Calculation method of plasma coating cleaning process parameters

Abstract

The present invention relates to the technical field of plasma cleaning process parameter calculation, and specifically to a method for calculating plasma coating cleaning process parameters, comprising the following steps: first, obtaining experimental data characterized by film material and cleaning rate and labeled with process parameters; then, inputting the experimental data into a deep neural network for training to obtain a trained deep neural network model; finally, inputting a sample to be predicted with known film material and cleaning rate into the trained deep neural network model to predict the corresponding process parameters; the present invention can provide accurate process parameters for the plasma coating cleaning process, thereby improving the quality and efficiency of cleaning.

Description

Technical Field

[0001] The present invention relates to the technical field of plasma cleaning process parameter calculation, in particular to a method for calculating plasma coating cleaning process parameters. Background Art

[0002] Plasma coating cleaning is a technology that uses the chemical reaction generated by plasma to remove pollutants, oxides and undesirable components on the surface of materials. It has a wide range of applications in materials science, electronic manufacturing, optics, aerospace and other fields.

[0003] In plasma coating cleaning technology, commonly used process methods mainly include redox cleaning, physical collision cleaning, discharge cleaning, chemical reaction cleaning and reactive ion beam cleaning. When using the above process methods for cleaning, some technicians will rely on work experience and determine the process parameters in the cleaning process based on the ultimate cleaning goal. Technicians can achieve cleaning through empirical judgment, but faced with a numerical operating system, empirical judgment cannot provide accurate and specific process parameters, resulting in certain errors in empirical judgment, which in turn causes a decrease in cleaning accuracy. Although during the cleaning process, technicians will promptly adjust the cleaning process parameters according to the real-time cleaning conditions to meet the cleaning requirements as much as possible.

[0004] To sum up, because the technicians did not give accurate process parameters at the beginning of cleaning, the process parameters needed to be constantly adjusted and corrected during the later cleaning process, which not only reduced the cleaning accuracy, but also reduced the cleaning speed and the quality of the cleaned surface. Summary of the Invention

[0005] In order to avoid and overcome the technical problems existing in the prior art, the present invention provides a method for calculating process parameters of plasma coating cleaning. The present invention can provide accurate process parameters for the plasma coating cleaning process, thereby improving the cleaning quality and efficiency.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for calculating the process parameters of plasma coating cleaning includes the following steps:

[0008] S1. Acquire experimental data characterized by film material and cleaning rate and labeled by process parameters;

[0009] S2. Input the experimental data into the deep neural network for training to obtain a trained deep neural network model;

[0010] S3. Input the samples to be predicted with known film material and cleaning rate into the trained deep neural network model to predict the corresponding process parameters.

[0011] As a further solution of the present invention: the film material includes the type of coating material and the coating thickness.

[0012] As a further solution of the present invention: the process parameters include microwave power, gas type, ratio between gases and gas intake volume of each gas.

[0013] As a further solution of the present invention: before inputting the experimental data into the deep neural network, the experimental data are sequentially subjected to feature scaling processing and normalization processing.

[0014] As a further solution of the present invention: the deep neural network model includes an input layer, two or more hidden layers and an output layer.

[0015] As a further solution of the present invention: after the experimental data are subjected to feature scaling and standardization processing, the experimental data are distributed into a training set, a test set, and a validation set according to a set ratio.

[0016] As a further solution of the present invention: the intake amount of each gas includes the intake amount of NF3, the intake amount of O2 and the intake amount of CF4.

[0017] As a further solution of the present invention: the input vector of the input layer is a 0 , a 0 The specific representation is as follows:

[0018] a 0 =[X1,X2,Y] T

[0019] Where X1 represents the input matrix of film material; X2 represents the input matrix of cleaning rate; Y represents the input matrix of known process parameters;

[0020] The output of the i-th hidden layer is a i , a i The specific representation is as follows:

[0021]

[0022] Among them, a i Represents the output of the i-th layer in the deep neural network model, i∈[1,4]; ω i-1 represents the weight coefficient matrix of the i-1th layer; b i-1 represents the bias vector of the i-1th layer; represents the output of the jth neuron in the i-1th layer; f represents the activation function, which is the ReLU function;

[0023] The output vector of the output layer is a 5 , a5 The specific representation is as follows:

[0024] a 5 =[P,S1,S2,S3] T

[0025] Wherein, P represents microwave power; S1 represents NF3 intake volume; S2 represents O2 intake volume; S3 represents CF4 intake volume.

