Self-adaptive optimization method and device for GaAs etching process

By constructing a process performance evaluation model and using reinforcement learning, the etching process parameters of GaAs are optimized, solving the problem of the difficulty in accurately controlling the etching process parameters and realizing adaptive optimization and performance improvement of the etching process.

CN119005026AInactive Publication Date: 2024-11-22JIANGSU ETERN

Patent Information

Application Number
CN202411493912.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-24
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The difficulty in precisely controlling the etching process parameters of GaAs results in low adaptability to etching process optimization.

Method used

By introducing process environment information, a process performance evaluation model is constructed, reinforcement learning is performed, and control strategies are optimized and adjusted to achieve adaptive optimization.

Benefits of technology

This improves the optimization adaptability of GaAs etching processes, ensuring the accuracy and stability of etching results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005026A_ABST
    Figure CN119005026A_ABST
Patent Text Reader

Abstract

The invention discloses a self-adaptive optimization method and device for a GaAs etching process, and relates to the related field of semiconductor processes, and the method comprises the steps: introducing process environment information, carrying out an experiment through the GaAs etching process according to the process environment information, obtaining a process experiment result, and synchronizing the process experiment result to a process performance evaluation model, obtaining a process performance evaluation result, defining an etching process parameter set to be optimized, analyzing the etching process parameter set to be optimized based on the process environment information, setting a parameter optimization target, establishing a multi-dimensional learning space, and carrying out the reinforcement learning of the etching process parameter set to be optimized. And determining an optimization adjustment control strategy to carry out adaptive optimization on the GaAs etching process. The technical problem that the GaAs etching process is low in optimization fitness due to the fact that etching process parameters are difficult to control accurately is solved, and the technical effects that reinforcement learning is carried out according to process performance evaluation in combination with the learning space, and the optimization fitness of the GaAs etching process is improved are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of semiconductor technology, and in particular to an adaptive optimization method and device for GaAs etching technology. Background Art

[0002] GaAs is a compound semiconductor material commonly used in optoelectronics and high-frequency electronic devices. Due to its excellent properties such as high electron mobility and wide bandgap, it is widely used in high-speed circuits, photodetectors, lasers and other fields. The etching process of GaAs materials is generally divided into two categories: dry etching and wet etching. With the development of integrated circuits and optoelectronic technology, the precision requirements of GaAs etching process are getting higher and higher. In the existing technology, it is difficult to accurately control the etching process parameters, resulting in the technical problem of low optimization adaptability of GaAs etching process. Summary of the invention

[0003] The present application provides an adaptive optimization method and device for GaAs etching process, which solves the technical problem that etching process parameters are difficult to accurately control, resulting in low optimization adaptability of GaAs etching process, and achieves the technical effect of improving the optimization adaptability of GaAs etching process by combining process performance evaluation with learning space for reinforcement learning.

[0004] The present application provides an adaptive optimization method for a GaAs etching process. The method is applied to an adaptive optimization device for a GaAs etching process, comprising: introducing process environment information, conducting experiments using a GaAs etching process according to the process environment information, and obtaining process experiment results; constructing a process performance evaluation model, synchronizing the process experiment results to the process performance evaluation model, and obtaining process performance evaluation results; defining an etching process parameter set to be optimized according to the process performance evaluation results; analyzing the etching process parameter set to be optimized based on the process environment information, and setting a parameter optimization target; establishing a multidimensional learning space according to the parameter optimization target, performing reinforcement learning on the etching process parameter set to be optimized based on the multidimensional learning space, determining an optimization adjustment control strategy, and executing the optimization adjustment control strategy to adaptively optimize the GaAs etching process.

[0005] In a possible implementation, the process performance evaluation model is constructed and the following processing is performed: the historical etching data record log of the GaAs etching process is retrieved, and parameters are integrated according to the historical etching data record log to obtain an etching process parameter space; a plurality of initial sampling points are determined by random selection based on the etching process parameter space, and the plurality of initial sampling points contain etching process performance information of each initial sampling point; the plurality of initial sampling points are traversed to match with the process experiment results to construct a process experiment matching list; a plurality of process experiment results are extracted based on the process experiment matching list, and there is a corresponding relationship between the plurality of process experiment results and the plurality of initial sampling points; an acquisition function is used to perform an experimental prediction on the first initial sampling point to obtain a first experimental prediction result, and it is iterated until the traversal of the plurality of initial sampling points is completed and the plurality of experimental prediction results are determined; the plurality of experimental prediction results are added to the process experiment matching list for combination to obtain a process experiment comparison list; the process performance evaluation model is constructed based on the process experiment comparison list combined with the etching process performance information of each initial sampling point.

[0006] In a possible implementation, the process experiment results are synchronized to the process performance evaluation model to obtain the process performance evaluation results, and the following processing is performed: the process experiment results are used as index data, the process experiment comparison list is traversed to extract the target experiment prediction results, and there is a corresponding relationship between the target experiment prediction results and the process experiment results; calculations are performed based on the target experiment prediction results to obtain multiple performance indicator prediction values; according to the multiple performance indicator prediction values, the process experiment results are compared with the target experiment prediction results in combination with the process cycle to generate process error parameters; error analysis is performed based on the process error parameters, and the process performance evaluation model is incrementally learned according to the error analysis results, and evaluation is performed in combination with the learning results to obtain the process performance evaluation results.

[0007] In a possible implementation, error analysis is performed based on the process error parameters, and incremental learning is performed on the process performance evaluation model according to the error analysis results, and the following processing is performed: determine whether the process error parameter is greater than a preset error threshold, and if so, generate a learning instruction; perform process analysis on the process experiment results through the learning instructions to generate a process experiment analysis result; perform error analysis on the process error parameters through the learning instructions to generate a process error result; use the process error result combined with the process experiment analysis result as the input of incremental learning, update the data distribution of the process performance evaluation model, and generate the learning result.

[0008] In a possible implementation, reinforcement learning is performed on the etching process parameter set to be optimized based on the multidimensional learning space to determine the optimization adjustment control strategy, and the following processing is performed: the multidimensional learning space is parsed to determine the learning state space and the learning action space; simulation verification is performed according to the parameter optimization target, performance feedback parameters are extracted to construct the etching process performance target, and a reward function is set according to the etching process performance target; the process environment information is initialized to obtain the etching process simulation environment information; the etching process simulation environment information is synchronized to the learning action space according to the time step for selection, and a first learning action is determined; according to the first learning action, the etching process simulation environment information is synchronized to the learning state space for calculation, the learning state of the etching process parameter set to be optimized is updated, a first learning state is generated, and it is iterated in combination with the reward function, and the learning state is stopped when it is stable at a preset change threshold, and parameter adjustment information is obtained, and the parameter adjustment information includes parameter adjustment direction information and parameter adjustment size information; adjustment verification is performed according to the parameter adjustment direction information, and when the verification passes, the optimization adjustment control strategy is constructed based on the parameter adjustment size information.

