CN119962621A - Wafer polishing surface control method and model training method, and related devices - Google Patents
Wafer polishing surface control method and model training method, and related devices Download PDFInfo
- Publication number
- CN119962621A CN119962621A CN202510447256.5A CN202510447256A CN119962621A CN 119962621 A CN119962621 A CN 119962621A CN 202510447256 A CN202510447256 A CN 202510447256A CN 119962621 A CN119962621 A CN 119962621A
- Authority
- CN
- China
- Prior art keywords
- process control
- parameters
- wafer polishing
- polishing surface
- wafers
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Finish Polishing, Edge Sharpening, And Grinding By Specific Grinding Devices (AREA)
Abstract
本申请公开了一种晶圆抛光面型控制方法及其模型训练方法、以及相关装置，其中，晶圆抛光面型控制模型为包括生成器和判别器的生成对抗网络模型，晶圆抛光面型控制模型的训练方法包括：获取历史加工数据；其中，历史加工数据至少包括不同阶段的实际工艺控制参数、抛光前面型曲线数据和期望面型曲线数据；将抛光前面型曲线数据和期望面型曲线数据输入生成器，输出预测工艺控制参数；将预测工艺控制参数和实际工艺控制参数输入判别器，分别输出对预测工艺控制参数和实际工艺控制参数的真假判别结果；基于真假判别结果对晶圆抛光面型控制模型的参数进行迭代更新。通过上述方式，能够提高晶圆抛光面型的控制精度和生产效率。
The present application discloses a wafer polishing surface control method and its model training method, and related devices, wherein the wafer polishing surface control model is a generative adversarial network model including a generator and a discriminator, and the training method of the wafer polishing surface control model includes: obtaining historical processing data; wherein the historical processing data at least includes actual process control parameters, front polishing profile curve data and expected profile curve data at different stages; inputting the front polishing profile curve data and the expected profile curve data into the generator, and outputting predicted process control parameters; inputting the predicted process control parameters and the actual process control parameters into the discriminator, and outputting the true and false discrimination results of the predicted process control parameters and the actual process control parameters respectively; and iteratively updating the parameters of the wafer polishing surface control model based on the true and false discrimination results. Through the above method, the control accuracy and production efficiency of the wafer polishing surface can be improved.
Description
Technical Field
The application relates to the technical field of semiconductor processing, in particular to a wafer polishing surface type control method, a wafer polishing surface type model training method and a related device.
Background
Wafer polishing is a key process in the semiconductor manufacturing process, and aims to remove tiny impurities and unevenness on the surface of a wafer through mechanical polishing and ensure the flatness and smoothness of the wafer. The surface shape of the wafer is an important index reflecting the surface shape of the wafer, wherein GBIR (TTV) refers to the total thickness deviation of the surface shape of the silicon wafer, SFQR refers to the local flatness of the silicon wafer, and the method can be used for judging the edge collapse condition of the polished silicon wafer.
However, the polishing process is a time-varying and highly nonlinear process, the product morphology is influenced by various factors such as working conditions, raw and auxiliary material types and the like, the scheme in the prior art is difficult to describe the association of each process parameter and the surface type change through a mechanism model, the process parameters need to be manually adjusted according to the predicted surface type in the wafer production process, the changes of different materials and different production batches of the product are difficult to cope with, the adaptability to the environment change is poor, and the control precision and the production efficiency of the wafer polishing surface type are seriously influenced.
Disclosure of Invention
The application provides a wafer polishing surface type control method, a model training method thereof and a related device, so as to improve the control precision and the production efficiency of the wafer polishing surface type.
The application provides a training method of a wafer polishing surface type control model, which comprises a generator and a discriminator, wherein the method comprises the steps of obtaining historical processing data, inputting the actual process control parameters, the polishing surface type curve data and the expected surface type curve data in different stages, inputting the polishing surface type curve data and the expected surface type curve data into the generator, outputting predicted process control parameters, inputting the predicted process control parameters and the actual process control parameters into the discriminator, respectively outputting true and false discrimination results of the predicted process control parameters and the actual process control parameters, and iteratively updating parameters of the wafer polishing surface type control model based on the true and false discrimination results.
The method comprises the steps of inputting the polishing front type curve data and the expected surface type curve data into the generator and outputting predicted process control parameters, wherein the step of inputting the polishing front type curve data and the expected surface type curve data into the convolutional neural network to extract polishing characteristic information, and inputting the polishing characteristic information and the auxiliary material life data into the fully-connected neural network to map the polishing characteristic information and the auxiliary material life data to the predicted process control parameters.
The convolutional neural network comprises a plurality of convolutional layers and pooling layers which are alternately connected, wherein the convolutional layers are used for extracting the characteristics of the input polished front surface type curve data and the expected surface type curve data;
And/or the fully-connected neural network comprises a plurality of first fully-connected layers which are sequentially connected, wherein the first fully-connected layers are used for carrying out nonlinear mapping on the polishing characteristic information extracted by the convolutional neural network and finally outputting the prediction process control parameters.
The arbiter comprises a deep neural network, wherein the deep neural network comprises a plurality of second full-connection layers which are sequentially connected, each neuron of each second full-connection layer is connected with all neurons of the second full-connection layer of the previous layer, the number of neurons of the adjacent second full-connection layer is gradually reduced, and the second full-connection layer is used for carrying out nonlinear mapping on the input predicted process control parameters and the actual process control parameters and finally outputting the true and false discrimination results.
The step of carrying out iterative updating on the parameters of the wafer polishing surface type control model based on the true and false judging results comprises the steps of determining a loss function of the generator according to the true and false judging results of the judging device on the predicted process control parameters, determining the loss function of the judging device according to the true and false judging results of the judging device on the predicted process control parameters and the actual process control parameters, and alternately optimizing network parameters of the generator and the judging device according to the loss function of the generator and the loss function of the judging device.
