Process debugging method and device based on transfer learning

Abstract

An embodiment of this specification provides a process debugging method and device based on transfer learning. A specific implementation of this method includes: obtaining a machine learning model trained by a material supplier, where the model parameters of the machine learning model include first weight factors for each production debugging parameter; constructing a training sample set according to the debugging data of the customer, where the training samples in the training sample set include production debugging parameters and product performance parameters; further training the machine learning model based on the training sample set to obtain second weight factors for each production debugging parameter; determining production debugging parameters as the parameters to be debugged from the production debugging parameters according to the first weight factors and the second weight factors of each production debugging parameter; and sending the parameters to be debugged to the material supplier for the material supplier to debug the production process based on the parameters to be debugged.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of material production processes, and particularly to a process debugging method and device based on transfer learning. Background Art

[0002] In the process of semiconductor integrated circuit manufacturing, a variety of materials are often used. Therefore, semiconductor integrated circuit manufacturers need to purchase materials from material providers. Facing different manufacturer customers, material providers in the semiconductor integrated circuit manufacturing industry chain usually face the problem that the product performance parameters to be debugged are different for different manufacturer customers. Taking a photoresist provider as an example, due to different process flows and process margins of different manufacturer customers, the provided photoresist often needs to be adjusted according to the specific process and supporting machine settings of the manufacturer customer. However, due to trade secrets, there may be obstacles to the sharing of sensitive data between integrated circuit manufacturers and material providers, that is, some sensitive data involving trade secrets cannot be informed to the other party. This will cause the material provider to lack direction and take too much time in the material adjustment process, which will generate additional time and economic costs for both the raw material provider and the integrated circuit manufacturer. Summary of the Invention

[0003] The embodiments of this specification describe a process debugging method and device based on transfer learning. This method first obtains a machine learning model trained by the material supplier, then further trains the machine learning model based on the debugging data of the customer side, and determines the parameters to be debugged according to the weight factors of the production debugging parameters before and after training, so that the material supplier can debug the production process of the material based on the parameters to be debugged. Since the degree of difference between the weight factors of the production debugging parameters before and after the further training of the model can, to a certain extent, reflect the difference degree of the production debugging parameters between the material supplier and the customer side, the parameters to be debugged determined by the first weight factor and the second weight factor can prompt the production debugging parameters that the material supplier may need to debug, thereby helping the material supplier quickly complete the debugging of the material formula that matches the process requirements of the customer side. In addition, there is no need for the material supplier and the customer side to inform each other of sensitive data, so it is beneficial to the protection of trade secrets.

[0004] According to a first aspect, a process debugging method based on transfer learning is provided, including: obtaining a pre-trained machine learning model of a material supplier, where the machine learning model is used to represent the functional relationship between production debugging parameters and product performance parameters in the production process of a target material, and the model parameters of the machine learning model include first weight factors for each production debugging parameter; constructing a training sample set according to the debugging data of a customer, where the training samples in the training sample set include production debugging parameters and product performance parameters; further training the machine learning model based on the training sample set to obtain second weight factors for each production debugging parameter; determining production debugging parameters as parameters to be debugged from the production debugging parameters according to the first weight factors and the second weight factors of each production debugging parameter; and sending the parameters to be debugged to the material supplier for the material supplier to debug the production process based on the parameters to be debugged.

[0005] In one embodiment, the further training the machine learning model based on the training sample set to obtain second weight factors for each production debugging parameter includes: for each production debugging parameter, taking the production debugging parameter as a target production debugging parameter and keeping the first weight factors of other production debugging parameters except the target production debugging parameter unchanged; retraining the machine learning model based on the training sample set to obtain a trained model; and determining the second weight factor of the target production debugging parameter according to the parameters of the trained model.

[0006] In one embodiment, the determining production debugging parameters as parameters to be debugged from the production debugging parameters according to the first weight factors and the second weight factors of each production debugging parameter includes: for each production debugging parameter, determining the difference information between the corresponding first weight factor and the second weight factor of the production debugging parameter; and determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the difference information.

[0007] In one embodiment, the determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the difference information includes: sorting the production debugging parameters according to the difference information corresponding to each production debugging parameter; and determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the sorting result.

