtering resistance, and a deposition film composition.

8. An exploration device (3) described in any one of claims 1 to 7, wherein the control unit: obtains each of the second parameter values ​​and each of the third parameter values ​​for each of the predicted values ​​of the first parameters by setting each of the predicted values ​​of the first parameters in the plasma processing apparatus; updates the learning data by adding data correlating each of the predicted values ​​of the first parameters, each of the obtained values ​​of the second parameters, and each of the obtained values ​​of the third parameters to the learning data; and updates the learning model using the updated learning data.

9. A plasma processing apparatus (1) for processing an object to be processed into a target object using a process gas, comprising: a processing chamber (70) in which the object to be processed is placed; a plasma generating apparatus (2) for generating the process gas in a plasma form in the processing chamber in accordance with the setting of each value of one or more first parameters, and performing plasma processing of the object to be processed; and a searching apparatus (3) for searching for each value of the first parameters to be set in the plasma generating apparatus, wherein the searching apparatus has a control unit (39) and a memory unit (31), and the control unit selects a plurality of first combinations constituted by values ​​that each of the first parameters can take, and by setting each of the first combinations in the plasma generating apparatus, obtains, for each of the first combinations, each value of one or more second parameters indicating a state of the process gas in a plasma form, and each value of one or more third parameters indicating an evaluation of the object to be processed after processing, the plasma processing apparatus (1) storing learning data (81) in the memory unit, the learning data associating the first combination with each of the acquired values ​​of the second parameters and each of the acquired values ​​of the third parameters, and generating a learning model (82) using the learning data, the first parameters and the second parameters being explanatory variables and the third parameter being a target variable; and predicting, using the learning model, each of the values ​​of the first parameters such that each of the values ​​of the third parameters becomes a predetermined value.

10. A search method for searching for values ​​of one or more first parameters to be set in a plasma processing apparatus (1) that processes an object to be processed into a target object using a process gas, comprising: a selection step (S2) of selecting a plurality of first combinations consisting of possible values ​​of each of the first parameters; an acquisition step (S4) of acquiring, for each of the first combinations, values ​​of one or more second parameters indicating a state of the process gas plasmatized and values ​​of one or more third parameters indicating an evaluation of the object to be processed after processing by setting each of the first combinations in the plasma processing apparatus; a generation step (S6) of storing learning data that associates the first combinations with the acquired values ​​of the second parameters and the acquired values ​​of the third parameters in the storage unit, and using the learning data to generate a learning model in which the first parameter and the second parameter are explanatory variables and the third parameter is a target variable; and a prediction step (S7, S8) of predicting, using the learning model, values ​​of the first parameters such that the values ​​of the third parameters are predetermined values. The searching method comprises:

Images