Gallium nitride-based diode epitaxial method and device with reverse voltage boosting

Abstract

The application discloses a gallium nitride-based diode epitaxial method and device for improving reverse voltage resistance, and relates to the technical field of semiconductor manufacturing. The method comprises the following steps: through reflection, curvature and light-emitting three-channel cooperative online monitoring, sequentially performing superlattice buffer stack construction, low-leakage buffer structure generation and epitaxial termination structure formation; then, based on the monitoring signal, using time sequence synchronization, noise filtering and threshold comparison to perform parameter linkage calculation, dynamically updating source flow and temperature and pressure parameters, and completing formula library write-back and batch record generation, forming a self-adaptive closed-loop control. Through full-process real-time monitoring and feedback regulation, the method effectively improves the reverse voltage resistance of the gallium nitride diode, significantly reduces the leakage current, enhances the epitaxial layer quality and batch consistency, and realizes the intelligentization and traceability of the process.

Description

Technical Field

[0001] This invention relates to the field of electronic component and equipment manufacturing technology, and in particular to a method and apparatus for epitaxial growth of gallium nitride-based diodes with improved reverse breakdown voltage. Background Technology

[0002] In the field of electronic component and equipment manufacturing technology, existing solutions for epitaxial growth of nitride devices typically revolve around the substrate, epitaxial formulation parameter set, and temperature ramp curve. Through degassing, surface reconstruction, and nucleation heating, coupled with graded aluminum content transition layer deposition and parallel online monitoring of superlattice periodic stacking, a superlattice buffer stack is formed. Subsequently, based on the carbon source target curve and silicon source target curve, source current setting, supply window arrangement, segmented layout, and continuous transition are performed, along with background impurity suppression to generate a low-leakage buffer structure. Afterwards, under the constraints of intrinsic growth formulation and thickness targets, intrinsic gallium nitride drift layer deposition, annealing, and hydrogen atmosphereing are carried out. The process involves replacing the lightly doped termination transition zone with growth and extension region reservation to complete the construction of the epitaxial termination layer and obtain the epitaxial termination structure. The above process generally suffers from limitations such as discrete acquisition links of reflection signal, curvature signal and light emission signal, inconsistent timing synchronization and noise filtering, and inconsistent threshold comparison rules and parameter sources. Common practices often rely on fixed thresholds and segmented empirical rules to promote source flow and temperature and pressure updates. The recipe library write-back and batch record generation links run independently. Under the constraints of the epitaxial termination structure and its adjacent processes, the decision link and control link are prone to breakage, making it difficult to meet the requirements for continuous recording and updating of epitaxial recipe parameter sets.

[0003] For the joint processing of reflected signals, curvature signals, and light-emitting signals, epitaxial termination structures, timing synchronization and parameter linkage calculations, source flow and temperature and pressure updates, formula library write-back and batch record generation, and offline process triggering, existing technologies generally suffer from shortcomings such as unclear interface definitions, inconsistent triggering relationships, and incomplete data traceability between acquisition and alignment, judgment and control, and recording and updating. It is difficult to form a continuous process of acquisition-alignment-judgment-control-recording in the process scenario, resulting in insufficient consistency and unstable links in the cross-stage transmission of epitaxial formula parameter sets and the reading of epitaxial formula parameter sets in previous steps. Summary of the Invention

[0004] The purpose of this invention is to provide a gallium nitride-based diode epitaxial method and apparatus with improved reverse withstand voltage, which mainly solves the technical defects in existing gallium nitride-based diode epitaxial processes, such as discrete multi-source signal acquisition, fragmented control and judgment, lag in parameter updates, and poor batch consistency.

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

[0006] An epitaxial method for gallium nitride-based diodes with improved reverse breakdown voltage includes:

[0007] Acquire substrate, epitaxial formulation parameter set and temperature ramp curve, perform online monitoring including three channels including reflection channel, curvature channel and light emission channel, perform degassing and surface reconstruction, nucleation heating and graded aluminum content transition layer deposition using a continuous transition strategy of aluminum composition from high to low or from low to high, superlattice periodic stacking process of alternating deposition of two or more types of sublayers in a fixed or quasi-fixed order, to generate superlattice buffer stack;

[0008] The background impurity suppression process involves setting the source flow, arranging and segmenting the supply window under dual references of time axis and growth depth, continuous transition and synergistic integrated measures for gas path and cavity, and generating a low-leakage buffer structure.

[0009] The process involves depositing an intrinsic gallium nitride drift layer, performing annealing and hydrogen replacement triggered at specified progress points, growing a lightly doped termination transition band by extracting the final progress point sequence from the intermediate state of the intrinsic drift layer structure, reserving an extension region, and constructing an epitaxial termination layer to obtain an epitaxial termination structure.

[0010] Perform time synchronization, noise filtering and threshold comparison using sliding median filtering and piecewise window function suppression, execute parameter linkage calculation including threshold comparison status table, source flow and temperature and pressure update, and perform recipe library write-back, batch record generation and offline process triggering operation to assemble deviation allocation results and read-back segments into change application entries, and generate extended recipe parameter set.

[0011] Furthermore, in this invention, the substrate, epitaxial formulation parameter set, and temperature ramp curve include:

[0012] The substrate includes crystal orientation, surface roughness, surface defect density, and contaminant spectrum;

[0013] The epitaxial formulation parameter set includes source flow settings, reaction chamber pressure range, carrier gas ratio, heating power cycle time, rotation speed range, and alarm threshold.

[0014] A temperature ramp curve is a continuous segment describing a heating section, an isothermal section, and a cooling section.

[0015] Furthermore, in this invention, timing synchronization includes: a set of actions that map the timestamps of each channel to a single main time axis and restore the sampling order without changing the original sampling content of the three signals; the synchronization starting point selection strategy adopts the first stable rhythm inflection point of the reflection stripe as the main anchor, and the curvature slope transition and the luminescence peak position stabilization window as the auxiliary anchor.

[0016] Furthermore, in this invention, noise filtering employs a combined strategy of sliding median filtering and piecewise window function suppression. Sliding median filtering targets pulse spikes, while piecewise window function suppression targets slow drifts. The window function length and weight values ​​are automatically derived based on the rhythm statistical labels of the aligned index.

[0017] Furthermore, in this invention, the parameter linkage calculation includes three types of rules: the basic rule defines the proportional relationship between the reflection stripe rhythm offset and the precursor source flow fine-tuning and turntable speed fine-tuning; the cumulative curvature change is mapped to the priority relationship between the reaction chamber pressure fine-tuning and the carrier gas ratio fine-tuning; the emission peak position and intensity texture change are mapped to the linkage relationship between the three-signal purification data and threshold comparison status table of the heating power beat fine-tuning and the doping source flow temporary threshold; the adaptive rule updates the proportional weight and priority order based on the deviation trajectory and stable section of historical batches; and the conflict arbitration rule handles the mutual entanglement caused by multiple adjustments.

[0018] Furthermore, in this invention, the source flow and temperature / pressure update include: a trial segment for establishing the action response direction, a confirmation segment for fixing the action level, and a readback segment for evaluating the changes in the three signals brought about by the action on a unified time axis.

[0019] Furthermore, in this invention, the recipe library write-back includes: assembling the deviation allocation result and the final state of the read-back segment and the threshold comparison status table together into a change application entry; the change application entry enters the pending area of ​​the recipe library to trigger a two-level review process; and after the review is passed, a new version of the extended recipe parameter set version draft is generated.

[0020] Furthermore, in this invention, the process of triggering the offline process also includes: the external termination structure entry reaches the completion mark, and there are no unclosed frozen entries in the batch event record; when both conditions are met simultaneously, an offline trigger order is generated and an unlock signal is formed at the device layer.

[0021] Furthermore, in this invention, the generated epitaxial formulation parameter set produces output products that form bidirectional pathways in both structural and data dimensions. In terms of structure, the generated structured parameter set object is output as the epitaxial formulation parameter set, which is read when the process returns to the nucleation stage. In terms of data, the three-signal purification data, threshold comparison status table, and readback fragments are stored as prior inventory for the three-signal stage before the next startup.

[0022] The present invention also provides a gallium nitride-based diode epitaxial device with reverse breakdown voltage enhancement for implementing the above method, comprising:

[0023] The timing synchronization module is used to timestamp and map the reflected signal, curvature signal and light emission signal to a single main time axis without changing the original sampling content of the three signals, and restore the sampling order. It also registers the deviation entries to the data layer and outputs the three-signal stage alignment data to the noise filtering and baseline stabilization module.

[0024] The noise filtering and baseline stabilization module is used to perform front-end hardware calibration and baseline stabilization, sliding median filtering and piecewise window function suppression on the three-signal stage alignment data. It derives parameters based on the rhythm statistical labels of the alignment index, generates three-signal cleaned data, and provides the three-signal cleaned data and noise profile list to the threshold comparison module.

[0025] The threshold comparison module is used to extract rhythm length sequence, slope sequence, peak position sequence and intensity texture sequence from the three-signal clean data, perform item-by-item comparison according to the static and dynamic parts of the threshold library, form out-of-bounds markers, coupling descriptions and conflict markers, generate a threshold comparison status table, and provide the threshold comparison status table to the parameter linkage calculation and conflict arbitration module.

[0026] The parameter linkage calculation and conflict arbitration module is used to decompose the deviation information in the threshold comparison status table and the three-signal purification data into precursor source flow, carrier gas ratio, reaction chamber pressure and heating power according to the basic rules, adaptive rules and conflict arbitration rules, generate deviation allocation results, and send the deviation allocation results to the source flow and temperature and pressure update module.

[0027] The source flow and temperature / pressure update module is used to perform progressive updates of the precursor valve group duty, evaporation source heating power, carrier gas ratio and reaction chamber pressure setpoint according to the deviation allocation results. It advances the action according to the trial segment, confirmation segment and readback segment, forming readback segments, execution failure entries and safety gear records, and returns the readback segments to the parameter linkage calculation and conflict arbitration module.