[0009] In a possible implementation, the learning state space performs the following processing: performing a state analysis on the GaAs etching process based on the parameter optimization target to determine multiple key parameters of the etching process; matching the multiple key parameters of the etching process to multiple state space dimensions, and the multiple state space dimensions correspond to the multiple key parameters of the etching process; traversing the multiple key parameters of the etching process and performing an impact analysis on the parameter optimization target, screening the multiple state space dimensions according to the impact analysis results, and constructing a discrete state space and a continuous state space; defining the learning state space based on the discrete state space and the continuous state space.

[0010] In a possible implementation, the learning action space performs the following processing: based on the parameter optimization target, an action analysis is performed on the GaAs etching process to determine a plurality of adjustable process parameters, wherein the plurality of adjustable process parameters include parameter adjustment amounts; the plurality of adjustable process parameters are matched to a plurality of action space dimensions, and there is a corresponding relationship between the plurality of action space dimensions and the plurality of adjustable process parameters; the plurality of adjustable process parameters are traversed to perform an impact analysis on the parameter optimization target to determine an action impact direction; the plurality of action space dimensions are screened according to the action impact direction in combination with the parameter adjustment amount to construct a discrete action space and a continuous action space; and the learning action space is defined based on the discrete action space and the continuous action space.

[0011] The present application also provides an adaptive optimization device for a GaAs etching process, comprising: a data experiment module, the data experiment module is used to introduce process environment information, conduct experiments using a GaAs etching process according to the process environment information, and obtain process experiment results; a performance evaluation module, the performance evaluation module is used to construct a process performance evaluation model, synchronize the process experiment results to the process performance evaluation model, and obtain process performance evaluation results; a parameter definition module, the parameter definition module is used to define an etching process parameter set to be optimized according to the process performance evaluation results; a parameter analysis module, the parameter analysis module is used to analyze the etching process parameter set to be optimized based on the process environment information and set a parameter optimization target; a process optimization module, the process optimization module is used to establish a multidimensional learning space according to the parameter optimization target, perform reinforcement learning on the etching process parameter set to be optimized based on the multidimensional learning space, determine an optimization adjustment control strategy, and execute the optimization adjustment control strategy to adaptively optimize the GaAs etching process.

[0012] One or more technical solutions provided in this application have at least the following technical effects or advantages: The adaptive optimization method and device for GaAs etching process provided by the present application relate to the field of semiconductor process technology, solve the technical problem that etching process parameters are difficult to accurately control, resulting in low optimization adaptability of GaAs etching process, and achieve the technical effect of improving the optimization adaptability of GaAs etching process by combining reinforcement learning with learning space based on process performance evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solution of the embodiment of the present application, the accompanying drawings of the embodiment of the present application will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the device according to the embodiment of the present application. It should be understood that the previous or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0014] Figure 1 A schematic flow chart of an adaptive optimization method for GaAs etching process provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an adaptive optimization device for GaAs etching process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0016] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0017] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, device, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0018] The embodiment of the present application provides an adaptive optimization method for a GaAs etching process, wherein the method is applied to an adaptive optimization device for a GaAs etching process, such as Figure 1 As shown, the method includes: Step A100, introducing process environment information, and conducting experiments using the GaAs etching process according to the process environment information to obtain process experiment results; in order to obtain more accurate experimental results, it is necessary to more accurately control the etching process by introducing environmental parameters and monitoring conditions. The introduced process environment information is an important external factor affecting the GaAs etching process results. The introduced process environment information may include gas pressure, that is, the working gas pressure in the etching equipment, which directly affects the density and reaction rate of the plasma, and temperature, that is, the temperature of the reaction chamber, which affects the etching rate, surface finish and reactants. The volatility of the gas, gas composition and gas flow, that is, the type of etching gas (such as Cl2, BCl3, SF6, etc.) and the flow ratio of each gas will affect the formation of reaction products and etching selectivity. The RF power, that is, the radio frequency power affects the kinetic energy of ions in the plasma, and then affects the etching rate and surface morphology. The etching time, that is, the etching depth and material removal amount, is further designed based on the above process environment information. The experimental variables in the designed experimental plan can include different gas flow rates, RF powers, gas pressures, etching times and temperatures, and the parameters in the experimental plan are combined according to the parameters. , using GaAs etching equipment to conduct experiments, and then collecting process experimental data after the experiment, the process test data may include etching rate, surface roughness, side wall morphology, material thickness change, etching uniformity, and finally analyzing the collected process experimental data, which may include analyzing which combination can achieve a higher etching rate while maintaining a reasonable surface quality according to different gas flow and power combinations, and obtaining etching rate change experimental results, observing whether the surface roughness meets the expected standard through changes in temperature, gas pressure and RF power, and analyzing the factors affecting the surface finish, obtaining surface roughness influence experimental results, evaluating anisotropic etching effects, analyzing which conditions can produce more vertical side wall morphology, and obtaining side wall angle experimental results, analyzing the etching ratio of mask materials to GaAs materials, judging which gas combinations can improve material selectivity and reduce mask material loss, and obtaining selective etching experimental results, evaluating the uniformity of etching at different positions, ensuring that the process is suitable for large-area uniform etching requirements, and obtaining uniformity analysis experimental results, and integrating the above experimental results to obtain process experimental results, which serve as a basis for subsequent process adaptive optimization.

[0019] Execute step A200, build a process performance evaluation model, synchronize the process experiment results to the process performance evaluation model, and obtain a process performance evaluation result; in a possible implementation method, step A200 further includes step A210, calling a historical etching data record log of the GaAs etching process, integrating parameters according to the historical etching data record log, and obtaining an etching process parameter space; first, the historical etching data record log can be called based on the database of the etching equipment or the experimental management system, the purpose of which is to obtain the historical parameters and performance of the etching process, the historical etching data record log may include process parameters, namely gas flow, RF power, etching time, chamber pressure, temperature, etc., process performance, namely etching rate, surface roughness, sidewall morphology, etching selectivity, etc., and then classify and integrate according to process parameters according to the historical etching data record log, and create an etching process parameter space at the same time, the etching process parameter space may include gas flow, RF power, time, pressure and temperature, etc.

