Sub-nano patterned substrate assisted MPCVD diamond heteroepitaxy system and sub-nano patterned substrate assisted MPCVD diamond heteroepitaxy method

Abstract

The invention discloses a sub-nano patterned substrate assisted MPCVD diamond heteroepitaxy system and a sub-nano patterned substrate assisted MPCVD diamond heteroepitaxy method. The method comprises the following steps: acquiring microwave power and spectral data in an MPCVD reaction cavity and surface temperature data of a heterogeneous substrate of which the surface is provided with a sub-nano pattern in advance; determining a current reaction stage according to the spectral data, calculating a current gas proportion, and generating a corresponding constraint condition; calculating through deep learning to obtain an expected value; judging whether the microwave power and the gas proportion meet expected values or not, if not, adjusting to the expected values, calculating the average temperature of the surface temperature data, and judging whether the average temperature meets the expected values or not; if yes, calculating the temperature difference between the surface temperature of the sub-nanometer pattern of the heterogeneous substrate and the temperature of the peripheral area to obtain a temperature difference, and judging whether the temperature difference is within a preset temperature interval or not, otherwise, changing the airflow distribution; and outputting an experimental analysis report. By monitoring spectral data and temperature distribution in real time and utilizing the sub-nanometer pattern, the defect problem caused by lattice mismatch is solved.

Description

Technical Field

[0001] This invention relates to the field of MPCVD diamond heteroepitaxial growth technology, specifically to a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth system and method. Background Technology

[0002] Microwave plasma chemical vapor deposition (MPCVD) is a technique for high-quality thin film epitaxial growth on material surfaces. In actual growth processes, the use of heterogeneous substrates has gradually become an important way to improve epitaxial quality, control crystal orientation, and reduce defect density.

[0003] In existing MPCVD systems, studies have attempted to use optical emission spectroscopy (OES) for process monitoring, but this is usually limited to qualitative assessment. During diamond heteroepitaxial growth, lattice mismatch between the heterosubstrate and the diamond crystal can easily lead to the formation of dislocations and twins, thus affecting the quality of the diamond single crystal. However, current methods cannot precisely control the stress field distribution in the initial nucleation region at the atomic or carbon cluster scale, which poses a challenge to achieving high-quality diamond single-crystal heteroepitaxial growth. Furthermore, the lack of real-time monitoring and precise control over the entire nucleation and epitaxial growth process cannot guarantee the uniformity and quality stability of the epitaxial growth. Summary of the Invention

[0004] To address the lack of real-time monitoring and precise control over the entire nucleation and epitaxial growth process in existing methods, this invention provides a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth system and method.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, this application discloses a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth method comprising the following steps:

[0007] Acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber;

[0008] The presence and concentration changes of active reactants in the spectral data are extracted to determine the current reaction stage and calculate the current gas ratio. In addition, the historical database is called to generate the constraints corresponding to the current reaction stage.

[0009] The expected values ​​are obtained by deep learning calculation using spectral data and constraints; the expected values ​​include microwave power, average surface temperature of heterogeneous substrate, and gas ratio.

[0010] Determine whether the microwave power and gas ratio meet the expected values; otherwise, adjust the microwave power and gas ratio to the expected values. Then calculate the average temperature of the surface temperature data and determine whether the average temperature meets the expected values.

[0011] Otherwise, the surface temperature of the heterogeneous substrate is finely adjusted using a substrate heating device until the desired value is met;

[0012] If so, the temperature difference is calculated by the difference between the surface temperature of the sub-nano patterned area of ​​the heterogeneous substrate and the temperature of the surrounding area of ​​the sub-nano pattern. It is then determined whether the temperature difference is within the preset temperature range. Otherwise, the airflow distribution is changed until the temperature difference is within the temperature range.

[0013] Analyze the data before and after the adjustment and output an experimental analysis report.

[0014] Secondly, this application discloses a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial system, which includes a data acquisition module, a data processing module, a data prediction module, a data judgment module, and a data output module.

[0015] The data acquisition module is used to acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber.

[0016] The data processing module is used to extract the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculate the current gas ratio. In addition, it calls the historical database to generate the constraints corresponding to the current reaction stage.

[0017] The data prediction module is used to calculate expected values ​​from spectral data and constraints using deep learning; the expected values ​​include microwave power, average surface temperature of the heterogeneous substrate, and gas ratio.

[0018] The data judgment module is used to determine whether the microwave power and gas ratio meet the expected values. Otherwise, it adjusts the microwave power and gas ratio to the expected values, then calculates the average temperature of the surface temperature data, and determines whether the average temperature meets the expected values.

[0019] Otherwise, the surface temperature of the heterogeneous substrate is finely adjusted using a substrate heating device until the desired value is met;

[0020] If so, the temperature difference is calculated by taking the difference between the surface temperature of the sub-nano patterned area on the heterogeneous substrate and the temperature of the surrounding area of ​​the sub-nano pattern. It is then determined whether the temperature difference is within the preset temperature range. Otherwise, the airflow distribution is changed until the temperature difference is within the temperature range.

[0021] The data output module is used to analyze the data before and after adjustment and output an experimental analysis report.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] This application effectively solves the defect problem caused by lattice mismatch by real-time monitoring of spectral data and temperature distribution, combined with deep learning to dynamically optimize microwave power, gas ratio and temperature parameters, and by using sub-nanometer patterns to control stress field distribution. It has the advantages of improving epitaxial quality, reducing dislocation and twin defects, achieving precise control of atomic-scale stress field and dynamic optimization of growth parameters. Attached Figure Description

[0024] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals refer to the same parts. Wherein:

[0025] Figure 1 This is a step diagram illustrating a sub-nano patterned substrate-assisted MPCVD diamond heteroepitaxial growth method described in this invention;

[0026] Figure 2 Based on Figure 1 One of the flowcharts for a sub-nano patterned substrate-assisted MPCVD diamond heteroepitaxialization method;

[0027] Figure 3 Based on Figure 3 Flowchart 2 of a sub-nano patterned substrate-assisted MPCVD diamond heteroepitaxialization method;

[0028] Figure 4 Based on Figure 1 The flowchart for generating constraints;