The method comprises the steps of alternately optimizing the network parameters of the generator and the discriminator according to the loss function of the generator and the loss function of the discriminator, wherein the step of alternately optimizing the network parameters of the generator and the discriminator comprises the steps of fixing the network parameters of the generator in the Nth round of training, minimizing the loss function of the discriminator, updating the network parameters of the discriminator through gradient descent, and fixing the network parameters of the discriminator in the (n+1) th round of training, minimizing the loss function of the generator and updating the network parameters of the generator through gradient descent.
The application further provides a wafer polishing surface type control method which is applied to a process control system of wafer polishing equipment, wherein the process control system is integrated with a generator of a wafer polishing surface type control model, the wafer polishing surface type control model is obtained through training of the training method of the wafer polishing surface type control model, the wafer polishing surface type control method comprises the steps of obtaining processing data corresponding to a current wafer when the current wafer is polished, the processing data corresponding to the current wafer at least comprise polishing front type curve data and expected surface type curve data of the current wafer, inputting the polishing front type curve data and expected surface type curve data of the current wafer into the generator, outputting predicted process control parameters of the current wafer, generating and sending optimized control instructions to the process control system based on the predicted process control parameters of the current wafer, and adjusting the processing parameters of the wafer according to the optimized control instructions by the process control system.
The method comprises the steps of inputting polishing front-type curve data and expected surface-type curve data of current batch wafers into the generator and outputting predicted process control parameters of the current batch wafers, wherein the step of inputting the polishing front-type curve data and the expected surface-type curve data of the current batch wafers into the convolutional neural network to extract polishing characteristic information, and the step of inputting the polishing characteristic information and the auxiliary material life data corresponding to the current batch wafers into the fully connected neural network to map the polishing characteristic information and the auxiliary material life data corresponding to the current batch wafers to the predicted process control parameters of the current batch wafers.
The wafer polishing surface type control method further comprises the steps of obtaining historical processing data corresponding to a previous wafer before polishing a next wafer, updating parameters of the wafer polishing surface type control model according to the historical processing data corresponding to the previous wafer, and calling a generator after parameter updating by the process control system to control polishing of the next wafer.
The training device comprises a first acquisition module, a first updating module and a first processing module, wherein the first acquisition module is used for acquiring historical processing data, the historical processing data at least comprise actual process control parameters, polishing front curve data and expected surface curve data in different stages, the first processing module is used for inputting the polishing front curve data and the expected surface curve data into the generator to output predicted process control parameters, the predicted process control parameters and the actual process control parameters are input into the discriminator to respectively output true and false discrimination results of the predicted process control parameters and the actual process control parameters, and the first updating module is used for iteratively updating parameters of the wafer polishing front curve control model based on the true and false discrimination results.
The application further provides a wafer polishing surface type control device, which is used for controlling wafer polishing equipment to polish wafers, wherein the wafer polishing surface type control device is integrated with a generator of a wafer polishing surface type control model, the wafer polishing surface type control model is obtained through training of the wafer polishing surface type control model by any one of the training methods, the wafer polishing surface type control device comprises a second acquisition module, the second acquisition module is used for acquiring processing data corresponding to the wafers in the current batch when the wafers in the current batch are polished, the processing data corresponding to the wafers in the current batch at least comprise polishing surface type curve data and expected surface type curve data of the wafers in the current batch, the second processing module is used for inputting the polishing surface type curve data and the expected surface type curve data of the wafers in the current batch into the generator and outputting predicted process parameters of the wafers in the current batch, and the second processing module is used for optimizing processing parameters of the wafers in the current batch according to a predicted process parameter updating command when the wafer in the current batch is updated, and the processing module is used for optimizing processing parameters in the current wafer processing equipment.
In order to solve the technical problems, the application also provides electronic equipment which comprises a memory and a processor which are mutually coupled, wherein the processor is used for executing program instructions stored in the memory so as to realize the training method of the wafer polishing surface type control model and/or the wafer polishing surface type control method.
To solve the above technical problem, the present application further provides a computer readable storage medium storing program instructions that can be executed to implement a training method of a wafer polishing surface type control model and/or a wafer polishing surface type control method as described in any one of the above.
The wafer polishing surface type control model has the advantages that the wafer polishing surface type control model is different from the situation of the prior art, the wafer polishing surface type control model comprises a generator and a discriminator, historical processing data are firstly obtained in the process of training the wafer polishing surface type control model, the historical processing data at least comprise actual process control parameters, polishing front surface type curve data and expected surface type curve data in different stages, then the polishing front surface type curve data and the expected surface type curve data are input into the generator to output predicted process control parameters, the predicted process control parameters and the actual process control parameters are input into the discriminator to respectively output true and false discrimination results of the predicted process control parameters and the actual process control parameters, and then the parameters of the wafer polishing surface type control model can be iteratively updated based on the true and false discrimination results. The application discloses a method for manufacturing a wafer polishing surface model, which comprises the steps of constructing a generating countermeasure network model comprising a generator and a discriminator, training the generating countermeasure network model by adopting historical processing data, wherein the historical processing data at least comprises actual process control parameters of different stages, polishing front surface profile data and expected surface profile data, then the generator can generate predicted process control parameters according to the polishing front surface profile data and the expected surface profile data of different stages, the discriminator is used for discriminating the difference between the generated predicted process control parameters and the actual process control parameters of corresponding stages, the discriminator can maximize the classification accuracy of the actual process control parameters and the generated predicted process control parameters by the countermeasure training of the generator and the discriminator, and the generator can minimize the difference between the generated predicted process control parameters and the actual process control parameters, so that after the wafer polishing surface profile control model is trained, when polishing any wafer, the wafer front surface profile curve data, the expected surface profile curve data and the wafer polishing front surface profile control data of the wafer can be utilized, the wafer can be accurately controlled according to the expected surface profile control parameters of the wafer, the wafer can be controlled according to the shape of the expected wafer, the wafer can be controlled according to the wafer polishing surface profile model, the wafer can be accurately processed according to the wafer control surface profile model, and the wafer can be controlled according to the expected wafer surface profile model, and the wafer can be manufactured by the wafer model after the wafer polishing surface model is accurately processed, thereby improving the precision and production efficiency of the polishing process.
