Semiconductor material film growth method and device based on machine learning

Abstract

The invention discloses a method and a device for growing a semiconductor material film based on machine learning. The method comprises the following steps: acquiring current process parameters; controlling the semiconductor material film to grow under the current process parameters according to the current process parameters; acquiring current film quality data of the semiconductor material film in real time, and predicting future growth quality of the semiconductor material film corresponding to the current process parameter by using a preset machine learning algorithm according to the current film quality data of the semiconductor material film; judging whether the future growth quality of the semiconductor material film is smaller than a preset growth quality threshold value or not, and if yes, ending growth; and if not, adjusting the current process parameters by using a machine learning algorithm, and continuing to execute the step of controlling the semiconductor material film to grow under the current process parameters according to the current process parameters. According to the method, the high controllability of the growth process and the stable improvement of the product quality are ensured, and the accuracy of material performance prediction is promoted.

Description

Technical Field

[0001] The present invention belongs to the field of semiconductor technology, and particularly relates to a method and device for growing a semiconductor material thin film based on machine learning. Background Art

[0002] Mist Chemical Vapor Deposition (Mist-CVD) epitaxial technology is a technology for preparing gallium oxide thin films, which has attracted much attention especially in the field of wide-bandgap semiconductors. By converting a solution containing the elements of the target thin film material into micron-sized droplets through ultrasonic atomization or other means, and using a carrier gas to transport these droplets to the surface of a high-temperature substrate. The droplets produce the Leidenfrost effect on the high-temperature substrate, and the substances in the droplets undergo chemical reactions such as thermal decomposition, oxidation, and reduction, gradually nucleating and islanding on the substrate surface, and finally growing a single-crystal thin film that matches the substrate lattice to complete epitaxial growth. Mist-CVD epitaxial technology can be applied to semiconductor processes.

[0003] In semiconductor processes, the quality of semiconductor thin films, such as the thickness uniformity of the thin films, is one of the key factors affecting the performance and reliability of devices. The non-uniformity of thin film deposition in the process may cause some areas to be too thick or too thin, which may in turn lead to structural defects and performance instability of the devices. Therefore, in the semiconductor manufacturing process, ensuring the quality of the thin film is a necessary condition for improving the performance and reliability of the devices. The inventors found in the process of implementing the present invention that although existing semiconductor processes use various methods to optimize thin film growth to ensure the quality of semiconductor thin films, there are still many problems such as poor uniformity and repeatability, and high process complexity.

[0004] Specifically manifested as:

[0005] 1. The growth rate is relatively low, the transport efficiency of the atomized precursor is limited, and the reaction kinetics is limited by the decomposition rate of the aerosol. 2. The crystallization quality of the thin film is limited. The size distribution of the atomized droplets is non-uniform, resulting in local growth condition fluctuations. The precursor may not decompose completely at high temperatures, introducing impurities or amorphous phases. 3. Doping control is difficult. When precisely doping gallium oxide (such as n-type or p-type doping), it is difficult to control the doping uniformity and concentration. The mixing uniformity, volatility, and thermal decomposition behavior of the dopant in the liquid precursor are difficult to precisely regulate. 4. Challenges in uniformity and repeatability. The distribution of the atomized droplets is significantly affected by factors such as air flow and temperature field, and high process stability is required. 5. High process complexity. It is necessary to finely regulate multiple factors such as atomization parameters (droplet size, carrier gas flow rate), temperature gradient, and reaction chamber pressure, and the process window is relatively narrow. Therefore, in order to solve the above problems of poor quality and repeatability of semiconductor material thin films, and high process complexity, further research and innovation on the growth optimization technology of semiconductor material thin films are still needed. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a method and device for growing a semiconductor material thin film based on machine learning, so as to solve the problems of poor quality and repeatability of semiconductor material thin films and high process complexity in the prior art. The specific technical solutions are as follows:

[0007] To achieve the above object, in a first aspect, the embodiments of the present invention provide a method for growing a semiconductor material thin film based on machine learning, the method comprising:

[0008] Obtain current process parameters; the current process parameters include: precursor concentration, atomization pressure, substrate temperature, and carrier gas flow rate;

[0009] Control the growth of the semiconductor material thin film under the current process parameters according to the current process parameters;

[0010] Real-time collect the current thin film quality data of the semiconductor material thin film, and according to the current thin film quality data of the semiconductor material thin film, use a preset machine learning algorithm to predict the future growth quality of the semiconductor material thin film corresponding to the current process parameters; the thin film quality data includes: thin film thickness, uniformity, crystal structure, and roughness;

[0011] Judge whether the future growth quality of the semiconductor material thin film is less than a preset growth quality threshold. If so, end the growth; if not, use the machine learning algorithm to adjust the current process parameters, and continue to execute the step of controlling the growth of the semiconductor material thin film under the current process parameters according to the current process parameters.

