CN112836853A - Gallium oxide preparation method and system based on deep learning and heat exchange method - Google Patents

Gallium oxide preparation method and system based on deep learning and heat exchange method Download PDF

Info

Publication number
CN112836853A
CN112836853A CN202011638983.3A CN202011638983A CN112836853A CN 112836853 A CN112836853 A CN 112836853A CN 202011638983 A CN202011638983 A CN 202011638983A CN 112836853 A CN112836853 A CN 112836853A
Authority
CN
China
Prior art keywords
data
gallium oxide
preparation
neural network
network model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011638983.3A
Other languages
Chinese (zh)
Inventor
齐红基
王晓亮
陈端阳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Fujia Gallium Technology Co Ltd
Original Assignee
Hangzhou Fujia Gallium Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Fujia Gallium Technology Co Ltd filed Critical Hangzhou Fujia Gallium Technology Co Ltd
Priority to CN202011638983.3A priority Critical patent/CN112836853A/en
Priority to PCT/CN2021/075545 priority patent/WO2022141750A1/en
Publication of CN112836853A publication Critical patent/CN112836853A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • C—CHEMISTRY; METALLURGY
    • C30—CRYSTAL GROWTH
    • C30B—SINGLE-CRYSTAL GROWTH; UNIDIRECTIONAL SOLIDIFICATION OF EUTECTIC MATERIAL OR UNIDIRECTIONAL DEMIXING OF EUTECTOID MATERIAL; REFINING BY ZONE-MELTING OF MATERIAL; PRODUCTION OF A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; SINGLE CRYSTALS OR HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; AFTER-TREATMENT OF SINGLE CRYSTALS OR A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; APPARATUS THEREFOR
    • C30B11/00—Single-crystal growth by normal freezing or freezing under temperature gradient, e.g. Bridgman-Stockbarger method
    • C—CHEMISTRY; METALLURGY
    • C30—CRYSTAL GROWTH
    • C30B—SINGLE-CRYSTAL GROWTH; UNIDIRECTIONAL SOLIDIFICATION OF EUTECTIC MATERIAL OR UNIDIRECTIONAL DEMIXING OF EUTECTOID MATERIAL; REFINING BY ZONE-MELTING OF MATERIAL; PRODUCTION OF A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; SINGLE CRYSTALS OR HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; AFTER-TREATMENT OF SINGLE CRYSTALS OR A HOMOGENEOUS POLYCRYSTALLINE MATERIAL WITH DEFINED STRUCTURE; APPARATUS THEREFOR
    • C30B29/00—Single crystals or homogeneous polycrystalline material with defined structure characterised by the material or by their shape
    • C30B29/10—Inorganic compounds or compositions
    • C30B29/16—Oxides
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The application relates to a gallium oxide preparation method and system based on deep learning and heat exchange methods, wherein the prediction method comprises the following steps: acquiring preparation data of the gallium oxide single crystal; the preparation data includes: seed data, environmental data, and control data; the control data includes: seed crystal cooling medium flow; preprocessing the preparation data to obtain preprocessed preparation data; inputting the preprocessed preparation data into the trained neural network model to obtain the corresponding prediction property data of the gallium oxide single crystal. Preprocessing preparation data to obtain preprocessed preparation data, inputting the preprocessed preparation data into a trained neural network model to obtain predicted property data corresponding to the gallium oxide single crystal, predicting the performance of the gallium oxide single crystal through the trained neural network model, and adjusting the preparation data to obtain the required performance of the gallium oxide single crystal, so that the performance of the gallium oxide single crystal is optimized.