[0029] Figure 5 Based on Figure 1 A flowchart for verifying the current reaction stage;

[0030] Figure 6 Based on Figure 5 A flowchart illustrating purification based on impurity concentration;

[0031] Figure 7 Based on Figure 6A flowchart illustrating the adjustment of the constraints based on the changing trend of impurity concentration after the purification operation;

[0032] Figure 8 Based on Figure 7 A flowchart for determining whether the purification operation has failed based on the trend of impurity concentration changes after purification;

[0033] Figure 9 Based on Figure 1 Flowchart of the drive and guide device;

[0034] Figure 10 Based on Figure 1 Scene illustration;

[0035] Figure 11 Block diagram of a sub-nano patterned substrate-assisted MPCVD diamond heteroepitaxial system. Detailed Implementation

[0036] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0037] Application Overview

[0038] In existing technologies, microwave plasma chemical vapor deposition (PCCVD) is relatively widely used for the epitaxial growth of diamond thin films. However, lattice mismatch between the heterostructure and the diamond crystal can easily lead to dislocations and twin defects. Existing methods rely on plasma emission spectroscopy for qualitative monitoring, which cannot control the stress field distribution at the atomic or carbon cluster scale. Furthermore, they lack the ability to monitor and dynamically adjust the entire nucleation and epitaxial growth process in real time, resulting in insufficient epitaxial uniformity and quality stability. For example, when epitaxially growing diamond on a silicon-based heterostructure, the inability to precisely control the stress distribution in the initial nucleation region easily leads to the formation of polycrystalline structures on the surface.

[0039] To address these challenges, research has revealed that sub-nanometer surface patterning can guide the ordered arrangement of carbon atoms, but this requires coordinated control with dynamic process parameters. By analyzing the concentration changes of active reactants in spectral data, reaction stages can be identified and constraints generated. Optimal process parameters can then be predicted using a deep learning model. Simultaneously, local stress field control in the sub-nanometer patterned region is achieved through temperature difference monitoring and airflow adjustment. This approach combines substrate microstructure design with real-time parameter feedback to form a closed-loop control mechanism.

[0040] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Exemplary methods

[0042] like Figure 1-3 As shown, this embodiment introduces a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth method, including the following steps:

[0043] 100. Acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber;

[0044] 200. Extract the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculate the current gas ratio. In addition, call the historical database to generate the constraints corresponding to the current reaction stage.

[0045] 300. The expected values ​​are obtained by deep learning calculation using the spectral data and constraints; the expected values ​​include microwave power expected value, average surface temperature of heterogeneous substrate expected value, and gas ratio expected value;

[0046] 400. Determine whether the microwave power and gas ratio meet the expected value; otherwise, adjust the microwave power and gas ratio to the expected value, then calculate the average temperature of the surface temperature data, and determine whether the average temperature meets the expected value.

[0047] 500. Otherwise, fine-tune the surface temperature of the heterogeneous substrate using a substrate heating device until the desired value is met;

[0048] 600. If yes, calculate the temperature difference by comparing the surface temperature of the sub-nano patterned area on the heterogeneous substrate with the temperature of the surrounding area of ​​the sub-nano patterned area, and determine whether the temperature difference is within a preset temperature range. Otherwise, change the airflow distribution until the temperature difference is within the temperature range.

[0049] 700. Analyze the data before and after adjustment and output an experimental analysis report.

[0050] This embodiment guides the orderly deposition of carbon atoms through a sub-nanometer patterned substrate, combined with dynamic parameter adjustment driven by spectral data to achieve atomic-scale stress field control. Local temperature difference monitoring and airflow distribution adjustment ensure directional energy supply to the microstructure region. The reaction progress is judged in real time by the concentration ratio of active groups, and multi-parameter synergistic optimization is achieved based on a deep learning model, forming an adaptive process control system. This effectively suppresses dislocations and twin defects during heteroepitaxial growth. Through the synergistic effect of the sub-nanometer patterning template and dynamic parameter control, directional and orderly growth of diamond single crystals is achieved. The combination of real-time spectral analysis and the deep learning model allows process parameters to adapt to the needs of different reaction stages, improving the uniformity of the epitaxial layer thickness. The temperature difference feedback mechanism ensures controllable local stress distribution in the microstructure region, avoiding lattice distortion caused by thermal gradients.

[0051] To facilitate a further understanding of the technical solution of this application, the specific steps for obtaining microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber are described in detail below.

[0052] Microwave power data is provided by a microwave generator. Spectral data is acquired in real-time from CH4 in the plasma using optical emission spectroscopy (OES). x Signals for H atoms and major impurities. Surface temperature data of the heterogeneous substrate is provided by a substrate heating device located inside the MPCVD reaction chamber.

[0053] The specific method for setting sub-nanometer patterns on heterogeneous substrates involves etching or imprinting periodic steps, holes, or lines with a size of less than 1 nm on the surface of heterogeneous substrates such as Si, Mo, or Ir.

[0054] In some specific implementations, for single-crystal silicon substrates, electron beam lithography combined with reactive ion etching can be used to form a periodic step array with a depth of 0.5 nm, and the step spacing can be set to 0.3 nm to match the diamond lattice constant. For copper substrates, nanoimprint lithography can be used to prepare a hexagonal close-packed hole array with a diameter of 0.8 nm, and the hole center-to-center distance can be controlled within the range of 1.2 nm. For gallium nitride substrates, atomic layer etching can be used to process a parallel line pattern structure with a width of 0.6 nm, and the line pattern direction maintains a specific angle with the substrate crystal orientation.

[0055] Atomic-level manipulation of the stress field during heteroepitaxial diamond growth enables diamond nuclei to align in an ordered manner along predetermined crystal orientations under the guidance of sub-nanostructures. This approach reduces stress concentration caused by lattice mismatch at the interface through geometric constraints, suppresses the formation of defects such as dislocations and twins, and improves the crystallinity and interfacial bonding strength of epitaxial diamond single crystals. Simultaneously, the periodic arrangement of the sub-nanostructures ensures spatial consistency during epitaxial growth, resulting in uniform crystal orientation and thickness distribution of the diamond film across different regions.