[0012] Optionally, before the step of real-time collecting the thin film quality data of the semiconductor material thin film and using a preset machine learning algorithm to predict the future growth quality of the semiconductor material thin film corresponding to the current process parameters, it further includes:

[0013] Introduce a preset three-dimensional atomization array into the semiconductor thin film growth technology, and use a multi-objective optimization model in the machine learning algorithm to construct a relationship model between the three-dimensional atomization array and the thin film quality of the semiconductor material thin film, and obtain an initial correlation model between process parameters and thin film quality;

[0014] Train the initial correlation model using a sample data set to obtain a target multi-objective optimization model, wherein the multi-objective optimization model is used to receive process parameters and calculate and output the future growth quality of the semiconductor material thin film corresponding to the received process parameters.

[0015] Optionally, the acquisition method of the sample data set is:

[0016] Obtain the historical process parameter data of the semiconductor material thin film in the semiconductor thin film growth equipment within a preset time period, and the thin film quality data corresponding to each historical process parameter data; wherein, a preset three-dimensional atomization array is provided in the semiconductor thin film growth equipment.

[0017] Perform data cleaning, noise filtering, and outlier removal on the historical process parameter data and the thin film quality data in sequence to obtain target historical process parameter data and target thin film quality data.

[0018] Label each historical process parameter with a thin film quality value according to the thin film quality corresponding to each historical process parameter in the historical process parameter data to obtain the sample data set.

[0019] Optionally, the step of training the initial correlation model using the sample data set to obtain a target multi-objective optimization model includes:

[0020] Input the sample data set with thin film quality value labels into the initial correlation model, and optimize the parameter optimization process of the initial correlation model using a preset high-throughput calculation and materials genomics algorithm during the training process.

[0021] Obtain the future growth quality of the semiconductor material thin film corresponding to each sample data set output by the initial correlation model.

[0022] Calculate a loss value using a preset loss function according to the output result of the initial correlation model and the label of the sample data set.

[0023] Adjust the parameters in the initial correlation model according to the loss value to obtain a target multi-objective optimization model, where the multi-objective optimization model is used to output the future growth quality of the semiconductor material thin film and the corresponding process parameters.

[0024] Optionally, the coefficient of variation value of the uniformity of the precursor concentration distribution in the reaction chamber of the preset three-dimensional atomization array is less than 3%.

[0025] Optionally, the step of collecting the current thin film quality data of the semiconductor material thin film in real time includes:

[0026] Obtain the current thin film quality data of the semiconductor material thin film collected in real time by the sensor device on the semiconductor thin film growth equipment.

[0027] Optionally, the semiconductor material thin film is a β-phase gallium oxide thin film.

[0028] In a second aspect, an embodiment of the present invention provides a growth device for a semiconductor material thin film based on machine learning. The device includes:

[0029] An acquisition module for acquiring current process parameters; the current process parameters include: precursor concentration, atomization pressure, substrate temperature, and carrier gas flow rate;

[0030] A control module for controlling the growth of a semiconductor material thin film under the current process parameters according to the current process parameters;

[0031] A prediction module for real-time collecting current thin film quality data of the semiconductor material thin film, and predicting the future growth quality of the semiconductor material thin film corresponding to the current process parameters according to the current thin film quality data of the semiconductor material thin film by using a preset machine learning algorithm; the thin film quality data includes: thin film thickness, uniformity, crystal structure, and roughness;

[0032] A judgment module for judging whether the future growth quality of the semiconductor material thin film is less than a preset growth quality threshold. If so, end the growth; if not, adjust the current process parameters by using the machine learning algorithm, and continue to execute the step of controlling the growth of the semiconductor material thin film under the current process parameters according to the current process parameters.

[0033] Optionally, the device further includes a model training module for:

[0034] Introducing a preset three-dimensional stereoscopic atomization array into the semiconductor thin film growth technology, and using a multi-objective optimization model in the machine learning algorithm to construct a relationship model between the three-dimensional stereoscopic atomization array and the thin film quality of the semiconductor material thin film, obtaining an initial correlation model between the process parameters and the thin film quality;

[0035] Training the initial correlation model by using a sample data set to obtain a target multi-objective optimization model, wherein the multi-objective optimization model is used to receive process parameters and calculate and output the future growth quality of the semiconductor material thin film corresponding to the received process parameters.

