Process parameter determination method and apparatus, and semiconductor process device

Abstract

Embodiments of the present application provide a process parameter determination method and apparatus, and a semiconductor process device. The process parameter determination method comprises: in response to target process information inputted by a user, inputting the target process information into a target network model, the target process information comprising a process result achievable by a process recipe to be determined; predicting, by means of the target network model, process parameters required to achieve the process result, the target network model comprising a capsule network layer constructed on the basis of a capsule network; and determining a prediction result outputted by the target network model as process parameters of the process recipe to be determined. The present application enables quick determination of process parameters that meet process requirements, thereby saving a significant amount of time. In addition, the target network model comprises a capsule network layer constructed on the basis of a capsule network, so that the target network model integrates the characteristics of the capsule network, thereby improving the accuracy of the process parameters.

Description

A method, apparatus, and semiconductor process equipment for determining process parameters. Technical Field

[0001] The embodiments described in this application relate to the field of manufacturing technology, specifically to process formulation processing technology in the field of manufacturing technology, and more specifically, to a method, apparatus and semiconductor process equipment for determining process parameters. Background Technology

[0002] A process formulation can be simply understood as a combination of various process parameters under the operating conditions of process equipment, and its accuracy directly affects the process outcome. In the process of process generation, the process requirement, i.e., the desired process result, usually emerges first. Then, based on this requirement, the process parameters are continuously adjusted to ensure that the adjusted parameters meet the process requirements.

[0003] However, the above process is not something that can be accomplished overnight. It requires continuous adjustment of process parameters so that the process results obtained after adjustment are closer to the desired process results, and finally process parameters that meet the process requirements are obtained. The whole process takes a long time. Summary of the Invention

[0004] This application provides a method, apparatus, and semiconductor process equipment for determining process parameters, in order to quickly determine process parameters that can meet process requirements.

[0005] In a first aspect, one embodiment of this application provides a method for determining process parameters, the method comprising:

[0006] In response to user-inputted target process information, the target process information is input into the target network model; wherein, the target process information includes the process results that the process formulation to be determined can achieve;

[0007] The target network model predicts the process parameters required to achieve the process result; wherein, the target network model includes: a capsule network layer constructed based on capsule networks;

[0008] The prediction results output by the target network model are determined as the process parameters of the process formula to be determined.

[0009] In some implementations, the capsule network layer includes: an input layer, a convolutional layer, a master capsule layer, a digital capsule layer, and an output layer;

[0010] The target network model predicts the process parameters required to achieve the process result, including:

[0011] The target process information is preprocessed through the input layer to generate input features;

[0012] The input features are extracted through the convolutional layer to generate an initial feature vector;

[0013] The initial feature vector is extracted through the main capsule layer to generate an initial capsule vector;

[0014] The initial capsule vector is processed by the digital capsule layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0015] Based on the resulting capsule vector, the output layer outputs the predicted value with the highest prediction probability for each process parameter.

[0016] In some implementations, feature extraction is performed on the initial feature vector through the main capsule layer to generate an initial capsule vector, including:

[0017] The initial capsule vector is generated by extracting features from the initial feature vector through dynamic routing in the main capsule layer.

[0018] In some implementations, before performing feature processing on the initial capsule vector through the digital capsule layer to generate the target number of result capsule vectors, the method further includes:

[0019] The execution chamber is determined based on the target process information using the target network model, wherein the execution chamber is a chamber in the process equipment that executes the process formula to be determined;

[0020] Based on the correspondence between the number of chambers and the number of process parameters in the process equipment, the number of process parameters corresponding to the execution chamber is determined as the target number using the target network model.

[0021] In some implementations, the target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, wherein the input of the long short-term memory network layer is the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0022] In some implementations, the process parameters required to achieve the process result are predicted by the target network model, including:

[0023] The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0024] The result capsule vector is processed by the forget gate, input gate, output gate, and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

[0025] In some implementations, the target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, and an attention module constructed based on an attention mechanism; wherein, the input of the attention module is the target process information, the input of the long short-term memory network layer is the fusion result of the output of the attention module and the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0026] In some implementations, the process parameters required to achieve the process result are predicted by the target network model, including:

[0027] The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0028] The attention module processes the target process information to generate a context vector corresponding to each result capsule vector.

[0029] The fusion result of the result capsule vector and the context vector is processed by the forget gate, input gate, output gate and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

[0030] In some implementations, before inputting target process information into the target network model in response to user input, the method further includes:

[0031] Acquire training data; wherein, the training data includes: historical process formulas and the process results that the historical process formulas can achieve;

[0032] The training data is used to train the network model constructed by the capsule network, and the trained network model is determined as the target network model.

[0033] In some implementations, during the training of the network model constructed using the training data, the sum of the interval loss and the reconstruction loss is used as the model loss.

[0034] Secondly, one embodiment of this application provides a process parameter determination apparatus, the process parameter determination apparatus comprising:

[0035] A response module is used to respond to user-inputted target process information and input the target process information into the target network model; wherein, the target process information includes the process results that the process formulation to be determined can achieve;

[0036] The model processing module is used to predict the process parameters required to achieve the process result through the target network model; wherein, the target network model includes: a capsule network layer constructed based on capsule network;

[0037] The determination module is used to determine the prediction results output by the target network model as the process parameters of the process formula to be determined.

[0038] In some implementations, the capsule network layer includes: an input layer, a convolutional layer, a master capsule layer, a digital capsule layer, and an output layer;

[0039] The model processing module includes:

[0040] The preprocessing unit is used to preprocess the target process information through the input layer to generate input features;

[0041] The first feature extraction unit is used to extract features from the input features through the convolutional layer to generate an initial feature vector;

[0042] The second feature extraction unit is used to extract features from the initial feature vector through the main capsule layer to generate an initial capsule vector;

[0043] The model processing unit is used to perform feature processing on the initial capsule vector through the digital capsule layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0044] The model output unit is used to output the predicted value with the highest prediction probability for each process parameter based on the result capsule vector through the output layer.

[0045] In some implementations, the second feature extraction unit is specifically used to extract features from the initial feature vector through the main capsule layer in a dynamic routing manner to generate an initial capsule vector.