[0056] The details of step 100 have been explained above. Below, we will combine... Figure 4 Step 200 involves extracting the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculating the current gas ratio. In addition, the historical database is called to generate the constraints corresponding to the current reaction stage, which are then described in detail.

[0057] The specific steps are as follows:

[0058] 201. Extract the characteristic peak intensities of the active reactants from the spectral data, perform normalization processing, and calculate the current gas ratio from the ratio of the processed characteristic peak intensities.

[0059] That is, from spectral data Characteristic spectral peaks of key active reactants extracted:

[0060] *Peak intensity (e.g., 431nm): ;

[0061] Peak intensity (e.g., 516 nm): ;

[0062] Peak intensity (e.g., 656nm): .

[0063] in, Indicates wavelength. Indicates time, This indicates the peak extraction window.

[0064] use As a reference spectral line, other peak intensities are normalized: .

[0065] Define the functional relationship between the normalized peak ratio and the gas ratio: .

[0066] 202. The current reaction stage is obtained by referring to a preset mapping table characterizing the relationship between gas ratio and stage.

[0067] That is, we define a mapping table representing the gas ratio and the reaction stage. Assuming there are four stages, it is represented by a piecewise function:

[0068]

[0069] in, , , The preset threshold is set based on experience or training data.

[0070] 203. Retrieve the temperature range, microwave power range, and gas ratio range of the current reaction stage from the historical database and use them as constraints.

[0071] This involves setting the range of process parameters corresponding to each reaction stage in the historical database:

[0072]

[0073] Combine them to form the constraint vector for the current reaction stage:

[0074] .

[0075] This embodiment extracts and normalizes the characteristic peak intensities of the active reactants, eliminating the influence of detection equipment differences and signal drift on concentration calculations and ensuring the accuracy of gas ratio calculations. The normalized characteristic peak intensity ratios are matched with a preset mapping table, mapping continuously changing concentration data to discrete reaction stages, overcoming the lag and ambiguity of traditional qualitative judgments. By accessing the temperature, microwave power, and gas ratio ranges for the corresponding stages in the historical database, multi-dimensional dynamic constraints are formed, providing boundary limits for subsequent parameter adjustments.

[0076] Therefore, this embodiment achieves accurate calculation of gas ratios by extracting and normalizing characteristic peaks from spectral data; establishes a correspondence between gas ratios and reaction stages through a preset quantization mapping table, enabling real-time and accurate judgment of process stages; and dynamically generates multi-dimensional constraint conditions from a historical database to ensure that process parameters at different stages are always within the optimal range. Based on this, this embodiment effectively solves the problem of insufficient parameter control precision in traditional methods by accurately identifying reaction stages in real time and dynamically generating multi-dimensional constraint conditions. Normalization and characteristic peak analysis of spectral data enhance the anti-interference capability of gas ratio calculations; the quantization mapping table enables rapid matching of reaction stages, shortening stage judgment time; and the dynamic constraint condition generation from a historical database ensures that process parameter adjustments are always within the optimal range for each stage, thereby improving the uniformity and quality stability of epitaxial growth.

[0077] The details of step 200 have been explained above. Below, we will combine them with… Figure 5 The document provides specific instructions on verification after the current reaction stage has been determined.

[0078] The specific steps are as follows:

[0079] 301. Obtain the reaction phases of the two previous cycles in the current cycle;

[0080] 302. Determine whether the reaction stage of the current cycle is consistent with the reaction stage of the three consecutive cycles consisting of the previous two cycles. If so, the current reaction stage is accurate.

[0081] 303. Otherwise, reacquire spectral data and determine the reaction stage until the reaction stages are consistent for three consecutive cycles or the number of iterations reaches the preset value, at which point an abnormal warning command will be issued.

[0082] The current period refers to the time window for the current reaction phase judgment, which can be divided using a fixed duration or a fixed data collection volume to establish phase continuity verification in a time series. The two preceding periods refer to the two consecutive judgment periods before the current period; their data is stored in a historical database, and phase continuity verification is achieved by calling adjacent time-series data. The three consecutive periods refer to a continuous time period including the current period and the two preceding periods; phase consistency judgment across these three periods can eliminate the impact of single data collection anomalies. The preset value refers to a pre-set maximum iteration threshold, for example, set to 3-5 times, to prevent system deadlocks caused by continuous anomalies.

[0083] Specifically, during the reaction phase verification process, the phase identifiers for three consecutive judgment periods are compared by calling time-series data stored in the historical database. When the three phase identifiers are found to be consecutively consistent, the current phase is determined to be in a stable state, allowing the subsequent deep learning computation process to proceed. If inconsistent phase identifiers are found, a spectral data resampling mechanism is triggered to eliminate instantaneous interference factors by updating the data input. This process is repeated until consecutively consistent phase judgment results are obtained, or the preset maximum number of iterations is reached. When the number of iterations exceeds the preset value, the system automatically generates an anomaly warning command, prompting the operator to perform manual intervention and inspection.

[0084] By introducing a multi-cycle data verification mechanism and combining it with the temporal correlation verification of historical data, the system effectively distinguishes between real-time transitions and instantaneous data anomalies. An iterative re-sampling mechanism enables self-correction, while setting an upper limit on the number of iterations avoids wasting system resources. The introduction of an anomaly alert mechanism forms a human-machine collaborative fault handling mode, reducing the risk of loss of control while ensuring automated operation efficiency.

[0085] The above describes the specific steps for verifying the current reaction stage. The following section combines... Figure 6The following is a detailed explanation of the purification process performed after verifying the current reaction stage, based on changes in the concentration of gaseous impurities within the MPCVD reaction chamber. The specific steps are as follows:

[0086] 401. Obtain the impurity concentration in the reaction atmosphere within the MPCVD reaction chamber;

[0087] 402. Determine whether the impurity concentration exceeds the impurity threshold corresponding to the current reaction stage;

[0088] 403. If so, the impurity concentration difference is calculated by the difference between the impurity concentration and the impurity threshold, and the purification level is determined by referring to the preset purification reference table. Then, the corresponding purification operation is performed according to the purification level.

[0089] The purification reference table represents the mapping relationship between the impurity concentration difference and the purification intensity, and different purification intensities have their corresponding purification levels.

[0090] Impurity concentrations can be extracted in real time from a gas phase analysis system (such as a mass spectrometer, laser Raman spectroscopy, or gas sensors), preferably a high-sensitivity quadrupole mass spectrometer (QMS): the QMS is integrated into the reaction chamber system for real-time monitoring of trace impurities in the reaction gas. Common impurities include N2, O2, H2O, Ar, CO, and metal vapors. Purification is performed by a dynamic gas purification device, such as a programmable multi-stage gas purifier installed in the gas input line, including molecular sieves, activated carbon filters, and catalyst beds. The purification intensity and method can be dynamically adjusted based on real-time feedback data from the QMS.

[0091] Define the target impurity concentration as Based on the current reaction stage, obtain the allowable impurity concentration threshold from the preset parameters. ,like Then it proceeds to the purification process.

[0092] Calculate the difference in impurity concentration: .

[0093] Find the purification level corresponding to the difference in impurity concentration in the "Purification Comparison Table":

[0094]

[0095] The specific operations performed by the dynamic gas purification device based on the impurity concentration difference, as illustrated in the examples above, can be found in the table below:

[0096] Table 1: Impurity Concentration Difference and Dynamic Purification Operation Control Table

[0097] Impurity concentration difference (ppm) Purification level Enable unit Control Strategy Description 0< ≤1 <![CDATA[L0 (mild)]]> Molecular sieve <![CDATA[Low-level moisture removal / trace O2]]> 1< ≤5 <![CDATA[L1 (Moderate)]]> Molecular sieve + activated carbon <![CDATA[Remove most of the organic matter and residual N2 and O2]]> 5< ≤10 <![CDATA[L2 (Enhanced)]]> Molecular sieve + activated carbon + catalyst bed <![CDATA[For reducing oxidizing impurities (such as H2O, O2)]]> >10 <![CDATA[L3 (highest strength)]]> All + Recycle Filtering The highest purification level is suitable for situations involving sudden or severe contamination.

[0098] Specifically, after the reaction stage verification is completed, impurity concentration data within the reaction chamber is collected in real time via online monitoring. The measured concentration value is compared with the preset impurity threshold for the current reaction stage. When an excess concentration is detected, the absolute difference between the actual concentration and the threshold is calculated. Based on this difference, the corresponding purification level is matched in a purification reference table. For example, when the difference is in the range of 5-10 ppm, a secondary purification procedure is triggered. Subsequently, the purification equipment combination corresponding to this level is activated, such as simultaneously activating the molecular sieve, activated carbon, catalyst bed adsorption device, and auxiliary vacuum pump. After continuous operation for a predetermined time, the impurity concentration is re-detected. Through this dynamic feedback mechanism, energy waste caused by over-purification can be avoided when the impurity concentration slightly exceeds the limit, while strong measures can be taken to ensure the quality of the epitaxial layer in the event of severe contamination.

[0099] It should be emphasized that dynamic gas purification devices are not limited to the structural composition described above. Dynamic gas purification devices with other structural compositions can also be used, as long as they can achieve multi-stage gas purification functions controlled by programming.

[0100] By establishing a staged threshold system and a gradient purification mechanism, precise impurity control and efficient resource utilization were achieved. Differential control can be implemented based on the impurity tolerance characteristics at different epitaxial growth stages, effectively reducing purification energy consumption while ensuring the quality of diamond crystal growth. Through dynamic matching of purification level and concentration difference, crystal defects caused by insufficient purification or process fluctuations caused by over-purification, as in traditional methods, were avoided, thus maintaining the stability of the heteroepitaxial process.

[0101] The above describes the specific procedures for performing corresponding purification steps based on changes in gaseous impurity concentration after verifying the current reaction stage. The following section will combine... Figure 7 The following is a detailed explanation of whether to include the impurity threshold in the constraints used to calculate the expected value after purification, and the specific steps are as follows:

[0102] 501. Obtain impurity concentration data for N consecutive periods after purification, and process the data to obtain the purified impurity concentration; where N > 3 and is an integer;

[0103] 502. Determine whether the concentration of the purified impurity exceeds the impurity threshold corresponding to the current reaction stage;

[0104] 503. If so, the impurity threshold corresponding to the current reaction stage is added to the constraint conditions, and then the expected value is calculated by deep learning based on the spectral data and the constraint conditions.

[0105] For processing impurity concentration data from N periods, the average concentration after purification can be obtained by averaging, or by averaging across segments to calculate the average concentration after purification from these N data points. .

[0106] Let the impurity threshold corresponding to the current reaction stage s be: Determine if the conditions are met: .

[0107] If it does not meet the requirements (i.e., exceeds the limit), then Add constraint set This allows them to be used as conditional constraints in subsequent deep learning processes to fine-tune various parameters.

[0108] Therefore, after the purification operation, when the impurity concentration is detected to continuously exceed the threshold of the current reaction stage, the impurity threshold will be automatically integrated into the constraint set. By inputting the impurity concentration data and spectral data into the trained deep learning model, an optimized parameter combination including microwave power, temperature, and gas ratio is generated. During this process, the dynamic updating of the constraints allows the model to actively avoid parameter combinations in regions where impurities exceed the limit. For example, when the nitrogen impurity concentration is detected to be continuously higher than 500 ppm during the methane decomposition stage, the gas ratio calculation logic will be adjusted to reduce impurity interference. By continuously monitoring data for N cycles, occasional fluctuations and persistent exceedances can be effectively distinguished, avoiding parameter adjustments caused by misjudgments.

[0109] This embodiment addresses the parameter matching failure issue caused by impurity concentration fluctuations after purification. By dynamically incorporating impurity thresholds into the constraints, the deep learning model generates process parameters that adapt to the actual impurity state, effectively suppressing the negative impact of excessive impurity concentration on the epitaxial crystal quality. This solution achieves real-time adaptation of process parameters to impurity states, ensuring the quality stability of the diamond heteroepitaxial process.

[0110] The above describes the specific steps for adding the impurity threshold to the constraints after purification. To further determine whether the excessive impurity concentration is related to the purification failure, the following section combines... Figure 8 This document details the process of determining whether purification has failed when the concentration of purified impurities exceeds the impurity threshold corresponding to the current reaction stage, based on the trend of impurity concentration changes. The specific steps are as follows:

[0111] 601. Obtain the trend of impurity concentration change after purification based on the impurity concentration data of N consecutive periods after purification.

[0112] 602. Determine whether the trend of change in the concentration of impurities after purification reaches the preset trend of decrease in the concentration of impurities after purification;

[0113] 603. Otherwise, check whether the purification point equipment is faulty. If it is determined that there is no fault, add the impurity threshold corresponding to the current reaction stage to the constraints.

[0114] The trend of impurity concentration change after purification can be obtained by fitting, that is, by fitting the impurity concentration data sequence of N consecutive periods after purification to obtain the trend of impurity concentration change after purification.

[0115] Define the impurity concentration detected in the i-th cycle: ;

[0116] Calculating the trend slope using the least squares method : .

[0117] Assume the expected downward trend slope is negative. ,Right now: (Meets the downward trend);

[0118] Otherwise, the following judgment is made: (The decline was not ideal).

[0119] If the downward trend does not meet the target and the equipment fault detection shows no faults, then... Add constraint set If the equipment malfunctions, a fault command will be issued, and the operation of adding the impurity threshold to the constraint conditions will not be performed immediately, because there may be large errors in this experiment. The relevant experiment can be repeated after the equipment is repaired.

[0120] Specifically, equipment malfunction checks at purification points can be performed using pressure sensors to monitor gas pipeline differential pressure or current sensors to detect the operating status of solenoid valves. This helps eliminate abnormal impurity concentrations caused by equipment malfunctions. For example, if an abnormal opening of the solenoid valve or filter blockage is detected in the gas purification unit, it is determined to be an equipment malfunction requiring maintenance. If the equipment is operating normally, it is confirmed that the excessive impurity concentration is due to process conditions, and the current impurity threshold is then included in the constraints. This process, through a dual verification mechanism of trend analysis and equipment status, effectively distinguishes between process parameter deviations and equipment malfunctions, avoiding incorrect adjustments to process parameters due to unidentified equipment malfunctions.

[0121] This embodiment effectively avoids the problem of impurity concentration control failure caused by undetected equipment malfunctions, ensuring that constraint adjustments only apply to actual process environment changes. Through a dual verification mechanism that dynamically monitors purification effect trends and equipment status, the reliability of impurity control is improved, thereby guaranteeing the quality stability of crystal growth during diamond heteroepitaxial growth.

[0122] The above details the steps of adding the impurity threshold corresponding to the current reaction stage before the constraints, and determining whether the purification operation has failed based on the trend of impurity concentration changes after purification. The following section details the process of using deep learning to calculate the expected value from the spectral data and constraints at step 300. The specific steps are as follows:

[0123] 310. Construct a mapping model and train it based on historical data;

[0124] 320. Encode the sub-nanometer pattern of the heterogeneous substrate into substrate structure parameters, and obtain preliminary expected values ​​by combining the spectral data and constraints through a trained mapping model;

[0125] 330. By comparing the initial expected value with the constraints, the expected value is obtained by trimming the initial expected value to the range of constraints.

[0126] The detailed calculation process is as follows:

[0127] The sub-nanometer patterns of the heterogeneous substrate are feature-encoded to construct a substrate structure parameter vector:

[0128] For example, dimensions, cycles, and material composition can be constructed into numerical codes.

[0129] A vector is composed of multiple feature peak intensities (normalized) associated with the active species:

[0130]

[0131] in, It is the first Each characteristic peak.

[0132] Constructing deep learning mapping functions:

[0133] in, This represents the model parameters, fitted using historical samples. This represents the initial predicted triplet, and the loss function is:

[0134]

[0135] in, Indicates the weight of power prediction error. Indicates the weight of temperature prediction error. Indicates the weight of the gas proportion prediction error. This represents the expected microwave power after trimming. This indicates the expected temperature after cutting. This represents the expected gas ratio after trimming.

[0136] Perform multiple rounds of iterative training to optimize parameters. Finally, a well-trained mapping model is obtained.

[0137] Then use the currently measured data. , , After obtaining the initial output values ​​of the model, the constraint interval of each output value is clipped to obtain the final expected value:

[0138] .

[0139] Specifically, in the diamond epitaxy process, the periodic distribution characteristics of sub-nanometer-scale steps on the surface of the heterogeneous substrate are first obtained through three-dimensional morphology scanning, and then transformed into a structural parameter vector containing spatial frequency and depth gradient. This vector, along with real-time acquired plasma spectral data, is input into a pre-trained deep neural network. The network outputs preliminary predicted values ​​for microwave power, temperature, and gas ratio through multi-layer nonlinear transformations. Subsequently, the predicted values ​​are compared with the upper limit of microwave power and the gas ratio fluctuation threshold allowed for the current reaction stage. If the predicted value exceeds the constraint range, it is corrected to the nearest safe boundary value. For example, when the model predicts a microwave power of 1500W, while the constraint limits the maximum power for this stage to 1400W, the expected value is automatically adjusted to 1400W.

[0140] This embodiment uses structural coding to precisely match the plasma modulation effect of sub-nanometer patterns with the control parameters, and eliminates out-of-bounds errors in model predictions through constraint trimming, thereby ensuring uniform growth of the epitaxial layer on complex microstructures.

[0141] The details of step 300 have been described above. Steps 400 and 500 will be explained in detail below.

[0142] (1) Determine whether the real-time microwave power and gas ratio of the current reaction stage meet the expected value. If not, do not adjust; if yes, adjust to the expected value.

[0143] (2) Calculate the average temperature of the substrate surface, i.e., sample N surface temperature points and calculate the average value:

[0144] .

[0145] (3) Determine the temperature Does it meet expectations, i.e. and If the difference is within the allowable error range, no adjustment is needed; otherwise, the surface temperature of the heterogeneous substrate is finely adjusted to the expected value using a substrate heating device.

[0146] A brief explanation of the substrate heating device is provided. For example, a layered coil array and a heat-conducting base can be used. A ring-shaped induction coil is embedded in the outer edge of the heat-conducting base to directly heat areas prone to heat loss. M sets of miniature thermocouples are embedded within the base, arranged in a hexagonal grid with spacing adjusted according to actual conditions, to monitor the temperature at various points on the sub-nanometer pattern in real time. Additionally, an infrared-transmitting window (transmittance >90%) can be placed at the top of the reaction chamber, used in conjunction with a short-wave infrared camera (3–5 μm band) to generate a real-time temperature field distribution map. Multiple sets of annular closed water tanks are set below the heating coil layer, circulated with deionized water (flow rate 0.5–5 L / min). Rapid cooling and temperature maintenance are achieved by changing the flow rate. Specifically, a metal expansion plate is installed in the water-cooling pipeline, which deforms upon heating, compressing the pipeline and changing its cross-sectional area.

[0147] 1. When the temperature is too high: the expansion plate squeezes the pipeline → the cross-sectional area decreases → the flow rate increases → the cooling is enhanced;

[0148] 2. When the temperature is too low: the expansion plate resets → the cross-sectional area increases → the flow rate decreases → heat dissipation is reduced.

[0149] The above is just a simple explanation of the specific implementation of the substrate heating device. In specific applications, other substrate heating devices with different structures can also be used, as long as they can achieve millimeter-level micro-region control of the surface temperature of the heterogeneous substrate.

[0150] Steps 400 and 500 have been explained in detail above. Step 600, in which the temperature difference is calculated between the surface temperature of the sub-nanoscale patterned area on the heterogeneous substrate and the temperature of the surrounding region of the sub-nanoscale pattern, is then determined. It is then determined whether the temperature difference is within a preset temperature range; otherwise, the airflow distribution is changed until the temperature difference is within the preset temperature range. The details are as follows:

[0151] (1) Temperature difference calculation

[0152]

[0153] in, This indicates the temperature of the central region of the sub-nanometer pattern. This indicates the average temperature of the area surrounding the pattern.

[0154] (2) Determine whether it is within the allowable temperature difference range.

[0155]

[0156] in, , This indicates the lower and upper limits of the permissible temperature difference range;

[0157] If the above conditions are not met, then adjustment will be initiated.

[0158] (3) Change the airflow distribution

[0159] like Then adjust the current airflow distribution parameters. .

[0160] Adjustment strategies could include increasing the local flow rate above the patterned area to enhance convective cooling or changing the incident angle to make heat exchange more uniform, etc.

[0161] The following is combined with Figure 9 The changes in airflow distribution are explained in detail below:

[0162] The relative positional relationship between the flow guiding device and the sub-nanometer pattern on the surface of the heterogeneous substrate is determined based on the distribution of the flow guiding device and the position of the heterogeneous substrate, thereby establishing a linear mapping relationship between the angle increment of the flow guiding device and the temperature difference change.

[0163] If the temperature difference is not within the temperature range, the following decision is made based on the difference between the temperature difference and the temperature range:

[0164] (a1) When the temperature difference is lower than the temperature range, the driving guide device increases the angle in positive steps;

[0165] (a2) When the temperature difference is higher than the temperature range, the driving guide device reduces the angle by a negative step.

[0166] Right now:

[0167] (1) Establishment of linear mapping relationship

[0168] Assuming local approximation linearity, within a certain range, we have: 1.

[0169] 1 indicates the current angle of the flow guiding device. 1 represents the amount of adjustment for the guide angle. This can be determined based on offline experiments or online data fitting. .

[0170] (2) Determine whether the temperature difference is within the allowable range: If so, then the angle adjustment will be triggered.

[0171] (3) Decision on adjusting the flow angle

[0172] The angle of the flow guide device is adjusted as follows: .

[0173] in, Represents a symbolic function. This indicates the temperature difference deviation, which is the deviation between the current temperature difference and the boundary of the temperature range.

[0174] like ,but Adjust the direction to a positive step size and increase the angle;

[0175] like ,but Adjust the direction to a negative step size to reduce the angle.

[0176] The flow guiding device refers to the mechanical structure used to regulate the gas flow direction within the MPCVD reaction chamber. Specifically, it can be implemented using a rotatable flow guide plate or blades. By changing the angle of the flow guiding device, the plasma distribution and thermal field uniformity can be adjusted. A positive step increase in angle means increasing the relative angle between the flow guiding device and the sub-nanometer pattern region by a preset fixed angle increment. A negative step decrease in angle means decreasing the relative angle between the flow guiding device and the sub-nanometer pattern region by a preset fixed angle increment. This can be achieved by using a stepper motor to drive the flow guiding device to rotate.

[0177] This embodiment can balance the thermal field distribution on the surface of the heterogeneous substrate in real time by dynamically adjusting the angle of the flow guiding device, ensuring that the temperature gradient of the sub-nanometer patterned region is always in the optimal range during the epitaxial growth process, thereby improving the crystallization quality and interface bonding strength of the diamond single crystal.

[0178] Step 600 has been explained in detail above. Step 700, which involves analyzing the data before and after adjustment and outputting an experimental analysis report, will be explained in detail below:

[0179] The experimental analysis report includes the original state record (before adjustment), adjustment strategy record, post-adjustment state record, and comparative analysis of key indicators. It may also include comparative analysis of parameters before and after adjustment. The format can be found in the table below:

[0180] Table 2: Key Process Parameter Recording Table

[0181] Parameter categories Before adjustment After adjustment Optimization range microwave power 5000W 5300W (Expected value) +6% <![CDATA[Gas ratio (CH4 / H2)]]> 5% 4.2% (Expected Value) -16% Substrate average temperature 1080℃ 1050℃ (Expected value) -30℃ Temperature difference in sub-nanometer patterned region >25K ≤15K (preset range) 40% decrease

[0182] In addition, it includes the problems that arose during the experiment and the analysis of the reasons for those problems, so that the experimenters can better understand the experimental process and make clearer plans for future experiments.

[0183] To facilitate understanding of the above embodiments, the following will be combined with... Figure 10 The following is an example of a specific application scenario from the above embodiments:

[0184] The equipment mainly consists of an MPCVD reaction chamber system equipped with an OES instrument, a QMS instrument, a microwave power source, a substrate heating device, and a flow guiding device, thereby achieving closed-loop control capabilities for temperature, microwave power, and gas ratio.

[0185] A 4H-SiC substrate with a surface-processed sub-nanometer groove pattern of 10nm spacing and 0.5nm depth was used, and the reaction gas was controlled by three gases, namely CH4 / H2 / CO2.

[0186] Assume the initial conditions are: microwave power set to 800W; initial substrate temperature set to 880°C; initial gas flow ratio: CH4 / H2=1:95, CH4 / CO2=1:3; and cavity pressure maintained at 50 Torr.

[0187] The system collects substrate surface temperature distribution, spectral data, microwave power and gas ratio settings, and impurity concentration data in real time.

[0188] The system extracts the characteristic peak ratio of active species CH* to C2, compares it with a preset mapping table, and determines that the current stage is "nucleation enhancement stage".

[0189] The system calls the deep learning model. Inputs include: substrate pattern encoding (in image or geometric parameter form); current spectral intensity matrix; and the current process parameter constraints.

[0190] Expected output value:

[0191] There is a difference between the current set value and the expected value, so the system issues a fine-tuning command; the microwave power is adjusted from 800W to 820W, and the real-time feedback shows that the peak ratio of CH* / C2 is close to the reference range after the gas ratio is adjusted. The average substrate temperature is still lower than the expected value, so the substrate heating device is used to compensate by raising the temperature by 5°C.

[0192] Pattern area =893℃, surrounding area =884℃, assuming the allowable temperature difference range is [3,5℃], the temperature difference is 9℃, which has exceeded the allowable range. The system establishes a linear gain model based on the position of the current guiding device and the center position of the substrate, adjusts the current guiding angle, and after multiple fine adjustments, the temperature difference drops to within the allowable range.

[0193] The current impurity concentration was measured to be 8. Exceeding the current stage threshold by 5 A moderate purification strategy was initiated, and after five consecutive monitoring cycles, the impurity concentration decreased to 5. The following meet the requirements of the purification trend.

[0194] The system outputs a report at the end of the cycle, including the original status record (before adjustment), adjustment strategy record, post-adjustment status record, and key indicator comparison analysis.

[0195] In summary, this embodiment calculates the micro-area temperature difference between the sub-nanometer patterned region and its surrounding area, enabling real-time identification of uneven heating or local anomalies. Dynamic corrections are then achieved using current-guiding control and localized heating methods, resulting in a uniform thermal field. This is particularly crucial for addressing microstructure issues such as edge overheating and pattern shading during patterned substrate growth. Ultimately, this improves grain orientation consistency, reduces dislocation density, increases epitaxial nucleation success rate, and reduces the risk of film warpage or peeling. The goal is to improve parameter control precision, enhance nucleation uniformity, achieve dynamic thermal field balance, and promote improved crystal quality.

[0196] Exemplary System

[0197] like Figure 11 As shown in the figure, this embodiment introduces a sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial system, which includes a data acquisition module, a data processing module, a data prediction module, a data judgment module, and a data output module.

[0198] The data acquisition module is used to acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber.

[0199] The data processing module is used to extract the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculate the current gas ratio. In addition, it calls the historical database to generate the constraints corresponding to the current reaction stage.

[0200] The data prediction module is used to calculate the expected values ​​from the spectral data and constraints using deep learning; the expected values ​​include the microwave power expected value, the average surface temperature of the heterogeneous substrate expected value, and the gas ratio expected value.

[0201] The data judgment module is used to determine whether the microwave power and gas ratio meet the expected value; otherwise, it adjusts the microwave power and gas ratio to the expected value, then calculates the average temperature of the surface temperature data, and determines whether the average temperature meets the expected value.

[0202] Otherwise, the surface temperature of the heterogeneous substrate is finely adjusted using a substrate heating device until the desired value is met;

[0203] If so, the temperature difference is calculated by measuring the difference between the surface temperature of the sub-nano patterned area on the heterogeneous substrate and the temperature of the surrounding area of ​​the sub-nano pattern. It is then determined whether the temperature difference is within a preset temperature range. Otherwise, the airflow distribution is changed until the temperature difference is within the temperature range.

[0204] The data output module is used to analyze the data before and after adjustment and output an experimental analysis report.

[0205] This embodiment effectively solves the defect generation problem caused by lattice mismatch during heteroepitaxial growth, ensuring the uniformity of epitaxial layer growth through real-time monitoring and dynamic adjustment. A multi-level tuning mechanism enables precise control of process parameters, avoiding quality fluctuations caused by parameter mismatch in traditional methods. The airflow distribution optimization function significantly reduces the temperature gradient in the sub-nanometer patterned region, suppressing the formation of dislocations and twins. Experimental analysis reports provide data support for process optimization, improving the repeatability and stability of the epitaxial process.

[0206] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for MPCVD diamond heteroepitaxial growth assisted by sub-nanometer patterned substrates, characterized in that, It includes the following steps: Acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber; The presence and concentration changes of active reactants in the spectral data are extracted to determine the current reaction stage and calculate the current gas ratio. In addition, the historical database is called to generate the constraints corresponding to the current reaction stage. The expected values ​​are obtained by deep learning calculation using the spectral data and constraints; the expected values ​​include microwave power expected value, average surface temperature of heterogeneous substrate expected value, and gas ratio expected value. Determine whether the microwave power and gas ratio meet the expected value; otherwise, adjust the microwave power and gas ratio to the expected value. Then, calculate the average temperature of the surface temperature data and determine whether the average temperature meets the expected value. Otherwise, the surface temperature of the heterogeneous substrate is finely adjusted using a substrate heating device until the desired value is met; If so, the temperature difference is calculated by the difference between the surface temperature of the sub-nano patterned area of ​​the heterogeneous substrate and the temperature of the surrounding area of ​​the sub-nano pattern. It is then determined whether the temperature difference is within the preset temperature range. Otherwise, the airflow distribution is changed until the temperature difference is within the temperature range. Analyze the data before and after the adjustment and output an experimental analysis report.

2. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 1, characterized in that, The specific steps for extracting the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculate the current gas ratio, and for calling the historical database to generate the constraints corresponding to the current reaction stage, are as follows: The characteristic peak intensities of the active reactants are extracted from the spectral data and normalized. The ratio of the processed characteristic peak intensities is used to calculate the current gas ratio. The current reaction stage is obtained by referring to a preset mapping table characterizing the relationship between gas ratios and stages; The temperature range, microwave power range, and gas ratio range of the current reaction stage are retrieved from the historical database and used as constraints.

3. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 1, characterized in that, The expected values ​​are calculated using deep learning based on the spectral data and constraints. These expected values ​​include the microwave power expected value, the average surface temperature of the heterogeneous substrate expected value, and the gas ratio expected value. Furthermore, the current reaction stage is verified, with the specific steps as follows: Obtain the reaction phases of the previous two cycles in the current cycle; Determine whether the current cycle is consistent with the reaction stage of the three consecutive cycles consisting of the previous two cycles. If so, the current reaction stage is determined to be accurate. Otherwise, the spectral data will be reacquired and the reaction stage will be determined until the reaction stages are consistent for three consecutive cycles or the number of iterations reaches a preset value, at which point an abnormal warning command will be issued.

4. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 3, characterized in that, After verifying the current reaction stage, purification is also performed based on the changes in the concentration of gaseous impurities within the MPCVD reaction chamber. The specific steps are as follows: To obtain the concentration of impurities in the reaction atmosphere within the MPCVD reaction chamber; Determine whether the impurity concentration exceeds the impurity threshold corresponding to the current reaction stage; If so, the impurity concentration difference is calculated by the difference between the impurity concentration and the impurity threshold, and the purification level is determined by referring to the preset purification reference table. Then, the corresponding purification operation is performed according to the purification level. The purification reference table represents the mapping relationship between the impurity concentration difference and the purification intensity, and different purification intensities have their corresponding purification levels.

5. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 4, characterized in that, After the purification operation, the constraints are adjusted based on the trend of impurity concentration changes. The specific steps are as follows: Obtain impurity concentration data for N consecutive periods after purification, and process the data to obtain the purified impurity concentration; where N > 3 and is an integer; Determine whether the concentration of the purified impurities exceeds the impurity threshold corresponding to the current reaction stage; If so, the impurity threshold corresponding to the current reaction stage is added to the constraint conditions, and then the expected value is calculated by deep learning based on the spectral data and the constraint conditions.

6. The sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth method according to claim 5, characterized in that, Adding the impurity threshold corresponding to the current reaction stage to the constraints also includes determining whether the purification operation failed based on the trend of impurity concentration changes after purification. The specific steps are as follows: The trend of impurity concentration change after purification is obtained based on the impurity concentration data of N consecutive periods after the purification operation. Determine whether the trend of the change in the concentration of impurities after purification reaches the preset trend of the decrease in the concentration of impurities after purification; Otherwise, check whether the purification point equipment is faulty. If it is determined that there is no fault, add the impurity threshold corresponding to the current reaction stage to the constraints.

7. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 1, characterized in that, The specific steps for calculating the expected value using deep learning based on the spectral data and constraints are as follows: Build a mapping model and train it based on historical data; The sub-nanometer pattern of the heterogeneous substrate is encoded as substrate structure parameters, and the preliminary expected values ​​are obtained by combining the spectral data and constraints through a trained mapping model. The expected value is obtained by comparing the initial expected value with the constraints and then trimming it to the range of constraints.

8. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 1, characterized in that, The specific steps for determining whether the temperature difference is within a preset temperature range, and otherwise changing the airflow distribution until the temperature difference is within the temperature range, are as follows: The distribution of the flow guiding device within the MPCVD reaction chamber and the position of the heterojunction substrate within the MPCVD reaction chamber are obtained. The relative positional relationship between the flow guiding device and the sub-nanometer pattern on the surface of the heterogeneous substrate is determined based on the distribution of the flow guiding device and the position of the heterogeneous substrate, thereby establishing a linear mapping relationship between the angle increment of the flow guiding device and the temperature difference change. If the temperature difference is not within the temperature range, the following decision is made based on the difference between the temperature difference and the temperature range: (a1) When the temperature difference is lower than the temperature range, the driving guide device increases the angle in positive steps; (a2) When the temperature difference is higher than the temperature range, the driving guide device reduces the angle by a negative step.

9. The method for sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth according to claim 1, characterized in that, Specifically, setting sub-nanometer patterns on a heterogeneous substrate involves etching or imprinting periodic steps, holes, or lines with a size less than 1 nm on the surface of the heterogeneous substrate.

10. A sub-nanometer patterned substrate-assisted MPCVD diamond heteroepitaxial growth system, characterized in that, It includes: The data acquisition module is used to acquire microwave power, spectral data, and surface temperature data of a heterogeneous substrate with pre-set sub-nanometer patterns within the MPCVD reaction chamber. The data processing module is used to extract the presence and concentration changes of active reactants from the spectral data to determine the current reaction stage and calculate the current gas ratio. In addition, it calls the historical database to generate the constraints corresponding to the current reaction stage. The data prediction module is used to calculate expected values ​​from the spectral data and constraints using deep learning; the expected values ​​include microwave power expected values, average surface temperature of the heterogeneous substrate expected values, and gas ratio expected values. The data judgment module is used to determine whether the microwave power and gas ratio meet the expected value; otherwise, it adjusts the microwave power and gas ratio to the expected value, then calculates the average temperature of the surface temperature data, and determines whether the average temperature meets the expected value. Otherwise, the surface temperature of the heterogeneous substrate is finely adjusted using a substrate heating device until the desired value is met; If so, the temperature difference is calculated by the difference between the surface temperature of the sub-nano patterned area of ​​the heterogeneous substrate and the temperature of the surrounding area of ​​the sub-nano pattern. It is then determined whether the temperature difference is within the preset temperature range. Otherwise, the airflow distribution is changed until the temperature difference is within the temperature range. The data output module is used to analyze the data before and after adjustment and output an experimental analysis report.

Images