A method, apparatus, device, and storage medium for adjusting the thickness of a silicon molten layer.

By real-time detection of the amount of crystal protrusions and geometric spacing at the interface, and by using laser displacement calculation methods and dual-channel decoupled closed-loop control, the problem of interface carbon concentration gradient change caused by silicon molten layer thickness drift was solved, which improved the growth uniformity and stability of single-crystal silicon carbide and reduced the probability of defects.

CN123108726APending Publication Date: 2026-11-17JIHUA LAB
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611602343.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

During the metastable solvent epitaxial growth of single-crystal silicon carbide, the thickness drift of the silicon molten layer leads to changes in the interface carbon concentration gradient and supersaturation, causing fluctuations in the growth rate, making it difficult to maintain the uniformity of the wafer over a large area, and insufficient material supply affects the long-term growth stability.

Method used

By real-time detection of the crystal protrusions and geometric spacing at the interface, the thickness of the silicon molten layer is obtained using a laser displacement calculation method. Combined with a preset deviation threshold and decomposition method, the lifting drive displacement and heating power bias are generated to achieve dual-channel decoupled closed-loop control, stabilize the interface carbon concentration gradient and supersaturation, and suppress polycrystalline and hillock defects.

Benefits of technology

This technology improves the uniformity of large-area wafers and the stability of long-term growth, reduces the probability of polycrystalline and various defects, and increases the yield of single-crystal silicon carbide products.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123108726A_ABST
    Figure CN123108726A_ABST
Patent Text Reader

Abstract

This invention relates to the field of silicon molten layer thickness adjustment technology, and discloses a method, apparatus, device, and storage medium for silicon molten layer thickness adjustment. The method involves calculating the interface crystal protrusion and measured geometric spacing based on a preset laser displacement calculation method to obtain the silicon molten layer thickness; analyzing the silicon molten layer thickness based on a preset target thickness, a preset deviation threshold, and a preset deviation decomposition method to obtain the lifting drive displacement and heating power bias; generating multiple limiting power adjustment increments based on preset irregularity identification conditions, preset partition power bias calculation formulas, and various crystal type discrimination values; generating silicon molten layer thickness adjustment commands based on the lifting drive displacement, heating power bias, each limiting power adjustment increment, and each channel radius; obtaining the lifting drive displacement and heating power bias through deviation decomposition; and fusing multiple parameters to generate adjustment commands to suppress radial defects and improve wafer uniformity and crystal yield.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of silicon molten layer thickness adjustment technology, and in particular to a method, apparatus, equipment and storage medium for adjusting silicon molten layer thickness. Background Technology

[0002] In the metastable solvent epitaxial growth of single-crystal silicon carbide, as epitaxy continues, the crystal thickness of the seed crystal increases and the surface layer of the feed crystal is continuously eroded, causing continuous drift in the seed-feed gap and the thickness of the silicon molten layer. This leads to changes in the interfacial carbon concentration gradient and supersaturation, resulting in fluctuations in the growth rate. It is difficult to maintain consistent spacing between radial positions of large-area seed wafers, and uneven drift amplifies local supersaturation differences, easily inducing defects such as polycrystalline material and hillocks, thus restricting the uniformity of the wafer over a large area. In addition, the consumption of the feed crystal cannot be monitored in real time, and insufficient feed in the later stages of growth can easily lead to concentration gradient failure and poor long-term growth stability. Existing technologies lack in-situ real-time detection and closed-loop feedback methods for the growth process, making it difficult to meet the stable growth requirements of high-quality, large-size single-crystal silicon carbide. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, apparatus, device and storage medium for adjusting the thickness of a silicon molten layer.

[0004] A method for adjusting the thickness of a silicon molten layer includes: acquiring the amount of interfacial crystal protrusions and the measured geometric spacing during the growth of single-crystal silicon carbide; calculating the amount of interfacial crystal protrusions and the measured geometric spacing according to a preset laser displacement calculation method to obtain the thickness of the silicon molten layer; analyzing the thickness of the silicon molten layer according to a preset target thickness, a preset deviation threshold, and a preset deviation decomposition method to obtain the lifting drive displacement and heating power bias; acquiring the crystal type discrimination value of each radial region in the seed crystal to obtain multiple crystal type discrimination values; generating multiple limiting power adjustment increments according to preset irregularity recognition conditions, preset partition power bias calculation formulas, and each crystal type discrimination value; acquiring multiple channel radii of the seed crystal, and generating a silicon molten layer thickness adjustment command according to the lifting drive displacement, heating power bias, each limiting power adjustment increment, and each channel radius.

[0005] Further, the step of analyzing the thickness of the silicon molten layer according to the preset target thickness, the preset deviation threshold, and the preset deviation decomposition method to obtain the lifting drive displacement and the heating power bias includes: calculating the difference between the preset target thickness and the silicon molten layer thickness to obtain the total deviation; determining whether the total deviation is greater than the preset deviation threshold; if the total deviation is greater than the preset deviation threshold, decomposing the total deviation according to the preset deviation decomposition method to obtain the thickness deviation component and the temperature deviation component; calculating the thickness deviation component according to the preset displacement conversion formula to obtain the lifting drive displacement; and calculating the temperature deviation component according to the preset heating power conversion formula to obtain the heating power bias.

[0006] Furthermore, the step of generating multiple limiting power adjustment increments based on preset irregularity identification conditions, preset partition power bias calculation formulas, and each crystal form discrimination value includes: performing discrimination analysis on each crystal form discrimination value based on preset irregularity identification conditions and preset irregularity identification thresholds to obtain discrimination analysis results; if the discrimination analysis results indicate that there are crystal form discrimination values ​​that meet the irregularity identification conditions, then collecting multiple radial channel deviations of the seed crystal; and analyzing each radial channel deviation based on preset partition power bias calculation formulas and preset limiting analysis conditions to obtain multiple limiting power adjustment increments.

[0007] Furthermore, the step of analyzing the deviation of each radial channel according to the preset partition power bias calculation formula and the preset limiting analysis conditions to obtain multiple limiting power adjustment increments includes: calculating the deviation of each radial channel according to the preset partition power bias calculation formula to obtain multiple power adjustment increments; and limiting each power adjustment increment according to the preset limiting analysis conditions and the preset limiting threshold to obtain multiple limiting power adjustment increments.

[0008] Further, the step of generating a silicon molten layer thickness adjustment command based on the lifting drive displacement, heating power bias, adjustment increments of each limiting power, and radius of each channel includes: calculating the radius of each channel according to a preset fitting radial first-order gradient calculation method and a preset tilt angle calculation formula to obtain the tilt angle; determining whether the tilt angle is less than or equal to a preset angle threshold; if the tilt angle is less than or equal to the preset angle threshold, obtaining the crystal thickening rate and the feeding erosion rate; analyzing the crystal thickening rate and the feeding erosion rate to obtain the feedforward feedback superposition amount and the remaining feeding thickness; and generating a silicon molten layer thickness adjustment command based on the lifting drive displacement, heating power bias, adjustment increments of each limiting power, tilt angle, feedforward feedback superposition amount, and remaining feeding thickness.

[0009] Furthermore, the step of calculating the radius of each channel according to the preset fitting radial first-order gradient calculation method and the preset tilt angle calculation formula to obtain the tilt angle includes: calculating the radius of each channel and the deviation of each radial channel according to the preset fitting radial first-order gradient calculation method to obtain the radial first-order gradient; and calculating the radial first-order gradient according to the preset tilt angle calculation formula to obtain the tilt angle.

[0010] Furthermore, the analysis of the crystal thickening rate and the feed erosion rate to obtain the feedforward feedback superposition amount and the remaining feed thickness includes: calculating the crystal thickening rate and the feed erosion rate according to a preset feedforward model to obtain the position feedforward amount; calculating the position feedforward amount and the preset feedback output amount according to a preset control quantity superposition calculation formula to obtain the feedforward feedback superposition amount; and calculating the feed erosion rate according to a preset feed lifetime prediction method to obtain the remaining feed thickness.

[0011] Furthermore, a silicon molten layer thickness adjustment device includes: a control unit for performing the silicon molten layer thickness adjustment method as described above, and a seed crystal, a feeder, a seed crystal clamping member, and a feeder clamping member electrically connected to the control unit; the feeder is disposed at the upper end of the feeder clamping member; the seed crystal is disposed below the seed crystal clamping member; and a silicon molten layer is formed between the seed crystal and the feeder.

[0012] Furthermore, a silicon molten layer thickness adjustment device is provided, the silicon molten layer thickness adjustment device comprising: a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the silicon molten layer thickness adjustment device to perform the various steps of the silicon molten layer thickness adjustment method as described above.

[0013] Furthermore, a computer-readable storage medium stores instructions that, when executed by a processor, implement the steps of the silicon molten layer thickness adjustment method as described above.

[0014] In the technical solution of this invention, during the growth of single-crystal silicon carbide, the amount of crystal protrusions at the interface and the measured geometric spacing are collected, and the thickness of the silicon molten layer is calculated based on a preset laser displacement calculation method. Thickness deviation analysis is performed by combining the target thickness, deviation threshold, and deviation decomposition method. The lifting drive displacement and heating power bias are decoupled and output. Mechanical displacement compensates for gap drift caused by crystal thickening and material feeding erosion, and heating power is used to quickly suppress thermal disturbances and stabilize the interface carbon concentration gradient and supersaturation. Simultaneously, crystal type discrimination values ​​in each radial region of the seed crystal are collected, and a partitioned power adjustment increment with amplitude limiting is generated based on the irregularity identification conditions and partitioned power bias calculation formula. Multiple parameters are fused to generate silicon molten layer thickness adjustment commands. This solution constructs a thickness control and radial crystal type partitioned compensation mechanism, realizing dual-channel decoupled closed-loop control, balancing local radial supersaturation of the wafer, suppressing radial irregularity defects such as polycrystalline and hillocks, sensing the growth status in real time, avoiding concentration gradient failure caused by insufficient material supply in the later stages of growth, and improving the uniformity of large-area wafers, long-term growth stability, and single-crystal yield. Attached Figure Description

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 A flowchart of a method for adjusting the thickness of a silicon molten layer provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a silicon molten layer thickness adjustment device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a silicon molten layer thickness adjustment device provided in an embodiment of the present invention.

[0016] Figure Labels 1-Control unit; 3-Seed crystal; 4-Feeding component; 2-Seed crystal clamping component; 5-Feeding clamping component. Detailed Implementation

[0017] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] This invention provides an apparatus and method for metastable solvent epitaxial growth of single-crystal silicon carbide. The purpose is to: suppress drift in the gap between the seed crystal and the feed element, or the thickness of the molten silicon layer, during crystal growth by in-situ detection and active control of the gap or the thickness of the molten silicon layer; stabilize the carbon concentration gradient and supersaturation at the interface; thereby improving the radial large-area uniformity of the wafer, enhancing long-term crystal growth stability, and reducing the probability of polycrystalline growth and various defects. As the crystal on the surface of the seed crystal continuously thickens, the seed crystal holder is controlled to retract upwards; and / or as the surface layer of the feed element is continuously eroded and consumed, the feed holder is controlled to advance upwards, maintaining the seed crystal-feed gap near a set target value. For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of a method for adjusting the thickness of a silicon molten layer according to the present invention includes: 101. Obtain the amount of interfacial crystal protrusions and the measured geometric spacing during the growth of single-crystal silicon carbide; In this embodiment, the detection unit acquires the measured geometric distance between the lower surface of the seed crystal and the upper surface of the feed piece using a laser displacement sensor. This measured geometric distance characterizes the physical spatial distance between the seed crystal and the feed piece. Simultaneously, it obtains the interface crystal protrusion amount by combining the crystal growth mechanism model or calibration curve. The interface crystal protrusion amount characterizes the height of the crystal protruding from the lower surface of the seed crystal within the silicon molten layer. By simultaneously acquiring the above two types of parameters, basic measurement inputs are provided for subsequent calculation of the silicon molten layer thickness. 102. Calculate the amount of crystal protrusion at the interface and the measured geometric spacing according to the preset laser displacement calculation method to obtain the thickness of the silicon molten layer; In this embodiment, the laser displacement calculation method is as follows: The measured geometric distance between the lower surface of the seed crystal and the upper surface of the feed piece is obtained using laser ranging; then, the amount of the interface crystal protrusion of the epitaxial crystal at the bottom of the seed crystal relative to the seed crystal reference surface is collected; and the thickness of the silicon molten layer is calculated using the following formula: , In the formula, The geometric distance between the lower surface of the seed crystal and the upper surface of the feed piece, as measured by laser. The interfacial crystal protrusion amount is the distance by which the silicon carbide crystal undergoing epitaxial growth protrudes downwards from the lower reference surface of the seed crystal. To determine the measured thickness of the obtained silicon molten layer; The seed crystal and the feed material are arranged opposite each other with a molten silicon layer between their surfaces. Single-crystal silicon carbide is epitaxially grown on the surface of the seed crystal by heating. During the growth process, the gap between the seed crystal and the feed material, or the thickness of the molten silicon layer, is monitored in real time, and the monitored values ​​are compared with set target values. Laser displacement or laser interferometry can be used to measure the distance between the lower surface of the seed crystal and the upper surface of the feed material, or the thickness of the molten silicon layer, through a transparent window. The target value planner outputs target trajectory values ​​for the thickness of the molten silicon layer according to different growth stages. and growth gap target trajectory value The two independent target parameters are sent to the controller; the controller sends control commands to the drive unit, which includes a seed crystal lifting mechanism, a feeding lifting mechanism, and a heating power circuit, to adjust the MSE growth device as the controlled object; the detection unit continuously collects growth state information such as gap, silicon molten layer thickness, and crystal form discrimination value output by the quality characterization channel, and feeds it back to the controller to form a complete control loop; where, This represents the target trajectory value for the thickness of the silicon molten layer, in μm; superscript. The target set value is represented by t, which is the growth time; it represents the expected thickness of the silicon molten layer dynamically planned according to the single crystal growth process. It is set in segments according to the nucleation segment, steady-state growth segment, and end segment. The S-curve is used to smoothly transition between the stages to match the crystal stability window and the target growth rate of different stages. The target trajectory value for the growth gap, in μm; superscript. The target setting value is represented by t, which is the growth time. It represents the expected geometric gap between the lower surface of the seed crystal and the upper surface of the feed crystal, which is dynamically planned as the single crystal growth process progresses. The controller acts on the controlled object, the MSE growth device, through the drive unit (seed crystal lifting mechanism, feeding lifting mechanism, heating power circuit); the detection unit collects growth status information such as gap, thickness, and crystal form discrimination value of the quality characterization channel and feeds it back to the controller. In this scheme, the measured geometric spacing refers to the geometric spacing from the lower surface of the seed crystal to the upper surface of the feed piece; the "silicon molten layer thickness" is the actual thickness of the liquid silicon layer, and the difference between the two represents the amount of crystal protrusion at the interface and the surface morphology; the "equivalent gap" is the spacing conversion value obtained after calibration and conversion of the original physical quantity detected; the measurement objects of different detection methods are different: laser displacement / interference method measures the gap, ultrasonic reflection method measures the molten layer thickness, and capacitance method measures the equivalent gap; the control unit converts all kinds of measurement signals into a feedback signal with the silicon molten layer thickness as the main control quantity according to the conversion relationship; among which the main control quantity is the silicon molten layer thickness. and equivalent gap Satisfy the conversion relationship, Thickness of the silicon molten layer, unit: ); represents the main control variable to be determined in this capacitive detection scheme; The protrusion amount is the height of the crystal protruding above the lower surface of the seed crystal within the molten layer. This protrusion amount can be estimated by the growth model according to the formula or obtained by looking up the calibration curve in a table. The detection unit acquires signals in real time at a sampling frequency of 1Hz to 100Hz. The control unit compares the detected value with the target value. When the deviation exceeds the set threshold (example ±5μm), it drives the seed crystal clamping component to generate displacement (servo / stepper drive, resolution up to 0.1μm level) so that the gap returns to the target range. When the deviation is within the threshold range, the current working state is maintained. This solution employs continuous measurement and dual-channel linkage decoupling control. The gap / molten layer thickness detection unit can utilize a laser interferometric sensor to continuously output measurement signals. The control unit implements dual-channel linkage control based on a PID control law. It should be noted that heating power linkage is an optional feature; this control architecture is not limited to laser interferometric detection, and capacitive and ultrasonic reflection detection schemes are also applicable. The dual-channel linkage is implemented as follows: the control unit decomposes the total deviation into thickness deviation and temperature deviation components, which are respectively fed into the displacement channel (seed crystal lifting drive and / or feeding lifting drive) and the heating power channel (heating unit) for adjustment. The heating power channel has a small time constant and fast thermal response, used to suppress transient deviations caused by feeding erosion and thermal disturbances. The displacement channel has a large time constant and slow mechanical response, used to maintain the steady-state value of the gap or molten layer thickness. The two channels adopt a frequency band separation decoupling approach, combined with a threshold scheduling strategy: when the deviation is large, the power channel quickly suppresses disturbances; when the deviation is small, the displacement channel finely tunes, thus avoiding single-channel saturation problems (displacement stroke has an upper limit, and heating power is constrained by hardware). PID control coefficients ( (Specific values ​​need to be adjusted) and determined by the adjustment experiment; The capacitive detection scheme features windowless operation and high-temperature resistance: Electrode pairs (electrode A and electrode B, which can be made of high-temperature resistant metal or graphite) are arranged laterally between the seed crystal and the feed element. An insulating dielectric layer is added between the electrode pairs and the molten silicon layer to achieve electrical isolation between the electrodes and the melt. Under high-temperature conditions, liquid silicon exhibits conductivity, rendering the ideal parallel-plate capacitor model inapplicable. Therefore, high-frequency excitation is used to acquire the complex impedance between the electrode pairs, utilizing the capacitive component in the complex impedance according to… Real-time calculation of molten layer thickness In the formula, The vacuum permittivity is a constant. ; The relative permittivity of the molten liquid silicon layer is affected by high temperature. It should be calibrated before use and can be corrected online according to the growth temperature. The effective facing area of ​​the electrode pair, in units of: This is the effective area of ​​the capacitor plates; The total equivalent capacitance measured under high-frequency excitation of the electrode pair, in units of: The total capacitance is calculated from the collected complex impedance signal; the total capacitance is the series connection of the capacitance of the insulating dielectric layers on both sides and the capacitance of the silicon melt. Capacitance of a single-sided insulating dielectric layer, unit: Since there is an insulating medium layer on both the top and bottom of the melt, the total dielectric term in series is: Simultaneously, the contact state between the melt and the electrode is monitored by relying on the resistance component; the correspondence between the capacitance value and the gap is established in advance through temperature calibration and can be corrected online according to the growth temperature; this embodiment does not require setting a transparent window, and is suitable for high temperature and closed atmosphere working environment. An ultrasonic transducer is installed at the lower end of the seed crystal holder, between the seed crystal holder and the seed crystal. Ultrasonic waves are incident on the silicon molten layer through the seed crystal. To resist the effects of high temperatures, the seed crystal holder is equipped with water-cooled or air-cooled cooling channels to ensure that the transducer's operating temperature is below the Curie point of the piezoelectric material. The upper interface (seed crystal side) and lower interface (feeding component side) of the silicon molten layer will generate reflected echoes respectively. The time difference between the two sets of echoes satisfies the following relationship: ; In the formula, The speed of sound in a silicon melt, satisfying ,parameter Reference temperature The speed of ultrasonic waves in a silicon melt under the specified conditions, in units of: ; for pre-calibrated reference constants, Pre-calibrated and corrected for temperature, Units: (per Kelvin) Real-time measured temperature of the molten silicon layer, unit: (Kelvin); This embodiment uses laser displacement detection; capacitive detection and ultrasonic reflection detection are optional alternative detection methods. The two schemes, capacitive detection and ultrasonic reflection detection, can be used as parallel detection channels that are mutually redundant / mutually verifying: one can be activated independently, and the thickness calculation result of the selected channel can be used as the subsequent value for the silicon molten layer thickness; when both channels are activated simultaneously, the capacitive detection result is used as the main measurement value, and the ultrasonic reflection detection result is used as the verification reference value to determine whether the thickness measurement is abnormal. The ultrasonic result is not used as the main control input; the core conversion formulas for each detection method are as follows: Capacitive detection: Complex impedance at high frequency Rongkang ,Right now , In the formula, The complex impedance of the electrode detection circuit, in units of: (Ohms); This is the complex impedance acquired by this capacitive thickness measurement system, containing both real resistance and imaginary reactance components. It is the fundamental measurement quantity for subsequent extraction of capacitive reactance and calculation of equivalent capacitance. The real part of the complex impedance, i.e., the equivalent resistance of the circuit, is expressed in units of: Characterizes the active power loss resistance caused by the detection electrode, molten silicon, and leads; The imaginary part of the complex impedance, i.e., the equivalent reactance of the circuit, is expressed in units of: In this embodiment, capacitive reactance is the primary component. Capacitive modulus value obtained from electrode pair measurement, unit: (Ohm), used for substitution into the formula Calculate the equivalent capacitance of the electrode circuit; The angular frequency of the high-frequency excitation signal, in units of: ; is the sinusoidal excitation frequency applied to the detection electrode pair; is the absolute value of the reactance component in the complex impedance, measured by the high-frequency impedance acquisition circuit; The equivalent capacitance of the electrode circuit, in units of (Farad), representing the capacitance value to be solved in this capacitive detection scheme, used for subsequent substitution into the silicon molten layer thickness calculation formula; considering the series connection of the insulating dielectric layers, it satisfies... molten capacitor The derivation yields ; In the formula, The vacuum permittivity, The relative permittivity of the melt (temperature calibrated). Effective electrode area; resistance component Used for monitoring the contact state of melt. Ultrasonic reflective testing: , The echo time difference between the upper and lower interfaces. The velocity of sound in molten silicon. , , Pre-calibrated and corrected for temperature; The reference temperature used for calibration experiments, in units of: (Kelvin); The control unit converts the raw measurement signals from each channel into master control quantities before participating in comparison and control. The conversion relationship is as follows: ; The measured thickness of the silicon molten layer is given by [value], and [value] is the output of this function; the unit can be [unit]. or This is used for subsequent calculations of radial deviation and first-order radial gradient; the conversion logic for the three detection methods: laser displacement / interference method for direct measurement of geometric spacing. Then through Convert to master control variable, Estimated by the growth model; Target value trajectory planning: The gap and melt layer thickness are no longer kept constant, but are used as process programming variables that change with the growth stage to adapt to the optimal thickness window for each stage of nucleation, steady state and termination. Control quantity caliber confirmation: The thickness h of the silicon molten layer is taken as the main control quantity, which directly determines the carbon concentration gradient. ;gap The geometric distance between the lower surface of the seed crystal and the upper surface of the feed piece is the interfacial crystal protrusion; the difference between the two is the interfacial crystal protrusion. The gap obtained by laser measurement is used as the surrogate quantity, which is converted and unified to a value in terms of... Feedback signal of main control quantity; target value trajectory planning logic: the control unit has a built-in target value planner ( Figure 3 Output time-varying trajectory target values ​​according to growth stage. Design example: Nucleation stage (0-0.5h) Maintain 60 μm to promote uniform nucleation; steady-state phase (0.5 h to 1 h before the end). The slope was gradually adjusted from 60 μm to 45 μm using a linear or gradually varying gradient to match the target growth rate and crystal form stability window; in the final stage, The sample was reduced to 40 μm and maintained thereafter; S-curves were used for smooth transitions between stages to avoid abrupt changes in target values ​​that could lead to sudden oversaturation. The trajectory parameters were determined jointly by the target growth rate and the target crystal form stability window. 103. Analyze the thickness of the silicon molten layer according to the preset target thickness, preset deviation threshold and preset deviation decomposition method to obtain the lifting drive displacement and heating power offset. In this embodiment, the thickness of the silicon molten layer is analyzed based on a preset target thickness, deviation threshold, and deviation decomposition method. The lifting drive displacement and heating power bias are output to achieve dual-channel decoupled control. The total thickness deviation is separated into steady-state deviation and transient disturbance components. The lifting displacement channel is used to complete the steady-state adjustment of the molten layer thickness, while the heating power channel is used to quickly suppress transient deviations caused by thermal disturbances and material feeding erosion, avoiding the saturation risk of single-channel actuators. At the same time, it supports the unified conversion of multiple detection signals such as laser interferometry, capacitive sensing, and ultrasonic reflection into the main control quantity. The thickness target trajectory can be dynamically planned according to the growth stage, with smooth transitions between stages, reducing supersaturation abrupt changes, stabilizing the crystal window, and improving the quality and process consistency of single-crystal silicon carbide epitaxial growth. 104. Obtain the crystal form discrimination value of each radial region in the seed crystal to obtain multiple crystal form discrimination values; In this embodiment, the detection unit arranges quality characterization channels along the radial direction of the seed crystal, with each quality characterization channel corresponding to a radial measurement point. In-situ detection methods such as reflectivity oscillation, Raman spectroscopy, and photoluminescence can be used to output the crystal form discrimination value of each radial region. Each crystal form discrimination value characterizes the quality of the crystal form of the single crystal silicon carbide at the corresponding radial position, as well as the presence of polycrystalline, defective, and other irregular growth characteristics. By simultaneously acquiring data from multiple points in the radial direction of the wafer, a set of crystal form discrimination values ​​covering the entire radial region of the wafer is obtained, providing raw quality detection input data for subsequent radial irregularity identification and incremental calculation of partitioned power adjustment. 105. Generate multiple limiting power adjustment increments based on preset irregularity identification conditions, preset partition power bias calculation formulas, and crystal form discrimination values; In this embodiment, the irregularity identification condition is used to determine whether irregular crystal growth defects occur at each radial point. The crystal form discrimination value corresponding to each radial point is traversed and compared with the irregularity identification condition to identify radial regions with crystal form abnormalities. Combined with the partition power bias calculation formula, the original power adjustment amount of the corresponding radial channel is solved, and the upper and lower power limits are introduced to limit the original adjustment amount to obtain the limited power adjustment increment corresponding to each radial channel. This limiting strategy prevents the single-channel power adjustment amount from being too large and causing severe thermal field disturbances, and provides partition power correction parameters for subsequent multivariate fusion to generate thickness adjustment commands. 106. Obtain the radii of multiple channels of the seed crystal, and generate silicon molten layer thickness adjustment instructions based on the lifting drive displacement, heating power bias, adjustment increment of each limiting power and each channel radius; In this embodiment, multiple channel radii represent the position coordinates of each quality characterization channel in the wafer radial direction, used to distinguish the effective range of different radial zones. The lifting drive displacement (mechanical displacement steady-state adjustment), heating power bias (global thermal disturbance compensation), and the limiting power adjustment increment of each radial channel (zone thermal field correction for crystal defects) are combined with the radial spatial distribution information represented by each channel radius to perform multivariate fusion calculation, generating a silicon molten layer thickness adjustment command adapted to the wafer radial thermal field distribution. This adjustment command simultaneously includes mechanical displacement control components and multi-zone heating power correction components, taking into account both steady-state control of molten layer thickness and local thermal field compensation for radial crystal defects, and is used to send to the drive unit and heating circuit to realize the control of the MSE growth device. In this embodiment, during the growth of single-crystal silicon carbide, the amount of crystal protrusions at the interface and the measured geometric spacing are collected. The thickness of the silicon molten layer is calculated using a preset laser displacement calculation method. Thickness deviation analysis is performed by combining the target thickness, deviation threshold, and deviation decomposition method. The lifting drive displacement and heating power bias are decoupled and output. Mechanical displacement is used to compensate for the gap drift caused by crystal thickening and material feeding erosion. Heating power is used to quickly suppress thermal disturbance and stabilize the interface carbon concentration gradient and supersaturation. At the same time, the crystal type discrimination value of each radial region of the seed crystal is collected. Based on the irregularity identification conditions and the partition power bias calculation formula, a partition power adjustment increment with amplitude limit constraint is generated. Multiple parameters are fused to generate silicon molten layer thickness adjustment instructions. This scheme constructs a thickness control and radial crystal type partition compensation mechanism to achieve dual-channel decoupled closed-loop control, balance the radial local supersaturation of the wafer, suppress radial irregularity defects such as polycrystalline and hillocks, perceive the growth status in real time, avoid the concentration gradient failure problem caused by insufficient material supply in the later stage of growth, and improve the uniformity of large-area wafers, long-term growth stability, and single-crystal product yield.

[0019] In a second embodiment of a method for adjusting the thickness of a molten silicon layer according to the present invention, step 103 includes: 201. Calculate the difference between the preset target thickness and the silicon molten layer thickness to obtain the total deviation; In this embodiment, the sensor body of the gap and silicon molten layer thickness detection unit is arranged in the low-temperature region of the chamber or outside the chamber, and optical coupling detection is achieved through a transparent window made of sapphire or quartz. The transparent window is equipped with an anti-deposition purge structure to reduce the detection signal attenuation caused by silicon vapor adhesion and deposition. The deposition amount of the window can be pre-calibrated and used for online measurement and correction. When using laser measurement, the short optical path design and purge airflow are combined to suppress the disturbance of optical signal by thermal convection in the chamber, ensuring the reliability of thickness detection. After acquiring the real-time thickness of the silicon molten layer, the difference between the real-time silicon molten layer thickness and the preset target thickness is calculated to obtain the total deviation. The total deviation characterizes the degree of deviation of the overall thickness of the molten layer from the target thickness, providing the original deviation input for subsequent threshold judgment, deviation decomposition and dual-channel control. 202. Determine whether the total deviation is greater than the preset deviation threshold; In this embodiment, a preset deviation threshold is used to implement threshold scheduling for large and small deviations, and hysteresis logic is configured to implement anti-jitter control; the determination logic compares the absolute value of the total deviation with the deviation threshold. For example, the threshold value is not less than 20% of the target thickness value; when If the condition is determined to be a large deviation, full-gain control of the power channel and output limiting strategy of the displacement channel are activated to avoid the risk of drive stroke saturation and prioritize the rapid suppression of transient disturbances; when If the condition is determined to be a small deviation steady-state condition, the power channel reduces the gain or stops regulating, and the displacement channel completes high-precision steady-state fine-tuning. The disturbance level is distinguished by the threshold judgment, the frequency band separation control is realized, the deviation condition judgment result is output, and the total deviation decoupling decomposition process is determined. 203. If the total deviation is greater than the preset deviation threshold, the total deviation is decomposed according to the preset deviation decomposition method to obtain the thickness deviation component and the temperature deviation component. In this embodiment, when the total deviation exceeds a preset deviation threshold, a transient condition indicating a significant deviation in the molten layer thickness is identified, and the deviation decoupling decomposition process is initiated. The decomposition principle is as follows: the thickness of the silicon molten layer is simultaneously affected by the coupling effect of the growth gap displacement and the internal thermal field. The displacement disturbance mainly causes a slow-changing, steady-state deviation in thickness, while the thermal field disturbance mainly causes a fast-changing, transient deviation in thickness. Based on the differences in the time scale and response frequency band of the two types of disturbances, the coupled total deviation is separated into components corresponding to different disturbance sources through frequency domain decoupling. The preset deviation decomposition method is used to split the coupled total deviation into a thickness deviation component and a temperature deviation component, realizing dual-channel decoupling control. Among them, the thickness deviation component is the low-frequency, slow-changing part of the total deviation, corresponding to the displacement channel; the temperature deviation component is the high-frequency, fast-changing part of the total deviation, corresponding to the heating power channel. The thickness deviation component is processed by the displacement channel, and the temperature deviation component is processed by the heating power channel. The power channel has a smaller time constant, suitable for quickly suppressing sudden transient disturbances; the displacement channel has a larger time constant, providing high-precision adjustment capabilities for maintaining the long-term steady-state thickness of the molten layer. The two types of deviation components obtained from the decomposition are independent of each other and are used as input signals for the displacement controller and power controller, respectively, realizing the division of labor between displacement adjustment and thermal field adjustment. The deviation decomposition method includes two feasible implementation methods: Method A (frequency domain separation): The total deviation is low-pass filtered (example cutoff frequency 0.1Hz), and the slow-changing component obtained after filtering is used as the thickness deviation component. The remaining fast-changing component after subtracting the thickness deviation component from the total deviation is used as the temperature deviation component. The total deviation is directly filtered in the frequency domain to separate the slow-changing component (thickness deviation component) and the fast-changing component (temperature deviation component). Both components are in the dimension of thickness. Method B (temperature sensing assistance): The temperature deviation ΔT is measured by thermocouple or radiation temperature measurement, and the calibration coefficient is used. The temperature deviation is converted into an equivalent thickness deviation, that is: , In the formula, This represents the total deviation in the thickness of the silicon molten layer. This is the thickness deviation component (the low-frequency, slowly varying deviation component corresponding to the displacement channel). The equivalent thickness deviation (μm) corresponding to the temperature is the equivalent thickness deviation of the molten layer calculated from the temperature deviation inside the cavity, which is used as the component corresponding to the heating power channel. The measured temperature deviation (K) inside the crystal growth cavity is the difference between the actual cavity temperature and the target set temperature, which is collected by a thermocouple or radiation temperature sensor. (μm / K) is the partial derivative coefficient of the molten layer thickness relative to temperature, which is a pre-calibrated coefficient representing the change in the thickness of the silicon molten layer caused by each unit change in the cavity temperature under steady-state conditions. It is a fixed coefficient obtained in advance through calibration experiments. The measured temperature deviation (multiplied by the partial derivative coefficient) is the equivalent thickness deviation, which is consistent with the dimensions of the total deviation. 204. Calculate the thickness deviation component according to the preset displacement conversion formula to obtain the lifting drive displacement. In this embodiment, the displacement channel is a slow-response, high-precision adjustment channel, employing PID control. The displacement conversion formula is as follows: ; In the formula, This is the output value of the displacement channel. The proportional coefficient of the displacement channel, The integral coefficients of the displacement channel, This is the differential coefficient of the displacement channel; the output of the displacement channel from the controller is converted into the lifting and lowering displacement of the seed crystal or the feeder, in micrometers. The drive unit can be a servo mechanism or a stepper mechanism, and the exemplary control resolution can reach 0.1μm. This channel is used to finely adjust the relative position between the seed crystal and the feeder under steady-state conditions, change the crystal growth gap, and provide position adjustment output for the steady-state maintenance of the molten layer thickness. 205. Calculate the temperature deviation component according to the preset heating power conversion formula to obtain the heating power offset. In this embodiment, the temperature deviation component is used as the input to the power controller. The heating power offset is obtained by solving a preset heating power conversion formula. The calculation principle is as follows: based on the mapping relationship between the thermal field and the erosion rate of the silicon molten layer, the temperature deviation component characterizes the degree of deviation of the actual thermal field in the cavity from the target thermal field. The preset heating power conversion formula adopts a PID control formula. , In the formula, This is the temperature deviation component. Power channel scaling factor, For integral coefficients, The differential coefficient represents the control output of the power channel output by the formula. As a bias for heating power The heating power conversion formula is used to convert the thermal field deviation into the power adjustment amount of the heating unit. By changing the heating output power, the thermal field distribution in the cavity is controlled, thereby changing the local melting rate of the silicon material and realizing the rapid correction of the molten layer thickness. The power bias is sent to the heating unit, and the local melting rate is changed by adjusting the thermal field to quickly correct the molten layer thickness. This is coordinated with the lifting and lowering drive displacement of the displacement channel to complete the fast and slow control. In this embodiment, an optical method is used to detect the thickness of the silicon molten layer. The sensor is external and combined with window purging, deposition calibration correction, and short optical path anti-convection design to improve the thickness detection accuracy in the high-temperature cavity. The total deviation is obtained by subtracting the measured silicon molten layer thickness from the target thickness. A threshold scheduling strategy with hysteresis anti-jitter is used to distinguish between large and small deviations. When the deviation exceeds the limit, the total deviation is separated into a thickness deviation component and a temperature deviation component based on the frequency domain decoupling principle. The thickness deviation component is converted into a lifting drive displacement by PID displacement conversion to finely adjust the growth gap. The temperature deviation component is converted into a heating power bias, which is used to change the silicon melting rate by thermal field control. The scheme constructs a fast and slow dual-channel control architecture. The power channel quickly suppresses transient large fluctuations, while the displacement channel achieves steady-state high-precision adjustment. At the same time, the displacement output is limited to avoid stroke saturation, effectively reducing detection interference in the high-temperature environment, stabilizing the silicon molten layer thickness and growth gap, ensuring the stability of the silicon carbide single crystal growth interface, and improving the quality of the finished crystal.

[0020] A third embodiment of a method for adjusting the thickness of a silicon molten layer in this invention includes step 105: 301. Based on the preset irregularity recognition conditions and preset irregularity recognition thresholds, perform discriminant analysis on each crystal form discrimination quantity to obtain the discriminant analysis results; In this embodiment, the crystal form discrimination value is used to characterize the degree to which the growth interface is within the target crystal form stability window; the preset irregularity identification conditions and irregularity identification thresholds are used to perform threshold comparison and logical discrimination on the crystal form discrimination values ​​of all radial measurement points to achieve early online identification of crystal form defects; the irregularity identification judgment logic is: if the conditions are met... If so, it is determined that there is a risk of heteromorphic or polycrystalline initiation in the corresponding radial region. The crystal form discrimination value is obtained in real time from the i-th radial measurement point, which is used to characterize the degree to which the growth interface at this radial position deviates from the target crystal form stability window; The crystal form discrimination value corresponding to the target crystal form. The non-deformation identification threshold was determined by offline calibration experiments. This partitioned quality feedback mechanism uses the crystal form competition state of each radial region, namely the competition between 3C and 4H crystal forms and the polycrystalline initiation state, as the feedback controlled variable. It can capture abnormal signals in the early stage of defect initiation and realize pre-defect intervention. By traversing all radial measurement points and comparing the crystal form discrimination value with the non-deformation identification threshold one by one, the whole radial crystal form state screening of the wafer is completed, and the discrimination analysis results are output. This provides a defect triggering basis for subsequent radial channel deviation acquisition, power increment calculation and partitioned process control. 302. If the discriminant analysis result shows that there is a crystal form discrimination quantity that meets the irregularity identification condition, then collect the radial channel deviations of the seed crystal. In this embodiment, when the discriminant analysis results determine that the crystal form discrimination value of at least one radial measuring point meets the irregular crystal identification condition, and irregular crystal or polycrystalline initiation signals are identified in the corresponding region, the radial channel deviation acquisition process is triggered; multiple measurement channels are arranged radially along the seed crystal, with 3 to 7 measurement points set for example, and the radial position of each measuring point is as follows: The system collects N radial channel deviations in real time. The radial channel deviations represent the difference between the actual thickness of the silicon molten layer at each measurement point and the target thickness. The collected radial channel deviations serve as input parameters for subsequent power calculations and can be called by the partitioned power control logic. Based on the collected radial channel deviations, the control unit can adjust the displacement of at least one of the seed crystal holder and the feeding holder through the drive unit, and / or adjust the heating power through the heating unit to control the crystal growth gap or the thickness of the silicon molten layer, keeping it stably within the process setting range. Each radial measurement channel is independent of each other, supporting independent partitioned adjustment of different radial regions of the wafer. This provides original deviation data support for targeted oversaturation control of defect areas and realizes the linkage between crystal quality feedback signals and geometric thickness feedback signals. 303. Analyze the deviation of each radial channel according to the preset calculation formula of the partition power bias and the preset amplitude limiting analysis conditions to obtain multiple amplitude limiting power adjustment increments. In this embodiment, the deviation of each radial channel is used as input, and the original power adjustment increment is solved zone by zone based on the partitioned power bias calculation formula. Then, the increment amplitude is constrained and clipped by preset limiting analysis conditions to output multiple limited power adjustment increments. This method realizes independent oversaturation control of each heating region in the radial direction of the wafer. It can apply a negative power bias to regions where the risk of aberration and polycrystalline initiation is identified. By reducing the local temperature, the erosion rate is reduced, and the growth interface is pulled back to the target crystal stability window. At the same time, the limiting constraint can limit the amplitude of a single power step, avoid violent fluctuations in the thermal field and instability of the molten layer. After the defect is eliminated, the parameters can be restored according to the gradual slope, reducing the process disturbance to adjacent regions. The limited power adjustment increment can be issued in conjunction with gap adjustment and global heating bias. While stabilizing the thickness of the silicon molten layer, it can suppress 3C / 4H crystal competition and polycrystalline generation in different zones, effectively improving the radial crystal uniformity of silicon carbide single crystal, the stability of the crystal growth interface, and the yield of the finished crystal product. In this embodiment, the crystal form discrimination value is obtained through radial multi-channel in-situ detection, and the early crystal form defect discrimination is completed using the irregularity identification threshold. After capturing the 3C / 4H crystal form competition and polycrystalline initiation signals, the radial channel deviation acquisition is triggered. The power adjustment increment is calculated by combining the partition power bias formula and the amplitude limiting processing is implemented to output a safe and controllable amplitude-limited power adjustment increment. The scheme adopts a control logic with defect suppression priority higher than geometric suppression priority, which can apply a negative power bias to the defect area and reduce local oversaturation with gap adjustment, so that the growth interface returns to the target crystal form stability window. The power increment limiting can suppress thermal field abrupt changes. After the defect is eliminated, the parameters are restored with a gradual slope to reduce cross-regional disturbances. This scheme realizes crystal form quality early warning and steady-state control of melt layer thickness, intervenes in the defect initiation stage in a partition, improves the radial crystal form uniformity of silicon carbide single crystal, the stability of the growth interface, reduces crystal defects, and improves the yield of finished products.

[0021] In a fourth embodiment of a method for adjusting the thickness of a molten silicon layer according to the present invention, step 303 includes: 401. Calculate the deviation of each radial channel according to the preset partition power offset calculation formula to obtain multiple power adjustment increments; In this embodiment, the formula for calculating the partitioned power bias is used to generate the power adjustment increment for partitioned heating based on the radial thickness deviation signal, thereby achieving oversaturation control of each independent radial heating zone. Independent partitioned P control refers to the use of independent proportional (P) control for each radial heating zone, with each zone's control loop decoupled from the others. The formula for calculating independent partitioned P control is as follows: , In the formula, For the power adjustment increment of the j-th radial partition, This refers to the power proportional control coefficient for the zone, in units of... The calibration was obtained through offline calibration experiments. Let $j$ be the radial channel deviation corresponding to the channel representing the j-th partition. The radial thickness deviation signal is a global characterization signal formed by the aggregation of all radial channel deviations. Each radial channel deviation is a discrete sampling component of the radial thickness deviation signal in space. Each radial channel deviation corresponds to the difference between the actual thickness of the silicon molten layer and the target thickness at a measurement point in the wafer's radial direction. The radial channel deviation of the channel representing the j-th partition is extracted from all radial channel deviations. As the input to the proportional controller (i.e., the partition P controller) set independently for the corresponding radial partition, it generates partition-independent power adjustment increments, and then synthesizes a complete radial thickness deviation feedback signal. The quality feedback signal and the geometric thickness feedback signal share the radial measuring points and actuators. The actuators include partition heating modules and gap adjustment modules. The two types of feedback dimensions are distinguished: the melt layer thickness deviation is used as the controlled variable to maintain the thickness steady state of each radial region; the quality uses the crystal quality state as the controlled variable to suppress the initiation of defects such as amorphous crystals and polycrystalline crystals. When the quality feedback signal is triggered, the geometric action of the corresponding radial region is downweighted or suspended, following the control strategy of first maintaining the crystal form and then maintaining the thickness. The defect suppression control actions include two types: applying a negative power bias to the current defect partition to bring the growth surface back to the target crystal stability window, reducing the local temperature to reduce the local dissolution rate, and thus reducing the composition supersaturation. And / or locally increase the gap h between the crystal and the feed source to reduce the carbon concentration gradient by increasing the gap value. Both control methods can reduce local supersaturation to the crystal form stability window; when the crystal form discrimination value... After returning to the threshold range, the controller gradually restores the power bias and gap parameters according to the preset gradual slope, suppressing the process disturbance caused by parameter mutation to the adjacent radial region; by substituting the radial channel deviation into each zone to complete the power increment solution, multiple independent power adjustment increments are generated, providing the original control amount for subsequent power output limiting processing, and realizing independent thermal field disturbance compensation in multiple radial zones. 402. Limit each power adjustment increment according to the preset limiting analysis conditions and preset limiting threshold to obtain multiple limited power adjustment increments; In this embodiment, the amplitude limiting analysis condition is used to constrain the maximum output amplitude of the partition power adjustment increment, avoiding excessive single power adjustment that could cause severe thermal fluctuations and melt layer instability; the amplitude limiting determination constraint is as follows: , for example ±5% of rated power; the original power adjustment increment obtained through the preceding steps. The signal is fed into the limiting stage, where incremental values ​​exceeding the positive and negative limiting thresholds are clamped within the threshold boundaries, and the limiting power adjustment increment is output. This stage adapts to the control logic of partition quality feedback: a quality characterization channel is arranged along the wafer radial direction as an online detection unit to identify the crystal form competition status of each radial region in real time, including 3C and 4H crystal form competition, polycrystalline initiation, and other conditions. When the online algorithm identifies amorphous or polycrystalline defect signals, it outputs a defect suppression command and performs an oversaturation reduction operation on the corresponding partition, that is, outputs a power negative bias command within the limiting range, or simultaneously issues a local gap increase command. Large command; after the defect signal disappears, the system gradually restores the limiting power adjustment increment and gap parameter to the steady-state set value with a gradually changing slope; throughout the entire process of quality feedback triggering, the geometric thickness of the corresponding radial partition is continuously reduced or paused to ensure that defect suppression is executed first; the power adjustment increment is safely trimmed by limiting constraints to prevent large power step disturbances from destroying the steady-state morphology of the silicon molten layer. The limited power increment can be sent to the partition heating execution unit, taking into account both the response capability of defect suppression and the stability of the single crystal growth process, and avoiding secondary process risks caused by thermal shock; In this embodiment, a radial partition control architecture for silicon carbide single crystals is constructed, reusing radial measuring points and partition heating and gap adjustment actuators. Using the deviation of each radial channel as input, the power adjustment increment is calculated using a partition-independent P control algorithm, and the increment is subject to amplitude limiting constraints to avoid severe thermal field disturbances caused by power step jumps. The scheme distinguishes the control priorities of thickness steady-state control and defect suppression. When defect signals such as 3C / 4H crystal form competition and polycrystalline initiation are detected, the oversaturation reduction control is performed first, and the geometric weighting of the corresponding partition is reduced or suspended. After the defect is eliminated, the process parameters are restored according to the gradual slope to reduce the disturbance of parameter abrupt changes to adjacent radial regions. This scheme can independently control the oversaturation at different radial positions of the wafer, stabilizing the thickness of the silicon molten layer and ensuring radial thickness uniformity while effectively suppressing the initiation of heteromorphic and polycrystalline defects, improving the stability of the single crystal growth interface and the yield of the finished crystal product.

[0022] In a fifth embodiment of a method for adjusting the thickness of a molten silicon layer according to the present invention, step 106 includes: 501. Calculate the radius of each channel according to the preset fitting radial first-order gradient calculation method and the preset tilt angle calculation formula to obtain the tilt angle. In this embodiment, the radial first-order gradient calculation method adopts the least squares fitting algorithm, using the channel radius and radial channel deviation corresponding to each radial measurement channel as input data. The channel radius represents the radial position of each measurement point from the seed crystal center, and the radial channel deviation is the deviation of the silicon molten layer thickness at the measurement point position relative to the target thickness. The radial first-order gradient is solved by fitting multiple sets of radial measurement point data. This gradient represents the degree of overall tilt change of the silicon molten layer thickness along the wafer radial direction. Then, the preset tilt angle calculation formula is substituted to complete the conversion and solve, and the tilt angle is output. The tilt angle is used to represent the tilt control amount of the tilt angle adjustment mechanism. In the case of small angle, an approximate simplified calculation can be performed. At the same time, an upper limit of the tilt angle is set as a limiting constraint to provide control parameters for subsequent clamping component tilt compensation. The tilt angle adjustment mechanism can drive the seed crystal clamping component or the feeding clamping component to tilt around the center to form a wedge gap to compensate for the radial overall thickness gradient deviation. This tilt angle control method is used to correct the global tilt of the entire wafer. It is independent of the local power bias adjustment and can be used in combination, providing tilt angle input for the tilt angle threshold judgment stage. 502. Determine whether the tilt angle is less than or equal to the preset angle threshold; In this embodiment, the angle threshold is the maximum allowable tilt boundary for the tilting mechanism to be safe and controllable. It is used to determine whether the current radial overall gradient deviation is within the range that the tilting mechanism can normally compensate for. The calculated tilt angle is compared with the preset angle threshold to complete the tilt angle adjustability determination. If the tilt angle exceeds the angle threshold, it means that the radial overall tilt exceeds the effective compensation range of the tilting mechanism, and the subsequent crystal rate parameter acquisition process is not entered. If the tilt angle meets the threshold constraint, it is determined that the radial overall gradient deviation can be compensated by tilting the clamping part, triggering the subsequent parameter acquisition process to ensure that the tilt angle adjustment action is within the controllable range of the mechanism and avoid problems such as abnormal seed crystal force and gap instability caused by excessive tilting. 503. If the tilt angle is less than or equal to the preset angle threshold, then obtain the crystal thickening rate and the feeding erosion rate; In this embodiment, when the determination result that the tilt angle meets the threshold constraint is met, it indicates that the radial overall thickness gradient deviation is within the controllable compensation range of the tilt angle mechanism, and the basic conditions for carrying out growth rate analysis and feedforward compensation calculation are met; then the acquisition process is triggered to capture and record the corresponding crystal thickening rate and feeding erosion rate under the single crystal silicon carbide growth condition in real time; at the same time, tilt angle limiting constraint is configured. The exemplary limiting range is 0.1° to 1°; the tilt adjustment mechanism drives the seed crystal clamping member or the feeding clamping member to tilt around the center, and relies on the formed wedge gap to compensate for the radial first-order gradient; 504. Analyze the crystal thickening rate and the feed dissolution rate to obtain the feedforward superposition amount and the remaining feed thickness; In this embodiment, the growth mechanism analysis criteria are used to conduct multi-dimensional mechanism analysis of crystal thickening rate and feed erosion rate. The analysis logic uses crystal thickening rate and feed erosion rate as model inputs. On the one hand, it solves the position feedforward quantity and fuses the feedback output to obtain the feedforward feedback superposition quantity, which is used to compensate for predictable gap drift and unknown disturbances. On the other hand, it relies on the feed consumption model and lifetime formula to solve the remaining feed thickness and estimated lifetime in real time. By substituting the rate parameters into the mechanism model for each calculation, the feedforward control quantity synthesis and feed consumption status assessment are completed, and the feedforward feedback superposition quantity and remaining feed thickness are generated, providing a basis for the generation of the final silicon molten layer thickness adjustment command. 505. Generate silicon molten layer thickness adjustment instructions based on lifting drive displacement, heating power offset, adjustment increment of each limiting power, tilt angle, feedforward feedback superposition, and remaining material thickness. In this embodiment, the lifting drive displacement, tilt angle, and feedforward feedback superposition jointly act on the seed crystal lifting drive and / or feeding lifting drive hardware. The feedforward feedback superposition is used to compensate for the predictable drift caused by crystal thickening and feeding erosion. The lifting drive displacement achieves steady-state correction of the molten layer thickness. The tilt angle is adjusted by the tilt angle adjustment mechanism to compensate for the radial thickness gradient by adjusting the tilt posture of the seed crystal holder / feeding holder. The heating power bias and each limiting power adjustment increment act on the heating unit. The heating power bias is a global heating power bias used to suppress transient thickness deviations caused by thermal disturbances. Each limiting power adjustment increment is output to the heating unit. Each radial zone heating loop achieves radial local thermal field compensation and suppresses crystal defect initiation. The remaining thickness of the feed material is not directly involved in the hardware adjustment output; it is only used for feed life warning. When the remaining thickness of the feed material is lower than the remaining thickness threshold, the control unit outputs an alarm or terminates the growth signal, without issuing adjustment actions to the displacement drive or heating unit. The multi-control quantity fusion strategy sets multiple constraints to perform fusion scheduling and logical analysis on the lifting drive displacement, heating power bias, each limit power adjustment increment, tilt angle, feedforward feedback superposition, and remaining thickness of the feed material. The judgment logic follows the priority of quality over geometry, and geometry over feedforward compensation. Quality relies on crystal form discrimination to suppress defect initiation, geometry maintains the steady-state thickness of the molten layer, and feedforward compensation predicts drift. These three types of compensation share radial measurement points and zoned heating, gap adjustment, and tilting actuators. The lifting drive displacement is responsible for axial gap correction, the heating power offset is the global temperature control amount, and each limit power adjustment increment is used for radial local thermal field correction. The heating power offset is converted into the global base power adjustment amount of the heating unit, uniformly adjusting the overall output power of the heating unit upwards or downwards to achieve global temperature control and compensate for overall thermal disturbances during the growth process. Each limit power adjustment increment is converted into the corresponding radial... The local power correction of the zoned heating loop is superimposed on the global base power to fine-tune the power of each radial zoned heating loop individually, thereby achieving radial local thermal field correction. The sum of the two forms the final power output value of each radial zoned heating loop, which together act on the heating unit to complete the thermal field adjustment. The tilt angle is used for global wedge gap compensation, the feedforward feedback superposition is used for gap drift comprehensive correction, and the remaining thickness of the feed is used for feed status monitoring. By integrating various control parameters and status parameters, the control signals of multiple actuators are synthesized, and the silicon molten layer thickness adjustment command is output and sent to the growth device controller to drive the action of each actuator. In this embodiment, the radial first-order gradient is obtained by least-squares fitting of radial multi-point data and converted into tilt angle. Combined with tilt angle threshold admission judgment, growth rate parameters are collected within the compensable range of the mechanism. The remaining thickness and lifetime of the feed are evaluated based on the linear consumption model of the feed. Then, multiple control parameters are fused according to the priority of mass, geometry, and feedforward compensation to generate adjustment commands. The scheme can separate the global radial tilt deviation and local thickness deviation of the silicon molten layer. The tilt mechanism forms a wedge gap to correct the overall tilt of the entire wafer. Combined with power bias, local thermal field control is achieved. The two compensate for the thickness gradient. At the same time, the gap drift feedforward correction is realized based on the growth rate. The feed consumption status is monitored in real time to avoid abnormal seed crystal stress and gap instability risks caused by over-range tilting, suppress crystal defects, stably maintain the target thickness of the silicon molten layer, and improve the thickness uniformity and process controllability of single crystal silicon carbide growth.

[0023] In a sixth embodiment of a method for adjusting the thickness of a molten silicon layer according to the present invention, step 501 includes: 601. Calculate the radius of each channel and the deviation of each radial channel according to the preset fitting radial first-order gradient calculation method to obtain the radial first-order gradient; In this embodiment, the method for calculating the radial first-order gradient uses a least-squares fitting algorithm to perform regression analysis on the spatial coordinates and thickness deviation data of each radial measurement point; the calculation logic employs the least-squares formula. Complete the calculation. In the formula, It is the radial first-order gradient; The radius of the channel at each measuring point, The mean of the channel radius. For the radial channel deviation at each measuring point, The radial channel deviation is the average value. Several radial measurement channels (example 3-7 measurement points) are arranged radially along the seed crystal. The channel radius represents the radial position of each measurement point from the center of the seed crystal, and the radial channel deviation represents the deviation of the silicon molten layer thickness at the measurement point from the target thickness. By performing least squares fitting on multiple sets of channel radius and radial channel deviation datasets, the first-order radial gradient is obtained. This gradient represents the overall tilt change of the silicon molten layer thickness along the wafer radial direction. The first-order radial gradient serves as a pre-input quantity for tilt angle conversion, providing gradient input parameters for subsequent tilt angle calculation, ensuring that the tilt angle adjustment output can match the overall tilt characteristics of the radial thickness, and providing a basic gradient basis for the wedge gap compensation of the clamping component tilt. By collecting channel radius and radial channel deviation at multiple radial measurement points, the first-order radial gradient is obtained by fitting using the least squares algorithm, establishing a gradient mapping relationship between the radial spatial position and the molten layer thickness deviation. This gradient quantifies the global tilt trend of the molten layer thickness of the entire wafer, which is different from the local deviation at a single point. It can be used for subsequent tilt angle command conversion and solution. The tilt angle command can be selected from... Under small angle conditions, it can be approximated as Two-point method can also be used. Solve for the tilt angle, where, The thickness deviation of the molten layer at the Nth radial measuring point; This refers to the thickness deviation of the molten layer at the first radial measuring point; The radial radius corresponding to the Nth radial measuring point. Given the radial radius corresponding to the first radial measurement point, this scheme is based on multi-radial measurement point data. It extracts the overall radial tilt feature of the thickness through least squares fitting, separates the global gradient deviation from the local single-point deviation, provides reliable gradient observation values ​​for the tilt angle compensation control of MSE growth, supports radial wedge gap correction, improves the radial uniformity of silicon molten layer thickness, and reduces the risk of crystal defects induced by radial thickness gradient. 602. Calculate the radial first-order gradient according to the preset tilt angle calculation formula to obtain the tilt angle; In this embodiment, the preset tilt angle calculation formula sets a fixed mapping operation relationship, which is used to convert and solve the input radial first-order gradient, and output the tilt angle for mechanism control; the radial first-order gradient y calculated in the previous step needs to be substituted to complete the calculation. The tilt angle calculation formula is as follows: In the formula, y is the radial first-order gradient. The tilt angle is calculated by substituting the first-order radial gradient obtained in the previous step into the tilt angle calculation formula. The MSE growth apparatus is equipped with a tilt angle adjustment mechanism, which can tilt the seed crystal holder or the feeding holder. By tilting the holder, a wedge-shaped gap can be formed to compensate for the overall thickness gradient deviation in the radial direction of the wafer. The heating unit of the apparatus is divided into inner, middle, and outer radial heating zones on the coplanar horizontal plane of the molten layer. The radial heating zones are matched one-to-one with the radial measurement channels. The radial zone adjustment is achieved by adjusting the heating power bias of each radial zone through the zone heating unit. The deviation of each zone's representative channel is calculated independently. Local temperature changes affect the local erosion rate and molten layer thickness, thereby correcting the local gap. Radial zone compensation includes two implementation paths: one is to use a zone heating unit to calculate the power adjustment increment of each zone based on the deviation of the representative channel, and rely on the local temperature to change the erosion rate and silicon molten layer thickness of the corresponding area to complete the local gap correction; the other is to use the tilt angle output by the tilt angle adjustment mechanism to execute the tilting action of the clamping part, generating a wedge gap to compensate for the overall radial first-order gradient deviation. The two compensation methods can be used individually or in combination, providing tilt angle control input for the radial gap uniformity correction of MSE growth. In this embodiment, by collecting the channel radius and radial channel deviation of multiple radial measurement channels, the first-order radial gradient is obtained using least-squares fitting. This gradient is then converted using the tilt angle calculation formula to obtain the tilt angle, providing control input for the tilt angle adjustment mechanism. A wedge-shaped gap is formed by tilting the clamping member, achieving overall radial thickness gradient deviation compensation for the wafer. Simultaneously, radial partition heating units corresponding to each measurement channel are set up. The power bias of each region is independently calculated based on the deviation of the representative channel in each partition. Local temperature control is used to change the etch rate and molten layer thickness, completing local gap correction. Tilt angle adjustment and partition heating compensation can be executed individually or in combination, taking into account both overall radial tilt deviation and local gap differences. This effectively improves the radial uniformity of the silicon molten layer thickness in large-area seed wafers, reduces crystal defects such as polycrystalline material and hillocks induced by radial thickness non-uniformity, and enhances the process stability and crystal product quality of MSE single-crystal silicon carbide growth.

[0024] In the seventh embodiment of a method for adjusting the thickness of a silicon molten layer according to the present invention, step 504 includes: 701. Calculate the crystal thickening rate and the feed dissolution rate based on the preset feedforward model to obtain the position feedforward amount; In this embodiment, the feedforward model is constructed based on the MSE growth mathematical model and is used to characterize the crystal thickening rate. Feed dissolution rate The intrinsic correlation between the same gap drift and the relevant model parameters were obtained through pre-experimental calibration; the model determination logic is as follows: the crystal thickening rate is solved according to the Arrhenius formula. With feed dissolution rate : , Feed dissolution rate: , Net drift rate of the gap: ; Based on this, the feedforward compensation speed command is obtained: , Then according to the sampling period The cumulative position feedforward is obtained as follows: ; The model can also be used to solve for the crystal protrusion at the interface through integration. Used for converting the thickness of the silicon molten layer; In the formula, For carbon concentration gradient, The Arrhenius precession coefficient for crystal growth is determined by preliminary experiments. The Arrhenius pre-correlation coefficient corresponding to the feed dissolution was determined by preliminary experiments; To provide activation energy for material dissolution, The activation energy for crystal growth. This is the universal gas constant. Thermodynamic temperature For feedforward compensation speed, For position feedforward, The discrete sampling time sequence number is a positive integer. Through feedforward model calculation, the displacement compensation amount (position feedforward amount) is output in advance for the predictable gap drift caused by crystal thickening and feeding erosion, providing feedforward dimension input for subsequent control quantity superposition, and realizing the prediction compensation of drift trend. Let be the amount of crystal protrusion at time t; Let be the crystal thickening rate at time t; For the integral operation over time, the physical meaning is that the crystal thickening rate accumulates over time to obtain the crystal growth protrusion length; under steady-state conditions... Approximately constant, can be simplified to Used for converting the thickness of the silicon molten layer (silicon molten gap); Through feedforward model calculation, the displacement compensation amount is output in advance for the predictable gap drift caused by crystal thickening and feeding erosion, providing feedforward dimension input for subsequent control quantity superposition, and realizing the prediction compensation of drift trend. 702. Calculate the position feedforward amount and the preset feedback output amount according to the preset control quantity superposition calculation formula to obtain the feedforward feedback superposition amount. In this embodiment, the control quantity superposition calculation formula is used to fuse the control outputs of the feedforward channel and the feedback channel. The calculation formula is as follows: Position feedforward in control logic The feedforward compensation speed command is integrally calculated to specifically compensate for predictable slow gap drift caused by crystal thickening and feed erosion; the feedback output is obtained from this calculation. The output of the PID controller is the gap deviation between the actual measured value of the molten gap and the target set value. The PID controller outputs a displacement correction quantity, specifically designed to handle model errors and unknown disturbances from the external environment; the position feedforward quantity... With feedback output Both are displacement quantities, with consistent dimensions (unit: μm), and can be directly superimposed; simultaneously, an integral separation constraint condition is configured, which is applied only when the absolute value of the clearance deviation is... Error integration is only performed when the error exceeds the dead zone threshold (example ±1μm) to avoid control saturation failure caused by repeated integration in the feedforward and feedback loops; during growth, the initial seed-feed gap is... As the crystal continues to thicken and the surface of the feed piece is eroded, causing the upper surface to shift downwards, if the clamping element remains stationary, the gap will decrease to [a smaller value]. This results in a corresponding reduction in the thickness of the silicon molten layer; , The difference between the two values ​​obtained through preliminary experimental calibration is used to determine the net drift rate and the feedforward compensation speed. The control unit adds the position feedforward and feedback output to obtain the total displacement control value, which drives the seed crystal holder to retract upward and / or drives the feeding holder to advance upward, correcting the drifted gap and restoring it to the target gap. The feedforward mechanism is responsible for predicting and compensating for drift trends, while the feedback mechanism is responsible for eliminating model residual disturbances. The two control signals work together to provide comprehensive control input for subsequent actuator displacement adjustment, ensuring the response speed and steady-state control accuracy of the gap adjustment. 703. Calculate the erosion rate of the feed material according to the preset feeding life estimation method to obtain the remaining thickness of the feed material. In this embodiment, the feeding life estimation method is based on the initial thickness of the feeding component. Feed dissolution rate As input parameters, according to the formula Calculate the remaining thickness of the material being supplied in real time. The remaining thickness of the feed piece at time t; The initial thickness of the feed part; and based on The estimated lifespan of the feed material is calculated. The total lifespan of the feed material is determined by the initial thickness of the feed material and the rate of erosion consumption per unit time. For example, if the initial thickness of the feed layer is 300 μm and the erosion consumption rate is 5 μm / h, the estimated total lifespan is 60 h. The warning threshold supports two configuration modes: a fixed thickness threshold and a remaining lifespan percentage threshold. In the fixed threshold mode, a warning is triggered when the remaining thickness of the feed material is ≤30 μm. In the percentage threshold mode, a warning is triggered when the remaining lifespan is ≤10% (approximately 6 h remaining under the example conditions). When the monitoring result is lower than the preset threshold, the control unit can output an audible and visual signal, a host computer prompt signal, or trigger a notification to indicate the end of growth. The separation process terminates crystal growth to prevent the feeder from being completely melted through and to prevent interruption of the carbon source supply to the molten layer. It should be understood that the displacement and propulsion of the feeder clamp is uniformly driven by the aforementioned feedforward-feedback control system. The feed life prediction logic does not generate independent displacement control commands or add independent movement rules. Its function is only to monitor the feed consumption status online and ensure the continuous supply of carbon source material during the growth process. It does not participate in maintaining the seed-feed gap. By continuously analyzing the feed erosion consumption status, the remaining thickness and remaining life of the feed are monitored online, providing a safety protection basis for the MSE single crystal silicon carbide growth process at the feed status level. In this embodiment, a pre-set feedforward model is used to solve for the crystal thickening rate and feed erosion rate using the Arrhenius growth equation, and the position feedforward is calculated to pre-compensate for predictable gap drift during the growth process. Then, the position feedforward and PID feedback output are fused through a control quantity superposition calculation formula, and integral separation constraints are used to avoid integral saturation. The output feedforward feedback superposition is used for gap correction. At the same time, based on the feed erosion rate, the remaining feed thickness is calculated in real time using a feed lifetime prediction method to achieve feed consumption status monitoring and threshold alarm. This scheme constructs a multi-layer control system of "feedforward prediction of drift, feedback correction of residual, and feed status early warning", which takes into account both pre-compensation for predictable drift and correction of unknown disturbances, while monitoring the feed consumption status, stabilizing the working range of the silicon molten layer, avoiding the growth failure risk caused by feed melt-through, and improving the control stability and process safety of the MSE single crystal silicon carbide growth process.

[0025] The above describes a method for adjusting the thickness of a silicon molten layer according to an embodiment of the present invention. The following describes a device for adjusting the thickness of a silicon molten layer according to an embodiment of the present invention. Please refer to [link to relevant documentation]. Figure 2 One embodiment of the silicon molten layer thickness adjustment device described in this invention includes: A silicon molten layer thickness adjustment device includes: a control unit 1 for performing the silicon molten layer thickness adjustment method as described above; a seed crystal 3, a feed element 4, a seed crystal clamping element 2, and a feed clamping element 5 electrically connected to the control unit 1; the feed element 4 is disposed at the upper end of the feed clamping element 2; the seed crystal 3 is disposed below the seed crystal clamping element 2; and a silicon molten layer is formed between the seed crystal 3 and the feed element 4.

[0026] Seed crystal 3 is a 4H-SiC single wafer (example: φ2 inch, 0.4 mm thick, growth surface is C-side), fixed to the lower end of seed crystal clamp 2; feed component 4 is a feed substrate with a SiC surface layer (example: polycrystalline SiC layer deposited on the surface of graphite support, thickness 50-500 μm), fixed to the upper end of feed clamp 5; seed crystal 3 and feed component 4 are opposite each other, and a high-purity Si wafer (example: 50 μm thick) is placed between them, which is heated and melted to form a silicon molten layer, with an initial target thickness of 50 μm in the example; the heating unit heats to the growth temperature (example: 1600-1900℃), and the chamber is inert atmosphere.

[0027] Figure 3This is a schematic diagram of a silicon molten layer thickness adjustment device 900 provided in an embodiment of the present invention. The silicon molten layer thickness adjustment device 900 can vary considerably due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the silicon molten layer thickness adjustment device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the silicon molten layer thickness adjustment device 900 to implement the steps of the silicon molten layer thickness adjustment method provided in the above-described method embodiments.

[0028] A silicon molten layer thickness adjustment device 900 may further include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, MacOSX, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated silicon molten layer thickness adjustment device structure does not constitute a limitation on the silicon molten layer thickness adjustment device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0029] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a silicon molten layer thickness adjustment method.

[0030] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0031] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0032] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of adjusting the thickness of a silicon melt layer, characterized by, include: The amount of interfacial crystal protrusions and the measured geometric spacing were obtained during the growth of single-crystal silicon carbide. The thickness of the silicon molten layer is obtained by calculating the amount of crystal protrusion at the interface and the measured geometric spacing according to the preset laser displacement calculation method. The thickness of the silicon molten layer is analyzed based on the preset target thickness, preset deviation threshold, and preset deviation decomposition method to obtain the lifting drive displacement and heating power offset. The crystal form discrimination value of each radial region in the seed crystal is obtained to obtain multiple crystal form discrimination values; Multiple limiting power adjustment increments are generated based on preset anomaly identification conditions, preset partition power bias calculation formulas, and crystal form discrimination values. The radii of multiple channels of the seed crystal are obtained, and the silicon molten layer thickness adjustment command is generated based on the lifting drive displacement, heating power bias, adjustment increment of each limiting power and each channel radius.

2. The method for adjusting the thickness of a silicon molten layer as described in claim 1, characterized in that, The analysis of the silicon molten layer thickness based on a preset target thickness, a preset deviation threshold, and a preset deviation decomposition method to obtain the lifting drive displacement and heating power offset includes: The difference between the preset target thickness and the silicon molten layer thickness is calculated to obtain the total deviation; Determine whether the total deviation is greater than the preset deviation threshold; If the total deviation is greater than the preset deviation threshold, the total deviation is decomposed according to the preset deviation decomposition method to obtain the thickness deviation component and the temperature deviation component. The thickness deviation component is calculated according to the preset displacement conversion formula to obtain the lifting drive displacement. The temperature deviation component is calculated based on the preset heating power conversion formula to obtain the heating power offset.

3. The method for adjusting the thickness of a silicon molten layer as described in claim 1, characterized in that, The process of generating multiple limiting power adjustment increments based on preset anomaly identification conditions, preset partition power bias calculation formulas, and crystal form discrimination values ​​includes: Based on the preset anomaly recognition conditions and preset anomaly recognition thresholds, the discrimination parameters of each crystal form are analyzed to obtain the discrimination analysis results; If the discriminant analysis result indicates that there is a crystal form discrimination value that meets the anomaly identification condition, then collect the radial channel deviations of the seed crystal. The deviation of each radial channel is analyzed based on the preset formula for calculating the partition power bias and the preset amplitude limiting analysis conditions to obtain multiple amplitude limiting power adjustment increments.

4. The method for adjusting the thickness of a silicon molten layer as described in claim 3, characterized in that, The method involves analyzing the deviation of each radial channel based on a preset formula for calculating the partition power bias and preset limiting analysis conditions to obtain multiple limiting power adjustment increments, including: The deviation of each radial channel is calculated according to the preset partition power bias calculation formula to obtain multiple power adjustment increments; Based on preset limiting analysis conditions and preset limiting thresholds, each power adjustment increment is limited to obtain multiple limited power adjustment increments.

5. The method for adjusting the thickness of a silicon molten layer as described in claim 3, characterized in that, The process of generating silicon molten layer thickness adjustment commands based on lifting drive displacement, heating power offset, adjustment increments of each limiting power, and radius of each channel includes: The radius of each channel is calculated based on the preset fitting radial first-order gradient calculation method and the preset tilt angle calculation formula to obtain the tilt angle. Determine whether the tilt angle is less than or equal to a preset angle threshold; If the tilt angle is less than or equal to the preset angle threshold, the crystal thickening rate and the feed dissolution rate are obtained. The crystal thickening rate and feed erosion rate were analyzed to obtain the feedforward superposition amount and the remaining feed thickness. The silicon molten layer thickness adjustment command is generated based on the lifting drive displacement, heating power offset, adjustment increment of each limiting power, tilt angle, feedforward feedback superposition, and remaining material thickness.

6. The method for adjusting the thickness of a silicon molten layer as described in claim 5, characterized in that, The calculation of the radius of each channel according to the preset fitting radial first-order gradient calculation method and the preset tilt angle calculation formula to obtain the tilt angle includes: The radial first-order gradient is calculated based on the preset fitting radial first-order gradient calculation method to obtain the radial first-order gradient. The radial first-order gradient is calculated according to the preset tilt angle calculation formula to obtain the tilt angle.

7. The method for adjusting the thickness of a silicon molten layer as described in claim 5, characterized in that, The analysis of the crystal thickening rate and feed erosion rate to obtain the feedforward superposition amount and the remaining feed thickness includes: The crystal thickening rate and feed dissolution rate are calculated based on the preset feedforward model to obtain the position feedforward amount; The position feedforward quantity and the preset feedback output quantity are calculated according to the preset control quantity superposition calculation formula to obtain the feedforward feedback superposition quantity. The feeding erosion rate is calculated based on the preset feeding life estimation method to obtain the remaining feeding thickness.

8. A device for adjusting the thickness of a silicon molten layer, characterized in that, include: A control unit for performing the silicon molten layer thickness adjustment method as described in any one of claims 1-7, and a seed crystal, a feeder, a seed crystal holder, and a feeder holder electrically connected to the control unit; the feeder is disposed at the upper end of the feeder holder; the seed crystal is disposed below the seed crystal holder; a silicon molten layer is formed between the seed crystal and the feeder.

9. A silicon molten layer thickness adjustment device, characterized in that, The silicon molten layer thickness adjustment device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the silicon molten layer thickness adjustment device to perform the steps of the silicon molten layer thickness adjustment method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the silicon molten layer thickness adjustment method as described in any one of claims 1-7.