Drawings
FIG. 1 is a flowchart of a first embodiment of a training method for a wafer polishing surface type control model according to the present application;
FIG. 2 is a flowchart of a second embodiment of a training method for a wafer polishing surface type control model according to the present application;
FIG. 3 is a schematic diagram of a control model of a polishing surface of a wafer in an application scenario of the present application;
FIG. 4 is a flowchart illustrating an embodiment of step S25 in FIG. 2;
FIG. 5 is a schematic diagram of a training method of a wafer polishing surface type control model in an application scenario of the present application;
fig. 6 is a schematic flow chart of a first embodiment of a wafer polishing surface type control method according to the present application;
Fig. 7 is a schematic flow chart of a second embodiment of a wafer polishing surface type control method according to the present application;
FIG. 8 is a schematic diagram of a method for controlling the polishing surface type of a wafer in an application scenario of the present application;
FIG. 9 is a schematic diagram of an embodiment of a training apparatus for a wafer polishing surface control model according to the present application;
FIG. 10 is a schematic view of an embodiment of a control apparatus for polishing a wafer according to the present application;
FIG. 11 is a schematic diagram of an embodiment of an electronic device according to the present application;
Fig. 12 is a schematic structural diagram of an embodiment of a computer readable storage medium provided by the present application.
Detailed Description
The following describes embodiments of the present application in detail with reference to the drawings.
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, interfaces, techniques, etc., in order to provide a thorough understanding of the present application.
The terms "system" and "network" are often used interchangeably herein. The term "and/or" is merely an association relationship describing the associated object, and means that three relationships may exist, for example, a and/or B may mean that a exists alone, while a and B exist together, and B exists alone. In addition, the character "/" herein generally indicates that the front and rear associated objects are an "or" relationship. Further, "a plurality" herein means two or more than two.
Referring to fig. 1, fig. 1 is a flowchart of a first embodiment of a training method for a wafer polishing surface type control model according to the present application. The wafer polishing surface type control model in the embodiment of the application is a generation countermeasure network model comprising a generator and a discriminator, and the training method of the wafer polishing surface type control model in the embodiment comprises the following steps:
And S11, acquiring historical processing data, wherein the historical processing data at least comprise actual process control parameters, polishing front surface type curve data and expected surface type curve data of different stages.
The wafer in the embodiment of the application can comprise workpieces such as a silicon wafer, a sapphire wafer, a silicon carbide wafer and the like, and the wafer polishing surface type control model can be used as a training sample by collecting historical processing data of wafer polishing equipment. The historical processing data comprises process control parameters and wafer surface profile data of different stages in the wafer processing process, wherein the process control parameters can comprise the rotating speed and the feeding speed of a grinding and polishing disc, the inclination angle of the grinding and polishing disc, the temperature parameter in the grinding process, the thickness measuring parameter of a sensor, the pressure parameter applied to the wafer, the polishing time and the like, and the process control parameters can influence the surface profile of the polished wafer. The wafer surface profile curve data comprise wafer polishing front profile curve data and expected surface profile curve data, wherein the wafer polishing front profile curve data refer to wafer surface shapes before polishing, are expressed as two-dimensional curve images, the main information contained in the wafer surface profile curve data is changes of wafer surface heights or shapes, the surface profile characteristics before polishing can be reflected, the expected surface profile curve data refer to target surface shapes of the wafer, namely, the surface profile expected to be achieved after ideal polishing treatment of the wafer, and are expressed as two-dimensional curve images.
And step S12, inputting the polishing front surface type curve data and the expected surface type curve data into the generator, and outputting predicted process control parameters.
It will be appreciated that, after the pre-polishing profile data and the corresponding desired profile data of the wafer are input into the generator, the generator is capable of simultaneously processing the pre-polishing profile data and the desired profile data, and generating the predicted process control parameters based on these inputs, i.e., the generator presumes that the wafer having the pre-polishing profile data is polished using the predicted process control parameters generated by the generator, so that the polished wafer has the corresponding desired profile data.
And S13, inputting the predicted process control parameters and the actual process control parameters into the discriminator, and respectively outputting the true and false discrimination results of the predicted process control parameters and the actual process control parameters.
It can be understood that the discriminator needs to learn the features extracted from the real data and the generated data of the generator and finally give out a distinguishing probability of the real data and the generated data, so that after the predicted process control parameter and the actual process control parameter are input into the discriminator, the predicted process control parameter is taken as the generated data of the generator, the discriminator can give out a true and false discriminating result of whether the predicted process control parameter is the real data, and likewise, the actual process control parameter is taken as the real data, and the discriminator can give out a true and false discriminating result of whether the actual process control parameter is the real data.
And S14, carrying out iterative updating on the parameters of the wafer polishing surface type control model based on the true and false discrimination result.
It can be understood that the predicted process control parameters generated by the generator for generating the countermeasure network are to be confused with the discriminators as much as possible, and the discriminators are to distinguish the predicted process control parameters and the actual process control parameters generated by the generator as much as possible, when performing countermeasure optimization training, the parameters of the optimized wafer polishing surface type control model are continuously updated according to whether the discriminators are true and false discrimination results of the predicted process control parameters and the actual process control parameters of the real data, so as to minimize the loss between the predicted value and the real output value of the model as much as possible, and the smaller the loss between the predicted value and the real output value is, the closer the model prediction is to the real value is.
According to the scheme, the generated countermeasure network model is used as the wafer polishing surface type control model by constructing the generated countermeasure network model comprising the generator and the discriminator, the historical processing data is adopted to train the generated countermeasure network model, and as the historical processing data at least comprises actual process control parameters, polishing front type curve data and expected surface type curve data of different stages, the generator can generate predicted process control parameters according to the polishing front type curve data and the expected surface type curve data of different stages, the discriminator is used for discriminating the difference between the generated predicted process control parameters and the actual process control parameters obtained at corresponding stages, and the discriminator can maximize the classification accuracy of the actual process control parameters and the generated predicted process control parameters through countermeasure training of the generator and the discriminator, and the generator can minimize the difference between the generated predicted process control parameters and the actual process control parameters; it can be understood that after the training of the wafer polishing surface type control model is finished, when polishing any batch of wafers, the polishing surface type curve data, the expected surface type curve data and the generator of the wafer polishing surface type control model of the batch of wafers can be utilized to generate the corresponding process control parameters of the batch of wafers, and after the process control parameters are adopted for processing, the batch of wafers can obtain the processing result close to the expected surface type curve data, the wafer polishing surface type control model adopts the end-to-end design scheme, inputs the polishing surface type curve data and the expected surface type curve data, outputs the process control parameters capable of obtaining the expected surface type curve data, namely, the control quantity is directly obtained by the current quantity and the expected quantity, and the whole process is a unified model, therefore, in the wafer polishing process, the trained wafer polishing surface type control model can be used for accurately adjusting the process control parameters according to the surface type data of wafers in any batch, and the final surface shape of the wafers is ensured to meet the expected target, so that the precision and the production efficiency of the polishing process can be improved.
