WO2024021259A1

WO2024021259A1 - Automatic batching system for buffered oxide etch production and batching method thereof

Info

Publication number: WO2024021259A1
Application number: PCT/CN2022/119567
Authority: WO
Inventors: 刘水星; 吴桂龙; 王茶英; 张永彪; 罗永春
Original assignee: 福建天甫电子材料有限公司
Priority date: 2022-07-29
Filing date: 2022-09-19
Publication date: 2024-02-01
Also published as: CN115240046B; CN115240046A

Abstract

An automatic batching system for buffered oxide etch production and a batching method thereof. The present invention uses artificial intelligence control technology, and on the basis of a deep neural network model, dynamic implicit features of hydrogen injection speed, a liquid chromatogram of a buffered oxide etch and a surface tension value of the buffered oxide etch are respectively extracted in time dimension. In this way, the hydrogen injection speed can be controlled on the basis of the real-time condition and dynamic change features of the surface tension of the etch, and thus the etching rate and etching quality of the buffered oxide etch in a using process can be ensured.

Description

Automatic batching system for buffered oxide etching liquid production and batching method thereof

Technical field

The present invention relates to the field of intelligent production, and more specifically, to an automatic batching system for the production of buffered oxide etching liquid and a batching method thereof.

Background technique

As the world's semiconductor industry manufacturing industry gradually shifts to mainland China, the domestic demand for buffered oxidation etchants is increasing year by year. Buffered oxidation etchants are mainly used in the microelectronics industry as cleaning agents and etchants. They are often used in the semiconductor industry to etch the oxide layer without photoresist shields. The acidic ammonium fluoride etching solution whose main components are hydrofluoric acid and ammonium fluoride is also called BOE etching solution. The surface tension of the etchant is one of the key factors affecting the etching rate and etching quality.

Hydrophilicity refers to the physical property of a molecule that can form temporary bonds with water molecules through hydrogen bonds. Hydrophobicity refers to the physical property of a molecule (hydrophobic substance) that repels water. Hydrophilicity and hydrophobicity can be collectively referred to as wettability, while the wettability of liquids can be characterized by surface tension. In order to improve the wetting performance of the etchant and reduce the surface tension, it is necessary to study the surface tension of the etching solution. The surface tension is very high and the wettability of the etched layer of the semiconductor silicon wafer is very poor, which can easily lead to serious deformation of the etched pattern in actual engineering applications. The lower surface tension can increase the permeability of the etching solution and etch into tiny pores. Therefore, in order to improve the wetting performance of the etchant and reduce the surface tension to achieve the desired effect, an optimized automatic batching system for the production of buffered oxide etching solutions is expected.

Contents of the invention

In order to solve the above technical problems, this application is proposed. Embodiments of the present application provide an automatic batching system and batching method for the production of buffered oxide etching liquid, which uses artificial intelligence control technology to control the hydrogen injection speed and buffered oxide etching based on a deep neural network model. The liquid chromatogram of the liquid and the surface tension value of the buffer oxide etching liquid are used to extract dynamic implicit features in the time dimension, so that hydrogen can be controlled based on the real-time status and dynamic change characteristics of the surface tension of the etching liquid. The injection speed can ensure the etching rate and etching quality of the buffered oxidation etchant during use.

According to one aspect of the present application, an automatic batching system for buffered oxide etching liquid production is provided, which includes:

The batching process data acquisition module is used to obtain the hydrogen injection rate at multiple predetermined time points within a predetermined time period, the liquid chromatogram of the buffered oxide etching solution, and the surface tension value of the buffered oxide etching solution; the timing encoding module, For passing the hydrogen gas injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution through a temporal encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature. Vector; a liquid chromatogram encoding module for passing the liquid chromatograms of the buffer oxide etching liquid at multiple predetermined time points within the predetermined time period through a first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatogram. phase chromatography feature map; a data dimensionality reduction module, used to reduce the dimensionality of the liquid chromatography feature map to obtain a liquid chromatography feature vector; a response module, used to calculate the injection feature vector relative to the liquid chromatography feature a vector transfer matrix; a fusion module for fusing the transfer matrix with the tension feature vector to obtain a classification feature vector; and

A batching control result generation module is used to pass the classification feature vector through a classifier to obtain a classification result. The classification result is used to indicate that the hydrogen injection rate at the current time point should be increased or decreased.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the timing encoding module includes: an input vector construction unit for combining the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the The surface tension values of the buffered oxide etching solution are arranged as injection speed input vectors and tension input vectors according to the time dimension; a fully connected encoding unit is used to use the fully connected layer of the timing encoder to respectively calculate the injection speed according to the following formula The input vector and the tension input vector undergo fully connected encoding to respectively extract the high-dimensional hidden features of the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,

Represents matrix multiplication; a one-dimensional convolution coding unit, used to perform one-dimensional convolution coding on the injection velocity input vector and the tension input vector using the following formula using the one-dimensional convolution layer of the temporal encoder to respectively Extract high-dimensional implicit correlation features between the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the liquid chromatogram encoding module is further used for: the first convolutional neural network using a three-dimensional convolution kernel in the forward transmission of the layer Perform three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel to obtain a convolution feature map; perform mean pooling processing on the convolution feature map to obtain a pooled feature map; and, Perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network is the liquid chromatography feature map, and the first convolutional neural network The input of the first layer is the liquid chromatogram of the buffered oxide etching solution at multiple predetermined time points within the predetermined time period.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the data dimensionality reduction module is further used to perform a global pooling based on explicit generalization of each feature matrix of the liquid chromatography feature map based on semantic reasoning information. to obtain the liquid chromatography feature vector, wherein the global pooling based on explicit generalization of semantic reasoning information is based on the natural exponential function value raised to the power of the sum of the eigenvalues of all positions of each feature matrix and The difference between the summed values of the eigenvalues at all positions of each eigenmatrix is performed.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the data dimensionality reduction module is further used to: perform dimensionality reduction on the liquid chromatography characteristic map using the following formula to obtain the liquid chromatography Feature vector; where, the formula is:

in

Represents the eigenvalues converted into probability space [0,1] at each position of the characteristic matrix of the k-th channel of the liquid chromatography characteristic map.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the response module is further used to: calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector according to the following formula; wherein , the formula is:

V ₂ =M*V ₁

Where V ₁ represents the injection eigenvector, M represents the transfer matrix, and V ₂ represents the liquid chromatography eigenvector.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the fusion module is further used to: fuse the transfer matrix and the tension feature vector according to the following formula to obtain the classification feature vector; wherein , the formula is:

Wherein, M represents the transfer matrix, V represents the tension feature vector, V' represents the classification feature vector,

Represents matrix multiplication.

