A method and system for adaptive control of silicon carbide etching sidewall steepness

Abstract

The application discloses a silicon carbide etching sidewall steepness self-adaptive regulation method and system, belongs to the field of semiconductor micro-nano processing technology, and aims to solve the problems of high artificial dependence, low precision and inability to adapt to long-term working condition changes of existing silicon carbide etching sidewall steepness regulation. The method comprises five steps of regulation benchmark system construction, abnormality monitoring and classification, target regulation parameter group matching and quantization, regulation instant verification and iteration, global effect verification and benchmark self-evolution. The application realizes the continuous self-evolution of the regulation benchmark through the double feedback dynamic updating mechanism combining instant link feedback and final global feedback, significantly improves the consistency and one-time regulation success rate of the silicon carbide etching sidewall steepness, greatly reduces the frequency of manual intervention, and effectively adapts to long-term working condition changes such as equipment aging, chamber state drift and substrate batch difference.

Description

Technical Field

[0001] This invention relates to the field of semiconductor micro-nano fabrication technology, and more specifically to a method and system for adaptively controlling the steepness of silicon carbide etching sidewalls. Background Technology

[0002] Silicon carbide, as a core material for third-generation semiconductors, has become a key basic material for manufacturing high-end electronic devices such as power devices, radio frequency devices, and AR waveguide chips due to its excellent properties such as wide bandgap, high breakdown electric field, and high thermal conductivity. Inductively coupled plasma dry etching is the core process for achieving high-precision micro-nano patterning of silicon carbide wafer surfaces. Among these processes, the steepness of the etching sidewalls is a core indicator for measuring etching quality, and its accuracy directly determines the electrical performance, long-term reliability, and mass production yield of the devices.

[0003] Currently, quality control in silicon carbide etching processes still heavily relies on human experience. When anomalies occur during processing, causing sidewall steepness to deviate from the target range, process engineers must manually adjust process parameters based on experience. This traditional control method has significant limitations: manual adjustment depends on individual technical skills and experience, leading to significant differences in adjustment results among different engineers, making it difficult to guarantee consistency in mass production. Furthermore, the adjustment process is time-consuming and labor-intensive, severely restricting production efficiency. Simultaneously, manual adjustment is mostly a reactive approach, unable to identify and intervene in the early stages of anomalies, easily leading to the gradual accumulation and amplification of abnormal errors, ultimately resulting in wafer scrap. In addition, existing process parameters are mostly static parameters calibrated in small-batch laboratory experiments, unable to adapt to long-term operating condition fluctuations such as equipment aging, changes in chamber conditions, and substrate batch differences. As the production cycle lengthens, process stability continuously declines, leading to increased fluctuations in sidewall steepness and making it difficult to maintain a high production yield. Therefore, to overcome these limitations, this invention proposes an adaptive control method and system for silicon carbide etching sidewall steepness. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an adaptive control method and system for the steepness of silicon carbide etching sidewalls, which solves the problems of high manual dependence, low precision, and inability to adapt to long-term operating condition changes in silicon carbide etching sidewall steepness control, thus leading to insufficient long-term production stability and yield.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An adaptive control method for the sidewall steepness of silicon carbide etching includes: S1: Based on the control parameters and monitoring indicators of each processing stage of silicon carbide etching, a control benchmark system for each processing stage is pre-constructed; S2: Collect monitoring index values ​​for each processing stage, identify abnormal monitoring indexes for abnormal processing stages, classify abnormal categories, and generate the actual abnormality degree of abnormal monitoring indexes. S3: Combining the anomaly type and actual anomaly degree of the anomaly monitoring indicators, based on the control benchmark system of the anomaly processing link, match the target control parameter group of the anomaly processing link, quantify the actual control degree of each control parameter in the target control parameter group, and generate control instructions. S4: After the control command is executed and the preset process stabilization time has elapsed, collect the abnormal monitoring index value, determine whether the control is effective immediately, if so, quantify the effective response parameter, if not, trigger the control update. S5: In response to the immediate effect of regulation, track the sidewall steepness index after processing, determine whether the regulation is globally effective, and if so, quantify the priority index of the target regulation parameter group, and then update the regulation priority and regulation benchmark system of the target regulation parameter group.

[0006] Specifically, the control benchmark system includes a set of control parameter groups for each abnormality type associated with each monitoring indicator, the initial control priority of each control parameter group within the set of control parameter groups, and a mapping table between the abnormality degree and the basic control degree of each control parameter group. The set of control parameters consists of multiple control parameter groups, each of which refers to a parameter combination that includes at least one control parameter and its corresponding adjustment direction.

[0007] Specifically, monitoring index values ​​are collected for each processing stage, abnormal monitoring indicators for abnormal processing stages are identified, and abnormality categories are classified. The actual abnormality degree of the abnormal monitoring indicators is generated, including: Collect all monitoring index values ​​of each processing stage, bind them one by one with the identity of the processed object, and record the collection time, processing stage number and equipment number to form a complete processing data chain. The collected monitoring index values ​​are compared point by point with the corresponding benchmark range of the monitoring index. If the monitoring index value of any monitoring index in a certain processing link exceeds the benchmark range for a preset number of consecutive times, the processing link is determined to be an abnormal processing link, and the monitoring index that exceeds the benchmark range is marked as an abnormal monitoring index. Extract the abnormal deviation characteristics of abnormal monitoring indicators, match them with the abnormal types corresponding to each set of control parameter groups in the control benchmark system, and classify the abnormal types of the current abnormal monitoring indicators. Based on the benchmark range of the anomaly monitoring indicators, the actual anomaly quantification calculation is performed on the identified anomaly monitoring indicators.

[0008] Specifically, the steps of matching the target control parameter set for abnormal processing stages, quantifying the actual control degree of each control parameter in the target control parameter set, and generating control instructions include: Based on the anomaly type identifier, the set of control parameter groups corresponding to the anomaly type is retrieved from the control benchmark system; the target control parameter group is selected according to the control priority of each control parameter group in the control parameter group set. Based on the actual anomaly degree, the anomaly degree interval to which the actual anomaly degree belongs is determined from the anomaly degree to basic control degree mapping table corresponding to the target control parameter group, and the basic control degree corresponding to the anomaly degree interval is obtained; the actual control degree of each control parameter in the target control parameter group is calculated by linear interpolation. The target control parameter set and the actual control degree of each control parameter are converted into control commands that can be recognized and executed by the corresponding equipment in the abnormal processing stage, and then sent to the abnormal processing stage.

[0009] Specifically, determining whether the regulation takes effect immediately includes: After the control command is executed and the preset process stabilization time has elapsed, obtain the abnormal monitoring index values ​​for a preset number of times and the monitoring index values ​​associated with the abnormal processing link. The collected abnormal monitoring index values ​​are compared point by point with the corresponding benchmark range of the monitoring index. If the monitoring index values ​​of the abnormal monitoring index are within the corresponding benchmark range for a consecutive preset number of times, and the monitoring index values ​​of the monitoring index associated with the abnormal processing link are within the corresponding preset fluctuation range for a consecutive preset number of times, then the current control is deemed to be effective immediately; otherwise, the current control is deemed to be invalid immediately.

[0010] Specifically, the steps for quantifying the priority indicators of the target control parameter group and then updating the control priority and control benchmark system of the target control parameter group include: Based on the effective response parameters and sidewall steepness index values ​​of this regulation, the single priority index score of the target regulation parameter group corresponding to this regulation is calculated, and the priority index of the target regulation parameter group is updated by the sliding weighted average method. After each preset update cycle of the same type of abnormality is controlled, the priority indicators of all control parameter groups corresponding to the abnormality type are sorted from high to low, and the control priority of each control parameter group is redistributed. At the same time, based on the global effective sample dataset of all regulation within this update cycle, the mapping table between the abnormality degree and the basic regulation degree of each regulation parameter group is simultaneously optimized.

[0011] Specifically, a silicon carbide etching sidewall steepness adaptive control system includes: The control benchmark system construction module is used to pre-build the control benchmark system for each processing step based on the control parameters and monitoring indicators of each processing step in silicon carbide etching. The anomaly monitoring module is used to collect monitoring index values ​​of each processing stage, identify abnormal monitoring indexes of abnormal processing stages, classify anomalies, and generate the actual anomaly degree of the anomaly monitoring indexes. The control decision module is used to combine the anomaly type and actual anomaly degree of the anomaly monitoring indicators, match the target control parameter group of the anomaly processing link based on the control benchmark system of the anomaly processing link, quantify the actual control degree of each control parameter in the target control parameter group, and generate control instructions. The control verification module is used to collect abnormal monitoring index values ​​after the control command is executed and a preset process stabilization time has elapsed, and to determine whether the control is effective immediately. If so, the effective response parameters are quantified; otherwise, the control is updated. The baseline self-evolution module responds to the immediate effect of regulation, tracks the sidewall steepness index after processing, determines whether the regulation is globally effective, and if so, quantifies the priority index of the target regulation parameter group, and then updates the regulation priority and regulation baseline system of the target regulation parameter group.

