Etching process parameter optimization method and system

Abstract

The application relates to the technical field of semiconductor manufacturing, and discloses an etching process parameter optimization method and system. The method comprises the following steps: constructing an initial parameter set based on historical process data and corresponding process result data, and training a parameter mapping model to quantify the correlation between parameters and the influence weight; a parameter exploration set is generated by adjusting the search range; a genetic algorithm is used for evaluation, candidate parameter combinations are screened and input into an etching simulation environment for verification; whether a preset process target is reached is determined according to a simulation result; if the target is not reached, cross mutation operation is iteratively updated and re-evaluated; if the target is reached, a final parameter set is output; a process deployment scheme is generated based on the final parameter set, and applicability verification is carried out in a target process execution environment, so that the best parameter combination is determined. The application realizes intelligent optimization of etching process parameters, and significantly improves the efficiency of etching process optimization and the reliability of parameter landing.

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing technology, and in particular to a method and system for optimizing etching process parameters. Background Technology

[0002] In semiconductor manufacturing, etching is a crucial process for pattern transfer and structure shaping, widely used in the fabrication of structures such as gates, contact holes, trenches, metal interconnects, and dielectric layers. The effectiveness of etching is typically influenced by a variety of process parameters, including reactive gas flow rate, RF power, chamber pressure, bias power, etching time, and temperature. These parameters often exhibit complex coupling relationships, and different combinations can significantly impact process outcomes such as etching rate, etching uniformity, selectivity, sidewall morphology, surface smoothness, and critical dimension control. Therefore, efficient and accurate optimization of etching process parameters directly affects the processing quality, manufacturing yield, and process stability of semiconductor devices.

[0003] From a technological development perspective, early etching process parameter optimization relied primarily on the experience of process engineers and repeated experiments, involving manual corrections through minor parameter adjustments combined with experimental results. While this approach is suitable for scenarios with a small number of parameters and a wide process window, it suffers from drawbacks such as long trial-and-error cycles, high R&D costs, strong reliance on experience, and difficulty in handling multi-parameter coupling issues. Subsequently, as the complexity of semiconductor processes continued to increase, optimization methods based on experimental design, statistical analysis, and mechanistic modeling were gradually applied. By establishing quantitative relationships between parameters and process results, the targeting of parameter adjustments was improved, and blind experimentation was reduced. Furthermore, with the development of data processing capabilities and intelligent algorithms, data-driven methods such as neural networks and genetic algorithms began to be introduced into the etching process optimization process to improve the optimization efficiency and result prediction capabilities in high-dimensional parameter spaces.

[0004] However, existing technologies still have many limitations. On the one hand, in semiconductor etching processes, various process parameters are often characterized by nonlinearity, strong coupling, and dynamic changes. Existing optimization methods often struggle to accurately extract the correlations between parameters and the degree of influence of each parameter on the process results, leading to unreasonable parameter search range settings, a strong degree of blindness in the optimization process, and a tendency to get trapped in local optima. On the other hand, while some existing methods can output candidate parameter combinations based on simulation or algorithms, they lack a complete closed-loop mechanism from historical data analysis, parameter space expansion, candidate parameter evaluation, simulation verification to process deployment verification. This results in insufficient applicability and stability of the obtained parameter combinations in the actual process execution environment, making it difficult to continuously meet the preset process objectives. Furthermore, as the process window continues to narrow in advanced processes, traditional optimization methods that rely on manual trial and error or single-model prediction are no longer able to balance optimization efficiency, prediction accuracy, and practical implementation results.

[0005] Therefore, how to construct an etching process parameter optimization method that can combine historical process data, parameter correlation modeling, parameter space optimization, simulation verification, and process deployment verification, so as to improve the parameter optimization efficiency, enhance the matching ability of parameter combinations to preset process objectives, and improve the executability and stability of optimization results in actual semiconductor manufacturing environment, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This application provides a method and system for optimizing etching process parameters to solve the technical problems existing in the optimization of etching process parameters, such as difficulty in accurately representing parameter correlations, unreasonable parameter search range settings, easy getting trapped in local optima during the optimization process, lack of effective connection between simulation verification and process deployment verification, and difficulty in stably meeting the preset process objectives with the obtained parameter combinations.

[0007] In a first aspect, this application provides a method for optimizing etching process parameters, the method comprising: S1. Extract historical process parameters and corresponding process result data from the preset database to construct an initial parameter set; S2. Train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process results based on the feature information. S3. Adjust the search range of each process parameter according to the correlation and influence weight, and sample the adjusted parameter search space to generate an expanded parameter exploration set; S4. Evaluate the parameter combinations in the parameter exploration set according to the preset process objectives, and select candidate parameter combinations. S5. Input the candidate parameter combination into the preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results; S6. Determine whether the optimized parameter configuration meets the preset process target. If not, update the parameter subset in the parameter exploration set according to the simulation results and return to S4. If yes, determine the optimized parameter configuration as the final parameter set. S7. Generate a process deployment plan based on the final parameter set, and verify its applicability to obtain the optimal parameter combination.

[0008] Secondly, this application provides an etching process parameter optimization system, the system comprising: The data extraction module is used to extract historical process parameters and corresponding process result data from a preset database to construct an initial parameter set; The modeling and quantification module is used to train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process result based on the feature information. The spatial expansion module is used to adjust the search range of each process parameter according to the correlation and influence weight, and to sample the adjusted parameter search space to generate an expanded parameter exploration set. The parameter evaluation module is used to evaluate the combinations of parameters in the parameter exploration set according to the preset process objectives and to select candidate parameter combinations. The simulation verification module is used to input the candidate parameter combination into the preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results; The iterative optimization module is used to determine whether the optimized parameter configuration meets the preset process target. If not, it updates the parameter subset in the parameter exploration set according to the simulation results and returns to S4. If yes, it determines the optimized parameter configuration as the final parameter set. The deployment verification module is used to generate a process deployment plan based on the final parameter set, perform applicability verification, and obtain the optimal parameter combination.

[0009] The beneficial effects of this application are at least as follows: By constructing an initial parameter set using historical process parameters and corresponding process result data, and establishing a mapping relationship between process parameters and process results based on a parameter mapping model, the nonlinear correlation between various process parameters and the influence weight of each parameter on the process results can be more accurately characterized. This provides a more reliable data foundation for subsequent parameter optimization and reduces the blindness caused by manual trial and error. Adjusting the parameter search range based on correlation and influence weights, and sampling the adjusted parameter search space to generate an expanded parameter exploration set, can narrow the invalid search area and increase attention to high-value parameter regions, thereby improving the search efficiency and targeting of the parameter optimization process. Evaluating parameter combinations in the parameter exploration set using a genetic algorithm, and updating the parameter exploration set that does not meet the preset process objectives using crossover and mutation operations, can enhance the screening capability of high-performance parameter combinations, improve the global optimization capability in complex multi-parameter scenarios, and reduce the risk of getting trapped in local optima. By inputting candidate parameter combinations into an etching simulation environment for simulation verification, and determining the optimized parameter configuration based on the simulation results, parameter combinations can be pre-verified before actual deployment, reducing the time and resource consumption of repeated trial and error in the real environment. By constructing a closed-loop mechanism encompassing data modeling, spatial expansion, evaluation and screening, simulation verification, iterative updates, and deployment verification, continuous optimization of etching process parameters can be achieved. This improves the matching degree between parameter configurations and preset process objectives, enhancing the stability and reliability of optimization results. Furthermore, by generating a process deployment plan based on the final parameter set and conducting applicability verification in the target process execution environment, the applicability of parameter combinations in actual process scenarios can be further confirmed, improving the application effect and process consistency of the optimal parameter combination in production. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of the etching process parameter optimization method of this application; Figure 2 This is a verification diagram showing the effect of the neural network model in this application on the prediction of etching rate; Figure 3 This is a graph showing the change in population fitness during the iteration process of the genetic algorithm in this application embodiment; Figure 4 This is a schematic diagram of the etching process parameter optimization system of this application. Detailed Implementation

[0012] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of the etching process parameter optimization method provided by the present invention. The flowchart specifically includes the following steps: S1. Extract historical process parameters and corresponding process result data from the preset database to construct an initial parameter set.