[0046] In some embodiments, the apparatus further includes:

[0047] The chamber determination module is used to determine the execution chamber based on the target process information through the target network model, wherein the execution chamber is a chamber in the process equipment that executes the process formula to be determined;

[0048] The quantity determination module is used to determine the number of process parameters corresponding to the execution chamber as the target quantity based on the correspondence between the number of chambers and the number of process parameters in the process equipment through the target network model.

[0049] In some implementations, the target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, wherein the input of the long short-term memory network layer is the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0050] In some implementations, the model processing module is specifically used for:

[0051] The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0052] The result capsule vector is processed by the forget gate, input gate, output gate, and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

[0053] In some implementations, the target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, and an attention module constructed based on an attention mechanism; wherein, the input of the attention module is the target process information, the input of the long short-term memory network layer is the fusion result of the output of the attention module and the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0054] In some implementations, the model processing module is specifically used for:

[0055] The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0056] The attention module processes the target process information to generate a context vector corresponding to each result capsule vector.

[0057] The fusion result of the result capsule vector and the context vector is processed by the forget gate, input gate, output gate and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

[0058] In some embodiments, the device further includes:

[0059] The model training module is used for:

[0060] Acquire training data; wherein, the training data includes: historical process formulas and the process results that the historical process formulas can achieve;

[0061] The training data is used to train the network model constructed by the capsule network, and the trained network model is determined as the target network model.

[0062] In some implementations, during the training of the network model constructed using the training data, the sum of the interval loss and the reconstruction loss is used as the model loss.

[0063] Thirdly, one embodiment of this application provides a semiconductor process apparatus, including: a host computer and a slave computer connected to the host computer;

[0064] The host computer is used to execute the process parameter determination method as described in any of the first aspects, to determine the process parameters of the process formula to be determined, and to send the process parameters of the process formula to be determined to the slave computer.

[0065] The lower-level machine is used to receive the process parameters of the process formula to be determined, and to carry out the process production according to the process parameters of the process formula to be determined.

[0066] Fourthly, one embodiment of this application provides an electronic device including a processor and a memory;

[0067] The memory is connected to the processor and is used to store computer programs;

[0068] The processor is configured to implement the method for determining process parameters as described in any of the first aspects by running a computer program stored in the memory.

[0069] Fifthly, one embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for determining process parameters as described in any of the first aspects.

[0070] In a sixth aspect, one embodiment of this application provides a computer program product or computer program, the computer program product including a computer program stored in a computer-readable storage medium; a processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program to implement the steps of the method for determining process parameters as described in any of the first aspects.

[0071] The multiple implementations provided in this application, for the process parameters of a process formulation to be determined, first determine the achievable or expected process result, i.e., the process requirement. Then, using a target network model, predict the process parameters that can achieve this process result, and finally use the predicted process parameters as the process parameters for the process formulation to be determined. Compared to using a self-consistent iterative method to back-calculate and find the optimal process parameters, this application can quickly determine the process parameters that can meet the process requirements, saving a significant amount of time. Simultaneously, the target network model includes a capsule network layer based on capsule networks, enabling the target network model to integrate the characteristics of capsule networks and better consider the correlation between different features, thereby improving the accuracy of the process parameters. Attached Figure Description

[0072] Figure 1 is a flowchart of a conventional method for determining process parameters provided in one embodiment of this application;

[0073] Figure 2 is a flowchart illustrating a method for determining process parameters according to one embodiment of this application;

[0074] Figure 3 is a schematic diagram of the dynamic routing process in a capsule network according to one embodiment of this application;

[0075] Figure 4 is a network structure diagram of a long short-term memory network in one embodiment of this application;

[0076] Figure 5 is a flowchart illustrating a method for determining process parameters provided in another embodiment of this application.

[0077] Figure 6 is a flowchart of the actual application of the method for determining process parameters provided in one embodiment of this application;

[0078] Figure 7 is one of the application architecture diagrams of the method for determining process parameters provided in one embodiment of this application;

[0079] Figure 8 is a second application architecture diagram of the method for determining process parameters provided in one embodiment of this application;

[0080] Figure 9 is a schematic diagram of a process parameter determination device provided in one embodiment of this application;

[0081] Figure 10 is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0082] Unless otherwise defined, the technical or scientific terms used in the embodiments of this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to avoid confusion of the constituent elements.

[0083] Unless the context otherwise requires, throughout this specification, "a plurality of" means "at least two," and "including" is interpreted as open-ended or encompassing, that is, "including, but not limited to." In the description of this specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this application. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example.

[0084] 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, and 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.

[0085] Overview

[0086] As described in the background section, process formulation is crucial in the production process. Determining suitable process parameters to meet process requirements is an unavoidable problem in process generation. As shown in Figure 1, after inputting process requirements, a set of process parameters is usually randomly assigned. The system then determines whether the current process parameters meet the requirements. If not, the process parameters need to be updated based on the deviation from the expected result until the requirements are met, at which point production will proceed according to the latest process parameters.

[0087] Although a neural network model is used to determine process parameters, the principle behind this model is still to gradually find the optimal process parameters through self-consistent iteration. To shorten the time required to determine process parameters, the inventors discovered that the required process parameters can be directly predicted from the desired process outcome. Furthermore, by combining the characteristics of capsule networks, the target network model can ultimately be used to quickly obtain the process parameters that meet the process requirements.

[0088] Based on the above concept, the method for determining process parameters provided in the embodiments of this application will be described exemplarily below.

[0089] Exemplary methods

[0090] As shown in Figure 2, this application embodiment provides a method for determining process parameters, which includes:

[0091] S201: In response to the target process information input by the user, input the target process information into the target network model.

