An optimized control method for semiconductor etching and an etching control cabinet

Abstract

The application provides a kind of semiconductor etching optimization control method and etching control cabinet, the method includes constructing at least two cascaded first type residual module, first etching error prediction module, and generating first etching error prediction result according to original feature map;Enhanced image acquisition module, at least two cascaded second type residual module, second etching error prediction module are constructed, and second etching error prediction result is generated according to enhanced feature map;Etching error weighting module is constructed, and etching error prediction result is generated;Etching parameter self-adaptive adjustment module is constructed, and etching control is carried out according to etching error prediction result.The application can be quickly and accurately adjusted etching parameter adaptively, improve the control precision and control efficiency of etching control cabinet, meet the demand of etching precision and efficiency for advanced process.

Description

Technical Field

[0001] This invention relates to the field of semiconductors, and more specifically to an optimized control method for semiconductor etching and an etching control cabinet. Background Technology

[0002] In the etching process of semiconductor manufacturing, the etching control cabinet is a core piece of equipment that controls etching parameters (such as gas flow rate, power, and temperature) to achieve semiconductor etching. The control precision of the etching control cabinet directly affects the uniformity of critical wafer dimensions, sidewall morphology quality, and device electrical performance.

[0003] Currently, etching control cabinets are mainly based on physical equations or empirical formulas, adjusting etching parameters according to preset programs. This makes it difficult to handle complex processes involving nonlinearity and multivariable coupling, resulting in large etching errors, low wafer yield, and increased manufacturing costs. At the same time, the closed-loop cycle of existing systems from etching error detection to parameter adjustment is generally large, which cannot meet the demand for timely response in advanced processes.

[0004] Therefore, providing an optimized control method for semiconductor etching, which can adaptively, quickly, and accurately adjust etching parameters to improve the control accuracy and efficiency of the etching control cabinet and meet the requirements of advanced processes for etching accuracy and efficiency, has become an urgent technical problem to be solved. Summary of the Invention

[0005] In view of the above problems, this application is proposed to provide an optimized control method for semiconductor etching, applied to an etching control cabinet, comprising:

[0006] Step S11: Construct at least two cascaded first-class residual modules to extract features from the image to be etched and generate the original feature map, wherein the first-class residual modules do not have pooling operations;

[0007] Step S12: Construct the first etching error prediction module and generate the first etching error prediction result based on the original feature map;

[0008] Step S13: Construct an enhanced image acquisition module to perform feature enhancement on the image to be etched, and generate an enhanced etched image;

[0009] Step S14: Construct at least two cascaded second-type residual modules to extract features from the enhanced etched image and generate an enhanced feature map, wherein the second-type residual modules have pooling operations;

[0010] Step S15: Construct a second etching error prediction module to generate a second etching error prediction result based on the enhanced feature map;

[0011] Step S16: Construct an etching error weighting module to weight the first etching error prediction result and the second etching error prediction result to generate the etching error prediction result.

[0012] Step S17: Construct an adaptive etching parameter adjustment module, adaptively adjust the preset etching parameters according to the etching error prediction results, generate the adjusted etching parameters, and perform etching control.

[0013] Furthermore, the first type of residual module includes: a first convolutional layer, a second convolutional layer, and a residual connection layer; the input of the first convolutional layer serves as the input of the first type of residual module, the output of the first convolutional layer is connected to the input of the second convolutional layer, the output of the second convolutional layer is connected to the first input of the residual connection layer, the second input of the residual connection layer is connected to the input of the first convolutional layer, and the output of the residual connection layer serves as the output of the first type of residual module.

[0014] Furthermore, the first etching error prediction module uses a fully connected layer and a softmax layer to predict the etching error of the original feature map, generating the first etching error prediction result.

[0015] Furthermore, the second type of residual module includes: a first convolutional layer, a second convolutional layer, a pooling layer, and a residual connection layer; the input of the first convolutional layer serves as the input of the second type of residual module, the output of the first convolutional layer is connected to the input of the second convolutional layer, the output of the second convolutional layer is connected to the input of the pooling layer, the output of the pooling layer is connected to the first input of the residual connection layer, the second input of the residual connection layer is connected to the input of the first convolutional layer, and the output of the residual connection layer serves as the output of the second type of residual module.