[0020] Execute step A220, randomly select based on the etching process parameter space, and determine multiple initial sampling points, wherein the multiple initial sampling points include etching process performance information of each initial sampling point; execute step A230, traverse the multiple initial sampling points to match with the process experiment results, and build a process experiment matching list; select a number of points from the etching process parameter space by random sampling or Latin hypercube sampling, and ensure that the multiple points can cover different parameter combination areas, so as to determine the multiple initial sampling points, and each sampling point needs to retain its corresponding etching process performance information, such as rate, roughness, etc., and then traverse the multiple initial sampling points to match with the process experiment results, which means traversing the sampling points one by one to search for corresponding experimental data from the experimental database. If there is a parameter combination of the sampling point that has appeared in the historical experiment, the matching relationship between the sampling point and the experimental result is recorded, and on this basis, a process experiment matching list is established, wherein the process experiment matching list includes the actual experimental result of each sampling point and its corresponding etching process performance.

[0021] Execute step A240 to extract multiple process experiment results based on the process experiment matching list, wherein the multiple process experiment results correspond to the multiple initial sampling points; execute step A250 to perform experimental prediction on the first initial sampling point using the acquisition function to obtain a first experimental prediction result, and iterate until the multiple initial sampling points are traversed and the multiple experimental prediction results are determined; execute step A260 to add the multiple experimental prediction results to the process experiment matching list for combination to obtain a process experiment comparison list; execute step A270 to construct the process performance evaluation model based on the process experiment comparison list combined with the etching process performance information of each initial sampling point.

[0022] Extract multiple experimental results from the process experiment matching list, the extracted multiple experimental results are the experimental results corresponding to the successfully matched sampling points, and then use the acquisition function to perform experimental prediction on the first initial sampling point, the first initial sampling point is any sampling point among the multiple initial sampling points, and experimental prediction is performed on each initial sampling point among the multiple initial sampling points at the same time, which means using acquisition functions such as expected improvement or confidence bound to determine the strategy for each prediction until all sampling points are traversed and predicted, and selecting the next sampling point to be predicted by using the acquisition function in Bayesian optimization, and determining the optimal experimental prediction direction at each selection according to the acquisition function, calculating the prediction result of each initial sampling point, and then iterating all the initial sampling points in turn until all the initial sampling points are predicted and stop, thereby obtaining multiple experimental prediction results.

[0023] Furthermore, combining multiple experimental prediction results with the process experiment matching list means merging multiple experimental prediction results with existing experimental matching data, comparing the actual results of each sampling point with the predicted results, and recording difference information such as errors, thereby forming a process experiment comparison list, which contains the actual experimental results and predicted results of each initial sampling point.

[0024] Further, the data in the process experiment comparison list is combined with the actual and predicted performance information to construct a process performance evaluation model, which means that the data in the process experiment comparison list is divided into a training set and a test set in proportion, and the model performance is evaluated using K-fold cross validation. The training data is further divided into K subsets, and one of the subsets is used for validation one by one, and the other subsets are used for training to ensure that the model has good generalization ability, and then the model is trained using the data in the training set. The model input is a combination of process parameters (such as gas flow, power, time, etc.), and the output is a process performance index (such as etching rate, roughness, etc.). At the same time, for linear models, the least squares method is used for parameter estimation. For nonlinear models such as neural networks, the back propagation algorithm and gradient descent method are used to adjust the model weights, minimize the error between the predicted value and the actual value, and define a suitable loss function to measure the prediction error of the model, and then tune the hyperparameters of the model, and then input the process parameter combination in the test set into the trained model to obtain the model's predicted value of the process performance under these parameter combinations, thereby establishing a performance evaluation model to evaluate the process experiment results and provide strong support for process control and improvement.

[0025] In a possible implementation, step A200 further includes step A280, taking the process experiment result as index data, traversing the process experiment comparison list to extract the target experiment prediction result, and the target experiment prediction result corresponds to the process experiment result; executing step A290, calculating according to the target experiment prediction result to obtain multiple performance index prediction values; executing step A2100, comparing the process experiment result with the target experiment prediction result in combination with the process cycle according to the multiple performance index prediction values, and generating a process error parameter; firstly, taking the process experiment result as index data means corresponding the process experiment result to a specific parameter combination, and the parameter combination can be obtained by combining the etching rate, surface roughness, side wall angle, etching selectivity, etc. according to the experimental content, and then traversing the process experiment comparison list, taking the process experiment result as an index, and finding the corresponding prediction result, and the parameter combination can be used as a matching condition to ensure that the extracted prediction result matches the experimental result in terms of process parameters, and after matching the corresponding experimental data, extracting the corresponding experimental prediction result, thereby determining the target experiment prediction result, so the target experiment prediction result corresponds to the process experiment result.

[0026] Furthermore, calculations based on the predicted results of the target experiment refer to using the predicted results of the target experiment to calculate performance indicators related to the etching process, such as etching rate, surface roughness, sidewall angle, and etching selectivity. Different calculation methods can be used for different process performance indicators. For example, the etching rate prediction value: the predicted rate value is obtained by dividing the etching depth by the time. The surface roughness prediction value: the predicted roughness value is obtained by simulation calculation of the surface morphology. The sidewall angle prediction value: the angle is calculated by analyzing the geometric characteristics of the sidewall shape in the model. The etching selectivity prediction value: the selectivity is calculated by the etching rate ratio of the material, thereby obtaining multiple performance indicator prediction values.

[0027] Furthermore, comparing the process experiment results with the target experiment prediction results in combination with the process cycle according to the predicted values ​​of multiple performance indicators means that at the end of each process cycle, the experimental results under the same conditions are compared with the corresponding prediction results one by one. The process cycle may refer to a data collection cycle after each experiment is completed. The error calculation may be obtained by subtracting the process experiment results from the target experiment prediction results.

[0028] Execute step A2110, perform error analysis based on the process error parameter, perform incremental learning on the process performance evaluation model according to the error analysis result, and evaluate in combination with the learning result to obtain the process performance evaluation result. In a possible implementation, step A2110 further includes step A2111, determining whether the process error parameter is greater than a preset error threshold, and if so, generating a learning instruction; executing step A2112, performing process analysis on the process experiment result through the learning instruction, and generating a process experiment analysis result; executing step A2113, performing error analysis on the process error parameter through the learning instruction, and generating a process error result; executing step A2114, using the process error result and the process experiment analysis result as the input of incremental learning, updating the data distribution of the process performance evaluation model, and generating the learning result.