[0036] Optionally, the device further includes a sample data set acquisition module for:

[0037] Acquiring historical process parameter data of the semiconductor material thin film in the semiconductor thin film growth device and thin film quality data corresponding to each historical process parameter data within a preset time period; wherein, the preset three-dimensional stereoscopic atomization array is provided in the semiconductor thin film growth device;

[0038] Successively performing data cleaning, noise filtering, and outlier removal on the historical process parameter data and the thin film quality data to obtain target historical process parameter data and target thin film quality data;

[0039] Based on the film quality corresponding to each historical process parameter in the historical process parameter data, label each historical process parameter with a film quality value to obtain the sample data set.

[0040] Optionally, the model training module is specifically configured to:

[0041] Input the sample data set with film quality value labels into the initial correlation model, and optimize the parameter optimization process of the initial correlation model by using a preset high-throughput calculation and materials genome algorithm during the training process;

[0042] Obtain the future growth quality of the semiconductor material film corresponding to each sample data set output by the initial correlation model;

[0043] According to the output result of the initial correlation model and the label of the sample data set, calculate the loss value by using a preset loss function;

[0044] Adjust the parameters in the initial correlation model according to the loss value to obtain a target multi-objective optimization model, where the multi-objective optimization model is used to output the future growth quality of the semiconductor material film and the corresponding process parameters.

[0045] Optionally, the coefficient of variation value of the precursor concentration distribution uniformity in the reaction chamber of the preset three-dimensional atomization array is less than 3%.

[0046] Optionally, the prediction module is specifically configured to:

[0047] Obtain the current film quality data of the semiconductor material film collected in real time by the sensor device on the semiconductor film growth equipment.

[0048] Optionally, the semiconductor material film is a β-phase gallium oxide film.

[0049] In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete mutual communication through the communication bus;

[0050] The memory is used to store a computer program;

[0051] The processor is configured to implement the method steps of the method for growing a semiconductor material film based on machine learning described in the first aspect when executing the program stored on the memory.

[0052] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and the computer program implements the method steps of the method for growing a semiconductor material film based on machine learning described in the first aspect when executed by a processor.

[0053] In a fifth aspect, an embodiment of the present invention further provides a computer program product containing instructions. When it runs on a computer, it causes the computer to execute the method steps of growing a semiconductor material thin film based on machine learning described in the first aspect above.

[0054] It can be seen that the method provided by the embodiment of the present invention, by introducing a machine learning algorithm and utilizing its unique high-efficiency computing power, accurate prediction accuracy, powerful self-learning mechanism, and high operating stability, can not only solve the problems existing in the existing thin film growth methods, such as poor quality and repeatability of thin films, high process complexity, etc., but also significantly improve the efficiency of thin film growth, shorten the experimental cycle, ensure the high controllability of the growth process and the stable improvement of product quality, promote the accuracy of material property prediction, optimize process parameters, enhance data utilization, drive innovation in material design, effectively reduce production costs, and greatly improve the consistency of production batches. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art.

[0056] Figure 1 It is a schematic flowchart of a method for growing a semiconductor material thin film based on machine learning provided by an embodiment of the present invention;

[0057] Figure 2 It is a schematic structural diagram of a honeycomb multi-nozzle atomizer provided by an embodiment of the present invention;

[0058] Figure 3 It is a schematic structural diagram of a device for growing a semiconductor material thin film based on machine learning provided by an embodiment of the present invention;

[0059] Figure 4 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The following will describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. It should be understood that the preferred embodiments described herein are only for illustrating and explaining the present invention, and are not used to limit the present invention.

[0061] Figure 1 It is a schematic flowchart of a method for growing a semiconductor material thin film based on machine learning provided by an embodiment of the present invention. As Figure 1 shown, the method includes:

[0062] S101: Obtain current process parameters;

[0063] The current process parameters include: precursor concentration, atomization pressure, substrate temperature, and carrier gas flow rate. Of course, in practical applications, users can also adjust the process parameters according to actual needs. For example, the process parameters can be adjusted to: precursor concentration, atomization pressure, substrate temperature, carrier gas flow rate, deposition rate, atomization intensity, atomized droplet size distribution, etc.

[0064] In practical applications, the current process parameters can be the process parameters simulated and calculated by a machine learning algorithm as the current process parameters.

[0065] S102: Control the growth of the semiconductor material thin film under the current process parameters according to the current process parameters;

[0066] In a specific embodiment, the semiconductor material thin film can be a β-phase gallium oxide thin film. The semiconductor material can be silicon carbide, gallium nitride, gallium arsenide, etc.

[0067] S103: Real-time collect the current thin film quality data of the semiconductor material thin film, and according to the current thin film quality data of the semiconductor material thin film, use a preset machine learning algorithm to predict the future growth quality of the semiconductor material thin film corresponding to the current process parameters;

[0068] The thin film quality data includes: thin film thickness, uniformity, crystal structure, and roughness. Of course, in practical applications, users can also select other data as the thin film quality data according to actual needs. For example, select thin film uniformity, defect density, growth rate, and phase purity, etc. as the thin film quality data.

