Semiconductor machine control method and device, medium and product

Abstract

The invention relates to the technical field of semiconductor manufacturing, in particular to a semiconductor machine control method and device, a medium and a product, and the method comprises the steps: training an initial weight prediction model according to training time sequence data until a weight prediction error is not greater than a preset target value, and obtaining a target weight prediction model; inputting the test time sequence data into the target weight prediction model, and predicting a target weight; the test time sequence data comprises process parameter time sequence data, machine state parameter time sequence data and process measurement result time sequence data; according to the predicted target weight and the actual feedback data and historical feedback data of the current batch of wafers, obtaining process parameter prediction data of the current batch of wafers; and controlling the semiconductor machine to execute the target technological process on the current batch of wafers according to the technological parameter prediction data. And at least according to the dynamic change of the actual working condition of the machine, the process parameters of the machine can be dynamically tracked and predicted in real time.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor manufacturing technology, and in particular to a semiconductor machine control method, apparatus, medium, and product. Background Technology

[0002] Advanced Process Control (APC) plays a central role in semiconductor manufacturing, and its application has evolved from single-point automation to end-to-end intelligent collaboration. Through multi-dimensional real-time control and intelligent decision-making, APC enables a leap in the precision and stability of semiconductor equipment processes.

[0003] However, APC requires relevant personnel to set the weights of historical data on the prediction results based on their experience and observations of historical data from semiconductor equipment. On the one hand, manually set fixed weights are difficult to reflect the dynamic changes in random process factors during actual equipment operation; on the other hand, they are highly dependent on the experience of the personnel and have a lag effect. Summary of the Invention

[0004] Therefore, it is necessary to address the problems mentioned in the background art by providing a semiconductor equipment control method, device, medium, and product that can at least dynamically track and predict the process parameters of the equipment in real time based on the dynamic changes in the actual operating conditions of the equipment, realize real-time online optimization of process parameters, improve the yield and accuracy of semiconductor equipment processes, and effectively avoid dependence on operator experience.

[0005] To achieve the above and other objectives, according to various embodiments of the present disclosure, one aspect of the present disclosure provides a semiconductor machine tool control method, comprising:

[0006] The initial weight prediction model is trained based on the training time series data until the weight prediction error is no greater than the preset target value, and the target weight prediction model is obtained. The training time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data.

[0007] The test time series data is input into the target weight prediction model to predict the target weight; the test time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data.

[0008] Based on the predicted target weights, the actual feedback data of the current batch of wafers, and the historical feedback data, the predicted process parameters of the current batch of wafers are obtained.

[0009] Based on the predicted data of process parameters, the semiconductor equipment is controlled to perform the target process on the current batch of wafers.

[0010] In some embodiments, the equipment includes an etching equipment; the initial weight prediction model includes a time-recurrent neural network; the process parameters include RF power, gas flow rate, chamber pressure, etching time, and electrostatic chuck temperature; the equipment status parameters include plasma impedance change rate and RF time; the plasma impedance change rate is used at least to characterize the thickness of the deposits accumulated on the chamber wall; the RF time is used at least to characterize the interval between two adjacent maintenance start times; training the initial weight prediction model includes: adjusting at least one of the learning rate and the number of hidden layer units of the time-recurrent neural network until the weight prediction error is not greater than a preset target value.

[0011] In some embodiments, the semiconductor equipment control method further includes: obtaining initial process parameter prediction data for the current batch of wafers based on an initial preset target value, actual feedback data of the current batch of wafers, and historical feedback data; controlling the semiconductor equipment to perform a target etching process on the current batch of wafers based on the initial process parameter prediction data to obtain an initial critical dimension; and determining a preset target value based on the initial critical dimension if the initial critical dimension is not greater than a critical dimension threshold.

[0012] In some embodiments, the time recurrent neural network includes an input layer, multiple network layers, and an output layer. The input layer includes time series data of process parameters, time series data of machine status parameters, and time series data of process measurement results at multiple time steps. Each network layer in the multiple network layers includes multiple units. The output layer includes a sigmoid activation function to generate dynamic prediction weights.

[0013] In some embodiments, the semiconductor machine control method further includes: generating a time series of target weights; and generating training time series data based on the time series of target weights.

[0014] In some embodiments, the time series of the target weight includes a first stage, a second stage, and a third stage distributed in chronological order; the minimum value of the first stage is greater than the maximum value of the second stage; and the maximum value of the second stage is less than the minimum value of the third stage.