[0016] Furthermore, the second etching error prediction module uses a fully connected layer and a softmax layer to predict etching errors in the enhanced feature map, generating a second etching error prediction result.

[0017] Furthermore, the kernel size of the first convolutional layer is N×1, and the kernel size of the second convolutional layer is 1×M, where N and M are positive integers greater than 1.

[0018] Furthermore, the enhanced image acquisition module includes a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, four attention weight generation modules, five multiplication modules, two addition modules, and a transpose module;

[0019] The third, fourth, and fifth convolutional layers receive the image to be etched as input. The output of the third convolutional layer is connected to the first attention weight generation module, the output of the first attention weight generation module is connected to the first input of the first multiplication module, the second input of the first multiplication module is connected to the output of the third convolutional layer, and the output of the first multiplication module is connected to the first input of the first addition module. The output of the fourth convolutional layer is connected to the second attention weight generation module, the output of the second attention weight generation module is connected to the first input of the second multiplication module, the second input of the second multiplication module is connected to the output of the fourth convolutional layer, and the output of the second multiplication module is connected to the second input of the first addition module. The output of the fifth convolutional layer is connected to the third attention weight generation module, the output of the third attention weight generation module is connected to the first input of the third multiplication module, and the third multiplication module... The second input of the multiplication module is connected to the output of the fifth convolutional layer, and the output of the third multiplication module is connected to the third input of the first addition module. The output of the first addition module is connected to the inputs of the sixth, seventh, and eighth convolutional layers, respectively. The output of the sixth convolutional layer is connected to the input of the transpose module, and the output of the transpose module is connected to the first input of the fourth multiplication module. The output of the seventh convolutional layer is connected to the second input of the fourth multiplication module, and the output of the fourth multiplication module is connected to the input of the fourth attention weight generation module. The output of the fourth attention weight generation module is connected to the first input of the fifth multiplication module, and the second input of the fifth multiplication module is connected to the output of the eighth convolutional layer. The output of the fifth multiplication module is connected to the first input of the second addition module, and the second input of the second addition module is connected to the output of the first addition module. The output of the second addition module generates the enhanced etching image.

[0020] Furthermore, the third, fourth, and fifth convolutional layers each employ different expansion rates, with the expansion rate of the third convolutional layer set to 2, the expansion rate of the fourth convolutional layer set to 3, and the expansion rate of the fifth convolutional layer set to 5.

[0021] Furthermore, the kernel of the sixth convolutional layer is 1×3×1, the kernel of the seventh convolutional layer is 3×1×1, and the kernel of the eighth convolutional layer is 1×1×3.

[0022] Furthermore, the optimized control method for semiconductor etching also includes training a first convolutional neural network prediction model composed of a cascaded first type of residual module and a first etching error prediction module, and training a second convolutional neural network prediction model composed of an enhanced image acquisition module, a cascaded second type of residual module, and a second etching error prediction module, so that the first etching error prediction result of the first convolutional neural network prediction model is... Second etching error prediction results of the second convolutional neural network prediction model The absolute value of the difference between satisfy:

[0023] .

[0024] This application also provides an etching control cabinet, including: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the optimized control method for semiconductor etching as described above.

[0025] This application utilizes a cascaded first type of residual module and a first etching error prediction module to construct a first convolutional neural network prediction model, and utilizes an enhanced image acquisition module, a cascaded second type of residual module, and a second etching error prediction module to construct a second convolutional neural network prediction model. The first convolutional neural network prediction model performs first etching error prediction on the global features of the image to be etched, and the second convolutional neural network prediction model performs second etching error prediction on the features of interest of the image to be etched. By combining the first etching error prediction results of the global features and the second etching error prediction results of the features of interest, a comprehensive etching error prediction result is generated, which improves the accuracy of etching error prediction of the image to be etched, and thus adaptively, quickly, and accurately adjusts the etching parameters, thereby improving the control precision of the etching control cabinet.

