Method and system for optimizing high frequency switching life of ald valves for semiconductors

Abstract

The application relates to the technical field of semiconductor manufacturing equipment control, and provides an ALD valve high-frequency switch service life optimization method and system for a semiconductor. The method comprises the following steps: collecting historical switch operation data of an atomic layer deposition valve, and constructing a service life prediction training data set; training a time sequence service life prediction model based on the service life prediction training data set, so that the time sequence service life prediction model has a residual service life output capacity; acquiring real-time operation data of the atomic layer deposition valve, and inputting the time sequence service life prediction model to obtain a real-time residual service life result; and according to the real-time residual service life result, calculating adjustment values of switch speed and air pressure parameters through a high-frequency switch action parameter self-adaptive optimization algorithm to complete service life optimization closed-loop control. The residual service life is predicted through real-time operation data, and the valve switch operation parameters are adjusted correspondingly, so that the valve loss is reduced under the premise of meeting process requirements, and the cyclic service life of the atomic layer deposition valve under the high-frequency switch working condition is effectively prolonged.

Description

Technical Field

[0001] This application relates to the field of semiconductor manufacturing equipment control technology, and in particular to a method and system for optimizing the high-frequency switching life of an ALD valve for semiconductors. Background Technology

[0002] Atomic layer deposition is a core thin film deposition process in semiconductor manufacturing. Atomic layer deposition valves are key actuators for controlling the flow of process gases, and the process requires them to support high-frequency switching operation.

[0003] The operating parameters of existing atomic layer deposition valves, such as switching speed and supply gas pressure, are fixed at the factory and cannot be dynamically adjusted according to the actual wear and tear of the valve components. Under high-frequency switching conditions, the valve sealing structure and actuator wear rapidly, resulting in a short overall cycle life. Furthermore, current valve life management primarily relies on periodic maintenance or reactive replacement, failing to incorporate proactive life optimization based on real-time operating conditions. This can easily lead to premature valve failure, increasing maintenance costs and downtime risks on semiconductor production lines. Summary of the Invention

[0004] This application provides a method and system for optimizing the high-frequency switching life of ALD valves for semiconductors, aiming to solve the problem that the operating parameters of existing atomic layer deposition valves, such as switching speed and gas supply pressure, are fixed at the factory and cannot be dynamically adjusted according to the actual wear and tear of the valve components.

[0005] In a first aspect, embodiments of this application provide a method for optimizing the high-frequency switching lifetime of an ALD valve for semiconductors, the method comprising: Historical on / off operation data of atomic layer deposition valves were collected to construct a lifetime prediction training dataset; a time-series lifetime prediction model was trained based on the lifetime prediction training dataset, enabling the time-series lifetime prediction model to output remaining lifetime. Real-time operating data of the atomic layer deposition valve is acquired and input into a time-series lifetime prediction model to obtain real-time remaining lifetime results. Based on the real-time remaining lifetime results, the adjustment values ​​of switching speed and gas pressure parameters are calculated using a high-frequency switching action parameter adaptive optimization algorithm. The adjustment value is sent to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; the operating loss data after the switching action parameters are adjusted is collected and fed back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

[0006] In some embodiments, the process of collecting historical switching operation data of atomic layer deposition valves to construct a life prediction training dataset includes: collecting the number of switching operations, action response time, sealing surface wear, air supply pressure fluctuation value, and action impact peak data of multiple atomic layer deposition valves throughout their entire life cycle; performing outlier removal and time series alignment processing on all collected data; labeling each set of time series data with the remaining cycle life label of the corresponding valve component; and dividing the labeled time series data into a training subset and a validation subset to form a life prediction training dataset.

[0007] In some embodiments, training a time-series lifetime prediction model based on a lifetime prediction training dataset to enable the time-series lifetime prediction model to output remaining lifetime includes: constructing an initial model of a prediction network containing a multi-layer time-series feature extraction structure; inputting a training subset from the lifetime prediction training dataset into the initial model of the prediction network to output remaining lifetime prediction values; iteratively updating the internal parameters of the initial model of the prediction network based on the deviation between the remaining lifetime prediction values ​​and the corresponding remaining cycle lifetime labels; verifying the accuracy of the iterated model using a validation subset from the lifetime prediction training dataset; and stopping the iteration when the model output accuracy reaches a preset accuracy threshold, thereby obtaining a time-series lifetime prediction model with remaining lifetime output capability.

[0008] In some embodiments, acquiring real-time operating data of the atomic layer deposition valve and inputting it into a time-series lifetime prediction model to obtain real-time remaining lifetime results includes: collecting real-time switching operation data of the currently operating atomic layer deposition valve according to a preset sampling period; performing outlier removal and time-series alignment preprocessing operations on the real-time switching operation data; inputting the preprocessed real-time switching operation data into the trained time-series lifetime prediction model; extracting time-series decay features from the real-time data through the time-series lifetime prediction model, and outputting the real-time remaining lifetime result of the current atomic layer deposition valve.

[0009] In some embodiments, the step of calculating the adjustment values ​​of switching speed and gas pressure parameters based on the real-time remaining lifetime results using an adaptive optimization algorithm for high-frequency switching action parameters includes: reading the range of switching speed and gas pressure parameters allowed by the atomic layer deposition process; inputting the real-time remaining lifetime results into the adaptive optimization algorithm for high-frequency switching action parameters to match the loss suppression target for the corresponding remaining lifetime range; and within the range allowed by the process, with the loss suppression target as a constraint, solving for the adjustment values ​​of switching speed and gas pressure parameters.

[0010] In some embodiments, the step of sending the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters includes: performing process compliance verification on the calculated switching speed adjustment value and air pressure parameter adjustment value; generating parameter control instructions for the corresponding actuator after the verification is passed; sending the parameter control instructions to the actuator of the atomic layer deposition valve; and having the actuator adjust the valve's switching speed and air supply pressure according to the parameter control instructions to complete the dynamic update of the switching action parameters.

[0011] In some embodiments, the step of collecting operational loss data after adjusting the switching action parameters and feeding it back to the time-series lifetime prediction model to update the remaining lifetime result and complete lifetime optimization closed-loop control includes: after the switching action parameters are adjusted, collecting operational loss data of the atomic layer deposition valve within a preset operating time; inputting the operational loss data into the time-series lifetime prediction model to refresh the current remaining lifetime result; and using the refreshed remaining lifetime result as the input basis for the next round of parameter optimization to form a continuously iterative lifetime optimization closed-loop control.

[0012] In some embodiments, the method further includes: real-time acquisition of thin film deposition yield data of the deposition chamber corresponding to the atomic layer deposition valve; inputting the thin film deposition yield data into an adaptive optimization algorithm for high-frequency switching action parameters, and adding yield compliance constraints; when parameter adjustment values ​​cause the yield to fall below a preset yield threshold, resolving the adjustment values ​​of switching speed and air pressure parameters to ensure that the lifetime optimization process meets the yield requirements of the semiconductor deposition process.