[0069] In a specific embodiment, one or more sensors can be deployed on the semiconductor thin film growth equipment to collect the thin film quality data in real time, that is, to monitor the growth process of the thin film in real time and provide real-time input data for the multi-objective optimization model.

[0070] In practical applications, before the step of real-time collecting the thin film quality data of the semiconductor material thin film and using a preset machine learning algorithm to predict the future growth quality of the semiconductor material thin film corresponding to the current process parameters, it can also include:

[0071] Introduce a preset three-dimensional stereoscopic atomization array into the semiconductor thin film growth technology, and use the multi-objective optimization model in the machine learning algorithm to construct a relationship model between the three-dimensional stereoscopic atomization array and the thin film quality of the semiconductor material thin film to obtain an initial correlation model between the process parameters and the thin film quality;

[0072] Use the sample data set to train the initial correlation model to obtain a target multi-objective optimization model, where the multi-objective optimization model is used to receive the process parameters and calculate and output the future growth quality of the semiconductor material thin film corresponding to the received process parameters.

[0073] In the method provided by the embodiments of the present invention, a three-dimensional atomization array is introduced into the semiconductor thin film growth technology. In a specific embodiment, the three-dimensional atomization array can be implemented by a honeycomb multi-nozzle atomizer.

[0074] Figure 2 It is a cross-sectional view of the reaction chamber of the three-dimensional atomization array. The mist nozzles are arranged in a hexagonal pattern, with a total of three circles. The red one is the first circle, with only one; the blue ones are the nozzles in the second circle, with 6; the green ones are the third circle, with 12. The entire design has a total of 19 nozzles. In practical applications, a honeycomb multi-nozzle atomizer can be used, such as Figure 2 The 19-channel annular array shown, in cooperation with a vortex gas flow generator, can achieve a coefficient of variation (CV value) of the precursor concentration distribution in the reaction chamber < 3%, so as to obtain a uniform and stable semiconductor material thin film. For example, a uniform and stable β-phase gallium oxide thin film of semiconductor material can be obtained.

[0075] The semiconductor material in a specific embodiment can be other materials such as silicon carbide, gallium nitride, and gallium arsenide. The method provided by the embodiments of the present invention can be applied to doping, single crystal control, and thin film thickness control, etc. The semiconductor thin film growth technology adopted by the embodiments of the invention can be Mist-CVD epitaxial technology, atomic layer deposition (ALD) technology, metal organic chemical vapor deposition (MOCVD) technology, molecular beam epitaxy (MBE) technology, and physical vapor deposition (PVD) technology, etc.

[0076] In a specific embodiment, the acquisition method of the sample data set can be:

[0077] Acquire the historical process parameter data of the semiconductor material thin film in the semiconductor thin film growth equipment within a preset time period, and the thin film quality data corresponding to each historical process parameter data; wherein, a preset three-dimensional atomization array is set in the semiconductor thin film growth equipment;

[0078] Perform data cleaning, noise filtering, and outlier removal on the historical process parameter data and the thin film quality data in sequence to obtain the target historical process parameter data and the target thin film quality data;

[0079] Label each historical process parameter with a thin film quality value according to the thin film quality corresponding to each historical process parameter in the historical process parameter data to obtain the sample data set.

[0080] The steps of training an initial correlation model using the sample data set to obtain a target multi-objective optimization model may include:

[0081] Input the sample data set with thin film quality value labels into the initial correlation model, and optimize the parameter optimization process of the initial correlation model by using the preset high-throughput calculation and materials genomics algorithm during the training process;

[0082] Obtain the future growth quality of the semiconductor material thin film corresponding to each sample data set output by the initial correlation model;

[0083] Calculate the loss value using the preset loss function according to the output result of the initial correlation model and the labels of the sample data set;

[0084] Adjust the parameters in the initial correlation model according to the loss value to obtain the target multi-objective optimization model, where the multi-objective optimization model is used to output the future growth quality of the semiconductor material thin film and the corresponding process parameters.

[0085] In practical applications, optimization algorithms in the materials genomics algorithm, such as genetic algorithm, particle swarm algorithm, etc., can be used to find the optimal process parameter combination within the set thin film quality threshold range and step size. These algorithms can search for the optimal solution within the given parameter range, so as to obtain the best thin film growth conditions.