[0026] On the other hand, the first type of residual module in the first convolutional neural network prediction model has no pooling operation, which prevents the loss of local features and accurately extracts global features. The second type of residual module in the second convolutional neural network prediction model has a pooling operation, which removes uninteresting local features and focuses on deep features of interest. This reduces the dimensionality of the features, reduces the feature map output by the second type of residual module, compensates for the processing time occupied by the enhanced image acquisition module, and improves the processing efficiency of the second convolutional neural network prediction model, matching the processing efficiency of the first convolutional neural network prediction model. This allows for rapid adjustment of etching parameters, improves the control efficiency of the etching control cabinet, and comprehensively meets the requirements of advanced processes for etching accuracy and efficiency.

[0027] In another aspect, this application also provides an enhanced image acquisition module that employs a multi-scale receptive field to extract hidden multi-scale contextual information in the image to be etched, efficiently aggregates multi-scale features to generate a multi-scale feature map, and utilizes spatial correlation characteristics to evaluate the importance of each feature in the multi-scale feature map to obtain an enhanced etched image, thereby improving the accuracy of subsequent predictions by a second convolutional neural network.

[0028] In another aspect, this application also provides a method for training a convolutional neural network prediction model. The first etching error prediction of the first convolutional neural network prediction model for global features is close to the second etching error prediction result of the second convolutional neural network prediction model for features of interest. This achieves the simultaneous optimization of the parameters of the first and second convolutional neural network prediction models, and enables accurate prediction of etching errors in the image to be etched. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 The execution flowchart of the optimized control method for semiconductor etching provided in this application;

[0031] Figure 2 The execution flowchart of the first type of residual module provided in this application;

[0032] Figure 3 The execution flowchart of the second type of residual module provided in this application;

[0033] Figure 4 The execution flowchart of the enhanced image acquisition module provided in this application;

[0034] Figure 5 The hardware structure diagram of the etching control cabinet provided in this application. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0037] See Figure 1 In one embodiment, this application provides an optimized control method for semiconductor etching, applied to an etching control cabinet, comprising:

[0038] Step S11: Construct at least two cascaded first-class residual modules to extract features from the image to be etched and generate the original feature map, wherein the first-class residual modules do not have pooling operations;

[0039] Specifically, in traditional convolutional feature extraction, a large number of convolutional layers are needed to enhance the nonlinear capabilities of the neural network. However, as the depth of the neural network increases, gradient vanishing and gradient explosion occur, leading to neural network degradation. Residual modules, by introducing skip connections, allow the input to skip some convolutional layers, thus preserving the input signal and alleviating the gradient problem in the neural network. Therefore, by constructing at least two cascaded Class I residual modules, multi-dimensional feature extraction is performed on the image to be etched, obtaining the global features of the image and generating the original feature map of the original image to be etched. Figure 1 The following example illustrates the cascading of three Type I residual modules. This application does not limit the specific number of Type I residual modules.

[0040] Furthermore, to avoid the loss of deep and edge features, the first type of residual module does not perform pooling operations, but only uses multiple convolutional kernels and skip connections to extract features and generate the original feature map.

[0041] Step S12: Construct the first etching error prediction module and generate the first etching error prediction result based on the original feature map;

[0042] Specifically, the first etching error prediction module uses a fully connected layer and a softmax layer to predict the etching error of the original feature map and generate the first etching error prediction result, which is between 0 and 1.

[0043] Step S13: Construct an enhanced image acquisition module to perform feature enhancement on the image to be etched, and generate an enhanced etched image;

[0044] Specifically, feature enhancement processing of the image to be etched can enhance the depth information of the image, expand the differences between different features in the image, suppress features of no interest, enhance features of interest, and improve the accuracy of subsequent feature extraction and error prediction.