In the above-mentioned automatic batching system for the production of buffered oxide etching solution, the batching control result generation module is further used to: use the classifier to process the classification feature vector with the following formula to obtain the classification result , where the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, and B ₁ to B _n are biases Vector, X is the classification feature vector.

According to another aspect of the present application, a batching method of an automatic batching system for buffered oxide etching liquid production includes:

Obtain the hydrogen injection rate at multiple predetermined time points within a predetermined time period, the liquid chromatogram of the buffer oxide etching solution, and the surface tension value of the buffer oxide etching solution; and combine the multiple predetermined time points within the predetermined time period. The hydrogen injection rate and the surface tension value of the buffer oxide etching solution are respectively passed through a timing encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector; multiple predetermined time points within the predetermined time period are The liquid chromatogram of the buffered oxide etching solution is passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatography characteristic map; the liquid chromatography characteristic map is dimensionally reduced to obtain the liquid chromatography characteristic vector. ; Calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector; fuse the transfer matrix with the tension feature vector to obtain a classification feature vector; and

The classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the hydrogen injection speed at the current time point should be increased or decreased.

In the above-mentioned batching method of the automatic batching system for the production of buffered oxide etching liquid, the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffered oxide etching liquid are respectively calculated by including The temporal encoder of the one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector includes: converting the hydrogen injection speed at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution according to The time dimension is arranged into injection speed input vector and tension input vector respectively; use the fully connected layer of the temporal encoder to perform fully connected encoding on the injection speed input vector and the tension input vector using the following formula to extract the respective The high-dimensional implicit features of the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

represents matrix multiplication; use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the injection velocity input vector and the tension input vector using the following formula to extract the injection velocity input vector and The high-dimensional implicit correlation features between the eigenvalues of each position in the tension input vector, where the formula is:

In the above-mentioned batching method of the automatic batching system for the production of buffered oxide etching liquid, the liquid chromatograms of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period are passed through the first three-dimensional convolution kernel. A convolutional neural network to obtain the liquid chromatography characteristic map, including: the first convolutional neural network using a three-dimensional convolution kernel performs on the input data respectively in the forward pass of the layer: based on the three-dimensional convolution kernel, The input data is subjected to three-dimensional convolution processing to obtain a convolution feature map; the convolution feature map is subjected to mean pooling processing to obtain a pooled feature map; and, the pooled feature map is subjected to non-linear activation to obtain activation Feature map; wherein, the output of the last layer of the first convolutional neural network is the liquid chromatography feature map, and the input of the first layer of the first convolutional neural network is the number of samples within the predetermined time period. Liquid chromatogram of the buffered oxide etching solution at a predetermined time point.

In the above-mentioned batching method of the automatic batching system for the production of buffered oxide etching solution, dimensionality reduction is performed on the liquid chromatography characteristic map to obtain a liquid chromatography characteristic vector, which includes: The feature matrix is subjected to global pooling based on explicit generalization of semantic reasoning information to obtain the liquid chromatography feature vector, wherein the global pooling based on explicit generalization of semantic reasoning information is based on all positions of each feature matrix. The sum of the eigenvalues is performed as the difference between the value of the natural exponential function raised to a power and the sum of the eigenvalues at all positions of the respective eigenmatrix.

In the above-mentioned batching method of the automatic batching system for the production of buffered oxide etching solution, the liquid chromatography characteristic map is dimensionally reduced to obtain the liquid chromatography characteristic vector, including: using the following formula to calculate the liquid chromatography characteristic vector Dimensionality reduction is performed on the graph to obtain the liquid chromatography feature vector; wherein, the formula is:

in

In the above-mentioned batching method of the automatic batching system for buffered oxide etching solution production, calculating the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector includes: calculating the relative injection feature vector with the following formula The transfer matrix in the liquid chromatography eigenvector; wherein, the formula is:

V ₂ =M*V ₁

In the above-mentioned batching method of the automatic batching system for buffered oxide etching solution production, fusing the transfer matrix with the tension feature vector to obtain a classification feature vector includes: combining the transfer matrix with the tension feature vector using the following formula The tension feature vectors are fused to obtain the classification feature vector; wherein, the formula is:

Represents matrix multiplication.

In the above-mentioned batching method of the automatic batching system for buffered oxide etching solution production, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate that the hydrogen injection rate at the current time point should be increased. Or should be reduced, including: using the classifier to process the classification feature vector with the following formula to obtain the classification result, where the formula is: softmax{(W _n ,B _n ):...:( W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, and X is the classification feature vector.

Compared with the existing technology, the automatic batching system and batching method provided by this application for the production of buffered oxide etching liquid adopts artificial intelligence control technology and uses a deep neural network model to control the hydrogen injection speed and buffer oxidation respectively. The liquid chromatogram of the etching liquid and the surface tension value of the buffer oxide etching liquid are used to extract dynamic implicit features in the time dimension, so that it can be based on the real-time status and dynamic change characteristics of the surface tension of the etching liquid. Controlling the hydrogen injection rate can ensure the etching rate and etching quality of the buffered oxidation etchant during use.

Description of drawings

The above and other objects, features and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The drawings are used to provide further understanding of the embodiments of the present application, and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation of the present application. In the drawings, like reference numbers generally represent like components or steps.

Figure 1 is an application scenario diagram of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application.

Figure 2 is a block diagram of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application.

Figure 3 is a flow chart of a batching method of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application.

FIG. 4 is a schematic structural diagram of a batching method of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application.

Detailed ways

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described here.