Referring to fig. 2, fig. 2 is a flowchart illustrating a second embodiment of a training method for a wafer polishing surface type control model according to the present application. The training method of the wafer polishing surface type control model in the embodiment comprises the following steps:
and S21, acquiring historical processing data, wherein the historical processing data at least comprise actual process control parameters, polishing front surface type curve data, expected surface type curve data and auxiliary material service life data of different stages.
It will be appreciated that the lifetime of the polishing liquid, polishing pad, etc. has a certain effect on the polishing surface type of the wafer, and thus, the difference from the previous embodiment is that the historical processing data in this embodiment further includes lifetime data of the auxiliary materials at different stages. The auxiliary material life data is life information of polishing liquid, polishing pad and other auxiliary materials, wherein the life information comprises time series data such as the use condition, abrasion life and the like of each auxiliary material, and the auxiliary material life data is expressed in a one-dimensional vector form.
In other embodiments, the historical process data may also include randomly generated noise data of the same dimensions as the profile curve, it being understood that the addition of noise to the model may make the model more challenging and generalizing.
And S22, inputting the polishing front type curve data and the expected surface type curve data into the convolutional neural network to extract polishing characteristic information.
And S23, inputting the polishing characteristic information and the auxiliary material service life data into the fully-connected neural network so as to map the polishing characteristic information and the auxiliary material service life data to the predicted process control parameters.
Referring to fig. 3, the generator of the wafer polishing surface type control model in this embodiment includes a convolutional neural network and a fully-connected neural network connected. The combined structure of the convolutional neural network and the fully-connected neural network can process the polishing front surface type curve data, the expected surface type curve data and the auxiliary material service life data simultaneously, and generate predicted process control parameters based on the inputs.
In one embodiment, the convolutional neural network comprises a plurality of convolutional layers and a pooling layer which are alternately connected, wherein the convolutional layers are used for extracting characteristics of the input polished front surface type curve data and the expected front surface type curve data, and the pooling layer is used for pooling the characteristics extracted by the convolutional layers.
Specifically, the convolution layers perform feature extraction on the input image data (i.e., the pre-polished profile data and the desired profile data in the present application) through a sliding window, and each convolution layer performs a convolution operation on the input image data through a set of convolution kernels (filters) to extract local features. Let the input image data of the first layer beThe convolution kernel isThe bias term isThe convolution operation may be expressed as:
。
Wherein, represents convolution operation, reLU is an activation function, and the form of ReLU function is that 。
The pooling layer is used for reducing and ensuring the stability of the extracted local features. In the embodiment of the application, a pooling mode of maximum pooling (Max Pooling) and average pooling (Average Pooling) can be adopted, for example, the pooling operation can be performed by window sliding, and the maximum value or the average value of each local area is taken, so that the size of the extracted feature map is reduced. The pooling operation can be expressed as:
。
It can be understood that after the polishing front-type curve data and the expected front-type curve data are subjected to convolution and pooling operations for many times, high-dimensional polishing characteristic information can be extracted, as shown in fig. 3, a flattening layer can be further arranged between the convolution neural network and the fully-connected neural network to serve as a transition layer, the flattening layer can flatten the high-dimensional characteristic extracted by the convolution neural network into one-dimensional data, namely one-dimensional polishing characteristic information is obtained, then the one-dimensional polishing characteristic information is input into the fully-connected neural network together with auxiliary material life data expressed in a one-dimensional vector form, and the polishing characteristic information extracted by the convolution neural network and the directly-input auxiliary material life data are mapped into the generated predicted process control parameters by the fully-connected neural network.
In one embodiment, the fully-connected neural network comprises a plurality of first fully-connected layers which are sequentially connected, wherein the first fully-connected layers are used for carrying out nonlinear mapping on polishing characteristic information extracted by the convolutional neural network and finally outputting predicted process control parameters.
The full-connection neural network carries out nonlinear mapping on the high-dimensional polishing characteristic information extracted by the convolutional neural network through the first full-connection layer, and finally the generated prediction process control parameters can be output. As shown in fig. 3, each layer of the fully-connected neural network is composed of a plurality of neurons, each of which is connected to all neurons of the upper layer. Assume that the input of the k-th first full connection layer isThe weight isBiased toOutput of the kth layerThe calculation formula is as follows:
。
The final output layer of the fully-connected neural network is also a first fully-connected layer for generating predicted process control parameters. Assuming that the final output is Representing the generated predicted process control parameters:
。
wherein, Is the weight of the output layer and,Is the output of the upper layer.
And step S24, inputting the predicted process control parameters and the actual process control parameters into the discriminator, and respectively outputting the true and false discrimination results of the predicted process control parameters and the actual process control parameters.
Referring to fig. 3, in an embodiment, the arbiter includes a deep neural network, where the deep neural network includes a plurality of second full-connection layers sequentially connected, each neuron of each second full-connection layer is connected to all neurons of a second full-connection layer above, and the number of neurons of adjacent second full-connection layers is gradually reduced, and the second full-connection layer is configured to perform nonlinear mapping on an input predicted process control parameter and an actual process control parameter, and finally output a true/false discrimination result.