Description

Gallium oxide preparation method and system based on deep learning and heat exchange method
Technical Field
The application relates to the technical field of gallium oxide preparation, in particular to a gallium oxide preparation method and system based on deep learning and a heat exchange method.
Background
Gallium oxide (Ga)2O3) The single crystal is a transparent semiconductor oxide, and belongs to a wide bandgap semiconductor material. Usually beta-phase gallium oxide (beta-Ga)2O3) Relatively stable, beta-Ga2O3The high band gap width enables the high breakdown voltage, and the characteristics of high saturated electron drift velocity, high thermal conductivity, stable chemical properties and the like enable the beta-Ga to have high breakdown voltage2O3The single crystal has wide application prospect in the field of electronic devices. The heat exchange method is one of the methods for preparing gallium oxide, and when the gallium oxide is prepared by adopting the heat exchange method, the preparation of the gallium oxide is difficult to control, and the influence factors on the performance of gallium oxide products are too many, so that the gallium oxide with better properties cannot be obtained.
Therefore, the prior art is in need of improvement.
Disclosure of Invention
The invention aims to solve the technical problem of providing a gallium oxide preparation method and system based on deep learning and a heat exchange method so as to optimize the performance of gallium oxide.
The embodiment of the invention provides a gallium oxide prediction method based on deep learning and a heat exchange method, which comprises the following steps:
acquiring preparation data of the gallium oxide single crystal; wherein the preparation data comprises: seed data, environmental data, and control data; the control data includes: seed crystal cooling medium flow;
preprocessing the preparation data to obtain preprocessed preparation data;
inputting the preprocessed preparation data into a trained neural network model, and obtaining the corresponding prediction property data of the gallium oxide single crystal through the trained neural network model.
The gallium oxide prediction method based on deep learning and heat exchange method is characterized in that the preprocessing is performed on the preparation data to obtain preprocessed preparation data, and comprises the following steps:
obtaining preprocessed preparation data according to the seed crystal data, the environment data and the control data; wherein the pre-processed preparation data is a matrix formed by the seed data, the environmental data, and the control data.
The gallium oxide prediction method based on the deep learning and heat exchange method comprises the following seed crystal data: the full width at half maximum of the diffraction peak of the seed crystal, the deviation value of the full width at half maximum of the diffraction peak of the seed crystal and the diameter of the seed crystal;
the environmental data includes: the thermal resistance value of the heat-insulating layer, the deviation value of the thermal resistance value of the heat-insulating layer and the shape factor of the heat-insulating layer;
the control data further comprises: coil input power and coil cooling power.
According to the gallium oxide prediction method based on the deep learning and heat exchange method, the preprocessed preparation data are obtained according to the seed crystal data, the environmental data and the control data, and the method comprises the following steps:
determining a preparation vector according to the seed crystal data, the environment data and the control data; the first element in the prepared vector is one of the seed crystal diffraction peak half-height width, the seed crystal diffraction peak half-height width deviation value and the seed crystal diameter, the second element in the prepared vector is one of the heat insulation layer thermal resistance value, the heat insulation layer thermal resistance value deviation value and the heat insulation layer shape factor, and the third element in the prepared vector is one of the coil input power, the coil cooling power and the seed crystal cooling medium flow;
and determining the preprocessed preparation data according to the preparation vector.
The gallium oxide prediction method based on the deep learning and heat exchange method comprises the following steps: predicting crack data, predicting mixed crystal data, predicting diffraction peak full width at half maximum radial deviation value and predicting diffraction peak full width at half maximum axial deviation value.
A gallium oxide preparation method based on deep learning and heat exchange method, the preparation method comprises:
acquiring target property data of a target gallium oxide single crystal;
determining target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model; wherein the target preparation data comprises: seed data, environmental data, and control data; the control data includes: coil input power, coil cooling power and seed crystal cooling medium flow;
and preparing the target gallium oxide single crystal according to the target preparation data based on a heat exchange method.
The gallium oxide preparation method based on deep learning and heat exchange method determines target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model, and comprises the following steps:
acquiring preset preparation data, and preprocessing the preset preparation data to obtain preprocessed preset preparation data;
inputting the preprocessed preset preparation data into a trained neural network model, and obtaining the corresponding prediction property data of the gallium oxide single crystal through the trained neural network model;
and correcting the preset preparation data according to the predicted property data and the target property data to obtain target preparation data corresponding to the target gallium oxide single crystal.
The gallium oxide prediction method based on the deep learning and heat exchange method is characterized in that the trained neural network model is obtained by training through the following steps:
acquiring training data of the gallium oxide single crystal and actual property data corresponding to the training data; wherein the training data comprises: seed crystal training data, environment training data and control training data; the control data includes: seed crystal cooling medium flow;
preprocessing the training data to obtain preprocessed training data;
inputting the preprocessed training data into a preset neural network model, and obtaining predicted generation property data corresponding to the preprocessed training data through the preset neural network model;
and adjusting the model parameters of the preset neural network model according to the predicted generation property data and the actual property data to correct so as to obtain the trained neural network model.
The gallium oxide prediction method based on the deep learning and heat exchange method, wherein the preset neural network model comprises: a feature extraction module and a full-connection module,
the inputting the training data into a preset neural network model, and obtaining the predictive generation property data corresponding to the training data through the preset neural network model, includes:
inputting the training data into the feature extraction module, and obtaining a feature vector corresponding to the training data through the feature extraction module;
and inputting the feature vector into the full-connection module, and obtaining the prediction generation property data obtained by the training data through the full-connection module.