[0013] In some embodiments, the method further includes: obtaining the real-time remaining lifetime results of all atomic layer deposition valves in the same semiconductor production line; calculating the dispersion of the remaining lifetime of all valves; when the dispersion exceeds a preset equilibrium threshold, adjusting the process load allocation corresponding to each atomic layer deposition valve; so that the remaining lifetime of all valves tends to be consistent, thereby achieving lifetime equilibrium optimization of the production line valve cluster.

[0014] Secondly, this application provides a high-frequency switching life optimization system for ALD valves in semiconductors, the system comprising: The data acquisition unit is used to collect historical switching operation data of the atomic layer deposition valve and construct a lifetime prediction training dataset; based on the lifetime prediction training dataset, the time-series lifetime prediction model is trained, enabling the time-series lifetime prediction model to output the remaining lifetime. The data acquisition unit is used to acquire real-time operating data of the atomic layer deposition valve, input the data into the time-series lifetime prediction model to obtain the real-time remaining lifetime result, and calculate the adjustment values ​​of the switching speed and gas pressure parameters based on the real-time remaining lifetime result through the high-frequency switching action parameter adaptive optimization algorithm. The control completion unit is used to send the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; collect the operating loss data after the switching action parameters are adjusted, and feed it back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

[0015] This application uses real-time operating data to predict the remaining lifespan and adjusts the valve switching operating parameters accordingly. Under the premise of meeting the process requirements, it reduces valve wear and effectively extends the cycle life of the atomic layer deposition valve under high-frequency switching conditions.

[0016] A closed-loop control mechanism for life decay feedback is constructed, which can dynamically adapt operating parameters according to the actual wear state of valve components, thereby achieving proactive and continuous optimization of life without the need for manual intervention.

[0017] This reduces the frequency of valve replacements and the probability of unplanned downtime, thereby reducing maintenance costs for semiconductor production lines and improving the operational stability of atomic layer deposition processes.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart illustrating the steps of a method for optimizing the high-frequency switching life of an ALD valve for semiconductors, provided in an embodiment of this application. Figure 2 This is a schematic diagram illustrating the principle of a method for optimizing the high-frequency switching life of an ALD valve for semiconductors, provided in an embodiment of this application. Figure 3 This is a schematic block diagram of a high-frequency switching life optimization system for an ALD valve in a semiconductor according to an embodiment of this application; Figure 4 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0024] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0025] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] Atomic layer deposition is a core thin film deposition process in semiconductor manufacturing. Atomic layer deposition valves are key actuators for controlling the flow of process gases, and the process requires them to support high-frequency switching operation.

[0028] The operating parameters of existing atomic layer deposition valves, such as switching speed and supply gas pressure, are fixed at the factory and cannot be dynamically adjusted according to the actual wear and tear of the valve components. Under high-frequency switching conditions, the valve sealing structure and actuator wear rapidly, resulting in a short overall cycle life. Furthermore, current valve life management primarily relies on periodic maintenance or reactive replacement, failing to incorporate proactive life optimization based on real-time operating conditions. This can easily lead to premature valve failure, increasing maintenance costs and downtime risks on semiconductor production lines.

[0029] Please refer to Figure 1, the embodiments of the present application provide a high-frequency switching life optimization method for ALD valves for semiconductors, which is applied to a computer device. The computer device may be deployed on a single server or a server cluster, and may also be deployed on a handheld terminal, a notebook computer, a wearable device, a robot, or the like. Meanwhile, it should be noted that each piece of information involved in the method provided by the present application is extracted under the authorization of the relevant user and in compliance with relevant regulations, and will not infringe on user privacy.

[0030] The provided high-frequency switching life optimization method for ALD valves for semiconductors includes steps S101 to S103, which are described in detail as follows: Step S101. Collect historical switching operation data of the atomic layer deposition valve, and construct a life prediction training dataset; train a time-series life prediction model based on the life prediction training dataset, so that the time-series life prediction model has the capability of outputting remaining life.

[0031] Specifically, the execution subject is the computer device deploying the present method, and the data sources are the historical operation and maintenance database of the semiconductor production line, and the storage records of the displacement sensing unit, the pressure sensing unit, and the impact sensing unit matched with the atomic layer deposition valve. The collection scope covers the full service cycles of multiple atomic layer deposition valves of the same specification from the first put into use to reaching the failure judgment standard, wherein the failure judgment standard is preset according to the precision requirements of the semiconductor process, including three indicators: the valve opening action response time exceeding 20% of the rated value, the valve closing action response time exceeding 20% of the rated value, and the leakage amount of the sealing structure exceeding the upper limit allowed by the process, and the valve is judged to reach the failure state when any one of the indicators is satisfied.

[0032] A life prediction training dataset is constructed based on the collected full-cycle historical data. As the basic input for model training, the dataset includes standardized time-series operation data and corresponding life labels. After the dataset is constructed, the dataset is input into a time-series prediction network with a preset structure, and the internal weight parameters of the network are iteratively adjusted through a supervised training method, so that the network can accurately output the remaining valid switching times of the corresponding valve according to the input continuous time-series operation data, and finally obtain a time-series life prediction model with accuracy meeting the use requirements and having the capability of outputting remaining life.

[0033] Step S102. Obtain real-time operation data of the atomic layer deposition valve, and input the data into the time-series life prediction model to obtain a real-time remaining life result; according to the real-time remaining life result, the adjustment values of the switching speed and air pressure parameters are calculated through a high-frequency switching action parameter adaptive optimization algorithm.

[0034] Specifically, the computer equipment reads real-time operating data of the atomic layer deposition valve's current service status from its associated sensing unit according to a pre-set sampling frequency. After performing preprocessing operations on the real-time data that are completely identical to those on the training data, the data is input into the trained time-series lifetime prediction model. The model extracts the time-series decay characteristics from the real-time data and combines them with the wear decay patterns learned during the training phase to output the real-time remaining lifetime result of the valve. The real-time remaining lifetime result is characterized by the number of effective switching cycles that the valve can continue to perform.

[0035] After obtaining the real-time remaining lifetime result, the result is input into the adaptive optimization algorithm for high-frequency switch action parameters. The algorithm uses the parameter range allowed by the atomic layer deposition process as the boundary constraint, and aims to reduce valve operating losses and extend cycle life as the optimization objective. It matches the corresponding optimization intensity with the current remaining lifetime level, and solves for the switch speed adjustment value and gas pressure parameter adjustment value within the boundary constraint. This ensures that the adjusted parameters not only meet the process requirements, but also effectively reduce the structural wear caused by the high-frequency switch.

[0036] Step S103. Send the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; collect the operating loss data after the switching action parameters are adjusted, and feed it back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

[0037] Specifically, the computer equipment converts the calculated parameter adjustment values ​​into control commands that the atomic layer deposition valve actuator can recognize, and sends them to the pneumatic actuator of the valve via the industrial control bus. After receiving the control commands, the actuator adjusts the air supply pressure and actuation logic of the pneumatic circuit accordingly to achieve dynamic adjustment of the switching speed.