[0029] By judging whether the current process error parameter exceeds the preset allowable range, it is decided whether to start the learning process. Therefore, the process error parameter obtained above is compared with the preset error threshold. The preset error threshold can be a reference value, which defines the maximum deviation allowed for the process result, and different performance indicators can have different thresholds. All process error parameters are traversed to judge whether each error parameter exceeds the preset error threshold. When it is detected that the error parameter exceeds the threshold, a learning instruction is generated to trigger subsequent process analysis and error analysis steps. The learning instruction can be a trigger to inform the system to enter the analysis and learning mode. The learning instruction drives the system to conduct in-depth analysis of the process experiment results and error parameters, and update the model.

[0030] Furthermore, based on the learning instructions, the process experiment data is analyzed, which refers to the process experiment data related to the learning instructions, including the process parameters in the historical experiments and their corresponding performance results, i.e., etching rate, surface roughness, etc., and the relevant experimental data are analyzed at the same time to evaluate which parameter combinations lead to large process errors. For example, when the gas flow rate is too high or the RF power is too low, it may lead to too fast etching rate or increased surface roughness. On this basis, the analysis results are generated, indicating that there are process parameters that need to be adjusted, and the process experiment analysis results are obtained.

[0031] Furthermore, based on the learning instructions, a detailed analysis of the generated process errors is performed, which refers to identifying the process conditions or process parameters where the process experiment results deviate from the target experiment prediction results. For example, if the error of the etching rate mainly comes from the RF power fluctuation, the RF power will be marked as a sensitive parameter, and then the error statistics of each performance indicator are calculated, which can be the mean, variance, standard deviation, etc. of the error. Through the error distribution, common error patterns are identified, and by analyzing the error sensitivity of different process parameters, the parameters that have the greatest impact on the error are identified. On this basis, the process error analysis results are generated to clarify the parameters that need to be solved first.

[0032] Finally, the process analysis results and error analysis results are combined and used as input data for incremental learning for model update, which means that the key process parameter adjustment suggestions identified in the process experimental analysis are combined with the error sources and sensitivity parameters identified in the error analysis to form complete input data, and the combined data is input into the incremental learning algorithm as the basis for model update. Since the online gradient descent method or recursive least squares method allows the model to be locally updated when new data is introduced without retraining the entire model, incremental updates such as online gradient descent method or recursive least squares method are used to fine-tune the model parameters based on the new experimental analysis and error analysis data. The learning algorithm will gradually reduce the error and improve the accuracy of the model in future predictions. The weights in the neural network model can be updated by introducing new experimental data. After the incremental learning is completed, new learning results are output, including the updated process performance evaluation model. This model can more accurately predict the etching performance under future process parameter combinations.

[0033] The process performance evaluation result is obtained by combining the learning results for evaluation, which means using the model updated by incremental learning to recalculate the performance prediction value under the process parameter combination, thereby outputting the etching rate, surface roughness, sidewall angle and selectivity prediction results corresponding to each process parameter combination, that is, the process performance evaluation result, to ensure that the prediction accuracy of the model improves with the increase of experimental data and the continuous deepening of learning, so as to better guide the optimization and adjustment of the GaAs etching process.

[0034] Execute step A300 to define a set of etching process parameters to be optimized based on the process performance evaluation results; first, perform multiple analyses on the process performance evaluation results obtained above, which may include sensitivity analysis and multi-factor analysis. The sensitivity analysis refers to performing sensitivity analysis on the evaluation results to determine which process parameters have a significant impact on changes in etching performance. For example, if the etching rate increases or decreases significantly with changes in gas flow rate, the gas flow rate is the key parameter affecting the etching rate. If the surface roughness is mainly affected by changes in temperature, the temperature is the key parameter affecting the surface quality. The multi-factor analysis refers to analyzing the interactive effects of multiple process parameters on performance using multi-factor analysis methods such as variance analysis or response surface methodology, identifying the comprehensive effects of multiple parameter combinations on process performance, and further selecting the parameters that have the greatest impact on etching performance based on the sensitivity analysis and multi-factor analysis results as the key parameters to be optimized to ensure that the etching process achieves the ideal performance target.

[0035] Execute step A400, analyze the etching process parameter set to be optimized based on the process environment information, and set a parameter optimization target; obtain important influencing factors on the etching process by analyzing different process environment information, the GaAs etching process includes a dry etching process and a wet etching process, and analyze the etching process parameter set to be optimized based on the process environment information, the process environment information includes dry etching environment information and wet etching environment information, and analyze the etching process parameter set to be optimized on this basis, which means that based on the dry etching environment information, the gas flow rate and gas composition, RF power, gas pressure and temperature can be analyzed, and the gas flow rate and gas composition refer to the ratio of gas flow rate and gas composition analyzed, Determine their influence on etching rate and selectivity. The RF power can improve the etching rate. By adjusting the gas pressure and temperature, the uniformity and sidewall morphology of etching can be optimized. Based on the wet etching environment information, the solution concentration, stirring speed and temperature control can be analyzed. That is, in wet etching, the concentration and stirring speed of the solution determine the etching rate and uniformity. At the same time, the solution temperature affects the reaction rate and etching uniformity. Furthermore, for each process parameter to be optimized (such as gas flow rate, solution concentration, etc.), a sensitivity analysis is performed to evaluate their influence on key performance indicators (etching rate, surface roughness, selectivity, etc.). Through numerical simulation or experimental analysis, the parameters with the most significant influence on the etching results are determined to provide a basis for setting optimization targets.

[0036] Furthermore, according to the analysis results of the process parameters to be optimized, specific parameter optimization targets are set. The parameter optimization targets can be divided into dry etching optimization targets and wet etching optimization targets. The dry etching optimization targets can include maximizing the etching rate, minimizing the surface roughness, optimizing the selectivity, etc. The wet etching optimization targets can include uniformity optimization, etching rate control, reducing the generation of by-products, etc. At the same time, when setting the optimization targets, multiple targets (such as rate and uniformity) need to be considered simultaneously. At this time, a multi-objective optimization method can be used, that is, through weight coefficients or Pareto frontier analysis, the optimal balance point between different targets can be found. For example, the gas flow rate and RF power can be adjusted to optimize the etching rate and selectivity at the same time, and the solution concentration and temperature can be adjusted to optimize the etching rate and surface uniformity at the same time. Setting the parameter optimization targets of the etching process parameter set to be optimized based on the dry etching optimization targets and the wet etching optimization targets plays a vital role in improving the efficiency, quality and accuracy of the etching process.