A gallium oxide production system based on deep learning and heat exchange method, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the prediction method of any of the above, or the steps of the production method of any of the above, when executing the computer program.
Compared with the prior art, the embodiment of the invention has the following advantages:
the method comprises the steps of preprocessing preparation data to obtain preprocessed preparation data, inputting the preprocessed preparation data into a trained neural network model, obtaining corresponding prediction property data of the gallium oxide single crystal through the trained neural network model, and predicting the performance of the gallium oxide single crystal through the trained neural network model, so that the required performance of the gallium oxide single crystal can be obtained by adjusting the preparation data, and the performance of the gallium oxide single crystal is optimized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a gallium oxide prediction method based on deep learning and heat exchange method in an embodiment of the present invention;
FIG. 2 is a schematic view showing the structure of a crystal growth furnace according to an embodiment of the present invention;
FIG. 3 is a schematic view showing the in-furnace position and temperature of a crystal in an embodiment of the present invention;
fig. 4 is an internal structure diagram of a gallium oxide production system based on deep learning and a heat exchange method in an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The heat exchange method utilizes a heat exchanger to take away heat, so that a longitudinal temperature gradient of a lower cooling part and an upper heating part is formed in a crystal growth area, and the temperature gradient is controlled by controlling the gas flow in the heat exchanger and changing the heating input power and the cooling power, thereby achieving the purpose that a melt in a crucible is slowly solidified from the lower part to the upper part to form crystals, as shown in figure 2.
Various non-limiting embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Referring to fig. 1-3, a gallium oxide prediction method based on deep learning and heat exchange method in an embodiment of the present invention is shown. In this embodiment, the prediction method may include the following steps:
s100, obtaining preparation data of gallium oxide single crystals; wherein the preparation data comprises: seed data, environmental data, and control data; the control data includes: and (4) seed crystal cooling medium flow.
Specifically, the production data refer to data for producing a gallium oxide single crystal. The method comprises the steps of obtaining preparation data of the gallium oxide single crystal, wherein the preparation data can be configured according to needs, for example, when the performance of the obtained gallium oxide single crystal needs to be predicted under certain preparation data, only the preparation data needs to be determined, preprocessing is carried out on the preparation data to obtain preprocessed preparation data, the preprocessed preparation data is input into a trained neural network model, and predicted property data is obtained through the trained neural network model.
The preparation data includes: seed data, environmental data, and control data. The seed crystal data refers to data of seed crystals adopted in the process of preparing the gallium oxide single crystal, the environment data refers to data of the environment where the crystals are located in the process of preparing the gallium oxide single crystal, and the control data refers to data for controlling the crystal growth in the process of preparing the gallium oxide single crystal; the flow rate of the seed crystal cooling medium is the flow rate of the gas for cooling the vicinity of the seed crystal at the bottom of the crucible.
S200, preprocessing the preparation data to obtain preprocessed preparation data.
Specifically, after the preparation data are obtained, the preparation data are preprocessed to obtain preprocessed preparation data, so that the preprocessed preparation data can be input into a trained neural network model, and the preprocessed preparation data can be processed through the trained neural network model.
In an implementation manner of the embodiment of the present application, the step S200 of preprocessing the preparation data to obtain preprocessed preparation data includes:
s210, obtaining preprocessed preparation data according to the seed crystal data, the environment data and the control data; wherein the pre-processed preparation data is a matrix formed by the seed data, the environmental data, and the control data.
Specifically, after the preparation data is obtained, the preparation data is preprocessed to obtain the preprocessed preparation data, and since sub-data (such as seed crystal data, environmental data, and control data) in the preparation data affect each other, but the degree of the interaction between the sub-data cannot be determined at present, the preparation data needs to be preprocessed, and the sub-data in the preparation data is rearranged and combined to form the preprocessed preparation data.
In one implementation manner of the embodiment of the present application, the seed crystal data includes: the full width at half maximum of the diffraction peak of the seed crystal, the deviation value of the full width at half maximum of the diffraction peak of the seed crystal and the diameter of the seed crystal; the environmental data includes: the thermal resistance value of the heat-insulating layer, the deviation value of the thermal resistance value of the heat-insulating layer and the shape factor of the heat-insulating layer; the control data further comprises: coil input power and coil cooling power.
Specifically, the full width at half maximum of the diffraction peak of the seed crystal can be tested by adopting an X-ray diffractometer, and the deviation value of the full width at half maximum of the diffraction peak of the seed crystal comprises: the radial deviation value of the full width at half maximum of the diffraction peak of the seed crystal and the axial deviation value of the full width at half maximum of the diffraction peak of the seed crystal. The radial direction is the direction on the horizontal plane, and the axial direction is the direction perpendicular to the horizontal plane, i.e. the axis of the vertical direction. The radial deviation value of the half-height width of the diffraction peak of the seed crystal can be obtained by testing the half-height width of the diffraction peak of the seed crystal on the two radial sides of the seed crystal and solving the difference value between the half-height widths of the diffraction peaks of the seed crystal on the two radial sides of the seed crystal. The axial deviation value of the half height width of the diffraction peak of the seed crystal can be obtained by testing the half height width of the diffraction peak of the seed crystal on two axial sides of the seed crystal and solving the difference value between the half height widths of the diffraction peak of the seed crystal on the two axial sides of the seed crystal.