[0026] As a further solution of the present invention: using the evaluation indicators in the regression algorithm: mean square error MSE and R 2 To evaluate the prediction effect of the deep neural network model; the mean square error MSE based on n samples is defined as:

[0027]

[0028] in, Represents the mean square error between the true value and the predicted value of the sample; Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample.

[0029] As a further solution of the present invention: the regression index R based on n samples 2 Defined as:

[0030]

[0031]

[0032] in, Represents the mean of the true values ​​of n samples.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. This invention achieves intelligent control of microwave plasma coating cleaning. By leveraging a DNN prediction model to establish a predictive relationship between cleaning rate and process parameters, it addresses existing issues such as difficulty optimizing process parameters, poor adaptability, and a lack of intelligent control. The system automatically outputs optimal process parameters based on the material being treated and the required cleaning efficiency, improving cleaning efficiency and consistency and enabling intelligent and personalized coating cleaning.

[0035] 2. The present invention combines a large amount of experimental data with deep neural networks to realize the automation of process parameter prediction, thereby improving cleaning efficiency and quality.

[0036] 3. The present invention takes into account multiple factors such as coating material, microwave power, gas type and ratio, and incorporates them into the DNN model for training, thereby achieving comprehensive optimization of process parameters.

[0037] 4. The operator can obtain the optimal process parameters through the DNN model based on the coating material to be processed, which makes the coating cleaning decision more scientific and accurate, avoiding the errors that may be caused by traditional rules of thumb.

[0038] 5. This invention uses a DNN model to predict process parameters, effectively reducing the time and resources required by traditional trial-and-error methods. Furthermore, the model can be trained on large amounts of data, improving prediction accuracy and ensuring consistent cleaning results. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the main calculation steps of the present invention.

[0040] Figure 2 Schematic diagram of the structure of the deep neural network in the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] See also Figures 1 and 2 This invention aims to optimize microwave plasma coating cleaning process parameters and improve cleaning efficiency and stability by training and inverting known process parameters using an artificial intelligence prediction model based on a deep neural network. The deep neural network model is used in this invention to predict the optimal coating cleaning process parameters. The output is the process parameters calculated by the model, which are used to control the operation of the microwave plasma cleaning device. The specific steps are as follows:

[0043] First, the material of the film to be treated and the desired cleaning rate requirements are input into the cleaning control module. The cleaning control module passes the input film material to be treated and the cleaning rate requirements to the deep neural network model. The model will calculate based on experimental data and trained weights to obtain the corresponding process parameters. The deep neural network model outputs a set of optimal process parameters, such as microwave power, the ratio between each gas, and the intake volume of each gas. These process parameters are believed to be able to achieve the desired cleaning rate and improve cleaning efficiency. The cleaning control module passes the process parameters output by the deep neural network model to the microwave plasma cleaning device. Based on the settings of these parameters, the device adjusts the power of the microwave generator, the ratio and flow of gas supply, etc. to achieve a precise coating cleaning process.

[0044] During the cleaning process, the system monitors cleaning results in real time and compares them with the expected cleaning rate. If the cleaning results do not meet expectations, the system automatically modifies the deep neural network model to improve its predictive accuracy. In this way, the output of the deep neural network model is transmitted as process parameters to the actual cleaning equipment, enabling intelligent cleaning control. This makes the cleaning process more efficient and consistent, and can automatically adjust process parameters according to different materials and requirements, thereby optimizing and personalizing the coating cleaning process.

[0045] The training of a deep neural network model consists of the following steps:

[0046] 1. Data preparation and preprocessing

[0047] First, obtain the coating cleaning rate of different materials and the effects of different process parameters. This includes the type of coating material, coating thickness, gas type, such as CF4, O2, NF3, H2, Ar, F2, etc., the ratio between the gases and the intake volume of each gas, microwave power, etc. These experimental data will be used as training samples to build a deep neural network model. Collect experimental data under different coating materials, including process parameters (such as microwave power, gas ratio, flow rate, etc.) and corresponding cleaning rates to ensure that the experimental data covers different coating materials and a wide range of process parameters. Preprocess the data, such as feature scaling and normalization, to facilitate neural network training.