Scenario overview

As mentioned before, with the gradual shift of the world's semiconductor industry manufacturing industry to mainland China, the domestic demand for buffered oxidation etchants is increasing year by year. Buffered oxidation etchants are mainly used in the microelectronics industry as cleaning agents and etchants. They are often used in the semiconductor industry to etch the oxide layer without photoresist shields. The acidic ammonium fluoride etching solution whose main components are hydrofluoric acid and ammonium fluoride is also called BOE etching solution. The surface tension of the etchant is one of the key factors affecting the etching rate and etching quality.

For example, the technical solution provided by patent CN 111892931B; wherein, the formula of the buffered oxidation etchant is: hydrofluoric acid, ammonium fluoride, nitric acid, acetic acid, ultrapure water and penetrant. In particular, a trace amount of hydrogen is introduced into the etchant to catalyze the etching solution, improve the wettability of the etching solution to the surface of the silicon oxide layer, and reduce the surface tension of the BOE etching solution. During the batching process, the injection speed of hydrogen is controlled so that the surface tension of the final BOE etching solution reaches about the effective value.

Based on this, the inventor of the present application found that in the above scheme, if the injection speed of the hydrogen gas is too fast, the hydrogen gas may not be fully dissolved in the etching liquid, and if the injection speed of the hydrogen gas is too slow, the etching liquid may not be fully dissolved. Preparation efficiency will be reduced. And after the hydrogen concentration in the etching liquid reaches a predetermined amount, the additional amount of hydrogen injected has a poor effect on increasing the surface tension of the etching liquid. Therefore, it is expected to intelligently control the hydrogen injection rate based on the real-time conditions and dynamic change characteristics of the surface tension of the etching solution to ensure the etching rate and etching quality of the buffered oxidation etchant during use.

Specifically, in the technical solution of the present application, first, the hydrogen gas injection rate and the surface tension value of the buffered oxide etching solution at multiple predetermined time points within a predetermined time period are obtained through various sensors, such as a tachometer and a tension detector, And use a liquid chromatograph to obtain a liquid chromatogram of the buffer oxide etching solution. Then, considering that the hydrogen injection rate and the surface tension value of the buffer oxide etching solution have dynamic patterns in the time dimension, in the technical solution of this application, in order to more fully realize the excavator output The implicit law of this dynamic change is further analyzed by using a temporal encoder containing a one-dimensional convolution layer to calculate the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution. Encoding to get the injection feature vector and the tension feature vector. In this embodiment of the present application, the timing encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the hydrogen injection speed and the buffer oxide etching liquid through one-dimensional convolutional coding. The surface tension values are correlated in the time series dimension, and the high-dimensional hidden features of the hydrogen injection rate and the surface tension value of the buffer oxide etching solution are extracted through fully connected encoding chalk.

Moreover, for the liquid chromatogram of the buffer oxide etching solution, considering that it has dynamic implicit change characteristics in the time dimension, in order to pay more attention to the characteristics of the buffer oxide etching solution during feature extraction, Correlation of dynamic changes. In the technical solution of the present application, a first convolutional neural network with a three-dimensional convolution kernel is further used to perform the liquid chromatogram of the buffered oxide etching solution at multiple predetermined time points within the predetermined time period. Feature mining is performed to extract the dynamic implicit change characteristics of local features in the liquid chromatogram of the buffer oxide etching solution in the time dimension, thereby obtaining a liquid chromatogram characteristic map.

Then, it should be understood that in order to reduce the number of parameters and prevent overfitting, the characteristics of downsampling-based forward propagation of features based on global mean pooling along the channel dimension are guided by a learnable normal sampling offset. The feature engineering of the integrated neural network is used to effectively model the long-range dependence in the spatial dimension within the feature matrix of the liquid chromatography feature map and the channel dimension between the feature matrices, and further performs along-line processing of the liquid chromatography feature map. Global mean pooling of the channel dimension is used to perform dimensionality reduction to obtain the liquid chromatography feature vector. However, when performing global mean pooling along the channel dimension on the liquid chromatography feature map to obtain the liquid chromatography feature vector, if the global mean of each feature matrix along the channel dimension is simply calculated, then the obtained The liquid chromatography feature vector may not be able to fully express all the effective information of the liquid chromatography feature map in the complex high-dimensional feature space. Therefore, preferably, in the technical solution of this application, explicit information based on semantic reasoning is performed Generalized global pooling, expressed as:

in

is the eigenvalue of each position of the characteristic matrix M _k of the k-th channel of the liquid chromatography characteristic map converted into the probability space [0,1].

In this way, the global pooling based on the explicit generalization of semantic reasoning information can explicitly generalize the semantic concepts corresponding to the eigenvalues of each feature matrix from bottom to top, thus forming the liquid phase in the channel direction. The grouping instances represented by each eigenvalue of the chromatography eigenvector are decoupled through information-based reasoning on the feature semantics, which improves the high-dimensional manifold corresponding to the eigenvalue representation of the liquid chromatography eigenvector. The plasticity of information under high spatial complexity in the semantic space improves the adequacy of information expression of the liquid chromatography feature vector to the liquid chromatography feature map, thereby improving the accuracy of classification.

Further, considering that the characteristic scales of the hydrogen injection rate data and the liquid chromatogram data of the buffer oxide etching liquid are different, and the dynamic change characteristics of the liquid chromatogram of the buffer oxide etching liquid are at high can be regarded as the responsiveness characteristics to the dynamic characteristics of the hydrogen injection speed in the dimensional space. Therefore, in order to better integrate the characteristic information of the two, the relationship between the injection characteristic vector and the liquid chromatography characteristic vector is further calculated. transfer matrix. Then, the transfer matrix and the tension feature vector can be fused to obtain the changing characteristics of the surface tension value of the buffered oxide etching solution and the dynamic characteristics of the hydrogen injection rate, as well as the liquid chromatogram of the buffered oxide etching solution. The local dynamic latent features are fused to obtain the classification feature vector. Furthermore, a classifier is used to perform classification processing on the classification feature vector to obtain a classification result indicating that the hydrogen injection speed at the current time point should be increased or decreased.

