Method and apparatus for predicting β-Ga2O3 epitaxial layers grown by MOCVD

Abstract

To provide a prediction method and a device for epitaxial layer of β-Ga2O3 grown by MOCVD.SOLUTION: The method includes acquiring preset growth parameters, and inputting the preset growth parameters into a trained epitaxial layer prediction model, the preset growth parameters include an O / Ga ratio, chamber pressure, temperature, and growth time; outputting predicted data of the β-Ga2O3 epitaxial layer corresponding to the preset growth parameters by the epitaxial layer prediction model. In the present disclosure, the machine learning model is used to predict the epitaxial layer and the preset growth parameters are adjusted according to the predicted data to select the growth parameters suitable for the process window of β-Ga2O3 epitaxial growth, and then the generated parameters are input into MOCVD for epitaxy, and the quality of β-Ga2O3 epitaxial layer grown by MOCVD is improved.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the technical field of β-Ga2O3 epitaxial layer growth, and more particularly to a method and apparatus for predicting the growth of β-Ga2O3 epitaxial layers grown by MOCVD. [Background technology]

[0002] When growing β-Ga2O3 using the conventional MOCVD epitaxial method, the high reaction temperature of TMGa (trimethylgallium, Ga(CH3)3) and the strong oxidizing properties of O2 can easily cause pre-reaction in a high-temperature environment. That is, the reactants may react before reaching the base, resulting in the formation of β-Ga2O3. The powder formed by this pre-reaction can become new nucleation points on the surface of the base, promoting the three-dimensional growth of the epitaxial layer and significantly affecting the quality of the epitaxial layer.

[0003] Therefore, there is a need for improvements and modifications to the conventional technology. Summary of the Invention [Problem to be solved by the invention]

[0004] The technical problem to be solved by the present invention is to provide a method and related apparatus for predicting the growth of a β-Ga2O3 epitaxial layer grown by MOCVD to overcome the drawbacks of the prior art. [Means for solving the problem]

[0005] To solve the above technical problem, a first embodiment of the present invention provides a method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, which includes the steps of obtaining preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model, and outputting predicted data of a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model, where the preset growth parameters include the O / Ga ratio, chamber pressure, temperature, and growth time, and the predicted data includes at least a FWHM ratio.

[0006] In the method for predicting the quality of a β-Ga2O3 epitaxial layer grown by MOCVD, the training process of the epitaxial layer prediction model is specifically as follows: obtaining a training data set, the training data set including a plurality of training data sets, each of which includes training growth data and label data of a β-GaO epitaxial layer; inputting the training growth parameters in the training data set into a preset machine learning model, and outputting training prediction data for the β-Ga2O3 epitaxial layer through the preset machine learning model; and obtaining a trained epitaxial layer prediction model by training the pre-configured machine learning model with the training prediction data and the label data.

[0007] In the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, the step of training the pre-defined machine learning model using the training prediction data and the label data specifically includes: calculating a predicted difference value between the training predicted data and the labeled data; When the predicted difference value is less than or equal to a preset difference value threshold, determining a loss function term according to the training predicted data and the label data, and training the preset machine learning model according to the loss function term; When the predicted difference value is greater than a preset difference value threshold, a loss function term is determined according to the training predicted data and the label data, and a penalty term is set for the training predicted data, and then the preset machine learning model is trained according to the loss function term and the penalty term.

[0008] In the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, the training prediction data and the label data both include FWHM ratios, and the step of calculating a predicted difference value between the training prediction data and the label data specifically includes: Reading out the predicted FWHM ratio in the training prediction data and the labeled FWHM ratio in the labeled data, wherein the predicted FWHM ratio and the labeled FWHM ratio are both used to reflect the difference between the FWHM after epitaxy and the FWHM before epitaxy; and calculating a difference between the predicted FWHM ratio and the labeled FWHM ratio to obtain a predicted difference value.