[0008] In one embodiment, the above method further includes: displaying the above sorting result for the user to view; and based on the above sorting result, determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged, including: receiving the parameter screening information input by the user based on the above sorting result; and determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged according to the above parameter screening information.

[0009] In one embodiment, the above sorting is performed in descending order according to the corresponding difference information; and based on the above sorting result, determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged, including: determining the production debugging parameters ranked in the top preset positions in the above sorting result as the parameters to be debugged.

[0010] In one embodiment, the above production debugging parameters include material components and usage conditions.

[0011] In one embodiment, the above target material is photoresist, the above production debugging parameters include material components, the coating rotation speed of the photoresist, and the light exposure dose, and the above product performance parameters include critical dimension, thickness, and aspect ratio.

[0012] According to a second aspect, there is provided a process debugging device based on transfer learning, including: an acquisition unit configured to acquire a machine learning model trained by a material supplier, where the machine learning model is used to represent the functional relationship between production debugging parameters and product performance parameters in the production process of a target material, and the model parameters of the machine learning model include first weight factors for each production debugging parameter; a construction unit configured to construct a training sample set according to the debugging data of the customer side, where the training samples in the above training sample set include production debugging parameters and product performance parameters; a training unit configured to further train the above machine learning model based on the above training sample set to obtain second weight factors for each production debugging parameter; a determination unit configured to determine production debugging parameters as the parameters to be debugged from the production debugging parameters according to the first weight factors and the second weight factors of each production debugging parameter; and a sending unit configured to send the above parameters to be debugged to the above material supplier for the above material supplier to debug the above production process based on the above parameters to be debugged.

[0013] According to a third aspect, there is provided a computer-readable storage medium, on which a computer program is stored, and when the above computer program is executed in a computer, the computer is made to execute the method described in any implementation manner of the first aspect.

[0014] According to a fourth aspect, there is provided a computing device including a memory and a processor. The memory stores executable code, and when the processor executes the executable code, the method described in any implementation manner of the first aspect is implemented.

[0015] Based on the method and device for process debugging based on transfer learning provided in the embodiments of the present specification, first, a machine learning model trained by a material supplier is obtained. The model parameters of the machine learning model include first weight factors for each production debugging parameter. Then, the machine learning model is further trained based on the debugging data of the customer side to obtain second weight factors for each production debugging parameter, and the parameters to be debugged are determined according to the weight factors of the production debugging parameters before and after the machine learning model is trained, so that the material supplier can debug the production process of the material based on the parameters to be debugged. Since the degree of difference between the weight factors of the production debugging parameters before and after the model is further trained can reflect to a certain extent the difference degree of the production debugging parameters between the material supplier and the customer side, therefore, the parameters to be debugged determined by the first weight factor and the second weight factor can prompt the material supplier of the production debugging parameters that may need to be debugged, thereby helping the material supplier quickly complete the debugging of the material formula that matches the process requirements of the customer side. In addition, there is no need for the material supplier and the customer side to inform each other of sensitive data, so it is beneficial to the protection of business secrets. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram showing an application scenario to which the embodiments of the present specification can be applied;

[0017] Figure 2 A schematic flowchart showing a method for process debugging based on transfer learning according to an embodiment;

[0018] Figure 3 A schematic diagram showing an example of the model structure of a machine learning model;

[0019] Figure 4 A schematic block diagram showing a device for process debugging based on transfer learning according to an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The technical solutions provided in the present specification will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention and are not intended to limit the invention. In addition, it should be noted that for the sake of description, only parts related to the relevant invention are shown in the drawings. It should be noted that without conflict, the embodiments of the present specification and the features in the embodiments can be combined with each other.

