WO2022141765A1 - 一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统 - Google Patents

一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统 Download PDF

Info

Publication number
WO2022141765A1
WO2022141765A1 PCT/CN2021/076071 CN2021076071W WO2022141765A1 WO 2022141765 A1 WO2022141765 A1 WO 2022141765A1 CN 2021076071 W CN2021076071 W CN 2021076071W WO 2022141765 A1 WO2022141765 A1 WO 2022141765A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
preparation
gallium oxide
crystal
target
Prior art date
Application number
PCT/CN2021/076071
Other languages
English (en)
French (fr)
Inventor
齐红基
陈端阳
赛青林
Original Assignee
杭州富加镓业科技有限公司
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 杭州富加镓业科技有限公司 filed Critical 杭州富加镓业科技有限公司
Priority to JP2022520490A priority Critical patent/JP7478819B2/ja
Priority to US17/760,964 priority patent/US12024791B2/en
Priority to EP21870470.8A priority patent/EP4050333A4/en
Publication of WO2022141765A1 publication Critical patent/WO2022141765A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • 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
    • C30B15/00—Single-crystal growth by pulling from a melt, e.g. Czochralski method
    • C30B15/34—Edge-defined film-fed crystal-growth using dies or slits
    • 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
    • C30B15/00—Single-crystal growth by pulling from a melt, e.g. Czochralski method
    • C30B15/20—Controlling or regulating
    • 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
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D21/00—Measuring or testing not otherwise provided for
    • G01D21/02—Measuring two or more variables by means not covered by a single other subclass
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/0095—Semiconductive materials
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/70—Machine learning, data mining or chemometrics