[0045] Step S14: Construct at least two cascaded second-type residual modules to extract features from the enhanced etched image and generate an enhanced feature map, wherein the second-type residual modules have pooling operations;

[0046] Specifically, while feature enhancement enhances the features of interest and increases the difference between them and features of no interest, it also reduces the efficiency of feature extraction in the second type of residual module. Therefore, pooling operations are added to the second type of residual module to further remove local features of no interest, focus on deeper features of interest, and perform feature dimensionality reduction to reduce the size of the feature map output by the second type of residual module. This compensates for the processing time occupied by the image acquisition module and improves the processing efficiency of the second type of residual module. Figure 1 The following example illustrates the cascading of three Type II residual modules. This application does not limit the specific number of Type II residual modules.

[0047] Step S15: Construct a second etching error prediction module to generate a second etching error prediction result based on the enhanced feature map;

[0048] Specifically, the structure of the second etching error prediction module is the same as that of the first etching error prediction module. It also uses a fully connected layer and a softmax layer to predict the etching error of the enhanced feature map and generate a second etching error prediction result. The second etching error prediction result is also between 0 and 1.

[0049] Step S16: Construct an etching error weighting module to weight the first etching error prediction result and the second etching error prediction result to generate the etching error prediction result.

[0050] Specifically, the cascaded first type residual module and the first etching error prediction module constitute the first convolutional neural network prediction model, and the enhanced image acquisition module, the cascaded second type residual module, and the second etching error prediction module constitute the second convolutional neural network prediction model. The first convolutional neural network prediction model performs first etching error prediction on the global features of the image to be etched, and the second convolutional neural network prediction model performs second etching error prediction on the features of interest of the image to be etched. By combining the first etching error prediction results of the global features and the second etching error prediction results of the features of interest, a comprehensive etching error prediction result is generated, which can improve the accuracy of etching error prediction of the image to be etched.

[0051] Step S17: Construct an adaptive etching parameter adjustment module, adaptively adjust the preset etching parameters according to the etching error prediction results, generate the adjusted etching parameters, and perform etching control.

[0052] Specifically, the etching parameter adaptive adjustment module adjusts the preset etching parameters in real time based on the comprehensive etching error prediction results to control the etching process.

[0053] In this embodiment of the application, a first convolutional neural network prediction model is constructed using a cascaded first type residual module and a first etching error prediction module. A second convolutional neural network prediction model is constructed using an enhanced image acquisition module, a cascaded second type residual module, and a second etching error prediction module. The first convolutional neural network prediction model performs first etching error prediction on the global features of the image to be etched, and the second convolutional neural network prediction model performs second etching error prediction on the features of interest of the image to be etched. By combining the first etching error prediction results of the global features and the second etching error prediction results of the features of interest, a comprehensive etching error prediction result is generated, which improves the accuracy of etching error prediction of the image to be etched. This allows for adaptive, rapid, and accurate adjustment of etching parameters, thereby improving the control precision of the etching control cabinet. Meanwhile, the first type of residual module in the first convolutional neural network prediction model has no pooling operation, which prevents the loss of local features and accurately extracts global features. The second type of residual module in the second convolutional neural network prediction model has a pooling operation, which removes uninteresting local features and focuses on deep features of interest. This reduces the dimensionality of the features, reduces the feature map output by the second type of residual module, compensates for the processing time occupied by the enhanced image acquisition module, and improves the processing efficiency of the second convolutional neural network prediction model, matching the processing efficiency of the first convolutional neural network prediction model. This allows for rapid adjustment of etching parameters, improves the control efficiency of the etching control cabinet, and comprehensively meets the requirements of advanced processes for etching accuracy and efficiency.

[0054] In another embodiment, this application provides a first type of residual module and a second type of residual module.

[0055] See Figure 2 The first type of residual module includes a first convolutional layer, a second convolutional layer, and a residual connection layer. The input of the first convolutional layer serves as the input of the first type of residual module. The output of the first convolutional layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the first input of the residual connection layer. The second input of the residual connection layer is connected to the input of the first convolutional layer. The output of the residual connection layer serves as the output of the first type of residual module. The residual connection layer uses an addition module. Y1 represents the output feature map of the first type of residual module, X1 represents the input of the first type of residual module, CON1 and CON2 represent the convolution processing of the first convolutional layer and the second convolutional layer, respectively. The kernel size of the multiple channels of the first convolutional layer is N×1, and the kernel size of the multiple channels of the second convolutional layer is 1×M, where N and M are positive integers greater than 1.