[0086] In practical applications, the semiconductor material thin film growth experiment can collect the historical data of the semiconductor material thin film growth generated by using the honeycomb multi-nozzle atomizer and the vortex gas generator. The historical data can include the chamber reaction temperature (such as: substrate temperature, atomizer temperature, etc.), pressure (such as: reaction chamber pressure, carrier gas pressure, etc.), gas flow rate (such as: oxygen, precursor solution atomization gas flow rate, etc.), deposition rate, atomization intensity, atomized droplet size distribution, precursor concentration, atomization frequency, fog gas flow rate, etc. Obtain the experimental data generated after the semiconductor material thin film growth experiment; the experimental data includes thickness (such as: ellipsometer, step profiler, etc.), uniformity (such as: standard deviation of thickness distribution, through multi-point measurement, etc.), crystal structure (such as: full width at half maximum of X-ray (X-Ray Diffraction, XRD) diffraction peak, grain size, etc.), surface morphology (such as: roughness of atomic force microscope (Atomic Force Microscopy, AFM), porosity of scanning electron microscope (Scanning Electron Microscope, SEM), etc.); electrical properties (such as: carrier concentration, mobility, etc.).

[0087] After obtaining the above historical data and experimental data, it is necessary to preprocess the historical data and experimental data respectively to obtain the preprocessed historical data and experimental data. For example, in a specific embodiment, the preprocessing process may be: first, clean the historical data and experimental data respectively to remove noise and outliers in the data; then perform noise filtering on the historical data and experimental data, and specifically, sliding window mean filtering or wavelet transform can be used to remove noise; then remove outliers from the historical data and experimental data; in a specific embodiment, the method for removing outliers may be: identify and remove outliers in the data based on the 3σ principle or the Isolation Forest algorithm; finally, perform data standardization: perform normalization processing on parameters with different dimensions such as temperature and pressure, such as Z-score normalization or Min-Max scaling processing, to obtain the preprocessed historical data and experimental data.

[0088] In a specific embodiment, the film thickness, uniformity, crystal structure, and roughness can be selected from the experimental data as characteristic parameters. Of course, in other embodiments, other parameters can also be selected as characteristic parameters, such as the deposition rate and atomization efficiency. Then, according to the correlation degree between the experimental optimization parameters and each characteristic parameter, different weights are configured for each characteristic parameter, and each characteristic parameter is weighted and sorted according to the configured weights. After that, other useful features can also be extracted from the historical data and experimental data, such as statistical features and frequency domain features. Among them, statistical features are indicators directly extracted from the historical data and experimental data (time domain or spatial domain) through mathematical calculations, and are used to describe the statistical characteristics such as the distribution, central tendency, and dispersion degree of the data. Frequency domain features are features extracted after converting the time domain or spatial domain signal into the frequency domain, and reflect the frequency component characteristics of the signal, such as combinations or transformation forms of process parameters such as temperature, pressure, and gas flow. In this way, a richer feature set can be constructed to improve the prediction ability of the model.

[0089] In a pre-established automated experimental platform, in a specific embodiment, the platform can be a Mist-CVD epitaxial technology growth platform for β-phase gallium oxide thin films designed using a three-dimensional stereoscopic atomization array. Among them, multiple machine learning models are pre-constructed in the automated experimental platform, such as random forest models, support vector machine models, neural network models, etc. After the platform obtains the characteristic parameters, it will select a matching model from the pre-constructed multiple machine learning models to train according to the relationship between the characteristic parameters. For example, according to the characteristic parameters of film thickness, uniformity, crystal structure, and roughness, a multi-task neural network (MTL) or Gaussian process regression (GPR) model is selected to predict film thickness, uniformity, crystal structure, and roughness. After selecting a suitable machine learning model, a non-linear relationship model between the input parameters (such as precursor concentration, atomization pressure, substrate temperature, carrier gas flow rate, etc.) and the output targets (such as film thickness, uniformity, crystal structure, and roughness, etc.) is established. In a specific embodiment, this is also the initial correlation model between process parameters and film quality. Then, the initial correlation model is trained using the sample data set, and the training process can be as follows:

[0090] First, perform data partitioning: In a specific embodiment, the training set and the validation set can be partitioned in chronological order, which is beneficial to avoiding future data leakage caused by random partitioning; for example, in a specific embodiment, for the sample data obtained during the period from January 1, 2024 to December 31, 2024, the sample historical data obtained during the period from January 1, 2024 to August 31, 2024 can be used as the training set, and the sample data obtained during the period from September 1, 2024 to December 31, 2024 can be used as the validation set. The initial correlation model is trained using the training set, and the training results are verified using the validation set. During the training process, hyperparameter optimization can also be performed, such as using Bayesian optimization to search for parameters such as the number of network layers and learning rate, and regularization can be performed, for example, adding L2 regularization or Dropout layers to prevent overfitting. Finally, the performance of the model can also be evaluated through methods such as cross-validation. Among them, cross-validation is a commonly used model evaluation method that can evaluate the performance of the model on different data sets, thereby ensuring the generalization ability of the model.