Definitions

  • the invention relates to the field of high resistance type gallium oxide crystal preparation, in particular to a quality prediction method, preparation method and system of high resistance type gallium oxide based on deep learning and guided mode method.
  • ⁇ -Ga 2 O 3 (high-resistance gallium oxide) is a direct bandgap wide bandgap semiconductor material with a band gap of about 4.8-4.9eV. It has many advantages such as large band gap, fast saturation electron drift speed, high thermal conductivity, high breakdown field strength, and stable chemical properties. In addition, it can also be used in LED chips, solar-blind ultraviolet detection, various sensor elements and camera elements, etc.
  • the large-scale high-resistance gallium oxide crystals are mainly prepared by the guided mode method.
  • the guided mode method is a mature single crystal preparation technology, which has the advantages of special-shaped crystal growth, fast growth speed and low growth cost.
  • the guided mode method is a crystal growth method improved on the basis of the guided mode method, and the exploration of crystal growth equipment and growth process is relatively more complicated.
  • the guided mode method requires a mold to be placed in the crucible, and the crystal growth interface is located on the upper surface of the mold. At high temperature, due to the action of surface tension, the melt rises along the capillary in the mold to the upper surface of the mold.
  • the rising height H of the melt along the capillary is given by the formula where, ⁇ is the surface tension of the melt, ⁇ is the contact angle between the melt and the capillary, ⁇ is the density of the melt, g is the acceleration of gravity, and r is the radius of the capillary.
  • the high resistance gallium oxide crystal prepared by the guided mode method has the following advantages:
  • the convection of the melt in the capillary of the mold is weak, and the impurity ions in the melt are difficult to return to the crucible after rising from the capillary to the solid-liquid interface. ;
  • the solid-liquid interface of crystal growth is located above the mold, and the shape of the interface can be controlled by the geometry of the mold surface.
  • the position of the solid-liquid interface in the temperature field remains unchanged, and is not affected by the melt disturbance in the crucible. Therefore, in the guided mode method, the crystal
  • the growth solid-liquid interface is more stable;
  • the crystal growth rate of the guided mode method is faster, which is beneficial to reduce energy consumption, can grow special-shaped crystals, and reduce crystal processing costs and processing losses.
  • the temperature field distribution near the die mouth the selection of the seed crystal, the pulling speed of the seed crystal rod, the crystal growth atmosphere, the heating power, the cooling power and other parameters
  • the change of the high resistance type gallium oxide crystal will have a great influence on the quality of the obtained high resistance type gallium oxide crystal.
  • the existing process of preparing high-resistance gallium oxide crystals by the existing guided mode method relies on the operator's experience to set parameters, which has poor repeatability, resulting in poor stability of the prepared high-resistance gallium oxide crystal products.
  • the technical problem to be solved by the present invention is to provide a quality prediction method, preparation method and system of high-resistance gallium oxide based on deep learning and guided mode method, aiming at solving the existing guided mode method.
  • the process of preparing high-resistance gallium oxide crystals all rely on the operator's experience to set parameters, which has poor repeatability, resulting in problems of poor quality and stability of the prepared high-resistance gallium oxide crystals.
  • a method for predicting the quality of high-resistance gallium oxide based on deep learning and guided mode method comprising the steps of:
  • preparation data for preparing a high-resistance gallium oxide single crystal by a guided mode method including seed crystal data, environmental data and control data, the control data including doping element concentration and doping element type;
  • the preprocessing preparation data is input into a trained neural network model, and predicted quality data corresponding to the high-resistance gallium oxide single crystal is obtained through the trained neural network model, and the predicted quality data includes predicted resistivity.
  • the steps of preprocessing the preparation data to obtain the preprocessing preparation data include:
  • pre-processing preparation data is obtained, and the pre-processing preparation data is a matrix formed by the seed crystal data, the environmental data and the control data.
  • the method for predicting the quality of high-resistance gallium oxide based on deep learning and guided mode method wherein the seed crystal data includes: seed crystal diffraction peak full width at half maximum, seed crystal diffraction peak full width at half maximum deviation value, seed crystal thickness and Seed width;
  • the environmental data also includes: the thermal resistance value of the upper thermal insulation cover, the deviation value of the thermal resistance value of the upper thermal insulation cover, the shape factor of the crystal growth channel, the shape factor of the crystal growth observation hole, the thermal resistance value of the lower thermal insulation cover, and the thermal resistance value deviation of the lower thermal insulation cover. value, the relative height of the crucible and the heating coil, the relative height of the heating ring and the heating coil, the width of the die gap, and the thickness of the die gap;
  • the control data includes: heating power, cooling power, atmosphere type, cavity pressure, gas flow, and seed rod pulling speed.
  • the method for predicting the quality of high-resistance gallium oxide based on deep learning and guided mode method wherein, according to the seed crystal data, environmental data and control data, the step of obtaining preprocessing preparation data includes:
  • a preparation vector is determined; wherein, the first element in the preparation vector is the half-width of the seed crystal diffraction peak, the deviation value of the half-width of the seed crystal diffraction peak, the seed crystal One of the thickness and the width of the seed crystal; the second element in the preparation vector is the thermal resistance value of the upper insulation cover, the deviation value of the thermal resistance value of the upper insulation cover, the shape factor of the crystal growth channel, the shape factor of the crystal growth observation hole, the lower The thermal resistance value of the insulation cover, the deviation value of the thermal resistance value of the lower insulation cover, the relative height of the crucible and the heating coil, the relative height of the heating ring and the heating coil, the width of the die mouth gap, the thickness of the die mouth gap, doping element concentration and doping
  • the types of doping elements include Fe, Ca, Zn, Co, Ti, Ni, Mg, Al and Cu
  • the third element in the preparation vector is the heating power
  • the preprocessing preparation data is determined.
  • the method for predicting the quality of high-resistance gallium oxide based on deep learning and guided mode method wherein the predicted quality data further includes: predicted crack data, predicted miscellaneous crystal data, predicted diffraction peak width at half maximum, and predicted diffraction peak half height Wide deviation value, predicted shoulder symmetry of high resistance gallium oxide crystal, predicted left edge retraction degree of high resistance gallium oxide crystal, predicted right edge retraction degree of high resistance gallium oxide crystal, predicted high resistance gallium oxide crystal thickness , Predict the thickness deviation of high-resistance gallium oxide crystal, predict the radial deviation value of resistivity, and predict the axial deviation value of resistivity.
  • the high-resistance gallium oxide quality prediction method based on deep learning and guided mode method wherein, the trained neural network model is obtained by the following training steps:
  • training data for preparing a high-resistance gallium oxide single crystal by a guided mode method, and corresponding actual quality data, wherein the training data includes: seed crystal training data, environmental training data, and control training data;
  • the model parameters of the preset neural network model are adjusted and corrected according to the quality data generated by the prediction training and the actual quality data, so as to obtain a trained neural network model.
  • the preset neural network model includes: a feature extraction module and a fully connected module,
  • the preprocessing training data is input into a preset neural network model, and the step of obtaining prediction training generation quality data corresponding to the preprocessing training data through the preset neural network model includes:
  • the feature vector is input into the full connection module, and the prediction training generation quality data corresponding to the preprocessing training data is obtained through the full connection module.
  • a method for preparing high-resistance gallium oxide based on deep learning and guided mode method comprising the steps of:
  • target quality data of the target high-resistance gallium oxide single crystal including the target resistivity
  • the target preparation data corresponding to the target high-resistance gallium oxide single crystal is determined, wherein the target preparation data includes: target seed crystal data, target environment data and target control data, the target environmental data includes target doping element concentration and target doping element type;
  • the target high-resistance gallium oxide single crystal is prepared according to the target preparation data.
  • the method for preparing high-resistance gallium oxide based on deep learning and guided mode method wherein, according to the target quality data and the trained neural network model, the target preparation data corresponding to the target high-resistance gallium oxide single crystal is determined , including steps:
  • the preset preparation data is corrected to obtain target preparation data corresponding to the target high-resistance gallium oxide single crystal.
  • a high-resistance gallium oxide preparation system based on deep learning and guided mode method which includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that, when the processor executes the computer program, the present invention is realized.
  • the present invention proposes a quality prediction method and preparation method of high-resistance type gallium oxide based on deep learning and guided mode method. Preprocessing the preparation data, then inputting the preprocessing preparation data into the trained neural network model, and obtaining the predicted quality data corresponding to the high-resistance gallium oxide single crystal through the trained neural network model.
  • the present invention can predict the quality of the high-resistance gallium oxide single crystal through the trained neural network model, so that the required performance of the high-resistance gallium oxide single crystal can be obtained by adjusting the preparation data, so that the high-resistance gallium oxide single crystal can be Performance is optimized.
  • FIG. 1 is a schematic structural diagram of a crystal growth furnace for preparing high-resistance gallium oxide crystals by a guided mode method provided by the present invention.
  • FIG. 2 is a flowchart of a preferred embodiment of a method for predicting the quality of high-resistance gallium oxide based on deep learning and guided mode method provided by the present invention.
  • FIG. 3 is a flowchart of a preferred embodiment of a method for preparing high-resistance gallium oxide based on deep learning and guided mode method provided by the present invention.
  • FIG. 4 is an internal structure diagram of a high-resistance gallium oxide preparation system based on deep learning and guided mode method provided by the present invention.
  • the present invention provides a method and system for preparing high-resistance gallium oxide crystal based on deep learning and guided mode method. Detailed description. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.
  • the present invention provides a crystal growth furnace for growing high-resistance gallium oxide crystals by a guided mode method.
  • FIG. 1 it includes a lower heat preservation cover 11, and a crucible 12 is arranged in the lower heat preservation cover 11.
  • a heating ring 14 is provided on the outside of the crucible 12;
  • the upper heat preservation cover 15 is provided above the lower heat preservation cover, and crystals are provided inside the upper heat preservation cover 15
  • a crystal growth observation hole 17 is provided on the side of the upper heat preservation cover, and an induction coil 18 is also provided on the periphery of the lower heat preservation cover 10 .
  • high-resistance gallium oxide crystals can be prepared.
  • a crystal growth furnace for preparing high-resistance gallium oxide crystals is installed, including the material selection of the upper heat preservation cover and the size of the axial opening of the upper heat preservation cover. , the shape and size of the opening window, the material selection of the lower insulation cover, the height position of the crucible relative to the induction coil, the relative height of the heating ring and the induction coil, etc., these will affect the thermal field distribution inside the device, thereby affecting the high resistance gallium oxide crystal.