[0015] In some embodiments, the first stage shows a downward trend; the third stage shows an upward trend.

[0016] In another aspect, this disclosure provides a semiconductor equipment control device, including a model training module, a weight prediction module, a process parameter prediction module, and a control module. The model training module is used to train an initial weight prediction model based on training time-series data until the weight prediction error is no greater than a preset target value, thereby obtaining a target weight prediction model. The training time-series data includes process parameter time-series data, equipment status parameter time-series data, and process measurement result time-series data. The weight prediction module is used to input test time-series data into the target weight prediction model to predict the target weight. The test time-series data includes process parameter time-series data, equipment status parameter time-series data, and process measurement result time-series data. The process parameter prediction module is used to obtain process parameter prediction data for the current batch of wafers based on the target weight, the actual feedback data of the current batch of wafers, and historical feedback data. The control module is used to control the semiconductor equipment to execute the target process on the current batch of wafers based on the process parameter prediction data.

[0017] In another aspect, this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described semiconductor machine control methods.

[0018] In another aspect, this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described semiconductor machine control methods.

[0019] The above-described embodiments can produce at least the following unexpected technical effects:

[0020] Based on training time-series data including process parameter time-series data and machine status parameter time-series data, an initial weight prediction model is trained until the weight prediction error is no greater than a preset target value, resulting in a target weight prediction model with expected weight prediction accuracy. Since the target weight prediction model has already fitted the specific correlation between the process parameter time-series data, machine status parameter time-series data, and the predicted target weights, inputting the test time-series data into the target weight prediction model allows for real-time and accurate online prediction of the target weights, avoiding the lag and reliance on experience inherent in manual prediction. Because the test time-series data includes process parameter time-series data and machine status parameter time-series data, encompassing the dynamic changes in both process parameters and machine status parameters over time, the predicted target weights already reflect the influence of random dynamic changes in the real-time operating state of the machine. Therefore, based on the predicted target weight, the actual feedback data of the current batch of wafers, and the historical feedback data, the process parameter prediction data of the current batch of wafers can be accurately predicted. This allows for precise control of the semiconductor equipment to execute the target process on the current batch of wafers based on the predicted process parameter data, achieving real-time online optimization of process parameters. This improves the yield and accuracy of the semiconductor equipment process while effectively avoiding reliance on operator experience. Attached Figure Description

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A time-series diagram illustrating the critical dimensions of the film etching process;

[0023] Figure 2 This is a schematic diagram of an application scenario in one embodiment of the present disclosure;

[0024] Figure 3 This is a schematic flowchart of a semiconductor machine control method provided in one embodiment of the present disclosure;

[0025] Figure 4 This is a schematic diagram of the architecture of a time-recurrent neural network provided in one embodiment of the present disclosure;

[0026] Figure 5 This is a schematic diagram illustrating the specific application process of a semiconductor machine control method provided in one embodiment of the present disclosure;

[0027] Figure 6This is a schematic diagram of the dynamic target weight time series generated in a semiconductor machine control method provided in one embodiment of the present disclosure;

[0028] Figure 7 A comparative diagram of prediction curves obtained by different prediction methods;

[0029] Figure 8 This is a schematic diagram of the structure of a semiconductor machine tool control device provided in one embodiment of the present disclosure;

[0030] Figure 9 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present disclosure.

[0031] Explanation of reference numerals in the attached figures:

[0032] 100. Semiconductor equipment control device; 10. Model training module; 102. Server; 104. Processor; 20. Weight prediction module; 30. Process parameter prediction module; 40. Control module. Detailed Implementation

[0033] To facilitate understanding of this disclosure, a more complete description will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the present disclosure. However, this disclosure can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure.

[0035] When using the terms “including,” “having,” and “comprising” as described herein, another component may be added unless explicitly qualifying terms such as “only,” “consisting of,” etc. are used. Unless otherwise stated, singular terms may include plural forms and should not be construed as having a quantity of one.

[0036] In this disclosure, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0037] Furthermore, the terms “first”, “second”, etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.

[0038] Advanced Process Control (APC) refers to the use of mathematical models, real-time data, and optimization algorithms to surpass traditional PID control, providing more precise and stable control over complex industrial processes (especially semiconductor manufacturing) to improve yield, consistency, and efficiency. APC typically includes feedforward and feedback control loops.