Based on this, this application proposes an automatic batching system for the production of buffered oxide etching liquid, which includes: a batching process data acquisition module, used to obtain the hydrogen injection rate and buffer oxidation rate at multiple predetermined time points within a predetermined time period. The liquid chromatogram of the etching liquid and the surface tension value of the buffer oxide etching liquid; a time sequence encoding module used to combine the hydrogen injection rate and the buffer oxide etching rate at multiple predetermined time points within the predetermined time period. The surface tension value of the liquid is passed through a time series encoder containing a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector; the liquid chromatogram encoding module is used to buffer the multiple predetermined time points within the predetermined time period. The liquid chromatogram of the oxide etching liquid is passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatography characteristic map; the data dimensionality reduction module is used to reduce the dimensionality of the liquid chromatography characteristic map to obtain Liquid chromatography feature vector; a response module, used to calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector; a fusion module, used to fuse the transfer matrix and the tension feature vector to obtain Classification feature vector; and, a batching control result generation module, used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate that the hydrogen injection rate at the current time point should be increased or decreased.

FIG. 1 illustrates an application scenario diagram of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application. As shown in Figure 1, in this application scenario, first, the hydrogen gas (for example, the tachometer T1 and the tension detector T2 illustrated in Figure 1) at multiple predetermined time points within a predetermined time period are obtained through various sensors (for example, the tachometer T1 and the tension detector T2 as shown in Figure 1). , H) injection speed as shown in Figure 1 and surface tension value of the buffer oxide etching solution (e.g., E as shown in Figure 1), and use a liquid chromatograph (e.g., as shown in Figure 1 Indicated L) acquires a liquid chromatogram of the buffered oxide etching solution. Then, the hydrogen injection rate at multiple predetermined time points within the predetermined time period, the liquid chromatogram of the buffered oxide etching solution, and the surface tension value of the buffered oxide etching solution are input into a computer configured for buffered oxide etching. in a server with an automatic dosing algorithm for buffer oxide etching liquid production (for example, the cloud server S as shown in Figure 1), wherein the server can use an automatic dosing algorithm for buffer oxide etching liquid production within the predetermined time period The hydrogen injection rate at multiple predetermined time points, the liquid chromatogram of the buffer oxide etching solution, and the surface tension value of the buffer oxide etching solution are processed to generate a representation that the hydrogen injection rate at the current time point should be increased. Or the classification result should be reduced.

After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced below with reference to the accompanying drawings.

Example system

FIG. 2 illustrates a block diagram of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application. As shown in Figure 2, an automatic batching system 200 for the production of buffered oxide etching liquid according to an embodiment of the present application includes: a batching process data acquisition module 210, used to obtain hydrogen injection at multiple predetermined time points within a predetermined time period. speed, the liquid chromatogram of the buffered oxide etching liquid and the surface tension value of the buffered oxide etching liquid; the timing encoding module 220 is used to combine the hydrogen injection speed and the hydrogen gas injection rate at multiple predetermined time points within the predetermined time period. The surface tension values of the buffered oxide etching solution are respectively passed through a time-series encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector; the liquid chromatogram encoding module 230 is used to convert the multiple parameters within the predetermined time period. The liquid chromatogram of the buffered oxide etching solution at a predetermined time point is passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatography characteristic map; the data dimensionality reduction module 240 is used to analyze the liquid chromatogram. The feature map is dimensionally reduced to obtain the liquid chromatography feature vector; the response module 250 is used to calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector; the fusion module 260 is used to combine the transfer matrix with The tension feature vectors are fused to obtain a classification feature vector; and a batching control result generation module 270 is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent the hydrogen at the current point in time. The injection rate should be increased or should be decreased.

Specifically, in the embodiment of the present application, the batching process data acquisition module 210 and the timing encoding module 220 are used to obtain the hydrogen injection rate and the liquid level of the buffer oxide etching liquid at multiple predetermined time points within a predetermined time period. Phase chromatogram and the surface tension value of the buffer oxide etching solution, and the hydrogen injection rate and the surface tension value of the buffer oxide etching solution at multiple predetermined time points within the predetermined time period are respectively represented by a one-dimensional The temporal encoder of the convolutional layer is used to obtain the injection feature vector and the tension feature vector. As mentioned above, in the production process of buffered oxide etching solution, if the injection speed of hydrogen gas is too fast, the hydrogen gas may not be fully dissolved in the etching solution, and if the injection speed of hydrogen gas is too slow, the etching solution may not be fully dissolved. The preparation efficiency will be reduced. And after the hydrogen concentration in the etching liquid reaches a predetermined amount, the additional amount of hydrogen injected has a poor effect on increasing the surface tension of the etching liquid. Therefore, in the technical solution of this application, it is expected to intelligently control the hydrogen injection rate based on the real-time conditions and dynamic change characteristics of the surface tension of the etching liquid to ensure the etching rate and etching quality of the buffered oxidation etchant during use. .

That is, specifically, in the technical solution of the present application, first, the hydrogen gas injection rate and the surface of the buffer oxide etching liquid at multiple predetermined time points within a predetermined time period are obtained through various sensors, such as a tachometer and a tension detector. Tension value, and use a liquid chromatograph to obtain a liquid chromatogram of the buffered oxide etching solution. Then, considering that the hydrogen injection rate and the surface tension value of the buffer oxide etching solution have dynamic patterns in the time dimension, in the technical solution of this application, in order to more fully realize the excavator output The implicit law of this dynamic change is further analyzed by using a temporal encoder containing a one-dimensional convolution layer to calculate the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution. Encoding to get the injection feature vector and the tension feature vector. In a specific example, the timing encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the hydrogen gas injection speed and the buffer oxide etching solution through one-dimensional convolutional coding. The surface tension values are correlated in the time series dimension, and the high-dimensional hidden features of the hydrogen injection rate and the surface tension value of the buffered oxide etching solution are extracted through fully connected encoding chalk.