[0009] In the method for predicting the quality of a β-Ga2O3 epitaxial layer grown by MOCVD, the predicted data further includes the thickness of the epitaxial layer and / or the surface roughness of the epitaxial layer.

[0010] In the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, after outputting predicted data of the β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model, the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD includes: Detecting whether the predicted data meets preset requirements; When the forecast data does not satisfy the preset requirements, modifying the preset growth parameters to obtain modified growth parameters; performing the steps of setting the modified growth parameters as preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model multiple times until the predicted data meets the preset requirements; The preset growth parameters corresponding to the prediction data that meets the prediction requirements are set as target growth parameters, and the target growth parameters are input into MOCVD to perform β-Ga2O3 epitaxial growth.

[0011] The method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD further includes generating a fine-tuning dataset based on target growth parameters and corresponding epitaxial data, and fine-tuning the epitaxial layer prediction model with the fine-tuning dataset.

[0012] A second embodiment of the present invention provides an apparatus for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, which includes: an acquisition module for acquiring preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model; and a prediction module for outputting predicted data of a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model, where the preset growth parameters include an O / Ga ratio, a chamber pressure, a temperature, and a growth time, and the predicted data includes at least a FWHM ratio.

[0013] In a third embodiment of the present invention, a computer-readable storage medium is provided, which stores one or more programs that can be executed by one or more processors to perform the steps of any one of the methods for predicting the quality of a β-Ga2O3 epitaxial layer grown by MOCVD.

[0014] In a fourth embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, and the memory stores a computer-readable program that can be executed by the processor, and the processor can execute the computer-readable program to perform each step of the method for predicting the quality of any one of the β-Ga2O3 epitaxial layers grown by MOCVD. [Effects of the Invention]

[0015] Compared with conventional techniques, the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD of the present invention achieves the following inventive advantages. The method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD includes the steps of acquiring preset growth parameters, inputting the preset growth parameters into a trained epitaxial layer prediction model, and outputting predicted data for a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model. The preset growth parameters include the O / Ga ratio, chamber pressure, temperature, and growth time. The present invention predicts the epitaxial layer using a machine learning model and selects growth parameters for a process window suitable for β-Ga2O3 epitaxial growth by adjusting the preset growth parameters based on the predicted data. The growth parameters are then input to MOCVD for epitaxy, thereby improving the quality of the β-Ga2O3 epitaxial layer grown by MOCVD epitaxy. [Brief explanation of the drawings]

[0016] In order to explain the technical matters relating to the embodiments of the present invention in more detail, the drawings used in the embodiments of the present invention will be briefly described below. It should be noted that the drawings are only a part of the embodiments of the present invention, and therefore, a person skilled in the art can imagine other drawings based on the drawings without creative research, and such drawings are also included in the present invention. [Figure 1] 1 is a flow chart showing a method for predicting a β-Ga 2 O 3 epitaxial layer grown by MOCVD according to an embodiment of the present invention. [Figure 2] 1 is a flow chart illustrating the principle of a specific implementation method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD according to an embodiment of the present invention. [Figure 3] FIG. 10 shows the coefficient of determination for epitaxial layer thickness of the epitaxial layer prediction model obtained by training on the training dataset. [Figure 4] FIG. 10 shows the coefficient of determination of the FWHM ratio of the epitaxial layer prediction model obtained by training on the training data set. [Figure 5] FIG. 10 shows the coefficient of determination of epitaxial layer thickness for the epitaxial layer prediction model in the test set. [Figure 6] FIG. 10 is a diagram showing the coefficient of determination of the FWHM ratio of the epitaxial layer prediction model in the test set. [Figure 7] FIG. 10 compares the training dataset with the augmented training dataset. [Figure 8] FIG. 1 is a diagram showing the principle of a prediction device for a β-Ga2O3 epitaxial layer grown by MOCVD according to an embodiment of the present invention. [Figure 9] FIG. 2 is a diagram illustrating the structure of a terminal device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0017] In an embodiment of the present invention, a method for predicting the quality of a β-Ga2O3 epitaxial layer grown by MOCVD and a related apparatus are provided. In order to clearly understand the objectives, technical features, and advantages of the present invention in detail, the following detailed description of the present invention is provided with reference to the accompanying drawings. It should be noted that the following specific examples are for illustrative purposes only and are not intended to limit the present invention.