  • the orientation of the seed crystal can be [010], [001] direction, etc.;
  • the crucible is heated by the heating element, so that the gallium oxide raw material and the doping element material placed in the crucible are completely melted, and the molten gallium oxide and the doping element material are transported to the
  • the top of the mold is expanded on the top until it is completely covered; then, the seed rod is slowly lowered, so that the seed is preheated at a position 3-5mm above the top of the iridium mold, and inoculation is started after 5-10 minutes; After the body is fully welded, the seeding necking operation is performed to prevent the original defects of the seed crystal from extending to the inside of the crystal to ensure the quality of the crystal.
  • the shoulder expansion is performed to expand the crystal laterally to the entire mold; then the equal diameter growth is performed. ; After the crystal growth is completed, the temperature is slowly lowered to room temperature, and the crystal is taken out to obtain a high-resistance gallium oxide crystal.
  • heating power, cooling power, atmosphere type, cavity pressure, gas flow and seed rod pulling speed will all affect the product performance of high-resistance gallium oxide crystal.
  • this embodiment provides a high-resistance gallium oxide quality prediction method based on deep learning and guided mode method, as shown in FIG. 2 , which includes the following steps:
  • preparation data for preparing a high-resistance gallium oxide single crystal by a guided mode method, where the preparation data includes seed crystal data, environmental data, and control data, where the control data includes doping element concentration and doping element type.
  • the preparation data refers to the data of preparing a high-resistance gallium oxide single crystal by the guided mode method
  • the preparation data is data that can be configured according to needs, for example, it is necessary to predict the high resistance obtained under a certain preparation data.
  • the predicted quality data can be obtained from the neural network model of the model, that is to say, the quality data of the high-resistance gallium oxide single crystal can be predicted by the trained neural network model after the preparation data is determined without conducting experiments.
  • the preparation data includes: seed crystal data, environmental data and control data
  • the seed crystal data refers to the data of the seed crystal used in the process of preparing the high-resistance type gallium oxide single crystal by the guided mode method
  • the environmental data refers to the data of the environment in which the crystal is located in the process of preparing the high-resistance gallium oxide single crystal by the guided mode method
  • the control data refers to the process of preparing the high-resistance gallium oxide single crystal by the guided mode method.
  • the doping element concentration refers to the concentration of doping elements in the high resistance type gallium oxide, and the doping element types include: Fe, Ca, Zn, Co, Ti, Ni, Mg, Al, Cu, and the like.
  • the preparation data is preprocessed first to obtain the preprocess preparation data, so that the preprocess preparation data can be input into the trained neural network model, so that the trained neural network model can be passed through the trained neural network model.
  • the network model processes the preprocessed data.
  • preprocessing the preparation data includes:
  • the preparation data is preprocessed to obtain the preprocessed preparation data. Since the sub-data (such as seed crystal data, environmental data and control data) in the preparation data will affect each other, but it is currently impossible to determine the degree of mutual influence between the sub-data, therefore, the preparation data needs to be analyzed. Preprocessing, rearranging and combining each sub-data in the preparation data to form preprocessing preparation data.
  • the sub-data such as seed crystal data, environmental data and control data
  • the seed crystal data includes: the full width at half maximum of the seed crystal diffraction peak, the deviation value of the half height width of the seed crystal diffraction peak, the thickness of the seed crystal, and the width of the seed crystal;
  • the environmental data further includes: The thermal resistance value of the upper insulation cover, the deviation value of the thermal resistance value of the upper insulation cover, the shape factor of the crystal growth channel, the shape factor of the crystal growth observation hole, the thermal resistance value of the lower insulation cover, the deviation value of the thermal resistance value of the lower insulation cover, the difference between the crucible and the heating coil.
  • the control data include: heating power, cooling power, atmosphere type, cavity pressure, gas flow, seed rod lift Pull speed.
  • the seed crystal diffraction peak width at half maximum can be tested by an X-ray diffractometer
  • the seed crystal diffraction peak half height width deviation value includes the seed crystal diffraction peak half height width radial deviation value and the seed crystal diffraction peak Half-height width axial deviation value
  • the radial direction is the direction on the horizontal plane
  • the axial direction is the direction perpendicular to the horizontal plane, that is, the axis of the vertical direction.
  • the radial deviation value of the full width at half maximum of the seed crystal diffraction peak can be obtained by testing the full width at half maximum of the seed crystal diffraction peak in the radial direction of the seed crystal, and obtaining the difference between the half height widths of the seed crystal diffraction peaks in the radial direction of the seed crystal, that is, The radial deviation value of the half height width of the seed crystal diffraction peak can be obtained.
  • the axial deviation value of the width at half maximum of the seed crystal diffraction peak can be obtained by testing the half height width of the seed crystal diffraction peak in the axial direction of the seed crystal, and obtaining the difference between the half height widths of the seed crystal diffraction peaks along the seed crystal axis, The axial deviation value of the half-width of the seed crystal diffraction peak can be obtained.
  • the thickness and width of the seed crystal can be directly measured.
  • both the upper and lower thermal insulation covers provide a stable thermal field for the growth of the high-resistance gallium oxide single crystal.
  • the thermal resistance value of the upper insulation cover refers to the temperature difference between the inner and outer sides of the insulation cover when unit heat per unit time passes through the upper insulation cover. The larger the thermal resistance value of the upper insulation cover, the stronger the ability of the upper insulation cover to resist heat transfer, and the better the insulation effect of the upper insulation cover.
  • the thermal resistance value deviation value of the upper thermal insulation cover includes a radial deviation value of the thermal resistance value of the upper thermal insulation cover and an axial deviation value of the thermal resistance value of the upper thermal insulation cover.
  • the radial deviation value of the thermal resistance value of the upper insulation cover can be measured by testing the thermal resistance value of the upper insulation cover on the radial sides of the upper insulation cover, and the difference between the thermal resistance values of the upper insulation cover on the radial sides of the upper insulation cover can be obtained.
  • the radial deviation value of the thermal resistance value of the upper insulation cover can be obtained.
  • the axial deviation value of the thermal resistance value of the upper thermal insulation cover can be measured by testing the thermal resistance value of the upper thermal insulation cover on the two axial sides of the upper thermal insulation cover, and the difference between the thermal resistance values of the upper thermal insulation cover on the two axial sides of the upper thermal insulation cover can be obtained.
  • the axial deviation value of the thermal resistance value of the upper insulation cover can be obtained.
  • the same method can be used to obtain the thermal resistance value of the lower insulation cover and the deviation value of the thermal resistance value of the lower insulation cover, and the deviation value of the thermal resistance value of the lower insulation cover includes the radial deviation value of the thermal resistance value of the lower insulation cover and the lower insulation cover. Thermal resistance value axial deviation value.
  • the crystal growth channel shape factor refers to the value of the size of the crystal growth channel.
  • the crystal growth channel shape factor includes: the diameter and height of the crystal growth channel;
  • the crystal growth channel shape factor includes the length, height and width of the crystal growth channel; the crystal growth channel shape factor also affects the thermal field distribution of the crystal growth environment, thereby affecting the crystal growth performance.
  • the crystal growth observation hole shape factor also refers to the value of the shape and size of the crystal growth observation hole, and the shape of the crystal growth observation hole also affects the thermal field distribution of the crystal growth environment, thereby affecting the crystal growth performance.
  • the relative height of the crucible and the induction coil and the relative height of the heating ring and the induction coil will affect the thermal field distribution of the crystal growth environment.
  • control data includes the width of the die gap and the thickness of the die gap, and the molten high-resistance gallium oxide is mainly transported to the top of the die through the capillary siphon effect of the die gap, so that in the seed crystal It gradually grows into a high-resistance gallium oxide crystal. Therefore, the width and thickness of the die gap have a great influence on the quality of the high-resistance gallium oxide single crystal.
  • the thermal resistance value of the upper insulation cover, the thermal resistance value of the lower insulation cover, the deviation value of the thermal resistance value of the upper insulation cover, and the deviation value of the thermal resistance value of the lower insulation cover will change.
  • the time does not change, and these environmental data can be retested after a certain number of crystal growths.
  • the heating power refers to the heating power of the heating ring to the crucible
  • the cooling power refers to the power used to cool the environment in the crystal growth furnace with liquid
  • the atmosphere type refers to the gas type passed into the crystal growth furnace , including O 2 , Ar, N 2 , CO 2 , etc.
  • the cavity pressure refers to the pressure in the crystal growth furnace
  • the gas flow refers to the gas flow into the crystal growth furnace
  • the crystal rotation speed refers to the speed at which the seed rod drives the crystal to rotate during the crystal growth process.
  • the crystallization rate is It refers to the proportion of melting high-resistance gallium oxide raw materials to form crystals, and the crystal to crucible diameter ratio refers to the ratio of the generated crystal diameter to the crucible diameter. These parameters are the control parameters that affect the crystal prepared by the guided mode method.
  • step S210 the step of obtaining preprocessing preparation data according to the seed crystal data, environmental data and control data includes:
  • the first element in the preparation vector is one of the half width of the seed crystal diffraction peak, the deviation value of the seed crystal diffraction peak at half width, the thickness of the seed crystal and the width of the seed crystal;
  • the second element in the preparation vector are the thermal resistance value of the upper insulation cover, the deviation value of the thermal resistance value of the upper insulation cover, the shape factor of the crystal growth channel, the shape factor of the crystal growth observation hole, the thermal resistance value of the lower insulation cover, the deviation value of the thermal resistance value of the lower insulation cover, the crucible and the One of the relative height of the heating coil, the relative height of the heating ring and the heating coil, the width of the die gap, the thickness of the die gap, the concentration of doping elements and the type of doping elements, the doping element types include Fe, Ca , Zn, Co, Ti, Ni, Mg, Al and Cu;
  • the third element in the preparation vector is the heating power, cooling power, atmosphere type, cavity pressure, gas flow, and seed rod pulling speed.
  • the preparation vector (A, B, C) is determined.
  • the seed crystal data A is selected from: seed crystal diffraction peak half-height width A1, seed crystal diffraction peak half-height width deviation value A2, seed crystal thickness A3 and seed crystal width A4.
  • Environmental data B is selected from: upper thermal resistance value B1, upper thermal resistance deviation value B2, crystal growth channel shape factor B3, crystal growth observation hole shape factor B4, lower thermal resistance value B5, lower thermal shield Thermal resistance deviation value B6, relative height B7 between crucible and induction coil, relative height B8 between heating ring and induction coil, width B9 of die gap, thickness B10 of die gap, doping element concentration 11 and doping element type 12.