[0039] Exponentially Weighted Moving Average (EWMA) is used to calculate the weighted average of time series data.

[0040] Lot typically refers to a group (e.g., 25 wafers) of wafers that are processed together on a machine using the same procedure.

[0041] In this embodiment of the disclosure, the current APC system predicts the feedback value for each batch of Lots using the EWMA algorithm. Before processing the next Lot, based on historical feedback values, the feedback value that the next Lot may generate under the current machine / recipe is predicted. For semiconductor etching machines, this embodiment of the disclosure aims to adjust the process parameters (such as temperature, time, gas flow rate, etc.) of the next Lot based on current and historical actual measurement values ​​(feedback) so that its output feedback value is as close as possible to a preset target value.

[0042] However, when using the EWMA algorithm to calculate the weighted average of time series data, the weights decay exponentially with the "age" of the data points. This means that more recent data points have a greater impact on the average. However, the condition of semiconductor manufacturing equipment (such as chamber cleanliness, component aging, and temperature drift) changes slowly over time (drift). This change is typically gradual. Simple arithmetic averages (with equal weights for all historical data) or moving averages (with equal weights within a fixed window) lag significantly in responding to slow changes. For some changing factors that require manual evaluation to influence the weights, this approach is highly dependent on worker experience and exhibits a lag.

[0043] Please refer to Figure 1In the time series diagram of the critical dimension of film etching, the horizontal axis represents the wafer batch number (Lot-ID), and the vertical axis represents the critical dimension (CD) corresponding to the Lot with different wafer batch numbers. The critical dimension value of the Lot shown in the rectangle HT is 35nm higher than the expected value. For the Feedback prediction value of the next Lot adjacent to the rectangle HT, if a simple arithmetic average (all historical data have the same weight) or a moving average (all data have the same weight within a fixed window) is used, it will take a long time to obtain an accurate Feedback prediction value, resulting in serious lag.

[0044] Therefore, the embodiments of this disclosure aim to provide a semiconductor equipment control method, apparatus, medium, and product, which can at least dynamically track and predict the process parameters of the equipment in real time according to the dynamic changes of the actual operating conditions of the equipment, realize real-time online optimization of process parameters, improve the yield and accuracy of semiconductor equipment processes, and effectively avoid dependence on operator experience.

[0045] Please refer to Figure 2 The semiconductor machine tool control method provided in this embodiment can be applied to a semiconductor machine tool control processor 104. The semiconductor machine tool control processor 104 communicates with a server 102 through a network. The server 102 is connected to a server receiving end. The semiconductor machine tool control processor can also be directly connected to the server receiving end. The communication connection can be wired or wireless.

[0046] For example, a semiconductor equipment control method, device, medium, and product are applied to a semiconductor equipment control processor 104. The semiconductor equipment control processor 104 can obtain test timing data and a target weight prediction model from the receiving end of the server 102. The test timing data includes process parameter time series data, equipment status parameter time series data, and process measurement result time series data. The weight prediction error of the target weight prediction model is not greater than a preset target value; for example, the weight prediction error of the target weight prediction model is less than or equal to 5%. The semiconductor equipment control processor 104 inputs the test timing data into the target weight prediction model to predict the target weight. Based on the predicted target weight, the actual feedback data of the current batch of wafers, and the historical feedback data, it obtains the process parameter prediction data of the current batch of wafers. Then, based on the process parameter prediction data, it controls the semiconductor equipment to execute the target process on the current batch of wafers.

[0047] For example, in a semiconductor equipment control method applied to server 102, server 102 obtains test timing data and a target weight prediction model from a receiving end. The test timing data includes time series data of process parameters, time series data of equipment status parameters, and time series data of process measurement results. The weight prediction error of the target weight prediction model is no greater than a preset target value; for example, the weight prediction error of the target weight prediction model is less than or equal to 5%. Server 102 inputs the test timing data into the target weight prediction model to predict the target weight. Based on the predicted target weight, the actual feedback data of the current batch of wafers, and the historical feedback data, it obtains the process parameter prediction data of the current batch of wafers. Then, based on the process parameter prediction data, it controls the semiconductor equipment to execute the target process on the current batch of wafers.

[0048] Please refer to Figure 3 In some embodiments, a semiconductor machine control method is provided, comprising the following steps:

[0049] Step S11: Train the initial weight prediction model based on the training time series data until the weight prediction error is no greater than the preset target value, and obtain the target weight prediction model. The training time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data.