[0021] The embodiments of this specification provide a process debugging method based on transfer learning, which helps the material supplier to quickly complete the debugging of the material formula that meets the process requirements of the customer. It can be understood that transfer learning is a series of general solutions in the current field of deep learning, rather than a specific algorithm model. The transfer learning method of Pre-training+fine-tuning is a very popular transfer learning method in current deep learning. Taking the target material as photoresist as an example, Figure 1 shows a schematic diagram of an application scenario to which the embodiments of this specification can be applied. As Figure 1 shown, first, obtain the machine learning model 102 pre-trained by the material supplier 101 of the photoresist. Among them, the machine learning model 102 can be used to characterize the functional relationship between the production debugging parameters and the product performance parameters in the production process of the photoresist. The model parameters of the machine learning model 102 include the first weight factors for each production debugging parameter. In this example, the production debugging parameters can include material components and usage conditions. For example, the coating rotation speed and light exposure measurement of the photoresist. The product performance parameters can include the critical dimension (CD), thickness, aspect ratio, etc. of the photoresist. Secondly, construct a training sample set 104 according to the debugging data of the customer 103, and use the training sample set 104 to further train the machine learning model 102, that is, adjust the parameters of the machine learning model 102. After parameter adjustment, obtain the second weight factors for each production debugging parameter. Then, determine at least one production debugging parameter as the parameter to be debugged 105 from multiple production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter. Finally, send the parameter to be debugged 105 to the material supplier 101 for the material supplier 101 to debug the production process based on the parameter to be debugged 105. Since the degree of difference between the weight factors of the production debugging parameters before and after the further training of the model can, to a certain extent, reflect the difference degree of the production debugging parameters between the material supplier and the customer, the parameter to be debugged determined by the first weight factor and the second weight factor can prompt the production debugging parameters that the material supplier may need to debug, thus helping the material supplier to quickly complete the debugging of the material formula that meets the process requirements of the customer. In addition, there is no need for the material supplier and the customer to disclose sensitive data to each other, so it is beneficial to the protection of business secrets.

[0022] Continue to refer to Figure 2 , Figure 2The flowchart of a process debugging method based on transfer learning according to an embodiment is shown. It can be understood that this method can be executed by any device, equipment, platform, or equipment cluster with computing and processing capabilities. To protect business secrets, the above process debugging method based on transfer learning can be executed by the customer, or by a third party jointly designated by the material supplier and the customer, which is not limited here. As Figure 2 shown, the process debugging method based on transfer learning may include the following steps:

[0023] Step 201, obtain the machine learning model trained by the material supplier.

[0024] In this embodiment, the machine learning model trained (or pre-trained) by the material supplier can be obtained. Here, the material supplier may refer to the material provider. The machine learning model can be trained based on the experimental data of the material supplier or the production data on the production line. For example, first, the data can be divided into input data and output data, and processed such as normalization, vectorization, and data desensitization. Among them, the input data x1, x2,..x m may refer to the production debugging parameters in the production process of the target material. For example, it can include material components, usage conditions, etc. The output data Y can be the product performance parameters. Then, use AI (Artificial Intelligence) algorithms, such as neural networks, to train the machine learning model in the form of Y = f v (x1, x2, ……, x m ). That is to say, the machine learning model pre-trained by the material supplier can be used to represent the functional relationship between the production debugging parameters and the product performance parameters in the production process of the target material. The model parameters of this machine learning model can include the first weight factors for each production debugging parameter. For example, the weight factor can refer to the weighting factor. It can be understood that the target materials applicable to this embodiment can include various materials used in the semiconductor integrated circuit manufacturing industry chain.

[0025] In practice, the model structure of the above machine learning model can be set according to actual needs. For example, it can include layer structures such as an input layer, a hidden layer, and an output layer. As Figure 3 shown, Figure 3 a schematic diagram showing an example of the model structure of the machine learning model is shown. In Figure 3 the example shown, a can be used to represent the nodes of the neural network, and W can be used to represent the weight matrix of the neural network. It can be understood that Figure 3The number of hidden layers included in the model structure shown, the number of nodes included in each layer, etc. are only illustrative and not a limitation on the model structure of the machine learning model of the present application. In practice, different numbers of layers, nodes, etc. can be set for the machine learning model according to actual needs.

[0026] In some optional implementation manners, the above-mentioned target material may be a photoresist. A photoresist, also known as a photo resist, refers to a corrosion-resistant thin film material whose solubility changes upon irradiation or radiation by ultraviolet light, electron beam, ion beam, X-ray, etc. It is a light-sensitive mixed liquid composed of three main components: a photosensitive resin, a sensitizer, and a solvent. It is used as an anti-corrosion coating material during the photolithography process. When processing the surface of semiconductor materials, if an appropriate selective photoresist is used, the desired image can be obtained on the surface. Photoresist is a material required in the manufacturing process of semiconductor integrated circuits. When the target material is a photoresist, the above-mentioned production debugging parameters may include material components, the coating rotation speed of the photoresist, light exposure dosage, etc., and the above-mentioned product performance parameters may include the critical dimension (CD), thickness, aspect ratio, etc. of the photoresist.