[0038] After parameter adjustments are completed and the system runs for the preset duration, the operational wear data of the atomic layer deposition valve during this period is collected. This wear data is then fed back to the time-series life prediction model, updating the current remaining life result of the valve. The updated remaining life result will serve as the input for the next round of parameter optimization. The process of life prediction, parameter solving, parameter adjustment, and data feedback is repeated, forming a continuously iterative closed-loop control mechanism. This mechanism continuously optimizes operating parameters throughout the valve's entire service life, extending its overall cycle life.

[0039] In some embodiments, the process of collecting historical switching operation data of atomic layer deposition valves to construct a life prediction training dataset includes: collecting the number of switching operations, action response time, sealing surface wear, air supply pressure fluctuation value, and action impact peak data of multiple atomic layer deposition valves throughout their entire life cycle; performing outlier removal and time series alignment processing on all collected data; labeling each set of time series data with the remaining cycle life label of the corresponding valve component; and dividing the labeled time series data into a training subset and a validation subset to form a life prediction training dataset.

[0040] This embodiment describes the specific implementation method of collecting historical on / off operation data of the atomic layer deposition valve and constructing a lifetime prediction training dataset in step S101. The specific process is as follows: The first step involved collecting data on the number of on / off cycles, response time, sealing surface wear, air supply pressure fluctuation, and peak impact of multiple atomic layer deposition (ALD) valves throughout their entire lifecycle. Data for each valve covered the complete operating cycle from initial commissioning to reaching the failure criteria. All data included a timestamp at the time of collection and a cumulative number of on / off cycles. The number of on / off cycles represents the total number of complete on / off cycles completed by the valve. The response time includes both opening and closing response times. Sealing surface wear data was obtained from periodically offline monitoring of sealing surface thickness attenuation. Air supply pressure fluctuation was the maximum deviation in air supply pressure during a single on / off cycle. Peak impact was the maximum structural impact value at the moment the valve fully opened and fully closed.

[0041] The second step involves outlier removal and time-series alignment of all collected data. Outlier removal is performed according to data type: for cumulative switch count data, outlier records with abrupt changes or inconsistent logic are removed; for three types of numerical data—action response time, gas supply pressure fluctuation, and action impact peak—the mean and standard deviation of the entire dataset are calculated, and outlier records exceeding three times the standard deviation of the mean are removed; for sealing surface wear data, outlier records with wear values ​​lower than the previous detection value are removed. Time-series alignment uses the cumulative switch count as the time base, setting one data sampling node for every 100 completed switch actions. For data with mismatched original sampling nodes, linear interpolation is used to calculate the corresponding sampling node value, ensuring that the number of nodes in each set of time-series data is completely consistent and the time base is completely aligned, forming a time-series data sequence of uniform length.

[0042] The third step is to label each set of time-series data with the remaining cycle life tag corresponding to the valve. For each time-series data sequence, first determine the cumulative total number of switching operations when the corresponding valve reaches the failure criterion. Then, using the cumulative number of switching operations corresponding to the current sampling node as a benchmark, subtract the current cumulative number of switching operations from the total number of switching operations to obtain the remaining effective number of switching operations corresponding to the current node. The remaining effective number of switching operations is used as the tag value of this set of time-series data, realizing a one-to-one binding between the time-series data sequence and the life tag.

[0043] The fourth step involves dividing the labeled time-series data into training and validation subsets to form a lifetime prediction training dataset. The labeled data is randomly divided in an 8:2 ratio, with 80% allocated to the training subset for iterative model training and the remaining 20% ​​allocated to the validation subset for accuracy verification during model training. After the division, the data distribution of the two subsets is validated to ensure consistency in lifetime range distribution and decay trend distribution, avoiding data distribution bias that could affect training performance, ultimately forming a complete lifetime prediction training dataset.

[0044] In some embodiments, training a time-series lifetime prediction model based on a lifetime prediction training dataset to enable the time-series lifetime prediction model to output remaining lifetime includes: constructing an initial model of a prediction network containing a multi-layer time-series feature extraction structure; inputting a training subset from the lifetime prediction training dataset into the initial model of the prediction network to output remaining lifetime prediction values; iteratively updating the internal parameters of the initial model of the prediction network based on the deviation between the remaining lifetime prediction values ​​and the corresponding remaining cycle lifetime labels; verifying the accuracy of the iterated model using a validation subset from the lifetime prediction training dataset; and stopping the iteration when the model output accuracy reaches a preset accuracy threshold, thereby obtaining a time-series lifetime prediction model with remaining lifetime output capability.

[0045] This embodiment describes the specific implementation method of training the time-series lifetime prediction model based on the lifetime prediction training dataset in step S101, enabling the time-series lifetime prediction model to have the ability to output remaining lifetime. The specific process is as follows: The first step is to construct an initial model of the prediction network, which includes a multi-layered temporal feature extraction structure. This initial model consists of three stacked temporal feature extraction units: the first layer extracts the instantaneous features of a single set of switching actions; the second layer extracts the short-term fluctuation features of one hundred consecutive switching actions; and the third layer extracts the long-term decay trend features of ten thousand cumulative switching actions. All internal weight parameters of the model are initially set using a random initialization method, and the output layer is set to output a single-node value corresponding to the remaining valid number of switches.

[0046] The second step involves inputting a training subset from the lifetime prediction training dataset into the initial model of the prediction network, which then outputs the remaining lifetime prediction value. The time-series data sequences in the training subset are grouped into batches and input into the initial model batch by batch. Each batch contains a fixed number of time-series data sequences, and the model outputs the remaining lifetime prediction value for each group of data through a three-layer feature extraction operation.

[0047] The third step involves iteratively updating the internal parameters of the initial model of the prediction network based on the deviation between the predicted remaining lifetime and the corresponding remaining cycle lifetime label. For each set of predicted values, the deviation from the corresponding label is calculated. Gradient descent is used iteratively to adjust the internal weight parameters layer by layer from the deepest to the shallowest layer of the model, reducing the deviation between the predicted values ​​and the true labels. Each iteration of training on all training subsets is considered a complete training epoch.

[0048] The fourth step is to validate the accuracy of the iterated model using a validation subset from the lifetime prediction training dataset. After every ten rounds of training, all time-series data from the validation subset are input into the model of the current iteration, and the remaining lifetime prediction results of the validation set are output. The average prediction bias of all validation samples is calculated as the accuracy evaluation index of the model, where the prediction bias is the percentage of the absolute value of the difference between the predicted value and the true label divided by the true label.

