Process parameter optimization method and system for growing silicon carbide by liquid phase method

Abstract

The invention relates to the technical field of semiconductor material preparation, in particular to a process parameter optimization method and system for growing silicon carbide single crystals by a liquid phase method, and the method comprises the following steps: constructing a macroscopic finite element model, and obtaining macroscopic physical field data of a seed crystal leading edge boundary layer according to coupling of multiple groups of process parameters; constructing a microscopic dynamic model, and operating microscopic simulation by taking the multiple groups of macroscopic physical field data as surface boundary conditions to obtain multiple groups of interface morphology results; extracting defect quantitative indexes of each group of interface morphology results; and constructing a mapping model of the process parameter group and the defect quantitative index, and solving an optimal process parameter group which enables the defect quantitative index to reach a preset target. Compared with a trial-and-error method, the constructed macroscopic and microscopic linked complete simulation closed loop has the advantages that the research and development period is remarkably shortened, the cost is reduced, the purity and integrity of silicon carbide crystals are improved by the optimized process, the electrical property is improved, and scientific and efficient process guidance is provided for preparing silicon carbide by a liquid phase method.

Description

Technical Field

[0001] This invention relates to the field of semiconductor material preparation technology. Specifically, this invention relates to a method and system for optimizing process parameters for liquid-phase growth of silicon carbide single crystals. Background Technology

[0002] Silicon carbide (SiC), as a core representative of third-generation wide-bandgap semiconductor materials, boasts superior physical properties such as high breakdown field strength, high thermal conductivity, and high electron saturation velocity, making it an ideal material for fabricating high-temperature, high-voltage, and high-frequency power devices and radio frequency devices. It holds irreplaceable application value in key areas such as new energy vehicles, photovoltaic power generation, rail transportation, smart grids, and 5G communications. While physical vapor transport (PVT) is currently the mainstream industrial production method for SiC single crystal fabrication, it requires growth at temperatures above 2100℃, resulting in a high concentration of intrinsic defects. In contrast, the liquid-phase method, grown at relatively low temperatures of 1700-2000℃, is closer to thermodynamic equilibrium and demonstrates significant technological potential in reducing the density of defects such as microtubes and dislocations, and achieving high-quality P-type heavy doping.

[0003] However, the liquid phase method faces the core challenge of a narrow and complex process window: crystal quality is affected by multiple process parameters coupled together, such as crystal growth temperature, temperature gradient distribution, seed crystal and crucible rotation parameters, background gas conditions, and seed crystal pulling speed, and there are highly nonlinear relationships between these parameters; the current mainstream trial-and-error optimization model in the industry suffers from long R&D cycles and high costs, and it is difficult to summarize the clear and repeatable essential laws between process parameters and crystal quality; although existing macroscopic finite element simulations have been attempted to guide experiments, they cannot directly predict the microscopic interface morphology of crystal growth, and it is even more difficult to predict the formation mechanism of key defects that are crucial to device performance, such as macroscopic steps, flux inclusions, and dislocations, making it difficult to accurately attribute experimental results.

[0004] Therefore, there is an urgent need in this field to develop an effective method that can establish a quantitative correlation between macroscopic process parameters and microscopic crystal defects, so as to provide efficient and accurate guidance and optimization for liquid phase crystal growth processes. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing process parameters of liquid-phase silicon carbide growth, in order to solve the problems of inefficiency of trial and error methods, inability of macroscopic simulation to predict microscopic defects, and difficulty in establishing quantitative relationships between macroscopic and microscopic processes in the existing technology.

[0006] The technical solution of this invention is: a method for optimizing process parameters of liquid-phase silicon carbide growth, comprising the following steps:

[0007] S1. Construct a macroscopic finite element model through a finite element simulation software platform, input the process parameter group into the model, and obtain the macroscopic physical field data of the silicon carbide crystal growth interface through calculation.

[0008] S2. Using macroscopic physical field data as boundary conditions on the surface, combined with microscopic dynamic models, the growth and evolution process of crystal micro-interfaces is simulated and calculated to obtain interface morphology results.