[0092] In this step, the target process information includes the process results achievable by the process formulation to be determined. The process formulation to be determined is the desired process formulation or a process formulation that meets the desired process requirements. The process result can be understood as the process requirement or the result of the process requirement. Specifically, during the process generation process, if it is necessary to determine a process formulation that can achieve a certain process result or meet a certain process requirement, then this process formulation can be considered as the process formulation to be determined, and the process result or process requirement can be considered as the target process information. For example, for semiconductor process equipment, suppose it is necessary to determine the specific parameter values ​​of process temperature, process pressure, and gas flow rate to achieve a deposition rate of 10 nanometers per minute. This deposition rate of 10 nanometers per minute is the target process information, and the specific parameter values ​​of process temperature, process pressure, and gas flow rate are the process formulation to be determined or the process parameters in the process formulation. It is understood that the target process information and the process formulation are related to the process production process involved in the process equipment, which includes, but is not limited to, the process production process of the semiconductor process equipment.

[0093] It should be noted that the method provided in this application can be directly applied to the process equipment to determine the process parameters. After determining the process parameters, production can be carried out directly according to these parameters. In this case, the user only needs to input the target process information into the process equipment. Alternatively, other electronic devices with higher computing power can be used to determine the process parameters using the method provided in this embodiment. The process parameters are then input into the process equipment so that the equipment performs production according to the input parameters.

[0094] S202: Predict the process parameters required to achieve the desired process results using the target network model.

[0095] S203: The prediction results output by the target network model are used as the process parameters for the process formulation to be determined.

[0096] It should be noted that the target network model is a pre-trained network model. Its input is the process requirement or process result, and its output is the predicted process parameters that can meet the process requirement or achieve the process result, i.e., the prediction result. Therefore, by using the target network model, the process parameters corresponding to the target process information can be obtained, that is, the process parameters that can achieve the process result in the target process information can be obtained. After determining the process parameters, the process formula to be determined can be determined. Subsequently, the process equipment can use the determined process formula to be determined for process production.

[0097] The target network model includes capsule network layers built upon capsule networks. Capsule networks, compared to convolutional neural networks (CNNs), offer significant advantages in several aspects. For instance, capsule networks extract richer object features through hierarchical feature extraction. In contrast, CNNs primarily extract features through tiling and pooling operations, potentially losing some detailed information. Capsule networks represent objects using vectors; this distributed representation better captures the internal structure and relationships of objects. CNNs generally use scalars to represent features, making it difficult to represent the complex structures of objects.

[0098] In this embodiment, for the process parameters of the process formulation to be determined, the achievable or expected process result, i.e., the process requirement, is first determined. Then, the process parameters that can achieve this process result are predicted using a target network model, and these predicted process parameters are ultimately used as the process parameters for the process formulation to be determined. Compared to using a self-consistent iterative method to back-calculate and find the optimal process parameters, this application can quickly determine the process parameters that meet the process requirements, saving a significant amount of time. Simultaneously, the target network model includes a capsule network layer based on capsule networks, integrating the characteristics of capsule networks into the target network model. This allows for better consideration of the correlation between different features, thereby improving the accuracy of the process parameters.

[0099] In some implementations, the capsule network layer includes: an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, and an output layer;

[0100] The target network model predicts the process parameters required to achieve the desired process results, including:

[0101] The target process information is preprocessed through the input layer to generate input features;

[0102] The input features are extracted using convolutional layers to generate an initial feature vector;

[0103] The initial capsule vector is generated by extracting features from the initial feature vector through the main capsule layer;

[0104] The initial capsule vector is processed by a digital capsule layer to generate the target number of result capsule vectors. Each result capsule vector corresponds to a process parameter and includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0105] The output layer outputs the predicted value with the highest prediction probability for each process parameter based on the resulting capsule vector.

[0106] It should be noted that the input of the input layer is the input of the target network model. It is used to preprocess the input data of the target network model, thereby processing it into features that can be directly recognized and processed. The input of the convolutional layer is the output of the input layer. It is used to perform convolution processing on the output of the input layer to generate an initial feature vector. The convolution processing will not be described in detail here. In some embodiments, the convolutional layer in this embodiment can adopt a three-layer convolutional structure to avoid the network structure being too complex and to be suitable for parameter tuning on large devices. The input of the main capsule layer is the output of the convolutional layer. It is used to process the output of the convolutional layer to generate an initial capsule vector. The main capsule layer is the PrimaryCaps layer in the capsule network, which can perform the data processing of the PrimaryCaps layer, and will not be described in detail here. The input of the digital capsule layer is the output of the main capsule layer. It is used to process the output of the main capsule layer to generate a result capsule vector. The digital capsule layer is the DigitCaps layer in the capsule network, which can perform the data processing of the DigitCaps layer, and will not be described in detail here. The output layer takes the output of the digital capsule layer as its input and processes that output to obtain the final output of the target network model. The output of the output layer is the output of the target network model.

[0107] Specifically, the input layer preprocesses the target process information, including standardizing the length of the target process information and performing IDX transformation. Standardizing the length involves adjusting the input data to a uniform length, which can be achieved by padding with null values ​​or truncation. IDX is a binary file format typically used to store multidimensional array data. It is converted into a tensor or matrix format suitable for the network model for input.

[0108] The processing steps of the convolutional layer include: processing the input features using a formula.

[0109] Formula 1: Conv i=σ(W conv ·x i +b conv );

[0110] Among them, Conv i W represents the initial eigenvector. conv Denotes the convolution kernel, b conv Let x represent the bias term, and σ be a non-linear activation function, such as the ReLU function, but not limited to it. i This represents the input features.

[0111] It is understandable that the larger the convolution kernel, the more information is contained in the initial feature vector.

[0112] The processing of the main capsule layer includes: processing the initial feature vector using Formula 2.

[0113] Formula 2: Cap conv =ρ(W cap ·Conv i +b cap );

[0114] Among them, Cap conv Let W represent the initial capsule vector. cap Represents the convolution kernel of the main capsule layer, Conv i Let b represent the initial eigenvector. cap ρ represents the bias term, and ρ is the squash function in the capsule network.

[0115] The main capsule layer can transform the multidimensional data output by the convolutional layer into several capsules or capsule vectors. Specifically, it performs in-depth feature extraction on the target process information.

[0116] The processing of the digital capsule layer includes: processing the initial capsule vector using Formula 3.

[0117] Formula 3: Cap full =φ(W full ·Cap conv +b full );

[0118] Among them, Cap full Cap represents the resulting capsule vector. conv Let W represent the initial capsule vector. full and b full These are the weights and biases, respectively. φ represents the transformation matrix.