[0037] Next, execute step A500, establish a multidimensional learning space according to the parameter optimization target, perform reinforcement learning on the etching process parameter set to be optimized based on the multidimensional learning space, determine the optimization adjustment control strategy, and execute the optimization adjustment control strategy to adaptively optimize the GaAs etching process.

[0038] In one possible implementation, step A500 further includes step A510, parsing the multidimensional learning space to determine a learning state space and a learning action space; in one possible implementation, step A510 further includes step A511, performing a state analysis on the GaAs etching process based on the parameter optimization target to determine a plurality of etching process key parameters; executing step A512, matching the plurality of etching process key parameters to a plurality of state space dimensions, wherein the plurality of state space dimensions correspond to the plurality of etching process key parameters; executing step A513, traversing the plurality of etching process key parameters and performing an impact analysis on the parameter optimization target, screening the plurality of state space dimensions according to the impact analysis results, and constructing a discrete state space and a continuous state space; executing step A514, defining the learning state space based on the discrete state space and the continuous state space.

[0039] First, the multidimensional learning space is analyzed to determine the learning state space and learning action space in the etching process, which provides a basis for the reinforcement learning model. The learning state space refers to all parameter combinations in the etching process. For the GaAs etching process, the state space contains different process parameters, such as gas flow, RF power, gas pressure, temperature, etching time, etc. The state formed by each combination of the parameters is used to reflect the current process conditions. The state analysis of the GaAs etching process based on the parameter optimization target refers to evaluating the combination of different process parameters through experiments or simulations, and analyzing the influence of each process parameter on the etching performance, that is, by analyzing the performance indicators such as etching rate, surface roughness, sidewall angle, selectivity, etc., identifying which parameters have the greatest impact on the results, and determining the key parameters in the etching process according to the state analysis results, determining multiple key parameters of the etching process, and then matching multiple key parameters of the etching process to multiple state space dimensions, which means first combining multiple key parameters of the etching process to form a multidimensional state space. Each point in the state space is a combination of process parameters and corresponds to a specific etching performance. According to the adjustability of the parameters and the actual process requirements, the state space is divided into discrete and continuous dimensions. For example, gas flow and RF power may be continuously changing parameters, and process modes (such as different etching process stages) may be discrete.

[0040] Furthermore, by traversing all key process parameter combinations and analyzing the influence of each parameter on the optimization target, it means traversing all possible process parameter combinations, recording the etching performance under each combination, and then evaluating the influence of each parameter combination on the performance target (such as etching rate, surface roughness) by using simulation tools or experimental data. According to the analysis and traversal results, it is evaluated which parameters have a significant impact on the optimization target, so as to obtain the impact analysis results, and according to the impact analysis results, the dimensions in the state space are screened, which means that according to the results of the impact analysis, the dimensions that have a greater impact on the optimization target are screened, and for discrete process conditions or modes set by different process stages and equipment, a discrete state space is constructed, and for process parameters with continuous values, such as gas flow, RF power, gas pressure, temperature, etc., a continuous state space is constructed, and finally the discrete state space and the continuous state space are combined into a unified learning state space, and the learning state space covers all key process parameters and can comprehensively reflect the etching process performance under each parameter combination.

[0041] In one possible implementation, step A510 further includes step A515, performing action analysis on the GaAs etching process based on the parameter optimization target to determine multiple adjustable process parameters, wherein the multiple adjustable process parameters include parameter adjustment amounts; executing step A516 to match the multiple adjustable process parameters to multiple action space dimensions, wherein the multiple action space dimensions correspond to the multiple adjustable process parameters; executing step A517 to traverse the multiple adjustable process parameters and perform impact analysis on the parameter optimization target to determine the direction of action impact; executing step A518 to screen the multiple action space dimensions according to the direction of action impact combined with the parameter adjustment amount to construct a discrete action space and a continuous action space; executing step A519 to define the learning action space based on the discrete action space and the continuous action space.

[0042] The learning action space represents a set of operations that can be taken by the reinforcement learning algorithm to adjust process parameters, which may include adjusting the increase or decrease of gas flow, increasing or decreasing RF power, adjusting gas pressure and temperature, etc., and each action corresponds to adjusting one or more parameters in the state space, thereby affecting the performance of the etching process. First, the adjustable parameters that can affect the performance of the GaAs etching process are analyzed, and their adjustment range and direction are determined. Through experiments or simulations, the process parameters that can be directly adjusted are analyzed, which may include gas flow, RF power, gas pressure, etching time, and temperature. At the same time, an adjustment amount is set for each adjustable process parameter, that is, each action The amplitude of the parameter can be adjusted, and multiple adjustable process parameters can be further matched to multiple action space dimensions, which means that each dimension of the action space corresponds to an adjustable process parameter, and each process parameter is mapped to a dimension of the action space, and the range of change of the dimension is determined by the parameter adjustment amount, thereby determining multiple action space dimensions, and then traversing different combinations of each adjustable process parameter to analyze its impact on the etching process performance (such as etching rate, surface roughness, etc.), which means that in order to optimize the etching process, the action needs to clearly increase, decrease or maintain a certain parameter value, and the optimization target is optimized by analyzing the change of each process parameter. The influence of the action is determined, and its positive or negative influence direction is determined, so that the adjustment direction of each adjustable process parameter is determined. Multiple action space dimensions are screened according to the action influence direction combined with the parameter adjustment amount, which means that according to the adjustment direction and adjustment amount of the parameter, the dimensions in the action space are screened, and it is divided into discrete and continuous dimensions, and a discrete action space is constructed based on process parameters with fewer discrete values. The discrete action space helps to simplify the optimization process, especially for some parameters with predefined values. Based on the process parameters that can be continuously adjusted, a continuous action space is constructed. The continuous action space can provide more refined adjustment capabilities and is suitable for parameter optimization that requires precise control. Finally, the discrete action space is combined with the continuous action space to construct a complete learning action space, and each action space dimension corresponds to an adjustable process parameter, and can be adjusted in a discrete or continuous manner. Boundary conditions can also be set for each action space dimension, that is, the maximum and minimum values ​​of each parameter. In the reinforcement learning process, the learning action space will be used as input, and the algorithm will select the optimal adjustment action (such as increasing gas flow, reducing RF power, etc.) according to the current state to optimize the etching process performance, thereby completing the definition of the learning action space to achieve continuous optimization of the etching performance.