[0027] Step 202: Construct a training sample set according to the debugging data of the customer side.

[0028] In this embodiment, the customer side may refer to a semiconductor integrated circuit manufacturing manufacturer. The manufacturer is a customer of the material supplier, so it is called the customer side here. Generally, during the process R & D stage, the customer side will conduct a series of debugging experiments on the processes related to the raw materials used, so as to obtain debugging data. For the obtained debugging data, processing such as normalization, vectorization, and data desensitization can be performed, and a training sample set can be constructed. The training samples in the training sample set may include generated debugging parameters and product performance parameters. It can be understood that the data included in the training samples constructed based on the debugging data of the customer side may be the same as the types, dimensions, etc. of the data included in the training samples used by the material supplier to train the machine learning model.

[0029] Step 203: Further train the machine learning model based on the training sample set to obtain a second weight factor for each production debugging parameter.

[0030] In this embodiment, based on the training sample set obtained in step 202, the machine learning model can be further trained to adjust the model parameters of the model. After adjusting the parameters, a second weight factor for each production debugging parameter can be obtained.

[0031] In some optional implementation manners, the above-mentioned step 203 can also be specifically carried out as follows:

[0032] First, for each production debugging parameter, use this production debugging parameter as the target production debugging parameter, and set the first weight factor of other production debugging parameters except the target production debugging parameter to remain unchanged.

[0033] Then, retrain the machine learning model based on the training sample set to obtain the trained model.

[0034] Finally, determine the second weight factor of the target production debugging parameter according to the parameters of the trained model.

[0035] Through this implementation method, training can be performed for only one production debugging parameter each time, that is, the first weight factor of other production debugging parameters is not changed during the training process. In this way, the difference in the weight factor corresponding to a single production debugging parameter before and after the machine learning model is further trained can be obtained. Compared with adjusting the weight factors of all production debugging parameters simultaneously during further training, adjusting the weight factor of only one production debugging parameter each time can more accurately obtain the difference between the first weight factor and the second weight factor corresponding to the production debugging parameter.

[0036] Step 204, determine a production debugging parameter as the parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter.

[0037] In this embodiment, at least one production debugging parameter can be determined as the parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter. For example, a production debugging parameter with a difference between the first weight factor and the second weight factor exceeding a preset difference threshold can be determined as the parameter to be debugged. In practice, since the degree of difference in the weight factors of the production debugging parameters before and after the further training of the model can, to a certain extent, reflect the difference degree of the production debugging parameters between the material supplier and the customer, the parameter to be debugged determined by the first weight factor and the second weight factor can prompt the production debugging parameters that the material supplier may need to debug.

[0038] In some alternative implementation methods, the above step 204 may further include the following steps:

[0039] Step S1, for each production debugging parameter, determine the difference information between the corresponding first weight factor and the second weight factor of this production debugging parameter. For example, calculate the difference value.

[0040] Step S2, based on the difference information, determine at least one production debugging parameter as the parameter to be debugged from the production debugging parameters. For example, select the production debugging parameter with the largest difference information as the parameter to be debugged.

[0041] Optionally, the above step S2 can be specifically performed as follows:

[0042] First, sort each production debugging parameter according to the difference information corresponding to each production debugging parameter.

[0043] In this implementation, multiple production debugging parameters can be sorted according to the difference information corresponding to each production debugging parameter in a preset order, for example, in ascending or descending order.

[0044] Then, based on the sorting result, determine at least one production debugging parameter from the production debugging parameters as the parameter to be debugged.

[0045] As an example, the sorting result can be displayed for the user to view. Here, the user can refer to a field engineer and / or a data scientist. The user can manually analyze which production debugging parameters may need to be debugged based on the displayed sorting result and send parameter screening information, which can be used to indicate which production debugging parameters are the parameters to be debugged. In this example, the above-mentioned determination of at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the sorting result can also be specifically carried out as follows: First, receive the parameter screening information input by the user based on the sorting result. Then, according to the received parameter screening information, determine at least one production debugging parameter from the production debugging parameters as the parameter to be debugged. Through this implementation, manual intervention can be carried out when determining the parameter to be debugged, so that the determined parameter to be debugged is more accurate.