[0119] Understandably, each capsule is associated with the previous capsule network through a transformation matrix. The transformation matrix is ​​used to predict the input. Through the calculation of the digital capsule layer, multiple capsule vectors containing feature information can be obtained, namely the result capsule vectors. The number of capsule vectors is equal to the number of process parameters in the process formulation to be determined.

[0120] In this embodiment, the capsule network layer is composed of an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, and an output layer. This allows for the effective integration of the characteristics of the capsule network into the target network model while avoiding an overly complex network structure.

[0121] In some implementations, initial capsule vectors are generated by feature extraction from the initial feature vector through the main capsule layer, including:

[0122] The initial capsule vector is generated by extracting features from the initial feature vector through dynamic routing in the main capsule layer.

[0123] It should be noted that in a capsule network, the main capsule layer can be considered as consisting of multiple capsules, with each capsule in each layer connected to each capsule in the layer below. Furthermore, each capsule performs the same dynamic routing process on the output of the capsule in the previous layer, and finally, the output of the last capsule layer is used as the initial capsule vector. During the process of lower-layer capsules passing the input vector to higher-layer capsules, the corresponding parameters are continuously adjusted to enable the higher-layer capsules to output more suitable data; this process can be called dynamic routing. Since the dynamic routing process is the same for each capsule, only one capsule is used as an example here. As shown in Figure 3, suppose the previous capsule layer includes two capsules, and the outputs of the two capsules are V1 and V2, with weights W1 and W2 respectively. Then, the coupling coefficient used by this capsule in the first processing is... and The result obtained is S1. After processing with the squeezing function, the coupling coefficient used in the update process is obtained. and Then, a second processing step is performed on the capsule, using the updated coupling coefficient. and The new result S2 is obtained. This is then processed using a squeezing function to update the coupling coefficients used in the processing. and Then, a third processing step is performed on the capsule, using the updated coupling coefficient. and A new result S3 is obtained. Given that S3 is suitable data, the true output V of the capsule is obtained through a squeezing function. It can be understood that when the capsule length represents the probability of process parameters, the capsule's modulus can represent the salience of the feature; the longer the modulus, the more salience of the feature. Furthermore, when the modulus is close to 0, it also has an amplification effect, eliminating the need for global compression using a squeezing function. The squeezing function is shown below:

[0124] Among them, the squash function squash(v j It mainly consists of two parts. The first part changes the input vector s through a squeezing function. j Length, without changing the input vector s j The direction of the first part is determined by a vector of length 1. The second part controls the size of the first part using a vector of length 1. This preserves the input vector s. j The direction can be controlled, and the size can also be controlled. For the input vector s j It can be determined by the following formula:

[0125] Among them, u i W represents the output of the lower capsule. ij c represents the weight parameter. ij denoted as the coupling coefficient, where there is a coupling coefficient between the upper-level capsule and the lower-level capsule, and the sum of the coupling coefficients is 1.

[0126] During the update of the coupling coefficients, an intermediate parameter b can be used. Specifically:

[0127] in,

[0128] Among them, b ij Indicates intermediate parameter, u i This represents the output of the lower-level capsule, v j This indicates the output of the high-level capsule. When and v j When they are in the same direction, it indicates that the lower-level capsules and higher-level capsules express a high degree of similarity in their features. ij The weight of b will increase; conversely, the features of lower-level capsules are highly unlikely to belong to higher-level capsules. ij The weight will decrease. The dynamic routing algorithm establishes the connection between lower-level capsules and higher-level capsules through weight updates.

[0129] Of course, besides dynamic routing, clustering algorithms can also be used to generate initial capsule vectors. It's understandable that when using clustering algorithms, the process from lower-level capsules to higher-level capsules is similar to the process of generating cluster centroids during clustering. Let's assume the features of the lower-level capsules are u1, u2, ... u... n Furthermore, they are divided into k classes. The process of determining the initial capsule vector can be understood as finding the central features v1, v2, ..., v of the k high-level capsules. k This minimizes the intra-class spacing. The intra-class spacing can be calculated based on Formula 4.

[0130] Formula 4:

[0131] Where L represents the minimum intra-class margin, u i This represents the output of the lower-level capsule, v j This represents the output of the high-level capsule. d represents the distance metric between the low-level and high-level capsules. The clustering result depends on the specific form of d.

[0132] The goal is to minimize the sum of the distances between categories of all high-level capsules. Where, v1, v2, ..., v k Let L represent the central features of the k high-level capsules, and L represent the minimum intraclass margin. Then, L can be calculated using an approximation formula. Specifically,

[0133] Approximate value formula 1:

[0134] Approximate value formula 2: min(λ1,λ2,...λ) n )=-max(-λ1,-λ2,...-λ n );

[0135] Wherein, λ1, λ2, ..., λ n It represents n numerical values.

[0136] Based on the above approximation formula, the formula for calculating the minimum intra-class margin can be obtained:

[0137] Through mathematical calculations, we can obtain:

[0138] make

[0139] Indicates v j The result of the r-th iteration, where <,> represents scalar multiplication, c ij Represents the coupling coefficient, u i This represents the output of the lower-level capsule, v iThe output of the higher-level capsules indicates the relationship between the lower-level and higher-level capsules. That is:

[0140] in, Let u represent the coupling coefficient in the r-th iteration. i This indicates the output of the lower capsule. This represents the output of the high-level capsule in the (r+1)th iteration.

[0141] Among them, the transformation matrix W is introduced. ij The dimensions of the centers of lower-level and higher-level capsules can be arbitrarily changed, ensuring the feature representation capability of the higher-level capsules. Correspondingly, the relationship between lower-level and higher-level capsules will change as follows:

[0142] Among them, W ij Represents the transformation matrix. Let u represent the coupling coefficient in the r-th iteration. i This indicates the output of the lower capsule. This represents the output of the high-level capsule in the (r+1)th iteration.

[0143] In this embodiment, by leveraging the characteristics of dynamic routing, the relationship between various parameters can be strengthened, thereby better capturing the internal structure and relationships of the processed object. Compared with other clustering algorithms, the network structure in this embodiment is more lightweight.