[0028] The recipe library write-back and batch record generation module is used to assemble the deviation allocation results and read-back segments and threshold comparison status table into change application entries, enter the recipe library pending review area, generate the extended recipe parameter set version draft, and aggregate the batch overview page, process event page, data index page and evidence chain page to form batch records, and output the version number and batch status to the downstream process trigger module.

[0029] The offline process trigger module is used to generate an offline trigger order when the extension termination structure item reaches the completion mark and there are no unclosed frozen items in the batch event record, send an unlock signal to the cooling and stress release and surface cleaning module, and write the start time and end time into the batch record.

[0030] The data layer traceability and evidence chain module is used to maintain the mapping relationship between batch event records and evidence chain case file numbers, and to establish a two-way traceability channel between three-signal stage aligned data, three-signal cleaned data, threshold comparison status table, deviation allocation results and batch records.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] (1) This invention achieves unified alignment by mapping the timestamps of each channel to a single main time axis through a three-channel collaborative online monitoring mechanism; at the same time, it adopts a combination strategy of sliding median filtering and piecewise window function suppression to accurately filter out two types of noise: pulse spikes and slow drift. This design solves the problems of timing disorder and noise interference of multi-source signals, provides a high-precision data foundation for subsequent threshold comparison and parameter control, and directly improves the monitoring accuracy of epitaxial layer growth status.

[0033] (2) This invention constructs a complete closed-loop control system of “threshold comparison - parameter linkage calculation - source current and temperature and pressure update”: by using multi-dimensional signals such as reflection stripe rhythm shift, curvature cumulative change, and emission peak position drift, combined with basic rules, adaptive rules and conflict arbitration rules, core process parameters such as precursor source current, carrier gas ratio, and reaction chamber pressure are dynamically adjusted. This mechanism realizes real-time optimization of process parameters, effectively improves the reverse withstand voltage characteristics of gallium nitride diodes, significantly reduces leakage current, and enhances the quality stability of epitaxial layers.

[0034] (3) This invention assembles the deviation allocation results, readback segments, and threshold comparison status table into change application entries, and realizes the version update of the extended formula parameter set through a two-level review process; at the same time, it aggregates the entire process data to generate batch records including an overview page, process event page, data index page, and evidence chain page, and establishes a two-way traceability channel from structure to formula and back to structure. This design realizes the traceability of the entire process link, greatly improves the production consistency between batches, and provides complete data support for intelligent process iteration. Attached Figure Description

[0035] Figure 1 This is a schematic flowchart illustrating a gallium nitride-based diode epitaxial method with reverse breakdown voltage enhancement provided in an embodiment of the present invention.

[0036] Figure 2 This is a structural block diagram of a gallium nitride-based diode epitaxial device with reverse breakdown voltage boosting, provided as an embodiment of the present invention.

[0037] The names corresponding to the reference numerals in the attached figures are as follows:

[0038] 101. Timing Synchronization Module; 102. Noise Filtering and Baseline Stabilization Module; 103. Threshold Comparison Module; 104. Parameter Linkage Calculation and Conflict Arbitration Module; 105. Source Flow and Temperature / Pressure Update Module; 106. Recipe Library Write-back and Batch Record Generation Module; 107. Offline Process Trigger Module; 108. Data Layer Traceability and Evidence Chain Module. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0040] Example 1

[0041] like Figure 1 As shown, this invention discloses a gallium nitride-based diode epitaxial growth method with improved reverse breakdown voltage. The method flow is as follows:

[0042] S1, acquire the substrate, epitaxial formulation parameter set and temperature ramp curve, perform online monitoring including three channels including reflection channel, curvature channel and light emission channel, perform degassing and surface reconstruction, nucleation heating and graded aluminum content transition layer deposition using a continuous transition strategy of aluminum composition from high to low or from low to high, superlattice periodic stacking process of alternating deposition of two or more types of sublayers in a fixed or quasi-fixed order, to generate superlattice buffer stack.

[0043] The substrate is a single-crystal substrate used for nitride epitaxy, containing basic descriptions such as crystal orientation, surface roughness, surface defect density, and contaminant spectrum. The epitaxial formulation parameter set is a complete set of process parameters from the previous round of parameter linkage control and formulation library write-back, including source flow settings, reaction chamber pressure range, carrier gas ratio, heating power cycle time, rotation speed range, and alarm thresholds, and is versioned with historical batch records for easy traceability and comparison. The temperature ramp curve is the temperature change trajectory over time during the nucleation and transition layer stages, including continuous segments describing the heating, isothermal, and cooling phases. Specifically, using the aforementioned substrate, epitaxial formulation parameter set, and temperature ramp curve as input, degassing and surface reconstruction operations are performed through loading, vacuuming, and heating triggering. Degassing is a treatment sequence that removes adsorbed water, organic residues, and weakly bonded adsorbed impurities under the combined action of vacuum and heating. Surface reconstruction is an operation process that rearranges surface atomic steps and reduces metastable structures under set temperature, carrier gas, and precursor reference flow. Upon completion of reconstruction, a nucleation interface parameter set is obtained, including surface step density, crystal orientation consistency markers, and reconstruction completion timestamps. Further, during the nucleation heating stage, the heating and isothermal sections of the temperature ramp curve are invoked. By setting the power and adjusting the carrier gas ratio segment by segment, the surface is brought to an activation state suitable for nucleation. Nucleation heating is defined as a preparatory process for establishing surface active sites using heating cycles before the precursor is fully introduced. The output usable information is recorded in the nucleation interface parameter set as the entry point for subsequent graded aluminum content transition layer deposition. In terms of anomaly handling, if the reflection baseline offset exceeds the threshold during reconstruction, the system pauses and enters a waiting window, and writes the offset and waiting time into the batch event record; if the pressure remains high during the vacuuming process, the system triggers a cavity leakage self-check and switches to the backup evacuation path, while adding a one-time correction mark to the formula parameter set for comparison and identification by subsequent online monitoring channels.

[0044] Specifically, nucleation state tags and surface step density ranges are extracted from the nucleation interface parameter set, and then deposited into a graded aluminum content transition layer. The graded aluminum content transition layer is defined as a stress-relieving transition structure constructed after nucleation using a continuous transition strategy of aluminum composition from high to low or low to high, used to coordinate stress accumulation caused by differences in lattice constants and thermal expansion. The corresponding implementation process involves establishing segmented source flow trajectories and segmented carrier gas ratios under the guidance of the epitaxial formulation parameter set, and performing brief steady-state maintenance at each segment boundary to ensure continuous and identifiable compositional transitions. To identify segment boundaries and intralayer periodic uniformity, online reflection, curvature, and emission channels are connected. The reflection channel is defined as a measurement path that monitors the change in the intensity of incident light reflected from the sample surface over time, used to determine the rhythm of surface interference fringes and the layer thickness growth beat. The curvature channel is defined as a real-time curvature trajectory constructed through the substrate bending changes caused by stress, used to determine stress release trends and abrupt changes. The emission channel is defined as a measurement path that records intrinsic or near-band edge emission intensity and peak position drift under stimulated conditions. All three channels of data are synchronously written to the data acquisition module with timestamps and associated by batch number. For the boundary determination of the graded aluminum content transition layer, the control process reads the inflection point of the reflection ripple and the change in curvature slope. When the two show a resonant change trend within the same time window, it is determined that a new component segment has been entered; if the two do not form a corresponding relationship, the peak position drift of the emission channel is used as the arbitration basis. For anomaly handling, if the curvature continuously increases and the reflection stripe spacing is abnormally compressed in any segment, the control process switches the segment formula to a safe formula and marks the segment as an abnormal segment. All data of the abnormal segment is recorded in the batch event log for subsequent parameter linkage calculation. The output of this stage is the graded aluminum content transition layer structure, which contains a description of the component gradient within the wafer, the start and end time of the segment, the average growth rate marker of each segment, and the three-channel alignment index. This structure serves as the basic reference for subsequent superlattice periodic stacking.

[0045] Furthermore, a superlattice periodic stacking is performed on the graded aluminum content transition layer structure. A superlattice is defined as a periodic buffer structure formed by the alternating deposition of two or more types of sublayers in a fixed or quasi-fixed order, with sublayer thicknesses at the nanometer level. However, this specification does not use numerical parameters to describe it, only defining its periodicity and alternation. Superlattice periodic stacking is a structured process that sequentially completes sublayer switching, maintains a stable interface, records cycle timestamps, and replicates the online monitoring baseline within each cycle. During implementation, the control flow reads the cycle list from the epitaxial formulation parameter set and performs source-current switching, reaction chamber pressure fine-tuning, and rotation speed correction on the sublayers within each cycle. The criterion for source-current switching is the starting signal of a new round of interference cycle appearing in the reflection channel, while simultaneously referencing the instantaneous slope of the curvature channel. If the slope shows a gradual decrease or a stable state after switching, the switching is considered complete; if it shows a continuous increase, a transition sublayer is inserted for sustained release. The emission channel triggers short-term sampling at the end of each cycle, and its peak drift trajectory is used to evaluate cycle repeatability. To handle potential formula drift during the cycle, the system resets the reflection baseline and curvature baseline to the average value at the end of each cycle and writes the target deviation for the next cycle into the real-time control cache. If multiple consecutive cycles are determined to be abnormal, the system enters protection mode without changing the overall cycle count. In protection mode, the source current and heating power cycle time are reduced while maintaining the rotational speed until the three channels return to a stable fluctuation range. The start and end times, sublayer sequence number, and switching time of each cycle are recorded in the cycle metadata table, which is linked to the batch event record via batch number. During the cycle interval, the equipment invokes the carrier gas ratio adjustment function, with the adjustment range set according to the cumulative curvature value of the previous cycle to prevent stress from accumulating further. If the emission peak position drift of the previous cycle shows a blue shift trend, it indicates that the sublayer band structure adjustment has not yet converged. The control process extends the sublayer thickness maintenance time of the next cycle by a minimum time segment, and all related actions are written into the extended fields of the cycle metadata table. When the superlattice periodic stacking is completed, a superlattice buffer stack is obtained. The description of the stack includes period count, sublayer switching sequence, three-channel alignment index of reflection, curvature and emission, period anomaly marker and period metadata table number.