[0050] Step S12: Input the test time series data into the target weight prediction model to predict the target weight; the test time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data;

[0051] Step S13: Based on the predicted target weight, the actual feedback data of the current batch of wafers and the historical feedback data, obtain the predicted process parameters of the current batch of wafers;

[0052] Step S14: Control the semiconductor equipment to perform the target process on the current batch of wafers based on the process parameter prediction data.

[0053] The above-described embodiments can produce at least the following unexpected technical effects:

[0054] Based on training time-series data including process parameter time-series data and machine status parameter time-series data, an initial weight prediction model is trained until the weight prediction error is no greater than a preset target value, for example, less than or equal to 3%, resulting in a target weight prediction model with expected weight prediction accuracy. Since the target weight prediction model has already fitted the specific correlation between the process parameter time-series data, the machine status parameter time-series data, and the predicted target weights, inputting the test time-series data into the target weight prediction model allows for real-time and accurate online prediction of the target weights, avoiding the lag and reliance on experience inherent in manual prediction. Because the test time-series data includes process parameter time-series data and machine status parameter time-series data, encompassing the dynamic changes in both process parameters and machine status parameters over time, the predicted target weights already reflect the influence of random dynamic changes in the real-time operating state of the machine. Therefore, based on the predicted target weight, the actual feedback data of the current batch of wafers, and the historical feedback data, the process parameter prediction data of the current batch of wafers can be accurately predicted. This allows for precise control of the semiconductor equipment to execute the target process on the current batch of wafers based on the predicted process parameter data, achieving real-time online optimization of process parameters. This improves the yield and accuracy of the semiconductor equipment process while effectively avoiding reliance on operator experience.

[0055] In some embodiments, the equipment includes an etching equipment; the initial weight prediction model includes a time-recurrent neural network; the process parameters include RF power, gas flow rate, chamber pressure, etching time, and electrostatic chuck temperature; the equipment status parameters include plasma impedance change rate and RF time; the plasma impedance change rate is used at least to characterize the thickness of the deposits accumulated on the chamber wall; the RF time is used at least to characterize the interval between two adjacent maintenance start times; training the initial weight prediction model includes: adjusting at least one of the learning rate and the number of hidden layer units of the time-recurrent neural network until the weight prediction error is not greater than a preset target value.

[0056] The principle behind prediction using Recurrent Neural Networks (RNNs) involves utilizing the network's internal "memory" (hidden states) to process temporal dependencies in sequential data, thereby predicting future values. Unlike ordinary feedforward neural networks, RNNs introduce recurrent connections. This means that when processing each element (time step) in the sequence, the network receives not only the current input but also "memory" information from the previous time step. By incorporating a Long Short-Term Memory (LSTM) network into the RNN, input gates, forget gates, output gates, and cell states are introduced. The cell state acts as a "highway," allowing information to be transmitted relatively losslessly over extended periods. The gating structure (especially the forget gate) precisely controls which information needs to be retained, which needs to be forgotten, which needs to be updated to the cell state, and which needs to be output to the hidden state. This enables the LSM network to effectively capture long-term dependencies.

[0057] In some embodiments, the time recurrent neural network includes an input layer, multiple network layers, and an output layer. The input layer includes time series data of process parameters, time series data of machine status parameters, and time series data of process measurement results at multiple time steps. Each network layer in the multiple network layers includes multiple units. The output layer includes a sigmoid activation function to generate dynamic prediction weights.

[0058] Please refer to Figure 4 Forget the door decides to start from cell state c i-1 Which information is discarded? The input gate determines which new candidate information is included. i Update to the cell state. The output gate is based on the current cell state c. i-1 Determine which information to output a i-1 to hidden state a i .

[0059] Please continue to refer to this. Figure 4 The activation function includes the Sigmoid activation function to generate dynamic prediction weights. Sequence data with time steps is used as training samples. For example, the input training time series data [x] 1 ,x 2 ,...,x i The corresponding target (true value) can be [x]. 2 ,x 3 ,...,x i+1 Define a loss function (e.g., mean squared error for regression prediction, cross-entropy for classification prediction) to measure the model's output y at each time step. i The difference between the predicted value and the actual preset target value. 'i' can be used to represent the time or data number.