[0056] For example, Figure 2 The kernel size of the first convolutional layer is 3×1, and the kernel size of the second convolutional layer is 1×7. This embodiment does not limit the sizes of N and M.

[0057] In this embodiment, the residual module connection allows the input signal to be directly transmitted to deeper layers of the network, rather than relying solely on the computation results of layer-by-layer forward propagation, thereby preserving the input signal and alleviating the gradient problem of the neural network. This residual module achieves N×M convolution processing across multiple channels through a first and second convolutional layer, decomposing the N×M rectangular dimension into two convolution processes: N×1 and 1×M. This is more suitable for extracting global features of the image to be etched, and compared to direct N×M convolution processing, it significantly reduces computation and improves prediction efficiency.

[0058] See Figure 3 The second type of residual module includes: a first convolutional layer, a second convolutional layer, a pooling layer, and a residual connection layer. The input of the first convolutional layer serves as the input of the second type of residual module. The output of the first convolutional layer is connected to the input of the second convolutional layer. The output of the second convolutional layer is connected to the input of the pooling layer. The output of the pooling layer is connected to the first input of the residual connection layer. The second input of the residual connection layer is connected to the input of the first convolutional layer. The output of the residual connection layer serves as the output of the second type of residual module. The residual connection layer uses an addition module. Y2 represents the output feature map of the second type of residual module, X2 represents the input of the second type of residual module, and Pool represents the pooling operation of the pooling layer.

[0059] The second type of residual module operates on the same principle as the first type, employing a joint first and second convolutional layer to achieve N×M convolution processing. This decomposes the N×M rectangular dimension into two convolutional processes, N×1 and 1×M, used for extracting features of interest from the image to be etched. Compared to direct N×M convolution, this significantly reduces computation and improves prediction efficiency. However, unlike the first type, it adds a pooling layer to further remove uninteresting local features, focusing on deeper features of interest. This dimensionality reduction of the features decreases the size of the feature map output by the second type of residual module, compensating for the processing time occupied by the image enhancement module and improving its overall processing efficiency.

[0060] See Figure 4 In another embodiment, this application also provides an enhanced image acquisition module, which employs a multi-scale receptive field to extract hidden multi-scale contextual information in the image to be etched, efficiently aggregates multi-scale features to generate a multi-scale feature map, and uses spatial correlation characteristics to evaluate the importance of each feature in the multi-scale feature map to obtain an enhanced etched image, thereby improving the accuracy of subsequent predictions by a second convolutional neural network.

[0061] The enhanced image acquisition module includes a third, fourth, fifth, sixth, seventh, and eighth convolutional layer, four attention weight generation modules, five multiplication modules, two addition modules, and a transpose module. The inputs of the third, fourth, and fifth convolutional layers receive the image to be etched. The output of the third convolutional layer is connected to the first attention weight generation module, the output of the first attention weight generation module is connected to the first input of the first multiplication module, the second input of the first multiplication module is connected to the output of the third convolutional layer, and the output of the first multiplication module is connected to the first input of the first addition module. The output of the fourth convolutional layer is connected to the second attention weight generation module, the output of the second attention weight generation module is connected to the first input of the second multiplication module, the second input of the second multiplication module is connected to the output of the fourth convolutional layer, and the output of the second multiplication module is connected to the second input of the first addition module. The output of the fifth convolutional layer is connected to the third attention weight generation module, the output of the third attention weight generation module is connected to the first input of the third multiplication module, and the third multiplication module... The second input of the module is connected to the output of the fifth convolutional layer, and the output of the third multiplication module is connected to the third input of the first addition module. The output of the first addition module is connected to the inputs of the sixth, seventh, and eighth convolutional layers, respectively. The output of the sixth convolutional layer is connected to the input of the transpose module, and the output of the transpose module is connected to the first input of the fourth multiplication module. The output of the seventh convolutional layer is connected to the second input of the fourth multiplication module, and the output of the fourth multiplication module is connected to the input of the fourth attention weight generation module. The output of the fourth attention weight generation module is connected to the first input of the fifth multiplication module, and the second input of the fifth multiplication module is connected to the output of the eighth convolutional layer. The output of the fifth multiplication module is connected to the first input of the second addition module, and the second input of the second addition module is connected to the output of the first addition module. The output of the second addition module generates an enhanced etched image.