When the gallium oxide single crystal is produced by the heat exchange method, usually, the cooling gas is blown from the bottom to the top, and the temperature above the crucible is higher than the temperature below the crucible, as shown in fig. 2, the cooling energy is transferred from the bottom to the top by continuously blowing the cooling gas from below the crucible, and the gallium oxide melt in the crucible gradually grows into the gallium oxide single crystal. The bottom of crucible 1 narrows down and forms the point portion, and the seed crystal is located the point portion, that is to say, in the crystal growth process, because from the crucible below constantly blow cooling gas, the temperature of crucible reduces from bottom to top gradually, and fuse-element 3 begins to grow from the seed crystal department of the bottom of crucible 1, grows gradually into crystal 2. Of course, the seed crystal may be placed at the tip of the crucible 1 after the gallium oxide in the crucible 1 is completely melted. The tip part is connected with a cooling medium conveying pipe, the cooling medium is conveyed through the cooling medium conveying pipe, the seed crystal cooling medium comprises water, gas and oil, and preferably, the gas is used as the seed crystal cooling medium.
As shown in fig. 2, an insulating layer is disposed outside the induction coil, and the insulating layer is used for maintaining temperature. The thermal resistance of the heat-insulating layer refers to the temperature difference between the two ends of the heat-insulating layer when unit heat passes through the heat-insulating layer in unit time. The larger the thermal resistance of the insulating layer is, the stronger the heat transfer resistance of the insulating layer is, and the better the heat preservation effect of the insulating layer is.
The thermal resistance deviation value of the heat preservation layer comprises: the radial thermal resistance deviation value of the heat-insulating layer and the axial thermal resistance deviation value of the heat-insulating layer. The heat preservation radial thermal resistance value deviation value can be obtained by testing the heat preservation thermal resistance values of the two radial sides of the heat preservation layer and calculating the difference value between the heat preservation thermal resistance values on the two radial sides of the heat preservation layer. The heat preservation axial thermal resistance value deviation value can be obtained by testing the heat preservation thermal resistance values of the two axial sides of the heat preservation layer and calculating the difference value between the heat preservation thermal resistance values on the two axial sides of the heat preservation layer.
The insulation layer form factor refers to the value of the shape and size of the insulation region, for example, when a cylindrical insulation layer is used, the insulation layer form factor includes: the diameter of the heat-insulating layer and the height of the heat-insulating layer. When a cubic insulating layer is used, the insulating layer form factor includes: the length of the heat-insulating layer, the width of the heat-insulating layer and the height of the heat-insulating layer.
After the crucible and the crystal growth furnace are determined, the shape factor of the insulating layer is determined. Along with the use of crystal growth stove, heat preservation thermal resistance value deviation value, heat preservation form factor can change, but can not change in the short time, can carry out the crystal growth of certain number of times after, again test these environmental data.
The coil input power refers to the input power of the induction coil 4 when crystals grow, the coil cooling power refers to the power corresponding to the induction coil during cooling, and the induction coil adopts a hollow induction coil, so that a cooling medium is introduced into the induction coil during cooling, the induction coil forms a cooling coil, and the cooling medium continuously flows in the cooling coil to cool. The cooling power of the high-temperature region and the low-temperature region can be determined according to the type of the cooling medium and the flow rate of the cooling medium, and the type of the cooling medium comprises the following components: the flow of water, oil, gas, cooling medium can be determined according to the flow rate of the cooling medium and the diameter of the cooling coil. The flow rate of the seed crystal cooling medium is the flow rate of the gas for cooling the vicinity of the seed crystal at the bottom of the crucible.
In an implementation manner of the embodiment of the present application, in step S210, obtaining pre-processed preparation data according to the seed crystal data, the environment data, and the control data includes:
s211, determining a preparation vector according to the seed crystal data, the environment data and the control data; the first element in the prepared vector is one of the seed crystal diffraction peak full width at half maximum, the seed crystal diffraction peak full width at half maximum deviation and the seed crystal diameter, the second element in the prepared vector is one of the heat insulation layer thermal resistance value, the heat insulation layer thermal resistance value deviation and the heat insulation layer shape factor, and the third element in the prepared vector is one of the coil input power, the coil cooling power and the seed crystal cooling medium flow.
S212, determining the preprocessed preparation data according to the preparation vector.
Specifically, from the seed data a, the environment data B, and the control data C, the preparation vector (a, B, C) is determined. Seed data a is selected from: the full width at half maximum of the seed crystal diffraction peak A1, the full width at half maximum deviation value of the seed crystal diffraction peak A2 and the seed crystal diameter A3. The environmental data B is selected from: the thermal insulation layer thermal resistance value B1, the thermal insulation layer thermal resistance value deviation value B2 and the thermal insulation layer form factor B3. The control data C is selected from: coil input power C1, coil cooling power C2 and seed crystal cooling medium flow C3. That is, in the preparation vector (a, B, C), a may be one of a1, a2, A3, B may be one of B1, B2, B3, and C may be one of C1, C2, C3. 27 prepared vectors can be formed.
And arranging all the preparation vectors according to the sequence numbers to form a matrix, thus obtaining the preprocessed preparation data.
Specifically, the preparation data for the pretreatment were as follows:
Figure BDA0002877721390000071
of course, other arrangements are also used to obtain the pre-processed preparation data.
S300, inputting the preprocessed preparation data into a trained neural network model, and obtaining the corresponding prediction property data of the gallium oxide single crystal through the trained neural network model.
Specifically, the predictive property data includes: predicting crack data, predicting mixed crystal data, predicting diffraction peak full width at half maximum radial deviation value and predicting diffraction peak full width at half maximum axial deviation value.
The crack data is crack grade data, and the predicted crack data is predicted crack grade data, for example, the crack may be classified into a plurality of grades, for example, if the crack is classified into 3 grades, the crack data are: 1. 2 and 3.
The mixed crystal data refers to mixed crystal grade data, and the predicted mixed crystal data refers to predicted mixed crystal grade data, for example, the mixed crystal can be divided into multiple grades, for example, the mixed crystal is divided into 3 grades, and then the mixed crystal data are respectively: 1. 2 and 3.
The predicted half-height width of the diffraction peak refers to the predicted half-height width of the diffraction peak, the predicted half-height width radial deviation value of the diffraction peak refers to the predicted difference value of the half-height width of the diffraction peak at the radial center and the edge, and the predicted half-height width axial deviation value of the diffraction peak refers to the predicted difference value of the half-height width of the diffraction peak at the two axial ends.