It will be appreciated that the deep neural network is the core of the arbiter, and that each neuron of each second fully connected layer in the deep neural network is connected to all neurons of the previous layer. The discriminator gradually extracts advanced features of the input data through a plurality of second full connection layers, and finally performs true and false classification. Similar to the convolutional neural network in the generator, the input of the first layer of the deep neural network is set asThe weight isThe bias term isOutput of the first layerCalculated from the following formula:
。
wherein the ReLU activation function is used to introduce nonlinear transformations to enhance the expressive power of the network.
Thus, the arbiter comprises a plurality of second fully connected layers, after each of which a ReLU activation function is applied to further capture the non-linear relationship in the input data. The number of neurons per second fully connected layer will gradually decrease until the last layer outputs a single probability value. Assuming that the arbiter has L layers, the output of the last layer isThe values of the output layer obtained by linear transformation are:
。
The output layer of the arbiter uses a sigmoid activation function to map the final output to a probability space representing the probability value that the input data is real data. Assuming that the final output is The output formula is:
。
wherein, Is a sigmoid activation function, compresses the output value to the [0,1] interval, and represents the probability that the data is true.
And S25, carrying out iterative updating on the parameters of the wafer polishing surface type control model based on the true and false discrimination result.
Referring to fig. 4, fig. 4 is a flowchart illustrating an embodiment of step S25 in fig. 2. In an embodiment, the step S25 specifically includes:
step S251, determining the loss function of the generator according to the true and false judging result of the predicted process control parameter by the discriminator, and determining the loss function of the discriminator according to the true and false judging result of the predicted process control parameter and the actual process control parameter by the discriminator.
It will be appreciated that the core of generating the countermeasure network model is its loss function, which generally includes two parts, a generator loss function and a arbiter loss function. Let the predicted process control parameter generated by the generator be G (z), the actual process control parameter be x, and the output of the discriminator be D (x) and D (G (z)), where D (x) represents the output probability of the discriminator for determining the actual process control parameter, and D (G (z)) represents the output probability of the discriminator for determining the predicted process control parameter generated by the generator. Wherein the generator loss function LG may be expressed as:
。
the optimization objective of the arbiter is to maximize its classification accuracy for the actual process control parameters and the predicted process control parameters. To this end, the arbiter may calculate the loss using a Cross entropy loss function (Binary Cross-Entropy Loss), the equation for the arbiter's loss function is as follows:
。
Wherein pdata represents the true data distribution and pz is the noise distribution of the generator input.
Step S252, alternately optimizing network parameters of the generator and the discriminator according to the loss function of the generator and the loss function of the discriminator.
The training for generating the countermeasure network model is a process of countermeasure training of the generator and the discriminator, and the optimization target is to optimize the loss function of the generator and the loss function of the discriminator through gradient descent.
In one embodiment, the network parameters of the generator are fixed during the Nth round of training, the loss function of the discriminator is minimized, the network parameters of the discriminator are updated through gradient descent, and the network parameters of the discriminator are fixed during the (n+1) th round of training, the loss function of the generator is minimized, and the network parameters of the generator are updated through gradient descent.
Referring to fig. 5, fig. 5 is a schematic diagram of a training method of a wafer polishing surface type control model in an application scenario of the present application. As shown in the figure, in the training process of one round, after the historical processing data is input into the generator, the generator can generate predicted process control parameters, then the predicted process control parameters and actual process control parameters are input into the discriminator for training, when the discriminator is trained, the parameters of the generator are required to be fixed, the parameters of the discriminator are updated by using a back propagation algorithm so as to minimize the loss function of the discriminator, in the training process of the next round, the parameters of the discriminator are required to be fixed, and the parameters of the generator are also required to be updated by using the back propagation algorithm so as to minimize the loss function of the generator. In the optimization process of the discriminator, the loss function LD of the discriminator needs to be minimized, and parameters of the discriminator are updated through gradient descent: . In the optimization process of the generator, the loss function LG of the generator needs to be minimized, and the parameters of the generator are updated through gradient descent: . The wafer can obtain the processing result which is close to the expected surface profile curve data through the countermeasure training of the generator and the discriminator and the prediction process control parameters generated by the generator.
It will be appreciated that the training goal of the wafer polishing surface type control model is to minimize the difference between the generated predicted process control parameters and the actual process control parameters. The measurement can be made by the following loss function:
。
wherein, Is the predicted process control parameter generated, ytarget is the actual data, and the loss function Lfinal evaluates the difference between the generated result and the desired result by minimizing the euclidean distance.
Referring to fig. 6, fig. 6 is a flowchart illustrating a first embodiment of a wafer polishing surface type control method according to the present application. The wafer polishing surface type control method is applied to a process control system of wafer polishing equipment, the process control system is integrated with a generator of a wafer polishing surface type control model, and the wafer polishing surface type control model is obtained through training by the training method of the wafer polishing surface type control model in any embodiment. Specifically, the wafer polishing surface type control method in this embodiment includes the following steps:
and step S61, when polishing the current batch of wafers, acquiring processing data corresponding to the current batch of wafers, wherein the processing data corresponding to the current batch of wafers at least comprises polishing front surface type curve data and expected surface type curve data of the current batch of wafers.
And S62, inputting the polishing front surface type curve data and the expected surface type curve data of the current batch of wafers into the generator, and outputting the predicted process control parameters of the current batch of wafers.
And step 63, generating and sending an optimized control instruction to the process control system based on the predicted process control parameters of the current batch of wafers.
And S64, the process control system adjusts the processing parameters of the wafer polishing equipment according to the optimized control instruction so as to polish the wafers in the current batch.
It is understood that the trained wafer polishing surface type control model may be used for wafer polishing surface type control, wherein the generator is integrated into a process control system of the wafer polishing apparatus. At this time, the generator does not rely solely on the historical processing data, but can generate a dynamically optimized control instruction according to the processing data corresponding to the latest input current batch of wafers. The generator can receive input signals from the process control system in real time, wherein the input signals of the process control system comprise processing data corresponding to the current batch of wafers, such as polishing front surface type curve data and expected surface type curve data of the current batch of wafers, so that the generator can output predicted process control parameters of the current batch of wafers in real time according to the signals, generate optimized control instructions and send the optimized control instructions to the process control system, for example, the optimized control instructions can adjust parameters of the wafer polishing equipment to predicted process control parameters of the current batch of wafers in real time, and the process control system can adjust the processing parameters of the wafer polishing equipment according to the optimized control instructions, and polish the current batch of wafers by adopting the predicted process control parameters so as to achieve that the surface types of the current batch of wafers are close to the expected surface type curve data and meet expected targets.
