A method for preparing p-type silicon carbide single crystals by multi-field time-coupling closed-loop controlled liquefaction.
By employing a multi-field time-series coupling closed-loop control method, the asymmetric warping of the solid-liquid interface is identified and predicted in real time, the linkage effect of aluminum loss is decoupled, and the hysteresis disturbance of convective solute transport in the molten pool is intervened in stages. This solves the problems of uneven crystal thickness, uneven doping, and performance dispersion in the preparation of P-type silicon carbide single crystals by the liquefaction method, and realizes the efficient preparation of large-size silicon carbide single crystals.
Patent Information
- Application Number
- CN202611308466.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies for preparing P-type silicon carbide single crystals by liquefaction, it is impossible to identify and predict the risk of asymmetric warping at the solid-liquid interface in real time, and it is impossible to analyze the linkage between aluminum component volatilization and adsorption loss on the inner wall of the crucible. Furthermore, the sluggish transport of solutes in the molten pool affects the crystal quality, resulting in uneven crystal thickness, uneven doping, and performance dispersion.
A multi-field time-series coupled closed-loop control method is adopted. By solving the temperature field, molten flow field, carbon solute transport field and aluminum doping multipath loss field in a time-series coupled manner, combined with the dynamic self-calibration of the solution boundary driven by external multi-measuring points, the prediction and dynamic control of the asymmetric warping of the solid-liquid interface are realized, the linkage effect of aluminum loss is decoupled, and the hysteresis disturbance of convective solute transport in the molten pool is intervened in a graded manner.
It effectively suppressed interface asymmetric warping distortion and aluminum inclusions, improved the thickness uniformity and doping uniformity of silicon carbide single crystals, reduced the performance differences between production batches, and improved the electrical performance consistency and industrial adaptability of large-size crystals.
Smart Images

Figure CN123108719A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of silicon carbide material processing, specifically relating to a method for preparing P-type silicon carbide single crystals by multi-field time-series coupled closed-loop controlled liquefaction. Background Technology
[0002] Silicon carbide is the core substrate of third-generation wide-bandgap semiconductors, and P-type silicon carbide single crystal substrates are indispensable basic materials for the manufacture of high-voltage bipolar power devices and ultraviolet optoelectronic devices. The liquefaction method, also known as the liquid-phase melting method, has a lower overall growth temperature and lower internal thermal stress compared to the mainstream physical vapor transport method. This effectively suppresses thermal stress-induced dislocation multiplication, making it a promising route for the industrialization of large-size silicon carbide single crystals. The liquefaction method for preparing P-type silicon carbide single crystals generally uses a silicon-based melt system, with aluminum as the P-type dopant source and graphite carbon source dissolving into the silicon melt under high-temperature conditions, continuously precipitating silicon carbide single crystals on the lower surface of the seed crystal.
[0003] Through a comprehensive search and review of publicly available patents and journal articles both domestically and internationally, current improvements to the preparation of P-type silicon carbide single crystals using the liquefaction method mainly focus on the following aspects: First, modification of the crucible body and lining materials to improve the molten wettability, reduce crystal adhesion to the crucible, and decrease the release of crucible impurities into the molten liquid; Second, optimization of the initial feed ratio of silicon, carbon, and aluminum to achieve basic P-type doping effects through static proportioning; Third, improvement of the seed crystal clamping mechanical structure and optimization of the seed crystal cooling method to improve the initial crystal formation quality; Fourth, use of offline simulation tools to complete a single-moment simulation of the molten pool's physical field before growth begins, with technicians manually compiling complete fixed heating, holding, and crucible rotation process curves based on the simulation results; Fifth, post-annealing treatment after growth to release residual thermal stress within the crystal. Existing liquefaction method simulations are all offline static simulations, only capable of calculating the internal temperature, flow field, and composition distribution of the molten pool at a fixed time slice. During continuous crystal growth, the crystal thickness continuously increases, the solid-liquid interface position continuously shifts downward, and the molten pool depth and geometric boundaries change simultaneously. Due to the combined effects of a non-uniform temperature gradient and asymmetric flow within the molten pool, the solid-liquid interface undergoes a slow, cumulative asymmetric warping deformation, with significant differences in the downward migration rate across different regions of the interface. This deformation is a slow evolutionary phenomenon that accumulates gradually over growth time and does not instantly cause crystal failure. Static simulations do not continuously iterate and update the boundary conditions of the solid-liquid interface over time, and therefore cannot capture this gradual asymmetric warping evolution trend. Actual production executes pre-programmed fixed process curves and does not dynamically adjust furnace process parameters based on the actual interface evolution. Only after crystal growth is complete and slicing is performed for characterization and testing can the consequences of defects be discovered: inconsistent growth thickness in different regions of the wafer, poor overall substrate thickness uniformity, asymmetric shift in P-type aluminum doping concentration along the wafer radial direction, and significant differences in electrical properties at different orientations of the same substrate. Existing technologies can only confirm defects after the fact and cannot predict the risk of asymmetric warping at the interface during crystal growth and implement timely process interventions.
[0004] In existing publicly available technical solutions, simulation calculations are all completed before growth begins, and the model boundaries are not updated during the growth process. The slow-developing interface asymmetric warping during growth can only be assessed retrospectively, lacking a complete technical path for real-time identification and control during growth. At the hardware level, modifications to the crucible and seed crystal structures alone cannot intervene in the dynamic evolution of interface warping trends during growth. In the high-temperature environment of the liquefaction method, aluminum components on the free surface of the molten liquid continuously volatilize outwards, while some aluminum components are adsorbed and retained by the inner wall of the high-temperature crucible. The effective aluminum component actually participating in crystal precipitation inside the molten pool continuously decreases with the growth process. The intensity of volatilization loss is affected by the flow state on the molten pool surface, and the intensity of adsorption loss on the inner wall of the crucible is affected by the flow state of the molten liquid adhering to the wall. There is a clear linkage between these two types of losses; changes in the molten pool convection state will synchronously change the loss rates of both types of losses. Existing processes all adopt a scheme of one-time addition of a fixed aluminum source during the feeding stage, lacking the means to perform time-series analysis of the linkage relationship between volatilization loss and adsorption loss on the inner wall of the crucible during the growth sequence. To offset aluminum losses during growth, process engineers can only increase the initial aluminum feed rate. However, excessive feed rate leads to the precipitation of a second phase of metallic aluminum at the solid-liquid interface, resulting in inclusion defects within the crystal. Insufficient feed rate results in insufficient effective doping flux at the interface during the later stages of growth, causing a significant drop in P-type doping concentration in the latter half of the crystal growth. Existing simulation models do not incorporate the coupled effects of volatilization and crucible adsorption losses into the iterative solution framework; these two types of losses are treated merely as independent random disturbances. Process parameters require repeated trial productions to refine, leading to significant fluctuations in the electrical properties of P-type silicon carbide crystals across different production batches, resulting in high material and time costs for research and development.