  • the control data C is selected from: heating power C1, cooling power C2, atmosphere type C3, cavity pressure C4, gas flow rate C5, and seed rod pulling speed C6. That is, in the preparation vector (A, B, C), A can be one of A1, A2, A3, and A4, and B can be B1, B2, B3, B4, B5, B6, B7, B8, B9, B10 , one of B11, B12, C can be one of C1, C2, C3, C4, C5, C6. Then 288 preparation vectors can be formed.
  • the preprocessed preparation data are as follows:
  • the predicted quality data includes: predicted crack data, predicted miscellaneous crystal data, predicted diffraction peak FWHM, predicted diffraction peak FWHM deviation, predicted shoulder symmetry of high resistance gallium oxide crystal, predicted high resistance The retraction degree of the left edge of the high resistance gallium oxide crystal, the retraction degree of the right edge of the predicted high resistance type gallium oxide crystal, the thickness of the high resistance type gallium oxide crystal, the thickness deviation of the predicted high resistance type gallium oxide crystal, the predicted radial deviation value of the resistivity and the Predicted resistivity axial deviation value.
  • the crack data refers to the crack grade data
  • the predicted crack data refers to the predicted crack grade data.
  • the cracks can be divided into multiple grades. For example, if the cracks are divided into 3 grades, the crack data are: 1, 2 and 3.
  • miscellaneous crystal data refers to the miscellaneous crystal grade data
  • predicted miscellaneous crystal data refers to the predicted miscellaneous crystal grade data.
  • miscellaneous crystals can be divided into multiple grades. are: 1, 2, and 3.
  • the predicted diffraction peak FWHM refers to the predicted diffraction peak FWHM
  • the predicted diffraction peak FWHM radial deviation value refers to the predicted difference in the radial diffraction peak FWHM
  • the predicted diffraction peak FWHM axial deviation Values refer to the predicted difference in the half-width of the diffraction peak in the axial direction.
  • the trained neural network model is obtained by using the following training steps:
  • training data for preparing a high-resistance gallium oxide single crystal by a guided mode method, and corresponding actual quality data, wherein the training data includes: seed crystal training data, environmental training data, and control training data.
  • the training data refers to the data of the high-resistance gallium oxide single crystal prepared by the guided mode method and used for training
  • the actual quality data refers to the data of the actual quality of the high-resistance gallium oxide single crystal prepared by the guided mode method.
  • a training set is formed by training data and actual quality data, and a preset neural network model is trained based on the training set to obtain a trained neural network model.
  • the high-resistance gallium oxide single crystal was prepared by the guided mode method, and the data of preparing the high-resistance gallium oxide single crystal was recorded as the training data.
  • the quality of the resistive gallium oxide single crystal is analyzed to obtain the actual quality data. In order to facilitate the training of the neural network model, as much data as possible can be collected to form a training set.
  • the training data is preprocessed to obtain the preprocessed training data.
  • the preprocessing process can refer to step S200.
  • the preprocessing training data is input into a preset neural network model, and prediction training generation quality data is obtained through the preset neural network model.
  • the quality data generated by the prediction training includes: prediction training to generate crack data, prediction training to generate miscellaneous crystal data, prediction training to generate diffraction peak full width at half maximum, prediction training to generate diffraction peak full width at half maximum radial deviation value, and prediction training to generate diffraction peak half width. Height and width axial deviation value.
  • the quality data and the actual quality data are generated according to the prediction training, the model parameters of the preset neural network model are modified, and the input of the preprocessing training data into the preset neural network model is continued. , obtaining through the preset neural network model the step of generating quality data by prediction training corresponding to the preprocessing training data (ie, step S03 ), until the preset training conditions are met, and a trained neural network model is obtained.
  • the quality data and the actual quality data are generated according to the prediction training, the model parameters of the preset neural network model are modified, and the input of the preprocessing training data into the preset neural network model is continued. , obtaining the preprocessed training data corresponding to the prediction training and generating quality data step by using the preset neural network model, until the preset training conditions are met, and a trained neural network model is obtained. That is, if the preset neural network model satisfies the preset training conditions, a trained neural network model is obtained. If the preset neural network model does not meet the preset training conditions, return to step S03 until the preset neural network model satisfies the preset training conditions, and a trained neural network model is obtained.
  • a preset loss function value of the neural network model is determined according to the quality data generated by the prediction training and the actual quality data, and the preset loss function value is determined according to the loss function value.
  • the model parameters of the neural network model are corrected.
  • a gradient-based method is used to modify the parameters of the preset neural network model, and after determining the loss function value of the preset neural network model, the preset neural network model is determined according to the loss function value.
  • the gradient of the parameters of the network model, the parameters of the preset neural network model, and the preset learning rate determine the modified parameters of the preset neural network model.
  • the preset training conditions include: the loss function value meets preset requirements and/or the preset number of training times of the neural network model reaches a preset number of times.
  • the 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 preset number of times is the preset maximum number of training times of the neural network model, for example, 4000 times.
  • the loss function of the preset neural network model includes: mean square error, root mean square error, mean absolute error, and the like.
  • the preset neural network model includes: a feature extraction module and a fully connected module.
  • the preset neural network model includes: a first convolution unit, a second convolution unit, a third convolution unit, a fourth convolution unit, and a fully connected unit.
  • the first convolution unit includes: two convolution layers and one pooling layer.
  • the second convolution unit, the third convolution unit, and the fourth convolution unit each include three convolution layers and one pooling layer.
  • the fully connected unit includes three fully connected layers.
  • the convolutional layer and the fully connected layer are responsible for mapping and transforming the input data. This process will use parameters such as weights and biases, and also need to use an activation function.
  • the pooling layer is a fixed function operation. Specifically, the convolutional layer plays the role of extracting features; the pooling layer performs a pooling operation on the input features to change their spatial size; and the fully connected layer connects all the data in the previous layer.
  • step S03 the preprocessing training data is input into a preset neural network model, and prediction training generation quality data corresponding to the preprocessing training data is obtained through the preset neural network model ,include:
  • S031 inputs the preprocessing training data into the feature extraction module, and obtains a feature vector corresponding to the preprocessing training data through the feature extraction module;
  • the preprocessing training data is input into a preset neural network model
  • the feature vector corresponding to the preprocessing training data is output through the feature extraction module in the preset neural network model
  • the The feature vector is input to the fully connected module
  • the prediction training generation quality data corresponding to the preprocessed training data output by the fully connected module is obtained.
  • this embodiment provides a high-resistance gallium oxide preparation method based on deep learning and guided mode method, as shown in FIG. 3 , the Preparation methods include:
  • the target quality data of the target high-resistance gallium oxide single crystal can be determined first, that is, the quality of the desired high-resistance gallium oxide single crystal can be determined. data.
  • the target quality data also includes: target crack data, target miscellaneous crystal data, target diffraction peak half-height width, target diffraction peak half-height width deviation value, target high-resistance gallium oxide crystal shoulder symmetry, target high-resistance type oxidation
  • target crack data target miscellaneous crystal data
  • target diffraction peak half-height width target diffraction peak half-height width deviation value
  • target high-resistance gallium oxide crystal shoulder symmetry target high-resistance type oxidation
  • target preparation data corresponding to the target high-resistance gallium oxide single crystal according to the target quality data and the trained neural network model, wherein the target preparation data includes: target seed crystal data and target environment data And target control data, the target environment data includes target doping element concentration, target doping element type.
  • the target preparation data corresponding to the target high-resistance gallium oxide single crystal is determined. It should be noted that since different preparation data can obtain the same quality data, the target preparation data corresponding to the target high-resistance gallium oxide single crystal is determined according to the target quality data and the trained neural network model. When the target preparation data is not unique, one target preparation data is determined according to the control difficulty of each data in the multiple target preparation data, so as to facilitate obtaining the target high-resistance type gallium oxide single crystal.
  • the step S20 determining the target preparation data corresponding to the target high-resistance gallium oxide single crystal according to the target quality data and the trained neural network model, includes:
  • the preparation data may be preset first, and the preset preparation data may be preprocessed to obtain the preprocessed preset preparation data.
  • the preset preparation data may be preprocessed to obtain the preprocessed preset preparation data.
  • step S200 Input the pre-processing preset preparation data into the trained neural network model, and obtain the predicted quality data that is modified by the pre-processing preset preparation data, and then prepare the preset preparation data according to the predicted quality data and the target quality data.
  • the data is corrected, and when the difference between the predicted quality data and the target quality data is smaller than a preset threshold, the corrected preset preparation data can be used as the target preparation data.
  • the target high-resistance gallium oxide single crystal can be prepared according to the target preparation data by the guided mode method.
  • the present invention provides a high-resistance gallium oxide preparation system based on deep learning and guided mode method.
  • the system can be computer equipment, and the internal structure is shown in FIG. 4 .
  • the system includes a processor, memory, a network interface, a display screen, and an input device connected by a system bus. Among them, the processor of the system is used to provide computing and control capabilities.
  • the memory of the system includes a non-volatile storage medium and an internal memory.
  • the nonvolatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the network interface of the system is used to communicate with external terminals through a network connection.
  • the display screen of the system can be a liquid crystal display screen or an electronic ink display screen
  • the input device of the system can be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the system shell, or An external keyboard, trackpad, or mouse, etc.
  • FIG. 4 is only a partial structure related to the solution of the present application, and does not constitute a limitation on the system to which the solution of the present application is applied. show more or fewer components, or combine certain components, or have a different arrangement of components.
  • a high-resistance gallium oxide preparation system based on deep learning and guided mode method including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program Carry out the steps of the prediction method, or the steps of the preparation method.