[0091] In the automated experimental platform provided by the embodiments of the present invention, devices such as sensors, controllers, and actuators are integrated, so that closed-loop control of the thin film growth process can be realized. That is, the experimental conditions can be automatically adjusted according to the real-time process parameters and the model prediction results, so as to achieve precise control of the thin film quality. At the same time, the closed-loop control can also improve the stability and repeatability of the experiment. For example, using the automated experimental platform provided by the embodiments of the present invention, the experimental parameters of the β-phase gallium oxide thin film of semiconductor material can be obtained, and the experimental parameters are input into the multi-objective optimization model of the trained machine learning algorithm for future growth quality prediction and parameter adjustment according to the prediction results of the model, so as to realize the closed-loop cycle feedback from data acquisition, model prediction to parameter adjustment.

[0092] On the automated experimental platform provided by the embodiments of the present invention, a machine learning multi-objective optimization plus intelligent system model can be established, which can be used to find the optimal value of the atomization intensity. A high-throughput calculation plus intelligent system model can also be established for phase stability control. For example, high-throughput calculation is used to predict the thermodynamic stability of the β-phase under a specific oxygen partial pressure, and the intelligent system dynamically adjusts the O2 / N2 ratio to inhibit the generation of the α-phase. A materials genomics + intelligent decision-making system model can also be established, which can be used to improve the doping uniformity of semiconductor materials. Materials genomics can be used to guide the design of dopant Si, and the intelligent decision-making system can be used to control the doping source injection concentration and timing to achieve uniformity in the carrier concentration space. A high-throughput calculation + materials genomics + intelligent decision-making system model can also be established, which can be used to quickly screen the influence of gallium oxide doping elements on the energy band structure and realize co-doped p-type gallium oxide thin films. For example, high-throughput calculation selects appropriate doping elements, materials genomics guides the design of co-dopants such as N+Si, Mg+Si, Li+Zn, etc., and the intelligent decision-making system controls the injection timing and concentration of the co-doping source to achieve the best element ratio for p-doping of gallium oxide. Therefore, the method provided by the embodiments of the present invention can be applied to the optimized semiconductor material thin film growth technology, and the industrial parameters can also be doping, single crystal control, thin film thickness control, etc.

[0093] Optionally, the step of collecting the current thin film quality data of the semiconductor material thin film in real time includes:

[0094] Obtain the current thin film quality data of the semiconductor material thin film collected in real time by the sensor device on the semiconductor thin film growth equipment.

[0095] In practical applications, sensors can be deployed on the thin film growth equipment to collect thin film quality data in real time. These sensors can monitor the thin film growth process in real time and provide real-time input data for the model.

[0096] S104: Determine whether the future growth quality of the semiconductor material thin film is less than a preset growth quality threshold. If so, execute step S105; if not, execute step S106;

[0097] S105: End the growth;

[0098] S106: Use a machine learning algorithm to adjust the current process parameters and return to continue executing step S102.

[0099] In summary, the method provided by the embodiment of the present invention can create a multi-objective optimization model during the thin film growth process, and use the created multi-objective optimization model to predict the future growth quality of the semiconductor material thin film according to the real-time process parameters, so as to realize the real-time monitoring of the semiconductor material thin film growth process. It can also adjust the process parameters in real time according to the prediction results of the model through a feedback control mechanism, such as the atomization intensity, etc., so as to realize the precise control of the future growth quality of the semiconductor material thin film.

[0100] Corresponding to the Figure 1 illustrated embodiment, the embodiment of the present invention also provides a growth device for a semiconductor material thin film based on machine learning. As Figure 3 illustrated, the device includes: an acquisition module 301, a control module 302, a prediction module 303, and a judgment module 304. Among them,

[0101] The acquisition module 301 is used to acquire the current process parameters; the current process parameters include: precursor concentration, atomization pressure, substrate temperature, and carrier gas flow rate;

[0102] The control module 302 is used to control the growth of the semiconductor material thin film under the current process parameters according to the current process parameters;

[0103] The prediction module 303 is used to collect the current thin film quality data of the semiconductor material thin film in real time, and predict the future growth quality of the semiconductor material thin film corresponding to the current process parameters according to the current thin film quality data of the semiconductor material thin film; the thin film quality data includes: thin film thickness, uniformity, crystal structure, and roughness;

[0104] The judgment module 304 is used to judge whether the future growth quality of the semiconductor material thin film is less than a preset growth quality threshold. If so, end the growth; if not, use a machine learning algorithm to adjust the current process parameters, and continue to execute the step of controlling the growth of the semiconductor material thin film under the current process parameters.