[0009] S3. Extract the defect quantification index of the interface morphology results;

[0010] S4. Establish a mapping model between process parameter sets and defect quantification indicators. Construct an objective function using defect quantification indicators. Combine the optimization algorithm and the objective function to perform global optimization on the mapping model and obtain the optimal process parameter set corresponding to the defect quantification indicators.

[0011] Preferably, the macroscopic finite element model includes the following steps in its establishment process:

[0012] Define the thermal properties of materials, including graphite, insulation materials, stainless steel, copper, and SiC seed crystals;

[0013] The thermal field geometry is constructed based on the furnace cavity structure;

[0014] Define the system composition and content of the auxiliary solution in the chemistry module, and calculate the carbon solubility in the system using a thermodynamic database.

[0015] Preferably, the macroscopic physical field data of the crystal growth interface is obtained from the post-processing results of the macroscopic finite element model simulation by calculating the thickness of the seed crystal growth boundary layer and extracting the macroscopic physical field data of this thickness boundary layer.

[0016] Preferably, the microscopic dynamic model is constructed based on any of the following principles, assumptions, or combinations thereof, including: the Temkin multilayer interface model, Gibbs free energy theory, coupling of three growth mechanisms, interfacial thermodynamic adsorption analysis, and interfacial atomic segregation and statistical distribution.

[0017] Preferably, when the Gibbs free energy theory is used in the micro-dynamic model for silicon carbide growth, the growth topological unit is defined as a polyhedron that matches the crystal symmetry of the target silicon carbide polymorph, so as to minimize the total free energy of the system and thus correspond to the equilibrium growth morphology of the target silicon carbide polymorph.

[0018] Preferably, the coupling of the three growth mechanisms specifically involves coupling the Kossel-Stransky two-dimensional nucleation growth mechanism, the spiral dislocation growth step mechanism, and the step flow growth mechanism.

[0019] Preferably, the defect quantification indicators include the number of macro steps and the average height of macro steps, wherein a macro step refers to a step with a height greater than 10 micrometers.

[0020] A process parameter optimization system for liquid-phase silicon carbide growth, based on the process parameter optimization method for liquid-phase silicon carbide growth according to any one of claims 1-7, characterized in that it includes:

[0021] The macroscopic simulation module is used to build finite element models and configure process parameter groups, and output macroscopic physical field data of the crystal growth interface.

[0022] The microscopic simulation module is used to construct microscopic dynamic models, simulate the growth and evolution process of crystal microscopic interfaces based on macroscopic physical field data, and output the interface morphology results of the crystal.

[0023] The defect quantification module is used to analyze the interface morphology results and extract the defect quantification indicators.

[0024] The parameter optimization module stores the corresponding data of process parameter sets and defect quantification indicators, establishes a mapping model between process parameter sets and defect quantification indicators, constructs an objective function based on defect quantification indicators, and performs global optimization on the mapping model using optimization algorithms to find the optimal process parameter set that satisfies the preset objective function value for defect quantification indicators.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) This invention establishes a complete simulation closed loop from macroscopic process parameters to microscopic defect quantification and then to optimal process solution, which solves the core pain point that macroscopic simulation cannot predict microscopic defects and cannot effectively guide process optimization.

[0027] (2) This invention replaces experiments with calculations, allowing R&D personnel to complete a large number of process screenings and optimizations on a computer, avoiding expensive and time-consuming trial-and-error experiments, which can greatly shorten the R&D cycle of new products and significantly reduce R&D costs.

[0028] (3) The process window found in this invention can suppress the formation of macroscopic steps and inclusions. Compared with the traditional optimization approach that only pursues high growth rate or uniform temperature field, this invention can fundamentally improve the crystallization quality of the final crystal. Attached Figure Description

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0030] Figure 1 This is a flowchart of a method for optimizing process parameters for growing silicon carbide single crystals in liquid phase, as provided in Embodiment 1 of the present invention.

[0031] Figure 2 This is a structural block diagram of a process parameter optimization system for liquid-phase growth of silicon carbide single crystals provided in Embodiment 2 of the present invention.