And inputting the preprocessed preparation data into a trained neural network model, and obtaining the predicted property data through the neural network model. It should be noted that the predictive property data may be one or more, e.g., only the predictive crack data is required.
In an implementation manner of the embodiment of the present application, the trained neural network model is obtained by training through the following steps:
a100, acquiring training data of gallium oxide single crystals and actual property data corresponding to the training data; wherein the training data comprises: seed crystal training data, environment training data and control training data; the control data includes: and (4) seed crystal cooling medium flow.
Specifically, the training data refers to data for preparing a gallium oxide single crystal and for training, and the actual property data refers to data of actual properties of the prepared gallium oxide single crystal. And forming a training set through the training data and the actual property data, and training a preset neural network model based on the training set to obtain the trained neural network model. The control data includes: coil input power and coil cooling power. The seed crystal training data comprises: seed crystal diffraction peak half-height width training data, seed crystal diffraction peak half-height width deviation value training data and seed crystal diameter training data; the environmental training data includes: thermal insulation layer thermal resistance value training data, thermal insulation layer thermal resistance value deviation value training data and thermal insulation layer shape factor training data; the control training data includes: seed crystal cooling medium flow, of course, the control training data further includes: coil input power training data, coil cooling power training data. The actual property data includes: actual crack data, actual mixed crystal data, actual diffraction peak full width at half maximum radial deviation value, and actual diffraction peak full width at half maximum axial deviation value.
When data are collected to obtain a training set, preparing the gallium oxide single crystal by adopting a heat exchange method, recording data of preparing the gallium oxide single crystal as training data, and analyzing the properties of the gallium oxide single crystal after the gallium oxide single crystal is obtained to obtain actual property data. In order to facilitate the training of the neural network model, data as much as possible can be collected to form a training set.
And A200, preprocessing the training data to obtain preprocessed training data.
Specifically, after the training data is obtained, the training data is preprocessed to obtain preprocessed training data. The preprocessing process may refer to step S200.
And A300, inputting the preprocessed training data into a preset neural network model, and obtaining predicted generation property data corresponding to the preprocessed training data through the preset neural network model.
Specifically, the preprocessed training data are input into a preset neural network model, and predicted generation property data are obtained through the preset neural network model. The predictive generation of property data includes: predicting generated crack data, predicting generated mixed crystal data, predicting generated diffraction peak half-height width radial deviation value and predicting generated diffraction peak half-height width axial deviation value.
And A400, adjusting the model parameters of the preset neural network model according to the predicted generated property data and the actual property data to correct so as to obtain the trained neural network model.
Specifically, according to the predicted generated property data and the actual property data, the model parameters of the preset neural network model are corrected, the preprocessed training data are input into the preset neural network model, and the predicted generated property data corresponding to the preprocessed training data are obtained through the preset neural network model (namely, step a300), until a preset training condition is met, so that the trained neural network model is obtained.
Specifically, according to the predicted generated property data and the actual property data, model parameters of the preset neural network model are corrected, the preprocessed training data are input into the preset neural network model, and the predicted generated property data corresponding to the preprocessed training data are obtained through the preset neural network model until preset training conditions are met, so that the trained neural network model is obtained. That is, if the preset neural network model meets the preset training condition, the trained neural network model is obtained. And if the preset neural network model does not meet the preset training condition, returning to the step A300 until the preset neural network model meets the preset training condition to obtain the trained neural network model.
In an implementation manner of the embodiment of the present invention, a loss function value of a preset neural network model is determined according to the predicted generated property data and the actual property data, and a model parameter of the preset neural network model is corrected according to the loss function value. Specifically, a gradient-based method is adopted to correct parameters of the preset neural network model, and after a loss function value of the preset neural network model is determined, the corrected parameters of the preset neural network model are determined according to the gradient of the loss function value to the parameters of the preset neural network model, the parameters of the preset neural network model and a preset learning rate.
The preset training conditions comprise: the loss function value meets a first preset requirement and/or the training times of the preset neural network model reach a first preset time.
The first preset requirement is determined according to the accuracy and efficiency of the preset neural network model, for example, the loss function value of the preset neural network model reaches a minimum value or does not change any more. The first preset number is a maximum training number of the preset neural network model, for example, 4000 times.
The loss function of the preset neural network model comprises the following steps: mean square error, root mean square error, mean absolute error, and the like.
In an implementation manner of the embodiment of the present application, the preset neural network model includes: the device comprises a feature extraction module and a full connection module.
For example, the preset neural network model includes: the device comprises a first convolution unit, a second convolution unit, a third convolution unit, a fourth convolution unit and a full connection unit. Specifically, the first convolution unit includes: two convolutional layers and one pooling layer. The second convolution unit, the third convolution unit and the fourth convolution unit all comprise three convolution layers and one pooling layer. The full-link unit includes three full-link layers.
The convolutional layer and the full link layer are responsible for mapping input data, and this process uses parameters such as weight and bias, and also needs to use an activation function. The pooling layer is a fixed and invariant function operation. Specifically, the convolutional layer functions to extract features; the pooling layer performs pooling operation on the input features to change the space size of the input features; and the full connection layer is fully connected to all data in the previous layer.