[0062] The third convolutional layer, an attention weight generation module, and a multiplication module constitute the first receptive field channel, generating the first receptive field feature map. The fourth convolutional layer, an attention weight generation module, and a multiplication module constitute the second receptive field channel, generating the second receptive field feature map. The fifth convolutional layer, an attention weight generation module, and a multiplication module constitute the third receptive field channel, generating the third receptive field feature map. The third, fourth, and fifth convolutional layers employ dilated convolution, expanding the receptive field by adding holes, thus focusing more on the detailed features of the image to be etched. Specifically, the third, fourth, and fifth convolutional layers use different dilation rates: the dilation rate of the third convolutional layer can be set to 2, the dilation rate of the fourth convolutional layer can be set to 3, and the dilation rate of the fifth convolutional layer can be set to 5. The attention weight generation module uses the sigmoid function.

[0063] In the first receptive field channel, the output of the third convolutional layer is processed by an attention weight generation module to generate the first receptive field weights, which are then multiplied element-wise with the output of the third convolutional layer to produce the first receptive field feature map. In the second receptive field channel, the output of the fourth convolutional layer is processed by an attention weight generation module to generate the second receptive field weights, which are then multiplied element-wise with the output of the fourth convolutional layer to produce the second receptive field feature map. In the third receptive field channel, the output of the fifth convolutional layer is processed by an attention weight generation module to generate the third receptive field weights, which are then multiplied element-wise with the output of the fifth convolutional layer to produce the third receptive field feature map. The first, second, and third receptive field feature maps are then element-wise added together using an addition module to aggregate multi-scale features, resulting in a multi-scale feature map. By setting three different receptive fields, hidden multi-scale contextual information in the image to be etched can be extracted, focusing on features of interest.

[0064] In the spatial dimension of the image to be etched, the sixth, seventh, and eighth convolutional layers extract features from three different spatial planes of the image. The kernel of the sixth convolutional layer is 1×3×1, the kernel of the seventh convolutional layer is 3×1×1, and the kernel of the eighth convolutional layer is 1×1×3. These three different kernels extract features from the XY, YZ, and XZ spatial planes of the image to be etched, thus focusing on the features of each spatial plane. Further, the transpose module transposes the output of the sixth convolutional layer and multiplies it element-wise with the output of the seventh convolutional layer. An attention weight generation module generates corresponding attention weights, which are then multiplied with the output of the eighth convolutional layer to produce a spatial attention feature map. Finally, an addition module element-wise adds the spatial attention feature map and the multi-scale feature map to produce an enhanced etched image. In the three interconnected spatial planes of the image to be etched, attention weights for the third spatial plane are generated by using features from two spatial planes to enhance the features of the third spatial plane and ultimately produce an enhanced etched image.

[0065] In another embodiment, this application also provides a neural network training method for training a first convolutional neural network prediction model composed of a cascaded first type residual module and a first etching error prediction module, and for training a second convolutional neural network prediction model composed of an enhanced image acquisition module, a cascaded second type residual module, and a second etching error prediction module.

[0066] Specifically, during the training phase, the first etch error prediction result of the first convolutional neural network prediction model is... Second etching error prediction results of the second convolutional neural network prediction model The absolute value of the difference between satisfy:

[0067]

[0068] In this embodiment, the first etching error prediction of the first convolutional neural network prediction model for global features is close to the second etching error prediction result of the second convolutional neural network prediction model for features of interest, so as to achieve the optimal parameters of the first convolutional neural network prediction model and the second convolutional neural network prediction model at the same time, and at the same time achieve accurate prediction of etching error of the image to be etched.