[0046] As another example, the above sorting can refer to sorting in descending order according to the corresponding difference information. In this example, the above-mentioned determination of at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the sorting result can also be specifically carried out as follows: Determine the production debugging parameters ranked in the top preset positions in the sorting result as the parameters to be debugged. Through this implementation, the parameter to be debugged can be automatically determined based on the sorting result, thus accelerating the selection efficiency of the parameter to be debugged.

[0047] Step 205: Send the parameter to be debugged to the material supplier for the material supplier to debug the production process based on the parameter to be debugged.

[0048] In this embodiment, the parameter to be debugged determined in step 204 can be sent to the material supplier for the material supplier to debug the production process based on the parameter to be debugged, thereby improving the debugging speed and efficiency.

[0049] Reviewing the above process, in the embodiments of this specification, first, a machine learning model trained by the material supplier is obtained. Then, the machine learning model is further trained based on the debugging data of the customer side, and the parameter to be debugged is determined according to the weight factors of the production debugging parameters before and after training, so that the material supplier can debug the production process of the material based on the parameter to be debugged. Since the difference degree of the weight factors of the production debugging parameters before and after the further training of the model can, to a certain extent, reflect the difference degree of the production debugging parameters between the material supplier and the customer side, therefore, the parameter to be debugged determined by the first weight factor and the second weight factor can prompt the production debugging parameters that the material supplier may need to debug, thereby helping the material supplier quickly complete the debugging of the material formula that matches the process requirements of the customer side. In addition, there is no need for the material supplier and the customer side to inform each other of sensitive data, so it is beneficial to the protection of business secrets.

[0050] According to an embodiment of another aspect, a process debugging device based on transfer learning is provided. The above process debugging device based on transfer learning can be deployed in any device, platform or device cluster with computing and processing capabilities.

[0051] Figure 4 The schematic block diagram of a process debugging device based on transfer learning according to an embodiment is shown. As Figure 4 shown, the process debugging device 400 based on transfer learning includes: an acquisition unit 401 configured to acquire a machine learning model trained by the material supplier, wherein the machine learning model is used to characterize the functional relationship between the production debugging parameters and the product performance parameters in the production process of the target material, and the model parameters of the machine learning model include the first weight factor for each production debugging parameter; a construction unit 402 configured to construct a training sample set according to the debugging data of the customer side, wherein the training samples in the training sample set include production debugging parameters and product performance parameters; a training unit 403 configured to further train the machine learning model based on the training sample set to obtain the second weight factor for each production debugging parameter; a determination unit 404 configured to determine the production debugging parameter as the parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter; a sending unit 405 configured to send the parameter to be debugged to the material supplier for the material supplier to debug the production process based on the parameter to be debugged.

[0052] In some alternative implementation manners of this embodiment, the above training unit 403 is further configured to: for each production debugging parameter, use the production debugging parameter as the target production debugging parameter, and keep the first weight factor of other production debugging parameters except the above target production debugging parameter unchanged; retrain the above machine learning model based on the above training sample set to obtain a trained model; and determine the second weight factor of the above target production debugging parameter according to the parameters of the above trained model.

[0053] In some alternative implementation manners of this embodiment, the above determining unit 404 includes: a first determining subunit (not shown in the figure), configured to determine the difference information between the first weight factor and the second weight factor corresponding to each production debugging parameter; and a second determining subunit (not shown in the figure), configured to determine at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the difference information.

[0054] In some alternative implementation manners of this embodiment, the second determining subunit is further configured to: sort the production debugging parameters according to the difference information corresponding to each production debugging parameter; and determine at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the above sorting result.

[0055] In some alternative implementation manners of this embodiment, the above device 400 further includes: a display unit (not shown in the figure), configured to display the above sorting result for the user to view; and the above determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the above sorting result includes: receiving the parameter screening information input by the user based on the above sorting result; and determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged according to the above parameter screening information.

[0056] In some alternative implementation manners of this embodiment, the above sorting is performed in descending order according to the corresponding difference information; and the above determining at least one production debugging parameter from the production debugging parameters as the parameter to be debugged based on the above sorting result includes: determining the production debugging parameters ranked in the top preset positions in the above sorting result as the parameters to be debugged.