[0012] The beneficial effects of this invention are: This invention pre-constructs a control benchmark system covering the entire silicon carbide etching process, solidifies experimentally verified effective control schemes, and replaces traditional trial-and-error based on manual experience. This fundamentally avoids conflicts arising from multi-parameter coupled control, significantly reducing reliance on manual intervention. Simultaneously, it extracts multi-dimensional abnormal deviation features linked to benchmark offset, time-domain changes, and related indicators, and performs precise classification. Combined with dimensionless anomaly quantification, it achieves accurate anomaly identification and unified characterization of severity, ensuring precise adaptation of subsequent control schemes to the causes of anomalies. Furthermore, based on control priorities, it matches target control parameter sets and quantifies the actual control degree, achieving precise control of the control amount. To avoid over- or under-regulation, a mechanism for real-time regulation effect verification and automatic iterative switching is set up to quickly detect regulation failures and adjust the scheme, effectively preventing the amplification of abnormal errors step by step. Finally, a dual feedback mechanism combining real-time process feedback and final global feedback is adopted to dynamically update the priority and abnormality-basic regulation degree mapping table of the regulation parameter group, so as to realize the continuous self-evolution of the regulation benchmark system. It can adapt to long-term operating condition changes such as equipment aging, chamber state drift, and substrate batch differences, significantly improving the consistency of silicon carbide etching sidewall steepness and the first-time regulation success rate, effectively improving the stability and yield of mass production. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for adaptively controlling the sidewall steepness of silicon carbide etching according to the present invention; Figure 2 This is a flowchart of step S2 of the present invention; Figure 3 This is a flowchart of step S3 of the present invention; Figure 4 This is a flowchart of step S4 of the present invention. Detailed Implementation

[0014] Example 1 Please see Figure 1 This embodiment introduces an adaptive control method for the steepness of silicon carbide etching sidewalls, the method specifically including the following steps: Step S1: Taking the target etching sidewall steepness as the final convergence target, based on the control parameters and monitoring indicators of each processing stage of silicon carbide etching, a control benchmark system for each processing stage is pre-constructed. The control benchmark system includes a set of control parameter groups under each anomaly type associated with each monitoring indicator, the initial control priority of each control parameter group within the control parameter group set, and a mapping table between the anomaly degree and the basic control degree of each control parameter group. Specifically, a silicon carbide etching production line refers to a semiconductor micro / nano fabrication production line with inductively coupled plasma (ICP) dry etching as its core process. It is used to achieve high-precision micro / nano patterning of silicon carbide wafer surfaces, providing core process support for the manufacturing of silicon carbide power devices, AR waveguide chips, and radio frequency devices. The production line includes mask preparation, pre-etching pretreatment, ICP dry etching, and post-etching processing. Control parameters refer to the quantitatively adjustable process control parameters in each processing step, serving as direct variables for controlling processing quality. For example, when the processing step is ICP dry etching, the control parameters include SF6 flow rate, O2 flow rate, C4F8 flow rate, ICP source power, substrate bias power, chamber pressure, and substrate temperature. When the processing step is mask preparation, the control parameters include photolithography exposure dose, development time, hard mask etching power, and hard mask etching time. Monitoring indicators refer to quantifiable and detectable parameters used to characterize the processing quality of each processing stage. They are the core basis for judging whether a processing stage is abnormal. For example, when the processing stage is mask preparation, the monitoring indicators include mask sidewall angle, mask linewidth deviation, and mask edge roughness. When the processing stage is ICP dry etching, the monitoring indicators include plasma F / C ratio, etching rate, substrate temperature uniformity, and sidewall passivation layer thickness. In addition, monitoring indicators are obtained through online detection devices deployed at the inlet and outlet of each processing stage, including online scanning electron microscopes (SEM), optical emission spectrometers (OES), infrared thermometers, contact angle meters, etc.

[0015] Furthermore, the set of control parameter groups refers to the set of all preset control parameter groups that can bring the abnormal monitoring index back to the benchmark range for a certain type of abnormality of a certain monitoring indicator. It consists of multiple independent control parameter groups with different control logics. Each control parameter group refers to a parameter combination that includes at least one control parameter and its corresponding adjustment direction, including single-parameter control groups, dual-parameter control groups and multi-parameter control groups. Its setting is based on the results of multiple sets of DOE orthogonal calibration etching experiments on silicon carbide standard wafers, and all parameter combinations that can effectively correct the corresponding abnormality without fatal process side effects are selected.

[0016] Furthermore, the initial control priority of each control parameter group within the control parameter group set refers to the static calling order determined based on the experimental control effect during the initial calibration stage for each control parameter group under the same anomaly type. This order is used to prioritize the control parameter group with better control effect when an anomaly occurs. The setting is based on the control success rate, convergence speed score, and global effect comprehensive score of each control parameter group in the initial calibration experiment. The higher the comprehensive score, the smaller the control priority value. Control priority 1 is the highest priority, and the larger the value, the lower the priority.

[0017] Furthermore, anomaly rate refers to a dimensionless index used to quantify the degree to which a monitoring indicator deviates from its baseline range. It is used to standardize the severity of anomalies in monitoring indicators of different dimensions and magnitudes. Its setting is based on the preset baseline range of the corresponding monitoring indicator, and the calculation formula is:

[0018] The allowable deviation of the reference range is the absolute value of the difference between the upper and lower limits of the reference range. The reference range of each monitoring indicator is set based on the target requirements for the final sidewall steepness of silicon carbide etching, device performance and yield thresholds, process stability boundaries, cumulative error tolerance of the entire process, industry-standard mass production standards and historical mass production data of the enterprise. In addition, the reference range of all monitoring indicators has been confirmed by multiple sets of repeated verification experiments on standard silicon carbide wafers to ensure that the processing quality within the reference range will not cause the final etched sidewall steepness to exceed the target requirements.

[0019] Furthermore, the baseline control degree refers to the adjustment amount of the control parameter corresponding to a unit anomaly degree. It is used to calculate the precise parameter adjustment amount based on the actual anomaly degree, avoiding over-control or under-control. Its setting is based on the gradient anomaly calibration experiment of silicon carbide standard wafers. By conducting multiple sets of repeated control experiments under different anomaly degree gradients, the baseline control degree values ​​of each control parameter in different anomaly degree intervals are fitted. The unit anomaly degree refers to the dimensionless standardized anomaly benchmark unit, defined as the critical anomaly state with an anomaly degree of 1, that is, the minimum exceeding state where the measured value of the monitoring indicator just touches the upper or lower limit of the benchmark range. Its setting is based on the benchmark range of the corresponding monitoring indicator and is the core measurement benchmark for unifying the severity of anomalies of different monitoring indicators.

[0020] Furthermore, the mapping table between the abnormality degree and the basic control degree of each control parameter group refers to a pre-constructed two-dimensional data table used to characterize the correspondence between the abnormality degree of each control parameter in the same control parameter group and the corresponding basic control degree. The mapping table is divided into multiple segments according to the abnormality degree interval, and each abnormality degree interval corresponds to a set of basic control degree values, which are used to calculate the actual control degree under any actual abnormality degree by linear interpolation method during actual control.

[0021] Specifically, step S1 is designed to address the shortcomings of existing silicon carbide etching control benchmark systems, such as fragmented processes and lack of multi-parameter coupling mapping. Silicon carbide materials are extremely chemically inert, and during ICP dry etching, the control parameters and monitoring indicators exhibit a much stronger many-to-many coupling characteristic than in silicon-based etching. Changes in a single control parameter can simultaneously affect multiple monitoring indicators, and anomalies in a single indicator require coordinated correction from multiple control parameters. Furthermore, errors in mask preparation, pretreatment, etching, and post-treatment accumulate and amplify at each stage, ultimately leading to excessive sidewall steepness. Existing technologies only construct static process parameter tables for a single etching stage, failing to establish a comprehensive anomaly tracing and control correspondence covering the entire process. When anomalies occur, they can only rely on manual experience for single-parameter trial and error, easily leading to control conflicts such as a sudden drop in etching rate due to steepness correction or drilling defects caused by rate increases, and the cumulative effect of errors cannot be predicted in advance. This step uses the final sidewall steepness as the unified convergence target and adopts a segmented construction approach to break down the entire process error control into each processing stage, thus avoiding error accumulation at its source. By pre-constructing a set of control parameter groups corresponding to each anomaly type, all experimentally verified effective control schemes are solidified, replacing manual trial and error and avoiding conflicts in multi-parameter coupled control from an architectural perspective. By setting initial control priorities, the optimal scheme is prioritized when an anomaly occurs. By establishing a segmented mapping table between anomaly degree and basic control degree, the control quantity is accurately quantified, avoiding over-control or under-control, and providing a unified and reliable benchmark for subsequent full-process adaptive control.