[0069] In another embodiment, this application also provides an etching control cabinet, which may include at least one processor, a memory (e.g., non-volatile memory), a RAM, and a communication interface, and the at least one processor, memory, RAM, and communication interface are connected together via an internal bus. The at least one processor executes at least one computer-readable instruction stored or encoded in the memory.

[0070] It should be understood that the computer-executable instructions stored in memory, when executed, cause at least one processor to perform the above-described combinations in the various embodiments of this specification. Figures 1 to 4 The description includes various operations and functions.

[0071] In another embodiment, this application also provides a program product, such as a machine-readable medium. The machine-readable medium may have instructions (i.e., the elements implemented in software as described above), which, when executed by a machine, cause the machine to perform the above-described combinations of the various embodiments of this specification. Figures 1-4 The various operations and functions described. Specifically, a system or apparatus equipped with a readable storage medium storing software program code that implements the functions of any of the embodiments described above, and enabling the computer or processor of the system or apparatus to read and execute the instructions stored in the readable storage medium.

[0072] In this case, the program code read from the readable medium itself can perform the functions of any of the above embodiments, and therefore the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of this specification.

[0073] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer or the cloud via a communication network.

[0074] Those skilled in the art will understand that the various embodiments disclosed above can be modified and varied without departing from the spirit of the invention. Therefore, the scope of protection of this specification should be defined by the appended claims.

[0075] It should be noted that not all steps and units in the above process and system structure diagrams are mandatory; some steps or units can be omitted according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in the above embodiments can be a physical structure or a logical structure. That is, some units may be implemented by the same physical client, or some units may be implemented by multiple physical clients, or they may be jointly implemented by certain components in multiple independent devices.

[0076] In the above embodiments, the hardware units or modules can be implemented mechanically or electrically. For example, a hardware unit, module, or processor may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit or processor may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0077] The specific embodiments described above with reference to the accompanying drawings are exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of the claims. The term "exemplary" as used throughout this specification means "serving as an example, instance, or illustration" and does not imply that it is "preferred" or "advantageous" compared to other embodiments. Specific details are included to provide an understanding of the described techniques. However, these techniques can be practiced without these specific details. In some instances, well-known structures and apparatuses are shown in block diagram form to avoid obscuring the concepts of the described embodiments.

[0078] The foregoing description of this disclosure is provided to enable any person skilled in the art to implement or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles applicable herein can be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. An optimized control method for semiconductor etching, applied to an etching control cabinet, characterized in that, include: Step S11: Construct at least two cascaded first-class residual modules to extract features from the image to be etched and generate the original feature map, wherein the first-class residual modules do not have pooling operations; Step S12: Construct the first etching error prediction module and generate the first etching error prediction result based on the original feature map; Step S13: Construct an enhanced image acquisition module to perform feature enhancement on the image to be etched, and generate an enhanced etched image; Step S14: Construct at least two cascaded second-type residual modules to extract features from the enhanced etched image and generate an enhanced feature map, wherein the second-type residual modules have pooling operations; Step S15: Construct a second etching error prediction module to generate a second etching error prediction result based on the enhanced feature map; Step S16: Construct an etching error weighting module to weight the first etching error prediction result and the second etching error prediction result to generate the etching error prediction result. Step S17: Construct an adaptive etching parameter adjustment module, adaptively adjust the preset etching parameters according to the etching error prediction results, generate the adjusted etching parameters, and perform etching control. The method further includes: training a first convolutional neural network prediction model composed of a cascaded first type residual module and a first etching error prediction module; and training a second convolutional neural network prediction model composed of an enhanced image acquisition module, a cascaded second type residual module, and a second etching error prediction module, so that the first etching error prediction result of the first convolutional neural network prediction model is... Second etching error prediction results of the second convolutional neural network prediction model The absolute value of the difference between satisfy: 。 2. The method according to claim 1, characterized in that, The first type of residual module includes: a first convolutional layer, a second convolutional layer, and a residual connection layer; the input of the first convolutional layer serves as the input of the first type of residual module, the output of the first convolutional layer is connected to the input of the second convolutional layer, the output of the second convolutional layer is connected to the first input of the residual connection layer, the second input of the residual connection layer is connected to the input of the first convolutional layer, and the output of the residual connection layer serves as the output of the first type of residual module.