[0018] As those skilled in the art will appreciate, unless otherwise specified, the terms "a," "one," "the," "the," and the like in this specification can include not only one item but also multiple items. The term "comprising" in this specification means that features, integers, steps, operations, components, and / or modules are included, but can also mean that one or more combinations of features, integers, steps, operations, components, and / or modules are present or added. It should be noted that when a component is "coupled" or "connected" to another component, the component can be directly coupled or connected to the other component, or indirectly coupled or connected to the other component via an intermediate component. In this specification, "coupled" or "connected" can be a wireless coupling or connection. In this specification, the term "and / or" means the inclusion of any one or more, any combination, or all of the associated component or components.

[0019] As known to those skilled in the art, unless otherwise specified, the terms in this specification (including technical and scientific terms) refer to the meanings of the terms commonly used by those skilled in the art. It should be noted that the meanings of the terms described in the present invention and prior art can refer to the meanings of the terms described in commonly used dictionaries, and if there is a special definition, the meaning of the special definition can be expressed.

[0020] The numbers or values ​​attached to each step in the embodiments of the present invention do not indicate the order in which each step is performed, but the order in which each step is performed is determined by its function and logic, that is, the present invention does not intend to limit the order in which each step in the embodiments of the present invention is performed by the numbers or values.

[0021] When growing β-Ga2O3 using the conventional MOCVD epitaxial method, the high reaction temperature of TMGa and the strong oxidizing properties of O2 can easily cause pre-reaction in a high-temperature environment. That is, reactants may react before reaching the substrate, resulting in the formation of β-Ga2O3. The powder formed by this pre-reaction can become new nucleation points on the surface of the substrate, promoting the three-dimensional growth of the epitaxial layer and significantly affecting the quality of the epitaxial layer.

[0022] To solve this problem, some researchers have proposed using TMGa and O2, which have low reaction dimensions. However, due to the low lateral thermal diffusivity of the atoms on the basal plane, it is necessary to artificially create a bevel cut angle and shorten the step width by beveling the basal plane. However, because the temperatures used in MOCVD epitaxial systems vary during epitaxy, the substrate must be beveled at various angles. This increases manufacturing costs and difficulty, making this method ineffective. Some researchers have also proposed using N2O, which has a higher cracking temperature, as the oxygen source, so that TMGa and N2O are activated and react only when close to the substrate. However, epitaxial growth of oxides at extremely high temperatures is difficult for MOCVD systems, and various technical challenges must be overcome to ensure the system's lifespan and the uniformity of the epitaxial layer.

[0023] According to the research of the present inventors, the reason why high-quality epitaxial growth cannot be achieved when TMGa and O2 are used as reactants is that the parameters of the conventional MOCVD process are adjusted based on the subjective experience of the operator when performing the MOCVD process. When adjusting the parameters of the MOCVD process based on the subjective experience of the operator, judgment errors can occur, making it difficult to obtain the optimal process window. In particular, when growing β-Ga2O3 epitaxially, it is very difficult to obtain the optimal process window for epitaxial growth, so it is necessary to completely eliminate judgment errors due to the subjective experience of the operator.

[0024] In this example, preset growth parameters are acquired and input into a trained epitaxial layer prediction model. The preset growth parameters include the O / Ga ratio, chamber pressure, temperature, and growth time. The epitaxial layer prediction model outputs predicted data for a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters. The present invention employs machine learning models to predict the epitaxial layer, and then adjusts the preset growth parameters based on the predicted data to select growth parameters within a process window suitable for β-Ga2O3 epitaxial growth. The growth parameters are then output to MOCVD for epitaxy. By combining machine learning and the MOCVD process, TMGa and O2 are used as reactants, and high-quality two-dimensional step flow growth is achieved on a substrate without intentional oblique cuts, ensuring the quality of the β-Ga2O3 epitaxial layer grown by MOCVD.