[0014] In one specific embodiment, the process of performing step S1 may specifically include the following steps: Extract historical process parameters, including gas flow rate, power, and chamber pressure, from a preset database, as well as process result data corresponding to the historical process parameters. The process result data includes at least etching rate and uniformity indicators. Based on historical process parameters and process result data, analyze the value distribution of each process parameter and the range of parameter combinations; Based on historical process parameters and their corresponding process results data, and combined with the value distribution and parameter combination range, an initial parameter set is constructed, and the initial distribution characteristics of each process parameter in the initial parameter set are extracted.

[0015] Specifically, a pre-defined database stores process recipe execution records, equipment operation records, and measurement result records associated with corresponding batches during the historical production process of the target etching process. Process recipe execution records include at least gas flow rate, power level, and chamber pressure, while measurement result records include at least etching rate and uniformity indicators. Process recipe execution records and measurement result records are linked through process recipe number, batch identifier, wafer identifier, or processing sequence identifier. Gas flow rate refers to the set flow rate value of each reactive gas or carrier gas during the etching process; power level refers to the set power or steady-state statistical power output by the RF power supply during the etching stage; chamber pressure refers to the statistical value of the steady-state pressure inside the chamber during the etching stage. The etching rate is calculated based on the pre-etch film thickness, post-etch film thickness, and corresponding process duration, such as (pre-etch film thickness - post-etch film thickness) ÷ process duration. Uniformity indices characterize the consistency of etching result distribution across different sampling locations on a wafer. They can be calculated based on the etching depth or remaining film thickness at multiple preset sampling points on the wafer surface. For example, the ratio of the standard deviation of the etching depth at each sampling point to the average etching depth can be used as a uniformity index; similarly, the ratio of the standard deviation of the remaining film thickness at each sampling point to the average remaining film thickness can be used as a uniformity index. By limiting the process result data to the physical quantities corresponding to the measurement results, the problem of a lack of clear physical correspondence between historical process parameters and output results is addressed. During execution of S1, historical batch records consistent with the target etching object, target material system, target chamber type, and target process mode are first selected from the preset database. Then, based on the associated fields, the corresponding historical process parameters and process result data are extracted, thereby forming the original process data table corresponding to each batch.

[0016] After the original process data table is generated, data preprocessing is performed. Data preprocessing includes at least missing value handling, abnormal batch removal, unit standardization, and steady-state interval extraction. Missing value handling involves marking batches lacking any key process parameter or corresponding process result data as invalid batches and removing them. Abnormal batch removal involves deleting batches with equipment alarms, mid-process shutdowns, vacuum anomalies, matching anomalies, temperature control anomalies, or measurement failure indicators from the original process data table. Unit standardization involves converting similar process parameters to the same unit of measurement. Steady-state interval extraction involves taking the statistical values ​​from the plasma stable phase as the historical process parameters for the batch when the equipment operation record contains time-series sampling values.

[0017] After obtaining preprocessed historical data, the distribution of values ​​and the range of parameter combinations for each process parameter are analyzed. For gas flow rate, power, and chamber pressure, the minimum, maximum, median, interquartile range, and frequency distribution are statistically analyzed for each historical batch. Historical batches are then grouped according to whether the process results meet the target requirements. Target requirements include at least a preset lower limit for etching rate and a preset upper limit for uniformity. For batches meeting the target requirements, the distribution range of the corresponding process parameters is statistically analyzed; for batches not meeting the target requirements, the deviation range of the corresponding process parameters is statistically analyzed. The distribution differences of the two types of batches for each process parameter are compared to determine the effective value range of each process parameter under the target etching result. For example, if a gas flow rate is concentrated in a sub-range in batches meeting the target requirements but discretely distributed in batches not meeting the target requirements, this sub-range is determined as the effective value range for that gas flow rate. Parameter distributions are differentiated according to process results to avoid including invalid ranges when determining parameter boundaries.

[0018] The parameter combination range is not a simple superposition of the upper and lower boundaries of individual parameters, but rather a range of combinations of multiple process parameters that occur together in historical production and yield corresponding process results. Gas flow rate, power level, and chamber pressure are divided into several intervals according to preset binning rules. The frequency of occurrence of each interval combination in historical batches and the distribution of its corresponding etching rate and uniformity indicators are statistically analyzed. For example, binning can be performed by equal width binning, equal frequency binning, or binning based on historical sample density. When the frequency of occurrence of a certain interval combination reaches a preset threshold, and the proportion of the corresponding batch meeting the target requirements reaches a preset proportion threshold, that interval combination is retained as a valid parameter combination range. When, although the individual parameters of an interval combination fall within their respective valid value ranges, the corresponding process results continuously deviate from the target requirements, that interval combination is excluded. For example, under high gas flow rate conditions, if the power level is not in the corresponding synergistic range, the etching rate decreases or the uniformity indicator deteriorates; therefore, this high flow rate and low power combination is not retained as a valid parameter combination range.

[0019] The initial parameter set consists of historical batches that meet the valid parameter combination range and have real measurement results. Each sample includes at least gas flow rate, power level, chamber pressure, etching rate, and uniformity index. When it is necessary to distinguish the process background, it may also include the etching object category, film thickness range, chamber status indicator, or batch temperature range. During construction, historical batches that fall within the valid parameter combination range and whose process results are reliable are directly written into the initial parameter set. Batches that have historical records but have measurement anomalies, equipment malfunctions, or parameters that exceed the valid parameter combination range are not included in the initial parameter set. The initial parameter set consists of historical data obtained from actual execution and measurement, and the sample is not expanded by interpolation based on nearest neighbor records or weighted averaging to generate synthetic process results.

[0020] The initial distribution characteristics include at least the central location characteristics, dispersion characteristics, and joint distribution characteristics of each process parameter. The central location characteristics are represented by the median or mean; the dispersion characteristics are represented by the interquartile range, standard deviation, or coefficient of variation; and the joint distribution characteristics are used to characterize the coordinated changes of multiple process parameters within the initial parameter set. For example, the joint distribution characteristics can be represented by a joint frequency distribution table or conditional distribution statistics, reflecting the dense regions of joint values ​​for each process parameter under the premise of meeting the target process results. Alternatively, the joint distribution characteristics can be represented by the parameter cluster boundaries obtained by density clustering of samples in the initial parameter set, with clustering inputs of gas flow rate, power magnitude, and chamber pressure, and clustering outputs of multiple dense parameter clusters and the corresponding process result distributions for each cluster.

[0021] S2. Train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process results based on the feature information.

[0022] In one specific embodiment, the process of performing step S2 may specifically include the following steps: A preliminary mapping relationship between process parameters and process results is constructed based on the initial parameter set; A neural network model is used to train the initial mapping relationship, capturing the interaction between various process parameters and the nonlinear correspondence between each process parameter and the process result; The feature vectors of each process parameter are extracted from the trained neural network model, and the correlation between each process parameter is quantified based on the feature vectors. Based on the quantified correlation, the influence weight of each process parameter on the process result is determined.

[0023] Specifically, the initial parameter set serves as the training samples for the parameter mapping model. Each sample includes at least parameters such as gas flow rate, power, and chamber pressure, along with corresponding etching rate and uniformity indices. When the target etching process is also affected by bias power, chamber temperature, substrate temperature, or etching time, the corresponding process parameters can also be included in the input samples. Normalization is performed on the input samples to ensure that process parameters of different dimensions are within similar numerical ranges, and the normalized parameters are retained so that subsequent prediction results can be restored to the actual process dimensions. For the output samples, while maintaining the physical meaning of the etching rate and uniformity indices, output terms with large numerical ranges are scaled to prevent any single output term from having an excessively high weight during training. Before the model output is used for comparison with process targets, the prediction results are restored to the actual process dimensions.

[0024] The initial mapping relationship is a one-to-one correspondence between combinations of process parameters and corresponding process results in historical samples. This involves using multiple process parameters from each sample as input records and their corresponding etching rate and uniformity indices as output records, organizing them into input and output datasets that the model can read. This initial mapping relationship characterizes the basic correspondence between process parameters and process results under historical etching conditions and is not output as the final quantization result. In this way, the scattered process records are transformed into trainable supervised learning samples.

[0025] A neural network model is used to train the initial mapping relationship. This model employs a multilayer perceptron structure, including an input layer, hidden layers, and an output layer. The input layer receives normalized process parameters, with the number of input nodes matching the number of process parameter terms. The hidden layer consists of at least two fully connected layers, connected by weight parameters, and each hidden layer uses a nonlinear activation function. The output layer outputs predicted etching rate and uniformity indices, with the number of output nodes matching the number of process result terms. For example, the hidden layer may have at least two fully connected nodes, using a modified linear unit function as the hidden layer activation function, and the output layer may use a linear output. If subsequent predictions of selectivity, sidewall angle, or critical dimension deviations are required, the number of output layer nodes can be expanded accordingly. By employing a multilayer nonlinear network structure, the problem of strong coupling of multiple parameters and the inadequacy of linear fitting to characterize complex process laws in etching processes is addressed.

[0026] During training, the training sample set is input into the neural network model in batches. Forward propagation yields predicted etching rate and uniformity index values. The loss value is calculated based on the difference between the predicted and actual values, and the connection weights and bias parameters of each layer are updated using the backpropagation algorithm. Since the output includes at least two process results—etching rate and uniformity index—the loss function adopts a multi-output joint loss form, i.e., the etching rate prediction error and the uniformity index prediction error are calculated separately and then summed according to preset weights. When the preset process target places higher requirements on uniformity than on etching rate, the weight of the uniformity error term in the joint loss can be increased. Model parameter updates can employ an adaptive moment estimation optimization algorithm. An initial learning rate, batch size, and number of training epochs are set. After each training epoch, the loss on the training sample set and the loss on the validation sample set are calculated separately. Training stops and the current model parameters are saved when the validation sample set loss no longer decreases within a preset number of epochs. If the samples meeting the target process requirements account for a low percentage of historical data, these samples can be assigned higher weights in the loss calculation to enhance the model's learning ability for the target process window.

[0027] After training, the neural network model is used to make predictions and verifications using a test sample set to obtain the etching rate prediction error and the uniformity index prediction error. When both are within the preset allowable deviation range, the trained neural network model is determined as the parameter mapping model. If the prediction error of at least one output term exceeds the preset allowable deviation range, the training phase is returned, and the number of hidden layers, the number of nodes per layer, the weight of the loss term, the learning rate or the sample weights are adjusted and retrained.

[0028] Instead of directly using the overall output of a hidden layer as the feature vector for a single process parameter, the feature information extracts the contribution information of each process parameter within the model's internal representation space. Specifically, perturbation analysis can be performed on each process parameter separately. That is, while keeping other process parameters constant, several perturbation points are selected within the historical valid value range of the target process parameter, and the changes in the hidden layer representation vector and output prediction result corresponding to each perturbation point are recorded. The changes in the hidden layer response and output response corresponding to each perturbation point are then concatenated in ascending order of the process parameter values, and the distribution information of the process parameter in the training samples is appended to the end of the concatenation result at a preset position to form the feature vector corresponding to that process parameter. For example, the same number of perturbation points are selected for each process parameter within its respective valid value range, and the same dimension of hidden layer response change vector and output response change vector is extracted for each perturbation point to ensure that the feature vector dimension corresponding to different process parameters is consistent.

[0029] Correlation relationships are used to characterize the degree of synergy, coupling, or mutual constraint among different process parameters when they influence the process outcome within the model. Specifically, the cosine similarity between the feature vectors of any two process parameters can be calculated and used as the correlation strength between them. A higher cosine similarity indicates that the two process parameters are closer in their direction of influence on the process outcome within the model; a lower or negative cosine similarity indicates that the two process parameters have different or constraining directions of influence within the model. When it is necessary to analyze different process outcomes separately, the corresponding parameter feature vectors can be extracted for the etching rate prediction path and the uniformity index prediction path respectively, and the correlation strength can be calculated separately to obtain the parameter correlation relationships for different process objectives.

[0030] Based on the quantified correlations, the influence weights of each process parameter on the process results are determined. Specifically, for the trained parameter mapping model, the gradient response of each process parameter to the predicted etching rate and uniformity index is calculated, and the absolute values ​​of the gradients over all training samples are averaged to obtain the basic influence value of the process parameter on the corresponding process result. Then, combined with the correlation strength between the process parameter and other process parameters, the basic influence value is corrected to obtain the final influence weight of the process parameter on the corresponding process result. For example, when a process parameter has a high gradient response to the etching rate and has a strong positive synergistic relationship with another key process parameter, the influence weight of the process parameter in the etching rate dimension is increased; when a process parameter responds to the uniformity index, but its effect depends on the cooperation of other process parameters, a correlation correction term is added while retaining its basic influence value to reflect the coupling effect. If subsequent steps require adjusting the search range for different process results, multiple sets of influence weights of each process parameter on etching rate and uniformity indicators should be retained. If subsequent steps use a single comprehensive indicator for spatial adjustment, the influence weights of each process parameter on different process results can be weighted and integrated according to the importance of different output items to the preset process objectives to obtain a comprehensive influence weight.

[0031] After the model training is completed, the neural network model is validated using a test sample set to obtain the prediction errors of etching rate and uniformity indices. For example... Figure 2 As shown, taking etching rate as an example, a scatter plot is created by comparing the actual etching rate of each sample in the test set with the model's predicted etching rate. Each point in the plot represents a test sample, the horizontal axis represents the actual measured value, the vertical axis represents the model's predicted value, and the dashed line represents the ideal prediction line. The data points are concentrated near the ideal prediction line, indicating that the neural network model has good predictive consistency for etching rate.

[0032] S3. Adjust the search range of each process parameter according to the correlation and influence weight, and sample the adjusted parameter search space to generate an expanded parameter exploration set.

[0033] In one specific embodiment, the process of performing step S3 may specifically include the following steps: Based on the correlation and influence weight, sensitivity analysis is performed on each process parameter to determine the sensitivity of each process parameter to the process results. The process parameters that affect sensitivity exceeding the preset sensitivity threshold are identified as target process parameters, and the search range of the target process parameters is adjusted. The adjusted parameter search space is sampled, and iterative sampling is performed until the parameter space coverage reaches the preset requirement, generating an expanded parameter exploration set. Extract the distribution characteristics of newly added parameter combinations in the parameter exploration set, and update the boundary conditions of the parameter search space based on the distribution characteristics.

[0034] Specifically, sensitivity analysis is first performed on each process parameter to determine its sensitivity to the process results. The sensitivity analysis is based on the parameter mapping model trained in S2. That is, while keeping the values ​​of other process parameters constant, the process parameter to be analyzed is subjected to a small perturbation within the effective value range determined in S1, and the changes in the predicted etching rate and uniformity index are calculated before and after the perturbation. These changes are normalized according to the fluctuation range of the corresponding process results in historical valid samples to obtain the etching rate response value and the uniformity response value. The etching rate response value is multiplied by the weight of the process parameter's influence on the etching rate to obtain the single-item influence sensitivity of the process parameter on the etching rate; the uniformity response value is multiplied by the weight of the process parameter's influence on the uniformity index to obtain the single-item influence sensitivity of the process parameter on the uniformity index. When the preset process objective simultaneously constrains the etching rate and uniformity index, the above product results are weighted and summed according to the preset objective weights to obtain the comprehensive impact sensitivity of the process parameter; when subsequent parameter searches need to be performed separately for different process results, the individual impact sensitivity for etching rate and the individual impact sensitivity for uniformity index are retained respectively.

[0035] The preset sensitivity threshold is set based on the distribution of sensitivity of each process parameter, and the mean, median, or high quantile of the sensitivity of all process parameters can be used as the judgment criterion. When multiple process parameters exceed the preset sensitivity threshold, they are collectively identified as target process parameters. When a process parameter does not exceed the preset sensitivity threshold individually, but its correlation strength with the target process parameter is higher than the preset correlation threshold, it can also be included as a linkage parameter in the target process parameter set. The search range of the target process parameters is adjusted not by proportionally enlarging the original upper and lower boundaries, but by determining the boundary offset and joint constraint conditions based on the magnitude of the influence sensitivity, the direction of action, and the correlation. For target process parameters with high influence sensitivity that still show benefits in improving the preset process objective at the edge of the current effective interval, their search boundary is appropriately extended in the favorable direction, and the extension range is determined according to the influence sensitivity of the process parameter. When the preset process objective simultaneously constrains multiple process results, the comprehensive influence sensitivity is used; when only a search is performed for a single process result, the single influence sensitivity is used. For target process parameters with high sensitivity but whose predicted etching rate is lower than a preset lower limit and / or uniformity index is higher than a preset upper limit after exceeding a certain range, the value range is defined as a restriction range, and the search boundary of the target process parameter is narrowed towards satisfying the preset process target. For non-target process parameters, the valid value range determined in S1 is maintained, or adjustments are made only when there is a strong coupling relationship with the target process parameter.

[0036] When adjusting the search range of target process parameters, the joint boundary is constrained based on the parameter correlation. When the correlation strength between two process parameters is higher than a preset correlation threshold, they are not considered as independent single-variable boundaries, but rather joint boundary conditions are established based on their corresponding cooperative or restrictive directions within the parameter mapping model. For example, when the gas flow rate increases, if S2 indicates that the power needs to be increased synchronously to keep the etching rate and uniformity within the target range, then the lower boundary of the power is increased while expanding the upper boundary of the gas flow rate; when increasing the chamber pressure will enhance the process fluctuation within a certain gas flow rate range, then the corresponding gas flow rate is restricted from entering the unfavorable range when expanding the chamber pressure search range. If necessary, the joint feasible region of multiple target process parameters can be represented as a parameter subspace that satisfies preset boundary constraints, and each boundary constraint is determined by the effective parameter combination range of S1 and the parameter correlation relationship of S2.

[0037] Sampling employs stratified random sampling or Latin hypercube sampling. When joint constraints exist between parameters, constraint sampling is performed within the parameter subspace that satisfies the joint boundary conditions. Specifically, the adjusted intervals of each target process parameter are divided into several sub-intervals according to a preset number of stratifications, and sampling points are selected within each sub-interval. For non-target process parameters, their original valid interval values ​​are maintained, or points are selected synchronously according to the linkage rules with the target process parameters. The sampling points of each parameter are combined into new parameter combinations, and parameter combinations that do not satisfy the joint boundary conditions are removed, resulting in a new parameter subset in the extended parameter exploration set. The new parameter subset is stored separately from the initial parameter set in S1, and each new parameter combination includes at least a process parameter item and a corresponding spatial source identifier.

[0038] During the sampling process, a coverage determination is performed on the adjusted parameter search space, and iterative sampling is conducted until the parameter space coverage reaches a preset requirement. Parameter space coverage represents the degree to which the currently added parameter combination covers the adjusted feasible parameter space. The adjusted parameter search space can be discretized into several feasible subspace units at a preset resolution (determined based on the target process window accuracy requirements and parameter interval width), and the proportion of feasible subspace units containing at least one sampling point to the total number of feasible subspace units is counted as the parameter space coverage. When the current coverage is lower than the preset coverage threshold, supplementary sampling is performed in uncovered areas or areas with low sampling density; when the current coverage reaches the preset coverage threshold, iterative sampling stops and an expanded parameter exploration set is output. If necessary, during the supplementary sampling process, the sampling probability can be increased for areas near the boundary, and the sampling probability can be decreased for areas that have formed high-density coverage but have not provided new boundary information, so that more new sampling points are distributed in the edge areas and low-coverage areas of the search space.

[0039] The distribution characteristics include at least the central location characteristics, dispersion characteristics, boundary clustering characteristics, and joint distribution characteristics of each process parameter in the newly added parameter subset. The median, quartile intervals, extreme value intervals, and stratified frequency distributions of the newly added parameter combinations across each process parameter dimension are statistically analyzed, and the combination frequency after joint stratification of different parameters is also statistically analyzed. When newly added parameter combinations form a significant cluster near a parameter boundary, it indicates that the area near the current boundary remains a key exploration area; when newly added parameter combinations are sparse or empty in a certain sub-interval for a long period, it indicates that the necessity of that sub-interval as a subsequent key search area is low; when newly added parameter combinations form a striped or clustered distribution within a certain joint interval, it indicates that multiple process parameters have a stable synergistic relationship within that joint interval.

[0040] The boundary conditions of the parameter search space are updated based on distribution characteristics. If a newly added parameter combination maintains a high distribution density near the expansion boundary of a target process parameter, and the boundary still satisfies the effective parameter combination range of S1 and the parameter association constraint of S2, then the boundary is retained or expanded further in that direction. If a newly added parameter combination remains sparse near an expansion boundary, or if the region where the boundary is located conflicts with the joint constraint conditions, then the boundary is reverted to the vicinity of a high-density region. If a newly added parameter combination forms a stable cluster within a certain linkage parameter combination interval, then the boundary of the cluster is updated as a new joint boundary condition to constrain the subsequent candidate parameter evaluation range and parameter subset update range.

[0041] S4. Evaluate the parameter combinations in the parameter exploration set according to the preset process objectives, and select candidate parameter combinations.

[0042] In one specific embodiment, the process of performing step S4 may specifically include the following steps: A genetic algorithm is used to evaluate the fitness of each parameter combination in the parameter exploration set according to the preset process target; Based on the evaluation results, calculate the performance score corresponding to each parameter combination; Parameter combinations that meet the preset performance requirements are selected as high-performance parameter combinations, and a candidate parameter combination list is generated. Sort the parameter combinations in the candidate parameter combination list to obtain candidate parameter combinations arranged by priority.

[0043] Specifically, the parameter exploration set generated in S3 is used as the set of parameters to be evaluated, and each parameter combination in the parameter exploration set is encoded as an individual in the genetic algorithm. The gene positions of the individual correspond sequentially to the values ​​of each process parameter; when the target etching process is also affected by bias power, temperature, or etching time, the corresponding process parameters can also be incorporated into the individual encoding, but should be consistent with the parameter combination field output by S3. Since S3 has completed the parameter search range adjustment and parameter space sampling, the genetic algorithm in S4 is used to perform fitness evaluation and optimal screening on the existing parameter combinations in the parameter exploration set, without generating a new parameter space without constraints.

[0044] The preset process objectives include at least etching rate and uniformity targets. In actual process scenarios, additional parameters such as ratio, sidewall angle, critical dimension deviation, or surface smoothness may be required. Therefore, a tiered approach can be adopted based on the predictability and evaluation stage of different process objectives: process objectives that can be stably predicted using the parameter mapping model trained in S2 are incorporated into the fitness evaluation in S4; process objectives that rely more on the morphological evolution of the etching process, surface contour changes, or mechanism simulation results are further evaluated in the subsequent etching simulation verification stage in S5. Thus, S4 is primarily used for rapid prediction and screening of parameter combinations, while S5 is primarily used for high-fidelity simulation verification of candidate parameter combinations, ensuring the hierarchy and coherence of the evaluation process.

[0045] Fitness evaluation is preferably performed based on the parameter mapping model trained by S2. Each parameter combination is input into the parameter mapping model to obtain the corresponding predicted process result. The predicted process result includes at least the predicted etching rate and uniformity index. When the parameter mapping model outputs other predicted process results, these can also be used as inputs for fitness evaluation. The predicted process result is compared with the preset process target to form the fitness value for that parameter combination.

[0046] When constructing the fitness function, the deviation between the predicted etching rate and the target etching rate, and the deviation between the predicted uniformity index and the target uniformity index are calculated separately. These deviations are then normalized according to their respective allowable deviation ranges to obtain normalized deviation values ​​for etching rate and uniformity. Based on the importance of the preset process objectives, the normalized deviation values ​​of different process outcome items are weighted and combined to form a single fitness value. If the predicted etching rate of a parameter combination is not lower than the preset lower limit and the predicted uniformity index is not higher than the preset upper limit, the fitness value of that parameter combination is increased. If a parameter combination only satisfies some process objectives, or causes the etching rate to be lower than the preset lower limit and / or the uniformity index to be higher than the preset upper limit, its fitness value is reduced through a constraint penalty term. If the preset process objective adopts a target interval control method, the fitness function consists of a target matching term and a constraint penalty term. The target matching term measures the closeness of the parameter combination to the preset process objective, while the constraint penalty term reduces the fitness of parameter combinations that exceed the allowable process range.

[0047] During fitness evaluation, the genetic algorithm uses the parameter combinations in the parameter exploration set as the initial population and calculates the fitness value for each individual in the initial population; then, it performs an individual selection operation within the population based on the fitness value. Since S6 is already limited to updating the parameter subset in the parameter exploration set through crossover and mutation operations of the genetic algorithm when the preset process target is not achieved, the genetic algorithm in S4 preferably only performs individual fitness calculation and selection operations, without performing crossover and mutation processing that changes the boundary conditions of the parameter exploration set; or, in one embodiment, a lightweight selection operation that does not change the boundary of the output space of S3 can be performed in S4, that is, only individuals with higher fitness are retained to enter the candidate set, while the crossover and mutation operations are reserved for S6.

[0048] The performance score characterizes the overall matching degree of parameter combinations to the preset process objectives. It can be directly mapped from the fitness value, or it can be further obtained by combining the fitness value with the process constraint satisfaction, boundary risk constraints, and process margin constraints. For any parameter combination, its corresponding fitness value is read, and it is determined whether the parameter combination satisfies the boundary conditions of the parameter search space updated in S3 and the valid parameter combination constraints determined in S1. If satisfied, the fitness value is determined as the base score, and it is further determined whether the parameter combination is located in the boundary proximity region or barely satisfies the preset process objectives. The boundary proximity region is determined based on the distance between the values ​​of each process parameter in the parameter combination and the boundary of the parameter search space updated in S3. When the distance is less than the preset boundary distance threshold, a boundary penalty is triggered. Barely satisfying the preset process objectives means that the difference between the predicted etching rate and the preset lower limit is less than the preset rate margin threshold and / or the difference between the predicted uniformity index and the preset upper limit is less than the preset uniformity margin threshold. In this case, a margin penalty is triggered. Boundary penalty items and margin penalty items are determined based on the boundary distance and target margin, respectively. The corresponding penalty items are deducted from the base score to obtain the performance score of the parameter combination. The boundary penalty term increases as the distance decreases, and the margin penalty term increases as the margin decreases. If the conditions are not met, the parameter combination is determined to be an invalid parameter combination and directly removed from the subsequent candidate parameter combination screening process. If necessary, the scores of parameter combinations that satisfy the constraints can be normalized to form performance scores under a unified dimension.

[0049] The preset performance requirements are set based on the distribution of performance scores for all parameter combinations. Parameter combinations that are above the preset performance threshold, rank among the top few, or are above the preset percentile can be used as high-performance parameter combinations. When the parameter exploration set is large, a screening method combining the preset performance threshold and the upper limit of the number of candidate parameter combinations is preferred. That is, first retain parameter combinations with performance scores above the threshold, and then limit the number of candidate parameter combinations within this set to no more than the preset upper limit to control the computational cost of etching simulation verification in S5. For parameter combinations with high performance scores but obvious risks of boundary conflicts or joint constraint conflicts, they can be removed from the high-performance parameter combinations or marked as risk in the candidate parameter combination list. Joint constraint conflict refers to a parameter combination that, although satisfying the single-parameter boundary conditions of each process parameter individually, does not satisfy the joint boundary conditions or linkage constraint conditions of multiple process parameters determined in the previous steps.

[0050] The ranking criteria include at least performance scores. When multiple parameter combinations have the same or similar performance scores, their deviations from the preset process targets are compared, and parameter combinations with smaller etching rate deviations and smaller uniformity deviations are prioritized. If the deviations are still similar, parameter combinations that better meet the boundary conditions of S3 and are closer to the historical high-quality parameter range in S1 are further prioritized. Thus, the candidate parameter combinations output by S4 are a set of parameter combinations that meet the basic process targets and have a high predictive matching degree, which are then used by S5 to further verify and screen morphology-related process targets such as flatness based on simulation results.

[0051] like Figure 3 The graph shows the change in population fitness during the iterative process of the genetic algorithm. The horizontal axis represents the number of iterations, and the vertical axis represents the fitness value. The solid line represents the optimal fitness in the current generation, and the dashed line represents the average fitness in the current generation. As the iteration progresses, the optimal fitness generally shows an upward trend, and the average fitness also gradually increases, indicating that the degree of matching between the parameter combinations and the preset process objective gradually improves in the individual population corresponding to the parameter exploration set. Based on this fitness change result, parameter combinations with higher fitness can be selected from the current population as candidate parameter combinations.

[0052] S5. Input the candidate parameter combination into the preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results.

[0053] In one specific embodiment, the process of performing step S5 may specifically include the following steps: The candidate parameter combinations are input into a preset etching simulation environment to perform virtual etching simulation and obtain the corresponding simulation results. Based on the simulation results, the uniformity index and flatness index corresponding to each candidate parameter combination are extracted. Based on the uniformity and flatness indices, the performance of each candidate parameter combination is analyzed. Based on the analysis results, the combination with the highest matching degree with the preset process target is selected from all candidate parameter combinations and used as the optimized parameter configuration.

[0054] Specifically, according to the priority obtained from S4, candidate parameter combinations are sequentially input into the preset etching simulation environment for virtual etching simulation. Candidate parameter combinations include at least gas flow rate, power level, and chamber pressure. When the target etching process is also affected by bias power, temperature, etching time, or gas ratio, the corresponding process parameters can also be included as simulation inputs. The preset etching simulation environment is used to perform virtual etching simulations on the target etching object, target material system, target chamber type, and target process mode. Its inputs include not only the process parameters in the candidate parameter combinations but also the initial state parameters of the workpiece corresponding to the actual process scenario and the simulation boundary conditions. The initial state parameters of the workpiece include at least the target film type, initial thickness of the target film, wafer size, pattern density, or target structural feature size. The simulation boundary conditions include at least the target chamber type, gas type configuration, initial pressure conditions, plasma interaction region setting, and simulation time step setting. The etching simulation environment is preferably a two-dimensional or three-dimensional etching simulation environment based on a physical mechanism model. It includes at least a plasma interaction solution unit, a surface reaction solution unit, and a morphology update unit. Based on the gas supply state, energy input state, and pressure state corresponding to the candidate parameter combination, it simulates the removal process of the target film, the surface contour change process, and the etching distribution change process at different locations on the wafer surface within a preset simulation time interval, thereby outputting simulation result data corresponding to the candidate parameter combination.

[0055] The simulation results include at least etching depth data at different sampling locations on the wafer surface, surface height distribution data after etching, and surface morphology data after the target film is removed. When evaluating sidewall angles, critical dimension deviations, or selectivity, corresponding contour geometry data and remaining material thickness data can be further output. The uniformity index characterizes the consistency of the etching depth or remaining film thickness distribution within the wafer. It can be calculated based on the etching depth data from multiple surface sampling points, taking the average, standard deviation, or range of the etching depth at each sampling point. The ratio of the standard deviation to the average etching depth, or the ratio of the range to the average etching depth, is determined as the uniformity index. The flatness index characterizes the degree of height undulation of the local contour of the surface after etching. It can be calculated based on the surface height distribution data in the simulation results; for example, the root mean square roughness of the surface height after etching can be used as the flatness index; or, the maximum peak-to-valley difference or the average height deviation of the surface height in a local area can be used as the flatness index.

[0056] Performance analysis uses the physical indices corresponding to the simulation results as the evaluation basis to re-evaluate the feasibility of each candidate parameter combination under the target etching scenario. First, the uniformity index corresponding to each candidate parameter combination is compared with the preset uniformity target, and the corresponding flatness index is also compared with the preset flatness target. The target deviations of the candidate parameter combinations in the uniformity and flatness dimensions are calculated respectively. When the uniformity index and flatness index of a candidate parameter combination are not higher than the preset uniformity upper limit and the preset flatness upper limit, it is considered a parameter combination that meets the simulation verification conditions. When a candidate parameter combination performs well in one index but exceeds the allowable range in another index, its overall matching degree is reduced. If necessary, the etching rate target deviation retained in S4 can also be used as an auxiliary analysis item, forming a comprehensive evaluation set together with the uniformity and flatness indices to avoid ignoring processing efficiency indices based solely on surface quality indices. To achieve a unified comparison among candidate parameter combinations, the deviations of the uniformity index from the preset uniformity target and the deviations of the flatness index from the preset flatness target are first normalized to obtain normalized deviation values ​​for uniformity and flatness. Based on the importance of the preset process target to uniformity and flatness, the normalized deviation values ​​are weighted and combined to form the simulation matching degree value for the corresponding candidate parameter combination. When the preset process target requires both stable intra-wafer distribution and smooth surface morphology, a dual-index joint evaluation is used; when the preset process target is more biased towards surface morphology control, the weight of the flatness index in the matching degree calculation is increased. If a candidate parameter combination falls within the preset target range for both the uniformity and flatness indices, its matching degree value is increased; if any index deviates from the target range, its matching degree value is reduced through a penalty term. If processing efficiency requirements need to be considered, the etching rate target deviation is included as an auxiliary term in the matching degree evaluation rules, but the evaluation structure of the uniformity and flatness indices as the main analysis objects remains unchanged.

[0057] First, candidate parameter combinations that meet the preset target requirements for both uniformity and flatness indices are selected. Within this set, they are sorted by simulation matching degree, with the parameter combination having the highest matching degree value being selected as the optimized parameter configuration. When multiple candidate parameter combinations have the same or similar simulation matching degree values, their process margins are compared, prioritizing the parameter combination with a larger distance from the uniformity and flatness target boundaries. If the process margins are still similar, the parameter combination with higher priority in S4 is prioritized.

[0058] In one implementation, the preset etching simulation environment includes a parameter input unit, an etching process solution unit, and a result extraction unit. The parameter input unit receives each process parameter from the candidate parameter combination, as well as the target film layer, target chamber, and initial boundary conditions. The etching process solution unit simulates the virtual etching process corresponding to the candidate parameter combination according to a preset etching mechanism or numerical solution rules, and outputs etching depth distribution data and surface contour data. The result extraction unit extracts uniformity and smoothness indices based on the etching depth distribution data and surface contour data, and outputs the corresponding index results. If the simulation environment uses a numerical mesh solution method, the simulation region division method, time progression method, and output sampling method are further recorded. If the simulation environment uses a pre-established mechanism model, the input and output items received by the mechanism model are further recorded.

[0059] S6. Determine whether the optimized parameter configuration meets the preset process target. If not, update the parameter subset in the parameter exploration set according to the simulation results and return to S4. If yes, determine the optimized parameter configuration as the final parameter set.

[0060] In one specific embodiment, the process of performing step S6 may specifically include the following steps: Obtain the simulation results corresponding to the optimized parameter configuration as the verification simulation results, and judge whether the optimized parameter configuration has achieved the preset process target based on the verification simulation results. If the target is not reached, the parameter subset in the parameter exploration set is updated by crossover and mutation operations of the genetic algorithm based on the verification simulation results. The parameter combination range of the updated parameter subset is adjusted, the parameter combination to be evaluated is regenerated, and S4 is returned to evaluate the parameter combination to be evaluated. If this is achieved, the optimized parameter configuration will be determined as the final parameter set.

[0061] Specifically, the simulation results corresponding to the optimized parameter configuration determined in S5 are obtained, and these simulation results are used as verification simulation results. The verification simulation results include at least the uniformity index and flatness index corresponding to the optimized parameter configuration. When processing efficiency needs to be taken into account, the etching rate index may also be included.

[0062] The preset process objectives include at least uniformity and flatness objectives. When processing efficiency needs to be considered, an etching rate objective is also included. The uniformity index in the verification simulation results is compared with the preset uniformity upper limit, and the flatness index is compared with the preset flatness upper limit. When etching rate needs to be constrained simultaneously, the etching rate index is compared with the preset etching rate lower limit. If the uniformity index is not higher than the preset uniformity upper limit, the flatness index is not higher than the preset flatness upper limit, and the etching rate is not lower than the preset lower limit when etching rate needs to be constrained, then the optimized parameter configuration is determined to have achieved the preset process objectives; otherwise, the optimized parameter configuration is determined to have not achieved the preset process objectives.

[0063] The parameter exploration set is a set of parameter combinations generated in S3 and continuously maintained in subsequent iterations. The parameter subset is a set of parameter combinations selected from the parameter exploration set for use in this iteration update. Each element in the parameter subset is a complete set of parameter combinations, not a single process parameter. The parameter subset preferably consists of at least some of the following: parameter combinations that are close to the currently optimized parameter configuration in the parameter space but have not yet met the preset process target; candidate parameter combinations with high priority in S4 and high simulation matching degree in S5 but not yet met the target; and parameter combinations that have a boundary proximity relationship or coupling relationship with the parameter combinations that have not met the target.

[0064] When it is determined that the current optimized parameter configuration does not meet the preset process target, in order to support subsequent stratification, screening, and genetic updates of the parameter subset, it is necessary to obtain the subset simulation results corresponding to each parameter combination in the parameter subset. The subset simulation results can at least partially come from the existing simulation results of the candidate parameter combinations in S5; for parameter combinations in the parameter subset that do not yet have corresponding simulation results, they are input into the etching simulation environment for supplementary simulation to obtain the corresponding subset simulation results. Therefore, the data foundation used for iterative updates in S6 includes not only the verification simulation results corresponding to the current optimized parameter configuration, but also the subset simulation results corresponding to each parameter combination in the parameter subset.

[0065] After determining the parameter subset and obtaining the corresponding subset simulation results for each parameter combination, the parameter combinations in the parameter subset are marked for retention, elimination, and update based on the subset simulation results. Specifically, the target deviation for each parameter combination in the parameter subset is calculated. The target deviation includes at least the deviation of the uniformity index relative to the preset uniformity target and the deviation of the flatness index relative to the preset flatness target; when processing efficiency needs to be considered, the deviation of the etching rate relative to the preset etching rate target is also included. The parameter combinations in the parameter subset are stratified according to the target deviation: the preset deviation stratification threshold is set according to the distribution of the target deviation for each parameter combination in the parameter subset, and the median, quartile, or preset quantile interval can be used as the stratification benchmark; parameter combinations with target deviations lower than the lower stratification threshold are marked as retained parameter combinations, parameter combinations with target deviations higher than the higher stratification threshold are marked as eliminated parameter combinations, and parameter combinations in between that have further search value are marked as parameter combinations to be updated. Retained parameter combinations will have their parameter structure retained in subsequent cross-operations; eliminated parameter combinations will no longer participate in this round of cross-mutation; parameter combinations to be updated will be the main update targets in this round of cross-mutation.

[0066] After the hierarchical processing is completed, the parameter subset is updated through crossover and mutation operations using a genetic algorithm. Crossover and mutation operations apply only to the retained parameter combinations and the parameter combinations to be updated. First, parent parameter combinations are selected from the retained and updated parameter combinations according to the rule that the simulation matching degree represented by the corresponding subset simulation results is high and the target deviation is small. A crossover operation is performed on the selected parent parameter combinations, causing the values ​​of corresponding process parameters in different parent parameter combinations to be recombinated according to a preset crossover probability to generate child parameter combinations. Then, a mutation operation is performed on the generated child parameter combinations, causing at least one process parameter in the child parameter combinations to undergo local perturbation under a preset mutation probability, thus forming the updated parameter subset. Local perturbation is constrained by the influence weight obtained in S2 and the influence sensitivity obtained in S3. That is, a larger local adjustment range is allowed for process parameters with high influence sensitivity, and a smaller local adjustment range is used for process parameters with low influence sensitivity. When the current optimized parameter configuration is close to the preset process target, the mutation amplitude is reduced to focus the update process on local fine-tuning. When the current optimized parameter configuration deviates significantly from the preset process target, the mutation amplitude is appropriately increased to expand the search range.

[0067] After crossover mutation, the parameter combination range of the updated parameter subset is further adjusted. This adjustment does not redefine the global search space of the entire parameter exploration set, but rather corrects the range of the local region containing the updated parameter subset based on the corresponding subset simulation results. Specifically, when subset simulation results indicate that a certain process parameter's value range will consistently cause the uniformity index to exceed the preset uniformity upper limit or the smoothness index to exceed the preset smoothness upper limit, the corresponding value range of that process parameter in the updated parameter subset is compressed. When subset simulation results indicate that a certain process parameter may still improve the uniformity or smoothness index near the current boundary, the local range of the process parameter is expanded in a limited manner along that direction. When there is a strong correlation between multiple process parameters obtained by S2 quantization, the joint boundary conditions of the related process parameters are simultaneously corrected when adjusting the parameter combination range to avoid generating invalid parameter combinations that do not satisfy the S3 constraint logic. By performing local range correction on the updated parameter subset, the problem of parameter combinations generated after crossover mutation potentially deviating from the effective process window is resolved.

[0068] After updating the parameter subset and adjusting the parameter combination range, the updated parameter subset is merged with the remaining parameter combinations that have not been eliminated from the parameter exploration set to form a new set of parameter combinations to be evaluated. If necessary, deduplication, boundary filtering, and constraint consistency checks can be performed on the set of parameter combinations to be evaluated based on the distribution characteristics of the updated parameter subset to ensure that the parameter combinations returned to S4 still satisfy the valid parameter combination range constraints of S1 and the boundary conditions updated in S3. The new set of parameter combinations to be evaluated is then re-entered into S4, and fitness evaluation, performance score calculation, candidate parameter combination screening, and priority ranking are performed according to the preset process objectives before proceeding to S5 for the next round of simulation verification.

[0069] When the optimized parameter configuration meets the preset process target, it is determined as the final parameter set. The final parameter set includes at least one or more parameter combinations that meet the preset process target; if the target etching process is also affected by bias power, temperature, etching time, or gas ratio, the corresponding process parameters can also be written into the final parameter set.

[0070] S7. Generate a process deployment plan based on the final parameter set, and verify its applicability to obtain the optimal parameter combination.

[0071] In one specific embodiment, the process of performing step S7 may specifically include the following steps: A process deployment plan is generated based on the final parameter set. The applicability of the process deployment plan is then verified in the target process execution environment, and the corresponding verification log data is obtained. Based on the verification log data, the applicability of the process deployment scheme in the target process execution environment is analyzed. When the analysis results show that the process deployment scheme meets the preset process objectives, the parameter combination corresponding to the process deployment scheme is determined as the optimal parameter combination.

[0072] Specifically, the final parameter set is the set of parameters retained after multiple rounds of evaluation, simulation verification and iterative optimization from S4 to S6. Each element is an executable parameter combination, and the optimal parameter combination is the specific parameter combination further confirmed from the final parameter set after the applicability verification is completed in the target process execution environment.

[0073] The process deployment plan should at least include the parameter combination to be deployed, the target process execution environment identifier, the target etching object identifier, the execution batch arrangement, equipment settings, and verification and monitoring items. Equipment settings should at least include gas flow rate settings, power settings, and chamber pressure settings. If the target etching process is also affected by bias power, temperature, etching time, or gas ratio, the corresponding process parameters can also be included in the process deployment plan. Verification and monitoring items should at least include etching rate monitoring, uniformity monitoring, flatness monitoring, and equipment operating status monitoring. The process deployment plan can be further organized into process recipe information executable by the etching equipment, including the process sequence, parameter settings, process duration, chamber temperature settings, and gas type and flow rate ratio.

[0074] The target process execution environment refers to the actual production equipment environment or pilot-scale verification environment that matches the target etching process. Its equipment type, chamber structure, film layer object, process mode, and monitoring conditions are consistent with or comparable to the simulation verification scenario in S5. During applicability verification, the parameter combinations in the process deployment plan are loaded into the target process execution environment. The target wafer or test wafer is loaded, and the etching process is executed according to the process deployment plan. During execution, equipment operation data and process result data are recorded in real time to form verification log data. Verification log data includes at least the process execution start time, execution batch identifier, actual gas flow rate output value, actual power output value, actual chamber pressure monitoring value, temperature change information, post-etching film thickness measurement results, etching depth measurement results at multiple sampling locations within the wafer, post-etching surface contour measurement results, and equipment alarm information or abnormal shutdown information. When monitoring process stability is required, result fluctuation data between different batches can also be recorded.

[0075] The applicability analysis includes at least execution consistency analysis, process result compliance analysis, and operational stability analysis. Execution consistency analysis determines whether the set parameters in the process deployment plan are correctly loaded and stably output in the target process execution environment. This is determined by comparing the deviations between the actual output gas flow rate, actual output power, and actual monitored chamber pressure in the verification log and the corresponding set values ​​in the process deployment plan. If the deviations are within the allowable error range, the execution consistency requirement is met. Process result compliance analysis determines whether the etching results after the target process execution meet the preset process objectives. Based on the film thickness measurement results, on-wafer etching depth measurement results, and surface profile measurement results in the verification log, the actual etching rate, actual uniformity index, and flatness index are extracted and compared with the preset lower limit of etching rate, preset upper limit of uniformity, and preset upper limit of flatness, respectively. If the actual etching rate is not lower than the preset lower limit, the actual uniformity index is not higher than the preset upper limit, and the flatness index is not higher than the preset upper limit, the process result is considered compliant. Operational stability analysis is used to determine whether there are abnormal fluctuations or equipment risks in the target process execution environment. It can be analyzed based on equipment alarm information, abnormal shutdown information and fluctuations of multiple batch measurement results in the verification log. If there are no alarm events of the preset type and the fluctuations of multiple batch results are within the allowable range, the operation stability is determined to meet the requirements.

[0076] When analyzing the applicability of process deployment schemes, applicability judgment rules can be constructed to achieve a unified comparison between different parameter combinations. Specifically, the deviations of the actual uniformity index from the preset uniformity target, the actual flatness index from the preset flatness target, and the actual etching rate from the preset etching rate target are calculated separately. These deviations are then normalized according to their respective allowable ranges to obtain normalized deviation values ​​for uniformity, flatness, and etching rate. Based on the importance of the preset process targets, the normalized deviation values ​​are weighted and combined to form the applicability matching degree value for the corresponding parameter combination. When the target process focuses more on intra-wafer distribution and surface morphology, the weights of the uniformity and flatness items are increased. When the target process also requires processing efficiency, the etching rate item is included as a parallel constraint or auxiliary evaluation item in the applicability matching degree calculation rules. If the parameter combination corresponding to a certain process deployment scheme meets the requirements in terms of execution consistency, process result achievement, and operational stability analysis, its applicability matching degree value is increased; if any one of these requirements is not met, its applicability matching degree value is decreased.

[0077] When a process deployment scheme meets the preset process objectives, its corresponding parameter combination is determined as the optimal parameter combination. From the final parameter set, parameter combinations that meet the requirements of execution consistency, process result compliance, and operational stability in the target process execution environment are selected. Within this set, they are sorted by their applicability matching value, and the parameter combination with the highest applicability matching value is selected as the optimal parameter combination. When multiple parameter combinations have the same or similar applicability matching values, their process margins are compared, and parameter combinations with larger distances from the preset uniformity target boundary, preset flatness target boundary, and preset etching rate target boundary are retained. If the process margins are still similar, parameter combinations that entered the final parameter set earlier in previous iterations are retained.

[0078] In one implementation, the process deployment scheme includes a parameter configuration unit, an execution deployment unit, and a log acquisition unit. The parameter configuration unit generates the combination of parameters to be deployed and their corresponding equipment settings based on the final parameter set. The execution deployment unit writes the equipment settings into the target process execution environment. The log acquisition unit collects equipment operation logs, measurement result logs, and abnormal event logs during the applicability verification process and outputs the log data as verification log data. If the target process execution environment includes a measurement equipment online interface, the verification log data can also directly read the film thickness measurement results and surface profile measurement results after etching. If the target process execution environment includes an equipment status monitoring module, it can also read power fluctuation, pressure fluctuation, and gas flow fluctuation logs to enhance the applicability analysis capabilities.

[0079] The etching process parameter optimization method in the embodiments of this application has been described above. The etching process parameter optimization system in the embodiments of this application is described below. Please refer to [link / reference]. Figure 4 The schematic diagram of the etching process parameter optimization system provided in this application shows that the system includes: The data extraction module 10 is used to extract historical process parameters and corresponding process result data from a preset database to construct an initial parameter set.

[0080] The modeling and quantification module 20 is used to train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process result based on the feature information.

[0081] The spatial expansion module 30 is used to adjust the search range of each process parameter according to the correlation and influence weight, and to sample the adjusted parameter search space to generate an expanded parameter exploration set.

[0082] The parameter evaluation module 40 is used to evaluate the parameter combinations in the parameter exploration set according to the preset process objectives and to screen out candidate parameter combinations.

[0083] The simulation verification module 50 is used to input the candidate parameter combination into a preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results.

[0084] The iterative optimization module 60 is used to determine whether the optimized parameter configuration meets the preset process target. If not, it updates the parameter subset in the parameter exploration set according to the simulation results and returns to S4. If yes, it determines the optimized parameter configuration as the final parameter set.

[0085] The deployment verification module 70 is used to generate a process deployment plan based on the final parameter set, perform applicability verification, and obtain the optimal parameter combination.

[0086] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for optimizing etching process parameters, characterized in that, The method includes: S1. Extract historical process parameters and corresponding process result data from the preset database to construct an initial parameter set; S2. Train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process results based on the feature information. S2 includes: Based on the initial parameter set, a preliminary mapping relationship between process parameters and process results is constructed. The preliminary mapping relationship is a one-to-one correspondence between the combination of process parameters in historical samples and the corresponding process results. Multiple process parameters in each sample are used as inputs, and the corresponding etching rate and uniformity index are used as outputs. The preliminary mapping relationship is trained using a neural network model to capture the interaction between various process parameters and the nonlinear correspondence between each process parameter and the process result; Feature vectors of each process parameter are extracted from the trained neural network model, and the correlation between each process parameter is quantified based on the feature vectors. Based on the quantified correlation, the influence weight of each process parameter on the process result is determined. For the trained parameter mapping model, the gradient response of each process parameter to the predicted etching rate and uniformity index is calculated, and the absolute value of the gradient on all training samples is averaged to obtain the basic influence value of the process parameter on the corresponding process result. Then, combined with the correlation strength between the process parameter and other process parameters, the basic influence value is corrected to obtain the final influence weight of the process parameter on the corresponding process result. S3. Adjust the search range of each process parameter according to the correlation and the influence weight, and sample the adjusted parameter search space to generate an expanded parameter exploration set; S3 includes: Based on the aforementioned correlation and influence weights, sensitivity analysis is performed on each process parameter to determine the sensitivity of each process parameter to the process results. The process parameters that affect sensitivity exceeding a preset sensitivity threshold are identified as target process parameters, and the search range of the target process parameters is adjusted. The adjusted parameter search space is sampled, and iterative sampling is performed until the parameter space coverage reaches the preset requirement, thereby generating the expanded parameter exploration set. Extract the distribution characteristics of newly added parameter combinations in the parameter exploration set, and update the boundary conditions of the parameter search space based on the distribution characteristics; S4. Evaluate each parameter combination in the parameter exploration set according to the preset process target, and screen to obtain candidate parameter combinations; S5. Input the candidate parameter combination into a preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results; S6. Determine whether the optimized parameter configuration meets the preset process target. If not, update the parameter subset in the parameter exploration set according to the simulation results and return to S4. If yes, determine the optimized parameter configuration as the final parameter set. S7. Generate a process deployment plan based on the final parameter set, and perform applicability verification to obtain the optimal parameter combination.

2. The method according to claim 1, characterized in that, S1 includes: Historical process parameters, including gas flow rate, power, and chamber pressure, and process result data corresponding to the historical process parameters are extracted from a preset database. The process result data includes at least etching rate and uniformity index. Based on the historical process parameters and the process result data, analyze the value distribution of each process parameter and the range of parameter combinations; Based on the historical process parameters and their corresponding process result data, and in combination with the value distribution and the parameter combination range, the initial parameter set is constructed, and the initial distribution characteristics of each process parameter in the initial parameter set are extracted.

3. The method according to claim 1, characterized in that, S4 includes: A genetic algorithm is used to evaluate the fitness of each parameter combination in the parameter exploration set according to the preset process objective. Based on the evaluation results, calculate the performance score corresponding to each parameter combination; The parameter combinations that meet the preset performance requirements are selected as high-performance parameter combinations, and then a candidate parameter combination list is generated. The parameter combinations in the candidate parameter combination list are sorted to obtain the candidate parameter combinations arranged by priority.

4. The method according to claim 1, characterized in that, S5 includes: The candidate parameter combinations are input into a preset etching simulation environment to perform virtual etching simulation and obtain the corresponding simulation results. Based on the simulation results, the uniformity index and flatness index corresponding to each candidate parameter combination are extracted. Based on the uniformity index and the flatness index, the performance of each candidate parameter combination is analyzed; Based on the analysis results, the combination with the highest matching degree with the preset process target is determined from each candidate parameter combination, and this combination is used as the optimized parameter configuration.

5. The method according to claim 1, characterized in that, S6 include: Obtain the simulation results corresponding to the optimized parameter configuration as the verification simulation results, and determine whether the optimized parameter configuration has achieved the preset process target based on the verification simulation results. If the target is not met, the parameter subset in the parameter exploration set is updated by crossover and mutation operations of the genetic algorithm based on the verification simulation results. The parameter combination range of the updated parameter subset is adjusted, the parameter combination to be evaluated is regenerated, and the evaluation of the parameter combination to be evaluated is returned to S4. If this is achieved, the optimized parameter configuration will be determined as the final parameter set.

6. The method according to claim 1, characterized in that, S7 includes: A process deployment scheme is generated based on the final parameter set, and the applicability of the process deployment scheme is verified in the target process execution environment to obtain the corresponding verification log data. Based on the verification log data, the applicability of the process deployment scheme in the target process execution environment is analyzed, and when the analysis results show that the process deployment scheme meets the preset process objective, the parameter combination corresponding to the process deployment scheme is determined as the optimal parameter combination.

7. An etching process parameter optimization system, used to implement the method as described in any one of claims 1 to 6, characterized in that, The system includes: The data extraction module is used to extract historical process parameters and corresponding process result data from a preset database to construct an initial parameter set; The modeling and quantification module is used to train the parameter mapping model based on the initial parameter set, establish the mapping relationship between process parameters and process results, extract the feature information of each process parameter, and quantify the correlation between each process parameter and the influence weight of each process parameter on the process result based on the feature information. The spatial expansion module is used to adjust the search range of each process parameter according to the correlation and the influence weight, and to sample the adjusted parameter search space to generate an expanded parameter exploration set. The parameter evaluation module is used to evaluate the combinations of parameters in the parameter exploration set according to the preset process objectives and to screen out candidate parameter combinations. The simulation verification module is used to input the candidate parameter combination into a preset etching simulation environment for simulation verification, and determine the optimized parameter configuration based on the simulation results; The iterative optimization module is used to determine whether the optimized parameter configuration meets the preset process target. If not, the parameter subset in the parameter exploration set is updated according to the simulation results and returned to S4. If yes, the optimized parameter configuration is determined as the final parameter set. The deployment verification module is used to generate a process deployment plan based on the final parameter set, perform applicability verification, and obtain the optimal parameter combination.

Images