[0060] In some embodiments, the semiconductor equipment control method further includes: obtaining initial process parameter prediction data for the current batch of wafers based on an initial preset target value, actual feedback data of the current batch of wafers, and historical feedback data; controlling the semiconductor equipment to perform a target etching process on the current batch of wafers based on the initial process parameter prediction data to obtain an initial critical dimension; if the initial critical dimension is not greater than a critical dimension threshold, for example, the error between the obtained initial critical dimension and the critical dimension threshold is greater than or equal to 0 and less than or equal to 5%, then determining a preset target value based on the initial critical dimension, for example, the initial critical dimension can be set as the preset target value.

[0061] Please refer to Figure 5 In some embodiments, after obtaining historical data of machine tool process parameters, abnormal data can be filtered out to obtain preparation data. The first 90% of the preparation data, distributed chronologically, is used as training data, and the last 10% is used as test data. Training time-series data is obtained based on the training data. An initial weight prediction model is trained according to the training time-series data until the weight prediction error is no greater than a preset target value, thus obtaining the target weight prediction model. The training time-series data includes time-series data of process parameters, time-series data of machine tool status parameters, and time-series data of process measurement results. The test time-series data is input into the target weight prediction model to predict the target weight, obtaining the dynamic weight coefficient α(t), and generating the time series of the target weight. The dynamic weight coefficient α(t) is applied to the APC EWMA algorithm to predict the feedback value of each batch of Lots. Training time-series data is generated based on the time series of the target weight to iteratively optimize the Long Short-Term Memory (LSTM) network model, further improving the accuracy and efficiency of LSTM model training.

[0062] In some embodiments, the time series of the target weight includes a first stage, a second stage, and a third stage distributed in chronological order; the minimum value of the first stage is greater than the maximum value of the second stage; and the maximum value of the second stage is less than the minimum value of the third stage.

[0063] In some embodiments, the first stage shows a downward trend; the third stage shows an upward trend.

[0064] Please refer to Figures 5-6 In some embodiments, dynamic weight coefficients α(t) are obtained, where t represents time, and the time series of the target weight is generated, as shown in the example. Figure 6As shown, the horizontal axis represents the wafer batch number arranged in chronological order, and the vertical axis represents the dynamic weighting coefficient α. In the early stage of the target process (early PM) of the semiconductor etching machine, the time series graph of the target weight shows a downward trend; in the middle stage of the target process (mid PM), the target weight value changes little over time, and the maximum value in the mid PM stage is less than the minimum value in the early PM stage; in the late stage of the target process (late PM), the time series graph of the target weight shows an upward trend, and the minimum value in the late PM stage is greater than the maximum value in the mid PM stage.

[0065] Because LSTM models can accurately identify the specific correlation between time-series data of machine process parameters and machine status parameters, and dynamic weighting coefficients, during normal production, the LSTM model automatically reduces the α value (e.g., α=0.6) based on the stability of the equipment status, retaining 40% of the historical data weight, which effectively suppresses random noise. When gas flow fluctuations (sudden increase in variance) are detected, the LSTM model increases α to 0.7, prioritizing the response to current data. During abnormal response phases, such as the first wafer after the machine executes the target process, the LSTM model forces α=0.9 (ignoring historical data) to avoid interference from old chamber states. If the RF power changes abnormally, an emergency reset of the LSTM model can be triggered, recalculating the historical mean using a rolling window.

[0066] The workflow of EWMA prediction in APC feedback control includes: processing Lot-N using the current process parameter recipe; after Lot-N is processed, measurement is performed to obtain the actual Feedback value _N (e.g., critical dimension CD_N); after the system obtains the latest Feedback value _N, it updates the predicted Feedback value for the next Lot-(N+1) using the EWMA formula, for example, predicted value _(N+1) = α * Feedback value _N + (1-α) * predicted value _N. The predicted value _(N+1) is compared with the preset target value to calculate the error. Based on this error, the process parameter offset that needs to be adjusted for Lot-(N+1) can be calculated. The goal is to make the actual Feedback value of Lot-(N+1) close to the preset target value. The adjusted process parameter recipe is applied to the machine to process Lot-(N+1). After Lot-(N+1) is processed, the above steps are repeated to process Lot-(N+2). N is the number or code of the wafer batch Lot.

[0067] In the current APC system, the EWMA algorithm is used to predict the Feedback value of each Lot. Through exponential decay weighting, it effectively captures the slow drift trend of process equipment status while smoothing out measurement noise. The generated predicted value represents the current process level estimated based on recent historical data, providing crucial input to the APC feedback controller. This enables the controller to calculate appropriate process parameter adjustments, driving the output of the next Lot towards the preset target value.

[0068] For example, process parameters include RF power, gas flow rate, chamber pressure, etching time, and electrostatic chuck temperature. Based on time-series data of RF power, gas flow rate, chamber pressure, etching time, and electrostatic chuck temperature, an initial time recurrent neural network is trained until the weight prediction error is no greater than a preset target value, thus obtaining the target EWMA algorithm prediction model. In the current APC system, the target EWMA algorithm prediction model is used to predict the target Feedback value of the current batch of Lots. Based on this target Feedback value, the actual feedback data of the current batch of wafers, and historical feedback data, the process parameter prediction data of the current batch of wafers is obtained. The process parameter prediction data includes at least one process parameter among the gas flow rate, etching time, and electrostatic chuck temperature of the etching equipment. Based on the aforementioned process parameter prediction data, the etching equipment is controlled to perform the target etching process on the current batch of wafers to obtain the semiconductor structure of the target critical dimension.

[0069] Please refer to Figure 7 In this embodiment of the disclosure, based on training time series data including process parameter time series data and machine status parameter time series data, an initial weight prediction model is trained until the weight prediction error is no greater than a preset target value, for example, less than or equal to 3%, to obtain a target weight prediction model whose weight prediction accuracy meets the expected value; a dynamic target weight is generated based on the target weight prediction model, and then the dynamic target weight is input into the EWMA algorithm to obtain the process parameter prediction data of the current batch of wafers. Figure 7 In the diagram, the horizontal axis represents the wafer batch number arranged in chronological order, and the vertical axis represents the critical dimension (CD) data in nm. The red curve represents the actual value; the green curve represents the preset target value; the blue curve represents the curve obtained by generating dynamic target weights using the target weight prediction model in this embodiment and combining it with the EWMA algorithm; and the yellow curve represents the curve obtained by inputting a fixed weight α=0.6 into the EWMA algorithm.

[0070] Table 1

[0071]

[0072] Figure 7The yellow and blue curves represent the errors at three different PM stages, as shown in Table 1. Table 1 reveals that the target weight prediction model used in this embodiment generates dynamic target weights, which, combined with the EWMA algorithm, reduces the prediction error by 73.2% in the early PM stage compared to the fixed-weight EWMA algorithm; by 9.0% in the middle PM stage; and by 33.3% in the late PM stage.

[0073] It should be understood that although Figure 3 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 3 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0074] like Figure 8 As shown, in some embodiments, a semiconductor equipment control device 100 is provided, including a model training module 10, a weight prediction module 20, a process parameter prediction module 30, and a control module 40. The model training module 10 is used to train an initial weight prediction model based on training time-series data until the weight prediction error is no greater than a preset target value, thereby obtaining a target weight prediction model. The training time-series data includes process parameter time-series data, equipment status parameter time-series data, and process measurement result time-series data. The weight prediction module 20 is used to input test time-series data into the target weight prediction model to predict the target weight. The test time-series data includes process parameter time-series data, equipment status parameter time-series data, and process measurement result time-series data. The process parameter prediction module 30 is used to obtain process parameter prediction data for the current batch of wafers based on the target weight, the actual feedback data of the current batch of wafers, and historical feedback data. The control module 40 is used to control the semiconductor equipment to execute the target process on the current batch of wafers based on the process parameter prediction data.

[0075] For example, please continue to refer to Figure 8The model training module 10 trains an initial weight prediction model based on training time-series data including process parameter time-series data and machine status parameter time-series data until the weight prediction error is no greater than a preset target value, thus obtaining a target weight prediction model with expected weight prediction accuracy. Since the target weight prediction model has already fitted the specific correlation between the process parameter time-series data, machine status parameter time-series data, and the predicted target weights, the weight prediction module 20 can accurately predict the target weights in real-time after inputting the test time-series data into the target weight prediction model, avoiding the lag and reliance on experience in manual prediction. Because the test time-series data includes process parameter time-series data and machine status parameter time-series data, it contains dynamic changes in both process parameters and machine status parameters over time; therefore, the predicted target weights already reflect the influence of random dynamic changes in the real-time operating state of the machine. Therefore, the process parameter prediction module 30 can accurately predict the process parameter prediction data of the current batch of wafers based on the predicted target weight, the actual feedback data of the current batch of wafers and the historical feedback data. This enables the control module 40 to accurately control the semiconductor equipment to perform the target process on the current batch of wafers based on the process parameter prediction data, thereby achieving real-time online optimization of process parameters, improving the process yield and accuracy of the semiconductor equipment, and effectively avoiding dependence on operator experience.

[0076] In some embodiments, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of any semiconductor machine control method in the embodiments of this disclosure.

[0077] In some embodiments, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of any semiconductor machine control method in the embodiments of this disclosure.

[0078] In some embodiments, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of any of the semiconductor machine control methods described above.

[0079] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 9As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a photolithography pattern calibration method. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0080] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0081] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this disclosure may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this disclosure may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these. The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features are not contradictory, they should be considered within the scope of this specification.

[0083] Please note that the above embodiments are for illustrative purposes only and are not intended to limit this disclosure.

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0085] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0086] The embodiments described above are merely illustrative of several implementation methods of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this disclosure, and these modifications and improvements all fall within the scope of protection of this disclosure.

Claims

1. A semiconductor machine tool control method, characterized in that, include: The initial weight prediction model is trained based on the training time series data until the weight prediction error is no greater than the preset target value, and the target weight prediction model is obtained. The training time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data. The test time series data is input into the target weight prediction model to predict the target weight; The test time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data. Based on the target weight, the actual feedback data of the current batch of wafers and the historical feedback data, the predicted process parameters of the current batch of wafers are obtained; Based on the predicted process parameters, the semiconductor equipment is controlled to perform the target process on the current batch of wafers.

2. The semiconductor machine control method according to claim 1, characterized in that, The equipment includes an etching machine; The initial weight prediction model includes a time-recurrent neural network; the process parameters include RF power, gas flow rate, chamber pressure, etching time, and electrostatic chuck temperature. The machine status parameters include plasma impedance change rate and radio frequency time; the plasma impedance change rate is used at least to characterize the thickness of the deposits accumulated on the chamber walls. The radio frequency time is used at least to characterize the interval between two adjacent maintenance start times; The training of the initial weight prediction model includes: By adjusting at least one of the learning rate and the number of hidden layer units of the time recurrent neural network, the weight prediction error is made no greater than the preset target value.

3. The semiconductor machine control method according to claim 2, characterized in that, Also includes: Based on the initial preset target value, the actual feedback data of the current batch of wafers and the historical feedback data, the initial process parameter prediction data of the current batch of wafers is obtained; Based on the initial process parameter prediction data, the semiconductor equipment is controlled to perform the target etching process on the current batch of wafers to obtain the initial critical dimensions; If the initial critical dimension is not greater than the critical dimension threshold, then the preset target value is determined based on the initial critical dimension.

4. The semiconductor machine control method according to claim 2, characterized in that, The time-recurrent neural network includes: The input layer includes time-series data of process parameters, time-series data of machine status parameters, and time-series data of process measurement results at multiple time steps. A multi-layer network, with each layer consisting of multiple units; The output layer includes a sigmoid activation function to generate dynamic prediction weights.

5. The semiconductor machine control method according to any one of claims 1-4, characterized in that, Also includes: Generate the time series of the target weights; The training time series data is generated based on the time series of the target weights.

6. The semiconductor machine control method according to claim 5, characterized in that, The time series of the target weights includes the first stage, the second stage, and the third stage, which are distributed in chronological order. The minimum value in the first stage is greater than the maximum value in the second stage; The maximum value in the second stage is less than the minimum value in the third stage.

7. The semiconductor machine control method according to claim 6, characterized in that, The first stage shows a downward trend; The third stage shows an upward trend.

8. A semiconductor machine tool control device, characterized in that, include: The model training module is used to train an initial weight prediction model based on training time series data until the weight prediction error is no greater than a preset target value, thereby obtaining a target weight prediction model. The training time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data. The weight prediction module is used to input test time series data into the target weight prediction model to predict the target weight; The test time series data includes process parameter time series data, machine status parameter time series data, and process measurement result time series data. The process parameter prediction module is used to obtain the process parameter prediction data for the current batch of wafers based on the target weight, the actual feedback data of the current batch of wafers, and the historical feedback data. The control module is used to control the semiconductor equipment to perform the target process on the current batch of wafers based on the process parameter prediction data.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1-7.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-7.

Images