[0025] The technical matters of the present invention will be explained in more detail below with reference to the drawings and examples.

[0026] In this embodiment, a prediction method for a β-Ga2O3 epitaxial layer grown by MOCVD is provided. As shown in Figure 1, the method includes: obtaining preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model; and outputting predicted data of a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model.

[0027] Specifically, the preset growth parameters are process parameters that affect the MOCVD epitaxial growth, and the preset growth parameters are the diffusion coefficient D of the MOCVD epitaxial growth. m and deposition coefficients F n It affects the diffusion coefficient D m is the lateral diffusion rate of atoms on the basis of the deposition coefficient F n is the vertical deposition rate on the atomic base. D m >F n When , the diffusion rate of atoms on the basal surface is greater than the deposition rate of the epitaxial layer, so the atoms attached to the basal surface are not bonded to the crystal lattice. The atoms undergo thermal movement on the basal surface, and the diffusion coefficient D and temperature have a positive correlation, so the epitaxial layer can grow into a 2D shape. D m <F n When the deposition rate is greater than the diffusion rate, the epitaxial layer can grow in 3D islands.

[0028] Diffusion coefficient D for MOCVD epitaxial growth m and the sedimentation coefficient F nThe process parameters that affect the growth of the GaN layer can include the O / Ga ratio, chamber pressure, temperature, growth time, and organic source flow rate. Therefore, in an embodiment of the present invention, the preset growth parameters can include the O / Ga ratio, chamber pressure, temperature, and growth time. In practical applications, the preset growth parameters can further include reactant flow rates. For example, in an embodiment of the present invention, the reactants are TMGa and O2, and the preset growth data includes the flow rate of TMGa. In this case, the O2 flow rate can be determined based on the O / Ga ratio.

[0029] The epitaxial layer prediction model can be a machine learning model, such as a polynomial regression model. The epitaxial layer prediction model is used to predict the epitaxial data of a β-Ga2O3 epitaxial layer. The β-Ga2O3 epitaxial layer is obtained by using TMGa and O2 as reactants and epitaxy on a base without intentional oblique cuts using an MOCVD epitaxial process. That is, when the epitaxial layer prediction model can predict the preset growth parameters, the epitaxial data of the β-Ga2O3 epitaxial layer can be obtained by using TMGa and O2 as reactants and growing β-Ga2O3 using an MOCVD epitaxial process.

[0030] When growing an epitaxial layer, a crystal quality code is required, and it is expected that the crystal's FWHM (Full Width at Half Maximum) will decrease. The predicted data includes the FWHM ratio, which can reflect the quality of the epitaxial layer. The FWHM ratio is used to reflect the difference between the FWHM after epitaxy and the FWHM before epitaxy. In practical applications, the quality of the epitaxial layer can be reflected not only by the FWHM ratio, but also by the thickness and surface roughness of the epitaxial layer. The predicted data can further include the thickness and / or surface roughness of the epitaxial layer. For example, the predicted data can include the thickness of the epitaxial layer, the surface roughness of the epitaxial layer, and the FWHM ratio.

[0031] For example, assuming that the preset growth data are O / Ga ratio = 300, chamber pressure = 100 mbar, temperature = 850°C, growth time = 3600 s, and TMGa = 500 sccm, the preset growth data can be input into the epitaxial layer prediction model, and the predicted data output by the epitaxial layer prediction model can be FWHM ratio = 36 arcsec, thickness = 7200 nm, and RMS = 12.3 nm.