[0144] To adapt to different process parameters in different execution chambers, in some implementation methods, before performing feature processing on the initial capsule vector through a digital capsule layer to generate the target number of result capsule vectors, the method further includes:

[0145] The execution chamber is determined based on the target process information using a target network model. The execution chamber is the chamber in the process equipment that executes the process formula to be determined.

[0146] Based on the correspondence between the number of chambers and the number of process parameters in the process equipment, the number of process parameters corresponding to the execution chamber is determined as the target number by using the target network model.

[0147] It should be noted that process equipment includes multiple execution chambers, and different execution chambers can perform different processes according to different process parameters or process formulations. For example, semiconductor process equipment includes MOW chambers and CVDW chambers. The process parameters executed by MOW chambers include process temperature, process pressure, and gas flow rate. The process parameters executed by CVDW chambers include soak time and process temperature.

[0148] The target network model can be applied to multiple execution chambers. Since the process results differ between different execution chambers, these results can be used to distinguish between them, thus determining which process parameters need to be output. Of course, the target network model can also be applied only to a single execution chamber. Regardless of which execution chamber's target process information is input to the target network model, its output will always be the process parameters for that specific execution chamber.

[0149] In this embodiment, the target network model can determine the execution chamber based on the target process information, thereby predicting process parameters for the determined execution chamber, so that the target network model can adapt to the process parameters in different execution chambers.

[0150] To further improve the accuracy of the model output and avoid gradient vanishing or gradient exploding problems, some implementation methods include a target network model that also includes a long short-term memory network layer based on a long short-term memory network. The input of the long short-term memory network layer is the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0151] It should be noted that the Long Short-Term Memory (LSTM) network integrates the gradient advantages of the LSTM network in the target network model, avoiding the problems of gradient vanishing and gradient exploding. In this embodiment, the LSTM layer is located after the capsule network layer, and the input of the LSTM layer is the output of the capsule network layer. The output of the LSTM layer is the output of the target network model.

[0152] In this embodiment, the advantages of LSTM networks in terms of gradients are integrated. By setting a long short-term memory network layer in the target network model, not only can the problems of gradient vanishing and gradient explosion be avoided, but the temporal characteristics of the target process information can also be taken into account, thereby improving the accuracy of the model output.

[0153] In some implementations, the process parameters required to achieve the desired process outcome are predicted using a target network model, including:

[0154] The target process information is processed by the capsule network layer to generate the target number of result capsule vectors; each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0155] The resulting capsule vector is processed by the forget gate, input gate, output gate, and output layer of the long short-term memory network to obtain the process parameters required to achieve the process results.

[0156] It should be noted that the network architecture of the Long Short-Term Memory (LSTM) network layer is shown in Figure 4. The forgetting gate is f. t The input gate is i t The output gate is 0 t Input gate i t The input consists of the historical state h t-1 and the current node state x t The decision is made. σ is the sigmoid function, which controls the output between 0 and 1. 1 indicates that historical information is retained, and 0 indicates that historical information is forgotten. The input gate can be calculated using the following formula: i t =σ(W ix ·x t +W ih ·h t-1 +b i );

[0157] Among them, i t W represents the output of the input gate. ix W represents the weight matrix of the current state at the input gate. ih The weight matrix b represents the weights of the historical states at the input gate. i This indicates the weight bias applied to the input gate.

[0158] Similarly, the forget gate and the output gate can be calculated using the following formula: f t =σ(W fx ·x t +W fh ·h t-1 +b f );o t =σ(W ox ·x t +W oh ·h t-1 +b o );

[0159] Among them, f t W represents the output of the forget gate. fx W represents the weight matrix of the current state in the forget gate. fh The weight matrix b represents the historical state in the forget gate. f Indicates the weight bias for the forget gate; o t W represents the output result of the output gate. ox W represents the weight matrix of the current state in the output gate. oh The weight matrix b represents the historical state in the output gate. o This indicates the weight bias applied to the output gate.

[0160] Then, based on the previous state g t and current state s tNew information is selected and passed to the next neuron. The specific process is as follows:

[0161] g t =σ(W ox ·x t +W oh ·h t-1 +b o );s t =g t ·i t +s t-1 ·f t h t =tanh(s t )·o t The current state is processed using the function tanh, and then multiplied with the output gate to realize the hidden vector h. t This allows for the connection with the next unit, enabling the extraction of temporal feature information related to preceding and following data packets during transmission. Finally, the softmax function is used to obtain the final prediction output.

[0162] In this embodiment, a long short-term memory network layer is formed by a forget gate, an input gate, an output gate, and an output layer to process the resulting capsule vector. By taking into account the temporal characteristics of the data, more accurate process parameters can be obtained.

[0163] To further improve the accuracy of the model output and avoid gradient vanishing or gradient exploding problems, some implementation methods include the following in the target network model: a long short-term memory network layer based on a long short-term memory network and an attention module based on an attention mechanism. The input of the attention module is the target process information, the input of the long short-term memory network layer is the fusion result of the output of the attention module and the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0164] It should be noted that the introduction of an LSTM network in the target network model is similar to that in the above embodiments, and will not be repeated here. The difference lies in the fact that this application also introduces an attention module based on an attention mechanism. The attention mechanism can be understood as a method that mimics the human visual and cognitive system, allowing the neural network model to focus on relevant parts when processing input data. By introducing the attention mechanism, the neural network model can automatically learn and selectively focus on important information in the input, thereby improving the model's performance and generalization ability. After model training, the output of the attention module can indicate the importance of each feature output by the capsule network layer. Furthermore, fusing the output of the attention module with the output of the capsule network layer can increase the proportion of features with higher importance, thereby focusing on important information. The fused result is then input into the LSTM network, and by continuously adjusting the weights between features, a more accurate result can be obtained.

[0165] In this embodiment, the advantages of LSTM networks in terms of gradients are integrated. By setting a long short-term memory network layer in the target network model, not only can the problems of gradient vanishing and gradient explosion be avoided, but the temporal characteristics of the target process information can also be taken into account. Furthermore, an attention mechanism is introduced, which can further improve the accuracy of the model output.