[0048] 2. Construction of deep neural network model

[0049] Based on the experimental data, a deep neural network model is constructed. The deep neural network model can contain multiple hidden layers, each with multiple neurons. The input layer receives the coating material to be processed and the desired cleaning rate. The output layer predicts the optimal process parameters, such as microwave power, the ratio between the gases, and the intake volume of each gas. The architecture of the deep neural network model is designed based on the parameter quantity, including the number of neurons and layers in the input layer, hidden layer, and output layer. The architecture of the deep neural network model is as follows: Figure 2 As shown, an represents the output vector of the nth layer, and bn represents the weight coefficient of the nth layer.

[0050] 3. Model initialization and training

[0051] Initialize the weights and biases of the deep neural network model. You can use random initialization or the weights of a pre-trained model. First, input the training data into the deep neural network model and calculate the predicted value through forward propagation. Then, compare the predicted value with the actual process parameters and calculate the loss function (such as mean square error) to measure the accuracy of the prediction. Then, use the backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters. Adjust the weights and biases of the model according to the gradient to reduce the value of the loss function. Use an optimization algorithm (such as gradient descent) to update the weights and biases of the model to minimize the loss function. Finally, repeat the above steps and iterate multiple times until the loss function converges or the predetermined number of training rounds is reached. The intake volume of each gas includes the NF3 intake volume, the O2 intake volume, and the CF4 intake volume. The input vector of the input layer is a 0 , a 0 The specific representation is as follows:

[0052] a 0 =[X1,X2,Y] T

[0053] Where X1 represents the input matrix of film material; X2 represents the input matrix of cleaning rate; Y represents the input matrix of known process parameters;

[0054] The output of the i-th hidden layer is a i , a i The specific representation is as follows:

[0055]

[0056] Among them, a i Represents the output of the i-th layer in the deep neural network model, i∈[1,4]; ω i-1 represents the weight coefficient matrix of the i-1th layer; b i-1 represents the bias vector of the i-1th layer; represents the output of the jth neuron in the i-1th layer; f represents the activation function, which is the ReLU function;

[0057] The output vector of the output layer is a 5 , a 5 The specific representation is as follows:

[0058] a 5 =[P,S1,S2,S3] T

[0059] Where P represents microwave power; S1 represents NF3 intake volume; S2 represents O2 intake volume; S3 represents CF4 intake volume. The evaluation indicators used in the regression algorithm are: mean square error (MSE) and R 2To evaluate the prediction effect of the deep neural network model; the mean square error MSE based on n samples is defined as:

[0060]

[0061] in, Represents the mean square error between the true value and the predicted value of the sample; Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample.

[0062] Regression index R based on n samples 2 Defined as:

[0063]

[0064]

[0065] in, Represents the mean of the true values ​​of n samples.

[0066] 4. Validation set verification and model adjustment

[0067] After each training iteration, the model is validated using a validation set. The data in the validation set is input into the model, the predicted values ​​are calculated, and compared with the actual process parameters. The performance of the model on unseen data is evaluated based on the loss function value of the validation set. Based on the performance of the validation set, the model's hyperparameters, such as the learning rate, batch size, and number of hidden layer nodes, can be adjusted to further optimize the model performance. By inputting the validation set data into the deep neural network model for verification, the model's prediction mean square error is less than 1.4%, indicating that the model has a high prediction accuracy and can more accurately predict the overall process parameters of the plasma cleaning process. The parameters of the deep neural network model are shown in Table 1.

[0068] Table 1. Deep neural network model parameters

[0069]

[0070] 5. Test set evaluation

[0071] After training is complete, the model is evaluated using an independent test set. The loss function and other evaluation metrics are calculated on the test set to assess the performance of the model on real data.

[0072] When the model performs satisfactorily, the trained deep neural network model is saved for subsequent use in practical applications. The model can predict the optimal process parameters based on the input coating material.

[0073] Through these steps, the deep neural network model can learn the relationship between coating cleaning process parameters and cleaning rate through training data, thereby achieving accurate prediction.

[0074] Based on this model, during the actual cleaning process, the operator inputs the film material to be processed and the desired cleaning rate.

[0075] The system automatically calculates the optimal process parameters using a deep neural network model. These process parameters will be used to control the operation of the microwave plasma cleaning device.

[0076] At the same time, the system will correct and optimize the model based on the actual cleaning effect.

[0077] Based on the above prediction process, the process parameters of the diamond film layer, SiNx layer and organic paint film layer in the plasma removal process were predicted respectively.

[0078] Some prediction results are shown in Tables 2, 3 and 4.