[0032] In an embodiment of the present invention, the training process of the epitaxial layer prediction model is specifically as follows: H10, which acquires the training dataset; inputting the training growth parameters in the training data group into a preset machine learning model, and outputting training prediction data for the β-Ga2O3 epitaxial layer through the preset machine learning model; and H30, which obtains a trained epitaxial layer prediction model by training the pre-configured machine learning model with the training prediction data and the labeled data.

[0033] Specifically, the training data set is pre-constructed and includes multiple training data sets. Each training data set includes training growth data and label data for the β-GaO epitaxial layer. The training growth data includes the O / Ga ratio, chamber pressure, temperature, and growth time, and the label data for the β-GaO epitaxial layer includes at least the FWHM ratio. In this embodiment, the label data for the β-GaO epitaxial layer includes the thickness of the epitaxial layer, the surface roughness of the epitaxial layer, and the FWHM ratio.

[0034] The preset machine learning model and the epitaxial layer prediction model have the same model structure, but differ from each other in that the model parameters of the preset machine learning model are different from those of the epitaxial layer prediction model. The model parameters of the preset machine learning model are initial model parameters, and the model parameters of the epitaxial layer prediction model are model parameters trained using a training dataset. The preset machine learning model may be a deep learning model, a polynomial regression analysis model, or a large-scale model. In an embodiment of the present invention, a polynomial regression analysis model is used as the preset machine learning model. The epitaxial layer prediction model is obtained by training the polynomial regression analysis model using a training dataset.

[0035] In an embodiment of the present invention, the step of training the pre-defined machine learning model using the training prediction data and the label data includes: calculating a predicted difference value between the training predicted data and the labeled data; When the predicted difference value is less than or equal to a preset difference value threshold, determining a loss function term according to the training predicted data and the label data, and training the preset machine learning model according to the loss function term; When the predicted difference value is greater than a preset difference value threshold, a loss function term is determined based on the training predicted data and the label data, and a penalty term is set for the training predicted data, and then the preset machine learning model is trained based on the loss function term and the penalty term.

[0036] Specifically, the training prediction data is the predicted data of the epitaxial layer output by the preset machine learning model, and the label data is the true value data corresponding to the training growth parameters. The label data is obtained by measuring the β-Ga2O3 epitaxial layer, and the epitaxial layer is obtained by MOCVD epitaxy using the training growth parameters. That is, after obtaining the training growth parameters, the β-Ga2O3 epitaxial layer is grown by MOCVD epitaxy to obtain the β-Ga2O3 epitaxial layer. Next, the label data is obtained by measuring the β-Ga2O3 epitaxial layer. The label data of each training data set in the training dataset is the epitaxial data of an epitaxial layer whose extravasation layer quality meets the preset requirements.

[0037] The predicted difference values ​​are the difference values ​​between each epitaxial data term in the label data and a predetermined epitaxial data term in the training prediction data. For example, the predicted difference values ​​may include a difference value of the FWHM ratio, or may be any one or more of a difference value of the FWHM ratio, a difference value of the epitaxial layer thickness, and a difference value of the epitaxial layer surface roughness. When the predicted difference values ​​include multiple difference values, the preset difference value threshold includes multiple preset difference value thresholds. The multiple preset difference value thresholds correspond to the multiple difference values ​​included in the predicted difference values ​​one by one. A predicted difference value greater than a preset difference value threshold means that each difference value includes at least one difference value greater than a predetermined difference value threshold. A predicted difference value less than or equal to a preset difference value threshold means that all of the multiple difference values ​​are less than or equal to a predetermined difference value threshold.

[0038] In an embodiment of the present invention, the predicted difference value includes a difference value of the FWHM ratio, and the preset difference value threshold is a threshold value of the FWHM ratio difference value. The step of calculating the predicted difference value of the training prediction data and the labeled data includes reading the predicted FWHM ratio in the training prediction data and the labeled FWHM ratio in the labeled data, and calculating the difference value between the predicted FWHM ratio and the labeled FWHM ratio in the labeled data to obtain the predicted difference value. Both the predicted FWHM ratio and the labeled FWHM ratio are used to reflect the change value of the FWHM. The change value refers to the difference value between the FWHM after epitaxy and the FWHM before epitaxy.