More specifically, in the embodiment of the present application, the timing encoding module includes: an input vector construction unit, used to combine the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the buffer oxide etching liquid The surface tension values are arranged into injection speed input vectors and tension input vectors according to the time dimension; a fully connected encoding unit is used to use the fully connected layer of the timing encoder to respectively encode the injection speed input vector and the The tension input vector is fully connected to extract the high-dimensional hidden features of the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

Represents a matrix multiplication; a one-dimensional convolution coding unit used to perform one-dimensional convolution coding on the injection velocity input vector and the tension input vector using the following formula using the one-dimensional convolution layer of the temporal encoder to respectively Extract high-dimensional implicit correlation features between the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

Specifically, in the embodiment of the present application, the liquid chromatogram encoding module 230 is used to encode the liquid chromatograms of the buffer oxide etching liquid at multiple predetermined time points within the predetermined time period by using three-dimensional convolution. Kernel the first convolutional neural network to obtain the liquid chromatography characteristic map. It should be understood that for the liquid chromatogram of the buffer oxide etching solution, considering that it has dynamic implicit change characteristics in the time dimension, in order to pay more attention to the buffer oxide etching during feature extraction In the technical solution of the present application, a first convolutional neural network with a three-dimensional convolution kernel is further used to analyze the liquid chromatography of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period. Feature mining is performed on the graph to extract the dynamic implicit change characteristics of the local features in the liquid chromatogram of the buffer oxide etching solution in the time dimension, thereby obtaining the liquid chromatography characteristic map.

More specifically, in the embodiment of the present application, the liquid chromatogram encoding module is further used to: the first convolutional neural network using a three-dimensional convolution kernel separately performs input data on the input data in the forward transmission of the layer. : Perform three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel to obtain a convolution feature map; perform mean pooling processing on the convolution feature map to obtain a pooling feature map; and, perform the pooling The feature map is nonlinearly activated to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network is the liquid chromatography feature map, and the output of the first layer of the first convolutional neural network is the liquid chromatography feature map. The input is a liquid chromatogram of the buffered oxide etching solution at multiple predetermined time points within the predetermined time period.

Specifically, in this embodiment of the present application, the data dimensionality reduction module 240 is used to perform dimensionality reduction on the liquid chromatography feature map to obtain a liquid chromatography feature vector. It should be understood that in order to reduce the number of parameters and prevent overfitting, the characteristics of forward propagation based on downsampling of features based on global mean pooling along the channel dimension are used to guide the convolutional neural network through learnable normal sampling offsets. Feature engineering of the network is used to effectively model the long-range dependencies in the spatial dimension within the feature matrix of the liquid chromatography feature map and the channel dimension between the feature matrices, and further the liquid chromatography feature map is further processed along the channel dimension. The global mean pooling process is used to perform dimensionality reduction to obtain the liquid chromatography feature vector. However, when performing global mean pooling along the channel dimension on the liquid chromatography feature map to obtain the liquid chromatography feature vector, if the global mean of each feature matrix along the channel dimension is simply calculated, then the obtained The liquid chromatography feature vector may not be able to fully express all the effective information of the liquid chromatography feature map in the complex high-dimensional feature space. Therefore, preferably, in the technical solution of this application, explicit information based on semantic reasoning is performed Global pooling for generalization. Correspondingly, in a specific example, global pooling based on explicit generalization of semantic reasoning information is performed on each feature matrix of the liquid chromatography feature map to obtain the liquid chromatography feature vector, wherein the semantic-based Global pooling for explicit generalization of inference information is based on the difference between the value of a natural exponential function raised to the power of the sum of the eigenvalues at all positions of the respective feature matrix and the sum of the eigenvalues at all positions of the respective feature matrix value to proceed.

More specifically, in the embodiment of the present application, the data dimensionality reduction module is further used to: perform dimensionality reduction on the liquid chromatography characteristic map using the following formula to obtain the liquid chromatography characteristic vector; wherein, The formula is:

in

Represents the eigenvalues converted into probability space [0,1] at each position of the characteristic matrix of the k-th channel of the liquid chromatography characteristic map. It should be understood that in this way, the global pooling based on the explicit generalization of semantic reasoning information can explicitly generalize the semantic concepts corresponding to the eigenvalues of each feature matrix from bottom to top, thus forming a channel-direction The grouping instances represented by each feature value of the liquid chromatography feature vector are decoupled through information-based reasoning of feature semantics, which improves the high-dimensional flow corresponding to the feature representation of the liquid chromatography feature vector. The plasticity of information under high spatial complexity in a high-dimensional semantic space improves the adequacy of information expression of the liquid chromatography feature vector to the liquid chromatography feature map, thereby improving the accuracy of classification.

Specifically, in this embodiment of the present application, the response module 250 is used to calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector. It should be understood that considering that the characteristic scales of the hydrogen injection rate data and the liquid chromatogram data of the buffer oxide etching liquid are different, and the dynamic change characteristics of the liquid chromatogram of the buffer oxide etching liquid are in The high-dimensional space can be regarded as a responsive feature to the dynamic characteristics of the hydrogen injection speed. Therefore, in order to better integrate the characteristic information of the two, the injection feature vector is further calculated relative to the liquid chromatography feature vector. transfer matrix.

More specifically, in the embodiment of the present application, the response module is further configured to: calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector with the following formula; wherein, the formula is :

V ₂ =M*V ₁

Specifically, in this embodiment of the present application, the fusion module 260 and the ingredient control result generation module 270 are used to fuse the transfer matrix and the tension feature vector to obtain a classification feature vector, and combine the The classification feature vector is passed through the classifier to obtain a classification result, which is used to indicate that the hydrogen injection speed at the current time point should be increased or decreased. That is to say, in the technical solution of the present application, the transfer matrix and the tension characteristic vector are further integrated to determine the changing characteristics of the surface tension value of the buffered oxide etching solution and the dynamic characteristics of the hydrogen injection rate and the buffered oxidation The local dynamic hidden features of the liquid chromatogram of the etching liquid are fused to obtain the classification feature vector. Furthermore, a classifier is used to perform classification processing on the classification feature vector to obtain a classification result indicating that the hydrogen injection speed at the current time point should be increased or decreased.

Correspondingly, in a specific example, the classifier is used to process the classification feature vector with the following formula to obtain the classification result, wherein the formula is: softmax{(W _n ,B _n ):… :(W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, and X is the classification feature vector.