[0105] Optionally, the device further includes a model training module, which is used for:

[0106] Introduce a preset three-dimensional atomization array into the semiconductor thin film growth technology, and use the multi-objective optimization model in the machine learning algorithm to build a relationship model between the three-dimensional atomization array and the thin film quality of the semiconductor material thin film, so as to obtain an initial correlation model between the process parameters and the thin film quality;

[0107] Use the sample data set to train the initial correlation model to obtain a target multi-objective optimization model. Among them, the multi-objective optimization model is used to receive process parameters and calculate and output the future growth quality of the semiconductor material thin film corresponding to the received process parameters.

[0108] Optionally, the device further includes a sample data set acquisition module for:

[0109] Obtain the historical process parameter data of the semiconductor material thin film in the semiconductor thin film growth equipment and the thin film quality data corresponding to each historical process parameter data within a preset time period; among them, a preset three-dimensional atomization array is set in the semiconductor thin film growth equipment;

[0110] Perform data cleaning, noise filtering and outlier removal on the historical process parameter data and the thin film quality data in sequence to obtain target historical process parameter data and target thin film quality data;

[0111] According to the thin film quality corresponding to each historical process parameter in the historical process parameter data, label each historical process parameter with a thin film quality value label to obtain a sample data set.

[0112] Optionally, the model training module is specifically used for:

[0113] Input the sample data set with the thin film quality value label into the initial correlation model, and optimize the parameter optimization process of the initial correlation model by using the preset high-throughput calculation and materials genome algorithm during the training process;

[0114] Obtain the future growth quality of the semiconductor material thin film corresponding to each sample data set output by the initial correlation model;

[0115] Calculate the loss value according to the output result of the initial correlation model and the label of the sample data set by using the preset loss function;

[0116] Adjust the parameters in the initial correlation model according to the loss value to obtain a target multi-objective optimization model. Among them, the multi-objective optimization model is used to output the future growth quality of the semiconductor material thin film and the corresponding process parameters.

[0117] Optionally, the coefficient of variation value of the uniformity of the precursor concentration distribution in the reaction chamber of the preset three-dimensional atomization array is less than 3%.

[0118] Optionally, the prediction module 303 is specifically configured to:

[0119] Obtain the current film quality data of the semiconductor material film collected in real time by the sensor device on the semiconductor thin film growth equipment.

[0120] Optionally, the semiconductor material film is a β-phase gallium oxide film.

[0121] In summary, the device provided by the embodiment of the present invention can create a multi-objective optimization model during the film growth process, and use the created multi-objective optimization model to predict the future growth quality of the semiconductor material film according to the real-time process parameters, so as to realize the real-time monitoring of the semiconductor material film growth process. It can also adjust the process parameters in real time according to the prediction results of the model through a feedback control mechanism, such as the atomization intensity, etc., so as to realize the precise control of the future growth quality of the semiconductor material film.

[0122] corresponding to Figure 1 the embodiment shown, the embodiment of the present invention also provides an electronic device, see Figure 4 , including a processor 401, a communication interface 402, a memory 403, and a communication bus 404. Among them, the processor 401, the communication interface 402, and the memory 403 complete communication with each other through the communication bus 404;

[0123] The memory 403 is used to store computer programs;

[0124] When the processor 401 is used to execute the program stored on the memory, it realizes any one of the method steps for the growth of the semiconductor material film based on machine learning in the above embodiment.

[0125] In summary, the electronic device provided by the embodiment of the present invention can create a multi-objective optimization model during the film growth process, and use the created multi-objective optimization model to predict the future growth quality of the semiconductor material film according to the real-time process parameters, so as to realize the real-time monitoring of the semiconductor material film growth process. It can also adjust the process parameters in real time according to the prediction results of the model through a feedback control mechanism, such as the atomization intensity, etc., so as to realize the precise control of the future growth quality of the semiconductor material film.

[0126] corresponding to Figure 1 the embodiment shown, the embodiment of the present invention also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it realizes any one of the method steps for the growth of the semiconductor material film based on machine learning in the above embodiment.

[0127] In summary, during the thin film growth process, the storage medium provided by the embodiments of the present invention can create a multi-objective optimization model, and use the created multi-objective optimization model to predict the future growth quality of the semiconductor material thin film according to real-time process parameters, so as to realize the real-time monitoring of the semiconductor material thin film growth process. It can also adjust the process parameters in real time according to the prediction results of the model through a feedback control mechanism, such as the atomization intensity, etc., so as to realize the precise control of the future growth quality of the semiconductor material thin film.