[0043] Execute step A520, perform simulation verification according to the parameter optimization target, extract performance feedback parameters to construct the etching process performance target, and set the reward function according to the etching process performance target; execute step A530, initialize the process environment information, and obtain the etching process simulation environment information; execute step A540, synchronize the etching process simulation environment information to the learning action space according to the time step for selection, and determine the first learning action; execute step A550, synchronize the etching process simulation environment information to the learning state space according to the first learning action for calculation, update the learning state of the etching process parameter set to be optimized, generate a first learning state, iterate in combination with the reward function, stop when the learning state is stable at a preset change threshold, and obtain parameter adjustment information, the parameter adjustment information includes parameter adjustment direction information and parameter adjustment size information; execute step A560, perform adjustment verification according to the parameter adjustment direction information, and when the verification passes, construct the optimization adjustment control strategy based on the parameter adjustment size information.

[0044] First, the learning space of the multi-dimensional etching process is analyzed, and the learning state space and learning action space are clarified as the basis for reinforcement learning. Then, simulation tools can be used to simulate dry etching and wet etching, and simulate the performance of etching rate, surface roughness, sidewall angle, etc. under different parameter combinations. For each simulation result, relevant feedback parameters such as etching rate, uniformity, selectivity, roughness, etc. are extracted as the basis for performance evaluation. According to the simulation results, the etching process performance target is defined, and the reward function is further set according to the etching process performance target. The reward function is used to evaluate the effect of each parameter adjustment. The reward function can be set according to the difference between the performance target and the current state, including the dry etching reward function and the wet etching reward function. The dry etching reward function means that when the etching rate and roughness reach the target, a high reward is given; when excessive adjustment causes surface damage, a negative reward is given. The wet etching reward function means that when the etching uniformity and depth accuracy meet the requirements, a positive reward is given; when excessive etching or unevenness occurs, a negative reward is given.

[0045] Furthermore, according to the setting of the experimental equipment or simulation system, the process environment parameters of dry or wet etching are initialized, so as to obtain the etching process simulation environment information, which is used to characterize the process settings and current state parameters in the initial state. At the same time, according to the time step of reinforcement learning, the current environment information is synchronized to the action space at each time point for action selection, which can be achieved by using Q-learning or Deep The reinforcement learning algorithm of Q-Network selects the optimal action according to the current learning state. The action selection can be based on the greedy strategy or the policy gradient method to determine how to adjust the process parameters. The selected learning action is used as the first learning action, and then the etching process simulation environment information is synchronized to the learning state space for calculation according to the first learning action. It means adjusting the process parameters according to the first learning action, and synchronizing the adjusted results to the learning state space. By updating the state space, combining simulation feedback or experimental data, the new process performance is evaluated, and the state information in the process parameter set to be optimized is updated. Then, according to the adjusted results, a new learning state is generated, that is, the first learning state. The first learning state includes the latest process parameters and corresponding performance results, and then combined with the reward function for iteration. When the learning state stabilizes at the preset change threshold, it stops. It means executing a new learning action at each time step, updating the process parameters, and generating a new learning state. According to the feedback of the reward function, the action selection strategy is adjusted, and after each action is executed, the process performance is evaluated according to the feedback of the reward function. If a positive reward is obtained, it indicates that the current process parameter adjustment is effective; otherwise, the direction or amplitude of the adjustment, when the amplitude of the change in the learning state is less than the preset threshold (such as parameter change <1%), it means that the learning process has converged and the iteration can be stopped. At the end of the learning process, the final parameter adjustment information is obtained. The parameter adjustment information includes parameter adjustment direction information, that is, increase or decrease, and parameter adjustment size information, that is, adjustment amplitude. Finally, the parameter adjustment direction and size are applied to the actual dry or wet etching experiment to verify whether the adjusted process meets the optimization goal. When the verification passes, the final control strategy is constructed based on the parameter adjustment size information to ensure that the etching process can automatically apply these adjustments in future operations, and generate an optimized control strategy for the etching process. The optimized control strategy for the etching process is used to adjust each parameter to achieve the best etching performance, ensuring the accuracy and effectiveness of process optimization.

[0046] Finally, the optimization and adjustment control strategy output by reinforcement learning is applied to adaptively optimize the actual GaAs etching process, adjust the process parameters in real time, and improve the etching effect. This means that the actual process parameters are gradually adjusted according to the optimization control strategy of the etching process. During the etching process, the system monitors the performance indicators such as etching rate, surface roughness, and sidewall morphology in real time, and adjusts the parameters according to the feedback information. If the process conditions change (such as pressure fluctuations or changes in material properties), the system can adaptively adjust the process parameters according to the current performance information. During the optimization process, real-time etching data is obtained through sensors or monitoring systems, fed back to the control system, and the process parameters are automatically adjusted according to the feedback information to maintain the best etching effect. At the same time, the etching effect after executing the optimization and adjustment control strategy can be evaluated. By comparing the effect before optimization, the effectiveness of the optimization scheme can be analyzed. By comparing the etching effects before and after optimization, it is confirmed whether the optimization scheme has improved the process performance. If both the etching rate and surface quality are improved, and the process stability is enhanced, it proves that the optimization scheme is effective and practical. After applying the optimization control strategy, detailed tuning can be performed to further improve process performance and adapt to a wider range of process conditions. That is, feedback data from the actual etching process is collected, and existing process fluctuations or unstable factors are analyzed. For certain special cases (such as different mask layer materials or changes in process conditions), the parameter adjustment amount in the action space can be fine-tuned to ensure the adaptability of the system. In the tuning process, the applicable range of process parameters is expanded. By tuning the optimization strategy, the accuracy and stability of process control are ensured, providing strong support for applications in actual production.

[0047] The embodiments of the present application solve the technical problem that etching process parameters are difficult to accurately control, resulting in low optimization adaptability of the GaAs etching process, and achieve the technical effect of improving the optimization adaptability of the GaAs etching process by combining reinforcement learning with learning space based on process performance evaluation.

[0048] In the above, refer to Figure 1 The adaptive optimization method for GaAs etching process according to the embodiment of the present application is described in detail. Figure 2 An adaptive optimization device for a GaAs etching process according to an embodiment of the present application is described.

[0049] According to the adaptive optimization device for GaAs etching process of the embodiment of the present application, it is used to solve the technical problem that the etching process parameters are difficult to accurately control, resulting in low optimization fitness of GaAs etching process, and realize the technical effect of improving the optimization fitness of GaAs etching process by combining reinforcement learning with learning space according to process performance evaluation. The adaptive optimization device for GaAs etching process includes: a data experiment module 10, a performance evaluation module 20, a parameter definition module 30, a parameter analysis module 40, and a process optimization module 50.