Landscapes

  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Organic Chemistry (AREA)
  • Metallurgy (AREA)
  • Materials Engineering (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Inorganic Chemistry (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Biochemistry (AREA)
  • Medicinal Chemistry (AREA)
  • Immunology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Analytical Chemistry (AREA)
  • Food Science & Technology (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Crystals, And After-Treatments Of Crystals (AREA)

Abstract

本发明公开了一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统，质量预测方法包括步骤：获取导模法制备高阻型氧化镓单晶的制备数据，所述制备数据包括籽晶数据、环境数据以及控制数据，所述控制数据包括掺杂元素浓度、掺杂元素类型；对所述制备数据进行预处理，得到预处理制备数据；将所述预处理制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到所述高阻型氧化镓单晶对应的预测质量数据，所述预测质量数据包括预测电阻率。本发明可通过训练好的神经网络模型对高阻型氧化镓单晶的质量进行预测，因此可以调整制备数据得到需要的高阻型氧化镓单晶的性能，使得高阻型氧化镓单晶的性能得到优化。

Description

一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统 技术领域
本发明涉及高阻型氧化镓晶体制备领域，特别涉及一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统。
背景技术
β-Ga 2O 3(高阻型氧化镓)是一种直接带隙宽禁带半导体材料，禁带宽度约为4.8～4.9eV。它具有禁带宽度大、饱和电子漂移速度快、热导率高、击穿场强高、化学性质稳定等诸多优点，在高温、高频、大功率电力电子器件领域有着广泛的应用前景。此外还可用于LED芯片，日盲紫外探测、各种传感器元件及摄像元件等。
目前，批量制备大尺寸高阻型氧化镓晶体主要采用导模法制备技术。导模法是一种成熟的单晶制备技术，其具有异形晶体生长、生长速度快、生长成本低的优点。导模法是在导模法基础上改良获得的一种晶体生长方法，晶体生长设备及生长工艺探索相对更加复杂。导模法需要在坩埚内放置模具，晶体生长界面位于模具上表面，高温下由于表面张力的作用，熔体沿模具中的毛细管上升到模具上表面。熔体沿毛细管上升高度H由公式
Figure PCTCN2021076071-appb-000001
决定，其中，γ为熔体表面张力，θ为熔体与毛细管之间的接触角，ρ为熔体密度，g为重力加速度，r为毛细管半径。
与导模法相比，采用导模法制备高阻型氧化镓晶体具有以下优点：
1、模具毛细管中熔体对流较弱，熔体中杂质离子从毛细管上升至固液界面后较难重新返回坩埚，晶体中的杂质离子分凝系数一般接近于1，晶体上下杂质分布一致性好；
2、晶体生长固液界面位于模具上方，界面形状可以通过模具表面几何形状调控，固液界面位于温场中的位置始终不变，且不受坩埚中熔体扰动影响，因此导模法中晶体生长固液界面更加稳定；
3、导模法晶体生长速度较快，有利于降低能耗，可以生长异形晶体，降低晶体加工成本及加工损耗。
然而，在采用导模法制备高阻型氧化镓晶体的过程中，模具口附近的温度场分布、籽晶的选择、籽晶杆提拉速度、晶体生长气氛环境以及加热功率、冷却功率等参数的变化均会对制得的高阻型氧化镓晶体质量产生较大影响。现有导模法制备高阻型氧化镓晶 体的过程均是依赖操作员的经验来设置参数，其重复性较差，从而导致制得的高阻型氧化镓晶体产品稳定性较差。
因此，现有技术还有待于改进和发展。
发明内容
本发明要解决的技术问题在于，针对现有技术的不足，提供一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统，旨在解决现有导模法制备高阻型氧化镓晶体的过程均是依赖操作员的经验来设置参数，其重复性较差，导致制得的高阻型氧化镓晶体质量以及稳定性较差的问题。
为了解决上述技术问题，本发明所采用的技术方案如下：
一种基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，包括步骤：
获取导模法制备高阻型氧化镓单晶的制备数据，所述制备数据包括籽晶数据、环境数据以及控制数据，所述控制数据包括掺杂元素浓度、掺杂元素类型；
对所述制备数据进行预处理，得到预处理制备数据；
将所述预处理制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到所述高阻型氧化镓单晶对应的预测质量数据，所述预测质量数据包括预测电阻率。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，对所述制备数据进行预处理，得到预处理制备数据的步骤包括：
根据所述籽晶数据、环境数据以及控制数据，得到预处理制备数据，所述预处理制备数据为由所述籽晶数据、环境数据以及控制数据形成的矩阵。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，所述籽晶数据包括：籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度；
所述环境数据还包括：上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度；
所述控制数据包括：加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，根据所述籽晶数据、环境数据以及控制数据，得到预处理制备数据的步骤包括：
根据所述籽晶数据、环境数据以及控制数据，确定制备向量；其中，所述制备向量中第一元素为所述籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度中的一个；所述制备向量中第二元素为所述上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度、掺杂元素浓度以及掺杂元素类型中的一个，所述掺杂元素类型包括Fe、Ca、Zn、Co、Ti、Ni、Mg、Al和Cu；所述制备向量中第三元素为所述加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度中的一个；
根据所述制备向量，确定所述预处理制备数据。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，所述预测质量数据还包括：预测裂纹数据、预测杂晶数据、预测衍射峰半高宽、预测衍射峰半高宽偏差值、预测高阻型氧化镓晶体放肩对称性、预测高阻型氧化镓晶体左边沿收放度、预测高阻型氧化镓晶体右边沿收放度、预测高阻型氧化镓晶体厚度、预测高阻型氧化镓晶体厚度偏差、预测电阻率径向偏差值以及预测电阻率轴向偏差值。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，所述训练好的神经网络模型采用如下训练步骤训练得到：
获取导模法制备高阻型氧化镓单晶的训练数据，以及对应的实际质量数据，其中，所述训练数据包括：籽晶训练数据、环境训练数据以及控制训练数据；
对所述训练数据进行预处理，得到预处理训练数据；
将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据；
根据所述预测训练生成质量数据以及所述实际质量数据对所述预设的神经网络模型的模型参数进行调整修正，得到训练好的神经网络模型。
所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其中，所述预设的神经网络模型包括：特征提取模块和全连接模块，
将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据的步骤包括：
将所述预处理训练数据输入所述特征提取模块，通过所述特征提取模块得到与所述预处理训练数据对应的特征向量；
将所述特征向量输入到所述全连接模块，通过所述全连接模块得到与所述预处理训 练数据对应的预测训练生成质量数据。
一种基于深度学习和导模法的高阻型氧化镓制备方法，其中，所述制备方法包括步骤：
获取目标高阻型氧化镓单晶的目标质量数据，所述目标质量数据包括目标电阻率；
根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，其中，所述目标制备数据包括：目标籽晶数据、目标环境数据以及目标控制数据，所述目标环境数据包括目标掺杂元素浓度，目标掺杂元素类型；
基于导模法，根据所述目标制备数据制备得到目标高阻型氧化镓单晶。
所述基于深度学习和导模法的高阻型氧化镓制备方法，其中，根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，包括步骤：
获取预设制备数据，对所述预设制备数据进行预处理，得到预处理预设制备数据；
将所述预处理预设制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到与所述预处理预设制备数据对应的预测质量数据；
根据所述预测质量数据以及所述目标质量数据，对所述预设制备数据进行修正，以得到所述目标高阻型氧化镓单晶对应的目标制备数据。
一种基于深度学习和导模法的高阻型氧化镓制备系统，其中，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现本发明任一项所述的预测方法的步骤，或本发明中任一项所述的制备方法的步骤。
有益效果：本发明提出了一种基于深度学习和导模法的高阻型氧化镓质量预测方法以及制备方法，先对导模法制备高阻型氧化镓单晶的制备数据进行预处理，得到预处理制备数据，然后将所述预处理制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到所述高阻型氧化镓单晶对应的预测质量数据。本发明可通过训练好的神经网络模型对高阻型氧化镓单晶的质量进行预测，因此可以调整制备数据得到需要的高阻型氧化镓单晶的性能，使得高阻型氧化镓单晶的性能得到优化。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中附图仅仅是本发明中记载的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下， 还可以根据这些附图获得其他的附图。
图1为本发明提供的一种导模法制备高阻型氧化镓晶体的晶体生长炉的结构示意图。
图2为本发明提供的一种基于深度学习和导模法的高阻型氧化镓质量预测方法较佳实施例的流程图。
图3为本发明提供的一种基于深度学习和导模法的高阻型氧化镓制备方法较佳实施例的流程图。
图4为本发明提供的一种基于深度学习和导模法的高阻型氧化镓制备系统的内部结构图。
具体实施方式
本发明提供一种基于深度学习和导模法的高阻型氧化镓晶体制备方法及系统，为使本发明的目的、技术方案及效果更加清楚、明确，以下参照附图并举实施例对本发明进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。
本技术领域技术人员可以理解，除非特意声明，这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是，本发明的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作、元件和/或组件，但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作、元件、组件和/或它们的组。应该理解，当我们称元件被“连接”或“耦接”到另一元件时，它可以直接连接或耦接到其他元件，或者也可以存在中间元件。此外，这里使用的“连接”或“耦接”可以包括无线连接或无线耦接。这里使用的措辞“和/或”包括一个或更多个相关联的列出项的全部或任一单元和全部组合。
本技术领域技术人员可以理解，除非另外定义，这里使用的所有术语(包括技术术语和科学术语)，具有与本发明所属领域中的普通技术人员的一般理解相同的意义。还应该理解的是，诸如通用字典中定义的那些术语，应该被理解为具有与现有技术的上下文中的意义一致的意义，并且除非像这里一样被特定定义，否则不会用理想化或过于正式的含义来解释。
下面结合附图，通过对实施例的描述，对发明内容作进一步说明。
本发明提供了一种导模法生长高阻型氧化镓晶体的晶体生长炉，如图1所示，其包括下保温罩11，所述下保温罩11内设置有坩埚12，所述坩埚内12设置有形成毛细虹 吸效应的模具13、所述坩埚12的外部设置有加热环14，；所述下保温罩的上方设置有所述上保温罩15，所述上保温罩15内部设置有晶体生长通道16，所述上保温罩的侧边设置有晶体生长观察孔17，所述下保温罩10的外围还设置有感应线圈18。
基于所述晶体生长炉可制备高阻型氧化镓晶体，首先，安装好用于制备高阻型氧化镓晶体的晶体生长炉，其中包括上保温罩材料选择、上保温罩的轴向开孔大小、开口窗形状及大小，下保温罩材料选择、坩埚相对感应线圈的高度位置、加热环与感应线圈的相对高度等，这些都会影响装置内部的热场分布，从而影响高阻型氧化镓晶体的产品性能。
将特定取向的β-Ga 2O 3籽晶放入籽晶夹具内并捆绑固定，籽晶的取向可以是[010]、[001]方向等；
首先，依次开启机械泵、扩散泵对设备内进行抽真空，当抽至预定真空度时，关闭真空设备，按照混合气体积比缓慢充气至设备内部；
然后，通过设置加热功率，通过所述发热体对坩埚进行加热，使放置在坩埚内的氧化镓原料以及掺杂元素材料完全融化，熔融氧化镓及掺杂元素材料通过毛细虹吸作用被输运至模具顶部并在顶部展开直至全部覆盖；随后，缓慢下降籽晶杆，使籽晶距离铱制模具顶端上方3-5mm位置进行籽晶预热，5-10分钟后开始接种；待籽晶与熔体充分熔接后进行引晶缩颈操作，避免籽晶的原有缺陷延伸至晶体内部，保证晶体质量，接着，进行扩肩生长，使晶体横向扩满至整个模具；再接下来进行等径生长；在晶体生长结束后，缓慢降至室温，取出晶体，即获得高阻型氧化镓晶体。在制备高阻型氧化镓晶体的过程中，加热功率、冷却功率、气氛类型、腔内压力、气体流量以及籽晶杆提拉速度，均会影响高阻型氧化镓晶体的产品性能。
基于此，本实施例提供了一种基于深度学习和导模法的高阻型氧化镓质量预测方法，如图2所示，其包括以下步骤：
S100、获取导模法制备高阻型氧化镓单晶的制备数据，所述制备数据包括籽晶数据、环境数据以及控制数据，所述控制数据包括掺杂元素浓度、掺杂元素类型。
具体来讲，所述制备数据是指采用导模法制备高阻型氧化镓单晶的数据，所述制备数据是可以根据需要进行配置的数据，例如，需要预测某一制备数据下得到的高阻型氧化镓单晶的性能时，只需要确定该制备数据，并对制备数据进行预处理得到预处理制备数据，再将所述预处理制备数据输入到训练好的神经网络模型，通过训练好的神经网络模型得到预测质量数据，也就是说，不需要进行实验，确定好制备数据后，就可以通过 训练好的神经网络模型预测高阻型氧化镓单晶的质量数据。
本实施例中，所述制备数据包括：籽晶数据、环境数据以及控制数据，所述籽晶数据是指采用导模法制备高阻型氧化镓单晶的过程中所采用的籽晶的数据，所述环境数据是指采用导模法制备高阻型氧化镓单晶的过程中晶体所处环境的数据，所述控制数据是指采用导模法制备高阻型氧化镓单晶的过程中控制晶体生长的数据。所述掺杂元素浓度是指高阻型氧化镓中掺杂元素的浓度，所述掺杂元素类型包括：Fe、Ca、Zn、Co、Ti、Ni、Mg、Al、Cu等。
S200、对所述制备数据进行预处理，得到预处理制备数据。
具体来讲，在得到制备数据后，先对所述制备数据进行预处理，得到预处理制备数据，从而可以将所述预处理制备数据输入训练好的神经网络模型中，以便通过训练好的神经网络模型对预处理制备数据进行处理。
在本申请实施例的一种实施方式中，步骤S200、对所述制备数据进行预处理，得到预处理制备数据的步骤包括：
S210、根据所述籽晶数据、环境数据以及控制数据，得到预处理制备数据，所述预处理制备数据为由所述籽晶数据、环境数据以及控制数据形成的矩阵。
具体来讲，在得到制备数据后，先对所述制备数据进行预处理，得到预处理制备数据。由于制备数据中各子数据(如籽晶数据、环境数据以及控制数据)之间是会相互影响的，但是目前无法明确各子数据之间相互影响的程度有多少，因此，需要对制备数据进行预处理，将制备数据中各子数据重新排列组合，形成预处理制备数据。
在本申请的另一实施方式中，所述籽晶数据包括：籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度；所述环境数据还包括：上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度；所述控制数据包括：加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度。
具体的，所述籽晶衍射峰半高宽可以采用X射线衍射仪对籽晶进行测试，籽晶衍射峰半高宽偏差值包括籽晶衍射峰半高宽径向偏差值以及籽晶衍射峰半高宽轴向偏差值，所述径向为位于水平面上的方向，轴向为垂直于水平面的方向，即竖直方向的轴线。所述籽晶衍射峰半高宽径向偏差值可以通过对籽晶径向测试籽晶衍射峰半高宽，并求得籽晶径向上籽晶衍射峰半高宽之间的差值，即可得到籽晶衍射峰半高宽径向偏差值。所述 籽晶衍射峰半高宽轴向偏差值可以通过对籽晶轴向测试籽晶衍射峰半高宽，并求得籽晶轴向上籽晶衍射峰半高宽之间的差值，即可得到籽晶衍射峰半高宽轴向偏差值。所述籽晶的厚度和宽度可直接测量得到。
如上所述，在采用图1所示的晶体生长炉制备高阻型氧化镓单晶时，所述上保温罩与下保温罩均为高阻型氧化镓单晶的生长提供稳定的热场。所述上保温罩热阻值是指单位时间内单位热量通过上保温罩时，保温罩内外两侧的温度差。上保温罩热阻值越大，表明上保温罩抵抗传热的能力越强，上保温罩的保温效果越好。
所述上保温罩热阻值偏差值包括上保温罩热阻值径向偏差值和上保温罩热阻值轴向偏差值。上保温罩热阻值径向偏差值可以通过对上保温罩径向两侧测试上保温罩热阻值，并求得上保温罩径向两侧上保温罩热阻值之间的差值，即可得到上保温罩热阻值径向偏差值。上保温罩热阻值轴向偏差值可以通过对上保温罩轴向两侧测试上保温罩热阻值，并求得上保温罩轴向两侧上保温罩热阻值之间的差值，即可得到上保温罩热阻值轴向偏差值。
相同地，采用同样的方法可以得到下保温罩热阻值以及下保温罩热阻值偏差值，所述下保温罩热阻值偏差值包括下保温罩热阻值径向偏差值和下保温罩热阻值轴向偏差值。
所述晶体生长通道形状因子是指晶体生长通道形状尺寸的值，例如，所述晶体生长通道为圆柱形时，则晶体生长通道形状因子包括：晶体生长通道的直径和高度；当所述晶体生长通道为立方体时，则晶体生长通道形状因子包括晶体生长通道的长度、高度和宽度；所述晶体生长通道形状因子也会影响晶体生长环境的热场分布，从而影响晶体生长性能。
所述晶体生长观察孔形状因子同样是指晶体生长观察孔形状尺寸的值，所述晶体生长观察孔的形状同样也会影响晶体生长环境的热场分布，从而影响晶体生长性能。
由于所述坩埚是通过感应线圈和加热环对其进行加热的，因此所述坩埚与感应线圈的相对高度、加热环与感应线圈的相对高度会影响晶体生长环境的热场分布。
本实施例中，所述控制数据包括模具口缝隙的宽度以及模具口缝隙的厚度，熔融高阻型氧化镓主要通过所述模具口缝隙的毛细虹吸作用被输运至模具顶部，从而在籽晶上逐渐生长成高阻型氧化镓晶体。因此，所述模具口缝隙的宽度以及厚度对高阻型氧化镓单晶的质量影响较大。
随着生长高阻型氧化镓晶体的装置的使用，上保温罩热阻值、下保温罩热阻值、上 保温罩热阻值偏差值、下保温罩热阻值偏差值会改变，但是短时间内不会改变，可以在进行一定次数的晶体生长后，再重新测试这些环境数据。
所述加热功率是指加热环对坩埚的加热功率，所述冷却功率是指采用液体对晶体生长炉内环境进行冷却的功率，所述气氛类型是指通入所述晶体生长炉内的气体类型，包括O 2，Ar，N 2，CO 2等，所述腔体压力是指所述晶体生长炉内的压力，所述气体流量是指通入所述晶体生长炉内的气体流量，所述籽晶杆提拉速度是指在晶体生长过程中籽晶杆向上提拉的速度，所述晶体转速是指在晶体生长过程中籽晶杆带动所述晶体旋转的速度，所述析晶率是指熔融高阻型氧化镓原料形成晶体的比例，所述晶体与坩埚直径比是指生成的晶体直径与坩埚直径的比例。这些参数都是影响导模法制备晶体的控制参数。
在本实施例的一种实施方式中，步骤S210、根据所述籽晶数据、环境数据以及控制数据，得到预处理制备数据的步骤包括：
S211、所述制备向量中第一元素为所述籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度中的一个；所述制备向量中第二元素为所述上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度、掺杂元素浓度以及掺杂元素类型中的一个，所述掺杂元素类型包括Fe、Ca、Zn、Co、Ti、Ni、Mg、Al和Cu；所述制备向量中第三元素为所述加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度中的一个；
S212、根据所述制备向量，确定所述预处理制备数据。
具体地，根据籽晶数据A、环境数据B以及控制数据C，确定制备向量(A，B，C)。籽晶数据A选自：籽晶衍射峰半高宽A1、籽晶衍射峰半高宽偏差值A2、籽晶厚度A3以及籽晶宽度A4。环境数据B选自：上保温罩热阻值B1、上保温罩热阻值偏差值B2、晶体生长通道形状因子B3、晶体生长观察孔形状因子B4、下保温罩热阻值B5、下保温罩热阻值偏差值B6，坩埚与感应线圈的相对高度B7，加热环与感应线圈的相对高度B8，模具口缝隙的宽度B9、模具口缝隙的厚度B10、掺杂元素浓度11以及掺杂元素类型12。控制数据C选自：加热功率C1、冷却功率C2、气氛类型C3、腔体压力C4、气体流量C5、籽晶杆提拉速度C6。也就是说，制备向量(A，B，C)中A可以是A1、A2、A3、A4中的一个，B可以是B1、B2、B3、B4、B5、B6、B7、B8、B9、B10、B11、B12中的一个，C可以是C1、C2、C3、C4、C5、C6中的一个。则可以形成288个制备向 量。
将所有制备向量按照序号排列形成矩阵，则得到了预处理的制备数据。
具体地，预处理的制备数据如下：
Figure PCTCN2021076071-appb-000002
当然，还采用其他的排列形式，得到预处理的制备数据。
S300、将所述预处理的制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到所述高阻型氧化镓单晶对应的预测质量数据。
具体地，所述预测质量数据包括：预测裂纹数据、预测杂晶数据、预测衍射峰半高宽、预测衍射峰半高宽偏差值、预测高阻型氧化镓晶体放肩对称性、预测高阻型氧化镓晶体左边沿收放度、预测高阻型氧化镓晶体右边沿收放度、预测高阻型氧化镓晶体厚度、预测高阻型氧化镓晶体厚度偏差、预测电阻率径向偏差值以及预测电阻率轴向偏差值。其中，裂纹数据是指裂纹等级数据，预测裂纹数据是指预测的裂纹等级数据，例如，可以将裂纹分为多个等级，举例说明，裂纹分为3级，则裂纹数据分别为：1、2以及3。
杂晶数据是指杂晶等级数据，预测杂晶数据是指预测的杂晶等级数据，例如，可以将杂晶分为多个等级，举例说明，杂晶分为3级，则杂晶数据分别为：1、2以及3。
预测衍射峰半高宽是指预测的衍射峰半高宽，预测衍射峰半高宽径向偏差值是指在径向衍射峰半高宽的预测差值，预测衍射峰半高宽轴向偏差值是指在轴向衍射峰半高宽的预测差值。
在一些实施方式中，所述训练好的神经网络模型采用如下训练步骤训练得到：
S01、获取导模法制备高阻型氧化镓单晶的训练数据，以及对应的实际质量数据，其中，所述训练数据包括：籽晶训练数据、环境训练数据以及控制训练数据。
具体地，训练数据是指采用导模法制备高阻型氧化镓单晶并用于训练的数据，实际质量数据是指采用导模法制备得到的高阻型氧化镓单晶的实际质量的数据。通过训练数据、实际质量数据形成训练集，基于该训练集训练预设的神经网络模型，得到训练好的神经网络模型。
在采集数据得到训练集时，采用导模法制备高阻型氧化镓单晶，并记录制备高阻型氧化镓单晶的数据作为训练数据，在得到高阻型氧化镓单晶后，对高阻型氧化镓单晶的质量进行分析得到实际质量数据。为了便于神经网络模型的训练，可以采集尽量多的数 据形成训练集。
S02、对所述训练数据进行预处理，得到预处理训练数据。
具体地，在得到训练数据后，对训练数据进行预处理，得到预处理训练数据。预处理过程可以参考步骤S200。
S03、将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据。
具体地，将所述预处理训练数据输入预设的神经网络模型，通过预设的神经网络模型得到预测训练生成质量数据。所述预测训练生成质量数据包括：预测训练生成裂纹数据、预测训练生成杂晶数据、预测训练生成衍射峰半高宽、预测训练生成衍射峰半高宽径向偏差值以及预测训练生成衍射峰半高宽轴向偏差值。
S04、根据所述预测训练生成质量数据以及所述实际质量数据对所述预设的神经网络模型的模型参数进行调整修正，得到训练好的神经网络模型。
具体地，根据所述预测训练生成质量数据以及所述实际质量数据，对所述预设的神经网络模型的模型参数进行修正，并继续执行将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到所述预处理训练数据对应的预测训练生成质量数据的步骤(即步骤S03)，直至满足预设训练条件，得到训练好的神经网络模型。
具体地，根据所述预测训练生成质量数据以及所述实际质量数据，对所述预设的神经网络模型的模型参数进行修正，并继续执行将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到所述预处理的训练数据对应的预测训练生成质量数据的步骤，直至满足预设训练条件，得到训练好的神经网络模型。也就是说，若所述预设的神经网络模型满足预设训练条件，则得到训练好的神经网络模型。若所述预设的神经网络模型不满足预设训练条件，则返回步骤S03，直至所述预设的神经网络模型满足预设训练条件，得到训练好的神经网络模型。
在本发明实施例的一种实现方式中，根据所述预测训练生成质量数据以及所述实际质量数据确定预设的神经网络模型的损失函数值，根据所述损失函数值对所述预设的神经网络模型的模型参数进行修正。具体地，采用基于梯度的方法对所述预设的神经网络模型的参数进行修正，确定所述预设的神经网络模型的损失函数值后，根据所述损失函数值对所述预设的神经网络模型的参数的梯度、所述预设的神经网络模型的参数以及预设学习率，确定所述预设的神经网络模型的修正的参数。
所述预设训练条件包括：损失函数值满足预设要求和/或所述预设的神经网络模型的 训练次数达到预设次数。
所述预设要求根据所述预设的神经网络模型的精度和效率确定，例如，所述预设的神经网络模型的损失函数值达到最小值或者不再变化。所述预设次数为所述预设的神经网络模型的最大训练次数，例如，4000次等。
预设的神经网络模型的损失函数包括：均方误差、均方根误差、平均绝对误差等。
在本申请实施例的一种实现方式中，所述预设的神经网络模型包括：特征提取模块和全连接模块。
举例说明，预设的神经网络模型包括：第一卷积单元、第二卷积单元、第三卷积单元、第四卷积单元以及全连接单元。具体地，第一卷积单元包括：两个卷积层和一个池化层。第二卷积单元、第三卷积单元以及第四卷积单元均包括三个卷积层和一个池化层。全连接单元包括三个全连接层。
卷积层和全连接层负责对输入数据进行映射变换，这个过程会用到权值和偏置等参数，也需要使用激活函数。池化层是一个固定不变的函数操作。具体地，卷积层起到提取特征的作用；池化层对输入特征进行池化操作，改变其空间尺寸；而全连接层是对前一次层中所有数据进行全部连接的。
在一些实施方式中，所述步骤S03、将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据，包括：
S031将所述预处理训练数据输入所述特征提取模块，通过所述特征提取模块得到与所述预处理训练数据对应的特征向量；
S032、将所述特征向量输入到所述全连接模块，通过所述全连接模块得到与所述预处理训练数据对应的预测训练生成质量数据。
具体地，将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型中的所述特征提取模块输出所述预处理训练数据对应的特征向量，并将所述特征向量输入所述全连接模块，得到所述全连接模块输出的所述预处理训练数据对应的预测训练生成质量数据。
基于上述基于深度学习和导模法的高阻型氧化镓质量预测方法，本实施例提供了一种基于深度学习和导模法的高阻型氧化镓制备方法，如图3所示，所述制备方法包括：
S10、获取目标高阻型氧化镓单晶的目标质量数据，所述目标质量数据包括目标电阻率。
具体地，如果需要得到目标高阻型氧化镓单晶时，可以先确定目标高阻型氧化镓单晶的目标质量数据，也就是说，确定想要得到的高阻型氧化镓单晶的质量数据。所述目标质量数据还包括：目标裂纹数据、目标杂晶数据、目标衍射峰半高宽、目标衍射峰半高宽偏差值、目标高阻型氧化镓晶体放肩对称性、目标高阻型氧化镓晶体左边沿收放度、目标高阻型氧化镓晶体右边沿收放度、目标高阻型氧化镓晶体厚度、目标高阻型氧化镓晶体厚度偏差、目标电阻率径向偏差值以及目标电阻率轴向偏差值。
S20、根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，其中，所述目标制备数据包括：目标籽晶数据、目标环境数据以及目标控制数据，所述目标环境数据包括目标掺杂元素浓度，目标掺杂元素类型。
具体地，根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据。需要说明的时，由于不同的制备数据可以得到相同的质量数据，因此，在根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据时，目标制备数据并不是唯一的，根据多个目标制备数据中各数据的控制难易程度确定一个目标制备数据，从而便于得到目标高阻型氧化镓单晶。
在一些实施方式中，所述步骤S20、根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，包括：
S21、获取预设制备数据，对所述预设制备数据进行预处理，得到预处理预设制备数据；
S22、将所述预处理预设制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到与所述预处理预设制备数据对应的预测质量数据；
S23、根据所述预测质量数据以及所述目标质量数据，对所述预设制备数据进行修正，以得到所述目标高阻型氧化镓单晶对应的目标制备数据。
具体地，可以先预设制备数据，并对预设制备数据进行预处理，得到预处理预设制备数据，具体预处理过程可以参考步骤S200。将预处理预设制备数据输入训练好的神经网络模型，可以得到对预处理预设制备数据进行修正的预测质量数据，然后根据所述预测质量数据以及所述目标质量数据对所述预设制备数据进行修正，当所述预测质量数据与所述目标质量数据之间的差值小于预设阈值时，则可以将该修正后的预设制备数据作为目标制备数据。
S30、基于导模法，根据所述目标制备数据制备得到目标高阻型氧化镓单晶。
具体地，在得到目标制备数据后，则可以导模法，根据所述目标制备数据制备得到目标高阻型氧化镓单晶。
基于上述预测方法或上述制备方法，本发明提供了一种基于深度学习和导模法的高阻型氧化镓制备系统，该系统可以是计算机设备，内部结构如图4所示。该系统包括通过系统总线连接的处理器、存储器、网络接口、显示屏和输入装置。其中，该系统的处理器用于提供计算和控制能力。该系统的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该系统的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现基于深度学习和导模法的高阻型氧化镓的预测方法或基于深度学习和导模法的高阻型氧化镓的制备方法。该系统的显示屏可以是液晶显示屏或者电子墨水显示屏，该系统的输入装置可以是显示屏上覆盖的触摸层，也可以是系统外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图4所示的仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的系统的限定，具体的系统可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种基于深度学习和导模法的高阻型氧化镓制备系统，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现所述的预测方法的步骤，或所述的制备方法的步骤。
最后应说明的是：以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (10)