[0166] In some implementations, the process parameters required to achieve the desired process outcome are predicted using a target network model, including:

[0167] The target process information is processed by the capsule network layer to generate the target number of result capsule vectors; each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0168] The attention module processes the target process information to generate a context vector corresponding to each result capsule vector;

[0169] The fusion results of the result capsule vector and context vector are processed by the forget gate, input gate, output gate and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process results.

[0170] It should be noted that the network architecture of the Long Short-Term Memory (LSTM) network layer is shown in Figure 4. The processing of the fusion result by the LSM network layer in the target network model is similar to the processing of the result capsule vector by the LSM network layer in the above embodiments, and will not be repeated here. However, the object processed in this application is different, namely, the fusion result of the result capsule vector and the context vector. The attention mechanism will not be detailed here, but it can be understood that the context vector is the output of the attention module. The value of the context vector corresponding to each result capsule vector is positively correlated with the importance of the result capsule vector. That is, the higher the importance of a certain result capsule vector in the process of predicting process parameters, the larger its corresponding context vector will be. This is the ability learned by the attention module based on the attention mechanism during model training. Specifically, multiple partially overlapping continuous subsequences can be extracted from the original input sequence, i.e., the target process information, as input to the attention module. The context vector output by the attention module can be regarded as a saliency feature, which can identify the importance of each output feature of the digital capsule layer. The higher the importance of the output features of the digital capsule layer, the closer the output of its corresponding attention module is to 1; conversely, the lower the importance of the output features of the digital capsule layer, the closer the output value of its corresponding attention module is to 0. Then, the output features of the digital capsule layer are fused with the salient features output by its corresponding attention module. The fusion method involves element-wise multiplication of the output features of the digital capsule layer with the salient features output by its corresponding attention module. Since the value of the attention module output reflects the importance of the output features of the digital capsule layer, the identification of important features is completed, which is more beneficial for subsequent processing by the LSTM network layers, thus further improving the model's accuracy. Introducing an attention mechanism during model training allows for high-weighting of important information and continuous adjustment of the weights between features, resulting in improved prediction accuracy compared to existing algorithms.

[0171] In this embodiment, a long short-term memory network layer is formed by a forget gate, an input gate, an output gate, and an output layer to process the fusion result. By taking into account the temporal characteristics of the data, more accurate process parameters can be obtained. At the same time, the introduction of an attention mechanism can reduce the interference of useless parameters, solve the problem of information overload, and weight the data with greater influence on the result among many input information, thereby improving the accuracy of the final prediction.

[0172] In some implementations, before inputting the target process information into the target network model in response to user input, the method further includes:

[0173] Acquire training data; the training data includes: historical process formulas and the process results that the historical process formulas can achieve;

[0174] The network model constructed using the training data is trained, and the trained network model is then used as the target network model.

[0175] It should be noted that historical process formulas are those used during the historical process generation process. The corresponding process result is the achievable process result. For example, during the historical process generation process, using historical process formula A achieves process result a. Using historical process formula B achieves process result b. Therefore, both historical process formula A and historical process formula B can be used as training data.

[0176] The training process for the capsule network model is the same as that for the neural network model, using pre-prepared training data to enable the trained model to accurately predict process parameters based on process results. In some implementations, the dataset for training can be prepared in advance and then divided into 80% offline training data and 20% real-time test data. This allows for both training and testing of the model. During testing, parameters are adjusted based on the accuracy of the prediction results. If overfitting occurs in the loss function or gradient descent function, the network architecture and parameters are further adjusted to improve generalization ability. The final adjusted model then yields the predicted process results.

[0177] In this embodiment, the training method of the neural network model can be used to train the network model constructed by the capsule network.

[0178] In some implementations, during the training of the network model constructed using the capsule network with training data, the sum of the interval loss and the reconstruction loss is used as the model loss.

[0179] It should be noted that the interval loss can be defined as the sum of the K types of interval losses. Specifically,

[0180] in, L margin This represents interval loss. T k T represents the loss value of the prediction result for the k-th class. If the prediction result matches the label, then T... k The value is 1 if the sum is less than 1, and 0 otherwise. Typically, λ is taken as 0.5, and m... + =0.9, m - =0.9, v k This represents the output of the k-th high-level capsule.

[0181] The reconstruction loss can be obtained by calculating the Euclidean distance between the prediction and the input. Therefore, in some implementations, when the intermediate margin loss is used as the dominant factor, the model loss is as follows:

[0182] L = L margin +0.005*L reconstruct Where L represents the model loss, L margin L represents the interval loss. reconstruct This represents the reconstruction loss.

[0183] In this embodiment, the sum of the interval loss and the reconstruction loss is used as the model loss, which combines the advantages of both loss functions and prevents overfitting.

[0184] In another embodiment of this application, the model is continuously adjusted and optimized during training using offline data. The modified model is then synchronized online in real time for real-time data processing. The data generated during real-time data processing is then used as offline training data to expand the training library, thereby improving the model's prediction accuracy. Specifically, as shown in Figure 5, this includes an offline data processing stage and a real-time data processing stage. The offline data processing stage includes: offline data training, adjusting the gradient optimization method and objective function, model parameter tuning, outputting and analyzing the accuracy rate, and modifying the model based on the analysis. This yields an optimized model, which is then updated in the real-time data processing stage. The real-time data processing stage involves: real-time data input, prediction of real-time data, outputting and analyzing the accuracy rate, and optimizing the model based on the results. Simultaneously, the generated data is synchronized to the offline training data to expand the training library.

[0185] The method for determining process parameters provided in this application is illustrated below with a specific example. As shown in Figure 6, this method is applied to process equipment and specifically includes:

[0186] Step S601: Input the original process requirement data into the process equipment. The original process requirement data is equivalent to the target process information in the aforementioned embodiments, which will not be described again here.

[0187] Step S602: Data Processing. This is equivalent to the input layer of the aforementioned capsule network layer, where the raw process requirement data is preprocessed.

[0188] Step S603: By using the capsule long short-term memory network model to predict the processed data, process parameters that meet the original process requirements can be obtained.

[0189] Step S604: Determine the process parameters output by the capsule long short-term memory network model as process parameters that can meet the original process requirements data.

[0190] It is understood that the application architecture of this embodiment, as shown in Figure 7, includes: an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, a long short-term memory network layer, and an output layer. Alternatively, the application architecture of this embodiment can also be as shown in Figure 8, including: an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, an attention module, a long short-term memory network layer, and an output layer. The input layer is used to execute step S602. The convolutional layer, main capsule layer, digital capsule layer, attention module, and long short-term memory network layer can be found in the descriptions of the convolutional layer, main capsule layer, digital capsule layer, attention module, and long short-term memory network layer in the preceding embodiments. The output layer is composed of a softmax function. For a description of each part of this application architecture, please refer to the descriptions of the corresponding parts in the preceding embodiments; they will not be described again here.

[0191] This implementation improves the efficiency of acquiring process parameters. Process engineers no longer need to manually try each parameter or guess its approximate range. It can also adapt to different inputs from different chambers to obtain different outputs, making it widely applicable. Furthermore, the capsule long short-term memory network, through its unique dynamic routing algorithm, can comprehensively obtain the characteristics of the process results, resulting in higher accuracy in prediction.

[0192] Exemplary device

[0193] Some embodiments of this application also provide a process parameter determination device, as shown in FIG9, which includes:

[0194] The response module 91 is used to respond to the target process information input by the user and input the target process information into the target network model; wherein, the target process information includes the process results that can be achieved by the process formulation to be determined;

[0195] The model processing module 92 is used to predict the process parameters required to achieve the process results through the target network model; wherein, the target network model includes: a capsule network layer constructed based on capsule network;

[0196] The determination module 93 is used to determine the prediction results output by the target network model as the process parameters of the process formula to be determined.

[0197] In some implementations, the capsule network layer includes: an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, and an output layer;

[0198] Model processing module 92 includes:

[0199] The preprocessing unit is used to preprocess the target process information through the input layer to generate input features;

[0200] The first feature extraction unit is used to extract features from the input features through the convolutional layer and generate an initial feature vector;

[0201] The second feature extraction unit is used to extract features from the initial feature vector through the main capsule layer to generate the initial capsule vector;

[0202] The model processing unit is used to perform feature processing on the initial capsule vector through the digital capsule layer to generate the target number of result capsule vectors; wherein, each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0203] The model output unit is used to output the predicted value with the highest prediction probability for each process parameter based on the result capsule vector through the output layer.

[0204] In some implementations, the second feature extraction unit is specifically used to extract features from the initial feature vector through the main capsule layer in a dynamic routing manner to generate the initial capsule vector.

[0205] In some embodiments, the device further includes:

[0206] The chamber determination module is used to determine the execution chamber based on the target process information through the target network model, wherein the execution chamber is the chamber in the process equipment that executes the process formula to be determined;

[0207] The quantity determination module is used to determine the number of process parameters corresponding to the execution chamber as the target quantity based on the correspondence between the number of chambers and the number of process parameters in the process equipment through the target network model.

[0208] In some implementations, the target network model further includes a long short-term memory network layer constructed based on a long short-term memory network, wherein the input of the long short-term memory network layer is the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0209] In some implementations, the model processing module 92 is specifically used for:

[0210] The target process information is processed by the capsule network layer to generate the target number of result capsule vectors; each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0211] The resulting capsule vector is processed by the forget gate, input gate, output gate, and output layer of the long short-term memory network to obtain the process parameters required to achieve the process results.

[0212] In some implementations, the target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, and an attention module constructed based on an attention mechanism; wherein, the input of the attention module is the target process information, the input of the long short-term memory network layer is the fusion result of the output of the attention module and the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

[0213] In some implementations, the model processing module 92 is specifically used for:

[0214] The target process information is processed by the capsule network layer to generate the target number of result capsule vectors; each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value.

[0215] The attention module processes the target process information to generate a context vector corresponding to each result capsule vector;

[0216] The fusion results of the result capsule vector and context vector are processed by the forget gate, input gate, output gate and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process results.

[0217] In some embodiments, the device further includes:

[0218] The model training module is used for:

[0219] Acquire training data; the training data includes: historical process formulas and the process results that the historical process formulas can achieve;

[0220] The network model constructed using the training data is trained, and the trained network model is then used as the target network model.

[0221] In some implementations, during the training of the network model constructed using the capsule network with training data, the sum of the interval loss and the reconstruction loss is used as the model loss.

[0222] The apparatus for determining process parameters provided in this application embodiment belongs to the same inventive concept as the method for determining process parameters provided in the above embodiments of this application. Technical details not described in detail in this embodiment can be found in the specific processing content of the method for determining process parameters provided in the above embodiments of this application, and will not be repeated here.

[0223] Exemplary semiconductor process equipment and electronic equipment

[0224] Another embodiment of this application also provides a semiconductor process apparatus, including: a host computer and a slave computer connected to the host computer;

[0225] The host computer is used to execute the process parameter determination method of various embodiments of this application to determine the process parameters of the process formula to be determined, and send the process parameters of the process formula to be determined to the slave computer.

[0226] The lower-level machine is used to receive the process parameters of the process formula to be determined, and to carry out the process production according to the process parameters of the process formula to be determined.

[0227] Another embodiment of this application also provides an electronic device. Referring to FIG10, an exemplary embodiment of this application also provides an electronic device, including: a memory and a processor. The memory stores a computer program, and the processor executes the steps in the method for determining process parameters according to various embodiments of this application described in the above embodiments.

[0228] The internal structure of the electronic device is shown in Figure 10. The electronic device includes a processor, memory, network interface, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it follows the steps in the method for determining process parameters according to various embodiments of this application as described in the above embodiments.

[0229] The processor may include the main processor, as well as baseband chips, modems, etc.

[0230] The memory stores a computer program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the computer program may include program code, which includes computer operation instructions. More specifically, the memory may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.

[0231] The processor can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0232] Input devices may include devices that receive data and information input by the user, such as keyboards, mice, cameras, scanners, light pens, voice input devices, touch screens, pedometers, or gravity sensors.

[0233] Output devices may include devices that allow information to be output to a user, such as displays, printers, speakers, etc.

[0234] The communication interface may include any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0235] The processor executes the computer program stored in the memory and calls other devices, which can be used to implement the various steps of any of the process parameter determination methods provided in the above embodiments of this application.

[0236] The electronic device may also include a display component and a voice component. The display component may be a liquid crystal display screen or an e-ink display screen. The input device of the electronic device may be a touch layer covering the display component, or a button, trackball or touchpad set on the casing of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0237] Those skilled in the art will understand that the structure shown in Figure 10 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0238] Exemplary computer program products and computer-readable storage media

[0239] In addition to the methods and devices described above, the method for determining process parameters provided in the embodiments of this application can also be a computer program product, which includes a computer program. When the computer program is run by a processor, it causes the processor to perform the steps in the method for determining process parameters according to various embodiments of this application as described in the "Exemplary Methods" section above.

[0240] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0241] Furthermore, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor of the steps in the method for determining process parameters according to various embodiments of this application as described in the "Exemplary Methods" section above.

[0242] It is understood that the specific examples in this document are only intended to help those skilled in the art better understand the implementation methods of this application, and are not intended to limit the scope of this application.

[0243] It is understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0244] It is understood that the various implementation methods described in this application can be implemented individually or in combination, and the implementation methods in this application are not limited in this respect.

[0245] Unless otherwise stated, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this application includes any and all combinations of one or more of the associated listed items. The singular forms "a," "the," and "the" as used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0246] It is understood that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0247] It is understood that the memory in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Specifically, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM). It should be noted that the memory in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0248] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0249] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the aforementioned method implementations, and will not be repeated here.

[0250] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0251] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0252] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0253] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0254] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included 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 determining process parameters, characterized in that, The method includes: In response to user-inputted target process information, the target process information is input into the target network model; wherein, the target process information includes the process results that the process formulation to be determined can achieve; The target network model predicts the process parameters required to achieve the process result; wherein, the target network model includes: a capsule network layer constructed based on capsule networks; The prediction results output by the target network model are determined as the process parameters of the process formula to be determined.

2. The method according to claim 1, characterized in that, The capsule network layer includes: an input layer, a convolutional layer, a main capsule layer, a digital capsule layer, and an output layer; The target network model predicts the process parameters required to achieve the process result, including: The target process information is preprocessed through the input layer to generate input features; The input features are extracted through the convolutional layer to generate an initial feature vector; The initial feature vector is extracted through the main capsule layer to generate an initial capsule vector; The initial capsule vector is processed by the digital capsule layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value. Based on the resulting capsule vector, the output layer outputs the predicted value with the highest prediction probability for each process parameter.

3. The method according to claim 2, characterized in that, The initial capsule vector is generated by extracting features from the initial feature vector through the main capsule layer, including: The initial capsule vector is generated by extracting features from the initial feature vector through dynamic routing in the main capsule layer.

4. The method according to claim 2, characterized in that, Before generating the target number of result capsule vectors by performing feature processing on the initial capsule vector through the digital capsule layer, the method further includes: The execution chamber is determined based on the target process information using the target network model, wherein the execution chamber is a chamber in the process equipment that executes the process formula to be determined; Based on the correspondence between the number of chambers and the number of process parameters in the process equipment, the number of process parameters corresponding to the execution chamber is determined as the target number using the target network model.

5. The method according to claim 1, characterized in that, The target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, wherein the input of the long short-term memory network layer is the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

6. The method according to claim 5, characterized in that, The target network model predicts the process parameters required to achieve the process result, including: The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value. The result capsule vector is processed by the forget gate, input gate, output gate, and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

7. The method according to claim 1, characterized in that, The target network model further includes: a long short-term memory network layer constructed based on a long short-term memory network, and an attention module constructed based on an attention mechanism; wherein, the input of the attention module is the target process information, the input of the long short-term memory network layer is the fusion result of the output of the attention module and the output of the capsule network layer, and the output of the long short-term memory network layer is the output of the target network model.

8. The method according to claim 7, characterized in that, The target network model predicts the process parameters required to achieve the process result, including: The target process information is processed by the capsule network layer to generate a target number of result capsule vectors; wherein each result capsule vector corresponds to a process parameter, and the result capsule vector includes multiple predicted values ​​of its corresponding process parameter and the predicted probability of each predicted value. The attention module processes the target process information to generate a context vector corresponding to each result capsule vector. The fusion result of the result capsule vector and the context vector is processed by the forget gate, input gate, output gate and output layer of the long short-term memory network layer to obtain the process parameters required to achieve the process result.

9. The method according to claim 1, characterized in that, In response to user-inputted target process information, before inputting the target process information into the target network model, the method further includes: Acquire training data; wherein, the training data includes: historical process formulas and the process results that the historical process formulas can achieve; The training data is used to train the network model constructed by the capsule network, and the trained network model is determined as the target network model.

10. The method according to claim 9, characterized in that, During the training process of the network model constructed using the aforementioned training data, the sum of the interval loss and the reconstruction loss is used as the model loss.

11. A device for determining process parameters, characterized in that, The device includes: A response module is used to respond to user-inputted target process information and input the target process information into the target network model; wherein, the target process information includes the process results that the process formulation to be determined can achieve; The model processing module is used to predict the process parameters required to achieve the process result through the target network model; wherein, the target network model includes: a capsule network layer constructed based on capsule network; The determination module is used to determine the prediction results output by the target network model as the process parameters of the process formula to be determined.

12. A semiconductor process apparatus, characterized in that, Includes: a host computer and a slave computer connected to the host computer; The host computer is used to execute the method for determining process parameters as described in any one of claims 1 to 10, to determine the process parameters of the process formula to be determined, and to send the process parameters of the process formula to be determined to the slave computer. The lower-level machine is used to receive the process parameters of the process formula to be determined, and to carry out the process production according to the process parameters of the process formula to be determined.

13. An electronic device, characterized in that, include: Processor and memory; The memory is connected to the processor and is used to store computer programs; The processor is configured to implement the method for determining process parameters as described in any one of claims 1 to 10 by running a computer program stored in the memory.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for determining process parameters as described in any one of claims 1 to 10.

Images