[0039] When the predicted difference value is greater than a preset difference value threshold, a loss function can be constructed by directly employing the predicted difference value, but this may result in overfitting of the preset machine learning model. Therefore, when constructing a loss function using the predicted difference value, a penalty term can be added to the loss function to prevent overfitting of the preset machine learning model due to the penalty term. A regular function, such as edge regression, least absolute shrinkage and selection operator, or elastic net, can be employed as the penalty term. When the predicted difference value is less than or equal to the preset difference value threshold, a loss function term can be constructed by directly employing the predicted difference value. For example, when the predicted FWHM ratio corresponding to the training growth parameter is -20 arcsec and the labeled FWHM ratio is +200 arcsec, the predicted difference value 220 is greater than the preset difference value threshold, and a penalty term is added to the loss function constructed by the predicted FWHM ratio. The loss function can be a coefficient of determination (R2), MSE, MAE, etc. In the embodiment of the present invention, the coefficient of determination is adopted as the coefficient of determination.

[0040] In an embodiment of the present invention, after the epitaxial layer prediction model outputs the predicted data of the β-Ga2O3 epitaxial layer corresponding to the preset growth parameters, the method further comprises: Detecting whether the predicted data meets preset requirements; When the forecast data does not satisfy the preset requirements, modifying the preset growth parameters to obtain modified growth parameters; performing the steps of setting the modified growth parameters as preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model multiple times until the predicted data meets the preset requirements; The preset growth parameters corresponding to the prediction data that meets the prediction requirements are set as target growth parameters, and the target growth parameters are input into MOCVD to perform β-Ga2O3 epitaxial growth.

[0041] Specifically, the preset requirements are set in advance and can be used to determine whether the preset growth data can be used as the actual growth data. For example, the preset requirement can be to reduce the FWHM ratio to 40 arcsec. If the predicted data meets the preset requirements, it means that the preset growth data can be used as the actual growth data. In this case, the preset growth parameters can be used as the target growth parameters, and the target growth parameters can be input into the MOCVD process to perform β-Ga2O3 epitaxial growth. Conversely, if the predicted data does not meet the preset requirements, the preset growth parameters can be modified to obtain modified growth parameters. When modifying the preset growth parameters, one or more parameter items among the preset growth parameters can be adjusted randomly, or the parameter items that need to be modified and the corresponding modification method can be determined based on preset experience items, and the parameter items can be modified based on the modification method, or the predicted data and preset requirements can be input into the modification network model based on a trained modification network model, and the parameter items that need to be modified and the corresponding modification method can be determined based on the modification network model, and the parameter items that need to be modified based on the modification method, such as increasing or decreasing random data, increasing or decreasing a set value, etc.

[0042] In an embodiment of the present invention, as shown in FIG. 2, the method includes: generating a refinement dataset with target growth parameters and corresponding epitaxial data; and fine-tuning the epitaxial layer prediction model with the fine-tuning data set.

[0043] After obtaining the target growth parameters, the target growth parameters are input into the MOCVD system to perform β-Ga2O3 epitaxy to obtain a β-Ga2O3 epitaxial layer. Next, the β-Ga2O3 epitaxial layer is measured to obtain label data corresponding to the target growth parameters, and the target growth parameters and the corresponding label data are combined into a fine-tuning data set. Finally, the epitaxial layer prediction model is fine-tuned using the fine-tuning data set to further improve the model accuracy of the epitaxial layer prediction model.