[0079] Table 2 Diamond film removal process parameters

[0080]

[0081] Table 3 SiNx layer removal process parameters

[0082]

[0083] Table 4 Process parameters for removing organic paint film

[0084]

[0085] From Tables 1, 2, and 3, it can be seen that the optimal process for removing the diamond film is: microwave power 2500W, gas NF3: 900sccm, O2: 100sccm, CF4: 0sccm, heating temperature 250°C, time 10min, and removal rate 0.202um / min.

[0086] The optimal process for removing the SiNx layer is: microwave power 3000W, using gases NF3: 0sccm, O2: 300sccm, CF4: 1800sccm, heating temperature 250℃, time 10min, and removal rate 0.173um / min.

[0087] The optimal process for removing the organic paint film layer is: microwave power 2500W, using gases NF3: 0sccm, O2: 1800sccm, CF4: 200sccm, heating temperature 250℃, time 10min, and removal rate 0.421um / min.

[0088] The deep neural network model that can be established can accurately predict the process parameters during the plasma removal process, thereby optimizing and personalizing the coating cleaning process and improving the accuracy and efficiency of the removal.

[0089] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for calculating process parameters of plasma coating cleaning, characterized in that: The following steps are involved: S1. Acquire experimental data characterized by film material and cleaning rate and labeled by process parameters. The film material includes coating material type and coating thickness, the cleaning rate is the coating thickness cleaned per minute, and the process parameters include microwave power, gas type, ratio between gases, and gas intake volume of each gas. S2. After performing feature scaling and standardization on the experimental data, the experimental data is input into a deep neural network for training to obtain a trained deep neural network model; S3. Input a sample to be predicted with known film material and cleaning rate into the cleaning control module. The cleaning control module transmits the input film material and cleaning rate requirements to the trained deep neural network model to predict the corresponding process parameters. The cleaning control module transmits the output process parameters to the microwave plasma cleaning device to adjust the microwave power of the microwave generator. During the cleaning process, the cleaning effect is monitored in real time and compared with the expected cleaning rate. If the cleaning effect does not meet the expectations, the deep neural network model is corrected according to the actual cleaning effect to achieve an accurate coating cleaning process.

2. The method for calculating process parameters of plasma coating cleaning according to claim 1, characterized in that: A deep neural network model consists of an input layer, two or more hidden layers, and an output layer.

3. The method for calculating process parameters of plasma coating cleaning according to claim 2, characterized in that: After the experimental data are feature scaled and standardized, they are divided into training set, test set and validation set according to the set ratio.

4. The method for calculating process parameters of plasma coating cleaning according to claim 3, characterized in that: The intake amount of each gas includes the intake amount of NF3, the intake amount of O2 and the intake amount of CF4.

5. The method for calculating process parameters of plasma coating cleaning according to claim 4, characterized in that: The input vector of the input layer is a 0 , a 0 The specific representation is as follows: to 0 =[X1,X2,Y] T Where X1 represents the input matrix of film material; X2 represents the input matrix of cleaning rate; Y represents the input matrix of known process parameters; The output of the i-th hidden layer is a i , a i The specific representation is as follows: Among them, a i Represents the output of the i-th layer in the deep neural network model, i∈[1,4]; ω i-1 represents the weight coefficient matrix of the i-1th layer; b i-1 represents the bias vector of the i-1th layer; represents the output of the jth neuron in the i-1th layer; f represents the activation function, which is the ReLU function; The output vector of the output layer is a 5 , a 5 The specific representation is as follows: <h2 style=";text-align:left;direction:ltr">a<h2 style=";text-align:left;direction:ltr"> 5 <h2 style=";text-align:left;direction:ltr"> =[P,S1,S2,S3]<h2 style=";text-align:left;direction:ltr"> T Wherein, P represents microwave power; S1 represents NF3 intake volume; S2 represents O2 intake volume; S3 represents CF4 intake volume.

6. The method for calculating process parameters of plasma coating cleaning according to claim 5, characterized in that: Evaluation indicators used in regression algorithm: mean square error MSE and R 2 To evaluate the prediction effect of the deep neural network model; the mean square error MSE based on n samples is defined as: in, Represents the mean square error between the true value and the predicted value of the sample; Represents the predicted value of the i-th sample, y i Represents the true value of the i-th sample.

7. The method for calculating process parameters of plasma coating cleaning according to claim 6, characterized in that: Regression index R based on n samples 2 Defined as: in, Represents the mean of the true values ​​of n samples.

Images