[0005] Existing technologies for dealing with aluminum loss focus on crucible coating modification and adjusting the initial feed ratio. These are all passive measures taken in advance and cannot dynamically analyze the linkage and coupling relationship between the two types of loss in accordance with the growth sequence, nor can they dynamically adjust the interface doping supply during the growth process.
[0006] Based on the two fundamental issues mentioned above, a third type of practical process problem arises: As crystal growth continues, the geometric boundary of the molten pool constantly changes, altering the convective motion within the molten pool, resulting in a significant hysteresis effect in solute transport within the melt. After the adjustment actions of mechanisms such as crucible rotation speed and hot zone power are completed, the molten flow state changes, but it takes a certain amount of time for the solute components to be re-transported to the solid-liquid crystallization interface, indicating a lack of synchronization between the adjustment actions and the actual solute response at the interface. Relying on preset fixed process parameters cannot adapt to this hysteresis disturbance in convective solute transport. The repeated hysteresis disturbance acting on the crystallization interface will generate periodic doping stripe defects within the crystal. These defects will extend throughout the crystal thickness direction and cannot be eliminated in subsequent wafer processing, directly causing device leakage current and performance dispersion, thus restricting the improvement of yield for large-size P-type silicon carbide substrates.
[0007] In summary, existing liquefaction methods for preparing P-type silicon carbide single crystals lack a multi-physics coupled progressive solution system that can continuously operate in tandem with crystal growth. This system cannot predict the risk of asymmetric warping at the solid-liquid interface in a time-sequential manner, cannot decouple the coupling effect of aluminum component volatilization and adsorption losses in a time-sequential manner, and cannot identify and intervene in the hysteresis disturbances of molten pool convection solute transport. These issues lead to practical bottlenecks such as interface warping distortion, aluminum inclusions, periodic doping stripes, poor substrate thickness and electrical consistency, and insufficient batch process reproducibility, hindering the large-size and industrial-scale application of liquefaction-based P-type silicon carbide single crystals. Summary of the Invention
[0008] The purpose of this invention is to overcome the aforementioned technical problems in the existing liquefaction method for preparing P-type silicon carbide single crystals, and to provide a method for preparing P-type silicon carbide single crystals using a multi-field time-series coupled closed-loop control liquefaction method. This invention constructs a multi-field time-series coupled closed-loop solution system, which performs time-series progressive coupling solutions for the temperature field, melt flow field, carbon solute transport field, aluminum doping multipath loss field, and solid-liquid interface evolution. All outputs of the solution system are mapped to physical hardware execution actions such as the furnace heating unit, crucible rotation mechanism, and seed crystal lifting mechanism, and do not belong to abstract intellectual activity rules. It relies on four algorithmic modeling innovations to achieve time-series evolution solutions for asymmetric warping of the solid-liquid interface, time-series decoupling solutions for aluminum double-loss linkage coupling, graded intervention for hysteresis disturbances in molten pool convection solute transport, and dynamic self-calibration of the solution boundary driven by external multi-measuring points. It is further supported by two physical process innovations: a layered gradient melt loading pretreatment process adapted to the time-series solution system, and a closed-loop execution process directly connected to the furnace body hardware of the solution system. It enables predictive dynamic control of the crystal growth process, suppresses interface asymmetric warping distortion, aluminum inclusions, and periodic doping stripe defects, improves the thickness uniformity and radial doping uniformity of P-type silicon carbide single crystals, reduces the differences in electrical properties between different batches of crystals, reduces the cost of repeated trial and error in the pilot production stage, and is suitable for mass production of 48-inch large-size P-type silicon carbide single crystals using the liquefaction method.
[0009] A method for preparing P-type silicon carbide single crystals by multi-field temporal coupling closed-loop controlled liquefaction method includes, in sequence, a melt pretreatment step, a liquid phase molten pool construction step, a multi-field temporal coupling closed-loop solution driven single crystal growth step, and a crystal in-situ annealing post-treatment step.
[0010] The multi-field temporal coupled closed-loop solution drives the single crystal growth process to run the multi-field temporal coupled closed-loop solution system; the multi-field temporal coupled closed-loop solution system includes a temperature field solution module, a melt flow field solution module, a carbon solute transport solution module, an aluminum doping multipath loss solution module, and a solid-liquid interface evolution solution module; the multi-field temporal coupled closed-loop solution system is configured with an iterative convergence judgment unit, which is used to complete the convergence judgment after each round of temporal progressive calculation.
[0011] This invention comprises six core elements, of which 1-4 are innovative algorithm modeling techniques not publicly reported in existing technologies. The algorithm's output maps all actions of the furnace body's physical hardware and does not belong to intellectual activity rules; 5-6 are supporting physical process innovations.
[0012] 1. The framework for solving the time-series evolution of asymmetric warpage at the solid-liquid interface abandons the traditional offline static snapshot simulation mode. It sets up an interface partition feature extraction unit and performs time-series progressively coupled iterative calculations on the temperature field, melt flow field, and carbon solute transport field. Progressive calculations are performed sequentially according to the crystal growth time steps. After each time step iteration, the solid-liquid interface is spatially partitioned and analyzed to extract the migration rate and morphological shift information of different partition interfaces, identifying the evolution trend of asymmetric warpage. The spatial position and partition morphological boundaries of the solid-liquid interface are updated synchronously, and the updated interface information is directly used as the boundary input condition for the next round of time step iterations. The boundary analysis logic distinguishes different orientation partitions of the interface, continuously tracking the evolution trend of the migration rate differences between each partition interface. When an asymmetric warpage risk is identified, the corresponding hardware process adjustment signal is directly output, correcting the growth rate of each region of the interface by adjusting the melt flow state and the thermal zone distribution state. The solution for the solid-liquid interface growth state also incorporates the effects of temperature gradient and the crystallization contribution from carbon and aluminum component transport, differing from traditional solutions that rely solely on temperature gradient to determine the interface growth state.