Therefore, the trained wafer polishing surface type control model is deployed in a process control system of the wafer polishing equipment, and real-time control of polishing process control parameters can be realized.
Referring to fig. 7, fig. 7 is a flowchart illustrating a second embodiment of a wafer polishing surface type control method according to the present application. The wafer polishing surface type control method in the embodiment comprises the following steps:
And step 71, acquiring processing data corresponding to the current batch of wafers when polishing the current batch of wafers, wherein the processing data corresponding to the current batch of wafers at least comprises polishing front type curve data, expected surface type curve data and auxiliary material service life data of the current batch of wafers.
It will be appreciated that the lifetime of the auxiliary materials such as polishing liquid and polishing pad has a certain influence on the polishing surface type of the wafer, so that the difference from the previous embodiment is that the processing data corresponding to the current batch of wafers in the present embodiment further includes lifetime data of the auxiliary materials corresponding to the current batch of wafers.
And S72, inputting the polishing front surface type curve data and the expected surface type curve data of the current batch of wafers into the convolutional neural network to extract polishing characteristic information.
And step 73, inputting the polishing characteristic information and the auxiliary material service life data corresponding to the current batch of wafers into the fully-connected neural network so as to map the polishing characteristic information and the auxiliary material service life data corresponding to the current batch of wafers to the predicted process control parameters of the current batch of wafers.
The generator of the wafer polishing surface type control model in the embodiment comprises a convolution neural network and a full-connection neural network which are connected, and the combination structure of the convolution neural network and the full-connection neural network can simultaneously process polishing surface type curve data, expected surface type curve data and auxiliary material service life data corresponding to the current batch of wafers and generate predicted process control parameters of the current batch of wafers based on the inputs.
And step S74, generating and sending an optimized control instruction to the process control system based on the predicted process control parameters of the current batch of wafers.
And step S75, the process control system adjusts the processing parameters of the wafer polishing equipment according to the optimized control instruction so as to polish the wafers in the current batch.
Step S74 and step S75 in this embodiment are substantially the same as step S63 and step S64 in the above embodiment, and are not described here again.
In an embodiment, the wafer polishing surface type control method further includes the following steps:
Step S76, before polishing the next wafer, acquiring historical processing data corresponding to the last wafer.
And step 77, updating parameters of the wafer polishing surface type control model according to the historical processing data corresponding to the previous wafer.
And step S78, the process control system calls the generator with updated parameters to carry out polishing control on the wafers in the next batch.
Referring to fig. 8, fig. 8 is a schematic diagram of a method for controlling a polishing surface type of a wafer in an application scenario of the present application. When the trained wafer polishing surface type control model is deployed in a process control system of wafer polishing equipment, the wafer polishing surface type control model can learn processing data of the last polishing before polishing each time, for example, the wafer polishing surface type control model can learn the latest processing data of the K time before polishing the K+1st time, update parameters of the wafer polishing surface type control model, then when polishing the K+1st time, the process control system calls a generator after parameter update to generate predicted process control parameters of a wafer corresponding to the K+1st time, generates and sends optimized control instructions to the process control system, so that the generated optimized control instructions can adapt to continuous changes of production environments, and the accuracy of the process control parameters of the K+1st time of polishing is ensured. The closed-loop feedback mechanism enables the polishing control strategy of each production process to be continuously optimized, and high-efficiency and accurate control strategy updating is realized in a dynamic and complex production environment.
In the embodiment of the application, in order to ensure that the generator is continuously optimized, the feedback of the discriminator needs to be timely transmitted to the generator for gradient update to form R2R control, namely, the future control strategy is adjusted and optimized through the history and real-time feedback information, the generator not only depends on static process control parameters, but also can flexibly adjust control signals through learning the history and real-time data, and is suitable for fluctuation of the production environment.
Referring to fig. 9, fig. 9 is a schematic structural diagram of an embodiment of a training apparatus for a wafer polishing surface type control model according to the present application. The training device 90 of the wafer polishing surface type control model in this embodiment includes a first acquisition module 900, a first processing module 902, and a first update module 904 that are connected to each other. The first obtaining module 900 is configured to obtain historical processing data, where the historical processing data includes at least actual process control parameters, pre-polishing profile data, and desired profile data at different stages. The first processing module 902 is configured to input the polishing front profile data and the desired front profile data into the generator, output a predicted process control parameter, input the predicted process control parameter and the actual process control parameter into the discriminator, and output a true and false discrimination result for the predicted process control parameter and the actual process control parameter, respectively. The first updating module 904 is configured to iteratively update parameters of the wafer polishing surface type control model based on the true and false discrimination result.
In one embodiment, the generator comprises a convolutional neural network and a fully-connected neural network connected, and the historical processing data further comprises adjuvant life data at different stages. The first processing module 902 performs the steps of inputting the pre-polishing profile data and the desired profile data to the generator and outputting predicted process control parameters, including inputting the pre-polishing profile data and the desired profile data to the convolutional neural network to extract polishing characteristic information, and inputting the polishing characteristic information and the accessory lifetime data to the fully-connected neural network to map the polishing characteristic information and the accessory lifetime data to the predicted process control parameters.
In one embodiment, the first update module 904 performs the step of iteratively updating parameters of the wafer polishing surface type control model based on the true and false discrimination results, including determining a loss function of the generator based on the true and false discrimination results of the discriminator for the predicted process control parameters, determining a loss function of the discriminator based on the true and false discrimination results of the discriminator for the predicted process control parameters and the actual process control parameters, and alternately optimizing network parameters of the generator and the discriminator based on the loss function of the generator and the loss function of the discriminator.