[0050] A data experiment module 10, wherein the data experiment module 10 is used to introduce process environment information, perform experiments using a GaAs etching process according to the process environment information, and obtain process experiment results; A performance evaluation module 20, wherein the performance evaluation module 20 is used to construct a process performance evaluation model, synchronize the process experiment results to the process performance evaluation model, and obtain a process performance evaluation result; A parameter definition module 30, wherein the parameter definition module 30 is used to define a set of etching process parameters to be optimized according to the process performance evaluation result; A parameter analysis module 40, the parameter analysis module 40 is used to analyze the etching process parameter set to be optimized based on the process environment information and set a parameter optimization target; The process optimization module 50 is used to establish a multidimensional learning space according to the parameter optimization target, perform reinforcement learning on the etching process parameter set to be optimized based on the multidimensional learning space, determine the optimization adjustment control strategy, and execute the optimization adjustment control strategy to adaptively optimize the GaAs etching process.

[0051] The specific configuration of the performance evaluation module 20 will be described in detail below. As described above, to construct the process performance evaluation model, the performance evaluation module 20 may further include: retrieving the historical etching data record log of the GaAs etching process, integrating parameters according to the historical etching data record log, and obtaining the etching process parameter space; performing random selection based on the etching process parameter space to determine multiple initial sampling points, wherein the multiple initial sampling points contain the etching process performance information of each initial sampling point; traversing the multiple initial sampling points to match with the process experiment results, and constructing a process experiment matching list; extracting multiple process experiment results based on the process experiment matching list, wherein the multiple process experiment results have a corresponding relationship with the multiple initial sampling points; performing experimental prediction on the first initial sampling point using the acquisition function to obtain the first experimental prediction result, and iterating until the traversal of the multiple initial sampling points is completed and the multiple experimental prediction results are determined; adding the multiple experimental prediction results to the process experiment matching list for combination to obtain a process experiment comparison list; constructing the process performance evaluation model based on the process experiment comparison list combined with the etching process performance information of each initial sampling point.

[0052] The specific configuration of the performance evaluation module 20 will be described in detail below. As described above, the process experiment results are synchronized to the process performance evaluation model to obtain the process performance evaluation results. The performance evaluation module 20 may further include: using the process experiment results as index data, traversing the process experiment comparison list to extract the target experiment prediction results, and there is a corresponding relationship between the target experiment prediction results and the process experiment results; performing calculations based on the target experiment prediction results to obtain multiple performance indicator prediction values; comparing the process experiment results with the target experiment prediction results in combination with the process cycle according to the multiple performance indicator prediction values ​​to generate process error parameters; performing error analysis based on the process error parameters, performing incremental learning on the process performance evaluation model according to the error analysis results, and performing evaluation in combination with the learning results to obtain the process performance evaluation results.

[0053] The specific configuration of the performance evaluation module 20 will be described in detail below. As described above, error analysis is performed based on the process error parameters, and incremental learning is performed on the process performance evaluation model according to the error analysis results. The performance evaluation module 20 may further include: determining whether the process error parameter is greater than a preset error threshold, and if so, generating a learning instruction; performing process analysis on the process experiment results through the learning instructions to generate process experiment analysis results; performing error analysis on the process error parameters through the learning instructions to generate process error results; using the process error results combined with the process experiment analysis results as the input of incremental learning, updating the data distribution of the process performance evaluation model, and generating the learning results.

[0054] Next, the specific configuration of the process optimization module 50 will be described in detail. As described above, based on the multidimensional learning space, the etching process parameter set to be optimized is reinforced learned to determine the optimization adjustment control strategy. The process optimization module 50 may further include: parsing the multidimensional learning space to determine the learning state space and the learning action space; performing simulation verification according to the parameter optimization target, extracting performance feedback parameters to construct the etching process performance target, and setting the reward function according to the etching process performance target; initializing the process environment information to obtain the etching process simulation environment information; synchronizing the etching process simulation environment information to the learning action space according to the time step for selection, and determining the first learning action; synchronizing the etching process simulation environment information to the learning state space according to the first learning action for calculation, updating the learning state of the etching process parameter set to be optimized, generating a first learning state, iterating in combination with the reward function, stopping when the learning state is stable at a preset change threshold, and obtaining parameter adjustment information, the parameter adjustment information including parameter adjustment direction information and parameter adjustment size information; performing adjustment verification according to the parameter adjustment direction information, and when the verification passes, constructing the optimization adjustment control strategy based on the parameter adjustment size information.

[0055] The specific configuration of the process optimization module 50 will be described in detail below. As described above, before the set of items to be used is classified according to the storage order, the process optimization module 50 may further include: performing state analysis on the GaAs etching process based on the parameter optimization target to determine multiple key parameters of the etching process; matching the multiple key parameters of the etching process with multiple state space dimensions, and the multiple state space dimensions correspond to the multiple key parameters of the etching process; traversing the multiple key parameters of the etching process and the parameter optimization target to perform impact analysis, screening the multiple state space dimensions according to the impact analysis results, and constructing a discrete state space and a continuous state space; defining the learning state space based on the discrete state space and the continuous state space.

[0056] The specific configuration of the process optimization module 50 will be described in detail below. As described above, before the set of items to be used is classified according to the storage order, the process optimization module 50 may further include: performing action analysis on the GaAs etching process based on the parameter optimization target, determining multiple adjustable process parameters, and the multiple adjustable process parameters include parameter adjustment amounts; matching the multiple adjustable process parameters with multiple action space dimensions, and the multiple action space dimensions have a corresponding relationship with the multiple adjustable process parameters; traversing the multiple adjustable process parameters and the parameter optimization target to perform impact analysis and determine the direction of action impact; screening the multiple action space dimensions according to the action impact direction combined with the parameter adjustment amount to construct a discrete action space and a continuous action space; defining the learning action space based on the discrete action space and the continuous action space.

[0057] The adaptive optimization device for GaAs etching process provided in the embodiment of the present application can execute the adaptive optimization method for GaAs etching process provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects of the execution method.

[0058] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present application.

[0059] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. An adaptive optimization method for GaAs etching process, characterized in that: The method comprises: Introducing process environment information, performing experiments using a GaAs etching process according to the process environment information, and obtaining process experiment results; Constructing a process performance evaluation model, synchronizing the process experiment results to the process performance evaluation model, and obtaining a process performance evaluation result; Defining an etching process parameter set to be optimized according to the process performance evaluation result; Analyzing the etching process parameter set to be optimized based on the process environment information, and setting a parameter optimization target; A multidimensional learning space is established according to the parameter optimization target, reinforcement learning is performed on the etching process parameter set to be optimized based on the multidimensional learning space, an optimization adjustment control strategy is determined, and the optimization adjustment control strategy is executed to adaptively optimize the GaAs etching process.