Step A300, inputting the preprocessed training data into a preset neural network model, and obtaining the preprocessed training data prediction generation property data through the preset neural network model, wherein the step A comprises the following steps:
a310, inputting the preprocessed training data into the feature extraction module, and obtaining a feature vector corresponding to the preprocessed training data through the feature extraction module;
and A320, inputting the feature vector into the full-connection module, and obtaining the prediction generation property data obtained by the preprocessed training data through the full-connection module.
Specifically, the preprocessed training data is input into a preset neural network model, a feature vector corresponding to the preprocessed training data is output through the feature extraction module in the preset neural network model, and the feature vector is input into the full-connection module in the pre-training model, so that the predicted generation property data corresponding to the preprocessed training data output by the full-connection module is obtained.
Based on the above gallium oxide prediction method based on deep learning and heat exchange method, this embodiment provides a gallium oxide preparation method based on deep learning and heat exchange method, where the preparation method includes:
and B100, acquiring target property data of the target gallium oxide single crystal.
Specifically, if it is desired to obtain a target gallium oxide single crystal, target property data of the target gallium oxide single crystal, that is, property data of the desired gallium oxide single crystal, may be determined first. The target property data includes: target crack data, target mixed crystal data, target diffraction peak full width at half maximum radial deviation value and target diffraction peak full width at half maximum axial deviation value.
B200, determining target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model; wherein the target preparation data comprises: seed data, environmental data, and control data; the control data includes: coil input power, coil cooling power and seed crystal cooling medium flow.
Specifically, target preparation data corresponding to the target gallium oxide single crystal is determined according to the target property data and a trained neural network model. When it is required to be mentioned that, since different preparation data can obtain the same property data, when the target preparation data corresponding to the target gallium oxide single crystal is determined according to the target property data and the trained neural network model, the target preparation data is not unique, and one target preparation data is determined according to the control difficulty of each data in a plurality of target preparation data, thereby facilitating obtaining the target gallium oxide single crystal.
In an implementation manner of this embodiment, determining, by B200, target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model includes:
and B210, acquiring preset preparation data, and preprocessing the preset preparation data to obtain preprocessed preset preparation data.
And B220, inputting the preprocessed preset preparation data into a trained neural network model, and obtaining the corresponding prediction property data of the gallium oxide single crystal through the trained neural network model.
And B230, correcting the preset preparation data according to the predicted property data and the target property data to obtain target preparation data corresponding to the target gallium oxide single crystal.
Specifically, the preparation data may be preset first, and the preset preparation data is preprocessed to obtain preprocessed preset preparation data, and the specific preprocessing process may refer to step S200. Inputting preset preparation data into a trained neural network model to obtain predicted property data, then correcting the preset preparation data according to the predicted property data and the target property data, and taking the preset preparation data as target preparation data when a difference value between the predicted property data and the target property data is smaller than a preset threshold value. When the preset preparation data is corrected, automatic correction or manual correction can be performed. Of course, the loss function value may also be determined according to the predicted property data and the target property data, and then the preset preparation data may be corrected according to the loss function value, and when the loss function value satisfies the preset correction condition, the preset preparation data may be used as the target preparation data. The preset correction conditions include: the loss function value meets a second preset requirement and/or the correction times of the preset preparation data reach a second preset times.
It should be noted that the preset preparation data includes: presetting seed crystal data, environment data and control data; the preset seed crystal data comprises: presetting the full width at half maximum of a seed crystal diffraction peak, presetting the deviation value of the full width at half maximum of the seed crystal diffraction peak and presetting the seed crystal diameter; the preset environment data includes: presetting thermal resistance values of the insulating layer, presetting deviation values of the thermal resistance values of the insulating layer and presetting form factors of the insulating layer; the preset control data includes: presetting coil input power, presetting coil cooling power and presetting seed crystal cooling medium flow.
And B300, preparing the target gallium oxide single crystal according to the target preparation data based on a heat exchange method.
Specifically, after the target production data is obtained, the target gallium oxide single crystal can be produced based on the target production data based on the heat exchange method.
Based on the prediction method or the preparation method, the invention provides a gallium oxide preparation system based on a deep learning and heat exchange method, which can be computer equipment, and the internal structure of the gallium oxide preparation system is shown in FIG. 4. The system comprises a processor, a memory, a network interface, a display screen and an input device which are connected through a system bus. Wherein the processor of the system is configured to provide computing and control capabilities. The memory of the system comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the system is used for communicating with an external terminal through network connection. The computer program is executed by a processor to realize a prediction method of gallium oxide based on deep learning and heat exchange method or a preparation method of gallium oxide based on deep learning and heat exchange method. The display screen of the system can be a liquid crystal display screen or an electronic ink display screen, and the input device of the system can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on a system shell, an external keyboard, a touch pad or a mouse and the like.
It will be appreciated by those skilled in the art that the block diagram of FIG. 4 is merely a partial block diagram of the structure associated with the disclosed aspects and is not intended to limit the systems to which the disclosed aspects apply, and that a particular system may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a system for preparing gallium oxide based on deep learning and heat exchange method is provided, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the prediction method or the steps of the preparation method when executing the computer program.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