[0032] Among them: 110, macroscopic simulation module; 120, microscopic simulation module; 130, defect quantification module; 140, parameter optimization module. Detailed Implementation

[0033] The present invention will be further described in detail below with reference to specific embodiments:

[0034] Example 1:

[0035] like Figure 1 As shown, a method for optimizing process parameters for liquid-phase growth of silicon carbide single crystals includes the following steps:

[0036] S1. Construct a macroscopic finite element model and obtain macroscopic physical field data of the seed crystal front boundary layer based on the coupling of multiple sets of process parameters.

[0037] Specifically, the model is first established, including: building a three-dimensional finite element model of the crystal growth furnace for growing 4-inch 4H-SiC crystals on a finite element simulation software platform, defining the thermal properties of the materials in the crystal growth furnace, including graphite, insulation materials, stainless steel, copper, and SiC seed crystals; at the same time, constructing the thermal field geometry based on the actual furnace cavity structure, determining the composition content of the fluxing solution system in the chemical module, and calculating the carbon solubility using a thermodynamic database.

[0038] After the model is established, the process parameters to be optimized and their ranges are defined. Specifically, the crystal growth temperature is controlled at 1700℃ to 1800℃, the seed crystal rotation speed is 50-200 rpm, the crucible is rotated in opposite directions at 10-50 rpm, and the inert gas argon pressure is 0.2-0.9 atm. At the same time, the fixed boundary conditions during simulation are determined, including the wall temperature of the water-cooled components, the seed crystal growth interface, and the radiation boundary.

[0039] The three-field coupling calculations were run 500 times using a parametric scanning method. Each calculation corresponds to a different set of process parameters, resulting in 500 different sets of macroscopic physical field data, including temperature field (T), flow field (V), and concentration field (C).

[0040] The thickness of the boundary layer at the seed crystal front was calculated based on the BPS formula. The temperature, carbon concentration, and flow rate data of the top and bottom surfaces of this boundary layer were extracted from the post-processing results of the macroscopic simulation. These extracted data were used as the surface boundary conditions for subsequent microscopic simulations.

[0041] The BPS formula is as follows:

[0042] ,

[0043] in, It is the boundary layer thickness; It is the solute diffusion coefficient; It is the kinematic viscosity of the melt; It is the angular velocity of the seed crystal rotation.

[0044] S2. Construct a microscopic dynamic model and run a microscopic simulation using multiple sets of macroscopic physical field data as boundary conditions on the surface to obtain multiple sets of interface morphology results.

[0045] In this embodiment, a three-dimensional step growth calculation model based on a non-Cartesian coordinate system, coupled with a Temkin multi-layer interface model, hexagonal topological units, and three growth mechanisms, is constructed as a microscopic dynamic model, specifically including:

[0046] The Temkin multilayer interface model is used to describe the interface extension. The traditional Jackson model only considers the interaction between the crystal surface layer and the interface layer in the derivation process. In order to eliminate the difference between this dual-interface growth model and the actual crystal-fluid interface as much as possible, this embodiment uses the more complex Temkin multilayer interface model.

[0047] Based on Gibbs free energy theory, the growth topological unit of 4H-SiC is summarized as "hexagonal". The growth morphology proposed by Hauy is composed of cubic unit cells, while 4H-SiC is hexagonal. According to Gibbs free energy theory, crystals grow in the direction of decreasing free energy. Therefore, in this embodiment, the growth topological unit of hexagonal SiC is summarized as hexagonal, so that the total free energy of the system is minimized, and the corresponding crystal morphology is the equilibrium morphology.

[0048] By coupling the Kossel-Stransky two-dimensional nucleation growth mechanism, the spiral growth mechanism of screw dislocations, and the step flow growth mechanism, a microscopic model of step growth under the combined influence of the three modes was constructed.

[0049] Calculate the thermodynamic adsorption coefficients of the tortuous region, step region, and planar region of the solid-liquid interface, and derive the change in Gibbs free energy of the interface layer caused by the transformation of atoms in the fluid into crystalline phase atoms.