[0128] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0129] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a machine for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 a device for the functions specified in one or more blocks.

[0130] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one or more flows and / or blocks Figure 1 a device for the functions specified in one or more blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 a device for the functions specified in one or more blocks.

[0132] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A method for growing a semiconductor material thin film based on machine learning, characterized in that: The method comprises: Acquiring current process parameters; the current process parameters include: precursor concentration, atomization pressure, substrate temperature and carrier gas flow rate; Controlling the growth of the semiconductor material film under the current process parameters according to the current process parameters; Collecting the current film quality data of the semiconductor material film in real time, and predicting the future growth quality of the semiconductor material film corresponding to the current process parameters using a preset machine learning algorithm based on the current film quality data of the semiconductor material film; the film quality data includes: film thickness, uniformity, crystal structure and roughness; Determine whether the future growth quality of the semiconductor material film is less than a preset growth quality threshold, and if so, terminate the growth; if not, use the machine learning algorithm to adjust the current process parameters, and continue to execute the step of controlling the growth of the semiconductor material film under the current process parameters according to the current process parameters.

2. The method according to claim 1, characterized in that: Before the step of collecting the film quality data of the semiconductor material film in real time and using a preset machine learning algorithm to predict the future growth quality of the semiconductor material film corresponding to the current process parameters, the method further includes: A preset three-dimensional atomization array is introduced into the semiconductor thin film growth technology, and a multi-objective optimization model in a machine learning algorithm is used to construct a relationship model between the three-dimensional atomization array and the film quality of the semiconductor material film, thereby obtaining an initial correlation model between the process parameters and the film quality; The initial correlation model is trained using a sample data set to obtain a target multi-objective optimization model, wherein the multi-objective optimization model is used to receive process parameters, and calculate and output the future growth quality of the semiconductor material film corresponding to the process parameters based on the received process parameters.

3. The method according to claim 2, characterized in that The sample data set is obtained as follows: Acquire historical process parameter data of a semiconductor material film in a semiconductor thin film growth device within a preset time period, and film quality data corresponding to each historical process parameter data; wherein the semiconductor thin film growth device is provided with the preset three-dimensional atomization array; The historical process parameter data and the film quality data are sequentially cleaned, noise filtered, and outlier eliminated to obtain target historical process parameter data and target film quality data; According to the film quality corresponding to each historical process parameter in the historical process parameter data, a film quality value label is added to each historical process parameter to obtain the sample data set.

4. The method according to claim 3, characterized in that The step of training the initial correlation model using the sample data set to obtain the target multi-objective optimization model includes: Inputting a sample data set with film quality value labels into the initial correlation model, and optimizing the parameter optimization process of the initial correlation model using a preset high-throughput computing and material genomics algorithm during the training process; Obtaining the future growth quality of the semiconductor material film corresponding to each sample data set output by the initial correlation model; Calculating a loss value using a preset loss function according to the output result of the initial correlation model and the label of the sample data set; According to the loss value, the parameters in the initial correlation model are adjusted to obtain a target multi-objective optimization model, wherein the multi-objective optimization model is used to output the future growth quality of the semiconductor material film and the corresponding process parameters.

5. The method according to claim 2, characterized in that: The coefficient of variation of the uniformity of the precursor concentration distribution in the reaction chamber of the preset three-dimensional atomization array is less than 3%.

6. The method according to claim 1, characterized in that The step of collecting the current film quality data of the semiconductor material film in real time comprises: Acquire the current film quality data of the semiconductor material film collected in real time by the sensor equipment on the semiconductor film growth equipment.

7. The method according to claim 1, characterized in that The semiconductor material film is a β-phase gallium oxide film.

8. A semiconductor material thin film growth device based on machine learning, characterized in that: The device comprises: An acquisition module is used to acquire current process parameters; the current process parameters include: precursor concentration, atomization pressure, substrate temperature and carrier gas flow rate; A control module, used for controlling the growth of a semiconductor material film under the current process parameters according to the current process parameters; A prediction module, for collecting current film quality data of the semiconductor material film in real time, and using a preset machine learning algorithm to predict the future growth quality of the semiconductor material film corresponding to the current process parameters based on the current film quality data of the semiconductor material film; the film quality data includes: film thickness, uniformity, crystal structure and roughness; A judgment module is used to judge whether the future growth quality of the semiconductor material film is less than a preset growth quality threshold. If so, the growth is terminated; if not, the current process parameters are adjusted using the machine learning algorithm, and the step of controlling the growth of the semiconductor material film under the current process parameters is continued.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; The memory is used to store computer programs; The processor is used to implement the method steps described in any one of claims 1-7 when executing the program stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 7 are implemented.

Images