  1. 一种基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，包括步骤：
    获取导模法制备高阻型氧化镓单晶的制备数据，所述制备数据包括籽晶数据、环境数据以及控制数据，所述控制数据包括掺杂元素浓度、掺杂元素类型；
    对所述制备数据进行预处理，得到预处理制备数据；
    将所述预处理制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到所述高阻型氧化镓单晶对应的预测质量数据，所述预测质量数据包括预测电阻率。
  2. 根据权利要求1所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，对所述制备数据进行预处理，得到预处理制备数据的步骤包括：
    根据所述籽晶数据、所述环境数据以及所述控制数据，得到预处理制备数据，所述预处理制备数据为由所述籽晶数据、所述环境数据以及所述控制数据形成的矩阵。
  3. 根据权利要求2所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，所述籽晶数据包括：籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度；
    所述环境数据还包括：上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度；
    所述控制数据包括：加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度。
  4. 根据权利要求3所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，根据所述籽晶数据、所述环境数据以及所述控制数据，得到预处理制备数据的步骤包括：
    根据所述籽晶数据、所述环境数据以及所述控制数据，确定制备向量；其中，所述制备向量中第一元素为所述籽晶衍射峰半高宽、籽晶衍射峰半高宽偏差值、籽晶厚度以及籽晶宽度中的一个；所述制备向量中第二元素为所述上保温罩热阻值、上保温罩热阻值偏差值、晶体生长通道形状因子、晶体生长观察孔形状因子、下保温罩热阻值、下保温罩热阻值偏差值、坩埚与加热线圈的相对高度、加热环与加热线圈的相对高度、模具口缝隙的宽度、模具口缝隙的厚度、掺杂元素浓度以及掺杂元素类型中的一个，所述掺杂元素类型包括Fe、Ca、Zn、Co、Ti、Ni、Mg、Al和Cu；所述制备向量中第三元素 为所述加热功率、冷却功率、气氛类型、腔体压力、气体流量、籽晶杆提拉速度中的一个；
    根据所述制备向量，确定所述预处理制备数据。
  5. 根据权利要求1所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，所述预测质量数据还包括：预测裂纹数据、预测杂晶数据、预测衍射峰半高宽、预测衍射峰半高宽偏差值、预测高阻型氧化镓晶体放肩对称性、预测高阻型氧化镓晶体左边沿收放度、预测高阻型氧化镓晶体右边沿收放度、预测高阻型氧化镓晶体厚度、预测高阻型氧化镓晶体厚度偏差、预测电阻率径向偏差值以及预测电阻率轴向偏差值。
  6. 根据权利要求1所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，所述训练好的神经网络模型采用如下训练步骤训练得到：
    获取导模法制备高阻型氧化镓单晶的训练数据，以及对应的实际质量数据，其中，所述训练数据包括：籽晶训练数据、环境训练数据以及控制训练数据；
    对所述训练数据进行预处理，得到预处理训练数据；
    将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据；
    根据所述预测训练生成质量数据以及所述实际质量数据对所述预设的神经网络模型的模型参数进行调整修正，得到训练好的神经网络模型。
  7. 根据权利要求6所述基于深度学习和导模法的高阻型氧化镓质量预测方法，其特征在于，所述预设的神经网络模型包括：特征提取模块和全连接模块，
    将所述预处理训练数据输入预设的神经网络模型，通过所述预设的神经网络模型得到与所述预处理训练数据对应的预测训练生成质量数据的步骤包括：
    将所述预处理训练数据输入所述特征提取模块，通过所述特征提取模块得到与所述预处理训练数据对应的特征向量；
    将所述特征向量输入到所述全连接模块，通过所述全连接模块得到与所述预处理训练数据对应的预测训练生成质量数据。
  8. 一种基于深度学习和导模法的高阻型氧化镓制备方法，其特征在于，所述制备方法包括步骤：
    获取目标高阻型氧化镓单晶的目标质量数据，所述目标质量数据包括目标电阻率；
    根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，其中，所述目标制备数据包括：目标籽晶数据、目标环境数 据以及目标控制数据，所述目标环境数据包括目标掺杂元素浓度，目标掺杂元素类型；
    基于导模法，根据所述目标制备数据制备得到目标高阻型氧化镓单晶。
  9. 根据权利要求8所述基于深度学习和导模法的高阻型氧化镓制备方法，其特征在于，根据所述目标质量数据以及训练好的神经网络模型，确定所述目标高阻型氧化镓单晶对应的目标制备数据，包括步骤：
    获取预设制备数据，对所述预设制备数据进行预处理，得到预处理预设制备数据；
    将所述预处理预设制备数据输入训练好的神经网络模型，通过所述训练好的神经网络模型得到与所述预处理预设制备数据对应的预测质量数据；
    根据所述预测质量数据以及所述目标质量数据，对所述预设制备数据进行修正，以得到所述目标高阻型氧化镓单晶对应的目标制备数据。
  10. 一种基于深度学习和导模法的高阻型氧化镓制备系统，其特征在于，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至7中任一项所述的预测方法的步骤，或权利要求8至9中任一项所述的制备方法的步骤。
PCT/CN2021/076071 2020-12-31 2021-02-08 一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统 WO2022141765A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2022520490A JP7478819B2 (ja) 2020-12-31 2021-02-08 ディープラーニング及びｅｆｇテクニック方法による高抵抗型酸化ガリウムの品質予測方法、製造方法及びそのシステム
US17/760,964 US12024791B2 (en) 2020-12-31 2021-02-08 Quality prediction method, preparation method and system of high resistance gallium oxide based on deep learning and edge-defined film-fed growth method
EP21870470.8A EP4050333A4 (en) 2020-12-31 2021-02-08 HIGH STRENGTH TYPE GALLIUM OXIDE QUALITY PREDICTION METHOD AND PREPARATION METHOD AND SYSTEM BASED ON DEEP LEARNING AND EDGE DEFINED FILM FED GROWTH TECHNIQUE