[0049] Fifth, when the model output accuracy reaches the preset accuracy threshold, the iteration stops, resulting in a time-series lifetime prediction model with remaining lifetime output capability. If the average prediction bias of the validation set is lower than the preset accuracy threshold of 5%, the model accuracy is determined to meet the usage requirements, training stops, and the current model parameters are saved; if the accuracy threshold is not reached, iterative training continues until the accuracy meets the requirements or the number of iterations reaches the preset maximum number of training rounds, ultimately yielding a usable time-series lifetime prediction model.

[0050] In some embodiments, acquiring real-time operating data of the atomic layer deposition valve and inputting it into a time-series lifetime prediction model to obtain real-time remaining lifetime results includes: collecting real-time switching operation data of the currently operating atomic layer deposition valve according to a preset sampling period; performing outlier removal and time-series alignment preprocessing operations on the real-time switching operation data; inputting the preprocessed real-time switching operation data into the trained time-series lifetime prediction model; extracting time-series decay features from the real-time data through the time-series lifetime prediction model, and outputting the real-time remaining lifetime result of the current atomic layer deposition valve.

[0051] This embodiment describes the specific implementation method of obtaining the real-time operating data of the atomic layer deposition valve and inputting it into the time-series lifetime prediction model to obtain the real-time remaining lifetime result in step S102. The specific process is as follows: The first step is to collect real-time switching operation data of the currently operating atomic layer deposition valve according to a preset sampling period. The sampling period is dynamically set based on the valve's current switching frequency. When the valve is in a high-frequency continuous operation state, a set of data is collected every 500 switching actions; when the valve is in an intermittent operation state, a set of data is collected every 2,000 switching actions. The data types collected are completely consistent with the data types of the training dataset, including five categories of data: number of switching actions, action response time, sealing surface wear, air supply pressure fluctuation value, and action impact peak value.

[0052] The second step involves performing outlier removal and time-series alignment preprocessing on the real-time switch operation data. The preprocessing logic is completely consistent with that of the training dataset: first, the data integrity is verified, and if any data set is missing any value of any type, the entire set is discarded; then, outlier data that exceeds the normal range is removed; finally, time-series alignment is completed through linear interpolation to ensure that the real-time data format and sequence length of the input model are completely matched with the training data format.

[0053] The third step is to input the preprocessed real-time switch operation data into the trained time-series lifetime prediction model. The aligned time-series data sequence is then organized into a fixed-length input sequence according to the model input requirements and input into the trained time-series lifetime prediction model.

[0054] The fourth step involves extracting time-series decay features from real-time data using a time-series lifetime prediction model, and outputting the real-time remaining lifetime result of the current atomic layer deposition valve. The model employs a three-layer feature extraction structure to extract three types of time-series features from real-time data: the valve's operational performance decay trend, the sealing structure wear rate, and the degree of cumulative impact damage. Combined with the loss-life mapping learned during the training phase, the model calculates the remaining effective switching cycles of the current valve, which is then output as the real-time remaining lifetime result.

[0055] In some embodiments, the step of calculating the adjustment values ​​of switching speed and gas pressure parameters based on the real-time remaining lifetime results using an adaptive optimization algorithm for high-frequency switching action parameters includes: reading the range of switching speed and gas pressure parameters allowed by the atomic layer deposition process; inputting the real-time remaining lifetime results into the adaptive optimization algorithm for high-frequency switching action parameters to match the loss suppression target for the corresponding remaining lifetime range; and within the range allowed by the process, with the loss suppression target as a constraint, solving for the adjustment values ​​of switching speed and gas pressure parameters.

[0056] This embodiment describes the specific implementation method of calculating the adjustment values ​​of switching speed and air pressure parameters in step S102 based on the real-time remaining lifetime results and through an adaptive optimization algorithm for high-frequency switching action parameters. The specific process is as follows: The first step is to read the allowable switching speed and gas pressure parameter ranges for the atomic layer deposition process. These ranges are pre-stored in the process parameter library and are set by production line personnel according to the quality requirements of thin film deposition. These ranges include the upper and lower limits of the switching speed and the upper and lower limits of the gas supply pressure. All parameter adjustments must not exceed these ranges to avoid affecting the film quality of the deposition process.

[0057] The second step involves inputting the real-time remaining lifetime results into an adaptive optimization algorithm for high-frequency switching action parameters, matching the loss suppression target for the corresponding remaining lifetime interval. The algorithm pre-divides the remaining lifetime into three intervals: the first interval is defined as remaining lifetime greater than 60% of the rated total lifetime, with the corresponding loss suppression target being to maintain the current loss rate; the second interval is defined as remaining lifetime between 20% and 60% of the rated total lifetime, with the corresponding loss suppression target being to reduce unit switching loss by 15%; and the third interval is defined as remaining lifetime less than 20% of the rated total lifetime, with the corresponding loss suppression target being to reduce unit switching loss by 30%.

[0058] The third step involves solving for the switching speed adjustment value and the gas pressure parameter adjustment value within the allowable range of process parameters, constrained by the loss suppression target. Within the allowable range of process parameters, the algorithm iterates through all feasible combinations of switching speed and gas pressure parameters at a fixed step size, calculates the estimated unit switching loss value for each set of parameters, selects the parameter combination that satisfies the loss suppression target for the corresponding interval and minimizes the decrease in switching speed, and outputs the difference between this combination and the current operating parameters as the switching speed adjustment value and the gas pressure parameter adjustment value.

[0059] In some embodiments, the step of sending the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters includes: performing process compliance verification on the calculated switching speed adjustment value and air pressure parameter adjustment value; generating parameter control instructions for the corresponding actuator after the verification is passed; sending the parameter control instructions to the actuator of the atomic layer deposition valve; and having the actuator adjust the valve's switching speed and air supply pressure according to the parameter control instructions to complete the dynamic update of the switching action parameters.

[0060] This embodiment describes the specific implementation method of sending the adjustment value to the atomic layer deposition valve actuator and dynamically adjusting the switching action parameters in step S103. The specific process is as follows: In the first step, process compliance verification is performed on the calculated adjustment value of switching speed and the calculated adjustment value of air pressure parameter. The verification is divided into two layers. The first layer is parameter range verification, in which the adjusted values of the two solved parameters are respectively compared with the pre-stored allowable value ranges of the process. If both parameters are within the value ranges, the first layer of verification is passed. The second layer is adjustment amplitude verification, in which it is judged whether the change amount of the switching speed in a single adjustment exceeds ten percent of the rated value and whether the change amount of the air pressure exceeds five percent of the rated value. If neither of the two items is exceeded, the second layer of verification is passed. If any layer of verification fails, the process returns to the parameter solving step to recalculate the adjustment values.

[0061] In the second step, a parameter control instruction for the corresponding actuator is generated after the verification is passed. The parameter adjustment values in numerical form are converted into digital control instructions recognizable by the actuator of the atomic layer deposition valve. The content of the instruction includes the adjustment amount of switching speed, the adjustment amount of supply air pressure and the execution effective time. The effective time is set to the gap period after the end of the current deposition batch, so as to prevent the adjustment action from affecting the ongoing deposition process.