[0013] Traditional static offline simulations can only output a snapshot of the physical field inside the molten pool at a fixed moment, failing to capture the slowly accumulating asymmetric warping evolution trend at the solid-liquid interface as the crystal thickens. Existing technologies can only detect wafer thickness inhomogeneity and radial doping asymmetric shift defects after crystal growth and slicing. This invention employs a time-progressive coupled iterative mode that advances according to the growth time step, with a dedicated interface partition feature extraction unit to perform spatial partitioning analysis of the solid-liquid interface, distinguishing the boundary analysis logic of different orientation partitions, and continuously tracking the differences in the interface migration rate and warping evolution trend of each partition. It can identify the warping evolution trend in advance, before asymmetric warping evolves into permanent crystal defects, and output corresponding furnace process adjustment signals, achieving predictive control. This reduces the risk of wafer thickness inhomogeneity and radial doping asymmetric shift induced by interface asymmetric warping from the source, transforming the simulation tool from a post-growth analysis tool to a pre-emptive predictive control tool for the entire crystal growth process. The entire calculation uses real physical data collected from the furnace molten pool as input and outputs process instructions executable by hardware, rather than simply abstract logical deduction.
[0014] 2: The dual-loss linkage coupling time-series decoupling solution module is set up inside the aluminum doping multi-path loss solution module. The two types of losses are the aluminum volatilization loss path on the free surface of the melt and the aluminum adsorption loss path on the inner wall of the crucible. The solution module follows the growth sequence and analyzes the linkage coupling relationship between the two types of losses in each round of iteration calculation, decoupling the loss contribution of each type of loss. The actual effective aluminum doping flux at the solid-liquid interface is obtained. The actual aluminum doping flux obtained is compared with the preset target doping flux to obtain the flux difference. Based on the flux difference, a compensation control signal is generated iteratively. The compensation control signal is directly mapped to the crucible rotation speed adjustment action, the seed crystal small lifting action, and the fine adjustment action of the upper and lower hot zones of the molten pool. The interface doping flux compensation is achieved by changing the internal flow and transport state of the melt and the solid-liquid interface contact state. No solid material is added to the inside of the crucible during the entire crystal growth process.
[0015] Existing processes treat aluminum volatilization loss and crucible inner wall adsorption loss as two independent sets of random interference factors. These losses can only be passively offset by a single, fixed aluminum source added during the feeding stage. This approach fails to analyze the interdependent coupling relationship between these two types of losses as they change with molten pool convection, easily leading to doping drift or aluminum inclusion defects in different crystal growth sections. This invention incorporates the interdependent coupling effect of volatilization loss and crucible adsorption loss into a time-series iterative calculation, decoupling the dual-loss linkage, distinguishing the loss contributions of different paths, and forming a complete closed-loop chain: time-series analysis of the dual-loss linkage coupling relationship – solving for the actual aluminum doping flux at the interface – comparing the target doping flux to obtain flux differences – generating a physical process compensation signal – changing the molten flow and interface contact state to correct the actual doping flux at the solid-liquid interface. This eliminates the need for multiple batch trials to determine the initial feeding ratio, offsetting flux drift caused by dynamically coupled changes in aluminum composition losses during growth, avoiding various doping defects caused by feeding imbalances, improving doping consistency in different crystal growth sections, and enhancing the consistency of crystal electrical performance between different production batches. The module's computational output directly corresponds to the hardware actions of crucible rotation, seed crystal movement, and hot zone adjustment, thus establishing a complete link from the physical effects of dual-loss linkage coupling to the adjustment of the physical process.
[0016] 3: A graded intervention solution unit for molten pool convection-solute transport hysteresis disturbance is set up inside the molten flow field solution module. Following the growth sequence, in each round of iteration, it analyzes the hysteresis evolution process of solute transport reaching the solid-liquid crystallization interface after the execution of the molten pool mechanism adjustment action; it identifies the strength level of the hysteresis disturbance and divides the intervention strategies corresponding to different disturbance levels; it does not immediately issue new adjustment commands after the mechanism adjustment action is issued, matching the time evolution characteristics of solute transport hysteresis; when the hysteresis disturbance is identified to accumulate to a risk level, it outputs the hardware intervention signal corresponding to the grade, and suppresses the continuous effect of hysteresis disturbance on the crystallization interface by adjusting the crucible rotation conditions and the power distribution of the hot zone.
[0017] Current processes directly issue commands to adjust crucible rotation speed and hot zone power, without considering the hysteresis effect of solute transport reaching the crystallization interface after changes in molten pool convection. This easily leads to frequent superimposed adjustment actions, with hysteresis disturbances repeatedly acting on the solid-liquid interface, inducing periodic doping stripe defects within the crystal. Traditional offline simulation models ignore the evolution of solute transport hysteresis, equating mechanical adjustment actions with instantaneous interface responses. This invention incorporates the evolution of convective solute transport hysteresis into time-series iterative calculations, classifies hysteresis disturbances into levels, and performs graded interventions matching the time characteristics of hysteresis evolution. This avoids excessive superposition of adjustment actions, suppresses the continuous effect of hysteresis disturbances on the crystallization interface, and reduces the probability of periodic doping stripe defects from the source. The module's calculation output directly corresponds to the hardware actions of crucible rotation and hot zone adjustment, establishing a complete link from the physical effect of convective solute transport hysteresis to the actual process adjustment.
[0018] 4. External Multi-Measuring Point Driven Dynamic Self-Calibration Algorithm for Solving Boundaries: The multi-field time-coupled closed-loop solving system receives measured data from multiple externally placed temperature sensors within the furnace body, eliminating the need for internal sensors inside the high-temperature, highly corrosive molten pool. At the end of each time-series progressive calculation cycle, the physical quantities output by the model solution are compared with the measured physical quantities from multiple external measuring points within the furnace body to obtain residual information. This residual information is used to perform time-series adaptive calibration and updates on multiple sets of solving boundaries of the solving system. The solving boundaries are continuously updated throughout the entire crystal growth time series, offsetting deviations between the model and actual working conditions caused by aging of furnace heating elements, wear and tear of insulation components, and erosion of the crucible inner wall by the molten metal. This ensures that the solving model closely matches the actual state of the molten pool within the furnace under long-cycle growth conditions.
[0019] Traditional simulations use fixed, pre-defined boundary conditions. However, the liquefaction crystal growth process is lengthy, and heating and insulation components within the furnace gradually age and wear down. The crucible's inner wall is continuously eroded by the high-temperature molten metal, causing the actual molten pool conditions inside the furnace to drift. Consequently, the calculation results of the fixed-boundary simulation model gradually deviate from the actual furnace conditions, leading to a continuous decline in simulation reliability. This invention incorporates measured physical data from multiple external measuring points within the furnace into each iteration of the time-series simulation. The residuals between the model's solution and the measured values are continuously used to calibrate and update the solution boundary, ensuring that the multi-physics solution results closely match the actual state of the molten pool under long-term crystal growth conditions. This eliminates the need to develop specialized sensors that can be placed inside the high-temperature, highly corrosive molten pool, lowering the hardware implementation threshold for the entire system. The algorithm uses measured physical data from the furnace as input, outputs updated solution boundaries, and further generates hardware control commands, seamlessly integrating with the physical growth furnace hardware.
[0020] 5: The layered gradient melt loading pretreatment process adapted to the time-series solution system involves graded high-temperature vacuum degassing and purification of the graphite carbon source material, layering graphite carbon sources of different particle sizes to construct a layered gradient carbon source dissolution and supply structure; inert coating protection is applied to the aluminum P-type doped source to reduce the instantaneous large-scale volatilization of aluminum during the heating and melting stage; the pretreated silicon matrix material, layered gradient-arranged graphite carbon sources, and inertly coated aluminum doped sources are loaded into the crucible in a layered loading manner; the layered loading structure matches the initial mesh generation of the multi-field time-series coupled closed-loop solution system, reducing the iteration oscillation amplitude during the initialization stage of the solution system and providing stable initial material boundary conditions for the time-series progressive coupled iterative computation.
[0021] 6: The solution system is directly connected to the local hardware of the furnace and executes a closed-loop process. The multi-field time-coupled closed-loop solution system is deployed inside the local control system of the silicon carbide growth furnace. The solution system continuously reads the measured physical data collected by various sensors in the furnace. After each round of time-series progressive calculation converges, the process control signal is output. The process control signal is directly sent to the furnace heating power supply, the crucible rotation drive mechanism, and the seed crystal lifting drive mechanism. The entire single crystal growth process does not load a pre-cured complete set of process curves. All key process control points in the growth process are dynamically generated by the time-series iterative calculation of the multi-field time-coupled closed-loop solution system, realizing that the solution calculation directly drives the physical furnace to complete the liquefaction growth of P-type silicon carbide single crystals.
[0022] Furthermore, the working logic of the iterative convergence determination unit of the multi-field time-coupled closed-loop solution system is as follows: compare the changes in the solid-liquid interface partition morphology, the migration rate of each partition, the aluminum doping flux, and the residual of the molten pool flow field obtained from the calculations of two adjacent time steps; if all the changes are less than the preset convergence threshold, the current round of time-series progressive calculation is determined to have converged, and steady-state process control parameters are output; if the convergence condition is not met, the solution boundary is updated, and the next time step iterative calculation is entered.
[0023] Furthermore, the solid-liquid interface evolution solution module automatically reconstructs the local solution mesh when the interface mesh nodes shift due to crystallization, thus avoiding temporal divergence caused by mesh distortion.
[0024] Furthermore, the cooling annealing process parameters for the in-situ annealing post-processing of the crystal are also output by the time-series iterative calculation of the multi-field time-series coupled closed-loop solution system.
[0025] Furthermore, after the liquid phase molten pool construction process is completed, the multi-field time-coupled closed-loop solution system first performs initialization time-series iterative calculations. After the initialization iterations converge, the seed crystal contact molten pool is then started to carry out single crystal growth.
[0026] Beneficial effects 1. This invention employs a time-series evolution solution framework for asymmetric warpage at the solid-liquid interface, abandoning the traditional static snapshot simulation mode. It performs progressively coupled iterations according to the crystal growth time step, and specifically sets up an interface partition feature extraction unit to perform spatial partitioning analysis of the solid-liquid interface. It distinguishes the boundary analysis logic of different orientation partitions, continuously tracks the differences in the interface migration rate and warpage evolution trend of each partition, and outputs process adjustment signals before the asymmetric warpage defect forms, correcting the growth rate of each region of the interface. This solves the technical problem in the background technology that static simulation cannot capture the progressive asymmetric warpage of the interface in a timely manner. It transforms the simulation tool from a post-analysis tool to a tool for predictive control throughout the entire growth process; all computational inputs and outputs are connected to the physical furnace hardware, avoiding the limitations of intellectual activity rules.
[0027] 2. This invention 2 sets up a dual-loss linkage coupling timing decoupling solution module, incorporating the linkage coupling effect of molten free surface volatilization loss and crucible inner wall adsorption loss into the timing iteration calculation, decoupling the contribution ratio of the two types of losses, and generating a closed-loop iterative signal to achieve doping flux compensation, without relying on excessive aluminum source addition during the feeding stage. This solves the doping concentration drift problem in different growth sections caused by dual-loss linkage coupling in the growth sequence, and avoids aluminum metal inclusion defects induced by aluminum feeding imbalance; it improves the uniformity of doping distribution along the growth direction and radial direction of P-type silicon carbide crystals, improves the reproducibility of crystal electrical performance in different production batches, and reduces the material and time costs caused by repeated feeding and trial and error during the trial production stage.
[0028] 3. The present invention's 3-stage intervention solution unit for hysteresis disturbance in molten pool convection incorporates the hysteresis evolution process of solute transport to the crystallization interface after mechanism adjustment into a time-series iterative calculation. It identifies the strength level of the hysteresis disturbance, executes a staged intervention strategy, matches the time evolution characteristics of solute transport hysteresis, avoids excessive superposition of adjustment actions, and suppresses the continuous action of hysteresis disturbance on the crystallization interface, thereby reducing the probability of periodic doping stripe defects inside the crystal from the source. This solves the problem of existing technologies directly issuing adjustment commands while ignoring stripe defects caused by solute transport hysteresis.
[0029] 4. This invention employs an external multi-measuring-point driven dynamic self-calibration algorithm for solving the boundary. Utilizing externally measured temperature data from multiple locations within the furnace body, the algorithm continuously calibrates the solution system boundary through residual comparison in each iteration cycle. This compensates for thermal and flow field drift caused by aging losses of furnace heating elements and insulation components, as well as crucible erosion. This ensures that the multi-physics solution results closely match the actual working conditions of the molten pool under long-cycle growth conditions. It eliminates the need for specialized sensors inside the high-temperature corrosion molten pool, reducing the difficulty of equipment hardware implementation. The solution chain is fully integrated with the physical hardware, effectively improving the reliability of long-term growth process control.
[0030] 5. The present invention has a layered gradient melt filling pretreatment physical process, which performs graded degassing and purification of carbon source, uses graphite of different particle sizes to construct a layered gradient carbon source supply structure, and provides inert coating protection for aluminum source. The layered filling of materials matches the initial grid of the solution system, reduces the oscillation of solution initialization iteration, provides stable initial boundary conditions for the whole time-series progressive solution, reduces the risk of divergence in early calculations, and ensures the stable operation of subsequent time-series coupled solutions.
[0031] 6. The present invention 6 has a closed-loop execution process with the solution system directly connected to the local hardware of the furnace body. The solution system is deployed in the local control system of the furnace body, continuously reads the measured data of the furnace body sensors, and directly sends control signals to the heating power supply, crucible rotation drive mechanism and seed crystal lifting drive mechanism after iterative convergence. The entire growth process does not use the pre-cured complete process curve. All key process control points are dynamically generated by time-series iterative calculation, truly realizing the closed-loop connection from the solution model to the solid growth process.
[0032] Figure 1 is a block diagram of the multi-field time-series coupled closed-loop solution-solid furnace closed-loop system of the present invention. 1-Data acquisition from multiple external sensors on the furnace body; 2-External multi-measuring-point driven dynamic self-calibration unit for the solution boundary; 3-Multi-field time-series progressive coupling iterative solution unit; 4-Temperature field solution submodule; 5-Molten flow field solution submodule, with a built-in molten pool convection-solute transport hysteresis disturbance hierarchical intervention unit; 6-Carbon solute transport solution submodule; 7-Aluminum doping multi-path loss solution submodule, with a built-in dual-loss linkage coupled time-series decoupling solution module; 8-Solid-liquid interface evolution solution submodule, with a built-in solid-liquid interface asymmetric warping time-series evolution solution framework; 9-Iterative convergence judgment unit; 10-Output solid process control signal; 11-Furnace body actuator; 12-Liquidation method P-type silicon carbide molten pool crystal growth. Detailed implementation method: Example 1
[0033] A method for preparing P-type silicon carbide single crystals by multi-field temporal coupling closed-loop controlled liquefaction method includes, in sequence, a melt pretreatment step, a liquid phase molten pool construction step, a multi-field temporal coupling closed-loop solution driven single crystal growth step, and a crystal in-situ annealing post-treatment step.
[0034] S1 Melt Pretreatment Process: High-purity silicon is selected as the molten matrix raw material, multi-size high-purity graphite particles are selected as the carbon source, and metallic aluminum is selected as the P-type dopant source. The graphite carbon source undergoes graded high-temperature vacuum degassing and purification treatment, and a layered gradient carbon source dissolution and supply structure is constructed using graphite of different particle sizes. The metallic aluminum P-type dopant source is subjected to inert coating protection treatment to reduce the instantaneous large-scale volatilization of aluminum during the heating and melting stage. The pretreated silicon matrix material, the layered gradient graphite carbon source, and the inertly coated aluminum dopant source are loaded into the crucible in a layered loading manner. The layered loading structure matches the initial mesh generation of the multi-field time-coupled closed-loop solution system, reducing the iteration oscillation amplitude during the initialization stage of the solution system.
[0035] S2 Liquid Phase Molten Pool Construction Process: The seed crystal is installed and fixed on the seed crystal clamping mechanism, the growth furnace cavity is sealed, and after the furnace cavity is evacuated, an inert protective gas is introduced; the furnace heating system is started to gradually increase the temperature, and the material inside the crucible is heated and melted to form a silicon-carbon-aluminum ternary molten pool; after the molten pool is constructed, the multi-field time-series coupled closed-loop solution system is started to perform the initialization time-series progressive calculation. After the initialization iteration converges, the single crystal growth process is then entered.
[0036] The S3 multi-field time-coupled closed-loop solution drives the single crystal growth process: controlling the seed crystal mechanism to move downwards, so that the lower surface of the seed crystal contacts the ternary molten pool surface, officially initiating the crystallization and growth of P-type silicon carbide single crystals. The multi-field time-coupled closed-loop solution system runs throughout the process, with the temperature field solution module, molten flow field solution module, carbon solute transport solution module, aluminum doping multipath loss solution module, and solid-liquid interface evolution solution module working collaboratively. The system calls the following: 1. The solid-liquid interface asymmetric warpage time-series evolution solution framework performs time-series progressive coupling iterative calculations according to the growth time steps. In each iteration, it performs spatial partitioning analysis of the solid-liquid interface, extracts the migration rate and morphological shift information of different partition interfaces, identifies the asymmetric warpage evolution trend of the interface, and updates the spatial position and partition morphological boundaries of the solid-liquid interface; it distinguishes the boundary analysis logic of different orientation partitions of the interface, and outputs hardware process adjustment signals after identifying warpage risks; 2. The dual-loss linkage coupling time-series decoupling solution module is called to perform time-series analysis of the linkage coupling relationship between two types of losses: molten free surface volatilization and crucible inner wall adsorption, decouples the contribution ratio of each type of loss, solves the actual aluminum doping flux of the interface, compares it with the target doping flux to generate compensation and control signals, and maps them to the crucible rotation speed, seed crystal slight rise and fall, and fine adjustment of the upper and lower hot zones of the molten pool; 3. The molten pool convection solute transport hysteresis disturbance graded intervention solution unit analyzes the hysteresis evolution process of solute transport to the crystallization interface after adjustment by the time-series analysis mechanism, identifies the strength level of hysteresis disturbance, executes graded intervention strategies, and outputs corresponding hardware intervention signals. It calls a dynamic self-calibration algorithm for the solution boundary driven by four external multi-measuring points, reads measured data from multiple external temperature sensors in the furnace, and compares the model output with the measured data at the end of each iteration cycle to obtain residuals. These residuals are used to calibrate and update the solution boundary of the solution system. After each time-series progression, it enters the iteration convergence judgment unit, comparing the interface partition morphology, the migration rate of each interface partition, the aluminum doping flux at the interface, and the change amplitude of the molten pool flow field residuals. If convergence is achieved, process control signals are output to the furnace heating power supply, the crucible rotation drive mechanism, and the seed crystal lifting drive mechanism. If convergence fails, the solution boundary is updated and the next time step iteration begins. The entire single crystal growth process does not load a pre-cured complete process curve; process control points are dynamically generated by time-series iteration calculations. Crystallization growth continues until the crystal reaches the target thickness.
[0037] S4 Crystal In-situ Annealing Post-processing: After crystal growth is completed, the multi-field time-coupled closed-loop solution system iteratively outputs cooling annealing process parameters. The furnace body completes in-situ annealing according to the output parameters to release residual thermal stress inside the crystal. After annealing, the furnace body is cooled down and the P-type silicon carbide single crystal is taken out.
[0038] Qualitative explanation of the algorithm modeling efficiency enhancement principle in this embodiment: This embodiment fully utilizes four algorithm modeling innovations, combined with two physical process innovations to complete single crystal preparation. 1. The solid-liquid interface asymmetric warpage time-series evolution solution framework performs spatial partitioning analysis on the solid-liquid interface, distinguishes the analysis logic of different orientation partitions, continuously tracks the difference in the migration rate of each partition of the interface and the asymmetric warpage evolution trend, and outputs process adjustment signals before the warpage forms permanent crystal defects, correcting the growth rate of each region of the interface, avoiding the drawbacks of traditional static simulation that can only detect uneven wafer thickness and radial doping asymmetric offset after the fact. 2. The dual-loss linkage coupling time-series decoupling solution module analyzes the linkage coupling relationship between aluminum volatilization loss and crucible adsorption loss, decouples the contribution ratio of the two types of losses, solves the actual effective doping flux of the interface, and achieves flux compensation by adjusting the melt flow and seed crystal interface contact state. No solid material is added during the growth process, avoiding doping drift and aluminum metal inclusion defects caused by material imbalance, and stabilizing the doping level of different growth sections of the crystal. 3. A graded intervention solution unit for hysteresis disturbances in molten pool convection solute transport analyzes the evolution process of solute transport hysteresis, classifies the hysteresis disturbances into levels, and performs graded interventions to match the hysteresis evolution time characteristics, avoiding excessive superposition of adjustment actions, suppressing the continuous effect of hysteresis disturbances on the crystallization interface, and reducing the probability of periodic doping stripe defects. 4. An external multi-measuring point-driven dynamic self-calibration algorithm for the solution boundary continuously calibrates the solution boundary in each iteration using measured data from external measuring points and residuals from the model output, offsetting the operational drift caused by furnace component aging and crucible erosion, ensuring that the model solution results for the long-cycle growth process closely match the actual working conditions of the molten pool. The entire solution system uses real physical data collected from the furnace as input, and the output directly drives the furnace hardware actuators; it does not involve abstract intellectual deduction. 5. A preprocessing step provides a stable initial boundary for the time-series iteration, reducing initialization calculation oscillations. 6. A closed-loop direct connection is achieved between the solution system and the local furnace hardware, dynamically generating process control points throughout the growth process. Multi-module collaborative operation improves wafer thickness uniformity, radial and growth direction doping uniformity, reduces the probability of various defects such as interface warping distortion, aluminum inclusions, and doping stripes, and improves process reproducibility between batches.
[0039] Example 2 A method for preparing P-type silicon carbide single crystals using a multi-field temporal coupling closed-loop controlled liquefaction process includes, sequentially, a melt pretreatment step, a liquid phase molten pool construction step, a multi-field temporal coupling closed-loop solution-driven single crystal growth step, and a crystal in-situ annealing post-treatment step. This embodiment is suitable for large-volume crucibles and long-cycle single crystal growth conditions. The large-volume molten pool has a larger convection scale, a more significant dual-loss linkage coupling effect, a longer solute transport lag time, and a more pronounced operating condition drift phenomenon due to the long growth cycle.
[0040] S1 Molten Material Pretreatment Process: High-purity silicon substrate raw materials, multi-particle-size high-purity graphite carbon source, and aluminum P-type doped source are selected. The graphite carbon source undergoes graded high-temperature vacuum degassing and purification to construct a layered gradient carbon source dissolution and supply structure; the aluminum doped source is inertly coated for protection; the silicon substrate material, the layered gradient-arranged graphite carbon source, and the coated aluminum doped source are layered and filled into the interior of a large-volume crucible. The layered filling matches the initial mesh of the solution system, reducing initialization iteration oscillations.
[0041] S2 Liquid Phase Molten Pool Construction Process: Install and fix the seed crystal, seal the furnace cavity, evacuate and then introduce inert protective gas; heat the furnace body to fully melt the material inside the crucible to form a large volume silicon-carbon-aluminum molten pool; after the molten pool is constructed, start the multi-field time-series coupled closed-loop solution system to perform initialization time-series progressive calculation. Only after initialization convergence can the single crystal growth stage be entered.
[0042] S3 Multi-Field Temporal Coupling Closed-Loop Solution Drives Single Crystal Growth Process: Drives the seed crystal to move down and contact the molten pool surface, initiating single crystal growth. The multi-field temporal coupling closed-loop solution system runs continuously, activating algorithm modules 1, 2, and 4. Targeting the large convection scale of the large-volume molten pool, it amplifies the temporal analysis interval of the dual-loss linkage coupling to adapt to the loss coupling characteristics of the large-volume molten pool. 1. The solid-liquid interface asymmetric warpage temporal evolution solution framework continuously iterates, performing spatial partitioning analysis of the solid-liquid interface, extracting the migration rate and morphological shift information of each partition, identifying the asymmetric warpage evolution trend, and outputting process adjustment signals; 2. The dual-loss linkage coupling temporal decoupling solution module adapts to the large-volume molten pool, amplifies the temporal analysis interval of the dual-loss linkage coupling, analyzes the linkage coupling relationship between volatilization and adsorption losses, decouples the loss contribution ratio, and generates compensation and control signals to map hardware actions; 4. An external multi-measuring-point driven dynamic self-calibration algorithm continuously reads measured data from multiple external sensors on the furnace body. At the end of each iteration, residual information is used to calibrate the solution boundary, offsetting the condition drift caused by furnace component aging and crucible erosion during long growth cycles. The molten pool convection solute transport hysteresis disturbance graded intervention solution unit continues to run. When the hysteresis disturbance intensity is within the normal range, the basic monitoring logic is maintained. Only when the accumulated hysteresis disturbance exceeds the preset range is the corresponding graded intervention command output. The iterative convergence judgment unit completes the convergence judgment for each iteration. After convergence, the control signal is output to the furnace heating power supply, crucible rotation drive mechanism, and seed crystal lifting drive mechanism. The entire growth process does not use the pre-cured complete process curve and continues to crystallize to the target crystal thickness.
[0043] S4 Crystal In-situ Annealing Post-processing: After growth is completed, the multi-field time-coupled closed-loop solution system outputs the annealing and cooling process parameters, the furnace body performs in-situ annealing, and after cooling, the large-size P-type silicon carbide single crystal is taken out.
[0044] Qualitative explanation of the algorithm modeling efficiency enhancement principle in this embodiment: This embodiment is designed for long-cycle growth conditions in large-volume crucibles, focusing on leveraging the control capabilities of algorithm modules 1, 2, and 4. Addressing the larger convection scale of large-volume molten pools, it expands the time-series analysis interval for the dual-loss linkage coupling, adapting to the loss coupling characteristics of large-volume molten pools. 1. The time-series progressive coupling iteration can continuously track the differences in the migration rate of different partitions of the solid-liquid interface under large-volume molten pool conditions, identifying asymmetric warping evolution trends in advance and outputting process adjustment signals to suppress wafer thickness unevenness and radial doping asymmetric offset defects. 2. For the more prominent dual-loss linkage coupling effect in large-volume molten pools, the time-series analysis interval is expanded to analyze the linkage coupling correlation between the two types of losses, decouple the loss contribution ratio, generate compensation control signals, and suppress doping drift and aluminum inclusion defects caused by aluminum component losses. 4. The external multi-measuring point driven dynamic self-calibration algorithm for solving the boundary is particularly valuable in long-cycle growth scenarios. Multiple external measuring points are used to jointly compare residuals and calibrate the solution boundary, continuously offsetting the operational deviations caused by heating, aging of insulation components, and molten metal erosion of the crucible. This ensures that the multiphysics solution results closely approximate the actual working conditions of the molten pool, preventing gradual model distortion. 3. A hysteresis-based perturbation hierarchical intervention unit continuously monitors the process, only outputting strong intervention actions when the accumulated perturbation exceeds the limit, avoiding frequent adjustments to the large-volume molten pool's steady state. 5. The preprocessing process ensures stable initialization of the solution under large feed volumes. 6. A closed-loop direct connection is maintained between the solution system and the furnace's local hardware. The entire computation is based on actual measured physical inputs from the furnace, outputting hardware-executable instructions. Under large-volume, long-cycle growth conditions, this ensures uniformity in the thickness, radial and growth direction doping of P-type silicon carbide single crystals, reducing batch performance fluctuations.
[0045] Example 3 A method for preparing P-type silicon carbide single crystals using a multi-field temporal coupling closed-loop controlled liquefaction process includes, sequentially, a melt pretreatment step, a liquid phase molten pool construction step, a multi-field temporal coupling closed-loop solution-driven single crystal growth step, and a crystal in-situ annealing post-treatment step. This embodiment addresses the stringent requirements of low-defect fabrication, fully utilizing all four innovative algorithm modules, improving the sensitivity for identifying interface asymmetric warping risks, and enhancing the accuracy of hierarchical analysis of convective solute transport hysteresis disturbances.
[0046] S1 Molten Material Pretreatment Process: High-purity silicon substrate, multi-particle-size high-purity graphite carbon source, and aluminum doped source are selected. The graphite carbon source is purified by graded high-temperature vacuum degassing to construct a layered gradient carbon source supply structure, and the aluminum doped source is inertly coated for protection. After pretreatment, the material is layered and filled into the crucible. The layered structure is matched to solve the initial mesh of the system to reduce initial iteration oscillations.
[0047] S2 Liquid Phase Molten Pool Construction Process: Install seed crystals, seal furnace cavity, evacuate and fill with inert protective gas; heat up and melt materials to form a silicon-carbon-aluminum ternary molten pool; after the molten pool construction is completed, start the multi-field time-series coupled closed-loop solution system to perform initialization time-series progressive calculations, and after initialization convergence, proceed to the single crystal growth process.
[0048] The S3 multi-field temporal coupled closed-loop solution drives the single crystal growth process: the seed crystal mechanism moves downward, the seed crystal contacts the molten pool surface, and single crystal precipitation and growth are initiated. The multi-field temporal coupled closed-loop solution system runs throughout the entire process, fully utilizing all algorithm modules 1, 2, 3, and 4, increasing the identification sensitivity of the solid-liquid interface asymmetric warping temporal evolution solution framework, and enhancing the analytical accuracy of the molten pool convection-solute transport hysteresis hierarchical intervention solution unit. 1. The framework for solving the asymmetric warpage evolution of the solid-liquid interface has improved recognition sensitivity. It performs iterative calculations step-by-step, spatially partitioning the solid-liquid interface to capture earlier differences in the migration rates of different partitions and early signs of asymmetric warpage evolution. It outputs hardware process adjustment signals to correct the growth rate of interface partitions. 2. The dual-loss linkage coupling and decoupling solution module continuously analyzes the linkage and coupling relationship between volatilization and adsorption losses, decouples the contribution ratio of the two types of losses, solves the actual aluminum doping flux at the interface, and generates compensation signals for crucible rotation speed, small increases / decreases in the seed crystal, and fine-tuning of the hot zone. 3. The molten pool convection-solute transport hysteresis disturbance graded intervention solution unit improves analytical accuracy. It analyzes the evolution process of solute transport hysteresis in a time sequence, accurately identifies the strength level of hysteresis disturbances, and executes corresponding graded intervention strategies to suppress the continuous effect of hysteresis disturbances on the crystallization interface. 4. An external multi-measuring-point driven dynamic self-calibration algorithm for solving the boundary conditions is used. At the end of each iteration, the measured data from multiple external sensors in the furnace body and the residual output of the model are used to calibrate and update the boundary conditions. The iteration convergence judgment unit completes the convergence judgment for each iteration. After convergence, the process control signal is output to the furnace heating power supply, the crucible rotation drive mechanism, and the seed crystal lifting drive mechanism. The entire growth process has a complete process curve without pre-curing and continues to grow until the crystal reaches the target thickness.
[0049] S4 Crystal In-situ Annealing Post-processing: After growth is completed, the multi-field time-coupled closed-loop solution system outputs the annealing and cooling process parameters, the furnace body completes in-situ annealing, and after cooling, the furnace is opened to take out the low-defect-level P-type silicon carbide single crystal.
[0050] Qualitative explanation of the algorithm modeling efficiency enhancement principle in this embodiment: This embodiment addresses the stringent low-defect fabrication requirements by increasing: 1. The sensitivity of identifying interface asymmetric warpage risk and enhancing the accuracy of hierarchical analysis of convective solute transport hysteresis disturbances. 1. Increasing the identification sensitivity allows for earlier detection of differences in the migration rates of different zones at the solid-liquid interface and early signs of asymmetric warpage evolution. Hardware intervention commands are output before warpage defects fully form, correcting the growth rates of each interface zone and suppressing wafer thickness unevenness and radial doping asymmetric shift from the source. 2. Continuous analysis of the linkage and coupling relationship between the aluminum volatilization crucible adsorption losses and decoupling the loss contribution ratio allows for dynamic compensation of interface doping flux, suppressing doping drift in different crystal growth sections and avoiding aluminum inclusion defects. 3. Improving the accuracy of hysteresis disturbance analysis accurately identifies the strength level of hysteresis disturbances, executes hierarchical intervention strategies, matches the solute transport hysteresis time characteristics, avoids excessive superposition of adjustment actions, and reduces the generation of periodic doping stripe defects. 4. The solution boundary is continuously calibrated by comparing residuals at multiple external measurement points to ensure consistency between the solution model and the actual molten pool conditions over long periods. 5. The preprocessing process provides a stable initial solution boundary. 6. A closed-loop direct connection is maintained between the solution system and the furnace's local hardware. The entire algorithm operates on the actual physical state of the molten pool and measured data from the furnace as input, with all outputs mapping to actions executed by the furnace hardware; it does not belong to abstract intellectual rules. Multi-module collaboration simultaneously suppresses interface asymmetric warping distortion, aluminum inclusion defects, and periodic doping stripe defects, resulting in P-type silicon carbide single crystals with uniform thickness, uniform radial and growth direction doping, and lower defect levels.
Claims
1. A method for preparing p-type silicon carbide single crystals using a multi-field time-coupling closed-loop controlled liquefaction method, characterized in that, The process sequentially includes a molten charge pretreatment step, a liquid phase molten pool construction step, a multi-field temporal coupled closed-loop solution-driven single crystal growth step, and a crystal in-situ annealing post-treatment step. The multi-field temporal coupled closed-loop solution-driven single crystal growth step operates a multi-field temporal coupled closed-loop solution system. This system includes a temperature field solution module, a molten flow field solution module, a carbon solute transport solution module, an aluminum doping multipath loss solution module, and a solid-liquid interface evolution solution module. The system is equipped with an iterative convergence determination unit. The system also features interface partitioning feature extraction. Taking a single unit, a time-sequential coupled iterative calculation is performed on the temperature field, melt flow field, and carbon solute transport field. The calculation is performed sequentially according to the crystal growth time step. After each time step iteration, the solid-liquid interface is spatially partitioned and analyzed to extract the migration rate and morphological shift information of different partition interfaces, identify the asymmetric warping evolution trend of the interface, and update the spatial position and partition morphological boundary of the solid-liquid interface. The updated interface information is used as the boundary input for the next round of iteration calculation. The boundary analysis logic of different orientation partitions of the interface is distinguished. The growth state of the solid-liquid interface simultaneously incorporates the crystallization contribution of temperature gradient and carbon and aluminum component transport.
2. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The aluminum doping multipath loss solving module is equipped with a dual-loss linkage coupling time-series decoupling solving module. The two types of losses are the aluminum volatilization loss path on the free surface of the melt and the aluminum adsorption loss path on the inner wall of the crucible. Following the growth sequence, the linkage coupling relationship between the two types of losses is analyzed in each round of iteration calculation, and the loss contribution of each type of loss is decoupled to solve the actual effective aluminum doping flux at the solid-liquid interface. The actual aluminum doping flux is compared with the preset target doping flux to obtain the flux difference. Based on the flux difference, a compensation control signal is generated iteratively. The compensation control signal is mapped to the crucible rotation speed adjustment action, the seed crystal small lifting action, and the fine adjustment action of the upper and lower hot zones of the molten pool. The interface doping flux compensation is achieved by changing the melt flow transport state and the solid-liquid interface contact state. No solid material is added to the crucible during the entire growth process.
3. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The melt flow field solution module is equipped with a graded intervention solution unit for hysteresis disturbance of solute transport in the molten pool; following the growth sequence, it analyzes the hysteresis evolution process of solute transport to the solid-liquid crystallization interface after the execution of the molten pool mechanism adjustment action in each round of iteration calculation; it identifies the strength level of the hysteresis disturbance, divides the intervention strategies corresponding to different disturbance levels, and matches the time evolution characteristics of solute transport hysteresis. When the identified hysteresis disturbance continues to accumulate and reaches the risk level, the corresponding hardware intervention signal is output.
4. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The multi-field time-series coupled closed-loop solution system receives measured data from temperature sensors located on multiple external parts of the furnace body, without placing built-in sensors inside the high-temperature molten pool. At the end of each time-series progressive calculation cycle, the physical quantities output by the model solution are compared with the measured physical quantities at multiple external measurement points on the furnace body to obtain residual information. The residual information is used to perform time-series adaptive calibration and update of multiple sets of solution boundaries of the solution system.
5. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The pretreatment process of the molten material includes: performing graded high-temperature vacuum degassing and purification treatment on the graphite carbon source; constructing a layered gradient carbon source dissolution and supply structure using graphite carbon sources of different particle sizes; implementing inert coating protection treatment on the aluminum P-type doped source; layering and filling the pretreated silicon substrate material, layered gradient-arranged graphite carbon source, and inertly coated aluminum doped source into the crucible; matching the initial mesh generation of the material layering and filling structure with the multi-field time-series coupled closed-loop solution system.
6. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The multi-field time-series coupled closed-loop solution system is deployed inside the local control system of the silicon carbide growth furnace. The solution system continuously reads the measured physical data of the furnace body sensors. After the time-series progressive calculation converges, it outputs the process control signal, which is then sent to the furnace body heating power supply, the crucible rotation drive mechanism, and the seed crystal lifting drive mechanism. The entire process of single crystal growth does not involve loading a pre-cured complete set of process curves, and the key process control points are dynamically generated by the time-series iterative calculation of a multi-field time-series coupled closed-loop solution system.
7. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The working logic of the iterative convergence determination unit is as follows: compare the changes in the solid-liquid interface partition morphology, the migration rate of each partition, the aluminum doping flux, and the residual of the molten pool flow field obtained from the calculations of two adjacent time steps; if all changes are less than the preset convergence threshold, iterative convergence is determined; if the convergence condition is not met, the solution boundary is updated and the next time step iterative calculation is entered.
8. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The solid-liquid interface evolution solution module automatically reconstructs the local solution mesh as the interface mesh nodes shift and move along with the crystallization.
9. The method for preparing P-type silicon carbide single crystals using the multi-field temporal coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, The cooling annealing process parameters for the post-in-situ annealing of crystals are output by the time-series iterative calculation of the multi-field time-series coupled closed-loop solution system.
10. The method for preparing P-type silicon carbide single crystals using the multi-field time-coupling closed-loop controlled liquefaction method according to claim 1, characterized in that, After the liquid phase molten pool construction process is completed, the multi-field time-coupled closed-loop solution system first performs the initialization time-series progressive calculation. After the initialization iteration converges, the seed crystal contact molten pool is started to carry out single crystal growth.