[0022] Step S2: By deploying online monitoring units at each processing stage, collect the monitoring index values ​​of each processing stage, identify abnormal monitoring indicators of abnormal processing stages, classify abnormal categories, and generate the actual abnormality degree of abnormal monitoring indicators. Please see Figure 2 Furthermore, step S2 includes: Step S21: By deploying online monitoring units at the entrance or exit of each processing stage, all monitoring index values ​​of the corresponding processing stage are collected synchronously; the collected monitoring index values ​​are bound one-to-one with the identity identifier of the processing object, and the collection time, processing stage number, and equipment number are recorded to form a complete processing data chain. The processing object refers to the silicon carbide wafer that has been prepared by the preceding processes such as silicon carbide single crystal growth, cutting, grinding, and polishing, and enters the silicon carbide etching processing production line of the present invention for micro-nano patterning processing, as well as silicon carbide wafers in different processing stages such as mask preparation, pre-etching pretreatment, ICP dry etching, and post-etching processing; each processing object has a unique identity identifier, which is used to realize the tracking and association of processing data throughout the entire life cycle of a single wafer.

[0023] Step S22: Compare the collected monitoring index values ​​with the corresponding benchmark range point by point. If the monitoring index value of any monitoring index in a certain processing stage exceeds the benchmark range for a preset number of consecutive times, the processing stage is determined to be an abnormal processing stage, and the monitoring index exceeding the benchmark range is marked as an abnormal monitoring index. Specifically, the preset number of times is determined based on a comprehensive balance between the duration of common transient interferences in the silicon carbide etching process, the abnormality misjudgment rate, and the response delay. During the silicon carbide ICP etching process, transient interferences such as plasma ignition initial fluctuations, equipment instantaneous voltage fluctuations, and detection device noise usually only last for 1 to 2 acquisition cycles. If anomalies are judged based solely on a single acquisition value exceeding the standard, the misjudgment rate is high. If the preset number of times is too high, it will lead to abnormal response delays and cause error accumulation. For example, the preset number of times is set to 3 times to balance identification accuracy and real-time performance.

[0024] Step S23: Extract the abnormal deviation characteristics of the abnormal monitoring indicators, match them with the abnormal types corresponding to each set of control parameter groups in the control benchmark system, and classify the abnormal types of the current abnormal monitoring indicators.

[0025] Furthermore, step S23 includes: Step S231: Based on the identified abnormal monitoring indicators, extract their abnormal deviation features, including baseline offset features, time-domain variation features, and linkage features of related indicators; Among them, the benchmark offset feature includes the offset direction and relative offset of the abnormal monitoring index relative to its benchmark range median value. The offset direction is divided into positive offset and negative offset, and the relative offset is the actual abnormality degree calculated in step S24. This benchmark offset feature is the core basis for distinguishing different abnormality types under the same monitoring index. The control logic corresponding to positive and negative abnormalities of the same monitoring index is completely opposite.

[0026] The temporal variation characteristics refer to the gradual and abrupt characteristics based on the changing trend, fluctuation amplitude, and degree of abrupt change of the abnormal monitoring indicators within a preset time period before the occurrence of an anomaly. Gradual characteristics refer to the continuous monotonic change of the abnormal monitoring indicator values ​​over time within the preset time period, with the change per unit time less than a preset rate of change threshold. Abrupt characteristics refer to the jump in the values ​​of the abnormal monitoring indicators within a single or no more than two consecutive acquisition cycles, with the change per unit time greater than or equal to a preset rate of change threshold. Gradual and abrupt characteristics are distinguished by calculating the rate of change per unit time of the abnormal monitoring indicators and comparing it with the preset rate of change threshold. This distinction is used to differentiate anomalies of different causes. Gradual characteristics typically correspond to slow-changing conditions such as equipment aging, carbon deposition on chamber walls, and batch differences in substrates, while abrupt characteristics typically correspond to sudden conditions such as sudden changes in gas flow, power supply voltage fluctuations, and plasma ignition anomalies. The adjustment range of the control parameters and the convergence strategies corresponding to these two types of characteristics differ significantly. The preset duration is set based on a comprehensive balance between the dynamic characteristics of the corresponding processing stage, the characteristic period of common slow-changing conditions, and the acquisition frequency. It must ensure complete capture of the changing trends of slow-changing anomalies while avoiding the introduction of excessive irrelevant historical data that could interfere with feature extraction. The preset duration for different processing stages can be set independently. For example, the preset duration for the ICP dry etching process is set to 30 seconds, and the preset duration for the mask preparation process is set to 5 minutes. The preset rate of change threshold is set based on the maximum allowable fluctuation range of the corresponding monitoring index under normal process conditions. For example, the preset rate of change threshold for the plasma F / C ratio is set to 0.05 / second.

[0027] The correlation index linkage characteristic refers to the classification of independent anomaly characteristics and coupled anomaly characteristics based on the correlation between the changes of the abnormal monitoring index and other monitoring indicators within the same time period. An independent anomaly characteristic is one where only the abnormal monitoring index changes beyond the baseline range, while all other associated monitoring indicators remain within a preset fluctuation range. A coupled anomaly characteristic is one where the abnormal monitoring index and at least one other associated monitoring indicator simultaneously change beyond the preset fluctuation range. The correlation index linkage characteristic is obtained by calculating the Pearson correlation coefficient between the abnormal monitoring index and each associated monitoring indicator within a preset time period before the anomaly occurs, and comparing it with a preset correlation coefficient threshold, used to distinguish between independent and coupled anomalies. The preset fluctuation range is set based on the inherent fluctuation amplitude of the corresponding monitoring index under normal process conditions; for example, the preset fluctuation range for etching rate is set to ±5%. The preset correlation coefficient threshold is set based on the coupling characteristic calibration experiment of silicon carbide standard wafers, used to distinguish between accidental synchronous fluctuations and true coupled anomalies; for example, the preset correlation coefficient threshold is set to 0.7, and a synchronous coupled change is determined when the correlation coefficient is greater than or equal to 0.7.

[0028] Step S232: All abnormal deviation features are classified and standardized by combining classification mapping, numerical normalization and Boolean feature mapping to construct an abnormal feature vector with uniform dimensions. Classification standardization means converting abnormal deviation features of different types, dimensions and value ranges into numerical features with uniform value ranges, eliminating the impact of differences in dimensions and value ranges on subsequent similarity calculations, and ensuring the accuracy and fairness of similarity calculations.

[0029] For example, for categorical feature mapping: categorical features such as offset direction and time-domain variation features are mapped to numerical features, where positive offset is mapped to +1 and negative offset is mapped to -1; gradual feature mapping is 0 and abrupt feature mapping is 1; for numerical feature normalization: numerical features such as actual anomaly degree are linearly normalized and mapped to the [0,1] interval; for Boolean feature mapping: Boolean features such as correlation index linkage features are mapped to numerical features, where coupled anomalies are mapped to 1 and non-coupled anomalies are mapped to 0.

[0030] Step S233: Calculate the similarity between the abnormal feature vector and the pre-calibrated standard feature vectors of each abnormal type in the control benchmark system. If the similarity with a certain abnormal type is greater than or equal to a preset similarity threshold, the current abnormal monitoring indicator is determined to be of that abnormal type. If the similarity with all abnormal types is less than the preset similarity threshold, it is determined to be an unknown abnormality, triggering a manual intervention process. Specifically, the preset similarity threshold is determined based on the abnormal calibration experiment of silicon carbide standard wafers and is used to balance the accuracy and recall of abnormal matching. For example, the preset similarity threshold is set to 0.85. The standard feature vectors of each abnormal type are obtained through multiple sets of DOE orthogonal calibration etching experiments in step S1, covering common abnormal types in the silicon carbide etching mass production process, and each standard abnormal type corresponds to a unique set of control parameters.

[0031] Specifically, step S23 is designed to address the core shortcomings of existing silicon carbide etching anomaly control technologies, such as coarse anomaly type classification, inability to distinguish anomaly causes and coupling characteristics, and susceptibility to errors in control direction and mismatch between control methods. Silicon carbide materials are extremely chemically inert, and its ICP dry etching process exhibits much stronger multi-parameter coupling characteristics than silicon-based etching. Anomalies in the same monitoring index may be caused by completely different operating conditions, and the optimal control strategies for anomalies with different causes are fundamentally different: the control directions for positive and negative anomalies are completely opposite; gradual anomalies require small, incremental adjustments to ensure process stability; sudden, abrupt anomalies require large, rapid adjustments to avoid irreversible defects; and coupled anomalies require multi-parameter coordinated control to simultaneously correct multiple related indices. Existing technologies only make simple anomaly judgments based on whether the monitoring index exceeds the baseline range, without distinguishing the direction of anomaly deviation, its cause, and coupling characteristics. When anomalies occur, they can only rely on manual experience and blind trial and error, leading to irreversible defects such as drilling and mask collapse. This step extracts three core anomaly deviation features—baseline offset, temporal variation, and linkage of related indicators—comprehensively covering the directional, causal, and coupling attributes of anomalies, fundamentally solving the problem of insufficient dimensionality in anomaly information. Through standardized processing of classification mapping and numerical normalization, different types of features are converted into numerical vectors of a unified dimension, laying the foundation for automated similarity matching. By calculating cosine similarity with pre-calibrated standard anomaly feature vectors, automated and high-precision anomaly type classification is achieved, ensuring that the set of control parameters subsequently invoked fully matches the control logic, adjustment range, and coordination requirements of the current anomaly. This fundamentally avoids problems of incorrect control direction and mismatched solutions, improving the success rate of a single control intervention.

[0032] Step S24: Based on the baseline range of the anomaly monitoring indicators, perform actual anomaly quantification calculation on the identified anomaly monitoring indicators. The calculation formula is as follows:

[0033] The calculated actual anomaly degree is bound to the anomaly type to generate an anomaly information package containing the anomaly processing stage number, anomaly monitoring indicator name, anomaly type identifier, actual anomaly degree, and anomaly occurrence time. This provides a quantitative basis for subsequent target control parameter group matching and actual control quantity calculation.

[0034] Specifically, step S2 is designed to address the core shortcomings of existing silicon carbide etching monitoring technologies, such as incomplete full-process coverage, inaccurate anomaly identification, and inability to quantify the severity of anomalies. Existing technologies typically only monitor a single point in the etching process and can only determine whether a monitored indicator exceeds a standard, failing to pinpoint the source of the anomaly, differentiate the anomaly type, or quantify its severity, resulting in a lack of precise input for subsequent adjustments. This step, through online monitoring devices deployed at the entry and exit points of each processing stage, achieves uninterrupted monitoring throughout the entire process from mask preparation to post-processing; an anomaly identification mechanism with continuous preset verification effectively filters false alarms caused by transient interference; multi-dimensional anomaly feature extraction and similarity matching achieve precise anomaly type classification; and dimensionless actual anomaly degree calculation unifies the severity of anomalies across different monitoring indicators, providing comprehensive, reliable, and quantitative decision input for subsequent precise adjustments, thus ensuring the realization of adaptive control throughout the entire process from a data perspective.

[0035] Step S3: Combining the anomaly type and actual anomaly degree of the anomaly monitoring indicators, based on the control benchmark system of the anomaly processing link, match the target control parameter group of the anomaly processing link, quantify the actual control degree of each control parameter in the target control parameter group, and generate control instructions; Please see Figure 3 Furthermore, step S3 includes: Step S31: Based on the anomaly type identifier, retrieve the set of control parameter groups corresponding to the anomaly type from the control benchmark system; select the target control parameter group according to the control priority of each control parameter group in the control parameter group set. Specifically, select the control parameter group ranked first as the target control parameter group; if the control of the target control parameter group fails subsequently, automatically switch to the control parameter group with the next control priority, and repeat this step and subsequent steps.

[0036] Step S32: Based on the actual anomaly degree calculated in step S24, determine the anomaly degree interval to which the actual anomaly degree belongs from the anomaly degree to basic control degree mapping table corresponding to the target control parameter group, and obtain the basic control degree corresponding to the anomaly degree interval; calculate the actual control degree of each control parameter in the target control parameter group using linear interpolation, the calculation formula is as follows:

[0037] The anomaly cause adjustment coefficient is set based on the following: using the time-domain variation characteristic of the silicon carbide etching sidewall steepness deviation as the independent variable, the anomaly cause adjustment coefficient value is generated through a preset linear mapping relationship. For example, the linear mapping relationship is set as follows: when the deviation change rate is lower than the process control threshold, the anomaly cause adjustment coefficient increases linearly within the range of 0.8 to 1.0 as the deviation change rate increases; when the deviation change rate is higher than the process control threshold, the anomaly cause adjustment coefficient increases linearly within the range of 1.1 to 1.3 as the deviation change rate increases. Specifically, using the sidewall steepness change rate as the quantified characteristic value, after normalizing it to the corresponding characteristic range, the anomaly cause adjustment coefficient value is calculated through linear interpolation. This anomaly cause adjustment coefficient is used to apply differentiated control intensity to anomalies of different causes, avoiding over-control of gradual anomalies or under-control of abrupt anomalies.

[0038] Step S33: Convert the determined target control parameter set and the calculated actual control degree of each control parameter into control instructions that can be recognized and executed by the corresponding equipment in the abnormal processing stage; issue the control instructions to the abnormal processing stage and adjust the corresponding control parameters synchronously according to the actual control degree; at the same time, record all information of this control, including the abnormal processing stage number, abnormal monitoring indicator name, abnormality type, actual abnormality degree, target control parameter set, actual control degree of each parameter, instruction issuance time, and execution time, in the control log database to provide data support for subsequent feedback verification and benchmark system iteration.

[0039] Specifically, step S3 is designed to address the core shortcomings of existing silicon carbide etching control technologies, including high reliance on manual intervention, mismatched control schemes, and inaccurate control quantities. Existing technologies rely entirely on the manual experience of process engineers for parameter adjustments after anomalies occur. This process is time-consuming and labor-intensive, with significant differences in results between engineers and poor batch consistency. Furthermore, manual adjustments cannot accurately quantify control quantities, easily leading to over- or under-control. This step, through a pre-set set of control parameters, solidifies all experimentally validated effective control schemes, replacing manual experience. By dynamically adjusting the priority of anomaly characteristics, it achieves precise adaptation between the control scheme and the cause of the anomaly. Through the mapping relationship between anomaly degree and basic control degree, it achieves precise quantitative calculation of control quantities. Through automated command generation and issuance, it achieves full automation of the control process, significantly improving control efficiency, accuracy, and batch consistency, and eliminating the high dependence on manual experience.

[0040] Step S4: After the control command is executed and the preset process stabilization time has elapsed, the online monitoring unit of the abnormal processing link collects the abnormal monitoring index value again to determine whether the control is effective immediately. If yes, the control is deemed effective immediately, and the effective response parameter is quantified. If no, the control is deemed ineffective immediately, triggering a control update and automatically switching to the control parameter group with the next control priority in the control parameter group set corresponding to the abnormal monitoring index. The actual control degree of each control parameter is requantified, and a control command is generated until the control is effective immediately or the preset maximum number of control times is reached.

[0041] Please see Figure 4 Furthermore, step S4 includes: Step S41: After the control command is executed, a preset process stabilization timer is started. After the preset process stabilization time, the online monitoring unit of the abnormal processing stage acquires the abnormal monitoring index values ​​for a preset number of times and the monitoring index values ​​associated with the abnormal processing stage according to the same acquisition sequence as in Step S21. These values ​​are then bound to the unique identifier of the processing object, and the time difference between the acquisition time and the completion time of the control command is recorded, providing a data basis for the subsequent quantification of effective response parameters. Specifically, the preset process stabilization time is determined based on the dynamic response characteristics of the corresponding processing stage, referring to the shortest time required for the process system to transition from the original equilibrium state to the new equilibrium state after the control parameters are adjusted. The preset process stabilization time for different processing stages can be set independently. For example, the preset process stabilization time for the ICP dry etching processing stage is set to 10 seconds, the preset process stabilization time for the mask preparation processing stage is set to 30 seconds, and the preset process stabilization time for the post-etching processing stage is set to 20 seconds.

[0042] Step S42: Compare the collected abnormal monitoring index values ​​with the corresponding benchmark ranges point by point; if the abnormal monitoring index values ​​are within the corresponding benchmark range for a preset number of consecutive times, and the monitoring index values ​​associated with the abnormal processing link are within the corresponding preset fluctuation range for a preset number of consecutive times, then the current control is deemed to be effective immediately; otherwise, the current control is deemed to be invalid immediately.

[0043] Step S43: If step S42 determines that the current control is effective immediately, then calculate the effective response parameters of this control, including the control convergence time, the final anomaly degree, and the maximum fluctuation amplitude of the monitoring indicator associated with the anomaly processing stage; wherein, the control convergence time is the time interval from the completion of the control command execution to the first return of the anomaly monitoring indicator to the baseline range; the final anomaly degree is the average anomaly degree after the anomaly monitoring indicator returns to the baseline range; the maximum fluctuation amplitude is the maximum deviation of the corresponding monitoring indicator from its baseline range median during the control process. The average anomaly degree refers to the arithmetic mean of the anomaly degrees of the sampling points for a consecutive preset number of sampling points after the anomaly monitoring indicator returns to the baseline range, which is calculated by summing the anomaly degrees of each sampling point and dividing by the number of sampling points; the maximum deviation refers to the maximum absolute value of the difference between the measured value of the associated monitoring indicator and its baseline range median during the control process, which is calculated by calculating the absolute difference between the measured value and the baseline median point by point and taking the maximum value.