In one embodiment, the first updating module 904 performs the step of alternately optimizing the network parameters of the generator and the arbiter according to the loss function of the generator and the loss function of the arbiter, and specifically includes fixing the network parameters of the generator during the nth round of training, minimizing the loss function of the arbiter, updating the network parameters of the arbiter through gradient descent, and fixing the network parameters of the arbiter during the n+1 th round of training, minimizing the loss function of the generator, and updating the network parameters of the generator through gradient descent.
Referring to fig. 10, fig. 10 is a schematic structural diagram of an embodiment of a wafer polishing surface type control device according to the present application. The wafer polishing surface type control device 100 in this embodiment is used for controlling a wafer polishing apparatus to polish a wafer, where the wafer polishing surface type control device 100 is integrated with a generator of a wafer polishing surface type control model, and the wafer polishing surface type control model is obtained by training the wafer polishing surface type control model in any one of the foregoing embodiments.
The wafer polishing surface type control device 100 includes a second acquisition module 1000, a second processing module 1002, and a second updating module 1004 that are connected to each other. The second obtaining module 1000 is configured to obtain processing data corresponding to a current batch of wafers when polishing the current batch of wafers, where the processing data corresponding to the current batch of wafers at least includes polishing front profile curve data and expected front profile curve data of the current batch of wafers. The second processing module 1002 is configured to input the polishing front profile data and the expected front profile data of the current batch wafer into the generator, output the predicted process control parameters of the current batch wafer, and generate the optimized control instruction based on the predicted process control parameters of the current batch wafer. The second updating module 1004 is configured to adjust a processing parameter of the wafer polishing apparatus according to the optimized control instruction, so as to perform polishing processing on the current batch of wafers.
In an embodiment, the generator includes a convolutional neural network and a fully connected neural network, and the processing data corresponding to the current batch of wafers further includes auxiliary material lifetime data corresponding to the current batch of wafers. The second processing module 1002 performs the steps of inputting polishing front profile curve data and expected front profile curve data of the current batch of wafers into the generator, and outputting predicted process control parameters of the current batch of wafers, including inputting the polishing front profile curve data and the expected front profile curve data of the current batch of wafers into the convolutional neural network to extract polishing characteristic information, and inputting the polishing characteristic information and auxiliary material lifetime data corresponding to the current batch of wafers into the fully-connected neural network to map the polishing characteristic information and the auxiliary material lifetime data corresponding to the current batch of wafers to the predicted process control parameters of the current batch of wafers.
In an embodiment, the second processing module 1002 is further configured to obtain historical processing data corresponding to a previous wafer before polishing a next wafer, update parameters of the wafer polishing surface type control model according to the historical processing data corresponding to the previous wafer, and call the generator after parameter update by the process control system to perform polishing control on the next wafer.
Referring to fig. 11, fig. 11 is a schematic structural diagram of an electronic device according to an embodiment of the application. The electronic device 110 in this embodiment includes a processor 1102 and a memory 1101 that are connected to each other, where the memory 1101 is configured to store program instructions, and the processor 1102 is configured to execute the program instructions stored in the memory 1101 to implement the steps of the training method embodiment of any one of the wafer polishing surface type control models and/or the steps of any one of the wafer polishing surface type control method embodiments. In one particular implementation scenario, electronic device 110 may comprise, but is not limited to, a microcomputer, a server.
Specifically, the processor 1102 is configured to control itself and the memory 1101 to implement the steps of any of the foregoing training method embodiments of the wafer polishing surface type control model, and/or the steps of any of the foregoing wafer polishing surface type control method embodiments. The processor 1102 may also be referred to as a CPU (Central Processing Unit ). The processor 1102 may be an integrated circuit chip with signal processing capabilities. The Processor 1102 may also be a general purpose Processor, a digital signal Processor (DIGITAL SIGNAL Processor, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), field-Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. In addition, the processor 1102 may be commonly implemented by an integrated circuit chip.
Referring to fig. 12, fig. 12 is a schematic diagram of a computer readable storage medium according to an embodiment of the application. The computer readable storage medium 120 of the present application has stored thereon program instructions 1200, which when executed by a processor, implement the steps of any one of the foregoing wafer polishing surface type control model training method embodiments and/or the steps of any one of the foregoing wafer polishing surface type control method embodiments.
The computer readable storage medium 120 may be a medium such as a usb (universal serial bus), a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, which may store the program instructions 1200, or may be a server storing the program instructions 1200, and the server may send the stored program instructions 1200 to another device for execution, or may also self-execute the stored program instructions 1200.
In the several embodiments provided in the present application, it should be understood that the disclosed method, apparatus and device may be implemented in other manners. For example, the above-described apparatus and device embodiments are merely illustrative, e.g., the division of modules or units is merely a logical functional division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute all or part of the steps of the methods of the embodiments of the present application. The storage medium includes a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes.
If the technical scheme of the application relates to personal information, the product applying the technical scheme of the application clearly informs the personal information processing rule before processing the personal information and obtains the autonomous agreement of the individual. If the technical scheme of the application relates to sensitive personal information, the product applying the technical scheme of the application obtains individual consent before processing the sensitive personal information, and simultaneously meets the requirement of 'explicit consent'. For example, a clear and obvious mark is set at a personal information acquisition device such as a camera to inform that the personal information acquisition range is entered, personal information is acquired, if the personal voluntarily enters the acquisition range, the personal information is considered as consent to acquire the personal information, or if a clear mark/information is used on a personal information processing device to inform that the personal information processing rule is used, personal authorization is obtained through popup information or a mode of requesting the personal information to upload the personal information by the personal, wherein the personal information processing rule can comprise information such as a personal information processor, a personal information processing purpose, a processing mode, a processed personal information type and the like.