[0062] In the third step, the parameter control instruction is issued to the actuator of the atomic layer deposition valve. The control instruction is transmitted to the pneumatic actuator matched with the atomic layer deposition valve through the industrial field control bus, and a data check code is attached during the transmission process to ensure accurate instruction transmission and avoid parameter abnormality caused by transmission errors.

[0063] In the fourth step, the actuator adjusts the switching speed of the valve member and the supply air pressure according to the parameter control instruction, so as to complete the dynamic update of the switching action parameters. After receiving the control instruction, the actuator adjusts the opening of the proportional valve of the pneumatic loop to change the supply air pressure at the preset effective time, and simultaneously adjusts the opening of the throttle valve of the pneumatic loop to change the switching action speed. After the adjustment is completed, a feedback signal indicating that the parameter adjustment is completed is returned.

[0064] In some embodiments, the collecting the operation loss data after the adjustment of the switching action parameters, feeding back the data to the time-series life prediction model to update the remaining life result, and completing the closed-loop control of life optimization comprises: after the adjustment of the switching action parameters is completed, collecting the operation loss data of the atomic layer deposition valve within a preset operation duration; inputting the operation loss data into the time-series life prediction model to refresh the current remaining life result; and taking the refreshed remaining life result as the input basis for the next round of parameter optimization, so as to form a continuously iterated closed-loop control of life optimization.

[0065] This embodiment describes the specific implementation manner of collecting the operation loss data after the adjustment of the switching action parameters, feeding back the data to the time-series life prediction model to update the remaining life result, and completing the closed-loop control of life optimization in step S103, and the specific process is as follows: In the first step, after the adjustment of the switch action parameters is completed, the operation loss data of the atomic layer deposition valve within the preset operation duration is collected. The preset operation duration is set as the operation period in which the valve completes one thousand switch actions in total. The collected operation loss data includes four types of data in this period: the average action response time, the seal wear increment per hundred switch actions, the average value of the action impact peak, and the gas supply pressure stability, which are used to characterize the actual loss state of the valve after parameter adjustment.

[0066] In the second step, the operation loss data is input into the time-series life prediction model to refresh the current remaining life result. The collected loss time-series data is updated to the input sequence of the model, and the sliding window method is adopted to replace the earliest historical data to keep the length of the input sequence fixed. The model recalculates the remaining effective switch cycle times of the valve in combination with the new loss data, and obtains the refreshed remaining life result.

[0067] In the third step, the refreshed remaining life result is used as the input basis for the next round of parameter optimization to form a continuously iterative closed-loop control for life optimization. The refreshed remaining life result will be input into the high-frequency switch action parameter adaptive optimization algorithm to start the next round of parameter adjustment calculation, and the complete process of life prediction, parameter solving, parameter issuance and data feedback is repeatedly executed. The trigger condition for closed-loop optimization is set to be triggered every time the cumulative number of switch actions of the valve increases by five thousand times, or triggered when the decrease range of the remaining life exceeds five percent. The operation parameters are continuously iterated and optimized throughout the entire service cycle of the atomic layer deposition valve to realize dynamic and continuous optimization of the service life.

[0068] In some embodiments, the method further comprises: collecting the thin film deposition yield data of the deposition chamber corresponding to the atomic layer deposition valve in real time; inputting the thin film deposition yield data into the high-frequency switch action parameter adaptive optimization algorithm, and adding a yield compliance constraint condition; when the parameter adjustment value will cause the yield to be lower than the preset yield threshold, the adjustment values of the switch speed and air pressure parameters are re-solved to ensure that the life optimization process meets the yield requirement of the semiconductor deposition process.

[0069] This embodiment provides an additional optimization control scheme, which adds a yield constraint in the parameter calculation process of step S102 to avoid the impact of life optimization on the quality of the deposition process, and the specific implementation process is as follows: In the first step, the thin film deposition yield data of the deposition chamber corresponding to the atomic layer deposition valve is collected in real time. The yield data after each batch of deposition is obtained from the film thickness detection system and the defect detection system matched with the deposition chamber. The yield judgment criteria include three indicators: the thin film thickness deviation is within plus or minus one percent, the thin film defect density is less than 0.5 per square centimeter, and the thin film composition deviation is within plus or minus 0.5 percent. The proportion of wafers that meet all three indicators at the same time is the yield value of the corresponding batch.

[0070] The second step involves inputting the thin film deposition yield data into the adaptive optimization algorithm for high-frequency switch action parameters, and adding yield compliance constraints. Based on the existing loss suppression targets and process parameter range constraints, a new yield constraint is added, requiring that the estimated thin film deposition yield after parameter adjustment must not be lower than a preset yield threshold. The preset yield threshold is set to 99.5% according to the production line process requirements.

[0071] The third step involves resolving the adjustment values ​​for the switching speed and air pressure parameters when the parameter adjustments cause the yield to fall below the preset yield threshold. This ensures that the lifetime optimization process meets the yield requirements of the semiconductor deposition process. After obtaining the initial adjustment values, the algorithm first queries the estimated yield level corresponding to the adjusted parameters based on the historical parameter-yield mapping table. If the estimated yield is lower than the preset threshold, the loss suppression target level is gradually reduced, and the parameter adjustment values ​​are resolved until the obtained adjustment values ​​simultaneously meet both the loss suppression target and the yield requirement, ensuring that the lifetime optimization operation does not reduce the finished product quality of the semiconductor process.

[0072] In some embodiments, the method further includes: obtaining the real-time remaining lifetime results of all atomic layer deposition valves in the same semiconductor production line; calculating the dispersion of the remaining lifetime of all valves; when the dispersion exceeds a preset equilibrium threshold, adjusting the process load allocation corresponding to each atomic layer deposition valve; so that the remaining lifetime of all valves tends to be consistent, thereby achieving lifetime equilibrium optimization of the production line valve cluster.

[0073] This embodiment provides a production line-level cluster optimization scheme, applicable to semiconductor production lines containing multiple atomic layer deposition valves. It can achieve balanced lifespan of multiple valves and reduce the risk of concentrated downtime. The specific implementation process is as follows: The first step is to obtain the real-time remaining lifetime results for all atomic layer deposition valves within the same semiconductor production line. The real-time remaining lifetime results for all atomic layer deposition valves in all deposition chambers are aggregated through the production line-level control system. All remaining lifetime results are calculated and judged using a unified standard to ensure the comparability of remaining lifetime data for different valves.

[0074] The second step is to calculate the dispersion of the remaining lifespan of all valves. The coefficient of dispersion is used as a characterization index of the dispersion. First, the standard deviation of the remaining lifespan of all valves is calculated. Then, the standard deviation is divided by the average of the remaining lifespan of all valves to obtain the coefficient of dispersion. The larger the coefficient of dispersion, the greater the difference in the remaining lifespan of different valves, and the more likely it is that a single valve will fail prematurely, leading to production line downtime.

[0075] The third step involves adjusting the process load distribution for each atomic layer deposition valve when the dispersion exceeds a preset equilibrium threshold. The preset equilibrium threshold is set at 15%. If the dispersion coefficient exceeds the threshold, the process scheduling logic of the production line is adjusted to increase the number of wafer processing batches for deposition chambers with valves having higher remaining lifespans and decrease the number of wafer processing batches for deposition chambers with valves having lower remaining lifespans. The load change ratio for a single adjustment should not exceed 10% of the rated load of a single chamber to avoid sudden load changes affecting the stability of the production line operation.

[0076] The fourth step is to bring the remaining lifespan of all valves closer together, achieving balanced optimization of the valve cluster's lifespan across the production line. Through continuous dynamic load adjustments, the remaining lifespan of all atomic layer deposition valves on the production line is gradually brought closer together, ultimately controlling the difference in remaining lifespan of all valves to within 10%. This facilitates unified scheduling of valve maintenance and replacement on the production line, reduces unplanned downtime, and improves the overall operational and maintenance efficiency of the production line.

[0077] In some embodiments, such as Figure 2 As shown, this embodiment provides a production line-level cluster optimization scheme, applicable to semiconductor production lines containing multiple atomic layer deposition valves. It can achieve balanced lifespan of multiple valves and reduce the risk of concentrated downtime. The specific implementation process is as follows: The first step is to obtain the real-time remaining lifetime results for all atomic layer deposition valves within the same semiconductor production line. The real-time remaining lifetime results for all atomic layer deposition valves in all deposition chambers are aggregated through the production line-level control system. All remaining lifetime results are calculated and judged using a unified standard to ensure the comparability of remaining lifetime data for different valves.

[0078] The second step is to calculate the dispersion of the remaining lifespan of all valves. The coefficient of dispersion is used as a characterization index of the dispersion. First, the standard deviation of the remaining lifespan of all valves is calculated. Then, the standard deviation is divided by the average of the remaining lifespan of all valves to obtain the coefficient of dispersion. The larger the coefficient of dispersion, the greater the difference in the remaining lifespan of different valves, and the more likely it is that a single valve will fail prematurely, leading to production line downtime.

[0079] The third step involves adjusting the process load distribution for each atomic layer deposition valve when the dispersion exceeds a preset equilibrium threshold. The preset equilibrium threshold is set at 15%. If the dispersion coefficient exceeds the threshold, the process scheduling logic of the production line is adjusted to increase the number of wafer processing batches for deposition chambers with valves having higher remaining lifespans and decrease the number of wafer processing batches for deposition chambers with valves having lower remaining lifespans. The load change ratio for a single adjustment should not exceed 10% of the rated load of a single chamber to avoid sudden load changes affecting the stability of the production line operation.

[0080] The fourth step is to bring the remaining lifespan of all valves closer together, achieving balanced optimization of the valve cluster's lifespan across the production line. Through continuous dynamic load adjustments, the remaining lifespan of all atomic layer deposition valves on the production line is gradually brought closer together, ultimately controlling the difference in remaining lifespan of all valves to within 10%. This facilitates unified scheduling of valve maintenance and replacement on the production line, reduces unplanned downtime, and improves the overall operational and maintenance efficiency of the production line.

[0081] Please see Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of an ALD valve high-frequency switching lifetime optimization system 200 for semiconductors provided in this application embodiment. The ALD valve high-frequency switching lifetime optimization system 200 for semiconductors is used to execute the steps of the ALD valve high-frequency switching lifetime optimization method for semiconductors shown in the above embodiments. The ALD valve high-frequency switching lifetime optimization system 200 for semiconductors can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, laptop computer, wearable device, or robot.

[0082] like Figure 3 As shown, the ALD valve high-frequency switching life optimization system 200 for semiconductors includes: The data acquisition unit 201 is used to collect historical switching operation data of the atomic layer deposition valve and construct a lifetime prediction training dataset; based on the lifetime prediction training dataset, the time-series lifetime prediction model is trained so that the time-series lifetime prediction model has the ability to output the remaining lifetime. The data acquisition unit 202 is used to acquire the real-time operating data of the atomic layer deposition valve, input the time-series lifetime prediction model to obtain the real-time remaining lifetime result; based on the real-time remaining lifetime result, the adjustment values ​​of the switching speed and air pressure parameters are calculated by the high-frequency switching action parameter adaptive optimization algorithm. The control completion unit 203 is used to send the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; collect the operating loss data after the switching action parameters are adjusted, and feed it back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

[0083] In some embodiments, the process of collecting historical switching operation data of atomic layer deposition valves to construct a life prediction training dataset includes: collecting the number of switching operations, action response time, sealing surface wear, air supply pressure fluctuation value, and action impact peak data of multiple atomic layer deposition valves throughout their entire life cycle; performing outlier removal and time series alignment processing on all collected data; labeling each set of time series data with the remaining cycle life label of the corresponding valve component; and dividing the labeled time series data into a training subset and a validation subset to form a life prediction training dataset.

[0084] In some embodiments, training a time-series lifetime prediction model based on a lifetime prediction training dataset to enable the time-series lifetime prediction model to output remaining lifetime includes: constructing an initial model of a prediction network containing a multi-layer time-series feature extraction structure; inputting a training subset from the lifetime prediction training dataset into the initial model of the prediction network to output remaining lifetime prediction values; iteratively updating the internal parameters of the initial model of the prediction network based on the deviation between the remaining lifetime prediction values ​​and the corresponding remaining cycle lifetime labels; verifying the accuracy of the iterated model using a validation subset from the lifetime prediction training dataset; and stopping the iteration when the model output accuracy reaches a preset accuracy threshold, thereby obtaining a time-series lifetime prediction model with remaining lifetime output capability.

[0085] In some embodiments, acquiring real-time operating data of the atomic layer deposition valve and inputting it into a time-series lifetime prediction model to obtain real-time remaining lifetime results includes: collecting real-time switching operation data of the currently operating atomic layer deposition valve according to a preset sampling period; performing outlier removal and time-series alignment preprocessing operations on the real-time switching operation data; inputting the preprocessed real-time switching operation data into the trained time-series lifetime prediction model; extracting time-series decay features from the real-time data through the time-series lifetime prediction model, and outputting the real-time remaining lifetime result of the current atomic layer deposition valve.

[0086] In some embodiments, the step of calculating the adjustment values ​​of switching speed and gas pressure parameters based on the real-time remaining lifetime results using an adaptive optimization algorithm for high-frequency switching action parameters includes: reading the range of switching speed and gas pressure parameters allowed by the atomic layer deposition process; inputting the real-time remaining lifetime results into the adaptive optimization algorithm for high-frequency switching action parameters to match the loss suppression target for the corresponding remaining lifetime range; and within the range allowed by the process, with the loss suppression target as a constraint, solving for the adjustment values ​​of switching speed and gas pressure parameters.

[0087] In some embodiments, the step of sending the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters includes: performing process compliance verification on the calculated switching speed adjustment value and air pressure parameter adjustment value; generating parameter control instructions for the corresponding actuator after the verification is passed; sending the parameter control instructions to the actuator of the atomic layer deposition valve; and having the actuator adjust the valve's switching speed and air supply pressure according to the parameter control instructions to complete the dynamic update of the switching action parameters.

[0088] In some embodiments, the step of collecting operational loss data after adjusting the switching action parameters and feeding it back to the time-series lifetime prediction model to update the remaining lifetime result and complete lifetime optimization closed-loop control includes: after the switching action parameters are adjusted, collecting operational loss data of the atomic layer deposition valve within a preset operating time; inputting the operational loss data into the time-series lifetime prediction model to refresh the current remaining lifetime result; and using the refreshed remaining lifetime result as the input basis for the next round of parameter optimization to form a continuously iterative lifetime optimization closed-loop control.

[0089] In some embodiments, the method further includes: real-time acquisition of thin film deposition yield data of the deposition chamber corresponding to the atomic layer deposition valve; inputting the thin film deposition yield data into an adaptive optimization algorithm for high-frequency switching action parameters, and adding yield compliance constraints; when parameter adjustment values ​​cause the yield to fall below a preset yield threshold, resolving the adjustment values ​​of switching speed and air pressure parameters to ensure that the lifetime optimization process meets the yield requirements of the semiconductor deposition process.

[0090] In some embodiments, the method further includes: obtaining the real-time remaining lifetime results of all atomic layer deposition valves in the same semiconductor production line; calculating the dispersion of the remaining lifetime of all valves; when the dispersion exceeds a preset equilibrium threshold, adjusting the process load allocation corresponding to each atomic layer deposition valve; so that the remaining lifetime of all valves tends to be consistent, thereby achieving lifetime equilibrium optimization of the production line valve cluster.

[0091] It should be noted that those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the ALD valve high-frequency switching life optimization system and its modules described above for semiconductors can be referred to the corresponding content in the various embodiments of the ALD valve high-frequency switching life optimization method for semiconductors, and will not be repeated here.

[0092] The aforementioned method for optimizing the high-frequency switching life of ALD valves for semiconductors can be implemented as a computer program, which can be used in applications such as... Figure 3 It runs on the system shown.

[0093] Please see Figure 4 , Figure 4 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0094] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any method for optimizing the high-frequency switching life of an ALD valve used in semiconductors.

[0095] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0096] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program can enable the processor to perform any method for optimizing the high-frequency switching life of ALD valves for semiconductors.

[0097] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0098] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0099] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: Historical on / off operation data of atomic layer deposition valves were collected to construct a lifetime prediction training dataset; a time-series lifetime prediction model was trained based on the lifetime prediction training dataset, enabling the time-series lifetime prediction model to output remaining lifetime. Real-time operating data of the atomic layer deposition valve is acquired and input into a time-series lifetime prediction model to obtain real-time remaining lifetime results. Based on the real-time remaining lifetime results, the adjustment values ​​of switching speed and gas pressure parameters are calculated using a high-frequency switching action parameter adaptive optimization algorithm. The adjustment value is sent to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; the operating loss data after the switching action parameters are adjusted is collected and fed back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

[0100] In some embodiments, the process of collecting historical switching operation data of atomic layer deposition valves to construct a life prediction training dataset includes: collecting the number of switching operations, action response time, sealing surface wear, air supply pressure fluctuation value, and action impact peak data of multiple atomic layer deposition valves throughout their entire life cycle; performing outlier removal and time series alignment processing on all collected data; labeling each set of time series data with the remaining cycle life label of the corresponding valve component; and dividing the labeled time series data into a training subset and a validation subset to form a life prediction training dataset.

[0101] In some embodiments, training a time-series lifetime prediction model based on a lifetime prediction training dataset to enable the time-series lifetime prediction model to output remaining lifetime includes: constructing an initial model of a prediction network containing a multi-layer time-series feature extraction structure; inputting a training subset from the lifetime prediction training dataset into the initial model of the prediction network to output remaining lifetime prediction values; iteratively updating the internal parameters of the initial model of the prediction network based on the deviation between the remaining lifetime prediction values ​​and the corresponding remaining cycle lifetime labels; verifying the accuracy of the iterated model using a validation subset from the lifetime prediction training dataset; and stopping the iteration when the model output accuracy reaches a preset accuracy threshold, thereby obtaining a time-series lifetime prediction model with remaining lifetime output capability.

[0102] In some embodiments, acquiring real-time operating data of the atomic layer deposition valve and inputting it into a time-series lifetime prediction model to obtain real-time remaining lifetime results includes: collecting real-time switching operation data of the currently operating atomic layer deposition valve according to a preset sampling period; performing outlier removal and time-series alignment preprocessing operations on the real-time switching operation data; inputting the preprocessed real-time switching operation data into the trained time-series lifetime prediction model; extracting time-series decay features from the real-time data through the time-series lifetime prediction model, and outputting the real-time remaining lifetime result of the current atomic layer deposition valve.

[0103] In some embodiments, the step of calculating the adjustment values ​​of switching speed and gas pressure parameters based on the real-time remaining lifetime results using an adaptive optimization algorithm for high-frequency switching action parameters includes: reading the range of switching speed and gas pressure parameters allowed by the atomic layer deposition process; inputting the real-time remaining lifetime results into the adaptive optimization algorithm for high-frequency switching action parameters to match the loss suppression target for the corresponding remaining lifetime range; and within the range allowed by the process, with the loss suppression target as a constraint, solving for the adjustment values ​​of switching speed and gas pressure parameters.

[0104] In some embodiments, the step of sending the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters includes: performing process compliance verification on the calculated switching speed adjustment value and air pressure parameter adjustment value; generating parameter control instructions for the corresponding actuator after the verification is passed; sending the parameter control instructions to the actuator of the atomic layer deposition valve; and having the actuator adjust the valve's switching speed and air supply pressure according to the parameter control instructions to complete the dynamic update of the switching action parameters.

[0105] In some embodiments, the step of collecting operational loss data after adjusting the switching action parameters and feeding it back to the time-series lifetime prediction model to update the remaining lifetime result and complete lifetime optimization closed-loop control includes: after the switching action parameters are adjusted, collecting operational loss data of the atomic layer deposition valve within a preset operating time; inputting the operational loss data into the time-series lifetime prediction model to refresh the current remaining lifetime result; and using the refreshed remaining lifetime result as the input basis for the next round of parameter optimization to form a continuously iterative lifetime optimization closed-loop control.

[0106] In some embodiments, the method further includes: real-time acquisition of thin film deposition yield data of the deposition chamber corresponding to the atomic layer deposition valve; inputting the thin film deposition yield data into an adaptive optimization algorithm for high-frequency switching action parameters, and adding yield compliance constraints; when parameter adjustment values ​​cause the yield to fall below a preset yield threshold, resolving the adjustment values ​​of switching speed and air pressure parameters to ensure that the lifetime optimization process meets the yield requirements of the semiconductor deposition process.