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011639219.8 2020-12-31
CN202011639219.8A CN112834700B (zh) 2020-12-31 2020-12-31 一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统

Publications (1)

Publication Number Publication Date
WO2022141765A1 true WO2022141765A1 (zh) 2022-07-07

Family

ID=75926817

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/076071 WO2022141765A1 (zh) 2020-12-31 2021-02-08 一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统

Country Status (5)

Country Link
US (1) US12024791B2 (zh)
EP (1) EP4050333A4 (zh)
JP (1) JP7478819B2 (zh)
CN (1) CN112834700B (zh)
WO (1) WO2022141765A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116525040A (zh) * 2023-04-26 2023-08-01 中南大学 钢纤维-二元骨料体系填充密度预测模型构建方法及应用

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117468083B (zh) * 2023-12-27 2024-05-28 浙江晶盛机电股份有限公司 降籽晶过程的控制方法、装置、长晶炉系统和计算机设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999063134A1 (de) * 1998-06-04 1999-12-09 Leybold Systems Gmbh Einrichtung und verfahren zum züchten von kristallen
CN109338456A (zh) * 2018-12-03 2019-02-15 上海交通大学 基于数值模拟与神经网络判断的单晶制品生产智能控制技术
CN110033486A (zh) * 2019-04-19 2019-07-19 山东大学 透明晶体生长过程边缘及体积实时监测方法及系统
CN111028893A (zh) * 2019-10-28 2020-04-17 山东天岳先进材料科技有限公司 一种晶体生长预测方法及装置
CN111695287A (zh) * 2020-04-22 2020-09-22 山东天岳先进材料科技有限公司 一种预测SiC单晶炉内整体温度场的方法及设备
CN111893563A (zh) * 2020-08-25 2020-11-06 连城凯克斯科技有限公司 可自动调节放肩工艺参数的单晶炉及控制方法

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI267012B (en) 2004-06-03 2006-11-21 Univ Nat Cheng Kung Quality prognostics system and method for manufacturing processes
JP4611103B2 (ja) * 2005-05-09 2011-01-12 株式会社光波 β−Ｇａ２Ｏ３結晶の製造方法
JP5786179B2 (ja) 2010-03-12 2015-09-30 並木精密宝石株式会社 酸化ガリウム単結晶及びその製造方法
JP5491483B2 (ja) * 2011-11-15 2014-05-14 株式会社タムラ製作所 β−Ｇａ２Ｏ３系単結晶の成長方法
JP2013237591A (ja) * 2012-05-16 2013-11-28 Namiki Precision Jewel Co Ltd 酸化ガリウム融液、酸化ガリウム単結晶、酸化ガリウム基板、および酸化ガリウム単結晶の製造方法
JP2016013933A (ja) 2014-06-30 2016-01-28 株式会社タムラ製作所 アンドープ半絶縁Ｇａ２Ｏ３系単結晶基板及びその製造方法
CN105603528B (zh) * 2016-03-04 2018-07-27 同济大学 一种具有热释光性能的氧化镓晶体及其制备方法
CN107541776A (zh) * 2017-08-14 2018-01-05 同济大学 一种大尺寸氧化镓单晶的生长设备及方法
CN109537055A (zh) * 2019-01-28 2019-03-29 山东大学 一种半绝缘氧化镓晶体及其制备方法
US11342484B2 (en) * 2020-05-11 2022-05-24 Silanna UV Technologies Pte Ltd Metal oxide semiconductor-based light emitting device
CN118369767A (zh) * 2021-11-10 2024-07-19 斯兰纳Uv科技有限公司 外延氧化物材料、结构和装置
WO2023084274A1 (en) * 2021-11-10 2023-05-19 Silanna UV Technologies Pte Ltd Epitaxial oxide materials, structures, and devices

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1999063134A1 (de) * 1998-06-04 1999-12-09 Leybold Systems Gmbh Einrichtung und verfahren zum züchten von kristallen
CN109338456A (zh) * 2018-12-03 2019-02-15 上海交通大学 基于数值模拟与神经网络判断的单晶制品生产智能控制技术
CN110033486A (zh) * 2019-04-19 2019-07-19 山东大学 透明晶体生长过程边缘及体积实时监测方法及系统
CN111028893A (zh) * 2019-10-28 2020-04-17 山东天岳先进材料科技有限公司 一种晶体生长预测方法及装置
CN111695287A (zh) * 2020-04-22 2020-09-22 山东天岳先进材料科技有限公司 一种预测SiC单晶炉内整体温度场的方法及设备
CN111893563A (zh) * 2020-08-25 2020-11-06 连城凯克斯科技有限公司 可自动调节放肩工艺参数的单晶炉及控制方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
MU, WENXIANG: "Study on the Single Crystal Growth, Process and Properties of β-Ga2O3", CHINESE DOCTORAL DISSERTATIONS FULL-TEXT DATABASE, ENGINEERING SCIENCE & TECHNOLOGY, no. 12, 15 January 2019 (2019-01-15), pages 1 - 183, XP055947411 *
See also references of EP4050333A4 *
SU WEIFENG: "Predictive Control of Crystal Growth Based on Neural Network ", MASTER THESIS, no. 2, 1 December 2004 (2004-12-01), China, pages 1 - 67, XP055948838 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116525040A (zh) * 2023-04-26 2023-08-01 中南大学 钢纤维-二元骨料体系填充密度预测模型构建方法及应用
CN116525040B (zh) * 2023-04-26 2024-03-22 湖南铁院土木工程检测有限公司 钢纤维-二元骨料体系填充密度预测模型构建方法及应用

Also Published As

Publication number Publication date
US12024791B2 (en) 2024-07-02
JP7478819B2 (ja) 2024-05-07
CN112834700A (zh) 2021-05-25
CN112834700B (zh) 2023-03-21
EP4050333A1 (en) 2022-08-31
US20230160098A1 (en) 2023-05-25
EP4050333A4 (en) 2022-12-07
JP2023512385A (ja) 2023-03-27

Similar Documents

Publication Publication Date Title
WO2022141768A1 (zh) 一种基于深度学习和导模法的氧化镓的质量预测方法、制备方法及系统
WO2022141765A1 (zh) 一种基于深度学习和导模法的高阻型氧化镓的质量预测方法、制备方法及系统
WO2022141752A1 (zh) 一种基于深度学习和提拉法的导电型氧化镓的质量预测方法、制备方法及系统
WO2022141758A1 (zh) 一种基于深度学习和提拉法的高阻型氧化镓的质量预测方法、制备方法及系统
WO2022141764A1 (zh) 一种基于深度学习和导模法的导电型氧化镓的质量预测方法、制备方法及系统
WO2022141756A1 (zh) 一种基于深度学习和坩埚下降法的氧化镓制备方法及系统
WO2022141759A1 (zh) 一种基于深度学习和提拉法的氧化镓的质量预测方法、制备方法及系统
WO2022141766A1 (zh) 一种基于深度学习和热交换法的导电型氧化镓制备方法
WO2022141751A1 (zh) 一种基于深度学习和坩埚下降法的高阻型氧化镓制备方法
WO2022141755A1 (zh) 一种基于深度学习和热交换法的高阻型氧化镓制备方法
CN115595667A (zh) 一种碲锌镉晶体智能生长方法、系统、设备及存储介质
WO2022141750A1 (zh) 一种基于深度学习和热交换法的氧化镓制备方法及系统
WO2022141760A1 (zh) 一种基于深度学习和坩埚下降法的导电型氧化镓制备方法
JPWO2022141752A5 (zh)
JPWO2022141758A5 (zh)

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2022520490

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021870470

Country of ref document: EP

Effective date: 20220328

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21870470

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE