Method and system for real-time regulation and control of crystal orientation in micron-sized single crystal growth process
By collecting and analyzing the physical signals of the crystal growth interface and using a micro heat source array for local thermal field control, the lag problem of micron-level crystal orientation control in single crystal growth was solved, achieving high-precision, non-destructive crystal orientation control and improving the quality and consistency of single crystal materials.
Patent Information
- Application Number
- CN202511401850.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-28
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies struggle to achieve real-time crystal orientation control at the micrometer level during single crystal growth. Macroscopic parameter adjustments suffer from lag and linkage effects, and online monitoring technologies lack sufficient resolution or interfere with the growth process in high-temperature, enclosed environments.
By acquiring the inherent physical signals of the crystal growth interface, generating raw signal data, extracting the crystal orientation shift feature vector, generating quantitative indicators using a preset mapping model, driving the micro heat source array to apply a local thermal field gradient for crystal orientation correction, and iteratively optimizing through closed-loop control.
It achieves instantaneous response and iterative correction to crystal orientation shift, avoids interference from traditional detection methods in the growth process, improves the internal quality and uniformity of crystal growth, and increases yield and performance consistency.
Smart Images

Figure CN120905767A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of crystal growth, and relates to a real-time crystal direction regulation method and system in a micron-level single crystal growth process. BACKGROUND
[0002] Single crystal materials, especially semiconductor, optical and laser crystals, are the cornerstone of modern high-tech industries. In the single crystal growth process, the crystal orientation, i.e. the crystal direction, is a key parameter that determines its physical and chemical properties. In order to obtain single crystals with specific properties, the accuracy and consistency of the crystal direction must be strictly controlled throughout the growth process. The real-time crystal direction regulation method in the micron-level single crystal growth process is a key process technology aimed at improving the level of high-quality single crystal manufacturing, and its core goal is to maintain the stable growth of the crystal along the preset crystal axis direction in the dynamic process of atoms or molecules continuously adhering to the growth interface.
[0003] Currently, in the practice of single crystal growth, the control of the crystal direction mainly relies on the accurate calibration of the seed crystal in the initial stage, and the indirect adjustment of macroscopic process parameters such as overall furnace temperature, pulling or rotating speed during the growth process. Some advanced processes will also be supplemented by X-ray diffraction or high-power optical microscopy for offline and periodic crystal direction detection. Once a deviation is found, manual intervention or adjustment of macroscopic parameters is used for rough correction. These methods can ensure the usability of the final product to a certain extent, but they are not sufficient in the context of pursuing higher crystal quality.
[0004] The existing technical means have obvious limitations in responding to the real-time crystal direction regulation demand of the micron level, which is specifically manifested as follows: 1. The adjustment of macroscopic parameters has significant hysteresis and linkage effect, making it difficult to achieve rapid and independent response to local minor deviations, and easily introducing new global thermal stress.
[0005] 2. The existing online monitoring technology applied to the high-temperature and closed growth environment is either difficult to capture micro changes due to insufficient resolution, or cannot obtain timely and accurate deviation information due to the interference of the detection means itself with the growth process. SUMMARY
[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the technical scheme of the present application is as follows: the first embodiment of the present application provides a real-time crystal direction regulation method in a micron-level single crystal growth process, comprising: Collecting inherent physical signals of the crystal growth interface region to generate original signal data.
[0007] Extracting feature parameters associated with crystal direction deviation from the original signal data to generate a crystal direction deviation feature vector.
[0008] The crystal orientation deviation feature vector is input into a preset mapping model to generate a crystal orientation deviation quantification index.
[0009] Based on the crystal orientation deviation quantification index, a local thermal field regulation instruction is calculated to drive a micro heat source array to apply a local thermal field gradient to a target region of the crystal growth interface to guide the crystal orientation back to normal.
[0010] The second embodiment of the present application provides a real-time crystal orientation regulation system for a micron-scale single crystal growth process, which comprises a signal acquisition module, a feature extraction module, an intelligent decision module, a thermal field regulation module and a closed-loop control module.
[0011] The signal acquisition module is connected to the feature extraction module, the feature extraction module is connected to the intelligent decision module, the intelligent decision module is connected to the thermal field regulation module, and the thermal field regulation module is connected to the closed-loop control module.
[0012] The signal acquisition module acquires inherent physical signals of the crystal growth interface region to generate original signal data.
[0013] The feature extraction module extracts feature parameters associated with the crystal orientation deviation from the original signal data to generate a crystal orientation deviation feature vector.
[0014] The intelligent decision module inputs the crystal orientation deviation feature vector into a preset mapping model to generate a crystal orientation deviation quantification index.
[0015] The thermal field regulation module calculates a local thermal field regulation instruction based on the crystal orientation deviation quantification index to drive a micro heat source array to apply a local thermal field gradient to a target region of the crystal growth interface to guide the crystal orientation back to normal.
[0016] The closed-loop control module performs iterative updating of the local thermal field regulation parameters based on real-time feedback data during the regulation process until the crystal orientation deviation converges to within a preset tolerance.
[0017] Compared with the prior art, the present application has the following advantages: (1) The present application realizes real-time microscopic monitoring of the growth state by high-sensitivity non-contact acquisition of inherent physical signals of the crystal growth interface and real-time conversion into a crystal orientation deviation feature vector, fundamentally avoids physical interference or damage to the crystal growth process that may be caused by traditional detection methods, guarantees the purity and stability of the growth environment, provides high-quality and non-destructive decision basis for subsequent precise regulation, and thus improves the regulation foundation reliability.
[0018] (2) The application innovatively uses a micro heat source array to apply a controllable thermal field gradient to a local area of a growth interface, and uses the anisotropy principle of crystal growth to generate a weak and accurate crystal orientation correction torque, the regulation means has accurate range and controllable strength, avoids global thermal stress fluctuation and new crystal defect hidden danger caused by traditional large-scale temperature field regulation, realizes flexible fine adjustment of the crystal orientation, and significantly improves the internal quality and uniformity of crystal growth.
[0019] (3) The application deeply integrates high-precision sensing, intelligent mapping decision and localized accurate execution, constructs a high-speed closed-loop control, can respond and iteratively correct the micron-level crystal orientation deviation in real time, continuously and stably keeps the crystal orientation deviation in a very small range, effectively suppresses various macro defects caused by long-term cumulative deviation of the crystal orientation, and significantly improves the yield and performance consistency of high-quality single crystal materials. BRIEF DESCRIPTION OF DRAWINGS
[0020] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0021] Figure 1 The method implementation step flow chart provided for the first embodiment of the application.
[0022] Figure 2 The logic flow chart for establishing the preset mapping model in the first embodiment of the application.
[0023] Figure 3 The module connection schematic diagram provided for the second embodiment of the application.
[0024] Figure 4 The crystal growth interface region device layout diagram in the first embodiment of the application. DETAILED DESCRIPTION
[0025] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.
[0026] As Figure 1 and Figure 4As shown, the first embodiment of the present application provides a real-time crystal orientation control method for micron-scale single crystal growth process, comprising: S11. Collecting inherent physical signals of the crystal growth interface region to generate original signal data.
[0027] In specific embodiments of the present application, the inherent physical signal collection process comprises: capturing specific frequency spectrum thermal fluctuation signals of the growth interface radiation through a non-contact sensor array.
[0028] Synchronously collecting acoustic emission elastic wave signals released by the growth interface.
[0029] Fusing the thermal fluctuation signals and the acoustic emission elastic wave signals to generate multi-modal original signal data.
[0030] S12. Extracting feature parameters associated with crystal orientation deviation from the original signal data to generate a crystal orientation deviation feature vector.
[0031] In specific embodiments of the present application, extracting feature parameters associated with crystal orientation deviation from the original signal data comprises: performing time-frequency joint analysis on the original signal data to extract signal intensity fluctuation patterns in a preset frequency band.
[0032] It should be noted that the above time-frequency joint analysis can exemplarily use continuous wavelet transform to process the multi-modal original signal data. Wavelet transform is a mathematical tool that can characterize the local features of signals in both time and frequency domains, overcoming the shortcomings of Fourier transform that cannot provide time localization information. Through this analysis, the pattern of signal energy change with time, i.e., the signal intensity fluctuation pattern, can be extracted in a series of preset feature bands closely related to lattice vibration and stress release.
[0033] Calculating the phase difference distribution of each unit signal in the non-contact sensor array.
[0034] It should be noted that the above phase difference reflects the time delay condition of signal wavefront arriving at different sensor units, thereby carrying the spatial position and orientation information of the growth interface signal source, and obtaining the phase difference distribution.
[0035] Fusing the signal intensity fluctuation pattern and the phase difference distribution to generate a crystal orientation deviation feature vector containing spatial dimension information.
[0036] It should be noted that the fusion of the signal intensity fluctuation pattern and the phase difference distribution is a structured combination of the two. This process is not simply a numerical addition, but rather a splicing of the two as independent dimension data to construct a multi-dimensional crystal orientation deviation feature vector. Its mathematical form can be represented as where P represents the final generated crystal orientation deviation feature vector, quantized values representing the signal intensity fluctuation pattern extracted in the kth preset frequency band, which can be obtained by integrating or averaging the signal energy in the specific frequency band, the number of each preset frequency band, , quantized values representing the signal phase difference distribution between sensor units i and j, both i and j referring to the number of each unit in the sensor array, The crystal orientation deviation feature vector generated according to this fusion method not only contains dynamic change information of the growth process, but also contains spatial pointing information of the interface deviation.
[0037] S13. Input the crystal orientation deviation feature vector into a preset mapping model to generate a crystal orientation deviation quantization index.
[0038] As shown in Figure 2 In specific embodiments of the present application, the preset mapping model is established by the following steps: in a crystal growth device, a preset angle tilt is actively applied to the seed holder by a control device, and a plurality of sets of crystal orientation deviation states covering the expected deviation range and direction are obtained.
[0039] The preset angle tilt applied to the seed holder can be exemplarily set to a deviation of 0 to 0.5 degrees in multiple azimuth angles with a step of 0.05 degrees.
[0040] Collect and process the signals during stable operation under each preset crystal orientation deviation state to generate corresponding crystal orientation deviation feature vector samples as experimental data.
[0041] Based on the crystal anisotropic growth physical equation, the theoretical feature vector under any virtual crystal orientation deviation state is quantized as simulation data.
[0042] It should be noted that the crystal anisotropic growth physical equation can be exemplarily expressed as: wherein is the crystal orientation angle is the local growth rate in the direction of the crystal orientation angle, T is the temperature, is the isotropic basic growth rate, which is positively correlated with the temperature, is a preset anisotropy coefficient reflecting the difference in growth rate of different crystal orientations, with a value range of [0, 1], g is a preset crystal symmetry factor, for example, the value corresponding to cubic crystal system is 4, and the value corresponding to hexagonal crystal system is 6, is the included angle between the actual crystal orientation and the target crystal orientation.
[0043] The physical basis of this example equation lies in the growth rate difference caused by the anisotropy of the crystal surface energy, i.e. the atomic arrangement density of the crystal in different crystal directions is different, resulting in the difference of specific surface energy and its corresponding growth kinetics coefficient, and then showing the periodic change of growth rate with the crystal direction angle, the cosine function term can reflect the effect of crystal symmetry on the modulation of growth rate.
[0044] The establishment of this example equation is based on the comprehensive consideration of crystallography, thermodynamics and kinetics, and the derivation process is as follows: a. The different atomic arrangement density and coordination number of the crystal in different crystal directions result in the change of specific surface energy with the crystal direction angle, which is called surface energy anisotropy. For a crystal with g-fold symmetry, the specific surface energy of the crystal direction can be expressed as: , where is the average value of the specific surface energy of all crystal directions, which is a preset constant related to the material itself and temperature, is the surface energy anisotropy intensity coefficient, which reflects the degree of change of surface energy with crystal direction.
[0045] b. The growth rate of the crystal depends on the kinetics of the atomic attachment process at the growth interface, and its macroscopic driving force is the Gibbs free energy difference between the melt and the solid . According to the classical reaction kinetics theory, there is an approximate exponential relationship between the growth rate and the driving force: , where is the preset Boltzmann constant.
[0046] c. For a crystal in actual growth, the local growth driving force will be affected by the interface curvature and surface energy. According to the Gibbs-Thomson effect, the local equilibrium melting point of a convex interface with anisotropic surface energy will change, resulting in the modification of the effective driving force , which can be approximately expressed as: , where K is the local curvature of the interface, is the atomic volume.
[0047] d. Substitute the above effective driving force into the growth rate kinetics formula, and assume that the anisotropy term is small, which can be simplified . Further, assuming that in the exponential term is small, it can be first-order Taylor expanded ( ) to get .
[0048] e. Substitute the specific surface energy expression of the crystal direction angle into the last formula in step d to get , where the first two terms in this formula represents a basic growth rate independent of the crystallographic orientation, which can be combined into a new constant the coefficient of the last term which can be defined as the crystallographic growth anisotropy coefficient by combining the thermodynamic and kinetic effects Therefore, the crystal anisotropic growth physical equation is finally simplified as: .
[0049] In summary, the crystal anisotropic growth physical equation adopted by the present application is not a pure empirical formula, but a simplified equation derived based on the surface energy anisotropy and the classical theory of Gibbs-Thomson effect. The core parameters all have clear physical meaning and can be calibrated through conventional experiments, for example, the growth rate of different crystal orientation seeds can be measured for fitting.
[0050] It should be further pointed out that the process of quantifying the theoretical characteristic vector in the state of virtual crystal orientation offset based on the crystal anisotropic growth physical equation is realized by multi-physical field numerical simulation, and the core is that the physical equation is embedded into a coupled simulation model including heat field and flow field calculation as a key constitutive relation describing the dynamic behavior of the crystal growth interface. The morphology, temperature distribution and stress state of the growth interface under different virtual crystal orientation offset states are simulated by calculation, and then the theoretical characteristic vector with the same structure as the aforementioned experimental signal is extracted therefrom. The specific simulation data generation steps are as follows: (1) A three-dimensional geometric model is established according to the structure of the real crystal growth equipment, and the physical fields related to heat transfer and fluid flow are defined in the model. The physical equation is set as the normal growth velocity of the interface in the crystal growth interface area, wherein the orientation angle is directly determined by the input virtual crystal orientation offset parameter, so as to parameterize the crystal orientation offset state into the simulation.
[0051] (2) For each set of preset virtual crystal orientation offset parameters, the coupled partial differential equation set is solved by using professional simulation software to obtain the stable temperature field, flow field and final growth interface morphology in the entire calculation domain.
[0052] (3) After the simulation calculation is completed, instead of directly outputting the growth rate v, the theoretical equivalent physical quantities corresponding to the thermal fluctuation signal and the acoustic emission elastic wave signal collected in the experiment are extracted from the simulation results, and the theoretical physical quantities are processed according to the same feature extraction process as the experimental data, so as to finally generate the theoretical characteristic vector consistent with the dimension and physical meaning of the experimental characteristic vector.
[0053] Specifically, for the above features: (31) Theoretical signal for extracting signal intensity fluctuation pattern: In the simulation model, it is equivalent to the experimental thermal fluctuation signal, and it is the temperature value of a series of virtual temperature measurement points near the growth interface and its fluctuation with time.
[0054] (32) Theoretical signal for extracting phase difference distribution: The simulation can calculate the local heat flux density distribution caused by the asymmetric growth interface morphology due to the crystal direction deviation or the change of the maximum principal stress direction near the interface. It can be analogous to the signal wavefront time delay received by different sensors in the experiment. By calculating the spatial gradient or the numerical difference between different positions, the theoretical phase difference distribution information is generated.
[0055] (4) Repeat steps 2 and 3 by traversing all pre-set virtual crystal direction deviation states, and a simulation data set covering the entire expected deviation range can be generated, where each theoretical feature vector corresponds to a known crystal direction deviation.
[0056] Integrate experimental data and simulation data to construct a training data set.
[0057] It needs to be supplemented that the integration process of the above experimental data and simulation data specifically refers to the allocation of the number ratio weight of experimental data and simulation data for the same crystal direction deviation state. By deeply fusing the real experimental measurement data and the simulation data based on the physical equation to construct the training data set, the synergistic technical effect beyond the training of a single data source is generated. The model trained purely relying on experimental data is limited by the number and accuracy of the calibration points, and has poor generalization ability for uncalibrated working conditions. The simulation data purely relying on the physical model may have systematic deviation from the actual situation due to model simplification or inaccurate parameters. The present application combines the two, calibrates and anchors the model using experimental data to ensure its high fidelity in real working conditions, and generates a large number of diversified simulation data using the physical model to fill the blank area of the experimental data, greatly enhancing the robustness and prediction ability of the final generated preset mapping model for unknown working conditions, achieving high precision and strong generalization ability.
[0058] The mapping relationship between the feature vector and the crystal direction deviation quantitative index is trained by a machine learning algorithm.
[0059] It needs to be supplemented that the specific form of the above crystal direction deviation quantitative index is the three-dimensional angle deviation value of the actual crystal direction based on the target crystal direction, which contains the deviation direction component and the deviation amount component. The machine learning algorithm can be exemplified as a deep neural network or a support vector regression.
[0060] Taking a deep neural network as an example, the mapping relationship is specifically represented as a nonlinear function where P is a feature vector fusing the fluctuation pattern of signal intensity and the distribution of phase difference, and O is a two-dimensional output vector containing a component of offset direction and a component of offset amount, the component of offset direction can be a azimuth angle specifically, and the component of offset amount can be a tilt angle specifically.
[0061] Suppose there is a trained small neural network model, whose structure and parameters are as follows: Input layer: 3 neurons, corresponding to an extremely simplified 3-dimensional feature vector P = [x1, x2, x3].
[0062] Hidden layer: 2 neurons, using ReLU activation function.
[0063] Output layer: 2 neurons, using linear activation function to output tilt angle and azimuth angle respectively.
[0064] Known trained model parameters: The weight matrix W1 from input layer to hidden layer is [[0.5, -0.2, 0.8], [0.1, 0.6, -0.3]], and the bias vector b1 is [0.1, -0.2].
[0065] The weight matrix W2 from hidden layer to output layer is [[1.5, -0.5], [0.8, 1.0]], and the bias vector b2 is [0.05, 0.1].
[0066] For a feature vector sample P = [0.8, 1.5, 0.4] collected and extracted in real time, the forward propagation calculation process in the model is as follows: By Calculate the input of the hidden layer, and apply the ReLU activation function because its original value is an integer, so the original , by Calculate the final mapping result, indicating that the crystal growth axis is tilted by 0.50 degrees, and the azimuth angle of the tilt is 1.176 radians.
[0067] It should be emphasized that the above example is a simplified illustration, the feature vector dimension in the actual model is higher, and the network structure is more complex, the weights and biases W1, b1, W2, b2, etc. are the optimal values obtained by a large number of experimental and simulation data training, not artificial setting. But its mathematical essence, that is, through layer-by-layer weighted summation and nonlinear transformation to map high-dimensional input to low-dimensional output, is exactly the same as this example.
[0068] The embodiment of the present application realizes real-time microscopic monitoring of the growth state by high-sensitivity non-contact collection of inherent physical signals of the crystal growth interface and real-time conversion into a crystal direction deviation feature vector, fundamentally avoids physical interference or damage to the crystal growth process that may be caused by traditional detection methods, ensures the purity and stability of the growth environment, provides high-quality and non-destructive decision basis for subsequent precise regulation and control, and thus improves the regulation and control foundation reliability.
[0069] S14. Calculate a local thermal field regulation instruction based on the crystal direction deviation quantification index, and drive a micro heat source array to apply a local thermal field gradient to a target region of the crystal growth interface to guide the crystal direction back to normal.
[0070] In specific embodiments of the present application, the local thermal field regulation instruction calculation process includes: analyzing the deviation direction component and the deviation amount component in the crystal direction deviation quantification index.
[0071] Determine the activation region of the micro heat source array according to the deviation direction component, and mark the activation region as the target region.
[0072] It should be noted that the process of determining the activation region of the micro heat source array according to the deviation direction component can be divided into the following specific steps: mapping the deviation direction component representing the azimuth angle of the crystal direction deviation from the target direction to the plane coordinate system of the micro heat source array, which is specifically an polar coordinate system with the projection of the center of the crystal growth interface on the array as the origin.
[0073] The regulation action point should be located at the position opposite to the deviation direction on the projection of the crystal growth interface onto the plane of the micro heat source array, and the reverse positioning activation direction ensures that the regulation action point and the deviation direction component form a hedge.
[0074] Project the physical boundary of the crystal growth interface to the plane of the micro heat source array to determine the array coordinate range corresponding to the interface target regulation point in the reverse direction.
[0075] According to the spatial resolution of the array, in the projection coordinate range of the reverse direction, filter out the micro heat source units that need to be activated, which can be a single unit or a local region composed of multiple adjacent units, and the specific number is determined by the deviation size, to form an activation region coordinate set.
[0076] According to the deviation amount component, analyze the temperature gradient value of the target region, convert the temperature gradient value into the configuration power intensity of each micro coil unit in the target region, and generate a local thermal field regulation instruction containing spatial position parameters and power intensity parameters.
[0077] It is to be noted that the temperature gradient value of the target region is specifically the product of the offset component and a preset control gain coefficient, wherein the preset control gain coefficient is obtained through experimental calibration, has a dimension of Kelvin per meter per radian, and is used to reflect the required thermal field gradient response intensity per unit angle offset.
[0078] The core logic of the conversion between the temperature gradient value and the configured power intensity is that, by establishing a power-thermal field mapping relationship, the power intensity required to be output by each coil unit is inversely deduced according to the spatial distribution requirement of the temperature gradient of the target region. For example, a measured database of power-temperature gradient is established through pre-experiment, an empirical formula or a lookup table is fitted, and the specific power intensity configuration is directly calculated by table lookup or formula substitution. The experimental calibration content for establishing the measured database includes activating different micro coil units alone or in combination, setting a series of power values, and measuring the temperature gradient of the corresponding region by using an infrared temperature measuring device or a thermocouple.
[0079] The conversion solving process of the temperature gradient value and the configured power intensity is as follows: A calibration experiment is performed before formal growth, each micro coil unit is activated alone, and a fixed unit reference power is applied When each unit is activated alone, a high spatial resolution infrared thermal imager or a micro thermocouple array is used to measure the temperature gradient field generated by the unit in the target region near the crystal growth interface, and the gradient value of the field at the preset observation points is recorded. The unit power influence results of all coil units are combined into an MxN dimensional unit power influence matrix A, wherein M is the number of observation points, N is the number of coil units, and the cth column of the matrix A represents the temperature gradient value vector generated by the cth coil unit at all M observation points under unit power.
[0080] In the online regulation, the calculated target temperature gradient value vector is represented as an Mx1 dimensional target gradient vector b.
[0081] The conversion problem is converted into solving a linear equation group: A*x=b, wherein x is an Nx1 dimensional vector, representing the multiple of the unit power required to be applied to each coil unit.
[0082] The equation is solved by the least square method to obtain the optimal power configuration vector of each coil unit The actual configured power intensity of each coil unit is: wherein r indicates the number of the coil unit, In this way, the conversion between the temperature gradient value and the configured power intensity is realized.
[0083] In specific embodiments of the present application, the arrangement of the micro heat source array satisfies the following constraint conditions: The distance between adjacent coil units is less than a preset proportion of the diameter of the growth interface.
[0084] The thermal response time is lower than the inverse of the crystal growth rate.
[0085] The power resolution reaches the milliwatt level of micro-area control accuracy.
[0086] It should be noted that the above requirement that the distance between adjacent coil units in the array is less than a preset proportion of the diameter of the growth interface being controlled is a limitation on the distribution density of the coil units, where the diameter of the growth interface can be measured before the experiment by optical microscopy or the like. This requirement ensures that the thermal field control has a high enough spatial resolution, so that a temperature gradient can be applied to any small area of the growth interface. The preset proportion can be exemplarily set to one-tenth.
[0087] The requirement that the thermal response time of the micro heat source array is lower than the inverse of the crystal growth rate is a limitation on the dynamic response capability of the coil units. The crystal growth rate can be measured by monitoring the change of the crystal length over time. The thermal response time refers to the time required for the heat source to reach a stable target value of the temperature field after receiving an instruction. This condition ensures that the speed of the thermal field control can keep up with or even exceed the dynamic changes of the crystal itself. If the thermal response is too slow, the control action will lag behind the actual deviation of the crystal orientation, resulting in the failure of closed-loop control.
[0088] The requirement that the power resolution of the micro heat source array needs to reach the milliwatt level is a limitation on the power control accuracy of the coil units. The power resolution refers to the smallest power adjustment step that the controller can output. This requirement ensures that the temperature gradient can be finely adjusted to respond to small crystal orientation deviations, avoiding control jitter or overshoot due to large power adjustment steps, and is the key to achieving smooth and stable control.
[0089] By optimizing and limiting the key physical parameters of the micro heat source array in space, time and energy, a decisive synergistic technical effect is produced. These three parameters are not isolated, but are interrelated and jointly determine the upper limit of the performance of the control mechanism. High spatial resolution makes precise positioning possible, high time response speed ensures the real-time nature of the control, and high power resolution guarantees the stability of the control. The combination of the three can achieve stable and accurate closed-loop control of micron-level and millisecond-level crystal orientation deviations. The overall effect is far superior to hardware that only meets a single parameter requirement, thereby ensuring the ultimate success of the entire control method.
[0090] In specific embodiments of the present application, the local thermal field gradient application process comprises: sending local thermal field regulation instructions to an independently addressable micro heat source array, positioning target coil units in the array according to spatial position parameters, and applying electrical power matching the power intensity parameters to establish a directional temperature field with controllable direction and gradient through non-contact thermal radiation.
[0091] It should be noted that the above non-contact thermal radiation is based on the fact that the target coil unit position is precisely designed within a micro space range of 0.5-20 microns from the growth interface side, which is close enough to produce effective thermal influence and avoids direct thermal contact.
[0092] According to the anisotropic growth rate characteristics of the crystal material in the directional temperature field, the atomic attachment rate in the target area is differentially distributed, forming a crystal orientation correction moment at the growth interface, which pushes the crystal tip to deflect towards the target crystal orientation to guide the correction.
[0093] It should be noted that the establishment of the directional temperature field is the key physical process for regulation. Since the driving force for crystal growth is highly sensitive to temperature, this local heating or cooling significantly changes the growth conditions in this small area. Crystal growth is anisotropic, i.e. the growth rate of different crystal orientations responds differently to temperature. This method takes advantage of this fundamental physical property. Under the action of the directional temperature field, the atomic or molecular attachment rate in different micro areas on the growth interface will be different, resulting in a change in the local growth rate. In the direction where growth needs to be suppressed, local heating slows down the growth rate, while in the direction where growth needs to be promoted, the relatively low temperature area maintains or accelerates the growth. This uneven distribution of local growth rate is equivalent to a moment that causes the orientation of the growth interface to deflect, i.e. a crystal orientation correction moment, which pushes the growing crystal tip to deflect towards the target crystal orientation to correct the crystal orientation deviation accurately and non-contactly.
[0094] To specifically illustrate the process of calculating local thermal field regulation instructions based on the crystal orientation deviation quantization index, driving the micro heat source array to apply a local thermal field gradient to the target area of the crystal growth interface to guide the crystal orientation correction, the following example shows how to convert the detected crystal orientation deviation into executable power instructions. The preset parameters in this example are assumed to be calibrated through previous experiments as follows: the preset regulation gain coefficient is 100 Kelvin per meter per radian, the unit power gradient efficiency is 0.25 (K / m) / W, which means that each 1 watt of power applied to a single coil unit can produce a temperature gradient of about 0.25 K / m at the interface, the micro heat source array is arranged in a ring array with a total of 36 units, the unit number is from 1 to 36, each unit covers a 10° azimuth angle, the maximum safe power of a single coil unit is 2.0 W, the maximum correctable deviation angle is 0.5°, and the active half-angle is 22.5°.
[0095] Suppose that the intelligent decision module outputs a crystal orientation offset quantization index in real time, the offset component of which is 0.0029 rad, and the offset direction component is 60°, and the calculated temperature gradient value of the target region is 0.29 K / m and the gradient direction is 240°, and the gradient direction 240° is located at the center of the unit 25 (covering 235°-245°), corresponding to the active range [240°-22.5°, 240°+22.5°] = [217.5°, 262.5°], which covers the coil units 22 to 26, and the ratio of the temperature gradient value of the target region to the unit power gradient efficiency is calculated as the estimated total power, and the specific value is 1.16 watts.
[0096] Then the power instruction details based on the Gaussian weight distribution can refer to the following table 1 Table 1 refers to the power instruction details based on the Gaussian weight distribution
[0097] The embodiment of the application innovatively uses a micro heat source array to apply a controllable thermal field gradient to a local area of a growth interface, uses the anisotropy principle of crystal growth to generate a weak and accurate crystal orientation correction torque, the regulation and control means has a precise action range and controllable strength, avoids global thermal stress fluctuations and new crystal defect risks caused by traditional large-scale temperature field adjustment, realizes flexible fine tuning of the crystal orientation, and significantly improves the internal quality and uniformity of crystal growth.
[0098] In specific embodiments of the application, the method further comprises a closed-loop optimization step: collecting a growth interface physical signal after applying a local thermal field gradient, extracting an updated crystal orientation offset feature vector to input a preset mapping model, and obtaining a current crystal orientation offset quantization index.
[0099] Analyzing the actual deviation value of the current crystal orientation offset quantization index and the target crystal orientation threshold value, if the actual deviation value exceeds the preset tolerance, triggering an iterative update of the local thermal field regulation and control parameters.
[0100] In specific embodiments of the application, the iterative update of the local thermal field regulation and control parameters comprises: dynamically adjusting the temperature gradient value using a proportional-integral control algorithm.
[0101] The updated crystal orientation offset feature vector and its actual offset value form an incremental training sample, and the preset mapping model weight is dynamically corrected through online back propagation.
[0102] Fusing the temperature gradient value dynamic regulation and control term and the model weight correction term, an enhanced regulation and control instruction containing feedback compensation is generated.
[0103] The embodiment of the present application deeply integrates high-precision perception, intelligent mapping decision and localized precise execution, constructs a high-speed closed-loop control, can respond to micron-level crystal direction deviation in real time and iteratively correct it, continuously and stably keep the crystal direction deviation in a very small range, effectively inhibit various macro defects caused by long-term cumulative deviation of the crystal direction, and significantly improve the yield and performance consistency of high-quality single crystal materials.
[0104] As shown in Figure 3 The second embodiment of the present application provides a micron-level single crystal growth process crystal direction real-time regulation system for executing the method of the first embodiment. The system is an integrated sensing, computing and execution physical hardware platform, and the specific components and connection relationships thereof are as follows. The system hardware includes a multi-modal sensor array, a signal conditioning and acquisition hardware, a real-time processing and decision unit, a multi-channel power drive source, a micro heat source array and a system integration bus.
[0105] The multi-modal sensor array serves as the sensing terminal of the system, is physically installed near the observation window of the crystal growth furnace and is directed to the crystal growth interface area. The array specifically includes an infrared thermal imager module for capturing specific spectrum thermal fluctuation signals of the growth interface and a laser Doppler vibrometer module or a high-frequency acoustic emission sensor array for non-contact acquisition of acoustic emission elastic wave signals or microscopic vibration signals released by the growth interface, wherein the infrared thermal imager module is selected to meet the requirements of a high frame rate of greater than 100 Hz and a high thermal sensitivity of less than 20 mK to distinguish small temperature fluctuations.
[0106] The signal conditioning and acquisition hardware is directly electrically connected with the sensor array and includes a high-precision data acquisition card for synchronously acquiring and digitizing analog signals from multiple sensors to generate the original signal data, and a preamplifier and filter for amplifying and anti-aliasing filtering weak signals from the sensors.
[0107] The real-time processing and decision unit is an industrial control computer or an embedded high-performance computing platform, which is connected with the data acquisition card through PCIe or gigabit Ethernet to receive the original signal data. A dedicated software algorithm is run in the unit to realize the functions of a feature extraction module, an intelligent decision module, a thermal field regulation module and a closed-loop control module.
[0108] The multi-channel power drive source is a precision power supply device capable of independently controlling multiple outputs. It receives control instructions from the computing unit through a communication interface, each output thereof can be independently programmed, has high power resolution and microsecond-level response speed to meet the requirements of micro-area precise regulation.
[0109] The micro heat source array is an execution terminal of the system, is physically installed in a growth furnace, surrounds a crystal growth interface side, and is connected with each output of the power driving source through high-temperature-resistant wires one by one, and the array is formed by a plurality of micro ceramic heating coils or laser diode units arranged in a two-dimensional grid.
[0110] The system communication bus provides power supply and data connection for all the components, including cables, connectors and communication protocols, to ensure that instructions and data are transmitted between hardware components at high speed and reliably.
[0111] The connection relationship of the system is as follows: the multi-modal sensor array senses the thermal wave and acoustic emission signals radiated by the crystal growth interface and converts them into electrical signals, the electrical signals are converted into digital signals by the data acquisition card after being processed by the signal conditioning and acquisition hardware, and the digital signals are transmitted to the real-time processing and decision calculation unit through the PCIe bus, the software in the calculation unit sequentially performs feature extraction, intelligent decision and control instruction calculation, and finally generates a set of digital control instructions, which are sent to the multi-channel power driving source through the EtherCAT bus, the driving source converts them into accurate analog power output, each output of the power driving source drives the corresponding coil unit in the micro heat source array, and a local and controllable thermal field gradient is established in the target area of the growth interface, the sensor array continuously monitors the growth interface after regulation and control, and new signal data are fed back to the calculation unit through the above path to form a closed loop control until the crystal orientation deviation converges.
[0112] The embodiment of the application discloses a complete and realizable crystal orientation real-time regulation system through the specific hardware selection, clear physical connection relationship and clear data flow design.
[0113] The system execution content includes: The signal acquisition module acquires the inherent physical signals of the crystal growth interface region and generates original signal data.
[0114] The feature extraction module extracts feature parameters associated with the crystal orientation deviation from the original signal data and generates a crystal orientation deviation feature vector.
[0115] The intelligent decision module inputs the crystal orientation deviation feature vector into a preset mapping model and generates a crystal orientation deviation quantitative index.
[0116] The thermal field regulation module calculates a local thermal field regulation instruction based on the crystal orientation deviation quantitative index, drives the micro heat source array to apply a local thermal field gradient to the target area of the crystal growth interface to guide the crystal orientation back to normal.
[0117] The closed-loop control module performs local thermal field control parameter iterative updating based on real-time feedback data in the regulation process until the crystal orientation deviation converges to within a preset tolerance.
[0118] The microscale single crystal growth process crystal orientation real-time regulation system provided in the embodiments of the present application has the same implementation principle and generated technical effects as the foregoing method embodiments, and for brevity of description, the system embodiments are not mentioned in the foregoing method embodiments, and the corresponding content in the foregoing method embodiments can be referred to.
[0119] It should be noted that: the above formulas can convert physical quantities of different properties into unitless standard values or same-dimension superimposable parameters through the principle of dimensional consistency and mathematical standardization means, so as to eliminate the interference of different dimensions on the operation logic, so that the formula has mathematical operation rationality and objective law adaptability while retaining the original data distribution characteristics. The above is only an exemplary embodiment of the present application, and cannot limit the scope of the present application.
[0120] The above content is only an example and description of the concept of the present application, and those skilled in the art can make various modifications or supplements or use similar ways to replace the described specific embodiments, as long as they do not deviate from the concept of the present application or exceed the scope defined by the present application, which shall belong to the protection scope of the present application.
Claims
1. A method for real-time control of crystal orientation in a process for growing a single crystal on a micron scale, characterized in that, The method comprises the following steps: Collecting intrinsic physical signals of the crystal growth interface region to generate original signal data; Extracting feature parameters associated with crystal orientation deviation from the original signal data to generate a crystal orientation deviation feature vector; Inputting the crystal orientation deviation feature vector into a preset mapping model to generate a crystal orientation deviation quantification index; Based on the crystal orientation deviation quantification index, a local thermal field regulation instruction is calculated to drive a micro heat source array to apply a local thermal field gradient to the target region of the crystal growth interface to guide the crystal orientation back to normal.
2. The method of claim 1, wherein the method comprises: The intrinsic physical signal collection process comprises: Capturing specific frequency spectrum thermal fluctuation signals of the growth interface radiation through a non-contact sensor array; Synchronously collecting acoustic emission elastic wave signals released by the growth interface; Fusing the thermal fluctuation signals and the acoustic emission elastic wave signals to generate multi-modal original signal data.
3. The method of claim 2, wherein the method comprises: The feature parameters associated with the crystal orientation deviation extracted from the original signal data comprise: Performing time-frequency joint analysis on the original signal data to extract signal intensity fluctuation patterns in a preset frequency band; Calculating the phase difference distribution of each unit signal in the non-contact sensor array; Fusing the signal intensity fluctuation patterns and the phase difference distribution to generate a crystal orientation deviation feature vector containing spatial dimension information.
4. The method for real-time control of crystal orientation in the micron-scale single crystal growth process according to claim 1, characterized in that, The preset mapping model is established by the following steps: In the crystal growth device, a preset angle tilt is actively applied to the seed crystal holder by a control device to obtain multiple sets of crystal orientation deviation states covering the expected deviation range and direction; Signals during stable operation under each preset crystal orientation deviation state are collected and processed to generate corresponding crystal orientation deviation feature vector samples as experimental data; Based on the crystal anisotropic growth physical equation, the theoretical feature vector under any virtual crystal orientation deviation state is quantified as simulation data; The experimental data and the simulation data are integrated to construct a training data set; The mapping relationship between the feature vector and the crystal orientation deviation quantification index is obtained through machine learning algorithm training.
5. The method for real-time control of crystal orientation in the micron-scale single crystal growth process according to claim 1, characterized in that, The local thermal field regulation instruction calculation process comprises: Analyzing the deviation direction component and the deviation amount component in the crystal orientation deviation quantification index; Determining the activation area of the micro heat source array according to the deviation direction component, and marking the activation area as a target area; Analyzing the temperature gradient value of the target area according to the deviation amount component, converting the temperature gradient value into the configuration power intensity of each micro coil unit in the target area, and generating a local thermal field regulation instruction containing spatial position parameters and power intensity parameters.
6. The method of claim 5, wherein the method further comprises: The arrangement of the micro heat source array satisfies the following constraint conditions: The spacing between adjacent coil units is less than a preset proportion of the diameter of the growth interface; The thermal response time is lower than the inverse of the crystal growth rate; The power resolution reaches the micro-area control accuracy.
7. The method of claim 5, wherein the method further comprises: determining a crystal orientation of the single crystal; and adjusting the crystal orientation of the single crystal in real time. The local thermal field gradient application process comprises: Sending the local thermal field regulation instruction to the independently addressable micro heat source array, positioning the target coil units in the array according to the spatial position parameters, and applying the electric power matched with the power intensity parameters to establish a directional temperature field with controllable direction and gradient through non-contact thermal radiation; According to the anisotropic growth rate characteristics of the crystal material in the directional temperature field, the atomic attachment rate of the target region is differentially distributed, forming a crystal orientation correction moment at the growth interface, which pushes the crystal tip to deflect towards the target crystal orientation to guide the correction.
8. The method for real-time control of crystal orientation in the micron-scale single crystal growth process according to claim 1, characterized in that, It also includes a closed-loop optimization step: Collect the physical signals of the growth interface after applying the local thermal field gradient, extract the updated crystal orientation deviation feature vector, input the preset mapping model, and obtain the current crystal orientation deviation quantization index; Analyze the actual deviation value of the current crystal orientation deviation quantization index and the target crystal orientation threshold value, if the actual deviation value exceeds the preset tolerance, trigger the local thermal field control parameter iterative update.
9. The method of claim 8, wherein the method further comprises: determining a crystal orientation of the single crystal growth process; and adjusting the crystal orientation of the single crystal growth process in real time. The local thermal field control parameter iterative update includes: Using a proportional-integral control algorithm to dynamically adjust the temperature gradient value; The updated crystal orientation deviation feature vector and its actual deviation value constitute an incremental training sample, and the preset mapping model weight is dynamically corrected through online back propagation; Fusion temperature gradient value dynamic control item and model weight correction item, generate enhanced control instruction containing feedback compensation.
10. A system for performing the method of real-time control of crystallographic orientation during microscale single crystal growth processes according to any one of claims 1-9, characterized in that, It includes: Signal acquisition module, collect the inherent physical signals of the crystal growth interface region, generate original signal data; Feature extraction module, extract feature parameters associated with crystal orientation deviation from the original signal data, generate crystal orientation deviation feature vector; Intelligent decision-making module, input the crystal orientation deviation feature vector into the preset mapping model, generate the crystal orientation deviation quantization index; Thermal field control module, based on the crystal orientation deviation quantization index, calculate the local thermal field control instruction, drive the micro heat source array to apply local thermal field gradient to the target region of the crystal growth interface to guide the crystal orientation correction; Closed-loop control module, based on real-time feedback data during the control process, perform local thermal field control parameter iterative update until the crystal orientation deviation converges within the preset tolerance.
Citation Information
Patent Citations
Rotation centering method and system based on online offset monitoring
CN118610143A
Indium phosphide growth monitoring system and method based on multi-parameter learning
CN119066519A
Monocrystalline silicon thermal field multi-section dynamic cooperative heating control method and system
CN119932698A
Semiconductor defect detection and process optimization method based on deep learning
CN120107239A
Automobile magnesium alloy pre-twin crystal deformation data analysis system
CN120257529A
Cited By
Multi-mode fusion control system of semiconductor single crystal growth equipment
CN122279730A