Claims (10)

1. A gallium oxide prediction method based on deep learning and heat exchange method is characterized by comprising the following steps:
acquiring preparation data of the gallium oxide single crystal; wherein the preparation data comprises: seed data, environmental data, and control data; the control data includes: seed crystal cooling medium flow;
preprocessing the preparation data to obtain preprocessed preparation data;
inputting the preprocessed preparation data into a trained neural network model, and obtaining the corresponding prediction property data of the gallium oxide single crystal through the trained neural network model.
2. The method for predicting gallium oxide based on deep learning and heat exchange method according to claim 1, wherein the preprocessing the preparation data to obtain preprocessed preparation data comprises:
obtaining preprocessed preparation data according to the seed crystal data, the environment data and the control data; wherein the pre-processed preparation data is a matrix formed by the seed data, the environmental data, and the control data.
3. The deep learning and heat exchange method based gallium oxide prediction method according to claim 2, wherein the seed data comprises: the full width at half maximum of the diffraction peak of the seed crystal, the deviation value of the full width at half maximum of the diffraction peak of the seed crystal and the diameter of the seed crystal;
the environmental data includes: the thermal resistance value of the heat-insulating layer, the deviation value of the thermal resistance value of the heat-insulating layer and the shape factor of the heat-insulating layer;
the control data further comprises: coil input power and coil cooling power.
4. The deep learning and heat exchange method-based gallium oxide prediction method according to claim 3, wherein the obtaining pre-processed preparation data from the seed data, the environmental data and the control data comprises:
determining a preparation vector according to the seed crystal data, the environment data and the control data; the first element in the prepared vector is one of the seed crystal diffraction peak half-height width, the seed crystal diffraction peak half-height width deviation value and the seed crystal diameter, the second element in the prepared vector is one of the heat insulation layer thermal resistance value, the heat insulation layer thermal resistance value deviation value and the heat insulation layer shape factor, and the third element in the prepared vector is one of the coil input power, the coil cooling power and the seed crystal cooling medium flow;
and determining the preprocessed preparation data according to the preparation vector.
5. The deep learning and heat exchange approach based gallium oxide prediction method according to claim 1, wherein the prediction property data comprises: predicting crack data, predicting mixed crystal data, predicting diffraction peak full width at half maximum radial deviation value and predicting diffraction peak full width at half maximum axial deviation value.
6. A gallium oxide preparation method based on deep learning and heat exchange method is characterized by comprising the following steps:
acquiring target property data of a target gallium oxide single crystal;
determining target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model; wherein the target preparation data comprises: seed data, environmental data, and control data; the control data includes: seed crystal cooling medium flow;
and preparing the target gallium oxide single crystal according to the target preparation data based on a heat exchange method.
7. The method for preparing gallium oxide based on deep learning and heat exchange method according to claim 6, wherein determining the target preparation data corresponding to the target gallium oxide single crystal according to the target property data and the trained neural network model comprises:
acquiring preset preparation data, and preprocessing the preset preparation data to obtain preprocessed preset preparation data;
inputting the preprocessed preset preparation data into a trained neural network model, and obtaining prediction property data corresponding to the gallium oxide single crystal through the trained neural network model;
and correcting the preset preparation data according to the predicted property data and the target property data to obtain target preparation data corresponding to the target gallium oxide single crystal.
8. The method for predicting gallium oxide based on deep learning and heat exchange method according to claim 6, wherein the trained neural network model is obtained by training with the following steps:
acquiring training data of the gallium oxide single crystal and actual property data corresponding to the training data; wherein the training data comprises: seed crystal training data, environment training data and control training data; the control data includes: seed crystal cooling medium flow;
preprocessing the training data to obtain preprocessed training data;
inputting the preprocessed training data into a preset neural network model, and obtaining predicted generation property data corresponding to the preprocessed training data through the preset neural network model;
and adjusting the model parameters of the preset neural network model according to the predicted generation property data and the actual property data to correct so as to obtain the trained neural network model.
9. The method for predicting gallium oxide based on deep learning and heat exchange method according to claim 8, wherein the preset neural network model comprises: a feature extraction module and a full-connection module,
the inputting the training data into a preset neural network model, and obtaining the predictive generation property data corresponding to the training data through the preset neural network model, includes:
inputting the training data into the feature extraction module, and obtaining a feature vector corresponding to the training data through the feature extraction module;
and inputting the feature vector into the full-connection module, and obtaining the prediction generation property data obtained by the training data through the full-connection module.
10. A gallium oxide production system based on a deep learning and heat exchange method, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor executes the computer program to implement the steps of the prediction method according to any one of claims 1 to 5 or the steps of the production method according to any one of claims 6 to 9.
CN202011638983.3A 2020-12-31 2020-12-31 Gallium oxide preparation method and system based on deep learning and heat exchange method Pending CN112836853A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202011638983.3A CN112836853A (en) 2020-12-31 2020-12-31 Gallium oxide preparation method and system based on deep learning and heat exchange method
PCT/CN2021/075545 WO2022141750A1 (en) 2020-12-31 2021-02-05 Gallium oxide preparation method and system based on deep learning and heat exchange method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011638983.3A CN112836853A (en) 2020-12-31 2020-12-31 Gallium oxide preparation method and system based on deep learning and heat exchange method