[0044] As described above, this embodiment provides a prediction method for a β-Ga2O3 epitaxial layer grown by MOCVD. The method includes the steps of obtaining preset growth parameters, inputting the preset growth parameters into a trained epitaxial layer prediction model, and outputting predicted data for a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model. The preset growth parameters include the O / Ga ratio, chamber pressure, temperature, and growth time. The present invention predicts the epitaxial layer using a machine learning model and selects growth parameters within a process window suitable for β-Ga2O3 epitaxial growth by adjusting the preset growth parameters using the predicted data. The growth parameters are then output to MOCVD for epitaxy. By combining machine learning and MOCVD processes, we can grow high-quality two-dimensional step-flow structures on a substrate without intentional oblique cuts, ensuring the quality of the β-Ga2O3 epitaxial layers grown by MOCVD, using TMGa and O2 as reactants.

[0045] In order to determine the effectiveness of the method for predicting the thickness of a β-Ga2O3 epitaxial layer grown by MOCVD according to the present invention, the performance of the epitaxial layer prediction model was measured. The results showed that the epitaxial layer prediction model was highly effective in overfitting the training data set. As shown in Figures 3 and 4, the coefficient of determination R 2 reached 0.996, and the coefficient of determination of the FWHM ratio, R 2 can reach 0.99. In the test set, as shown in Figures 5 and 6, the coefficient of determination R 2 reached 0.993, and the coefficient of determination R 2 can reach 0.962.

[0046] Furthermore, by expanding the training data set with the fine-tuning data set and continuously fine-tuning the epitaxial layer prediction model, the accuracy of the epitaxial layer prediction model can be further improved. During the prediction process using the epitaxial layer prediction model, target growth parameters were obtained that reduced the FWHM ratio by approximately 40 arcsec. β-Ga2O3 epitaxial growth was achieved using these target growth parameters. Measurement of the resulting epitaxial layer revealed a 50 arcsec reduction in the FWHM of the epitaxial layer, as shown in Figure 7, and the sample surface was smooth and transparent. AFM measurement revealed that the surface roughness of the epitaxial layer within a 1 x 1 μm area was 0.144 nm.

[0047] Based on the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, an embodiment of the present invention further provides an apparatus for predicting a β-Ga2O3 epitaxial layer grown by MOCVD. As shown in Figure 8, the apparatus for predicting a β-Ga2O3 epitaxial layer grown by MOCVD includes: an acquisition module 100 for acquiring preset growth parameters and inputting the preset growth parameters into a trained epitaxial layer prediction model; and a prediction module 200 for outputting predicted data of a β-Ga2O3 epitaxial layer corresponding to the preset growth parameters using the epitaxial layer prediction model, where the preset growth parameters include the O / Ga ratio, chamber pressure, temperature, and growth time, and the predicted data includes at least the FWHM ratio.

[0048] Based on the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, an embodiment of the present invention further provides a computer-readable storage medium having one or more programs stored therein, the one or more programs being executed by one or more processors to implement each step of the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD according to the embodiment of the present invention.

[0049] Based on the method for predicting a β-Ga2O3 epitaxial layer grown by MOCVD, an embodiment of the present invention also provides a terminal device. As shown in Figure 9, the terminal device includes a processor 20, a display panel 21, a memory 22, a communications interface 23, and a bus 24. The processor 20, the display panel 21, the memory 22, and the communications interface 23 can communicate with each other via the bus 24. The display panel 21 is configured to display a user guidance interface preset in an initial setup mode. The communications interface 23 can transmit information. The processor 20 can implement the method of the embodiment by using logical instructions stored in the memory 22.

[0050] When the logic instructions stored in the memory 22 are in the form of a software functional unit, and the software functional unit can be sold or used as an independent product, the logic instructions can be stored on a computer-readable storage medium.

[0051] The memory 22 is a computer-readable storage medium, and may store software programs, computer-executable programs, such as program instructions or program modules corresponding to the methods of the embodiments of the present invention. The processor 20 may execute the software programs, instructions, or modules stored in the memory 22 to perform functions or applications and process data, i.e., to implement the methods of the embodiments.

[0052] The memory 22 includes a program storage area and a data storage area. The program storage area stores an operating system and an application program required for at least one function, and the data storage area stores data generated by the terminal device's usage. The memory 22 includes a high-speed random access memory and may further include a non-volatile memory. For example, the memory 22 may be a medium for storing various program codes, such as a USB flash drive, a mobile hard disk drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or may be a volatile storage medium.

[0053] The specific process of inputting the instructions in the memory and the terminal device into the processor for execution has been described in detail above, so it will not be described again here.

[0054] It should be noted that the above examples are merely illustrative of the technical aspects of the present invention and do not limit the present invention. While the technical aspects of the present invention have been described in detail above using examples of the present invention, the examples are merely illustrative of the present invention and the present invention is not limited to the configurations of the above examples. Those skilled in the art may modify or substitute some of the technical aspects described in the above examples without departing from the spirit of the present invention, and such design modifications and substitutions are of course within the scope of the claims of the present invention.

Claims

1. β-Ga grown by MOCVD ２ O ３ 1. A method for predicting an epitaxial layer, comprising: obtaining preset growth parameters and inputting them into a trained epitaxial layer prediction model; and outputting prediction data corresponding to the preset growth parameters using the epitaxial layer prediction model; the preset growth parameters include O / Ga ratio, chamber pressure, temperature, and growth time, and the predicted data include at least FWHM ratio, thickness, and surface roughness; If the predicted data does not satisfy the preset requirements, the preset growth parameters are modified and re-input, and the process is repeated until the preset requirements are satisfied to determine target growth parameters. Further, the target growth parameters are used to grow β-Ga by MOCVD. ２ O ３ growing an epitaxial layer, generating a fine-tuning data set based on the obtained epitaxial data and target growth parameters, and fine-tuning the epitaxial layer prediction model using the fine-tuning data set. ２ O ３ Methods for predicting epitaxial layers.

2. The training process of the epitaxial layer prediction model is specifically as follows: obtaining a training data set, the training data set including a plurality of training data sets, each of the training data sets being a combination of training growth data and β-Ga; ２ O ３ including label data in the epitaxial layer; The training growth parameters in the training data set are input to a preset machine learning model, and the preset machine learning model is used to generate β-Ga ２ O ３ outputting training prediction data for the epitaxial layer; and obtaining a trained epitaxial layer prediction model by training the pre-defined machine learning model with the training prediction data and the label data. ２ O ３ Methods for predicting epitaxial layers.

3. The process of training the pre-defined machine learning model using the training prediction data and the label data specifically includes: calculating a predicted difference value between the training predicted data and the labeled data; When the predicted difference value is less than or equal to a preset difference value threshold, determining a loss function term according to the training predicted data and the label data, and training the preset machine learning model according to the loss function term; 3. The method of claim 2, further comprising: determining a loss function term based on the training prediction data and the label data when the predicted difference value is greater than a preset difference value threshold; setting a penalty term for the training prediction data; and training the preset machine learning model based on the loss function term and the penalty term. ２ O ３ Methods for predicting epitaxial layers.

4. The training prediction data and the label data both include FWHM ratios, and the step of calculating a predicted difference value between the training prediction data and the label data specifically includes: Reading out predicted FWHM ratios in the training predicted data and labeled FWHM ratios in the labeled data, where the predicted FWHM ratios and labeled FWHM ratios are both used to reflect the difference between the FWHM after epitaxy and the FWHM before epitaxy; 4. The method of claim 3, further comprising: calculating a difference between the predicted FWHM ratio and the labeled FWHM ratio to obtain a predicted difference value. ２ O ３ Methods for predicting epitaxial layers.

5. A computer-readable storage medium having one or more programs stored therein, the one or more programs being executed by one or more processors to produce the MOCVD-grown β-Ga according to any one of claims 1 to 4. ２ O ３ A computer-readable storage medium embodying the steps of a method for predicting epitaxial layers.

6. The method includes a processor and a memory, and the memory stores a computer-readable program that is executed by the processor. The method includes the steps of: ２ O ３ A terminal device characterized in that it carries out the steps of the method for predicting an epitaxial layer.

Images