More specifically, in the embodiment of the present application, the fusion module is further configured to: fuse the transfer matrix and the tension feature vector with the following formula to obtain the classification feature vector; wherein, the formula is :

Represents matrix multiplication.

In summary, based on the embodiment of the present application, the automatic batching system 200 for the production of buffered oxide etching solution is clarified. It uses artificial intelligence control technology to control the hydrogen injection speed and buffered oxide based on a deep neural network model. The liquid chromatogram of the etching solution and the surface tension value of the buffer oxide etching solution are used to extract dynamic implicit features in the time dimension, so that control can be based on the real-time status and dynamic change characteristics of the surface tension of the etching solution. The hydrogen injection rate can ensure the etching rate and etching quality of the buffered oxidation etchant during use.

As mentioned above, the automatic batching system 200 for the production of buffered oxide etching liquid according to the embodiment of the present application can be implemented in various terminal devices, such as a server of the automatic batching algorithm for the production of buffered oxide etching liquid, etc. In one example, the automatic dispensing system 200 for producing buffered oxide etching solution according to an embodiment of the present application can be integrated into a terminal device as a software module and/or a hardware module. For example, the automatic batching system 200 for the production of buffered oxide etching solution can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the system for The automatic batching system 200 for the production of buffered oxide etching solution can also be one of the many hardware modules of the terminal equipment.

Alternatively, in another example, the automatic batching system 200 for buffer oxide etching liquid production and the terminal equipment may also be separate devices, and the automatic batching system 200 for buffer oxide etching liquid production may be Connect to the terminal device through a wired and/or wireless network, and transmit interactive information according to the agreed data format.

Example methods

Figure 3 illustrates a flow chart of a dosing method of an automatic dosing system for buffered oxide etching solution production. As shown in Figure 3, the batching method of the automatic batching system for the production of buffered oxide etching solution according to the embodiment of the present application includes the step: S110, obtaining the hydrogen injection rate, buffer oxidation rate and buffer oxidation rate at multiple predetermined time points within a predetermined time period. The liquid chromatogram of the etching liquid and the surface tension value of the buffer oxide etching liquid; S120, calculate the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension of the buffer oxide etching liquid. The values are respectively passed through a temporal encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector; S130, use the liquid chromatogram of the buffered oxide etching solution at multiple predetermined time points within the predetermined time period. The first convolutional neural network of the three-dimensional convolution kernel to obtain the liquid chromatography feature map; S140, perform dimensionality reduction on the liquid chromatography feature map to obtain the liquid chromatography feature vector; S150, calculate the injection feature vector relative to The transfer matrix of the liquid chromatography feature vector; S160, fuse the transfer matrix with the tension feature vector to obtain a classification feature vector; and, S170, pass the classification feature vector through a classifier to obtain a classification result, The classification result is used to indicate that the hydrogen injection rate at the current time point should be increased or decreased.

FIG. 4 illustrates a schematic diagram of the architecture of a batching method of an automatic batching system for buffered oxide etching solution production according to an embodiment of the present application. As shown in Figure 4, in the network architecture of the batching method of the automatic batching system for buffered oxide etching liquid production, first, the obtained hydrogen injection rate at multiple predetermined time points within the predetermined time period ( For example, P1 as shown in Figure 4) and the surface tension value of the buffer oxide etching solution (for example, P2 as shown in Figure 4) are respectively passed through a temporal encoder including a one-dimensional convolution layer (for example, E) as shown in Figure 4 to obtain the injection feature vector (for example, VF1 as shown in Figure 4) and the tension feature vector (for example, VF2 as shown in Figure 4); then, the predetermined time The liquid chromatograms of the buffered oxide etching solution at multiple predetermined time points within the segment (for example, P3 as shown in Figure 4) are passed through a first convolutional neural network using a three-dimensional convolution kernel (for example, as shown in Figure 4 CNN as shown) to obtain the liquid chromatography feature map (for example, F as shown in Figure 4); then, perform dimensionality reduction on the liquid chromatography feature map to obtain the liquid chromatography feature vector (for example, as shown in Figure 4 VF3 as shown in Figure 4); then, calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector (for example, MF as shown in Figure 4); then, compare the transfer matrix with the The tension feature vectors are fused to obtain a classification feature vector (for example, VF as shown in Figure 4); and, finally, the classification feature vector is passed through a classifier (for example, a circle S as shown in Figure 4) To obtain a classification result, the classification result is used to indicate that the hydrogen injection speed at the current time point should be increased or decreased.

More specifically, in steps S110 and S120, the hydrogen injection rate at multiple predetermined time points within a predetermined time period, the liquid chromatogram of the buffered oxide etching liquid, and the surface tension value of the buffered oxide etching liquid are obtained, And pass the hydrogen gas injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution through a timing encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector. . It should be understood that during the production process of the buffered oxide etching solution, if the injection speed of hydrogen gas is too fast, the hydrogen gas may not be fully dissolved in the etching solution, and if the injection speed of the hydrogen gas is too slow, the etching solution may not be fully dissolved. Preparation efficiency will be reduced. And after the hydrogen concentration in the etching liquid reaches a predetermined amount, the additional amount of hydrogen injected has a poor effect on increasing the surface tension of the etching liquid. Therefore, in the technical solution of this application, it is expected to intelligently control the hydrogen injection rate based on the real-time conditions and dynamic change characteristics of the surface tension of the etching liquid to ensure the etching rate and etching quality of the buffered oxidation etchant during use. .

That is, specifically, in the technical solution of the present application, first, the hydrogen gas injection rate and the surface of the buffer oxide etching liquid at multiple predetermined time points within a predetermined time period are obtained through various sensors, such as a tachometer and a tension detector. Tension value, and use a liquid chromatograph to obtain a liquid chromatogram of the buffered oxide etching solution. Then, considering that the hydrogen gas injection rate and the surface tension value of the buffer oxide etching solution both have dynamic patterns in the time dimension, in the technical solution of this application, in order to more fully extract the excavator The implicit law of this dynamic change is further analyzed by using a temporal encoder containing a one-dimensional convolution layer to calculate the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching solution. Encoding to get the injection feature vector and the tension feature vector. In a specific example, the timing encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the hydrogen gas injection speed and the buffer oxide etching solution through one-dimensional convolutional coding. The surface tension values are correlated in the time series dimension, and the high-dimensional hidden features of the hydrogen injection rate and the surface tension value of the buffered oxide etching solution are extracted through fully connected encoding chalk.

More specifically, in step S130, the liquid chromatograms of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period are passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatogram. Feature map. It should be understood that for the liquid chromatogram of the buffer oxide etching solution, considering that it has dynamic implicit change characteristics in the time dimension, in order to pay more attention to the buffer oxide etching during feature extraction In the technical solution of the present application, a first convolutional neural network with a three-dimensional convolution kernel is further used to analyze the liquid chromatography of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period. Feature mining is performed on the graph to extract the dynamic implicit change characteristics of local features in the liquid chromatogram of the buffer oxide etching solution in the time dimension, thereby obtaining a liquid chromatography characteristic map.

More specifically, in step S140, the liquid chromatography feature map is dimensionally reduced to obtain a liquid chromatography feature vector. It should be understood that in order to reduce the number of parameters and prevent overfitting, the characteristics of forward propagation based on downsampling of features based on global mean pooling along the channel dimension are used to guide the convolutional neural network through learnable normal sampling offsets. Feature engineering of the network is used to effectively model the long-range dependencies in the spatial dimension within the feature matrix of the liquid chromatography feature map and the channel dimension between the feature matrices, and further the liquid chromatography feature map is further processed along the channel dimension. The global mean pooling process is used to perform dimensionality reduction to obtain the liquid chromatography feature vector. However, when performing global mean pooling along the channel dimension on the liquid chromatography feature map to obtain the liquid chromatography feature vector, if the global mean of each feature matrix along the channel dimension is simply calculated, then the obtained The liquid chromatography feature vector may not be able to fully express all the effective information of the liquid chromatography feature map in the complex high-dimensional feature space. Therefore, preferably, in the technical solution of this application, explicit information based on semantic reasoning is performed Global pooling for generalization. Correspondingly, in a specific example, global pooling based on explicit generalization of semantic reasoning information is performed on each feature matrix of the liquid chromatography feature map to obtain the liquid chromatography feature vector, wherein the semantic-based Global pooling for explicit generalization of inference information is based on the difference between the value of a natural exponential function raised to the power of the sum of the eigenvalues at all positions of the respective feature matrix and the sum of the eigenvalues at all positions of the respective feature matrix value to proceed.

More specifically, in step S150, a transfer matrix of the injection feature vector relative to the liquid chromatography feature vector is calculated. It should be understood that considering that the characteristic scales of the hydrogen injection rate data and the liquid chromatogram data of the buffer oxide etching liquid are different, and the dynamic change characteristics of the liquid chromatogram of the buffer oxide etching liquid are in The high-dimensional space can be regarded as a responsive feature to the dynamic characteristics of the hydrogen injection speed. Therefore, in order to better integrate the characteristic information of the two, the injection feature vector is further calculated relative to the liquid chromatography feature vector. transfer matrix.

More specifically, in steps S160 and S170, the transfer matrix and the tension feature vector are fused to obtain a classification feature vector, and the classification feature vector is passed through a classifier to obtain a classification result. The classification result Used to indicate whether the hydrogen injection rate at the current point in time should be increased or decreased. That is to say, in the technical solution of the present application, the transfer matrix and the tension characteristic vector are further integrated to determine the changing characteristics of the surface tension value of the buffered oxide etching solution and the dynamic characteristics of the hydrogen injection rate and the buffered oxidation The local dynamic hidden features of the liquid chromatogram of the etching liquid are fused to obtain the classification feature vector. Furthermore, a classifier is used to perform classification processing on the classification feature vector to obtain a classification result indicating that the hydrogen injection speed at the current time point should be increased or decreased.

In summary, based on the embodiments of the present application, the batching method of the automatic batching system for the production of buffered oxide etching liquid has been clarified, which uses artificial intelligence control technology to separately control the hydrogen injection speed and buffering based on a deep neural network model. The liquid chromatogram of the oxide etching solution and the surface tension value of the buffered oxide etching solution are used to extract dynamic implicit features in the time dimension, so that the real-time conditions and dynamic change characteristics of the surface tension of the etching solution can be extracted. To control the hydrogen injection rate, thereby ensuring the etching rate and etching quality of the buffered oxidation etchant during use.

The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, advantages, effects, etc. mentioned in this application are only examples and not limitations. These advantages, advantages, effects, etc. cannot be considered to be Each embodiment of this application must have. In addition, the specific details disclosed above are only for the purpose of illustration and to facilitate understanding, and are not limiting. The above details do not limit the application to be implemented using the above specific details.

The block diagrams of the devices, devices, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, devices, equipment, and systems may be connected, arranged, and configured in any manner. Words such as "includes," "includes," "having," etc. are open-ended terms that mean "including, but not limited to," and may be used interchangeably therewith. As used herein, the words "or" and "and" refer to the words "and/or" and are used interchangeably therewith unless the context clearly dictates otherwise. As used herein, the word "such as" refers to the phrase "such as, but not limited to," and may be used interchangeably therewith.

It should also be pointed out that in the device, equipment and method of the present application, each component or each step can be decomposed and/or recombined. These decompositions and/or recombinations shall be considered equivalent versions of this application.