Publications (1)

Publication Number Publication Date
CN112836853A true CN112836853A (en) 2021-05-25

Family

ID=75926831

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011638983.3A Pending CN112836853A (en) 2020-12-31 2020-12-31 Gallium oxide preparation method and system based on deep learning and heat exchange method

Country Status (2)

Country Link
CN (1) CN112836853A (en)
WO (1) WO2022141750A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111695287A (en) * 2020-04-22 2020-09-22 山东天岳先进材料科技有限公司 Method and equipment for predicting overall temperature field in SiC single crystal furnace

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2006273684A (en) * 2005-03-30 2006-10-12 Koha Co Ltd Method for producing group iii element oxide-based single crystal
CN103541008B (en) * 2013-11-12 2016-01-06 上海硅酸盐研究所中试基地 A kind of growth method of large size gallium oxide single crystal and growing apparatus
EP3042986A1 (en) * 2015-01-09 2016-07-13 Forschungsverbund Berlin e.V. Method for growing beta phase of gallium oxide (ß-Ga2O3) single crystals from the melt contained within a metal crucible by controlling the partial pressure of oxygen.
CN107541776A (en) * 2017-08-14 2018-01-05 同济大学 A kind of growth apparatus and method of large scale gallium oxide single crystal
KR102549196B1 (en) * 2018-02-07 2023-06-30 어플라이드 머티리얼즈 이스라엘 리미티드 Inspection method based on deep learning of semiconductor specimen and system thereof
CN109487334A (en) * 2018-11-22 2019-03-19 太原理工大学 A kind of kyropoulos sapphire based on random distribution melts brilliant inoculation state control method
JP7115688B2 (en) * 2019-01-25 2022-08-09 株式会社デンソー Film forming apparatus and semiconductor device manufacturing method
KR102012809B1 (en) * 2019-04-05 2019-08-21 충남대학교산학협력단 Method for Evaluating Twin Defect Density

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111695287A (en) * 2020-04-22 2020-09-22 山东天岳先进材料科技有限公司 Method and equipment for predicting overall temperature field in SiC single crystal furnace

Also Published As

Publication number Publication date
WO2022141750A1 (en) 2022-07-07

Similar Documents

Publication Publication Date Title
CN112853468A (en) Conductive gallium oxide preparation method based on deep learning and heat exchange method
CN112734009A (en) Gallium oxide preparation method and system based on deep learning and Bridgman method
CN105046030A (en) Method for obtaining quenching process heat transfer coefficient of aluminum alloy component under three-dimensional heat transfer condition based on finite element method
CN112834700B (en) Quality prediction method, preparation method and system of high-resistance gallium oxide based on deep learning and guided mode method
US20230160096A1 (en) Quality prediction method, preparation method and system of high resistance gallium oxide based on deep learning and czochralski method
CN112863617A (en) High-resistance gallium oxide preparation method based on deep learning and Bridgman-Stockbarge method
WO2022141752A1 (en) Deep learning and czochralski method-based quality prediction method and preparation method and system for conductive gallium oxide
CN112836853A (en) Gallium oxide preparation method and system based on deep learning and heat exchange method
CN112859771A (en) Quality prediction method, preparation method and system of gallium oxide based on deep learning and model-guiding method
CN107092724A (en) One kind considers probabilistic thermal protection system model modification method
CN112863618A (en) High-resistance gallium oxide preparation method based on deep learning and heat exchange method
CN112863619A (en) Conductive gallium oxide preparation method based on deep learning and Bridgman method
CN115595667B (en) Intelligent growth method, system, equipment and storage medium for tellurium-zinc-cadmium crystals
Ren et al. Data-driven and mechanism-based hybrid model for semiconductor silicon monocrystalline quality prediction in the czochralski process
CN114117675B (en) Numerical simulation method and system for temperature and humidity field of operating mechanism
WO2022141764A1 (en) Quality prediction method for conductive gallium oxide on the basis of deep learning and edge-defined film-fed growth method, and preparation method and system
CN112853470A (en) Quality prediction method, preparation method and system of gallium oxide based on deep learning and Czochralski method
Tian et al. Optimizing Temperature and Flow Fields of 4H-SiC Epitaxial Growth by Integrating CFD Simulation with Multi-objective Particle Swarm Optimization
CN115331742A (en) Method for determining heater power of ammonothermal gallium nitride growth system
JPWO2022141758A5 (en)

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination