Method for growing silicon carbide crystal by PVT method

Abstract

The invention discloses a method for growing silicon carbide crystals through a PVT method, and relates to the technical field of semiconductor material preparation, and the method comprises the following steps: S100, establishing a coupling model of a pressure perturbation spectrum and a gas phase migration process, and inverting a stable interval of a carbon-silicon component ratio based on long-period operation data to obtain a stable interval of a carbon-silicon component ratio; calibrating an interface energy critical value and a temperature gradient boundary range; and S200, under the constraint of an interface energy critical value and a temperature gradient boundary range, constructing a risk identification feature template, extracting an early signal of crystal growth interface step evolution by using in-situ Raman detection and reflection interference monitoring, and generating a space thermal stress distribution diagram according to signal distribution. According to the method, disturbance identification, thermal stress mapping and dual-channel regulation and control in crystal growth are realized, a model is adaptively updated in combination with stress feedback, a carbon-silicon proportion stable interval and a risk threshold are dynamically optimized, the stability, the yield and the automation level of silicon carbide crystal growth are improved, and the bottleneck of existing control lag and model rigidity is broken through.

Description

Technical Field

[0001] This invention relates to the field of semiconductor material preparation technology, specifically to a method for growing silicon carbide crystals using the PVT method. Background Technology

[0002] The PVT method for growing silicon carbide crystals refers to a process that uses physical vapor transport (PVT) to prepare high-quality silicon carbide single crystals. Its core principle is to sublimate silicon carbide raw materials into gaseous components (mainly volatile molecules such as Si, Si₂C, and SiC₂) at high temperatures. Driven by a temperature gradient, these gaseous components migrate from the high-temperature region to the low-temperature region, recrystallizing and depositing on the seed crystal surface in the low-temperature region, thus growing layer by layer a silicon carbide single crystal with complete volume and regular structure. This method avoids the difficulties in melt growth caused by excessively high melting points, and can obtain SiC crystals with low defect density and high purity. It is currently the mainstream process for preparing silicon carbide substrate wafers and is widely used in the preparation of core materials for high-power electronic devices and high-frequency devices.

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, while the overall temperature field and atmospheric conditions can maintain relative equilibrium during long-term stable growth of silicon carbide single crystals using the PVT method, the process is still susceptible to minor disturbances within the sealed cavity. For example, minute pressure fluctuations within the cavity can cause a sudden shift in the carbon-to-silicon atom ratio in the gas phase, disrupting the original deposition equilibrium. This abrupt change in ratio directly affects the interface between the seed crystal and the gas phase, transforming the previously stable layered growth mode into an unstable three-dimensional dendritic deposition state. This process is a transient phase transition instability phenomenon. Once it occurs, numerous irregular growth steps rapidly appear on the crystal surface, forming step aggregation zones. This leads to the accumulation of significant internal stress in localized areas, ultimately causing macroscopic cracks or even overall crystal cracking. While this type of problem is uncommon in existing technologies, its occurrence can render the entire silicon carbide crystal unusable, severely impacting single crystal quality and production yield.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for growing silicon carbide crystals using the PVT method, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for growing silicon carbide crystals using the PVT method, comprising the following steps:

[0008] S100, establish a coupled model of pressure perturbation spectrum and gas phase migration process, invert the stable range of carbon and silicon component ratio based on long-period operation data, and calibrate the critical value of interface energy and the boundary range of temperature gradient accordingly.

[0009] S200, under the constraints of the critical value of interface energy and the boundary range of temperature gradient, constructs a risk identification feature template, uses in-situ Raman detection and reflection interferometry monitoring to extract early signals of the evolution of crystal growth interface steps, and generates a spatial thermal stress distribution map based on the signal distribution;

[0010] S300 constructs a virtual sensing system for the carbon and silicon component ratio based on the spatial thermal stress distribution map. By integrating crucible temperature field data, raw material sublimation rate and component partial pressure trajectory, it continuously tracks the supply ratio offset and locates the offset path and key timing nodes.

[0011] S400 performs dual-channel closed-loop control operation based on the offset path and key timing nodes. On the one hand, it fine-tunes the temperature gradient to suppress the offset of the carbon and silicon composition ratio, and on the other hand, it adjusts the micro-valve damping to reduce the intensity of pressure fluctuation, thereby stabilizing the deposition flux on the seed crystal surface.

[0012] S500 applies small-amplitude compositional pulse perturbations under the condition that the deposition flux tends to be stable, monitors the change amplitude of the interface residual stress field, and adaptively adjusts the control gain parameters and time delay compensation model according to the residual change trend.

[0013] After the stability boundary of the residual stress field is verified, the residual distribution, offset trigger time and control response effect are used as feedback data input to the coupled model to update the stable range of carbon and silicon component ratio, and simultaneously update the risk identification threshold, thus realizing a closed-loop evolution process of modeling, identification and control.

[0014] Preferably, step S100 includes:

[0015] Based on a silicon carbide crystal growth experiment that has been running continuously and stably for more than 72 hours, data on sublimation zone temperature, seed crystal zone temperature, static pressure in the chamber, partial pressure of carbon atoms, partial pressure of silicon atoms, crystal growth rate, and crystal surface morphology were collected. The amplitude, duration, frequency, and composition change rate of each pressure disturbance event were calibrated.

[0016] Each set of perturbation samples is correlated with the partial pressure change data of carbon and silicon components in the gas phase. The net migration of carbon and silicon in the axial path per unit time before and after the perturbation is calculated, and the relationship between perturbation frequency and carbon-silicon ratio change is plotted to identify perturbation intervals with nonlinear abrupt change characteristics.

[0017] Based on the changes in the step structure on the crystal surface, the change in free energy per unit area during the deposition process at the crystal interface is derived. The minimum free energy value that causes the crystal interface stability to be destroyed is extracted as the critical value of the interface energy. Under the condition of constant perturbation intensity, the temperature gradient is adjusted to obtain the temperature gradient boundary range that suppresses proportional offset.

[0018] All parameters are input into the gas-phase migration coupling model, and simulation calculations are performed for each set of disturbance conditions. Based on the carbon-silicon ratio fluctuation, free energy fluctuation, and temperature gradient, it is determined whether it belongs to the stable range. Finally, a continuous range spectrum of whether the carbon-silicon supply is stable is generated.

[0019] Preferably, step S200 includes:

[0020] Based on the critical value of interface energy and the boundary range of temperature gradient, a detection window is set at the top of the reaction chamber, a Raman laser beam with an incident angle of 10 degrees and a wavelength of 532 nanometers is arranged, and a white light interference probe is arranged simultaneously to collect the Raman scattering signal of the crystal surface and the step height change data respectively.

[0021] The cumulative change in the Raman displacement peak exceeds 1.5 cm. -1 When the spectral width increases and the step abruptness in the interference signal exceeds three times and disappears periodically, it is marked as an unstable signal of interface deposition. Three-dimensional information units are generated by combining the detection coordinates and a crystal front unfolding diagram is constructed.

[0022] Extract regions with dense signal abrupt changes from the crystal front development diagram, and compare the temperature gradient, free energy change, and stress abrupt change frequency. If the boundary conditions are met simultaneously, they are marked as high-risk regions, and a risk identification feature template containing coordinates, time period, stress amplitude, and step jump rate is formed.

[0023] The feature template is mapped onto the three-dimensional crystal structure to generate a thermal stress spatial distribution map with a resolution of 0.2 mm and a step size of 0.5 GPa, which is then used as the input condition for subsequent proportional offset tracking.

[0024] Preferably, step S300 includes:

[0025] A cross-sectional thermal field measurement surface centered on the crystal growth axis was constructed, a 5×3 lattice network was set, an S-type platinum-rhodium-platinum thermocouple was installed, three-dimensional temperature field data were collected, and the region with the steepest temperature gradient was identified as the supply distribution sensitive area.

[0026] The sublimation rate of raw materials was determined using a thermoelectric balance, and the actual sublimation rates of carbon and silicon were calculated by combining thermal field data. A sublimation source term distribution map was generated and high-risk source areas with sublimation rates exceeding 20% ​​of the average value were marked.

[0027] Infrared spectrophotometers were set up at the top of the raw material area, the convection area and the seed crystal area to measure the partial pressure of Si, Si2C and C2, and a two-dimensional curve of the carbon-silicon partial pressure ratio was plotted to determine whether a supply ratio shift had occurred and to conduct preliminary verification by comparing the coordinates of the thermal stress diagram.

[0028] Based on the temporal and spatial evolution relationship between the initial point of partial pressure mutation and the intersection point of thermal stress mutation, key time segments are defined, spatial coordinates and thermal field anomaly locations are extracted, and a coupling correspondence between the supply offset path and the deposition response is established.

[0029] Preferably, step S400 includes:

[0030] Based on the starting point and spatial location of the carbon and silicon component ratio offset path, regions with temperature gradients below the boundary range are identified in the three-dimensional temperature field. By adjusting the power of the raw material heating section and the convection zone heating section, the local temperature gradient is increased to the set range.

[0031] Infrared thermal imaging and in-situ interferometry were used to simultaneously monitor the temperature difference and step changes on the crystal surface. Raman spectroscopy was combined to confirm whether the stress had recovered to the stable range and to judge the effect of temperature regulation.

[0032] The opening of the electric needle valve on the atmosphere inlet guide tube is gradually reduced to reduce the static pressure fluctuation of the cavity to below ±0.2Pa, thereby ensuring the stability of the gas phase component migration process.

[0033] The four indicators of sedimentary step advancement rate, stress change, temperature difference distribution and pressure fluctuation are jointly judged to determine whether they meet the standards. After confirming that the sedimentation flux has recovered and stabilized, the control parameters are locked and maintained until the next offset path appears.

[0034] Preferably, step S500 includes:

[0035] Under stable deposition flux conditions, a temperature disturbance was implemented by increasing the electric heating power of the raw material zone from 8.0 kW to 8.2 kW, resulting in a slightly enriched silicon composition that was maintained for 20 minutes.

[0036] Raman spectra were collected every 2 minutes during the disturbance process. The amplitude of the interface stress response was determined and the stability of the sedimentary morphology was evaluated by combining the step height change data.

[0037] The adjustment step size of the raw material heating power was optimized to 4% based on the stress response intensity, the control response cycle was shortened to 10 minutes, and the micro-valve adjustment step size and pressure refresh frequency were increased.

[0038] After the disturbance ends, continue recording the deposition flux and stress curves for 30 minutes. If the step advancement speed and stress value fluctuations remain within ±5% of the pre-disturbance range, then record this state as the steady-state recovery boundary and update the control strategy reference dataset.

[0039] Preferably, step S600 includes:

[0040] After a small-amplitude perturbation trial is completed, continuous Raman spectrum data of the pre-, mid- and post-perturbation stages are extracted to clarify the changing trends of the main peak displacement, half-width at half-maximum and step advance rate, and to form a perturbation-response-recovery dataset.

[0041] The dataset was embedded into the original coupled model, the stable range of carbon and silicon component ratio was corrected, the upper limit of partial pressure ratio was updated to 1.07, the upper limit of sublimation temperature was updated to 2267℃, and boundary validity confirmation conditions were set.

[0042] Based on the revised model, the risk identification criteria have been updated, and the primary warning threshold for main peak displacement has been raised to ±1.5 cm. -1 Furthermore, a continuous cycle judgment mechanism and a spatial risk clustering judgment standard were introduced;

[0043] In the next growth cycle, the initial settings are executed according to the updated parameters, and the control channel is called in real time according to the stress fluctuation trend. At the same time, the control results are written back to continuously optimize the model, realizing a closed-loop update of modeling, identification and control.

[0044] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0045] This invention constructs a coupled model of pressure perturbation and gas phase migration, and combines in-situ optical monitoring to obtain crystal interface step evolution and residual stress signals. It establishes a correlation mechanism between the spatial thermal stress distribution map and the supply offset path. Based on this, it achieves highly responsive steady-state control through dual-channel fine-tuning of temperature and pressure. Furthermore, it adaptively corrects the control parameters through small-amplitude perturbation trials and stress recovery judgments, and writes the process response data back into the model to achieve dynamic evolution of the stable range and risk identification threshold. This effectively avoids irreversible structural instability caused by a single perturbation, fundamentally improving the yield, quality consistency, and automation level of silicon carbide single crystal growth, and overcoming the technical bottlenecks caused by monitoring delay, control lag, and model rigidity in existing technologies. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0047] Figure 1 This is a flowchart of a method for growing silicon carbide crystals using the PVT method according to the present invention. Detailed Implementation

[0048] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0049] This invention provides, for example Figure 1 The method for growing silicon carbide crystals using the PVT method, as shown, includes the following steps:

[0050] S100, establish a coupled model of pressure perturbation spectrum and gas phase migration process, invert the stable range of carbon and silicon component ratio based on long-period operation data, and calibrate the critical value of interface energy and the boundary range of temperature gradient accordingly.

[0051] To address the problem of abrupt changes in the carbon-to-silicon atom ratio in the gas phase caused by minute pressure disturbances within the cavity, leading to a shift from a stable layered mode to an unstable dendritic mode at the crystal growth interface, a method is proposed to establish a gas-phase disturbance response model based on long-term operational data. This method establishes the causal relationship between disturbances and component migration behavior step by step, and forms a quantitative basis for stability determination through parameter inversion and boundary calibration. The method includes the following steps:

[0052] Data from a silicon carbide crystal growth experiment that ran continuously and stably for over 72 hours was used as the basis for analysis. Key parameter data within the reaction chamber were collected throughout the crystal growth process. The collected data included: ① Sublimation zone temperature, sampled once per second with an accuracy of ±1℃; ② Seed crystal zone temperature, sampled at the same frequency as the sublimation zone; ③ Static pressure changes within the chamber, collected using a pressure sensor with an accuracy of 0.1 Pa; ④ Partial pressure data of carbon and silicon, periodically collected every 1 minute using an infrared spectrophotometer; ⑤ Crystal growth rate and surface morphology data, acquired every 10 minutes using an online laser confocal scanner to obtain crystal surface profile data. The above data were precisely synchronized using timestamps to identify pressure disturbance events occurring during the growth period. The characteristic parameters of each event included disturbance amplitude (Pa), duration (seconds), disturbance frequency (Hz), and the rate of change of atmosphere composition before and after the disturbance. Statistical analysis was used to generate multiple typical disturbance samples for subsequent modeling.

[0053] The perturbation samples obtained in the first step were correlated with the changes in the partial pressures of carbon and silicon components in the gas phase during the corresponding time periods. The rate of change of the partial pressures of carbon and silicon atoms in the gas phase before and after the perturbation was calculated using a piecewise integration method. By comparing the component migration rates before and after the perturbation, and combining this with the temperature gradient values ​​of the crucible and seed crystal areas, the net migration of carbon and silicon in the axial gas flow path per unit time was further calculated. Based on multiple perturbation events, a two-dimensional relationship graph between the perturbation frequency and the change in the carbon-silicon ratio was plotted, identifying perturbation combinations with obvious nonlinear abrupt changes. The changing trend of the crystal surface deposition morphology under each perturbation condition was further analyzed, and the turning point from smooth steps to convex dendrites in the deposition process was used as the criterion for growth instability. Based on comparative analysis, the range of perturbation parameters that could maintain the carbon-silicon ratio fluctuation within ±5% and maintain layered growth of the deposition morphology was extracted, and this range was defined as the initial value of the carbon-silicon supply stability boundary.

[0054] Repeated growth experiments were conducted under multiple perturbation conditions near the stable boundary of the carbon-silicon supply ratio. After each experiment, X-ray diffraction and scanning electron microscopy imaging analysis were performed on the surface step structure of the crystal slices. Combining temperature distribution data, carbon-silicon supply ratio data, and surface stress trends at various time points during growth, the change in free energy per unit area of ​​the interface during deposition was derived based on the principles of energy and mass conservation. Using the energy point where a significant abrupt change in interface morphology occurred as the criterion, the lowest energy value that caused interface instability under perturbation conditions was extracted as the interface energy critical value. Simultaneously, the temperature gradient was gradually adjusted under the same perturbation intensity. By observing the crystal morphology stability under different gradient values, the minimum and maximum temperature gradient values ​​that could effectively suppress the amplification effect of carbon-silicon ratio perturbation were identified as the temperature gradient boundary range. After completing this step, the two physical criteria for determining whether carbon-silicon ratio perturbation will cause interface instability—the energy threshold value and the transport buffering capacity—can be clearly defined.

[0055] All perturbation conditions with calibrated parameters are re-inputted into the aforementioned gas-phase migration coupling model to simulate the migration trend of the carbon-silicon ratio after perturbation triggering, the evolution trajectory of crystal surface free energy, and the buffering effect corresponding to the temperature gradient. For each set of perturbation trigger points, it is determined whether the following conditions are met: ① carbon-silicon ratio fluctuation does not exceed ±5%; ② free energy does not exceed the interface energy critical value; ③ temperature gradient is within the defined boundary range. If all three conditions are met, it is considered an acceptable perturbation, and its parameter combination is added to the stable interval. If any condition is not met, the combination is removed, and the energy or temperature gradient boundary is fine-tuned based on actual deposition failure characteristics to update the model's judgment range. Ultimately, a clear set of correspondences between perturbation frequency, perturbation amplitude, gas phase component change rate, and temperature gradient is formed, serving as a continuous interval spectrum for judging whether the carbon-silicon supply is stable. This spectrum can be directly used as a real-time judgment standard in the actual crystal growth process, ensuring that any subsequent perturbation behavior will have clear physical boundary support for whether it will cause crystal instability, thus ensuring the targetedness and effectiveness of the control measures.

[0056] This step aims to construct a physical mechanism model based on real process operation data to quantify the impact of minute pressure disturbances on the carbon-to-silicon ratio in the gas phase, thereby identifying and determining potential triggers for interfacial instability during crystal growth. By establishing the coupling relationship between the pressure disturbance spectrum and the migration behavior of carbon and silicon components, and combining multiple sets of disturbance samples for parameter inversion and interfacial behavior analysis, this step clarifies the two key physical boundary conditions upon which interfacial deposition stability depends: the critical interfacial energy value and the temperature gradient boundary range. Furthermore, through the systematic identification and determination of the disturbance response, a criterion for determining a stable range can be established to assess whether the gas phase supply is in a safe state, avoiding the risk of growth interface instability caused by sudden component shifts. This step not only provides a clear physical basis for subsequent risk identification, shift tracking, and control operations but also significantly improves the predictability and control accuracy of the crystal growth process, serving as a crucial foundation for shifting from passive response to proactive prevention.

[0057] S200, under the constraints of the critical value of interface energy and the boundary range of temperature gradient, constructs a risk identification feature template, uses in-situ Raman detection and reflection interferometry monitoring to extract early signals of the evolution of crystal growth interface steps, and generates a spatial thermal stress distribution map based on the signal distribution;

[0058] To achieve interface risk identification during crystal growth, based on the established critical values ​​of interface energy and the boundary range of temperature gradients, a risk identification method combining in-situ detection technology, interface morphology monitoring, and spatial stress mapping is proposed. This method establishes a risk identification feature template to extract early signals of potential depositional instability at the growth interface and generates a thermal stress distribution map of the crystal surface through spatial projection, providing precise input for subsequent control decisions. The specific steps include:

[0059] The critical interfacial energy value obtained in the previous embodiment (e.g., the critical free energy change is 0.25 J / m) 2 The precise detection of the microscopic morphological evolution and stress changes on the crystal surface during growth is crucial, given the temperature gradient boundaries (e.g., within the seed crystal region at 2000℃, a temperature gradient between 3.5K / mm and 5.0K / mm is permissible). To address this, a 25mm diameter detection window is created at the top of the reaction chamber, and a continuous laser beam with a wavelength of 532nm and a power of 100mW is positioned at a 10° angle to form the Raman beam illumination path. The beam is focused by a lens onto the central region of the seed crystal surface, with an incident spot diameter no greater than 50μm. Raman scattering signals from the crystal surface are collected using a backscattering light collection device. A complete Raman spectrum is acquired every 3 minutes, and the Raman shift change and the full width at half maximum (FWHM) of the spectral peak are used as indicators of stress change. Simultaneously, a white light interference probe is positioned 0.5mm beside the Raman measurement point. Using the Fresnel reflection interference principle, the periodic changes in the reflected interference fringes are used to determine the dynamic changes in the height of the surface deposition steps. The interference signal sampling frequency is 1 Hz, and the resolution is controlled within ±10 nm. The two types of detection signals mentioned above need to undergo temperature-protected encapsulation and infrared filtering to ensure that the detection process is not interfered with by high-temperature radiation from the cavity.

[0060] In the acquired Raman spectral data, the displacement peak is at the initial reference state (e.g., 776 cm⁻¹). -1 When a red shift (peak value decreases) or blue shift (peak value increases) occurs near the peak, and the peak width (FWHM) changes from the initial 8cm... -1 Expand to 10cm -1 The above information allows for a preliminary assessment of changes in the surface stress of the crystal. The cumulative displacement change exceeding 1.5 cm over two consecutive periods is considered a positive result. -1 This is considered a Raman stress signal abrupt change event. Simultaneously, if the step height in the interference signal exhibits at least three abrupt changes within 10 consecutive minutes, with each change exceeding 80 nanometers in amplitude, and the step periodically disappears, then the crystal surface deposition state is considered to have transitioned from a stable step to unstable dendrite growth.

[0061] To achieve spatial positioning of the risk area, the coordinates of the detection point are unified with the crystal coordinate system inside the cavity, and spatial coordinate labels are set according to the diameter and axial directions. Each detection event is marked as a three-dimensional information unit (r, z, t), where r is the radial position (in millimeters), z is the axial distance (in millimeters), and t is the detection time (in minutes). Based on this, a two-dimensional unfolded diagram of the crystal growth front interface is constructed. Each signal abrupt change is precisely marked on the diagram, and the corresponding Raman stress value and step height change amplitude are recorded.

[0062] In each round of crystal growth, the signal mutation units recorded in the previous step are continuously extracted, and their density distribution per unit time and per unit area is statistically analyzed. For example, taking 1 square millimeter as the area unit, if the area experiences more than 3 Raman shift mutations within 30 minutes, accompanied by highly abrupt changes in the interference signal, then the area is identified as a risk feature area. For each risk area, its corresponding temperature gradient value (obtained by thermocouple array recording) and the free energy value of the crystal surface at that time (calculated in real time by free energy estimation model) are further retrieved. If the area meets the following three conditions: (1) the temperature gradient value is less than 5% of the lower limit of the temperature gradient boundary; (2) the change in free energy exceeds 0.25 J / m of the critical free energy change value. 2 (3) If the frequency of stress mutation exceeds the threshold of 3 times / hour set by the Raman spectrum, the area is classified as a high-risk area, and its spatial coordinates, corresponding time period, stress mutation amplitude, and step jump rate are marked in the risk identification feature template. This template is ultimately used as a standard library for interface instability prediction and for subsequent instability trend judgment.

[0063] High-risk areas identified in the template are mapped back to the actual 3D crystal model structure one by one, and the points are restored according to the coordinate labels (r, z, t). Combining the Raman stress value of each point, a spatial distribution map of thermal stress on the crystal's leading edge surface is generated using a color gradient method. The color ranges from dark blue (zero stress) to dark red (high stress), with a gradient step size of 0.5 GPa. The plotting granularity is controlled at 0.2 mm, and the map is updated every 10 minutes, continuously generated until the crystal growth process is complete. Finally, a set of thermal stress distribution maps that evolve synchronously with the time axis is obtained. This is used to analyze stress concentration areas, determine possible crack initiation points, and serve as a key input for subsequent proportional offset tracking, assisting in tracing the offset source, formulating compensation strategies, and determining the target adjustment path.

[0064] The purpose of this step is to, under the premise of knowing the critical value of the interface energy and the boundary range of the temperature gradient, use in-situ Raman detection and reflection interferometry to perceive the minute changes in the interface deposition state during the growth of silicon carbide crystals in real time. By extracting stress abrupt changes and step structure anomaly signals, a risk identification feature template for the growth process is constructed. This template can be used to identify which regions on the crystal surface have entered a potentially unstable state, thereby predicting the dendrite growth trend and the location of stress concentration areas in advance. Furthermore, by mapping these high-risk feature regions in space, a spatial distribution map of thermal stress at the crystal front is generated, visualizing the stress evolution path during growth. This map not only provides accurate spatial input for subsequent control processes but also significantly improves the response speed and accuracy of crystal defect risk assessment, effectively overcoming the limitations of existing technologies that cannot achieve real-time, visual, and quantitative risk identification, and providing a forward-looking guarantee for crystal quality control.

[0065] S300 constructs a virtual sensing system for the carbon and silicon component ratio based on the spatial thermal stress distribution map. By integrating crucible temperature field data, raw material sublimation rate and component partial pressure trajectory, it continuously tracks the supply ratio offset and locates the offset path and key timing nodes.

[0066] In the physical vapor transport method for silicon carbide crystal growth, the carbon-to-silicon ratio in the supply atmosphere directly determines the stability of the crystal surface deposition process. When this ratio deviates, especially when the critical interface energy value and temperature gradient boundary are exceeded, it easily induces the crystal growth interface to transform from a two-dimensional step mode to a three-dimensional dendritic mode, leading to step accumulation, stress concentration, structural distortion, and even crystal cracking. To achieve non-contact continuous tracking of the carbon-to-silicon supply ratio, this step constructs a virtual sensing method for the carbon-to-silicon component ratio based on the previously obtained spatial thermal stress distribution map. This method dynamically estimates the actual supply ratio of carbon and silicon components by integrating the spatial distribution of the crucible temperature field, the real-time changes in the raw material sublimation rate, and the partial pressure trajectory of carbon and silicon components in the gas phase, and identifies the deviation path and its key timing nodes. The method includes the following steps:

[0067] Inside the reaction chamber, a vertical cross-section centered on the crystal growth axis was selected as the thermal field measurement surface. Five measuring points were established along this cross-section from bottom to top: the center point at the bottom of the raw material zone, the center point at the upper edge of the raw material zone, the lower edge point of the central convection zone, the upper edge point of the seed crystal support structure, and the center point of the seed crystal. Each measuring point had three radial sub-points at distances of 0 mm, 5 mm, and 10 mm from the central axis, forming a 5×3 point lattice network. High-temperature thermocouples (Type S platinum-rhodium / platinum thermocouples) were installed at each point, with a temperature range of 0℃ to 2500℃ and a measurement accuracy controlled within ±0.5℃. Each thermocouple had a sampling period of 1 second and a continuous operating time of more than 72 hours. The point temperature data obtained from each sampling were expanded into a three-dimensional temperature field matrix using coordinate interpolation, and a complete thermal field update was performed every 10 minutes. By performing spatial gradient analysis on the three-dimensional thermal field matrix, the region with the steepest temperature change was identified as a sensitive region for supply distribution, providing thermal driving force boundary conditions for subsequent sublimation behavior analysis.

[0068] The sublimation rate of silicon carbide raw material at the bottom of the crucible was measured using a high-temperature weighing device. The initial loading of the raw material was 100.00 grams, evenly spread on the graphite lining surface at the bottom of the crucible with a diameter of 50 mm. Preheating calibration was performed using a thermoelectric balance structure mounted on the external base of the crucible before the cavity was closed. This balance device had an accuracy of ±0.01 grams, a sampling frequency of once every 20 seconds, and could operate stably for more than 90 hours. During the growth process, the relationship between the reduction in raw material mass and time was continuously recorded, and combined with the aforementioned thermal field data, the actual sublimation rate of the raw material under each temperature condition was calculated. Considering that the silicon component in the decomposition products of silicon carbide at high temperatures is easily volatile while the carbon component has a low sublimation efficiency, the sublimation behavior needs to be refined into two components: silicon sublimation rate and carbon sublimation rate. The contribution of the temperature field to the release of different components was compared with existing experimental data on pyrolysis reaction rates. The sublimation rate is weighted according to the spatial heat distribution to form a sublimation source term distribution map in three-dimensional space. Areas with sublimation rates exceeding the average by more than 20% are highlighted as high-risk sources of supply shift.

[0069] Three sets of non-contact infrared spectrophotometers are installed at three locations: the top of the raw material zone, the middle convection zone, and directly above the seed crystal zone. Multi-component infrared absorption spectroscopy is used to measure the partial pressures of Si, Si₂C, and C₂ molecules in the gas phase in real time. Each set of detectors operates over a wavelength range of 2000 cm⁻¹. -1 Up to 5000cm -1 The absorption peak positions correspond to Si (4500 cm⁻¹) and Si (4500 cm⁻¹) -1 ), Si2C (3100cm) -1 ) and C2 (2800cm -1The sampling period is once every 30 seconds, and the partial pressure estimation error is less than ±5%. Combining the spatial coordinates corresponding to each detector, the partial pressure change trend is plotted as a two-dimensional time-space curve for carbon pressure and silicon pressure, respectively. The slope, peak value change, and stabilization time of each curve are recorded as evaluation indicators of the proportional change trend. In each 10-minute cycle, if the carbon-silicon partial pressure ratio at any measuring point continuously deviates from the previous cycle's stable value by more than ±5%, and this change coincides with the coordinates of the stress concentration area in the thermal stress distribution diagram, it is judged that a supply offset event has initially occurred.

[0070] After identifying the supply offset event, the partial pressure trajectory is traced back along the time series. The location where the earliest change in partial pressure ratio occurs is set as the starting point of the supply offset, and its coordinates in three-dimensional space are recorded as (x0, y0, z0), with the occurrence time recorded as T0. Starting from this starting point, the spatial points where the carbon-silicon partial pressure ratio continues to deviate at each subsequent moment are recorded along the time axis, forming a three-dimensional coordinate sequence of the supply offset path. Furthermore, the intersection point of this path with the stress abrupt change contour line in the thermal stress distribution map is extracted, and the time of this intersection point is recorded as T1. If T1-T0 is within 30 minutes, and the free energy change amplitude in the intersection area reaches 0.25 J / m, then the supply offset is considered complete. 2 Therefore, the T0-T1 time period is defined as the critical time segment, serving as the main risk window for the migration evolution process. Combined with the spatial distribution data of the temperature field at this time, it is determined whether there is an abnormal temperature drop or rise exceeding the boundary value range in a certain area. This further identifies the direct coupling relationship between supply migration behavior and interface deposition response, providing spatial and temporal coordinates for target setting and path selection in subsequent intervention operations.

[0071] The purpose of this step is to eliminate the need for direct measurement of carbon and silicon content during crystal growth. By integrating the three-dimensional temperature field distribution inside the crucible, changes in the raw material sublimation rate, and the partial pressure trajectories of carbon and silicon components, a virtual supply ratio tracking method is constructed, enabling real-time identification and dynamic analysis of carbon and silicon component ratio shifts. This method allows for the early detection of unbalanced carbon and silicon supply trends before interface instability occurs, identifying the starting position, evolution path, and key temporal nodes of the supply shift. It also allows for spatial linkage judgment with stress anomaly regions in the thermal stress distribution map, accurately pinpointing potential instability sources. This step effectively solves the problem of real-time monitoring of component ratios in high-temperature, sealed environments in existing technologies. It possesses the advantages of non-contact, high-resolution, and continuous monitoring, providing clear adjustment targets and feedback for subsequent closed-loop control operations, and is one of the core prerequisites for achieving stable crystal quality control.

[0072] S400 performs dual-channel closed-loop control operation based on the offset path and key timing nodes. On the one hand, it fine-tunes the temperature gradient to suppress the offset of the carbon and silicon composition ratio, and on the other hand, it adjusts the micro-valve damping to reduce the intensity of pressure fluctuation, thereby stabilizing the deposition flux on the seed crystal surface.

[0073] In the PVT process for growing silicon carbide crystals, the aforementioned steps have achieved continuous tracking of the carbon-to-silicon composition ratio shift and clarified the spatial path and key timing nodes of the shift. To prevent the interface deposition instability caused by the carbon-to-silicon ratio shift from further expanding, this step proposes a dual-channel closed-loop control method based on physical causal path response. This method stabilizes the actual deposition flux on the crystal surface by jointly controlling the temperature gradient and cavity pressure perturbation, ensuring continuous crystal growth in a step-by-step propagation mode. The entire process includes the following steps:

[0074] Based on the starting point and evolution direction of the offset path identified in the previous step, the coordinate information of the spatial region where the path is located is extracted in the three-dimensional crucible thermal field. Assuming the offset starting point is located in the region between z = 120 mm in the height direction of the crucible axis and r = 0 to r = 10 mm in the radial direction, combined with the current thermal field measurement data, the temperature gradient in this region is found to be 3.6 K / mm, lower than the calibrated lower limit of stability of 4.5 K / mm. To improve the temperature gradient in this region, the heating power configuration needs to be adjusted. The power controller output of the raw material heating section is set to 8kW, adjusted to 8.6kW; the convection zone heating section is increased from 6.2kW to 6.6kW, while the seed crystal heating section remains at the original setting of 5.5kW to ensure a difference in heating contrast. The adjustment sequence is as follows: first, the raw material section is increased to raise the sublimation end temperature by approximately 15°C, then the convection zone power is finely adjusted to further widen the axial temperature difference. Each round of power adjustment lasts for 10 minutes, after which the power is kept constant and the system response is observed.

[0075] During the heating power adjustment process, an infrared thermal imaging device positioned 5mm above the seed crystal was used to capture thermal images of the interface area. The thermal imager had a field of view radius of 20mm, a spatial resolution of 0.3mm, and a thermal sensitivity not exceeding 0.3℃, acquiring one frame every 10 seconds. Simultaneously, the deposition step height variation spectrum was obtained using the aforementioned in-situ interferometer. Each frame corresponded to one deposition flux measurement, with a repetition interval of 1 minute. If the step advancement rate increased from less than 0.05μm / min to 0.09μm / min in three consecutive images, and there were no local fractures in the step uniformity, it indicated that the temperature adjustment was initially effective. In-situ Raman detection was simultaneously activated, with 532nm laser excitation at the center point of the crystal surface, recording Raman shift changes. If the spectral line shift remained within ±1.0cm over two cycles... -1 Within, the peak width is controlled at 8–10 cm. -1If the above standard is not met, the interfacial stress is considered to be in a stable range. If the above standard is not met, the power in the raw material zone needs to be further increased or the power distribution in the heating section needs to be reconfigured until the temperature difference in the thermal image reaches the set value.

[0076] An electrically operated micro-needle valve is installed on the atmosphere inlet guide tube. The valve body is made of high-temperature resistant molybdenum alloy, with a diameter of 2.5 mm, a stroke range of 0.1 mm to 1.2 mm, a step adjustment accuracy of 0.01 mm, and a response delay of no more than 0.2 seconds per response. The micro-valve is connected to a feedback driver, with an initial set opening of 0.80 mm. During temperature adjustment, if the pressure sensor shows that the static pressure fluctuation range inside the chamber exceeds ±0.5 Pa, the needle valve opening is immediately reduced to 0.70 mm via step adjustment, and the adjustment time is recorded. If the pressure fluctuation range drops to ±0.2 Pa within the next 10 minutes, this opening is maintained for stable operation for 30 minutes; if the fluctuation is still above ±0.5 Pa, it is further reduced to 0.65 mm, and three rounds of feedback are observed. The pressure sensor is located 20 mm above the seed crystal area, with a measurement resolution of 0.05 Pa and a sampling frequency of once per second. This method weakens the source of low-frequency disturbances, making the gas phase migration process inside the chamber more stable and helping to maintain a stable transport rate of carbon silicon components.

[0077] After the dual-channel control is completed, the observation phase begins. The judgment criteria are: (1) the deposition step advancement rate is maintained above 0.10 μm / min for 3 consecutive rounds, and the step thickness fluctuation is less than 100 nm; (2) the Raman displacement is maintained at ±1.0 cm. -1 Within this range, the peak width fluctuation does not exceed 2cm. -1 (3) The temperature difference in the infrared thermogram is controlled within ±5% of the set gradient, and there are no local cold spot areas; (4) The pressure fluctuation in the cavity is consistently less than ±0.2 Pa. If any of the above four indicators fails to meet the standard, the temperature power or the micro-valve opening needs to be finely adjusted until all indicators meet the standard within two consecutive 10-minute cycles. After the conditions are met, the power and valve position are fixed and maintained until the next offset path appears or the crystal growth cycle ends.

[0078] The purpose of this step is to proactively intervene in the main physical causes of the shift in carbon and silicon component supply—uneven temperature gradients and cavity pressure disturbances—after accurately identifying the path and key timing nodes of the shift, thereby restoring the stability of the gas phase component supply and preventing the crystal growth interface from entering an unstable deposition state. By setting adjustable heating power, the temperature difference between the sublimation source and the deposition region is precisely adjusted, allowing the migration rates of carbon and silicon in the gas phase to rebalance. Simultaneously, by adjusting the damping degree of the needle valve, the amplitude of low-frequency pressure fluctuations inside the cavity is reduced, preventing disturbances from being transmitted to the crystal surface and causing further fluctuations in the supply ratio. The synergistic effect of these two factors restores the deposition flux on the seed crystal surface to a uniform and continuous step-by-step progression mode, suppressing dendrite formation and local stress concentration. Unlike existing technologies that rely on empirical temperature adjustment or post-defect repair, this dual-channel closed-loop control mechanism can achieve real-time response and physical quantity feedback control during the growth process, exhibiting higher process adaptability and stability. It is a key operational step to ensure high-quality and stable growth of silicon carbide single crystals.

[0079] S500 applies small-amplitude compositional pulse perturbations under the condition that the deposition flux tends to be stable, monitors the change amplitude of the interface residual stress field, and adaptively adjusts the control gain parameters and time delay compensation model according to the residual change trend.

[0080] To further confirm the stability of the interfacial stress state and the responsiveness of the control system after the crystal deposition flux has stabilized, and to effectively optimize the feedback characteristics of the temperature gradient and pressure regulation channels in the previous stage, this step involves applying controlled component perturbation to guide a predictable slight stress response at the crystal interface. Based on the stress change trend, the gain parameters and response period of the regulation channels are optimized and adjusted, thereby improving the system's ability to quickly respond to the next round of supply offset events. This process has the advantages of minimal intervention, rapid feedback, and clear regulation, and includes the following steps:

[0081] After the dual-channel control is completed in the previous stage, the deposition state on the surface of the seed crystal area needs to be continuously monitored for at least 60 minutes, and the following indicators must be confirmed to meet the standards: (1) The step advancement speed is stable between 0.10 μm / min and 0.12 μm / min, with fluctuations not exceeding ±0.01 μm / min; (2) The Raman displacement variation range is maintained within ±1.0 cm⁻¹, and the half-width at half-maximum is maintained between 8.0 and 10.5 cm⁻¹; (3) The cavity pressure fluctuation range is less than ±0.2 Pascals. After meeting the above three conditions, the disturbance stage is entered. In this stage, a small temperature increase is applied to the raw material heating zone, and the power of the electric heater in the raw material zone is increased from the initial 8.0 kW to 8.2 kW, corresponding to a temperature increase of about 15 degrees Celsius, from 2250℃ to 2265℃. The rise time is controlled within 15 minutes, and the power change rate is 0.013 kW / min. This heating process does not change the set temperature of the seed crystal region or convection region; it only affects the sublimation source release rate, causing a slight increase in silicon molecule content and creating a "silicon bias" disturbance on the supply side. The entire disturbance is maintained for 20 minutes, and no further heating is performed.

[0082] Starting from the first minute of the heating disturbance, an in-situ Raman scattering detection device was activated. The Raman source wavelength was set to 532 nm, the excitation power to 100 mW, the incident angle to 90 degrees, and the detection point to the geometric center of the seed crystal surface. The detection cycle was set to acquire a complete Raman spectrum every 2 minutes, for a total of 6 sets of data. For each set of spectra, the main peak shift value and the full width at half maximum (FWHM) were extracted and compared with the baseline data before the disturbance. If the shift value continuously increased, with a cumulative change exceeding +0.8 cm⁻¹, and the FWHM increased by more than 0.5 cm⁻¹, it was judged as an enhanced interface stress response; if the shift value decreased, with a cumulative change within -1.0 cm⁻¹, and the FWHM decreased by more than 0.5 cm⁻¹, it was considered as stress relief. To enhance the accuracy of the judgment, the change in the step height of the crystal surface in the interferometric measurement data was also recorded, once per minute, with a resolution not exceeding ±10 nm. If the step advance rate remains unchanged or decreases slightly after the disturbance, but there are no abnormal morphology such as step collapse or aggregation, it is confirmed that the deposition process has not been disrupted.

[0083] Based on the stress response characteristics obtained in the previous step, if the Raman shift fluctuation range is less than ±1.0 cm⁻¹, and the change in the full width at half maximum (FWHM) of the spectral peak is less than 1.0 cm⁻¹, while the interface morphology remains uniform and flat, it indicates that the current gain level of the control system is appropriate and no adjustment is needed. If the stress fluctuation exceeds ±1.5 cm⁻¹, or the peak width amplifies by more than 1.5 cm⁻¹, and step accumulation or sudden velocity changes occur on the deposition surface, the feedback response is considered insufficient, and the gain of the control channel should be increased. In the temperature channel, the adjustment step size of the raw material heating power is increased from 3% to 4%, meaning that the original power of 8.0 kW can be increased to 8.32 kW in one adjustment; and the control feedback response cycle is shortened from 15 minutes to 10 minutes to improve the response rate. In pressure control, the microvalve opening adjustment step size is adjusted from 0.02 mm to 0.03 mm, while the pressure sensor data refresh cycle is shortened from 10 seconds to 5 seconds. After the above optimizations, the current disturbance state is maintained for another 15 minutes to monitor whether the stress curve trend tends to stabilize.

[0084] After the optimized parameters are implemented, the crystal deposition process data is recorded for another 30 minutes. If the crystal surface step-advance speed is confirmed to be stable within ±5% of the pre-disturbance range, the deposition morphology shows no dendrite tendency, the Raman shift value fluctuation is stable within ±0.8 cm⁻¹, and the peak width remains between 8.5 and 10.5 cm⁻¹, then this is confirmed as a new steady-state boundary point. The disturbance intensity (temperature rise), response characteristics (stress fluctuation value, morphology change), and gain settings (power step size, adjustment period, microvalve step size) corresponding to this boundary point are archived as a control strategy reference set. This strategy set will be used for rapid recall in the next growth cycle should similar disturbance signals occur, avoiding the need to re-explore parameters from the initial state, thereby improving control efficiency, reducing response delay, and enhancing crystal quality stability.

[0085] The purpose of this step is to actively stimulate a weak stress response at the crystal growth interface by introducing small-amplitude compositional perturbations after the crystal deposition flux has reached initial stability. This verifies whether the current deposition state has the ability to maintain a steady state and evaluates the feedback response effect of the temperature and pressure control channels. By frequently monitoring the changes in Raman shift, peak half-width at half-maximum (FWHM), and step morphology induced by the perturbation, it is possible to determine whether there is potential residual stress accumulation or control hysteresis at the interface. Based on these real-time response signals, key control parameters such as the heating power adjustment step size, feedback response interval, and micro-valve adjustment amplitude are adjusted to more closely match the control strategy with the current physical state of the crystal, thereby improving the sensitivity and robustness of the control. Finally, by verifying whether the deposition morphology after perturbation is stable and whether the stress has returned to equilibrium, a new steady-state boundary identification standard is established, and the optimization results are used as a reference for rapid response in subsequent growth stages. This step realizes the transition from stability detection to enhanced control and is an important bridge link in forming an adaptive closed-loop control system and achieving continuous and stable growth of high-quality silicon carbide crystals.

[0086] After the stability boundary of the residual stress field is verified, the residual distribution, offset trigger time and control response effect are used as feedback data input to the coupled model to complete the update of the stable range of carbon and silicon component ratio, and simultaneously update the risk identification threshold to realize the closed-loop evolution process of modeling, identification and control.

[0087] After completing the initial small-scale perturbation and feedback control in the previous stage, the crystal deposition interface achieved stress stability within the set perturbation range. The deposition step morphology maintained a continuous and regular progression, and the main peak shift and full width at half maximum (FWHM) in the Raman spectrum were within acceptable fluctuation ranges, indicating that the current combined strategy of temperature gradient control and pressure perturbation slow release has good dynamic adaptability. Based on this, to ensure the control strategy can continuously adapt to potential perturbation conditions in different growth cycles in the future, and to improve the accuracy of migration event identification and the robustness of intervention response, it is necessary to use the residual stress distribution, migration trigger time, and the effect of control intervention during the current perturbation process as feedback information to update the stability range parameters of the original carbon and silicon component ratio, and simultaneously refresh the risk identification threshold, thereby achieving iterative evolution of the control model. This process includes the following steps:

[0088] Using the start time of the perturbation as a reference point, spectral information acquired continuously before, during, and after the perturbation in the in-situ Raman detection data was extracted. The amplitude, duration, and rate of return of the main peak displacement value before and after the perturbation peak were clearly recorded. For example, after a 3% increase in heating power, the maximum stress fluctuation occurred in the 4th minute, with the main peak displacement reaching +1.6 cm. -1 It recovered to +0.9cm in the 10th minute. -1 At the 16th minute, it returned to the baseline ±0.5cm. -1Within the fluctuation range, and the full width at half maximum (FWHM) of the spectral peak decreased from 9.8 cm before the disturbance. -1 It rose to 11.1 cm at one point during the disturbance. -1 It eventually stabilized at 9.9cm. -1 Furthermore, the step height variation values ​​obtained using interferometry confirm that the deposition flux remained between 0.10 and 0.12 μm / min before and after the disturbance, with no step collapse or aggregation, and the crystal morphology remained regular. Regarding the control response, the temperature adjustment action was maintained at T+3℃ for 10 minutes before ceasing, and the microvalve control did not initiate any new interventions, indicating that the temperature channel independently achieved a closed-loop control. The above data were numbered, classified, and associated according to timestamps and spatial coordinates to form a complete "disturbance-response-recovery" dataset, which will serve as the basis for subsequent model corrections.

[0089] The original coupled model defined a stable carbon-silicon supply range as follows: the source region sublimation temperature maintained between 2240℃ and 2260℃, the cavity pressure controlled between 90Pa and 95Pa, and the silicon-to-carbon partial pressure ratio fluctuating between 0.96 and 1.04 without triggering a stress overshoot warning. In this perturbation experiment, the sublimation temperature was increased to 2265℃, the cavity pressure stabilized at 93.2Pa, and the silicon partial pressure slightly increased. However, even with an actual partial pressure ratio of 1.06, dendrite formation or interface cracking did not occur, indicating that the original model's defined range was overly conservative. Therefore, the temperature, pressure, and component partial pressure parameters corresponding to the current perturbation test results were used as new boundary conditions input into the original model. A new stable range node was inserted into the model, and the upper limit of the stable partial pressure ratio was revised to 1.07, and the upper limit of temperature adjustment was revised to 2267℃. Simultaneously, to avoid erroneous boundary expansion, the stress response time at this boundary point was required to recover to ±1.0cm within 15 minutes. -1 The step advancement rate is no less than 0.095 μm / min, which is written into the model as a boundary validity confirmation condition. All parameter updates are integrated into the original model after curve fitting, so that the stable interval is in a continuous and dynamic evolution state, completing one closed-loop update.

[0090] In the original identification strategy, the displacement of the main Raman stress peak was set to exceed ±1.2 cm. -1 Half-height and width exceed 2.0cm -1 A Level 1 warning was triggered; if the disturbance did not subside within 3 minutes, intervention procedures were initiated. In this test, the disturbance caused a peak displacement of up to 1.6 cm. -1 However, it naturally dropped back to 0.9cm after 10 minutes. -1 The following did not trigger actual sedimentary instability. Therefore, the primary warning threshold for main peak displacement in risk identification is increased to ±1.5 cm. -1 The level-two warning threshold is set at ±1.7cm. -1 Furthermore, only when the full width at half maximum (FWHM) of the spectral peak expands to 2.3 cm in three consecutive samplings.-1 Only after these steps are taken does the intervention process begin, enhancing the anti-interference capability of the early warning logic. A "continuous cycle judgment mechanism" is introduced, requiring three consecutive sampling cycles to meet the deviation condition before triggering an early warning, avoiding false triggers due to occasional fluctuations. Simultaneously, a spatial distribution factor is added to the updated judgment logic. If stress concentration points continuously appear within a radius of 5mm in the same coordinate region, this region is marked as a high-risk area, and its flux is subsequently prioritized for control in supply management. These risk judgment criteria are output through model evolution and serve as the online judgment basis for the next round of crystal growth.

[0091] The updated stability range and risk identification criteria were written into the initial parameters of the crystal growth control program as initial settings. During the initial stage of crystal growth, the sublimation temperature was set to 2255℃, the partial pressure ratio control target value was set to 1.03, and the initial microvalve opening was 0.72mm, all according to the corrected parameters. Raman stress data and crystal surface step morphology were continuously collected during the first 48 hours of growth. The real-time data was compared with the model's set boundaries to determine whether control needed to be initiated. If the main peak displacement showed an increasing trend but did not exceed ±1.5cm... -1 If the stress fluctuations are concentrated in the same area frequently and last for more than 12 minutes, then a slight temperature reduction operation will be immediately initiated, and the power output in the sublimation temperature zone will be reduced by 3%. During the operation, if a significant delay in control response is found (such as stress fallback time exceeding 20 minutes), the current data will be extracted and rewritten into the model to form a new round of data nodes, updating the current stability boundary and risk threshold, thereby achieving dynamic evolution and closed-loop continuous optimization of the model.

[0092] This step aims to construct a dynamically evolving crystal growth control mechanism. By using the residual stress distribution obtained during perturbation testing, the specific time points of supply offset triggering, and the actual effects of the control response as feedback information, this information is written back in real-time to the previously established carbon-silicon composition ratio coupling model. This enables continuous updating of model parameters and recalibration of the stable range. This step not only corrects the boundary conservatism issues exposed by the early model in actual operation but also dynamically adjusts the judgment criteria for temperature, partial pressure, and stress identification based on the latest process performance, forming a true closed loop of "identification-control-modeling" in the entire crystal growth process. In this way, the system can gradually improve the accuracy of the model and the adaptability of the control strategy as it accumulates perturbation and response data. This avoids misjudgments or delayed interventions caused by unreasonable parameter settings or excessively low identification thresholds, effectively improving growth stability and crystal quality consistency. It is a key guarantee for achieving adaptive, high-precision silicon carbide single crystal growth control.

[0093] The above scheme enables precise identification, dynamic tracking, and closed-loop control of the sudden shift in the carbon-to-silicon composition ratio caused by pressure perturbations during the PVT method for growing silicon carbide crystals, significantly improving the stability and robustness of the crystal growth process. This method constructs a coupled model of pressure perturbation and gas phase migration, combines in-situ optical monitoring to acquire crystal interface step evolution and residual stress signals, establishes a correlation mechanism between the spatial thermal stress distribution map and the supply shift path, and, based on this, achieves highly responsive steady-state control through dual-channel fine-tuning of temperature and pressure. Furthermore, adaptive correction of control parameters is achieved through small-amplitude perturbation trials and stress recovery judgments, and the process response data is written back into the model to achieve dynamic evolution of the stable range and risk identification threshold. This effectively avoids irreversible structural instability caused by a single perturbation, fundamentally improving the yield, quality consistency, and automation level of silicon carbide single crystal growth, overcoming the technical bottlenecks caused by monitoring delay, control lag, and model rigidity in existing technologies.

[0094] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for growing silicon carbide crystals using the PVT method, characterized in that, Includes the following steps: S100, establish a coupled model of pressure perturbation spectrum and gas phase migration process, invert the stable range of carbon and silicon component ratio based on long-period operation data, and calibrate the critical value of interface energy and the boundary range of temperature gradient accordingly. S200, under the constraints of the critical value of interface energy and the boundary range of temperature gradient, constructs a risk identification feature template, uses in-situ Raman detection and reflection interferometry monitoring to extract early signals of the evolution of crystal growth interface steps, and generates a spatial thermal stress distribution map based on the signal distribution; S300 constructs a virtual sensing system for the carbon and silicon component ratio based on the spatial thermal stress distribution map. By integrating crucible temperature field data, raw material sublimation rate and component partial pressure trajectory, it continuously tracks the supply ratio offset and locates the offset path and key timing nodes. S400 performs dual-channel closed-loop control operation based on the offset path and key timing nodes. On the one hand, it fine-tunes the temperature gradient to suppress the offset of the carbon and silicon composition ratio, and on the other hand, it adjusts the micro-valve damping to reduce the intensity of pressure fluctuation, thereby stabilizing the deposition flux on the seed crystal surface. S500 applies small-amplitude compositional pulse perturbations under the condition that the deposition flux tends to be stable, monitors the change amplitude of the interface residual stress field, and adaptively adjusts the control gain parameters and time delay compensation model according to the residual change trend. After the stability boundary of the residual stress field is verified, the residual distribution, offset trigger time and control response effect are used as feedback data input to the coupled model to update the stable range of carbon and silicon component ratio, and simultaneously update the risk identification threshold, thus realizing a closed-loop evolution process of modeling, identification and control.

2. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S100 includes: Based on a silicon carbide crystal growth experiment that has been running continuously and stably for more than 72 hours, data on sublimation zone temperature, seed crystal zone temperature, static pressure in the chamber, partial pressure of carbon atoms, partial pressure of silicon atoms, crystal growth rate, and crystal surface morphology were collected. The amplitude, duration, frequency, and composition change rate of each pressure disturbance event were calibrated. Each set of perturbation samples is correlated with the partial pressure change data of carbon and silicon components in the gas phase. The net migration of carbon and silicon in the axial path per unit time before and after the perturbation is calculated, and the relationship between perturbation frequency and carbon-silicon ratio change is plotted to identify perturbation intervals with nonlinear abrupt change characteristics. Based on the changes in the step structure on the crystal surface, the change in free energy per unit area during the deposition process at the crystal interface is derived. The minimum free energy value that causes the crystal interface stability to be destroyed is extracted as the critical value of the interface energy. Under the condition of constant perturbation intensity, the temperature gradient is adjusted to obtain the temperature gradient boundary range that suppresses proportional offset. All parameters are input into the gas-phase migration coupling model, and simulation calculations are performed for each set of disturbance conditions. Based on the carbon-silicon ratio fluctuation, free energy fluctuation, and temperature gradient, it is determined whether it belongs to the stable range. Finally, a continuous range spectrum of whether the carbon-silicon supply is stable is generated.

3. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S200 includes: Based on the critical value of interface energy and the boundary range of temperature gradient, a detection window is set at the top of the reaction chamber, a Raman laser beam with an incident angle of 10 degrees and a wavelength of 532 nanometers is arranged, and a white light interference probe is arranged simultaneously to collect the Raman scattering signal of the crystal surface and the step height change data respectively. The cumulative change in the Raman displacement peak exceeds 1.5 cm. -1 When the spectral width increases and the step abruptness in the interference signal exceeds three times and disappears periodically, it is marked as an unstable signal of interface deposition. Three-dimensional information units are generated by combining the detection coordinates and a crystal front unfolding diagram is constructed. Extract regions with dense signal abrupt changes from the crystal front development diagram, and compare the temperature gradient, free energy change, and stress abrupt change frequency. If the boundary conditions are met simultaneously, they are marked as high-risk regions, and a risk identification feature template containing coordinates, time period, stress amplitude, and step jump rate is formed. The feature template is mapped onto the three-dimensional crystal structure to generate a thermal stress spatial distribution map with a resolution of 0.2 mm and a step size of 0.5 GPa, which is then used as the input condition for subsequent proportional offset tracking.

4. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S300 includes: A cross-sectional thermal field measurement surface centered on the crystal growth axis was constructed, a 5×3 lattice network was set, an S-type platinum-rhodium-platinum thermocouple was installed, three-dimensional temperature field data were collected, and the region with the steepest temperature gradient was identified as the supply distribution sensitive area. The sublimation rate of raw materials was determined using a thermoelectric balance, and the actual sublimation rates of carbon and silicon were calculated by combining thermal field data. A sublimation source term distribution map was generated and high-risk source areas with sublimation rates exceeding 20% ​​of the average value were marked. Infrared spectrophotometers were set up at the top of the raw material area, the convection area and the seed crystal area to measure the partial pressure of Si, Si2C and C2, and to plot a two-dimensional curve of the carbon-silicon partial pressure ratio. This was used to determine whether a supply ratio shift had occurred and to conduct preliminary verification by comparing the coordinates of the thermal stress diagram. Based on the temporal and spatial evolution relationship between the initial point of partial pressure mutation and the intersection point of thermal stress mutation, key time segments are defined, spatial coordinates and thermal field anomaly locations are extracted, and a coupling correspondence between the supply offset path and the deposition response is established.

5. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S400 includes: Based on the starting point and spatial location of the carbon and silicon component ratio offset path, regions with temperature gradients below the boundary range are identified in the three-dimensional temperature field. By adjusting the power of the raw material heating section and the convection zone heating section, the local temperature gradient is increased to the set range. Infrared thermal imaging and in-situ interferometry were used to simultaneously monitor the temperature difference and step changes on the crystal surface. Raman spectroscopy was combined to confirm whether the stress had recovered to the stable range and to judge the effect of temperature regulation. The opening of the electric needle valve on the atmosphere inlet guide tube is gradually reduced to reduce the static pressure fluctuation of the cavity to below ±0.2Pa, thereby ensuring the stability of the gas phase component migration process. The four indicators of sedimentary step advancement rate, stress change, temperature difference distribution and pressure fluctuation are jointly judged to determine whether they meet the standards. After confirming that the sedimentation flux has recovered and stabilized, the control parameters are locked and maintained until the next offset path appears.

6. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S500 includes: Under stable deposition flux conditions, a temperature disturbance was implemented by increasing the electric heating power of the raw material zone from 8.0 kW to 8.2 kW, resulting in a slightly enriched silicon composition that was maintained for 20 minutes. Raman spectra were collected every 2 minutes during the disturbance process. The amplitude of the interface stress response was determined and the stability of the sedimentary morphology was evaluated by combining the step height change data. The adjustment step size of the raw material heating power was optimized to 4% based on the stress response intensity, the control response cycle was shortened to 10 minutes, and the micro-valve adjustment step size and pressure refresh frequency were increased. After the disturbance ends, continue recording the deposition flux and stress curves for 30 minutes. If the step advancement speed and stress value fluctuations remain within ±5% of the pre-disturbance range, then record this state as the steady-state recovery boundary and update the control strategy reference dataset.

7. The method for growing silicon carbide crystals using the PVT method according to claim 1, characterized in that, Step S600 includes: After a small-amplitude perturbation trial is completed, continuous Raman spectrum data of the pre-, mid- and post-perturbation stages are extracted to clarify the changing trends of the main peak displacement, half-width at half-maximum and step advance rate, and to form a perturbation-response-recovery dataset. The dataset was embedded into the original coupled model, the stable range of carbon and silicon component ratio was corrected, the upper limit of partial pressure ratio was updated to 1.07, the upper limit of sublimation temperature was updated to 2267℃, and boundary validity confirmation conditions were set. Based on the revised model, the risk identification criteria have been updated, and the primary warning threshold for main peak displacement has been raised to ±1.5 cm. -1 Furthermore, a continuous cycle judgment mechanism and a spatial risk clustering judgment standard were introduced; In the next growth cycle, the initial settings are executed according to the updated parameters, and the control channel is called in real time according to the stress fluctuation trend. At the same time, the control results are written back to continuously optimize the model, realizing a closed-loop update of modeling, identification and control.

Images