[0046] Online monitoring is integrated throughout the entire process of nucleation heating, graded aluminum content transition layer deposition, and superlattice periodic stacking. Online monitoring is defined as a set of acquisition, judgment, and feedback actions that do not interrupt the growth continuity. This step employs a three-channel coordinated approach: reflection, curvature, and emission channels. This three-channel coordination refers to collecting data according to a unified clock process and forming an alignment index within the data acquisition module. Data alignment is achieved by using the phase marker of the reflection fringes as the main time anchor, projecting the curvature slope change points and the inflection points of emission peak position drift onto the main time axis. If a projection error exceeds a threshold, the system invokes the data repair function. Data repair is defined as generating interpolation segments and label segments for each abnormal window without altering the original record, and attaching the repair segment number to the batch event record. To ensure data link continuity, the online monitoring module is configured with connection status inspection. When the inspection detects a short-term disconnection of the acquisition channel, the system prioritizes retaining the curvature channel and suspends the emission channel, while simultaneously recording the disconnection time and recovery time. In this step, the primary purpose of online monitoring data is for process-wide judgment and cycle switching. The data is also marked as a deliverable object for subsequent main steps. The three-channel raw data serves as the input for the three-signal raw data in the online characterization and parameter linkage control of the subsequent main steps. This step does not directly perform parameter linkage calculations; instead, key events are written to a temporary area of ​​the epitaxial formula parameter set only when necessary to prevent drift between the formula and records. For triggering conditions, online monitoring initiates pre-acquisition after the cavity pressure reaches the upper limit of the set range, and formal acquisition begins after the isothermal phase of nucleation heating is completed. A staged archive is generated when the superlattice periodic stacking is complete, with each archived file corresponding to a batch number. If a missing page or timestamp jump occurs during archive verification, the system triggers a re-archiving process. This process reads data back from the multi-channel buffer, repairs the alignment index, and rewrites the archived file. Related operations are recorded in the monitoring sub-volume of the batch event record.

[0047] During this step, the connection between data structures and fields is established through natural links. The nucleation interface parameter set originates from the state acquisition during the degassing, surface reconstruction, and nucleation heating stages, including interface state labels, step density ranges, and reconstruction completion timestamps. This parameter set subsequently drives the selection of the initial segment formulation for the graded aluminum content transition layer deposition. The graded aluminum content transition layer structure carries the segment boundaries, composition trajectories, and three-channel alignment indexes, providing a reference for the cycle length and switching time for superlattice periodic stacking. The cycle metadata table and batch event records support anomaly identification and protection mode switching during the cycle. After this step, the three-channel data and alignment indexes form callable entries, named "Three-Channel Stage Data," which are marked as deliverable objects for subsequent main steps to read. The superlattice buffer stack is the output field name for this step. It contains a periodic structure description, anomaly markers, alignment indexes, and periodic metadata table number. This output appears as an input in subsequent steps. Specifically, the superlattice buffer stack field is read in the first sub-step of doping cooperative gradient construction, and the source flow trajectory varying with depth is set in conjunction with the carbon source target curve and the silicon source target curve. The interface between the two is mapped by the batch number and the periodic metadata table number. Meanwhile, the key events and recipe temporary region rewriting content generated during this step will be incorporated into the official version of the epitaxial recipe parameter set in the recipe write-back action of the subsequent main step, forming a closed-loop record across main steps.

[0048] In this embodiment, substrate loading and positioning are achieved using a robotic arm and positioning pins in tandem. The contact pose between the conical surface of the positioning pin and the substrate edge follows a fixed standard in the loading statistics. If the deviation exceeds the loading window, the system automatically ejects the wafer and reloads it. After loading, the turntable speed is slowly increased to the lower limit within the range specified in the epitaxial formulation parameter set to ensure stable airflow distribution within the cavity. Degassing is triggered when the cavity pressure enters the low-pressure range and remains stable for a minimum time interval. Then, the heating power cycle is activated. The power cycle has two phases: an increase and a maintenance phase. During the maintenance phase, a carrier gas is introduced, which dilutes the precursor residue and carries away escaped molecules. Surface reconstruction is completed within the pre-nucleation heating window. The window length is identified by the initial segment of the reflection channel fringes from zero to one. Once the fringes stabilize, reconstruction is considered complete, and the reconstruction completion timestamp is written into the nucleation interface parameter set. Nucleation heating enters the isothermal zone, where nucleation precursors are introduced. The introduction rate of the precursors gradually increases, and the process ends at the intersection of the period of the reflection stripes gradually shortening and the slope of the curvature gradually decreasing. At the end, a nucleation completion marker is written into the nucleation interface parameter set.

[0049] The deposition of the graded aluminum content transition layer is triggered by the completion of nucleation. The control flow reads the segment list from the epitaxial formulation parameter set, sets the source flow and carrier gas ratio combinations sequentially according to the list, and references the phase transition and curvature slope short-term reversal of the reflective fringes at the segment switching point. At the beginning of each segment, short-term sampling is performed in the luminescent channel. The peak position and half-peak width obtained from the sampling are written into the three-channel stage data and linked to the current segment number. At the end of each segment, a segment end timestamp and an average growth rate marker are written. The average growth rate marker is obtained by mapping the fringe period and the turntable speed, without involving a formula expression, only recording the mapped grade label. If a segment exhibits a merging phenomenon of fringe distortion and rapid curvature rise, the system inserts a transition segment at the end of that segment. The formulation of the transition segment comes from the spare entry in the epitaxial formulation parameter set. After the transition segment ends, the target formulation of the next segment is restored. After graded deposition, a graded aluminum content transition layer structure is formed. The structure includes segment trajectories, intra-segment sampling summaries, and three-channel alignment indexes. This structure is labeled as the transition layer structure and published externally on the data bus.

[0050] Superlattice periodic stacking is triggered by a transition layer structure release event. Within a cycle, two types of sublayers are deposited sequentially. Each sublayer switch is accompanied by a short-term steady state. The length of the steady state is graded by the cumulative curvature value at the end of the previous sublayer; a higher grade indicates a longer steady state. A fringe phase jump occurs in the reflection channel during sublayer switching, serving as the first criterion for switching completion. If the curvature channel shows a negative slope within a short time window after switching, this serves as the second criterion. If neither criterion is met, a delay window is entered. If the criterion is still not met after the delay, a protection mode is triggered, reducing the precursor introduction rate and prolonging the steady state. After each cycle, the system determines the stability of the material's band structure properties within the cycle based on the peak position changes of the emission channels. If the direction of change is consistent with historical cycles, the cycle is considered to have good consistency. If the direction is reversed, the cycle is marked as a suspicious cycle, and all parameters of the suspicious cycle are written to the cycle metadata table, prompting subsequent main steps to pay attention to the corresponding depth position. After the periodic stacking is completed, the system generates a superlattice buffer stack containing the time series of all periods, sublayer sequence number, switching time, alignment index of reflection, curvature and luminescence, and abnormal period markers.

[0051] Online monitoring and processing also handles data availability verification and data archiving. Data availability verification includes three parts: clock synchronization verification, sampling frequency verification, and data integrity verification. Clock synchronization verification compares the timestamp differences of the three channels to see if they are within the allowable range. If they exceed the range, a clock correction segment is inserted into the data layer and a correction label is added. Sampling frequency verification identifies sampling omissions in a channel due to instantaneous load. When an omission occurs, an interpolation strategy is invoked to generate a placeholder segment. The interpolation strategy is stored separately from the original data. Data integrity verification is performed before archiving, checking the consistency between the length of the three-channel data and the alignment index length. If there is a discrepancy, a re-archiving process is triggered. The archiving task is executed at the end of this step. The archive includes the original data, alignment index, periodic metadata table, and batch event record summary. The archived file forms a subscribing object on the data bus, which can be subscribed to and read by the online representation and parameter linkage control modules in the subsequent main steps.

[0052] The output of this step is a superlattice buffer stack, and this output field name is read downstream after being registered on the data bus. This field is subsequently called in the first sub-step of the doping cooperative gradient construction, with the input location name listed as the superlattice buffer stack in the calling interface. This superlattice buffer stack, along with the carbon source target curve and the silicon source target curve, serves as one of the three inputs for calculating the source current trajectory. Simultaneously, the three-channel stage data, periodic metadata table, and batch event record summary generated in this step are subscribed to by the downstream online characterization and parameter linkage control module, becoming evidence of the deviation allocation results in subsequent linkage calculations. The rewrite flag and safe recipe call flag written in the temporary area of ​​the epitaxial recipe parameter set during this step will be integrated into a new version of the epitaxial recipe parameter set in the subsequent recipe write-back stage, and will again become input in the nucleation stage, forming a natural closed loop.

[0053] Then, proceed to step S2, where source flow settings are performed, supply window arrangement and segmented layout are carried out under dual references of time axis and growth depth, background impurity suppression is performed through continuous transition and synergistic integrated measures for gas path and cavity, and a low-leakage buffer structure is generated.

[0054] Among them, the carbon source target curve is a set describing the supply intensity trajectory and boundary constraints of the carbon source precursor in the growth depth and process time axis, including the source flow setting baseline, the gradual rise and fall cycle, the forbidden zone, and the abnormal switching mark; the silicon source target curve is used to characterize the synergistic relationship between the silicon source precursor's continuous deployment with depth and the carbon source, including items such as same-phase or different-phase superposition, alternating duty and delayed overlap; the superlattice buffer stack comes from the deposition results of the preceding main step, containing period count, sublayer switching time, three-channel alignment index and abnormal period mark, and the three-channel alignment index corresponds to the unified time axis of reflection signal, curvature signal and emission signal. Specifically, the aforementioned carbon source target curve, silicon source target curve, and superlattice buffer stack are used as inputs to enter the source flow setting stage under the guidance of the epitaxial formulation parameter set. The source flow setting is defined as a combined action sequence of initial power, introductory duty cycle, and steady-state flow rate based on the precursor vapor pressure characteristics and supply path inertia. The combined action sequence is achieved through the coordinated opening and closing of valve groups, carrier gas dilution, and fine-tuning of reaction chamber pressure. The first setting time is aligned to the nearest sublayer switching time of the superlattice buffer stack, forming a time reference consistent with the periodic structure. To monitor the offset during the setting process, this step calls the Residual Gas Analysis (RGA) channel and the Optical Emission Spectroscopy (OES) channel to perform in-situ observations on the gas phase and plasma sides. RGA provides the relative intensity trajectories of hydrocarbon fragments and nitrogen-hydrogen fragments, while OES provides the emission spectral intensity and peak position drift trajectory. Both, along with the three-channel alignment index, are written to the extended page of the batch event record. If, during the execution of the source flow setting, the RGA trajectory is too high and exhibits a covariant trend in the same direction as the OES peak shift, it is determined to be an instantaneous overload. The control process will automatically revert the source flow setting to the previous steady-state setting and record the revert time and reason entry.

[0055] In the supply window arrangement stage, supply window arrangement is defined as generating staggered or overlapping open and closed windows for carbon and silicon sources under dual references of time axis and growth depth, and inserting constant-time holding pressure sections at the window boundaries to stabilize the interface reaction. Specifically, the system divides each period into several sub-segments based on the periodic metadata of the superlattice buffer stack. Each sub-segment establishes a one-to-one mapping with the first segment of the next period to carry the segmented layout of the doping cooperative gradient. The trigger condition for establishing the mapping is when the reflected signal shows the starting point of a new round of interference period and the slope of the curvature signal is in the low variation range. If the starting point does not coincide with the low variation range, the peak inflection point of the emission signal is used as the arbitration anchor point. The specific generation of the supply windows is completed by a window generator. Based on the nodes of the carbon source target curve and the silicon source target curve, the window generator arranges the first carbon source window and the first silicon source window according to the same segment and in-phase or out-of-phase relationship. When the out-of-phase arrangement is adopted, the carbon source window opens slightly earlier and maintains a short overlap after the silicon source window opens to establish the initial state of electrical compensation. When the in-phase arrangement is adopted, the two windows open synchronously and a short delay is inserted before closing to lock them, in order to suppress the instantaneous non-uniformity caused by the separation of the interface charge. The opening and closing of the supply windows are jointly achieved by the valve group pulse control and the small amplitude swing of the carrier gas ratio. The valve group pulse produces a clear step on the time axis, and the carrier gas swing produces a recognizable slope on the pressure axis. The timing of the two is uniformly written into the window timing table. In abnormal situations, if the window boundary conflicts with the switching of the superlattice sublayer, the system will delay the opening of the window and insert a short-term dilution segment within the window. The dilution segment is completed in coordination with the increase of the carrier gas ratio and the slight decrease of the source flow ratio. After the conflict is resolved, the window timing table will record a rearrangement event and provide a correction reference for subsequent segment deployment.

[0056] Segmented deployment involves dividing the supply intensity of carbon and silicon sources into several level segments along the growth depth after the supply windows are arranged, and establishing a smooth splicing strategy at the segment boundaries. Specifically, the system reads the aforementioned window time series table, maps the time axis within each window to the growth depth axis, and the mapping ratio is jointly given by the three-channel alignment index and the periodic average growth rate label to avoid introducing unobservable variables. Then, level labels for carbon and silicon sources are set within each window, using monotonic or quasi-monotonic sequences to create predictable cooperative gradients between segments. To achieve smooth splicing, the system inserts transition segments at the segment boundaries. These transition segments are short-time steady states composed of a lower source-to-flow ratio and a higher carrier gas dilution ratio. The length of the transition segment is labeled based on the cumulative curvature value of the previous segment and the emission peak position drift direction; the higher the label, the longer the transition. During segmented deployment and operation, if a divergent trend occurs where the background hydrocarbon debris in the RGA display suddenly decreases while the OES spectral line intensity increases, it indicates a sudden change in gas-phase reaction efficiency within the window. The system will maintain the current segment's level label and extend the transition segment until all three channels exhibit a consistent rhythm again. After all segments are completed, the system generates a set of doped layer configuration segments. This set contains the segment level, transition segment time, and segment duration for each window. The set is marked as being under construction and pushed to the continuous transition module.

[0057] A continuous transition establishes a continuous gradient between multiple doped layer configuration segments, without creating visible supply steps. Specifically, the system overlaps adjacent configuration segments along the time axis and growth depth axis, with two types of overlap: forward overlap and reverse overlap. Forward overlap refers to the start window of the later segment starting earlier at the end stage of the end window of the previous segment, forming a short overlap; reverse overlap refers to the end window of the previous segment opening later and closing later, forming a delayed overlap. The overlap amplitude is automatically initialized by the window generator based on the three-channel alignment index and fine-tuned according to real-time RGA and OES data during the online cruise of this step. To avoid instantaneous enrichment caused by overlap, the system introduces a micro-amplitude oscillation strategy in the overlap region. The oscillation strategy is achieved by superimposing low-frequency, small-amplitude perturbations on the source-current ratio. The frequency and amplitude of the perturbation are level parameters, not expressed in formulas, but stored in the recipe library as levels and tags. When the curvature signal monotonically rises and the emission peak position continuously blue shifts during the continuous transition, the system determines that the stress release rhythm and the band modulation rhythm are inconsistent. At this time, the new overlapping action is paused, and a dilution plug is added to the existing overlapping area. After the plug is inserted, the overlapping is resumed. If multiple overlapping areas trigger plugs consecutively, the system marks the batch as an overlapping abnormal batch. The overlapping abnormal batch forms a separate entry in the batch event record for use in the parameter linkage calculation of the subsequent main steps.

[0058] Background impurity suppression is integrated throughout the entire process, from source / flow settings and supply window arrangement to segmented deployment and continuous transition. The targets of background impurity suppression include oxygen-based debris, water-based debris, and metal vapor traces. The treatment strategy is a comprehensive, coordinated approach at both the gas path and chamber ends. At the gas path end, multi-stage purging and low-duty-hour pre-supply are employed. Purging is performed using a carrier gas pulse, while low-duty-hour pre-supply opens the precursor valve assembly to clear stagnant sections without forming effective deposits. At the chamber end, a two-step sequence of actions is used: high-temperature baking and wall renitriding. High-temperature baking increases the heating power to a preset level and maintains it in the absence of precursors. Wall renitriding introduces a nitrogen source into the carrier gas and maintains it briefly, occupying the active sites on the wall. The state determination of background impurities is jointly performed by RGA and quadrupole mass spectrometry (QMS). RGA provides species identification and relative intensity, while QMS provides the stability and drift trend of mass number sites. If an excessively high metal vapor trace is detected, the system will trigger a wall re-adsorption process. Wall re-adsorption is completed by briefly introducing an inert precursor and maintaining low-power heating, and the re-adsorption event number will be marked in the batch event log. All background suppression actions do not change the timing of the doped layer configuration segments; they only insert non-deposited segments within the window without altering the level identifier. The information for these non-deposited segments is located in an appendix to the window timing table, and the appendix is ​​linked to the main table via page number indexing.

[0059] To enhance the auditability of the doping synergistic gradient, this step correlates and records supporting data from the gas and solid phases. On the solid phase side, secondary ion mass spectrometry (SIMS) sampling measurements are performed during process intervals, with the sampling location consistent with the growth depth mapping in the window time series table. Sampling data is archived in the data layer with anonymous tags for comparing the consistency of segmented layouts across different batches. Continuous recordings on the gas phase side are provided by RGA and OES, and the data uses a unified time anchor identifier consistent with the three channels. To avoid production cycle time being affected by offline measurements, sampling is performed only on specific batches. The sampling batch number is pre-set in the recipe library and indicated before this step begins. When there is a significant discrepancy between the sampling data and online data, the system will trigger a recipe review entry. The review entry only records the event and suggestions, without changing the current batch recipe. Through the above data correlation, the doping synergistic gradient forms a consistent three-dimensional mapping relationship in the time domain, depth domain, and evidence domain. The key set of this mapping relationship is stored in the evidence chain dossier in the recipe library, providing searchable entries for subsequent main step calculations.

[0060] Once the source flow setting, supply window arrangement, segmented deployment, continuous transition, and background impurity suppression processing are all completed sequentially, the system generates a low-leakage buffer structure. The low-leakage buffer structure is the output field name for this step, and its structure description includes a segment level sequence, a window overlap list, a transition segment list, a background suppression action log, and an evidence chain dossier number. This output field name, after being registered on the data bus, is called by subsequent main steps and appears as an input item in the first sub-step of constructing the main drift layer and the termination transition zone. The input location name is "low-leakage buffer structure," and the input and this step are bidirectionally mapped in the batch number and evidence chain dossier number dimensions. Simultaneously, source flow setting deviations, window rearrangement events, overlap anomaly batches, and background suppression action logs are subscribed to by the online characterization and parameter linkage control module, used as the evidentiary basis for deviation allocation results, and transformed into new extended recipe parameter set version entries during subsequent recipe library write-back. The three-channel stage data left by the previous main steps are used as time anchors for window arrangement and segmentation in this step. The low-leakage buffer structure generated in this step will form a forward and backward bidirectional data path to the subsequent main steps and online modules. This path forms a natural closed loop in the process chain.

[0061] Next, proceed to step S3, where intrinsic gallium nitride drift layer deposition, annealing and hydrogen replacement triggered at specified progress points, growth of lightly doped termination transition bands from the intermediate state of the intrinsic drift layer structure, extension region reservation, and epitaxial termination layer construction are performed to obtain the epitaxial termination structure.

[0062] The intrinsic growth formulation is a set of parameters for nitride epitaxy, including source flow settings, reaction chamber pressure range, carrier gas ratio cycle time, turntable speed range, and heating power curve. It also includes batch numbers and version identifiers to form a traceable execution path within the same production line. The thickness target is an abstract description of the depth target and interface coverage target of the intrinsic gallium nitride drift layer. It uses segmented labels to express the layer thickness range and unfolding order, without involving numerical values. The low-leakage buffer structure comes from the preceding main step and is composed of a segmented level sequence, a window overlap list, a transition segment list, a background suppression log, and an evidence chain dossier number. It maintains a key association with the three-channel aligned index in the data layer. Specifically, the intrinsic growth formula, thickness target, and low-leakage buffer structure are used as inputs to enter the intrinsic gallium nitride drift layer deposition stage under the trigger of the process scheduler. During the deposition stage, the upper surface state label and alignment index of the low-leakage buffer structure are first read to construct an interface transition window. This window consists of three segments: low-power preheating, carrier gas micro-oscillation, and precursor slow start, used to establish the initial growth surface for intrinsic deposition without disturbing the internal stress of the buffer. Subsequently, the baseline setting of the intrinsic growth formula is introduced. This baseline setting is latched as the starting reference after the reflection signal shows regular fringes and the curvature signal enters the low-variation zone. The latching time is written to the batch event record. The deposition process proceeds according to the segmented labels of the thickness target. Segmented progress is given using time calibration and alignment index. Each progress point triggers a short-term emission sampling. The sampling results and the start and end times of the interface transition window form paired entries for subsequent comparison. Regarding boundary constraints, if the reflection fringes exhibit continuous disorder and an upward curvature slope, the process scheduler inserts a thin-layer steady-state segment at the end of the current segment. During the steady-state segment, the carrier gas is increased and the source flow is slightly decreased until the fringes return to an evenly spaced rhythm. If the emission peak position continuously drifts and is inconsistent with historical batch difference labels, it is recorded as a material bandgap anomaly. Material bandgap anomalies do not trigger formulation modifications; they are only annotated in the data layer to avoid affecting the growth continuity of this batch. Intermediate products during the deposition stage are recorded as intrinsic drift layer structure intermediate states, containing a progress point sequence, short-term sampling trajectory, and interface transition window number, and are mapped to the low-leakage buffer structure.

[0063] Annealing is a heat treatment subprocess triggered at a specified progress point in the deposition stage. Its main function is to rearrange point defects and weak bonds in the intrinsic drift layer and repair local micro-stress caused by uneven interface transitions. Specifically, after the intermediate state of the intrinsic drift layer structure reaches the mid-section label of the thickness target, the control unit smoothly increases the heating power to the annealing level and maintains a static window without precursor introduction while keeping the carrier gas stable. The changes in the fringe amplitude of the reflected signal and the cumulative value of the curvature signal within the static window are recorded synchronously. If the fringe amplitude decreases while the cumulative value remains flat, it indicates that the surface roughness tends to stabilize, and the window is closed as planned. If the fringe amplitude decreases while the cumulative value increases sharply, it indicates that the micro-stress is released in a short time, the control unit extends the static window, and writes a delay entry to the data layer. After annealing, hydrogen atmosphere replacement begins. Hydrogen atmosphere replacement involves introducing a diluted hydrogen source and maintaining it for a short time under the same temperature and pressure conditions as the annealing setting, to remove residual debris adsorbed at the interface without forming new deposits. The replacement process ends when a stable near-band edge signal appears in the luminescence sampling. The start and end times and duration of the hydrogen atmosphere replacement are recorded in the batch event log. To prevent the transient effects introduced by the replacement from persisting into subsequent growth, a short recovery phase is inserted after the replacement. During this phase, the carrier gas ratio is gradually restored, and the intrinsic precursor is introduced in a slow, gradual manner until the reflective fringes regain a regular rhythm. The execution trajectories of annealing and hydrogen atmosphere replacement are incorporated into the thermal history entry for the intermediate state of the intrinsic drift layer structure. This entry establishes a pointing relationship with the version identifier of the intrinsic growth formulation in the data layer, facilitating subsequent formulation retrospection and cross-sectional comparison.

[0064] The light-doped termination transition zone growth is a localized doping subprocess initiated when the intrinsic drift layer deposition reaches the final stage near its thickness target. This subprocess is used to establish a gradient region of charge distribution on the upper surface of the drift layer. Specifically, the progress point sequence and short-time sampling trajectory of the final stage are extracted from the intermediate state of the intrinsic drift layer structure. The segmented level sequence and window overlap list within the low-leakage buffer structure are also read, and the process scheduler uses this information to schedule the light-doped windows. The opening of the light-doped window employs a symmetrical structure of a short-time leader and a short-time tail sweep. The leader stage introduces a very low level of doping source current and maintains weak carrier gas dilution. The tail sweep stage maintains a short intrinsic precursor after the doping source current gradually withdraws, forming a natural closure of the doping gradient. To avoid conflicts between the doping window and the intrinsic deposition progress points, the system uses alignment indices as anchor points and sets buffer sub-segments before and after the progress points. These buffer sub-segments do not change the precursor type, only adjusting the carrier gas ratio to maintain deposition continuity. If the emission sampling shows a brief blue shift trend within the lightly doped window and the cumulative curvature value increases synchronously, the system determines that the doped window is too narrow and automatically extends the tail scan duration. If the emission sampling shows no significant change within the lightly doped window but the reflection fringe spacing is compressed, the system determines that the doping source current is interfering with the interface prematurely, automatically advances the leader segment, and lowers the leader segment level. All of the above actions are written into the timing record of the lightly doped window and form a segment label for the lightly doped termination transition band. After the lightly doped termination transition band is completed, the intrinsic drift layer structure intermediate state is updated to the intrinsic drift layer structure, which includes the start and end markers of the lightly doped segment, window parameters, and sampling labels, and is annotated with the evidence chain dossier number.

[0065] The extension region is reserved to construct an undoped or very weakly doped extension region after the lightly doped termination transition band. Its purpose is to provide a structural buffer on the epitaxial side for subsequent device termination design. Specifically, after the completion marker of the termination transition band is written, the control unit reads the coverage target label in the thickness target and activates the reserved segment of the extension region. The reserved segment does not introduce new doping sources, maintains a low level of the intrinsic precursor and a stable carrier gas ratio, and latches a baseline once after a new rhythm of the reflective fringes appears, used as a criterion for the continuation of the subsequent epitaxial termination layer. If the curvature signal rises significantly within the reserved segment while the emission sampling remains stable, it is judged as a stress response delay. The system inserts a short-term dilution segment to alleviate the surface reaction rate and then returns to the reserved segment. If a sudden phase jump occurs in the reflective fringes within the reserved segment and the emission sampling changes accordingly, it is recorded as an interface energy level disturbance event. The process scheduler inserts an intrinsic steady-state segment after this event to restore the fringe rhythm. The start and end times reserved in the extension region and the baseline latch time are incorporated into the extension entries of the intrinsic drift layer structure and are adjacent to the lightly doped termination transition zone segment label to limit the start timing of the epitaxial termination layer.

[0066] The epitaxial termination layer is constructed on the baseline reserved in the extension region, providing an epitaxial structural base for the terminal electric field spread of this invention. Specifically, the system reads the termination layer template, the segment label of the lightly doped termination transition zone, and the baseline reserved in the extension region from the intrinsic growth formula. These three elements jointly determine the activation time of the termination layer, the precursor combination, and the carrier gas swing strategy. When the termination layer is activated, the control unit performs short-term interface finishing, which is an extremely short steady state maintained by a low-level intrinsic precursor to eliminate the activity difference at the end of the extension region. After finishing, the precursor combination of the termination layer is introduced, and the slow swing rhythm of the carrier gas is set. The slow swing rhythm is phase-locked with the rhythm of the reflection stripes to form periodic micro-perturbations to avoid a flat boundary between the termination layer and the terminal intrinsic layer. During the growth of the termination layer, whenever a stable repetitive rhythm of the reflective fringes appears and the near-band edge signal of the emission sampling is stable, the system records this segment as a periodic unit of the termination layer. If there is an overlap between the periodic units where the cumulative curvature value rises and the spacing of the reflective fringes shrinks, the system inserts a low-level easing segment in the first half of the next periodic unit. After the easing segment is completed, the system returns to the termination layer template. The completion of the termination layer is marked by the periodic unit count reaching the template requirement and the difference between the emission sampling trajectory and the leader baseline converging. After reaching this mark, the construction of the termination layer ends. At the end, a short-term hydrogen atmosphere purge is performed on the top surface of the termination layer. The purge segment is used to remove weakly adsorbed molecules and stabilize the surface states. The start and end times of the termination layer, the periodic unit count, the easing segment insertion record, and the purge segment information are all written into the termination layer entry and incorporated into the version of the intrinsic drift layer structure to form the encapsulated structural description.

[0067] The operational and structural data of this step form a clear input-output link at the data layer. On the input side, the intrinsic growth recipe, thickness target, and low-leakage buffer structure are received from the upstream module. The low-leakage buffer structure, along with the three-channel alignment index, provides an anchor point for the timing of deposition and sub-processes. On the processing side, intrinsic gallium nitride drift layer deposition, annealing, hydrogen atmosphere replacement, light-doped termination transition band growth, extension region reservation, and epitaxial termination layer construction are executed sequentially. All sub-processes are judged based on the rhythm of the three channels, and events are recorded without immediate modification of the upper-layer recipe when anomalies occur, maintaining process continuity. On the output side, after completing the termination layer construction, this main step generates a structured description object with the output field name "Epithelial Termination Structure." This output field name contains information such as the intrinsic drift layer structure entry, the light-doped termination transition band segment label, the extension region reservation baseline, the termination layer entry, and the evidence chain dossier number, and is registered as a subscribing object on the data bus. In the cross-step transition, the extensional termination structure is called as an input item in the subsequent main step S4, used to establish a structural reference for recipe rewriting in online characterization and parameter linkage control. Simultaneously, the thermal history entries, abnormal event records, and sampling trajectories generated during this main step are subscribed to by the online module and used as evidence for deviation allocation results. The three-channel stage data left by the preceding main step is repeatedly referenced in this main step for interface continuity, window arrangement, and periodic unit identification. Furthermore, the extensional termination structure output by this main step is fed back into the subsequent recipe library rewriting and batch record generation, forming a closed-loop pathway from structure to recipe and back to structure.

[0068] Subsequently, step S4 is performed, including timing synchronization, noise filtering and threshold comparison using sliding median filtering and piecewise window function suppression, parameter linkage calculation including threshold comparison status table, source flow and temperature and pressure update, and formula library write-back to assemble deviation allocation results and read-back segments into change application entries, batch record generation and offline process triggering operation, generating an extended formula parameter set.

[0069] This main step receives data and structural products from the preceding main steps. The three signals originate from the online measurement channels of the superlattice buffer stack stage, the low-leakage buffer structure stage, and the epitaxial termination structure stage. The raw data for the three signals refers to the time-series variation trajectories of reflection intensity, curvature, and emission peak position and intensity, obtained under a unified acquisition framework. The alignment index is the cross-channel time anchor mapping table generated in the preceding steps. The batch event record is a compilation of time-series events and anomaly entries covering the entire process from loading to the termination layer construction. The evidence chain dossier number is the retrieval key associated with all process data and structural descriptions. Specifically, the reflection signal, curvature signal, emission signal, alignment index, and batch event record are used as input to trigger the timing synchronization process. Timing synchronization is defined as a set of actions that map the timestamps of each channel to a single main time axis and restore the sampling order without changing the original sampling content of the three signals. The synchronization start point selection strategy uses the first stable rhythm inflection point of the reflection stripe as the main anchor, and the curvature slope transition and the emission peak position stabilization window as auxiliary anchors. If the deviation between the auxiliary anchor and the main anchor exceeds the allowable band specified by the alignment index, a deviation entry is registered in the data layer and synchronization fine-tuning is initiated. Fine-tuning is accomplished by interpolating to fill short-term drop points and extracting equal-length windows before and after burst pulses, while archiving the fine-tuning window number in a one-to-one correspondence with the original window number. After timing synchronization is completed, the original data of the three signals with a unified time axis is obtained and recorded in the data layer as the three-signal stage alignment data. This object plays the role of the benchmark for judgment and calculation in subsequent processes.

[0070] In the noise filtering stage, the system performs multi-level processing against three common interference sources: electrical interference from power supply ripple and grounding loops, optical interference from light source drift and cavity window contamination, and mechanical interference from turntable micro-vibration and dynamic fluctuations in the pumping system. Specifically, the aligned data from the three signal stages first enters the front-end hardware calibration and baseline stabilization stage. Hardware calibration is achieved through calibration light source preheating, detector dark current level correction, and sampling channel gain consistency. Baseline stabilization is achieved through a carrier gas steady-state sampling segment and a short-time observation segment without a precursor. These two types of actions generate calibration segment entries and steady-state segment entries in the batch event record. Subsequently, the system enters the software denoising stage. The denoising process adopts a combination strategy of sliding median filtering and piecewise window function suppression. Sliding median filtering targets pulse spikes, while piecewise window function suppression targets slow drift. The window function length and weight values ​​are automatically derived based on the rhythm statistical labels of the alignment index, without involving formula descriptions. To monitor the stability of the denoising process, the system introduces a control chart framework based on Statistical Process Control (SPC). An outlier marker is generated at the end of each window and written into the noise profile list. If an outlier marker remains high in the same direction for several consecutive windows, it is defined as noise skewness, triggering a hardware inspection request. These actions form the noise-filtered data object, called the three-signal cleaned data, and establish a bidirectional traceability channel with the data aligned to the aforementioned three-signal stages. The traceability channel key value comes from the combination of the evidence chain file number and the window number, facilitating subsequent review of the original evidence during the parameter linkage calculation stage.

[0071] The threshold comparison process is initiated after the three-signal cleanup data is obtained. Threshold comparison involves comparing the key statistics of the current batch with the threshold library item by item and generating a set of actions for boundary judgment. The threshold library consists of two parts: a static part derived from the process access baseline, including the average rhythm band of reflective stripes, the cumulative curvature change band, the emission peak position drift band, and the intensity texture band; and a dynamic part derived from the temporary area and abnormal entries written in the previous main step, where dynamic entries reflect temporary biases caused by batch characteristics or material differences. Specifically, the system extracts the rhythm length sequence, slope sequence, peak position sequence, and intensity texture sequence from the three-signal cleanup data, forms the current batch comparison vector through windowed statistics, and maps this comparison vector to the corresponding entries in the threshold library. The mapping process calls the alignment index to complete the window position correspondence. If the mapping result shows a single item exceeding the boundary, the system generates a single item exceeding the boundary marker and writes it to the batch event record; if multiple items exceed the boundary simultaneously and the direction of the exceedance is the same, the system generates a coupled exceedance marker and adds a coupling description; if the directions of the exceedance conflict with each other, the system generates a conflict marker and enters a secondary review. The secondary verification process utilizes the sampling retest channel and the data self-consistency verification channel. The sampling retest channel involves short-term retesting within a gap that does not disrupt the continuity of the deposition. The data self-consistency verification channel performs differential consistency checks on key indicators of the three-signal purified data and the data before synchronization. After all comparisons are completed, the threshold comparison result object is obtained and recorded in the data layer as a threshold comparison status table. This table concatenates all markers and corresponding evidence keys according to window number, for querying during the next stage of parameter linkage calculation.

[0072] After the threshold comparison status table is formed, the parameter linkage calculation enters the execution state. The task of this step is to decompose the deviation information of the three-signal purification data and the threshold comparison status table into adjustable process variables and generate deviation allocation results. The deviation allocation results are the calculation intermediate objects of this step. The parameter linkage calculation includes three types of rules. The basic rules define the proportional relationship between the reflection stripe rhythm offset and the precursor source flow fine-tuning and turntable speed fine-tuning; the cumulative curvature change is mapped to the priority relationship between the reaction chamber pressure fine-tuning and the carrier gas ratio fine-tuning; and the emission peak position and intensity texture change are mapped to the linkage relationship between the heating power beat fine-tuning and the doping source flow temporary threshold. The adaptive rules are based on the deviation trajectory and stable section of historical batches. They use the online loop concept of Advanced Process Control (APC) to update the proportional weights and order. The update action only modifies the rule weights and does not change the main entries of the process formula. The conflict arbitration rule handles the mutual entanglement caused by multiple adjustments. The arbitration first reads the threshold comparison and the coupling description and offset mark in the status table, and then refers to the historical similar fragments corresponding to the evidence chain dossier number. If two rules give opposite suggestions for the same variable, a temporary frozen entry is generated and transferred to the manual review channel. Mutual cancellation actions are not performed within the same window. The calculation process proceeds within the instruction queue of the Distributed Control System (DCS). The queue is ordered by window number, and any window containing a freeze entry is skipped within the queue, with a skip entry added to the batch event log. After all rules are solved, the deviation allocation result is obtained. The deviation allocation result includes variable names, adjustment amplitude levels, duration levels, and execution order labels, without involving formula expressions, and establishes a reference relationship with the three-signal cleanup data.

[0073] After the deviation allocation results are sent to the equipment side, the source flow and temperature / pressure updates enter the action state. This stage completes the mapping and execution of parameters from the computational domain to the physical domain, and performs readback verification of the execution process. Specifically, source flow updates refer to the coordinated adjustment of the duty cycle of the precursor valve group and the ratio of evaporation source heating power to carrier gas. Temperature / pressure updates refer to the coordinated adjustment of the heating power curve and the reaction chamber pressure setpoint. Both follow a progressive update strategy, which consists of three stages: a trial stage to establish the action response direction, a confirmation stage to fix the action level, and a readback stage to evaluate the changes in the three signals brought about by the action on a unified time axis. At the start of the action, the control unit sends the minimum level adjustment command in the trial phase and monitors the response of the three-signal purification data. If the response direction is consistent with the deviation allocation result, it enters the confirmation phase. If the direction is inconsistent, a reverse response entry is recorded and the variable weights are slightly adjusted downward in the calculation domain. The confirmation phase executes level fixation and maintenance. If the threshold comparison status table still shows out of bounds in the adjacent window during the maintenance period, the control unit extends the maintenance and requests a higher level deviation allocation in the next calculation cycle. The readback phase starts after the maintenance ends. During the readback period, the rhythm, slope, and peak of the three signals are packaged into a readback segment, which is written into the deviation allocation result object as closed-loop evidence. If the action fails or the device returns a rejection in any window, the control unit generates an execution failure entry and switches to the safety mode. At the same time, the failure reason and device return code are recorded in the batch event log. The safety mode does not change the deposition continuity, only freezes the current variable, and other variables are executed normally.

[0074] Recipe library write-back and batch record generation proceed in parallel after source flow and temperature / pressure updates enter a stable phase. The recipe library is a structured storage entity that stores process parameter sets, version relationships, and approval trajectories. Each recipe entry consists of parameter key-value pairs, boundary labels, and applicable scope labels. Version relationships employ a parallel strategy of chained numbering and branch numbering. The approval trajectories record the responsible parties and timestamps for compilation, review, and approval. Specifically, the system assembles the deviation allocation results and the final status of the read-back segment and threshold comparison status table into a change request entry. The change request entry enters the pending review area of ​​the recipe library, triggering a two-level review process. The first-level review involves the process responsible party confirming the entry's correctness, while the second-level review involves the quality responsible party confirming the completeness of the evidence chain. After the review is passed, a draft version of the extended recipe parameter set is generated, and the draft status is marked as readable upon the next startup within the recipe library. The batch record generation aggregates all data objects for this batch, from loading to termination layer construction, linkage control, and write-back triggering. The record structure includes a batch overview page, a process event page, a data index page, and an evidence chain page. The overview page lists key entries and version numbers; the process event page connects all time-sequential events; the data index page provides retrieval keys for the three-signal purification data, read-back segments, and threshold comparison status table; and the evidence chain page binds external metrology reports and sampling measurement numbers. After the recipe library write-back is completed, the work order extended fields of the Manufacturing Execution System (MES) are updated synchronously, and the new version number is pushed to the process scheduler's startup list as the source for reading the next wafer.

[0075] The offline process is triggered when the recipe library is written back and all actions within the deviation allocation results are in the completed state. This step sends a unified trigger signal to the cooling, stress release, and surface cleaning modules and registers the trigger key in the data layer. The trigger conditions include two factors: the external termination structure item reaches the completion mark, and there are no unclosed frozen items in the batch event record. When both conditions are met, an offline trigger order is generated, and an unlock signal is formed at the equipment layer. The unlock signal is transmitted to the cooling, stress release, and surface cleaning modules, which then perform actions according to their local recipes. The start and end times of these actions are written back to the batch record. If the offline trigger is delayed, the system generates a delay entry in the batch record, indicating that the delay is due to equipment occupancy or safety inspection. Once the delay is lifted, the trigger order automatically takes effect without manual re-issuance. To support subsequent cross-batch review, the system also packages all key objects of this main step into an archive package. The archive package includes three-signal stage alignment data, three-signal purification data, threshold comparison status table, deviation allocation results, readback segments, change application items, and new version numbers. The archive package number is synchronized with the evidence chain dossier number and can be retrieved and compared at the data layer.

[0076] At the cross-step connection level, the output of this main step forms a two-way path forward and backward in both structural and data dimensions. In terms of structure, after completing parameter linkage calculation, source flow and temperature and pressure update, recipe library write-back, batch record generation, and offline process triggering, the system generates a structured parameter set object. The output field name is epitaxial recipe parameter set. This object is read when the process returns to the nucleation stage. Subsequent main steps complete the input loading in the epitaxial recipe parameter set in S1 and enter the nucleation interface for purification and temperature ramp control, thus forming a closed loop from process control to recipe reconstruction and back to front-end construction. In terms of data, the three-signal purification data and threshold comparison status table and readback segments exist as the prior inventory of the three-signal stage before the next start. They are used as rhythm anchors and deviation weight calculation entries, and maintain a stable mapping with the evidence chain dossier number of the superlattice buffer stack, low-leakage buffer structure, and epitaxial termination structure. Any node can be traced back to the upstream structural description and downstream parameter change history. This concludes the complete process description of the main step, from input source to processing link to output product. The epitaxial formulation parameter set is registered as a subscribed object in the data layer, and will be read and injected into the preceding main step by the process scheduler when the next batch starts.

[0077] Example 2

[0078] like Figure 2 As shown in the figure, an embodiment of the present invention discloses a gallium nitride-based diode epitaxial device with reverse breakdown voltage enhancement, comprising:

[0079] The timing synchronization module 101 is used to map the reflected signal, curvature signal, and emission signal to a single main time axis and restore the sampling order without changing the original sampling content of the three signals. It registers the deviation entries in the data layer and outputs the three-signal stage alignment data to the noise filtering and baseline stabilization module. Specifically, it receives the reflected signal, curvature signal, emission signal, and their timestamp and sampling order information. According to the preset main time axis rules, it performs timestamp standardization, duplicate point folding, missing point filling, burst segment pruning, and sequence rearrangement to construct a unified timing structure. Based on the preset anchor point strategy, it identifies the deviation between the main anchor and the auxiliary anchor to trigger fine-tuning. The fine-tuning stage performs short-term interpolation and equal-length window truncation to generate deviation entries and register their numbers and source positions in the data layer. The three-signal stage alignment data formed after unified processing is output to the noise filtering and baseline stabilization module as input. At the same time, it retains the deviation entries and alignment position relationship for the threshold comparison module to review and verify in subsequent stages.

[0080] The noise filtering and baseline stabilization module 102 is used to perform front-end hardware calibration and baseline stabilization, sliding median filtering and piecewise window function suppression on the three-signal stage alignment data. Based on the rhythm statistical label of the alignment index, it derives parameters to generate three-signal clean data and provides the three-signal clean data and noise profile list to the threshold comparison module. Specifically, it receives the three-signal stage alignment data and the corresponding alignment index, first performs the front-end hardware calibration and baseline stabilization process. The calibration stage completes the zero-point reset and gain consistency of the sampling channel. The baseline stage arranges blank observation segments and steady-state observation segments and records the baseline drift trajectory. Then, it performs sliding median filtering to process pulse spikes and piecewise window function suppression to process slow drift. In the parameter derivation stage, it selects the window length and weight according to the rhythm statistical label of the alignment index. At the boundary, it triggers the transition segment splicing strategy. When there is an anomaly, it registers the calibration segment entry and the steady-state segment entry and sends back the source information. After the above processing, the three-signal clean data and noise profile list are obtained. Both are called by the threshold comparison module, and the temporal position relationship of the three-signal clean data is mapped with the upstream three-signal stage alignment data for the data layer traceability and evidence chain module to record.

[0081] The threshold comparison module 103 is used to extract rhythm length sequence, slope sequence, peak position sequence, and intensity texture sequence from the three-signal cleaned data. It performs item-by-item comparisons according to the static and dynamic parts of the threshold library, forming boundary crossing markers, coupling descriptions, and conflict markers, generating a threshold comparison status table, and providing the threshold comparison status table to the parameter linkage calculation and conflict arbitration module. Specifically, it receives the three-signal cleaned data and a noise profile list, extracts the rhythm length sequence, slope sequence, peak position sequence, and intensity texture sequence according to window order, and establishes corresponding entries with the static and dynamic parts of the threshold library. The system records the direction and magnitude of the boundary crossing when comparing items one by one. When multiple trends in the same direction appear, a coupling description is generated. When mutual cancellation trends appear, an offsetting mark is generated and a review subprocess is triggered. The review subprocess performs differential verification on the segments before and after the same window and registers the review items. After all windows are processed, the data is summarized into a threshold comparison status table. The boundary crossing mark, coupling description and offsetting mark are connected in sequence according to time. The threshold comparison status table is output to the parameter linkage calculation and conflict arbitration module. At the same time, a bidirectional reference key for the three-signal purification data is attached to the data layer for the data layer traceability and evidence chain module to record.

[0082] The parameter linkage calculation and conflict arbitration module 104 is used to decompose the deviation information in the threshold comparison status table and the three-signal purification data into adjustable process variables such as precursor source flow, carrier gas ratio, reaction chamber pressure, and heating power according to basic rules, adaptive rules, and conflict arbitration rules. It generates deviation allocation results and sends them to the source flow and temperature and pressure update modules. Specifically, it receives the threshold comparison status table and the three-signal purification data, establishes the mapping relationship between rhythm offset and precursor source flow and rotation beat according to basic rules, establishes the priority order of curvature change and reaction chamber pressure and carrier gas ratio according to basic rules, establishes the linkage relationship between peak position and texture change and heating power and doping threshold according to basic rules, adjusts the weights and execution order of each mapping according to adaptive rules, and calls the conflict arbitration rules to generate frozen entries and mark the frozen window when mutual entanglement occurs. Frozen entries do not perform conflict actions in this window. The deviation allocation results obtained after decomposition contain variable names, amplitude levels, duration periods, and sequence labels. The deviation allocation results are sent to the source flow and temperature and pressure update modules, and the frozen entries and out-of-bounds sources are backfilled to the data layer for subsequent recording.

[0083] The source flow and temperature / pressure update module 105 is used to progressively update the pre-valve group duty cycle, evaporator heating power, carrier gas ratio, and reaction chamber pressure setpoint based on the deviation allocation results. It proceeds through trial, confirmation, and readback phases, generating readback segments, execution failure entries, and safety gear records. The readback segments are then returned to the parameter linkage calculation and conflict arbitration module. Specifically, it receives the deviation allocation results, issues a minimum amplitude command in the trial phase, and monitors the response direction of the three-signal purification data. If the response direction matches the label, it enters the confirmation phase and locks the gear; if they do not match, it registers a reverse response entry and requests upstream weight adjustment. The confirmation phase maintains the gear across several windows; if it still exceeds the limit, it extends the maintenance and requests a higher amplitude. After completion, it enters the readback phase, which aggregates response segments of rhythm, slope, and peak position to form readback segments. It also records execution failure entries and safety gear records. The readback segments are returned to the parameter linkage calculation and conflict arbitration module for verification. Related records are collected by the formula library write-back and batch record generation module.

[0084] The formula library write-back and batch record generation module 106 is used to assemble the deviation allocation results and the read-back fragments and threshold comparison status table into change application entries, enter the formula library pending review area, generate an extended formula parameter set version draft, and simultaneously aggregate the batch overview page, process event page, data index page, and evidence chain page to form a batch record, and output the version number and batch status to the downstream process triggering module. Specifically, it receives the deviation allocation results, read-back fragments, and threshold comparison status table, completes the binding of change application entries and the attachment of source keys, and sends them to the formula library pending review area for hierarchical review. After the review is passed, it generates an extended formula parameter set version draft and registers the version number; it also aggregates the batch overview page, process event page, data index page, and evidence chain page in parallel to form a batch record and records the status. The version number and batch status are transmitted to the downstream process triggering module, and the index key set is submitted to the data layer traceability and evidence chain module for subsequent retrieval.

[0085] The offline process trigger module 107 is used to generate an offline trigger order when the extension termination structure item reaches the completion mark and there are no unclosed frozen items in the batch event record. It sends an unlock signal to the cooling, stress release and surface cleaning module and writes the start time and end time into the batch record. Specifically, it receives the version number and batch status output by the recipe library write-back and batch record generation module, monitors the status of the extension termination structure item and the closed status of the frozen items in the batch event record, issues an offline trigger order when the trigger conditions are met, sends an unlock signal to the cooling, stress release and surface cleaning module, records the start time and end time to the batch record in local order, and transmits the trigger order and time record synchronously to the data layer traceability and evidence chain module to form a correspondence with the version number.

[0086] The data layer traceability and evidence chain module 108 is used to maintain the mapping relationship between batch event records and evidence chain file numbers, establish a bidirectional traceability channel between three-signal stage alignment data, three-signal purification data, threshold comparison status table, deviation allocation results and batch records, and register the extended formula parameter set as a subscribing object for subsequent steps to read. Specifically, it receives batch event records and evidence chain file numbers, registers the mapping relationship and saves the source path, establishes a bidirectional traceability channel around three-signal stage alignment data, three-signal purification data, threshold comparison status table, deviation allocation results and batch records, and saves the time position key, source key and version number in the channel. The query action horizontally connects all objects. After registration, the extended formula parameter set is registered to the data layer directory as a subscribing object, providing an extended formula parameter set reading channel for subsequent steps, and recording the subscription behavior and reading time in the batch record, forming a closed loop from parameter generation to reading.

[0087] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A method for epitaxial growth of gallium nitride-based diodes with improved reverse breakdown voltage, characterized in that, include: The substrate to be epitaxially grown is loaded, and the substrate parameters, epitaxial formulation parameter set, and temperature ramp curve corresponding to the substrate are obtained. Online monitoring with three channels including reflection channel, curvature channel, and light emission channel is performed. Degassing and surface reconstruction, nucleation heating, and deposition of a graded aluminum content transition layer using a continuous transition strategy of aluminum composition from high to low or from low to high are performed. The superlattice periodic stacking process, in which two or more types of sublayers are deposited alternately in a fixed or quasi-fixed order, generates a superlattice buffer stack. Among them, the online monitoring with three channels means collecting data according to a unified clock process and forming an alignment index in the data acquisition module. The alignment index is implemented by using the phase marker of the reflection stripes as the main time anchor and projecting the curvature slope change points and the inflection points of the emission peak position drift onto the main time axis. Once the projection error exceeds the threshold, the system calls the data repair function. Data repair is defined as generating interpolation segments and label segments for each abnormal window without changing the original record, and attaching the repair segment number to the batch event record. The three-channel alignment index corresponds to a unified time axis for the three signals, including the reflection signal, curvature signal, and emission signal. The background impurity suppression process involves setting the source flow, arranging and segmenting the supply window under dual references of time axis and growth depth, continuous transition and synergistic integrated measures for gas path and cavity, and generating a low-leakage buffer structure. The process involves depositing an intrinsic gallium nitride drift layer, performing annealing and hydrogen replacement triggered at specified progress points, growing a lightly doped termination transition band by extracting the final progress point sequence from the intermediate state of the intrinsic drift layer structure, reserving an extension region, and constructing an epitaxial termination layer to obtain an epitaxial termination structure. The system performs time synchronization of three signals, noise filtering and threshold comparison using sliding median filtering and piecewise window function suppression, parameter linkage calculation including threshold comparison status table, source flow and temperature and pressure update, and recipe library write-back to assemble deviation allocation results and read-back segments into change application entries, batch record generation and offline process triggering operation to generate an extended recipe parameter set.

2. The method for epitaxial growth of gallium nitride-based diodes with improved reverse breakdown voltage according to claim 1, characterized in that, The substrate parameters, epitaxial formulation parameter set, and temperature ramp curve include: Substrate parameters include crystal orientation, surface roughness, surface defect density, and contaminant spectrum; The epitaxial formulation parameter set includes source flow settings, reaction chamber pressure range, carrier gas ratio, heating power cycle time, rotation speed range, and alarm threshold. A temperature ramp curve is a continuous segment describing a heating section, an isothermal section, and a cooling section.

3. The method for epitaxial growth of gallium nitride-based diodes with reverse breakdown voltage enhancement according to claim 2, characterized in that, Timing synchronization includes: Without changing the original sampling content of the three signals, the action set that maps the timestamps of each channel to a single main time axis and restores the sampling order is used; the synchronization start point selection strategy adopts the first stable rhythm inflection point of the reflection stripe as the main anchor, and the curvature slope inflection and the luminous peak position stabilization window as the auxiliary anchor.

4. The method for epitaxial growth of gallium nitride-based diodes with reverse breakdown voltage enhancement according to claim 3, characterized in that, Noise filtering employs a combination strategy of sliding median filtering and piecewise window function suppression. Sliding median filtering targets pulse spikes, while piecewise window function suppression targets slow drifts. The window function length and weight values ​​are automatically derived based on the rhythm statistical labels of the aligned index.

5. The method for epitaxial growth of gallium nitride-based diodes with reverse breakdown voltage enhancement according to claim 4, characterized in that, The parameter linkage calculation includes three types of rules: the basic rule defines the ratio of the reflection stripe rhythm offset to the precursor source flow fine-tuning and the turntable speed fine-tuning; the curvature cumulative change is mapped to the priority relationship between the reaction chamber pressure fine-tuning and the carrier gas ratio fine-tuning; and the emission peak position and intensity texture change are mapped to the linkage relationship between the three-signal purification data and threshold comparison status table of the heating power beat fine-tuning and the doping source flow temporary threshold. The adaptive rules update the proportional weights and order of the deviation trajectories and stable segments based on the historical batches. The conflict arbitration rules address the interconnectedness resulting from multiple mediations.

6. The method for epitaxial growth of a gallium nitride-based diode with improved reverse breakdown voltage according to claim 5, characterized in that, The source flow and temperature / pressure updates include: a trial phase for establishing the direction of the action response, a confirmation phase for fixing the action level, and a readback phase for evaluating the changes in the three signals brought about by the action on a unified time axis.

7. The method for epitaxial growth of gallium nitride-based diodes with reverse breakdown voltage enhancement according to claim 6, characterized in that, The recipe library write-back process includes: assembling the deviation allocation results and the final state of the read-back segment and the threshold comparison status table together into a change application entry. The change application entry enters the pending area of ​​the recipe library to trigger a two-level review process. After the review is passed, a new version of the extended recipe parameter set version draft is generated.

8. The method for epitaxial growth of gallium nitride-based diodes with reverse breakdown voltage enhancement according to claim 7, characterized in that, The process of triggering the offline operation also includes: the external termination structure entry reaches the completion mark, and there are no unclosed frozen entries in the batch event record; when both conditions are met, an offline trigger order is generated and an unlock signal is formed at the equipment layer.

9. The method for epitaxial growth of a gallium nitride-based diode with reverse breakdown voltage enhancement according to claim 8, characterized in that, The generated epitaxial formulation parameters are set, and the output product forms a bidirectional pathway in both structural and data dimensions, moving forward and backward. In terms of structure, the generated structured parameter set object is output as an extensional recipe parameter set, which is read when the process returns to the nucleation stage; in terms of data, the three-signal purification data and threshold comparison status table and readback fragment exist as the prior inventory of the three-signal stage before the next startup.

10. A gallium nitride-based diode epitaxial device with reverse breakdown voltage boosting, applied to the method of any one of claims 1-9, characterized in that, include: The timing synchronization module is used to timestamp and map the reflected signal, curvature signal and light emission signal to a single main time axis without changing the original sampling content of the three signals, and restore the sampling order. It also registers the deviation entries to the data layer and outputs the three-signal stage alignment data to the noise filtering and baseline stabilization module. The noise filtering and baseline stabilization module is used to perform front-end hardware calibration and baseline stabilization, sliding median filtering and piecewise window function suppression on the three-signal stage alignment data. It derives parameters based on the rhythm statistical labels of the alignment index, generates three-signal cleaned data, and provides the three-signal cleaned data and noise profile list to the threshold comparison module. The threshold comparison module is used to extract rhythm length sequence, slope sequence, peak position sequence and intensity texture sequence from the three-signal clean data, perform item-by-item comparison according to the static and dynamic parts of the threshold library, form out-of-bounds markers, coupling descriptions and conflict markers, generate a threshold comparison status table, and provide the threshold comparison status table to the parameter linkage calculation and conflict arbitration module. The parameter linkage calculation and conflict arbitration module is used to decompose the deviation information in the threshold comparison status table and the three-signal purification data into precursor source flow, carrier gas ratio, reaction chamber pressure and heating power according to the basic rules, adaptive rules and conflict arbitration rules, generate deviation allocation results, and send the deviation allocation results to the source flow and temperature and pressure update module. The source flow and temperature / pressure update module is used to perform progressive updates of the precursor valve group duty, evaporation source heating power, carrier gas ratio and reaction chamber pressure setpoint according to the deviation allocation results. It advances the action according to the trial segment, confirmation segment and readback segment, forming readback segments, execution failure entries and safety gear records, and returns the readback segments to the parameter linkage calculation and conflict arbitration module. The recipe library write-back and batch record generation module is used to assemble the deviation allocation results and read-back segments and threshold comparison status table into change application entries, enter the recipe library pending review area, generate the extended recipe parameter set version draft, and aggregate the batch overview page, process event page, data index page and evidence chain page to form batch records, and output the version number and batch status to the downstream process trigger module. The offline process trigger module is used to generate an offline trigger order when the extension termination structure item reaches the completion mark and there are no unclosed frozen items in the batch event record, send an unlock signal to the cooling and stress release and surface cleaning module, and write the start time and end time into the batch record. The data layer traceability and evidence chain module is used to maintain the mapping relationship between batch event records and evidence chain case file numbers, and to establish a two-way traceability channel between three-signal stage aligned data, three-signal cleaned data, threshold comparison status table, deviation allocation results and batch records.

Images