The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, this application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for the purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the form disclosed herein. Although various example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

An automatic batching system for the production of buffered oxide etching liquid, characterized in that it includes: a batching process data acquisition module, used to obtain the hydrogen injection rate at multiple predetermined time points within a predetermined time period, the buffered oxide etching liquid Liquid chromatogram and the surface tension value of the buffer oxide etching solution; a time sequence encoding module used to combine the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension of the buffer oxide etching solution The values are respectively passed through a temporal encoder including a one-dimensional convolution layer to obtain the injection feature vector and the tension feature vector; a liquid chromatogram encoding module is used to convert the buffer oxide etching liquid at multiple predetermined time points within the predetermined time period. The liquid chromatogram is passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatography feature map; the data dimensionality reduction module is used to reduce the dimension of the liquid chromatography feature map to obtain the liquid chromatography feature vector; a response module, used to calculate the transfer matrix of the injection feature vector relative to the liquid chromatography feature vector; a fusion module, used to fuse the transfer matrix with the tension feature vector to obtain a classification feature vector; and a batching control result generation module for passing the classification feature vector through a classifier to obtain a classification result. The classification result is used to indicate that the hydrogen injection rate at the current time point should be increased or decreased.
The automatic batching system for buffered oxide etching solution production according to claim 1, characterized in that the timing encoding module includes: an input vector construction unit for converting multiple predetermined times within the predetermined time period. The hydrogen injection speed of the point and the surface tension value of the buffer oxide etching solution are arranged as an injection speed input vector and a tension input vector according to the time dimension; a fully connected encoding unit is used to use the fully connected layer of the timing encoder to The following formula performs fully connected encoding on the injection speed input vector and the tension input vector respectively to extract the high-dimensional hidden features of the eigenvalues of each position in the injection speed input vector and the tension input vector, where, The formula is:
where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,
Represents a matrix multiplication; a one-dimensional convolution coding unit used to perform one-dimensional convolution coding on the injection velocity input vector and the tension input vector using the following formula using the one-dimensional convolution layer of the temporal encoder to respectively Extract high-dimensional implicit correlation features between the eigenvalues of each position in the injection velocity input vector and the tension input vector, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.
The automatic batching system for the production of buffered oxide etching solution according to claim 2, characterized in that the liquid chromatogram encoding module is further used for: the first convolution neural network using a three-dimensional convolution kernel The network performs three-dimensional convolution processing on the input data in the forward pass of the layer based on the three-dimensional convolution kernel to obtain a convolution feature map; performs mean pooling processing on the convolution feature map to obtain Obtain a pooling feature map; and perform nonlinear activation on the pooling feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network is the liquid chromatography feature map, so The input of the first layer of the first convolutional neural network is the liquid chromatogram of the buffer oxide etching solution at multiple predetermined time points within the predetermined time period.
The automatic batching system for the production of buffered oxide etching liquid according to claim 3, characterized in that the data dimensionality reduction module is further used to perform semantic reasoning on each feature matrix of the liquid chromatography feature map. Global pooling of information explicit generalization to obtain the liquid chromatography feature vector, wherein the global pooling of information explicit generalization based on semantic reasoning is based on the summation value of the eigenvalues of all positions of each feature matrix It is performed as the difference between the value of a natural exponential function raised to a power and the sum of the eigenvalues at all positions of the respective eigenmatrix.
The automatic batching system for the production of buffered oxide etching liquid according to claim 4, characterized in that the data dimensionality reduction module is further used to: reduce the liquid chromatography characteristic map with the following formula Dimension to obtain the liquid chromatography characteristic vector; wherein, the formula is:

in
Represents the eigenvalues converted into probability space [0,1] at each position of the characteristic matrix of the k-th channel of the liquid chromatography characteristic map.
The automatic batching system for the production of buffered oxide etching solution according to claim 5, characterized in that the response module is further used to: calculate the injection characteristic vector relative to the liquid chromatography characteristic according to the following formula The transfer matrix of the vector; wherein, the formula is:

V 2 =M*V 1

Where V 1 represents the injection eigenvector, M represents the transfer matrix, and V 2 represents the liquid chromatography eigenvector.
The automatic batching system for the production of buffered oxide etching solution according to claim 6, characterized in that the fusion module is further used to: fuse the transfer matrix and the tension feature vector according to the following formula to Obtain the classification feature vector; wherein, the formula is:

Wherein, M represents the transfer matrix, V represents the tension feature vector, V' represents the classification feature vector,
Represents matrix multiplication.
The automatic batching system for the production of buffered oxide etching liquid according to claim 7, characterized in that the batching control result generation module is further used to: use the classifier to classify the classification feature vector according to the following formula Processing is performed to obtain the classification result, where the formula is: softmax{(W n ,B n ):...:(W 1 ,B 1 )|X}, where W 1 to W n are weight matrices, B 1 to B n are bias vectors, and X is the classification feature vector.
A batching method of an automatic batching system for the production of buffered oxide etching liquid, which is characterized in that it includes: obtaining the hydrogen injection rate at multiple predetermined time points within a predetermined time period, the liquid chromatogram of the buffered oxide etching liquid, and The surface tension value of the buffer oxide etching liquid; passing the hydrogen injection rate at multiple predetermined time points within the predetermined time period and the surface tension value of the buffer oxide etching liquid through a time series including a one-dimensional convolution layer encoder to obtain the injection feature vector and the tension feature vector; pass the liquid chromatograms of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period through the first convolutional neural network using a three-dimensional convolution kernel to obtain Liquid chromatography characteristic diagram; perform dimensionality reduction on the liquid chromatography characteristic diagram to obtain liquid chromatography characteristic vector; calculate the transfer matrix of the injection characteristic vector relative to the liquid chromatography characteristic vector; compare the transfer matrix with The tension feature vectors are fused to obtain a classification feature vector; and the classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the hydrogen injection speed at the current time point should be increased or decreased.
The batching method of an automatic batching system for the production of buffered oxide etching liquid according to claim 9, wherein the liquid phase of the buffered oxide etching liquid at multiple predetermined time points within the predetermined time period is The chromatogram is passed through the first convolutional neural network using a three-dimensional convolution kernel to obtain the liquid chromatography characteristic map, including: the first convolutional neural network using the three-dimensional convolutional kernel separately processes the input data in the forward pass of the layer. Perform: perform three-dimensional convolution processing on the input data based on the three-dimensional convolution kernel to obtain a convolution feature map; perform mean pooling processing on the convolution feature map to obtain a pooling feature map; and perform the pooling The feature map is nonlinearly activated to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network is the liquid chromatography feature map, and the output of the first layer of the first convolutional neural network is the liquid chromatography feature map. The input is a liquid chromatogram of the buffered oxide etching solution at multiple predetermined time points within the predetermined time period.