[0044] Step S44: If step S42 determines that the current adjustment is ineffective, check whether the number of adjustments already executed has reached the preset maximum number of adjustments. If not, automatically switch to the next adjustment priority adjustment parameter group in the adjustment parameter group set corresponding to the anomaly monitoring indicator. Based on the latest measured value of the anomaly monitoring indicator, recalculate the actual anomaly degree and the actual adjustment degree of the corresponding adjustment parameter, generate a new adjustment instruction and issue it for execution, and repeat steps S41 to S44. If the preset maximum number of adjustments has been reached, the automatic adjustment is determined to have failed, triggering a manual intervention process. Specifically, the preset maximum number of adjustments is determined based on the fault tolerance capability of the silicon carbide etching process and the risk of wafer scrapping, and is set to 3 times for example. If three consecutive automatic adjustments are ineffective, it means that the current anomaly exceeds the coverage of the preset adjustment scheme. Continuing automatic adjustment may cause irreversible morphological defects, leading to wafer scrapping, thus triggering manual intervention.

[0045] Specifically, step S4 is designed to address the core shortcomings of existing silicon carbide etching control technologies, such as the lack of real-time closed-loop verification, the inability to promptly detect control failures, and the absence of an automatic iterative control mechanism. Existing technologies typically do not perform immediate effect verification after issuing control commands; instead, they wait until the entire processing stage is completed before conducting checks. If control fails at this point, the anomaly has already been amplified and propagated to subsequent stages, easily leading to wafer scrap. Furthermore, existing technologies lack an automatic iterative control mechanism; after a single control failure, manual intervention is required, severely impacting production efficiency. This step, through real-time control effect verification, can quickly detect control failures and automatically switch control schemes, avoiding error accumulation. Synchronous verification of monitoring indicators associated with the abnormal processing stage resolves the control conflict problem caused by multi-parameter coupling. A protection mechanism with a preset maximum number of control cycles effectively reduces the risk of wafer scrap.

[0046] Step S5: In response to the immediate effect of the control, based on the identity of the processed object, the side wall steepness index after the processing is tracked to determine whether the control is globally effective. If so, based on the effective response parameters of the target control parameter group of the abnormal monitoring index, the priority index of the target control parameter group is quantified, and then the control priority and control benchmark system of the target control parameter group are updated to achieve the self-evolution of the control benchmark.

[0047] Furthermore, step S5 includes: Step S51: If step S4 determines that the adjustment is effective immediately, then based on the identification of the processing object, the full-process processing tracking mechanism for the processing object is activated. After the processing object completes all processing steps, the sidewall steepness indicators of the processing object are collected according to preset detection points using scanning electron microscopy (SEM) and atomic force microscopy (AFM), including etched sidewall steepness, sidewall taper, sidewall roughness, drilling depth, and top corner radius. The collected sidewall steepness indicators are then bound to the identification of the processing object. Specifically, the preset detection points are determined based on wafer size and process uniformity requirements. For example, for a 6-inch silicon carbide wafer, the preset detection points are five points: the wafer center, top, bottom, left, and right. Three sets of parallel data are collected at each point, and the arithmetic mean is taken as the final detection result to ensure the representativeness and accuracy of the detection results.

[0048] Step S52: Compare the sidewall steepness index value collected in step S51 with the preset qualification standard point by point; if the sidewall steepness index value of all detection points is within the qualification standard, the overall control is deemed to be effective and the processing object is qualified; otherwise, the overall control is deemed to be invalid and the processing object is unqualified.

[0049] Step S53: Based on the effective response parameters and sidewall steepness index values ​​of this control, calculate the single priority index score of the target control parameter group corresponding to this control, and update the priority index of the target control parameter group by means of the sliding weighted average method. The priority index is a dimensionless index used to quantify the overall control success capability of the target control parameter group under the current production line conditions, which corrects the corresponding anomaly and ensures that the final etching sidewall steepness meets the standard and there are no secondary defects. The higher the index value, the higher the control success rate, the better the control effect, and the faster the control efficiency of the target control parameter group. It should be called first when an anomaly occurs in the future.

[0050] Furthermore, step S53 includes: Step S531: Based on the ratio of the baseline convergence time to the actual convergence time of this adjustment, quantify the convergence speed score of this adjustment; the larger the ratio, the faster the convergence speed, the higher the adjustment efficiency, the smaller the impact on the production line cycle time, and the higher the corresponding score; at the same time, a maximum upper limit is set to avoid interference from accidental outliers on the overall evaluation. Specifically, the baseline convergence time is the average convergence time of this adjustment parameter group under the corresponding anomaly type in the initial calibration experiment of step S1, which is pre-stored in the adjustment baseline system; the maximum upper limit is set to 1.2, and when the convergence speed score is greater than 1.2, it is uniformly taken as 1.2 to avoid excessively inflating the score due to abnormally fast convergence results caused by instantaneous operating condition fluctuations.

[0051] Step S532: Retrieve all control records of the target control parameter group within the past preset statistical period, and count the number of successful controls that simultaneously meet the requirements of immediate control effectiveness and global control effectiveness, along with the total number of controls, to calculate the control success rate of the target control parameter group. This control success rate directly reflects the long-term stable control success capability of the parameter group and is a core component of the priority indicator. Specifically, the preset statistical period is set based on a comprehensive balance between the rate of change of the silicon carbide etching process conditions, the significance of the statistical results, and the production line cycle time. The length of the preset statistical period must simultaneously meet the constraints of sample size and timeliness. It must be sufficient to cover the typical drift cycle of the process chamber to ensure the significance of the statistical results, while limiting the time span to avoid including outdated historical data, which would cause the statistical results to be unable to match the current operating conditions. For example, all control records within the past preset statistical period can be set to the most recent 100 controls of the same abnormal type, or all control records of the most recent 30 days, taking the shorter time span. The upper limit of the sample size for the preset statistical period is set at the most recent 100 adjustments to ensure that the statistical sample size is large enough to meet the statistical significance requirements. The upper limit of the preset statistical period is set at the most recent 30 days to adapt to the typical maintenance cycle of the silicon carbide etching chamber, avoid outdated data caused by chamber aging from affecting the evaluation results, and ensure that the adjustment records within the preset statistical period can reflect the true performance of the parameter group under the current operating conditions.

[0052] Step S533: Based on the relative deviations of the sidewall steepness index values ​​at all test points from the corresponding median of the pass standard, calculate the individual score for each sidewall steepness index, and obtain the overall comprehensive score through weighted calculation; specifically, first calculate the measured average value of each index at all test points, then calculate the absolute deviation of the measured average value from the median of the corresponding target range, and divide the absolute deviation by the allowable deviation of the pass standard for that index to obtain the relative deviation value of that index; the allowable deviation of the pass standard refers to the acceptable range of silicon carbide etching sidewall steepness. The half-width value, which is half the difference between the upper and lower limits of the pass standard, is used to normalize the measured deviation to the 0-1 range to calculate the individual score. The individual score of the indicator is obtained by subtracting the relative deviation value from 1. The highest individual score is 1.0 and the lowest is 0. Then, the weights are assigned according to the degree of influence of each sidewall steepness index value on the final device performance. Among them, the core index, etching sidewall steepness, is assigned the highest weight. The other auxiliary indexes are assigned corresponding weights according to their degree of influence. The individual scores of each index are multiplied by their corresponding weights and then summed to obtain the overall performance score. For example, the weight of sidewall steepness is set to 70%, and the weights of sidewall taper, sidewall roughness, drilling depth, and top fillet radius are each set to 7.5%. If the final individual score for sidewall steepness of a wafer is 0.95, the individual score for sidewall taper is 0.9, the individual score for sidewall roughness is 0.92, the individual score for drilling depth is 0.88, and the individual score for top fillet radius is 0.9, then the overall performance score is 0.95×0.7+0.9×0.075+0.92×0.075+0.88×0.075+0.9×0.075=0.935.

[0053] Step S534: Combining the control success rate, convergence speed score, and global effect comprehensive score, calculate the single-time priority index score of the target control parameter group corresponding to this control through weighted summation; the weighting of the control success rate, convergence speed score, and global effect comprehensive score is set according to the importance ranking of each index to the long-term stability of silicon carbide etching process, control effectiveness, and production line cycle time: the control success rate directly reflects the long-term stable control capability of the parameter group, is the core index, and is given the highest weight; the global effect comprehensive score reflects the improvement effect of control on sidewall steepness, is directly related to device performance, and is given the second highest weight; the convergence speed score reflects the impact of control on production line cycle time, is an auxiliary index, and is given a lower weight; for example, the single-time priority index score is obtained by weighted summation according to the weight ratio of 50% for control success rate, 30% for global effect comprehensive score, and 20% for convergence speed score; the higher the score, the stronger the comprehensive control success capability of the target control parameter group under the current operating conditions. If the control success rate of a certain control parameter group is 0.92, the convergence speed score is 1.1, and the overall global effect score is 0.934, then the single priority index score = 0.92×0.5 + 1.1×0.2 + 0.934×0.3 = 0.9602.

[0054] Step S535: Using a sliding weighted average method, the single priority index score calculated this time is integrated with the historical priority index of the target control parameter group to update the latest priority index of the target control parameter group. Specifically, the weight of the historical priority index is set to 0.7 to ensure the long-term stability of the control benchmark and avoid large fluctuations in priority caused by a single control result. The weight of the single priority index score this time is set to 0.3 to ensure that the priority index can adapt to long-term operating condition changes such as equipment aging, chamber carbon deposition, and substrate batch differences in a timely manner. The calculated latest priority index is then used to cover the original historical priority index of the control parameter group in the control benchmark system, completing the effect feedback and data update of this control, and providing a quantitative basis for the dynamic ranking of the control parameter group in the subsequent step S54.

[0055] Specifically, step S53 is designed to address the core shortcomings of existing silicon carbide etching control technologies, such as the static solidification of control parameter group priorities, the inability to objectively reflect the actual control success capability of the parameter groups, and the low priority call rate of the optimal solution. In existing technologies, the priority of control parameter groups is usually a static value determined once during the initial calibration stage, and it is not dynamically updated according to the actual control effect during subsequent production. This static priority mechanism has three fatal problems: First, it cannot adapt to long-term changes in operating conditions such as equipment aging, chamber carbon deposition, and substrate batch differences, which may cause the initially calibrated optimal parameter group to gradually become ineffective in actual production, or even become the worst solution; Second, the priority ranking is based only on small-batch experimental data in the laboratory, without combining real control records in large-scale mass production, and cannot objectively reflect the long-term stable control success capability of the parameter group under complex and ever-changing actual operating conditions; Third, the evaluation dimension is singular, usually only considering whether it can correct the current anomaly, without comprehensively considering multiple dimensions such as control efficiency, final product quality, and secondary defects, resulting in the priority ranking being seriously deviated from actual needs.

[0056] This step constructs a multi-dimensional comprehensive priority evaluation system that combines historical long-term performance with current actual performance. Historical control success rate is used as the core evaluation dimension, directly quantifying the long-term stability of the parameter set in correcting anomalies and ensuring the final product's qualification under current operating conditions. Simultaneously, the convergence speed score and overall effect score of this control are combined to comprehensively evaluate the control efficiency and final processing quality of the parameter set. The priority index is updated using a sliding weighted average method, ensuring the long-term stability of the control benchmark, avoiding significant fluctuations in priority due to single accidental results, and responding promptly to changes in operating conditions. This ensures that the priority index always reflects the latest control success capability of the parameter set, significantly improving the accuracy and stability of control, and effectively reducing the frequency of manual intervention and wafer scrap rate.

[0057] Step S54: After completing the control of the same anomaly type in each preset update cycle, sort the priority indicators of all control parameter groups corresponding to that anomaly type from high to low. If the priority indicator of a control parameter group with a low control priority is higher than that of a control parameter group with a high control priority, then reallocate the control priorities of each control parameter group according to the order of priority indicators from high to low. At the same time, based on the globally effective sample dataset of all controls within this update cycle, simultaneously optimize the anomaly degree and basic control degree mapping table of each control parameter group. Specifically, the preset update cycle is determined comprehensively based on the rate of change of operating conditions in batch production and the process stability requirements. If the update cycle is too short, the control priority will fluctuate frequently, affecting process stability. If the update cycle is too long, it will be unable to adapt to long-term operating condition changes such as equipment aging and chamber state drift in a timely manner. For example, the preset update cycle is set to 20 times, which can achieve timely adaptive updates of the control benchmark while ensuring process stability. The sample dataset refers to the complete set of control records that are determined to be globally effective for control within the current preset update period. It consists of the abnormality type identifier, actual abnormality degree, target control parameter group, actual control degree of each control parameter, effective response parameter, and final sidewall steepness index corresponding to each control record.

[0058] Further steps in optimizing the mapping table between the anomaly degree and the basic control degree for each control parameter group include: Step S541: From the sample dataset of this update cycle, classify and filter samples according to anomaly type and target control parameter group, removing invalid samples with missing data or equipment malfunction interference; group the filtered valid samples according to the original anomaly degree interval division rules of the control parameter group in the control benchmark system, ensuring that the grouping intervals are completely consistent with the intervals of the original mapping table, guaranteeing the continuity and compatibility of the mapping table. Specifically, the anomaly degree interval division rules are consistent with the initial calibration stage in Step S1. For example, the anomaly degree is divided into 5 continuous intervals: 0.1 to 0.3, 0.3 to 0.5, 0.5 to 0.7, 0.7 to 1.0, and 1.0 to 1.5; when the number of valid samples in a certain anomaly degree interval is less than 5, the basic control degree of that interval is not updated temporarily, and the original value is retained to avoid statistical bias caused by small samples.

[0059] Step S542: For all valid samples within each anomaly interval, extract the actual control degree values ​​of each control parameter in the control parameter group recorded in the samples; calculate the arithmetic mean of all actual control degrees for each control parameter within the interval, and use it as the candidate basic control degree for the control parameter corresponding to the anomaly interval. For example, if a control parameter has 8 valid samples within an anomaly interval, and its actual control degrees are 2.1, 2.3, 2.2, 2.4, 2.2, 2.3, 2.1, and 2.2 respectively, then the candidate basic control degree for the control parameter in that interval is (2.1+2.3+2.2+2.4+2.2+2.3+2.1+2.2) / 8=2.225.

[0060] Step S543: Using a sliding weighted average method, the candidate basic control degree calculated in step S542 is merged with the basic control degree of the corresponding interval in the original mapping table to obtain the updated basic control degree. The weight of the original basic control degree can be set to 0.6, and the weight of the candidate basic control degree can be set to 0.4, ensuring both the long-term stability of the mapping table and timely absorption of operating condition changes reflected in mass production data. For example, if the original basic control degree of a certain interval is 2.0 and the candidate basic control degree is 2.225, then the updated basic control degree = 2.0 × 0.6 + 2.225 × 0.4 = 2.09, rounded to two decimal places as the final updated value.

[0061] Step S544: Calculate the control success rate of each anomaly interval within the current update cycle. If the control success rate of an anomaly interval is lower than the preset success rate threshold for three consecutive update cycles, the basic control degree correction mechanism for that interval is automatically triggered. Calculate the control degree correction coefficient based on the average actual anomaly degree and average residual anomaly degree of all failed control samples within the anomaly interval. Multiply the basic control degree of the interval by the correction coefficient for adjustment and mark it as an interval to be verified. Formal optimization will be performed after accumulating no less than 10 valid samples. Residual anomaly degree refers to the final deviation of the anomaly monitoring index relative to its baseline range median value after the control command is executed and the process reaches a stable state. It is obtained by collecting the measured values ​​of the anomaly monitoring index within the preset stable time window after the control is completed, calculating its deviation from the baseline range median value, and combining it with the actual anomaly degree calculation method in step S24. A positive residual anomaly degree indicates insufficient control, and a negative residual anomaly degree indicates over-control. Specifically, the preset success rate threshold is set based on a quantitative balance between the fault tolerance of the silicon carbide etching process, production line yield requirements, the reliability requirements of the control system, and the risk of false triggering of the correction mechanism: based on the process yield target, a minimum acceptable level of control success rate is set within the update cycle, while considering the cost of false triggering of the correction mechanism to avoid frequent corrections due to short-term fluctuations; for example, the preset success rate threshold is set to 80%; the control degree correction coefficient is calculated based on the ratio of the average actual anomaly of failed samples within the interval to the anomaly eliminated by this control, used to quantify the insufficiency or excess of the current basic control degree, specifically the control degree correction... The positive coefficient = average actual anomaly / (average actual anomaly - average residual anomaly); for example, if the control success rate for three consecutive cycles in a certain interval is 72%, and the average actual anomaly of the failed samples is 2.2 and the average residual anomaly is 0.2, then the control degree correction coefficient = 2.2 / (2.2 - 0.2) = 1.1, increasing the original basic control degree by 10% to solve the problem of insufficient control; if the average actual anomaly of the failed samples is 1.35 and the average residual anomaly is -0.15, then the control degree correction coefficient = 1.35 / (1.35 - (-0.15)) = 0.9, decreasing the original basic control degree by 10%.

[0062] Step S545: Replace the corresponding values ​​in the original anomaly and basic control degree mapping table of the control parameter group in the control benchmark system with the basic control degree of all updated anomaly degree ranges; at the same time, back up the original mapping table completely to the process history version library, and record the version number, update time, sample size and main adjustment content based on the update, for subsequent process review, anomaly tracing and version rollback.

[0063] Specifically, this optimization step addresses the shortcoming of the initially calibrated mapping table between anomaly degree and basic control degree, which cannot adapt to long-term changes in operating conditions. The initial mapping table is built only based on small-batch experimental data in the laboratory. As operating conditions change, such as equipment aging and chamber state drift, the calculated actual control degree will gradually deviate from the optimal value, leading to a decrease in control accuracy. This step uses globally effective control samples accumulated during mass production, employing a combination of incremental updates and adaptive corrections to dynamically optimize the basic control degree for each anomaly degree interval, ensuring that the mapping table always reflects the optimal control value correspondence under the current operating conditions.

[0064] Specifically, step S5 is designed to address the core shortcomings of existing silicon carbide etching control technologies, such as the lack of global effect verification, static fixation of the control benchmark, and inability to adapt to long-term operating conditions. Existing technologies only perform immediate verification of a single step after control, failing to use the final device quality as the final criterion for judging the control effect. This easily leads to local optima where local step indicators are normal, but the final product is unqualified. Furthermore, the priority of the control parameter group is based on initially calibrated static values, which cannot be dynamically adjusted with long-term operating conditions such as equipment aging, chamber carbon deposition, and substrate batch differences. This results in the optimal control scheme not being prioritized in long-term production, leading to a continuous decline in control accuracy and yield. This step completely solves the local optima problem by establishing a dual-feedback verification mechanism of immediate step feedback and final global feedback, using the final sidewall steepness as the final criterion for judging the control effect. Through quantified priority indicators and a dynamic control priority adjustment mechanism, the control benchmark system can adapt to long-term operating conditions, always maintaining the optimal control effect, significantly improving the stability and reliability of long-term production.

[0065] Example 2 This embodiment introduces a silicon carbide etching sidewall steepness adaptive control system for executing the silicon carbide etching sidewall steepness adaptive control method described in Embodiment 1 above. The system includes: The control benchmark system construction module takes the target etching sidewall steepness as the final convergence target. Based on the control parameters and monitoring indicators of each processing stage of silicon carbide etching, it pre-constructs a control benchmark system covering the entire processing flow. It stores the set of control parameter groups under each anomaly type associated with each monitoring indicator, the initial control priority of each control parameter group within the control parameter group set, and the mapping table between the anomaly degree and the basic control degree of each control parameter group. It supports the import, export, and version management of the control benchmark system, providing a unified and reliable benchmark basis for subsequent full-process adaptive control.

[0066] The anomaly monitoring module, through online monitoring units deployed at the entry and exit points of each processing stage, synchronously collects all monitoring indicator values ​​for the corresponding processing stage according to a preset collection sequence. It binds each collected monitoring indicator value to a unique identifier of the processing object, and records the collection time, processing stage number, and equipment number, forming a complete single-wafer processing data chain. It compares each collected monitoring indicator value with its corresponding benchmark range point by point to identify abnormal processing stages and abnormal monitoring indicators. It extracts the benchmark offset features, time-domain variation features, and linkage features of related indicators from the abnormal monitoring indicators, performs standardization processing to construct an anomaly feature vector, calculates its similarity with the standard feature vector in the control benchmark system, and classifies the anomaly type of the current anomaly monitoring indicator. Finally, it calculates the actual anomaly degree of the anomaly monitoring indicator, generates an anomaly information package containing complete anomaly information, and sends it to the control decision and instruction generation module.

[0067] The control decision module receives anomaly information packets from the full-process anomaly monitoring and classification module. Based on the anomaly type identifier in the anomaly information packet, it retrieves the unique control parameter set corresponding to that anomaly type from the control benchmark system. According to the current control priority of each control parameter set in the control parameter set, it selects the highest-ranked control parameter set as the target control parameter set. Based on the actual anomaly degree in the anomaly information packet, it determines the anomaly degree interval to which the actual anomaly degree belongs from the anomaly degree-to-basic control degree mapping table corresponding to the target control parameter set, obtains the corresponding basic control degree, and calculates the actual control degree of each control parameter in the target control parameter set in conjunction with the anomaly cause adjustment coefficient. It then converts the target control parameter set and the actual control degree of each parameter into digital control instructions that can be recognized and executed by the corresponding equipment in the anomaly processing stage, and sends them to the equipment execution unit. Simultaneously, all information from this control operation is completely recorded in the data storage and management module.

[0068] The control verification module, after the control command is executed, starts a preset process stabilization timer. After the preset process stabilization time for the corresponding processing stage, it controls the online monitoring unit of the abnormal processing stage to continuously collect real-time data of the abnormal monitoring indicators and all related monitoring indicators for a preset number of times, following the same collection sequence as the abnormal monitoring stage. The collected verification data is compared point by point with the benchmark range of the corresponding monitoring indicators to determine whether the control is effective immediately. If the control is effective immediately, the module calculates and records the effective response parameters such as the control convergence time, final abnormality, and maximum fluctuation range of related indicators. If the control is ineffective immediately, it checks whether the number of control operations executed has reached the preset maximum number of control operations. If not, it automatically switches to the control parameter group with the next control priority in the control parameter group set corresponding to the abnormality type, recalculates the actual abnormality and actual control degree, and generates a new control command for execution. If the preset maximum number of control operations has been reached, the automatic control is determined to have failed, triggering the manual intervention process and sending an alarm message to the manual interaction and intervention module.

[0069] The baseline self-evolution module, responding to the immediate effect signal from the real-time control effect verification and iteration module, initiates a full-process processing tracking mechanism for the processed object based on its unique identifier. This mechanism continuously records the process parameters and monitoring data for the processed object in all subsequent processing stages. After the processed object completes all etching processes, it receives core morphological indicators such as final etched sidewall steepness, sidewall taper, sidewall roughness, drilling depth, and top corner radius from scanning electron microscopy and atomic force microscopy. These final morphological indicators are then compared point-by-point with preset device qualification standards to determine if the control is globally effective. If globally effective, ... Based on the effective response parameters and final morphological indicators of this regulation, the single priority index score of the target regulation parameter group is calculated, and the latest priority index of the regulation parameter group is updated by the sliding weighted average method. After each regulation of the same abnormality type is completed in a preset update cycle, the latest priority index of all regulation parameter groups corresponding to the abnormality type is sorted from high to low, and the regulation priority of each regulation parameter group is redistributed. At the same time, based on the global effective sample dataset of all regulation in this update cycle, the mapping table between the abnormality degree and the basic regulation degree of each regulation parameter group is optimized by combining incremental update and adaptive correction, so as to realize the self-evolution of the regulation benchmark system.

[0070] Working principle and its effects: Based on the core concept of closed-loop control and self-evolution, this invention constructs a complete adaptive control system covering silicon carbide etching from anomaly identification and precise control to benchmark iteration. By transforming human experience into standardized control benchmarks and automated decision-making logic, it fundamentally solves the industry pain points of high dependence on traditional manual control, insufficient precision, and inability to adapt to long-term operating condition changes.

[0071] A pre-constructed control benchmark system covering all processing stages, including mask preparation, etching, and post-processing, solidifies effective control schemes validated through extensive experiments into a set of callable control parameter groups and quantified mapping relationships. This replaces traditional manual trial-and-error, architecturally avoiding conflicts arising from multi-parameter coupled control and significantly reducing the frequency of manual intervention. By collecting monitoring data throughout the entire process, multi-dimensional abnormal deviation features linked to benchmark offsets, time-domain changes, and related indicators are extracted and standardized for matching and classification. Combined with dimensionless anomaly quantification, accurate anomaly identification and unified severity characterization are achieved, ensuring precise adaptation of subsequent control schemes to the causes and severity of anomalies. Based on control priorities, the optimal target control parameter group is automatically matched, combined with anomalies... The constant-to-basic control degree mapping table quantifies and calculates the actual control amount and generates executable instructions to achieve precise control of the control amount and avoid over-control or under-control. After the control is executed, the effectiveness of the control is quickly judged through real-time effect verification. If it is ineffective, the control scheme is automatically iterated and switched to effectively avoid the amplification of abnormal errors step by step. Finally, a dual feedback mechanism combining real-time process feedback and final global feedback is adopted. Based on the historical control success rate, single control effect and final morphology quality, the priority ranking of the control parameter group and the abnormality-to-basic control degree mapping table are dynamically updated. This enables the control benchmark system to continuously adapt to long-term operating condition changes such as equipment aging, chamber state drift and substrate batch differences, and always maintain the optimal control effect.

[0072] In summary, this invention achieves fully automated and adaptive control of the sidewall steepness of silicon carbide etching, significantly improving the consistency of sidewall steepness and the success rate of first-time control, effectively enhancing the stability and yield of mass production, and providing reliable process support for the large-scale mass production of silicon carbide semiconductor devices.

[0073] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for adaptively controlling the steepness of silicon carbide etching sidewalls, characterized in that, include: S1: Based on the control parameters and monitoring indicators of each processing stage of silicon carbide etching, a control benchmark system for each processing stage is pre-constructed; The control benchmark system includes a set of control parameter groups for each abnormality type associated with each monitoring indicator, the initial control priority of each control parameter group within the set of control parameter groups, and a mapping table between the abnormality degree and the basic control degree of each control parameter group. The set of control parameter groups consists of multiple control parameter groups, each of which refers to a parameter combination that includes at least one control parameter and its corresponding adjustment direction. S2: Collect monitoring index values ​​for each processing stage, identify abnormal monitoring indexes for abnormal processing stages, classify abnormal categories, and generate the actual abnormality degree of abnormal monitoring indexes. The process of collecting monitoring index values ​​for each processing stage, identifying abnormal monitoring indicators for abnormal processing stages, classifying abnormality categories, and generating the actual abnormality degree of abnormal monitoring indicators includes: Collect all monitoring index values ​​of each processing stage, bind them one by one with the identity of the processed object, and record the collection time, processing stage number and equipment number to form a complete processing data chain. The collected monitoring index values ​​are compared point by point with the corresponding benchmark range of the monitoring index. If the monitoring index value of any monitoring index in a certain processing link exceeds the benchmark range for a preset number of consecutive times, the processing link is determined to be an abnormal processing link, and the monitoring index that exceeds the benchmark range is marked as an abnormal monitoring index. Extract the abnormal deviation characteristics of abnormal monitoring indicators, match them with the abnormal types corresponding to each set of control parameter groups in the control benchmark system, and classify the abnormal types of the current abnormal monitoring indicators. Based on the baseline range of the anomaly monitoring indicators, the actual anomaly quantification calculation is performed on the identified anomaly monitoring indicators; S3: Combining the anomaly type and actual anomaly degree of the anomaly monitoring indicators, based on the control benchmark system of the anomaly processing link, match the target control parameter group of the anomaly processing link, quantify the actual control degree of each control parameter in the target control parameter group, and generate control instructions. The steps of matching the target control parameter group for the abnormal processing stage, quantifying the actual control degree of each control parameter in the target control parameter group, and generating control instructions include: Based on the anomaly type identifier, the set of control parameter groups corresponding to the anomaly type is retrieved from the control benchmark system; the target control parameter group is selected according to the control priority of each control parameter group in the control parameter group set. Based on the actual anomaly degree, the anomaly degree interval to which the actual anomaly degree belongs is determined from the anomaly degree to basic control degree mapping table corresponding to the target control parameter group, and the basic control degree corresponding to the anomaly degree interval is obtained; the actual control degree of each control parameter in the target control parameter group is calculated by linear interpolation. The target control parameter set and the actual control degree of each control parameter are converted into control commands that can be recognized and executed by the corresponding equipment in the abnormal processing stage, and then sent to the abnormal processing stage. S4: After the control command is executed and the preset process stabilization time has elapsed, collect the abnormal monitoring index value, determine whether the control is effective immediately, if so, quantify the effective response parameter, if not, trigger the control update. S5: In response to the immediate effect of regulation, track the sidewall steepness index after processing, determine whether the regulation is globally effective, and if so, quantify the priority index of the target regulation parameter group, and then update the regulation priority and regulation benchmark system of the target regulation parameter group.

2. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The mapping table between the abnormality degree and the basic control degree of each control parameter group is a two-dimensional data table used to characterize the correspondence between the abnormality degree of each control parameter in the same control parameter group and the corresponding basic control degree. The mapping table between anomaly degree and basic control degree is divided according to the anomaly degree interval, and each anomaly degree interval corresponds to a set of basic control degree values.

3. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The abnormal deviation characteristics of the extracted abnormal monitoring indicators are matched with the abnormal types corresponding to each set of control parameter groups in the control benchmark system to classify the abnormal types of the current abnormal monitoring indicators, including: Based on the identified abnormal monitoring indicators, their abnormal deviation characteristics are extracted, including baseline offset characteristics, time-domain variation characteristics, and linkage characteristics of related indicators. The baseline offset features include offset direction and relative offset amount; the time-domain variation features include gradual change features and abrupt change features; the linkage features of related indicators include independent anomaly features and coupled anomaly features; All abnormal deviation features are classified and standardized by combining classification mapping, numerical normalization and Boolean feature mapping to construct an abnormal feature vector; The similarity between the abnormal feature vector and the standard feature vector of each abnormal type in the control benchmark system is calculated; if the similarity with a certain abnormal type is greater than or equal to the preset similarity threshold, the current abnormal monitoring indicator is determined to be that abnormal type.

4. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The aforementioned adjustment update includes: Automatically switch to the next control parameter group with the next control priority in the set of control parameter groups corresponding to the abnormal monitoring indicators, requantify the actual control degree of each control parameter, generate control instructions, until the control takes effect immediately or the preset maximum number of control times is reached.

5. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The determination of whether the regulation takes effect immediately includes: After the control command is executed and the preset process stabilization time has elapsed, obtain the abnormal monitoring index values ​​for a preset number of times and the monitoring index values ​​associated with the abnormal processing link. The collected abnormal monitoring index values ​​are compared point by point with the corresponding benchmark range of the monitoring index. If the monitoring index values ​​of the abnormal monitoring index are within the corresponding benchmark range for a consecutive preset number of times, and the monitoring index values ​​of the monitoring index associated with the abnormal processing link are within the corresponding preset fluctuation range for a consecutive preset number of times, then the current control is deemed to be effective immediately; otherwise, the current control is deemed to be invalid immediately.

6. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The effective response parameters include the control convergence time, final anomaly degree, and maximum fluctuation amplitude; The control convergence time is the time interval from the completion of the control command execution to the first return of the abnormal monitoring index to the benchmark range; The final anomaly degree is the average anomaly degree after the anomaly monitoring indicators return to the benchmark range; the maximum fluctuation amplitude is the maximum deviation of the monitoring indicators associated with the anomaly processing stage during the regulation process relative to the median of their benchmark range.

7. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 1, characterized in that, The steps of quantifying the priority index of the target control parameter group and then updating the control priority and control benchmark system of the target control parameter group include: Based on the effective response parameters and sidewall steepness index values ​​of this regulation, the single priority index score of the target regulation parameter group corresponding to this regulation is calculated, and the priority index of the target regulation parameter group is updated by the sliding weighted average method. After each preset update cycle of the same type of abnormality is controlled, the priority indicators of all control parameter groups corresponding to the abnormality type are sorted from high to low, and the control priority of each control parameter group is redistributed. At the same time, based on the global effective sample dataset of all regulation within this update cycle, the mapping table between the abnormality degree and the basic regulation degree of each regulation parameter group is simultaneously optimized.

8. The method for adaptively controlling the steepness of silicon carbide etching sidewalls as described in claim 7, characterized in that, The priority indicators for updating the target control parameter group include: The convergence speed score of this regulation is quantified based on the ratio of the benchmark convergence time to the actual convergence time. Retrieve all control records of the target control parameter group within the past preset statistical period, count the number of successful controls that simultaneously meet the conditions of immediate control effect and global control effectiveness, and calculate the control success rate of the target control parameter group. Based on the relative deviation between the sidewall steepness index values ​​of all test points and the corresponding qualified standard median, the individual score of each sidewall steepness index is calculated, and the overall score of the global effect is obtained through weighted calculation. By combining the control success rate, convergence speed score, and global effect comprehensive score, the single priority index score of the target control parameter group corresponding to this control is calculated. The single priority index score obtained from this calculation is integrated with the historical priority index of the target control parameter group to update the latest priority index of the target control parameter group.

9. A silicon carbide etching sidewall steepness adaptive control system, used to execute the silicon carbide etching sidewall steepness adaptive control method according to any one of claims 1-8, characterized in that, include: The control benchmark system construction module is used to pre-build the control benchmark system for each processing step based on the control parameters and monitoring indicators of each processing step in silicon carbide etching. The anomaly monitoring module is used to collect monitoring index values ​​of each processing stage, identify abnormal monitoring indexes of abnormal processing stages, classify anomalies, and generate the actual anomaly degree of the anomaly monitoring indexes. The control decision module is used to combine the anomaly type and actual anomaly degree of the anomaly monitoring indicators, match the target control parameter group of the anomaly processing link based on the control benchmark system of the anomaly processing link, quantify the actual control degree of each control parameter in the target control parameter group, and generate control instructions. The control verification module is used to collect abnormal monitoring index values ​​after the control command is executed and a preset process stabilization time has elapsed, and to determine whether the control is effective immediately. If so, the effective response parameters are quantified; otherwise, the control is updated. The baseline self-evolution module responds to the immediate effect of regulation, tracks the sidewall steepness index after processing, determines whether the regulation is globally effective, and if so, quantifies the priority index of the target regulation parameter group, and then updates the regulation priority and regulation baseline system of the target regulation parameter group.

Images