3. The method according to claim 2, characterized in that, The first etching error prediction module uses a fully connected layer and a softmax layer to predict the etching error of the original feature map and generate the first etching error prediction result.

4. The method according to claim 1, characterized in that, The second type of residual module includes: a first convolutional layer, a second convolutional layer, a pooling layer, and a residual connection layer; the input of the first convolutional layer serves as the input of the second type of residual module, the output of the first convolutional layer is connected to the input of the second convolutional layer, the output of the second convolutional layer is connected to the input of the pooling layer, the output of the pooling layer is connected to the first input of the residual connection layer, the second input of the residual connection layer is connected to the input of the first convolutional layer, and the output of the residual connection layer serves as the output of the second type of residual module.

5. The method according to claim 4, characterized in that, The second etching error prediction module uses a fully connected layer and a softmax layer to predict etching errors in the enhanced feature map, generating the second etching error prediction result.

6. The method according to any one of claims 2-5, characterized in that, The kernel size of the first convolutional layer is N×1, and the kernel size of the second convolutional layer is 1×M, where N and M are positive integers greater than 1.

7. The method according to claim 1, characterized in that, The enhanced image acquisition module includes a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, four attention weight generation modules, five multiplication modules, two addition modules, and a transpose module; The third, fourth, and fifth convolutional layers receive the image to be etched as input. The output of the third convolutional layer is connected to the first attention weight generation module, the output of the first attention weight generation module is connected to the first input of the first multiplication module, the second input of the first multiplication module is connected to the output of the third convolutional layer, and the output of the first multiplication module is connected to the first input of the first addition module. The output of the fourth convolutional layer is connected to the second attention weight generation module, the output of the second attention weight generation module is connected to the first input of the second multiplication module, the second input of the second multiplication module is connected to the output of the fourth convolutional layer, and the output of the second multiplication module is connected to the second input of the first addition module. The output of the fifth convolutional layer is connected to the third attention weight generation module, the output of the third attention weight generation module is connected to the first input of the third multiplication module, and the third multiplication module... The second input of the multiplication module is connected to the output of the fifth convolutional layer, and the output of the third multiplication module is connected to the third input of the first addition module. The output of the first addition module is connected to the inputs of the sixth, seventh, and eighth convolutional layers, respectively. The output of the sixth convolutional layer is connected to the input of the transpose module, and the output of the transpose module is connected to the first input of the fourth multiplication module. The output of the seventh convolutional layer is connected to the second input of the fourth multiplication module, and the output of the fourth multiplication module is connected to the input of the fourth attention weight generation module. The output of the fourth attention weight generation module is connected to the first input of the fifth multiplication module, and the second input of the fifth multiplication module is connected to the output of the eighth convolutional layer. The output of the fifth multiplication module is connected to the first input of the second addition module, and the second input of the second addition module is connected to the output of the first addition module. The output of the second addition module generates the enhanced etching image.

8. The method according to claim 7, characterized in that, The third, fourth, and fifth convolutional layers each employ different expansion rates, with the expansion rate of the third convolutional layer set to 2, the expansion rate of the fourth convolutional layer set to 3, and the expansion rate of the fifth convolutional layer set to 5.

9. The method according to claim 8, characterized in that, The kernel of the sixth convolutional layer is 1×3×1, the kernel of the seventh convolutional layer is 3×1×1, and the kernel of the eighth convolutional layer is 1×1×3.

10. An etching control cabinet, characterized in that, include: At least one processor; And a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the optimized control method for semiconductor etching as described in any one of claims 1 to 9.

Images