2. The adaptive optimization method for GaAs etching process according to claim 1, characterized in that: Constructing the process performance evaluation model, the method includes: Retrieving a historical etching data record log of a GaAs etching process, performing parameter integration according to the historical etching data record log, and obtaining an etching process parameter space; Based on the etching process parameter space, a random selection is performed to determine a plurality of initial sampling points, wherein the plurality of initial sampling points include etching process performance information of each initial sampling point; Traversing the multiple initial sampling points and matching them with the process experiment results to construct a process experiment matching list; Extracting a plurality of process experiment results based on the process experiment matching list, wherein the plurality of process experiment results correspond to the plurality of initial sampling points; Performing experimental prediction on the first initial sampling point using the acquisition function to obtain a first experimental prediction result, and iterating therefrom until the traversal of the plurality of initial sampling points is completed and the plurality of experimental prediction results are determined; Adding the plurality of experimental prediction results to the process experiment matching list for combination to obtain a process experiment comparison list; The process performance evaluation model is constructed based on the process experiment comparison list combined with the etching process performance information of each initial sampling point.

3. The adaptive optimization method for GaAs etching process according to claim 2, characterized in that: The process experiment result is synchronized with the process performance evaluation model to obtain a process performance evaluation result, the method comprising: Using the process experiment results as index data, traversing the process experiment comparison list to extract target experiment prediction results, wherein the target experiment prediction results correspond to the process experiment results; Calculate according to the target experiment prediction results to obtain multiple performance indicator prediction values; According to the predicted values ​​of the multiple performance indicators, the process experiment results are compared with the predicted results of the target experiment in combination with the process cycle to generate a process error parameter; An error analysis is performed based on the process error parameters, incremental learning is performed on the process performance evaluation model according to the error analysis results, and an evaluation is performed in combination with the learning results to obtain the process performance evaluation result.

4. The adaptive optimization method for GaAs etching process according to claim 3, characterized in that: Performing error analysis based on the process error parameters and performing incremental learning on the process performance evaluation model according to the error analysis results, the method comprising: Determine whether the process error parameter is greater than a preset error threshold, and if so, generate a learning instruction; Performing process analysis on the process experiment results through the learning instructions to generate process experiment analysis results; Performing error analysis on the process error parameters through the learning instructions to generate a process error result; The process error result is combined with the process experiment analysis result as an input for incremental learning, the data distribution of the process performance evaluation model is updated, and the learning result is generated.

5. The adaptive optimization method for GaAs etching process according to claim 1, characterized in that: Based on the multi-dimensional learning space, the etching process parameter set to be optimized is subjected to reinforcement learning to determine an optimization adjustment control strategy, and the method includes: Analyzing the multidimensional learning space to determine a learning state space and a learning action space; Perform simulation verification according to the parameter optimization target, extract performance feedback parameters to construct an etching process performance target, and set a reward function according to the etching process performance target; Initializing the process environment information to obtain etching process simulation environment information; Synchronizing the etching process simulation environment information to the learning action space according to the time step for selection, and determining a first learning action; According to the first learning action, the etching process simulation environment information is synchronized to the learning state space for calculation, the learning state of the etching process parameter set to be optimized is updated, a first learning state is generated, and iteration is performed in combination with the reward function. When the learning state is stable at a preset change threshold, the learning state is stopped, and parameter adjustment information is obtained, wherein the parameter adjustment information includes parameter adjustment direction information and parameter adjustment size information; Adjustment verification is performed according to the parameter adjustment direction information, and when the verification passes, the optimization adjustment control strategy is constructed based on the parameter adjustment size information.

6. The adaptive optimization method for GaAs etching process according to claim 5, characterized in that: The learning state space method comprises: Performing state analysis on the GaAs etching process based on the parameter optimization target to determine multiple key parameters of the etching process; Matching the multiple key etching process parameters to multiple state space dimensions, where the multiple state space dimensions correspond to the multiple key etching process parameters; Traversing the multiple key etching process parameters and the parameter optimization targets to perform impact analysis, screening the multiple state space dimensions according to the impact analysis results, and constructing a discrete state space and a continuous state space; The learning state space is defined based on the discrete state space and the continuous state space.

7. The adaptive optimization method for GaAs etching process according to claim 5, characterized in that: The learning action space method comprises: Performing an action analysis on the GaAs etching process based on the parameter optimization target to determine a plurality of adjustable process parameters, wherein the plurality of adjustable process parameters include parameter adjustment amounts; Matching the plurality of adjustable process parameters to a plurality of action space dimensions, wherein the plurality of action space dimensions correspond to the plurality of adjustable process parameters; Traversing the plurality of adjustable process parameters and the parameter optimization target to perform an impact analysis and determine an action impact direction; The plurality of action space dimensions are screened according to the action influence direction and the parameter adjustment amount to construct a discrete action space and a continuous action space; The learning action space is defined based on the discrete action space and the continuous action space.

8. An adaptive optimization device for GaAs etching process, characterized in that: The device is used to implement the adaptive optimization method for GaAs etching process according to any one of claims 1 to 7, and the device comprises: A data experiment module, wherein the data experiment module is used to introduce process environment information, perform experiments using a GaAs etching process according to the process environment information, and obtain process experiment results; A performance evaluation module, wherein the performance evaluation module is used to construct a process performance evaluation model, synchronize the process experiment results to the process performance evaluation model, and obtain a process performance evaluation result; A parameter definition module, wherein the parameter definition module is used to define a set of etching process parameters to be optimized according to the process performance evaluation result; A parameter analysis module, the parameter analysis module is used to analyze the etching process parameter set to be optimized based on the process environment information and set a parameter optimization target; A process optimization module, wherein the process optimization module is used to establish a multidimensional learning space according to the parameter optimization target, perform reinforcement learning on the etching process parameter set to be optimized based on the multidimensional learning space, determine an optimization adjustment control strategy, and execute the optimization adjustment control strategy to adaptively optimize the GaAs etching process.

Citation Information

Patent Citations

  • Manufacturing process technological parameter optimization method and system based on reinforcement learning

    CN117420800A

  • GaAs etching process simulation process and system

    CN118656991A

Cited By

  • Method and system for controlling semiconductor manufacturing machine

    CN119472448A

  • A control method and system for a semiconductor manufacturing machine

    CN119472448B

  • Body-in-white structure process parameter optimization model training and application method and device

    CN119538421A

  • A body-in-white structure process parameter optimization model training, application method and device

    CN119538421B

  • Acid mist absorption tower intelligent control system based on big data

    CN119847027A