[0107] In some embodiments, the method further includes: obtaining the real-time remaining lifetime results of all atomic layer deposition valves in the same semiconductor production line; calculating the dispersion of the remaining lifetime of all valves; when the dispersion exceeds a preset equilibrium threshold, adjusting the process load allocation corresponding to each atomic layer deposition valve; so that the remaining lifetime of all valves tends to be consistent, thereby achieving lifetime equilibrium optimization of the production line valve cluster.

[0108] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method for optimizing the high-frequency switching life of an ALD valve for semiconductors as provided in any embodiment of this application.

[0109] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing the high-frequency switching life of an ALD valve for semiconductors, characterized in that, include: Historical on / off operation data of atomic layer deposition valves were collected to construct a lifetime prediction training dataset; a time-series lifetime prediction model was trained based on the lifetime prediction training dataset, enabling the time-series lifetime prediction model to output remaining lifetime. Acquire real-time operating data of the atomic layer deposition valve and input it into the time-series lifetime prediction model to obtain real-time remaining lifetime results; Based on the real-time remaining lifetime results, the adjustment values ​​of the switching speed and air pressure parameters are calculated by the adaptive optimization algorithm of the high-frequency switching action parameters. The adjustment value is sent to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; the operating loss data after the switching action parameters are adjusted is collected and fed back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

2. The method according to claim 1, characterized in that, The historical on / off operation data of the atomic layer deposition valve is collected to construct a lifetime prediction training dataset, including: Data on the number of opening and closing times, action response time, sealing surface wear, air supply pressure fluctuation, and peak action impact of multiple atomic layer deposition valves throughout their entire life cycle were collected. All collected data were processed for outlier removal and time-series alignment. Label each set of time series data with the remaining cycle life tag of the corresponding valve; The labeled time-series data is divided into a training subset and a validation subset to form a lifetime prediction training dataset.

3. The method according to claim 1, characterized in that, The process of training a time-series lifetime prediction model based on a lifetime prediction training dataset, enabling the model to output remaining lifetime, includes: Construct an initial model of a prediction network that includes a multi-layer temporal feature extraction structure; Input a training subset from the lifetime prediction training dataset into the initial model of the prediction network, and output the remaining lifetime prediction value; The internal parameters of the initial model of the prediction network are iteratively updated based on the deviation between the predicted remaining lifetime and the corresponding remaining cycle lifetime label. The accuracy of the iterative model was verified using a validation subset from the lifetime prediction training dataset. When the model output accuracy reaches the preset accuracy threshold, the iteration stops, and a time-series lifetime prediction model with the ability to output remaining lifetime is obtained.

4. The method according to claim 1, characterized in that, The process of acquiring real-time operating data of the atomic layer deposition valve and inputting it into a time-series lifetime prediction model to obtain real-time remaining lifetime results includes: Collect real-time on / off operation data of the currently operating atomic layer deposition valve according to the preset sampling period; Perform outlier removal and timing alignment preprocessing operations on real-time switch operation data; The preprocessed real-time switch operation data is input into the completed time-series lifetime prediction model; The time-series decay characteristics in real-time data are extracted by a time-series lifetime prediction model, and the real-time remaining lifetime result of the current atomic layer deposition valve is output.

5. The method according to claim 1, characterized in that, The step of calculating the adjustment values ​​of switching speed and air pressure parameters based on the real-time remaining lifetime results using a high-frequency switching action parameter adaptive optimization algorithm includes: Read the allowable switching speed range and gas pressure parameter range for the atomic layer deposition process; The real-time remaining lifetime results are input into the adaptive optimization algorithm for high-frequency switch action parameters to match the loss suppression target for the corresponding remaining lifetime range; Within the range of values ​​allowed by the process, and with the loss suppression target as a constraint, the switching speed adjustment value and the air pressure parameter adjustment value are obtained by solving.

6. The method according to claim 1, characterized in that, The step of sending the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters includes: Verify the process compliance of the calculated switching speed adjustment value and air pressure parameter adjustment value; After successful verification, the corresponding actuator parameter control instructions are generated. The parameter control commands are sent to the actuator of the atomic layer deposition valve; The actuator adjusts the valve's opening and closing speed and the air supply pressure according to the parameter control instructions, thereby completing the dynamic update of the opening and closing action parameters.

7. The method according to claim 1, characterized in that, The operational loss data after the acquisition switch action parameters are adjusted is fed back to the time-series lifetime prediction model to update the remaining lifetime result, completing the lifetime optimization closed-loop control, including: After the switching action parameters are adjusted, the operating loss data of the atomic layer deposition valve is collected within the preset running time. Input the operational loss data into the time-series lifetime prediction model and refresh the current remaining lifetime result; The updated remaining lifetime results are used as input for the next round of parameter optimization, forming a continuously iterative lifetime optimization closed-loop control.

8. The method according to claim 1, characterized in that, The method further includes: Real-time acquisition of thin film deposition yield data for the deposition chamber corresponding to the atomic layer deposition valve; Input the thin film deposition yield data into the adaptive optimization algorithm for high-frequency switch action parameters, and add yield target constraint conditions; When parameter adjustments cause the yield to fall below the preset yield threshold, the adjustment values ​​for switching speed and air pressure parameters are recalculated to ensure that the lifetime optimization process meets the yield requirements of the semiconductor deposition process.

9. The method according to claim 1, characterized in that, The method further includes: Obtain the real-time remaining lifetime results for all atomic layer deposition valves within the same semiconductor production line; Calculate the dispersion of the remaining life of all valve components; When the dispersion exceeds the preset equalization threshold, adjust the process load distribution corresponding to each atomic layer deposition valve; This ensures that the remaining lifespan of all valve components is consistent, achieving balanced optimization of the lifespan of the valve component cluster on the production line.

10. A high-frequency switching lifetime optimization system for ALD valves in semiconductors, for implementing the method as described in any one of claims 1-9, characterized in that, include: The data acquisition unit is used to collect historical switching operation data of the atomic layer deposition valve and construct a lifetime prediction training dataset; based on the lifetime prediction training dataset, the time-series lifetime prediction model is trained, enabling the time-series lifetime prediction model to output the remaining lifetime. The data acquisition unit is used to acquire real-time operating data of the atomic layer deposition valve and input it into the time-series lifetime prediction model to obtain real-time remaining lifetime results. Based on the real-time remaining lifetime results, the adjustment values ​​of the switching speed and air pressure parameters are calculated by the adaptive optimization algorithm of the high-frequency switching action parameters. The control completion unit is used to send the adjustment value to the atomic layer deposition valve actuator to dynamically adjust the switching action parameters; collect the operating loss data after the switching action parameters are adjusted, and feed it back to the time-series lifetime prediction model to update the remaining lifetime result, thus completing the lifetime optimization closed-loop control.

Images