[0050] Consider the segregation effect of atoms within the interface layer, and assume that all crystalline and fluid phase atoms contained in the monolayer are located at lattice positions, and that the adsorbed atoms enter the interface randomly, following a statistical distribution law.

[0051] Based on the above principles and assumptions, this embodiment constructs a three-dimensional step growth theoretical model in a non-Cartesian coordinate system.

[0052] In some other embodiments, the microdynamic model may also be constructed based on one or more of the above principles and assumptions.

[0053] After the microscopic dynamic model is constructed, 500 sets of macroscopic physical field data are read sequentially. Microscopic simulation is run for each set of data to simulate the crystal interface growth process. After the microscopic simulation corresponding to a single set of macroscopic physical field data is completed, the final interface morphology result under that set of data is automatically generated, which is a snapshot of the overall morphology of the seed crystal surface at the end of the simulation. It clearly presents the key features such as the final retained step distribution density, the maximum and minimum height of a single step, and the interface flatness.

[0054] After all 500 sets of microscopic simulations were completed, the step morphology evolution sequence and interface morphology results of each set of macroscopic physical field data were associated and stored according to the process parameter group number corresponding to each set of data, forming a complete data chain. This ensures that when extracting defect quantification indicators, the corresponding original process parameters and macroscopic physical field conditions can be accurately traced, providing coherent and traceable microscopic data support for subsequent parameter optimization solutions.

[0055] S3. Extract the defect quantification index for each group of interface morphology results.

[0056] The defect quantification indicators include the number of macro steps (N) and the average height of macro steps (H). In this embodiment, steps with a height greater than 10 micrometers are defined as macro steps. Their number is counted and the average height is calculated to obtain the number of macro steps and the average height of macro steps, which provides data support for the subsequent construction of the mapping relationship between process parameters and defect quantification indicators.

[0057] S4. Construct a mapping model between the process parameter set and the defect quantification index, and solve for the optimal process parameter set that makes the defect quantification index reach the preset target.

[0058] In this embodiment, a neural network model is used as the mapping model, and a four-layer fully connected neural network is constructed. During the model training process, 500 sets of process parameters and corresponding defect quantification index datasets obtained in the previous steps are used as training data, and mean squared error is used as the loss function to train the model, thus obtaining the trained mapping model.

[0059] A target function containing N and H is defined, with the minimum value of this function as the optimization objective. Within a preset process range, the trained mapping model is optimized using a genetic algorithm to obtain the optimal set of process parameters for suppressing defects.

[0060] Define the objective function:

[0061] ,

[0062] in, , Here, N and H are weighting coefficients, respectively, representing the number of macroscopic steps and the average height of the macroscopic steps; in this embodiment, , They are 0.6 and 0.4 respectively. It is the objective function value; the ultimate goal is to find the solution that makes... The optimal set of process parameters corresponding to the minimum value.

[0063] The steps for optimizing the mapping model using a genetic algorithm include:

[0064] Within the range of process parameters, 500 individuals are randomly generated to form an initial population; the parameter interval is controlled during generation to ensure that individuals are evenly distributed in the parameter space and to avoid local optima caused by the concentration of the initial population.

[0065] After population initialization, the fitness calculation stage begins: the process parameters of each individual in the population are input into the mapping model to quickly predict the corresponding N and H; then the J value of each individual is calculated according to the objective function. Considering that the smaller the J value, the fewer the defects, the fitness is defined as the reciprocal of J, so as to amplify the advantages of high-quality individuals and facilitate subsequent selection operations.

[0066] Pair up the parent pool individuals, rotate and cross each pair of continuous process parameter vectors by 5° to 15° to generate offspring, and truncate parameters that are out of range; then select offspring with a 10% probability, add a Gaussian perturbation of ±5% to the random parameters, and truncate parameters that are out of range as well, to complete the offspring processing.

[0067] When the optimal fitness no longer significantly improves after 30 consecutive generations, the iteration stops, and the individual with the highest fitness in the population is determined as the optimal process parameter set. In this embodiment, the final optimal process parameter set A is (1850℃, 125 rpm, -20 rpm, 0.5 atm). Specifically, the crystal growth temperature is 1850℃, the seed crystal rotation speed is 125 rpm, the crucible rotation speed is 20 rpm in the opposite direction, and the background gas pressure is 0.5 atm.

[0068] In some other embodiments, Bayesian algorithms or other optimization algorithms may be used to find the optimal set of process parameters.

[0069] To verify the actual beneficial effects of the optimal process parameter set found by the method of the present invention, this embodiment employs experimental verification, as detailed below:

[0070] The same seed crystal processing technology is used, including the seed crystal bonding process and the seed crystal remelting process;

[0071] A set of conventional process parameters, group B, was used as a control group for crystal growth experiments. The parameters of process parameter group B were (1900℃, 100 rpm, 50 rpm, 0.5 atm).

[0072] Using the optimal process parameter group A as the experimental group, crystal growth experiments were conducted. After the crystal growth was completed and cooled, the two groups of crystals were taken out and subjected to the same characterization tests: Raman spectroscopy was used to test the crystal form; X-ray diffraction was used to test the full width at half maximum (FWHM); molten potassium hydroxide wet etching was used to test the microtube density; and the resistivity of P-type silicon or N-type silicon was tested using the van der Bauer method on the Hall test platform.

[0073] Comparative test results: The surface morphology of the crystals grown using experimental group A was significantly better than that of the control group B.

[0074] Specifically, the experimental group had a monomorphic ratio of 98% or higher, while the control group had a monomorphic ratio of 95% or higher.

[0075] The experimental group had a half-peak width of less than 30 arcseconds; the control group had a half-peak width of greater than 50 arcseconds.

[0076] The microtubule density in the experimental group was less than 0.05 cells / cm³. 2， The microtubule density in the control group was less than 0.1 cells / cm². 2 ;

[0077] The resistivity of the experimental group was less than 0.1 Ω·cm; the resistivity of the control group was less than 0.2 Ω·cm.

[0078] This invention overcomes the technical bottlenecks of inefficiency in traditional trial-and-error methods and the inability of macroscopic simulation to correlate microscopic defects by constructing a complete technical closed loop encompassing macroscopic finite element simulation, microscopic dynamic modeling, defect quantification analysis, and mapping optimization. It uses macroscopic process parameters as input, obtains boundary layer physical field data through multi-physics coupling calculations, accurately simulates interface growth evolution through a multi-principle coupled microscopic dynamic model, and finally achieves efficient solution for optimal process parameters using a mapping model and optimization algorithm. Experimental verification shows that the process parameter set optimized by this invention significantly outperforms conventional processes in key indicators such as crystal purity, crystal integrity, defect suppression, and electrical performance. This not only fundamentally improves crystal quality but also significantly shortens the R&D cycle and reduces experimental costs, providing a scientific, accurate, and efficient process optimization solution for growing high-quality silicon carbide single crystals using the liquid-phase method.

[0079] Example 2:

[0080] like Figure 2 As shown, a process parameter optimization system for liquid-phase growth of silicon carbide single crystals includes:

[0081] The macroscopic simulation module 110 constructs a three-dimensional macroscopic model of the crystal growth furnace based on the finite element simulation software platform, determines fixed boundary conditions and configures 500 sets of process parameters to be optimized, and obtains macroscopic physical field data of the seed crystal front growth boundary layer through 500 coupled calculations and the BPS formula, including temperature, carbon concentration and flow rate data.

[0082] The microscopic simulation module 120, based on the microscopic simulation software platform, uses the Temkin multi-layer interface model, hexagonal topological units, and three growth mechanisms to construct a three-dimensional step growth dynamic model in a non-Cartesian coordinate system. It receives 500 sets of macroscopic physical field data output from the macroscopic simulation module as boundary conditions on the surface, simulates and calculates the growth evolution process of the crystal microscopic interface, and finally outputs 500 sets of interface morphology results.

[0083] The defect quantification module 130 analyzes the interface morphology results to obtain the defect quantification indicators for each group, including the number of macro steps and the average height of macro steps.

[0084] The parameter optimization module 140 establishes a dataset containing 500 data points, each including a set of process parameters and a corresponding defect quantification index. Through a trained mapping model between the process parameter set and the defect quantification index, and a preset objective function, the genetic algorithm is used to optimize the mapping model to obtain the optimal set of process parameters for suppressing defects.

[0085] The above embodiments are merely illustrative of the technical concept and features of the present invention, intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be construed as limiting the scope of protection of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the present invention is defined by the appended claims rather than the foregoing description, and thus all changes falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention.

Claims

1. A method for optimizing process parameters for growing silicon carbide by a liquid phase method, characterized by, The method comprises the following steps: A macro finite element model is constructed by a finite element simulation software platform, a process parameter group is input into the model, and macro physical field data of a silicon carbide crystal growth interface is obtained through calculation; The macro physical field data is taken as a surface boundary condition, a micro kinetic model is combined, a growth evolution process of a crystal micro interface is simulated and calculated, and an interface morphology result is obtained; A defect quantitative index of the interface morphology result is extracted; A mapping model of the process parameter group and the defect quantitative index is established, a target function is constructed by the defect quantitative index, a global optimization of the mapping model is performed by combining an optimization algorithm and the target function, and an optimal process parameter group corresponding to the defect quantitative index is obtained.

2. The method of claim 1, wherein the method is characterized by: The macro finite element model comprises the following steps in the establishment process: Material thermal physical parameters are defined, including graphite, heat preservation material, stainless steel, copper and SiC seed crystal; A thermal field geometric structure is constructed according to a furnace cavity structure; System components and contents of a dissolving solution in a chemical module are defined, and carbon solubility under the system is calculated by using a thermodynamic database.

3. The method of claim 1, wherein the method is characterized by: The macro physical field data of the crystal growth interface is obtained by calculating a seed crystal front growth boundary layer thickness and extracting macro physical field data of the thickness boundary layer from a post-processing result of the macro finite element model simulation.

4. The method of claim 1, wherein the method is characterized by: The micro kinetic model is constructed based on any one of the following principles, assumptions or various combinations thereof, including a Temkin multi-layer interface model, a Gibbs free energy theory, a three growth mechanism coupling, an interface thermodynamic adsorption analysis and an interface atom segregation and statistical distribution.

5. The method of claim 4, wherein the method is characterized by: When the Gibbs free energy theory is used in the micro kinetic model for silicon carbide growth, a growth topology unit is defined as a polyhedron matched with the crystal symmetry according to the crystal symmetry of a target silicon carbide polytype.

6. The method of claim 4, wherein the method is characterized by: The three growth mechanism coupling specifically couples a Kossel-Strandsky two-dimensional nucleation growth mechanism, a screw dislocation spiral growth step mechanism and a step flow growth mechanism.

7. The method of claim 1, wherein the method is characterized by: The defect quantitative index includes a macro step number and a macro step average height, wherein the macro step refers to a step with a height greater than 10 microns.

8. A process parameter optimization system for liquid phase method growth of silicon carbide, based on the process parameter optimization method for liquid phase method growth of silicon carbide according to any one of claims 1 to 7, characterized in that: The method comprises the following steps: A macro simulation module is configured to construct a finite element model and configure a process parameter group, and output macro physical field data of a crystal growth interface; A micro simulation module is configured to construct a micro kinetic model, simulate a growth evolution process of a crystal micro interface according to the macro physical field data, and output an interface morphology result of the crystal; A defect quantitative module is configured to analyze the interface morphology result and extract a defect quantitative index therefrom; A parameter optimization module is configured to store corresponding data of the process parameter group and the defect quantitative index, and establish a mapping model of the process parameter group and the defect quantitative index; A target function is constructed by the defect quantitative index, a global optimization of the mapping model is performed by combining an optimization algorithm, and an optimal process parameter group satisfying a preset target function value of the defect quantitative index is solved.

Images