[0057] In some alternative implementation manners of this embodiment, the above production debugging parameters include material components and usage conditions.

[0058] In some alternative implementation manners of this embodiment, the above target material is photoresist, the above production debugging parameters include material components, the coating rotation speed of the photoresist, and the light exposure dose, and the above product performance parameters include critical dimension, thickness, and aspect ratio.

[0059] According to an embodiment of another aspect, there is also provided a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer, causes the computer to execute Figure 2 the described method.

[0060] According to an embodiment of still another aspect, there is also provided a computing device including a memory and a processor. It is characterized in that an executable code is stored in the memory, and when the processor executes the executable code, the Figure 2 described method is implemented.

[0061] Those of ordinary skill in the art should further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0062] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0063] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A process debugging method based on transfer learning, comprising: Obtaining a machine learning model trained by a material supplier, wherein the machine learning model is used to characterize the functional relationship between production debugging parameters and product performance parameters in the production process of the target material, and the model parameters of the machine learning model include first weight factors for each production debugging parameter; Constructing a training sample set according to the debugging data of the customer, wherein the training samples in the training sample set include production debugging parameters and product performance parameters; Further training the machine learning model based on the training sample set to obtain second weight factors for each production debugging parameter, including: for each production debugging parameter, taking this production debugging parameter as the target production debugging parameter, and setting the first weight factors of other production debugging parameters except the target production debugging parameter to remain unchanged; retraining the machine learning model based on the training sample set to obtain a trained model; determining the second weight factor of the target production debugging parameter according to the parameters of the trained model; Determining a production debugging parameter as a parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter; Sending the parameter to be debugged to the material supplier for the material supplier to debug the production process based on the parameter to be debugged.

2. The method according to claim 1, wherein The determining a production debugging parameter as a parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter includes: For each production debugging parameter, determining the difference information between the corresponding first weight factor and the second weight factor of this production debugging parameter; Based on the difference information, determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters.

3. The method according to claim 2, wherein The determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the difference information includes: Sorting the production debugging parameters according to the difference information corresponding to each production debugging parameter; Based on the sorting result, determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters.

4. The method according to claim 3, wherein, The method further includes: Displaying the sorting result for the user to view; and The determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the sorting result includes: Receiving parameter screening information input by the user based on the sorting result; Determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters according to the parameter screening information.

5. The method according to claim 3, wherein, The sorting is performed in descending order according to the corresponding difference information; and The determining at least one production debugging parameter as a parameter to be debugged from the production debugging parameters based on the sorting result includes: Determining the production debugging parameters ranked in the top preset positions in the sorting result as the parameters to be debugged.

6. The method according to claim 1, wherein The production debugging parameters include material components and usage conditions.

7. The method according to claim 1, wherein The target material is photoresist, the production debugging parameters include material components, the coating rotation speed of the photoresist, and the light exposure dosage, and the product performance parameters include critical dimension, thickness, and aspect ratio.

8. A process debugging device based on transfer learning, comprising: An acquisition unit configured to acquire a machine learning model trained by a material supplier, wherein the machine learning model is used to characterize the functional relationship between production debugging parameters and product performance parameters in the production process of the target material, and the model parameters of the machine learning model include first weight factors for each production debugging parameter; A construction unit configured to construct a training sample set according to the debugging data of the customer side, wherein the training samples in the training sample set include production debugging parameters and product performance parameters; A training unit configured to further train the machine learning model based on the training sample set to obtain second weight factors for each production debugging parameter, wherein the training unit is further configured to: for each production debugging parameter, use the production debugging parameter as the target production debugging parameter, and keep the first weight factors of other production debugging parameters except the target production debugging parameter unchanged; retrain the machine learning model based on the training sample set to obtain a trained model; determine the second weight factor of the target production debugging parameter according to the parameters of the trained model; A determination unit configured to determine a production debugging parameter as a parameter to be debugged from the production debugging parameters according to the first weight factor and the second weight factor of each production debugging parameter; A sending unit configured to send the parameter to be debugged to the material supplier for the material supplier to debug the production process based on the parameter to be debugged.

9. A computing device, comprising a memory and a processor, characterized in that An executable code is stored in the memory, and when the processor executes the executable code, the method described in any one of claims 1-7 is implemented.

Images