Claims (13)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510447256.5A CN119962621B (en) | 2025-04-09 | 2025-04-09 | Wafer polishing surface control method and model training method, and related devices |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510447256.5A CN119962621B (en) | 2025-04-09 | 2025-04-09 | Wafer polishing surface control method and model training method, and related devices |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119962621A true CN119962621A (en) | 2025-05-09 |
| CN119962621B CN119962621B (en) | 2025-07-15 |
Family
ID=95591217
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510447256.5A Active CN119962621B (en) | 2025-04-09 | 2025-04-09 | Wafer polishing surface control method and model training method, and related devices |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119962621B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6540591B1 (en) * | 2001-04-18 | 2003-04-01 | Alexander J. Pasadyn | Method and apparatus for post-polish thickness and uniformity control |
| KR20210017431A (en) * | 2019-08-08 | 2021-02-17 | 연세대학교 산학협력단 | Apparatus and method for generating control data of polishing process and control device comprising the same |
| US20210303757A1 (en) * | 2020-03-30 | 2021-09-30 | Kla Corporation | Semiconductor fabrication process parameter determination using a generative adversarial network |
| CN116305658A (en) * | 2023-03-30 | 2023-06-23 | 上海天岳半导体材料有限公司 | Method for constructing wafer processing control model |
| CN117032111A (en) * | 2023-08-10 | 2023-11-10 | 江苏道达智能科技有限公司 | Error control method for factory automation production line |
| CN117601013A (en) * | 2023-12-22 | 2024-02-27 | 华海清科(北京)科技有限公司 | Polishing pressure control method and device, electronic equipment and wafer polishing system |
| US20240066657A1 (en) * | 2022-08-23 | 2024-02-29 | Kctech Co., Ltd. | Control method, control apparatus, training method, and training apparatus based on thickness estimation of wafer substrate |
| CN119337804A (en) * | 2023-07-20 | 2025-01-21 | 三星电子株式会社 | Device, method and system for transforming measurement data |
| CN119719568A (en) * | 2024-12-10 | 2025-03-28 | 盖泽精密科技(苏州)有限公司 | Wafer film thickness calculation method and device based on GAN and adaptive feedback mechanism |
-
2025
- 2025-04-09 CN CN202510447256.5A patent/CN119962621B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6540591B1 (en) * | 2001-04-18 | 2003-04-01 | Alexander J. Pasadyn | Method and apparatus for post-polish thickness and uniformity control |
| KR20210017431A (en) * | 2019-08-08 | 2021-02-17 | 연세대학교 산학협력단 | Apparatus and method for generating control data of polishing process and control device comprising the same |
| US20210303757A1 (en) * | 2020-03-30 | 2021-09-30 | Kla Corporation | Semiconductor fabrication process parameter determination using a generative adversarial network |
| US20240066657A1 (en) * | 2022-08-23 | 2024-02-29 | Kctech Co., Ltd. | Control method, control apparatus, training method, and training apparatus based on thickness estimation of wafer substrate |
| CN116305658A (en) * | 2023-03-30 | 2023-06-23 | 上海天岳半导体材料有限公司 | Method for constructing wafer processing control model |
| CN119337804A (en) * | 2023-07-20 | 2025-01-21 | 三星电子株式会社 | Device, method and system for transforming measurement data |
| CN117032111A (en) * | 2023-08-10 | 2023-11-10 | 江苏道达智能科技有限公司 | Error control method for factory automation production line |
| CN117601013A (en) * | 2023-12-22 | 2024-02-27 | 华海清科(北京)科技有限公司 | Polishing pressure control method and device, electronic equipment and wafer polishing system |
| CN119719568A (en) * | 2024-12-10 | 2025-03-28 | 盖泽精密科技(苏州)有限公司 | Wafer film thickness calculation method and device based on GAN and adaptive feedback mechanism |
Non-Patent Citations (2)
| Title |
|---|
| HASNAA OUIDADI, ET.AL: "MPS-GAN: A multi-conditional generative adversarial network for simulating input parameters\' impact on manufacturing processes", JOURNAL OF MANUFACTURING PROCESSES, vol. 131, 12 December 2024 (2024-12-12), pages 1030 - 1045 * |
| 王子睿，等: "基于激光技术的硬脆材料SiC表面抛光的研究进展", 激光与光电子学进展, vol. 61, no. 23, 10 December 2024 (2024-12-10), pages 1 - 13 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119962621B (en) | 2025-07-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230117033A1 (en) | Method and apparatus for generating fixed-point quantized neural network | |
| CN110689109B (en) | Neural network method and apparatus | |
| US12026611B2 (en) | Method and apparatus for quantizing parameters of neural network | |
| WO2017157112A1 (en) | Method and system for bit-depth reduction in artificial neural networks | |
| CN107330355B (en) | Deep pedestrian re-identification method based on positive sample balance constraint | |
| WO2018124918A1 (en) | Training of deep neural networks on the basis of distributions of paired similarity measures | |
| CN115393779B (en) | Control system and control method for laser cladding metal ball manufacturing | |
| KR20200086581A (en) | Method and apparatus for neural network quantization | |
| CN110084221A (en) | A kind of serializing face critical point detection method of the tape relay supervision based on deep learning | |
| CN118567240B (en) | Adaptive control method and system for aluminum smelting process parameters based on big data modeling | |
| CN119962621B (en) | Wafer polishing surface control method and model training method, and related devices | |
| EP4392935A1 (en) | Robustifying nerf model novel view synthesis to sparse data | |
| CN1212453A (en) | A method of maximizing the chip yield of semiconductor wafers | |
| CN118915671A (en) | Ceramic production process parameter control method | |
| JP7493380B2 (en) | Machine learning system, and method, computer program, and device for configuring a machine learning system | |
| CN117568585A (en) | Gear heat treatment control system and method | |
| US12361275B2 (en) | Neural network device, operation method thereof, and neural network system including the same | |
| CN110728303A (en) | Dynamic Adaptive Computing Array Based on Convolutional Neural Network Data Complexity | |
| CN111931698B (en) | Image deep learning network construction method and device based on small training set | |
| CN117315720B (en) | Cross-modal pedestrian re-recognition method based on multi-feature joint progressive learning | |
| WO2024072948A1 (en) | Optimization of fabrication processes | |
| CN118096695A (en) | A defect detection method and device for steel surface defect grinding | |
| WO2024230170A1 (en) | Quantization precision selection method and device for deep network model | |
| US20250036922A1 (en) | Generative adversarial networks-based method for component anomaly detection and device thereof | |
| CN113076971A (en) | Method and device for establishing material removal rate model based on transfer learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |