A Deep Learning-Based Lithography Hot Spot Detection Method and Device

Abstract

The present invention discloses a lithography hot spot detection method and device based on deep learning, including collecting circuit layout images and constructing a sample data set containing lithography layout hot spots and non-hot spots; using a generative adversarial network to perform mixed data augmentation on hot spot samples; using the improved data set to train a hot spot detection network; and inputting the layout to be detected into the trained hot spot detection network to obtain a hot spot detection result. The generative adversarial network of the present invention proposes a mixed data augmentation method for generating more lithography hot spot layout samples, effectively solving the problem of serious imbalance in the sample quantity distribution, improving the training stability of the GAN network and the ability to generate auxiliary samples with complete features and high quality; the improved hot spot detection network reduces the calculation amount, and can solve the problems of overfitting and gradient dispersion that are likely to occur when the model parameters are too many and the training data set is limited, making it difficult to optimize the model.

Description

Technical Field

[0001] The present invention relates to the field of lithography hot spot detection, and in particular to a lithography hot spot detection method and device based on deep learning. Background Art

[0002] In semiconductor manufacturing, the lithography process is a key technology for pattern transfer, and its control accuracy and resolution directly determine the yield of semiconductors. Among the many factors affecting the lithography process, the quality of the optical devices of the lithography machine plays a decisive role, and the quality of the imaging is directly determined by the optical devices. During the lithography process, due to the influence of the optical proximity effect, the mask pattern will be distorted when transferred to the surface of the silicon wafer. For example: corner rounding, line end shortening, line width deviation, and partial area missing. The circuit fault area generated by the lithography of the mask layout is called a lithography hot spot. Lithography layout hot spots can cause defects such as circuit short circuits, open circuits, and poor ohmic contacts, seriously affecting the electrical characteristics of integrated circuits and the product yield of integrated circuit board manufacturing.

[0003] To improve the manufacturing quality of integrated circuits, the industry has proposed to use resolution enhancement technologies to repair hot spot layouts, such as sub-resolution assist feature technology and optical proximity correction technology. This technology can improve the quality of the circuit board after imaging, but it cannot completely eliminate layout hot spots. Therefore, it is very necessary to perform lithography layout hot spot detection before the circuit layout is burned onto the silicon wafer. The complexity of integrated circuits increases in accordance with Moore's law, and hot spot detection faces problems such as an imbalance in the ratio of hot spot samples to non-hot spot samples, complex hot spot patterns, and high false alarm rates in detection. Lithography layout hot spot detection has become a huge challenge in the field of integrated circuit manufacturing.

[0004] Lithography hot spot detection can be defined as accurately locating the lithography hot spot areas existing in the layout within an acceptable time range. As an important link in the design for manufacturability of integrated circuits, the research on lithography hot spot detection has received wide attention. The existing lithography hot spot detection methods can be roughly divided into three types: hot spot detection based on lithography simulation, hot spot detection based on pattern matching, and hot spot detection based on machine learning.

[0005] To shorten the feedback time of lithography hot spot detection for layout design, machine learning has been applied to lithography hot spot detection. The hot spot detection based on machine learning defines lithography hot spot detection as a classification problem. During the training stage, a large number of lithography hot spot and non-lithography hot spot layout samples are used as the training set, and the feature vectors of each layout are extracted as the input of the machine learning model. The lithography hot spot classifier is trained through supervised learning. In the testing stage, the trained classifier can effectively determine whether the layout is a hot spot and locate the positions with a relatively high probability of belonging to the lithography hot spot area in the layout.

[0006] However, machine learning methods require manual extraction of image features, and the effectiveness of manual feature extraction requires further exploration and verification. As a subcategory of end-to-end learning, deep learning possesses superior learning capabilities. Compared to machine learning methods, deep learning eliminates the tedious manual feature extraction process and can automatically extract image features using convolutional neural networks. Deep learning can effectively improve feature learning capabilities through a massive data-driven approach, enabling precise fitting of complex functional relationships and capturing the rich information inherent in the data.

[0007] While using GAN models to generate data samples offers advantages such as simplicity, unsupervised autonomous learning, and high-quality generated samples, they also suffer from issues such as vanishing gradients, unstable training, and mode collapse, making them unsuitable for processing discrete data. Mode collapse occurs when GANs generate unstable and poor-quality samples during training. This occurs when the generator generates an auxiliary sample that is not realistic, but the discriminator correctly identifies it. The generator then continues to generate similar auxiliary samples, leading to mutual deception between the two. This results in feature loss and incomplete information in the final generated auxiliary sample. Summary of the invention

[0008] The purpose of the present invention is to address the deficiencies of the existing technology and propose a lithography hotspot detection method and device based on deep learning.

[0009] The objective of the present invention is achieved through the following technical solution: a method for detecting lithography hotspots based on deep learning, the method comprising the following steps:

[0010] S1. Collect circuit layout images and construct a sample dataset containing hot spots and no hot spots in the photolithography layout;

[0011] S2. Use generative adversarial networks to perform hybrid data enhancement on hotspot samples;

[0012] S3, use the improved dataset to train the hotspot detection network;

[0013] S4. Input the to-be-detected pattern into the trained hotspot detection network to obtain the hotspot detection result.

[0014] Furthermore, the S1 specifically includes: using an optical nano-level camera to capture an integrated circuit layout image, segmenting the acquired image, and generating a sample data set including hot spots and no hot spots in the lithography layout.

[0015] Furthermore, the hybrid data enhancement includes: processing layout samples containing lithography hotspots through geometric transformation; constructing a hybrid loss based on Wassersteun loss and binary cross entropy loss, and optimizing the generative adversarial network model to enhance the generated sample data.

[0016] Further, the layout sample containing lithography hotspots processed by geometric transformation includes: increasing the number of samples by using two methods of rotation and mirror transformation;

[0017] Further, the construction of the hybrid loss based on Wasserstein loss and binary cross-entropy loss includes:

[0018] =

[0019]

[0020]

[0021] Among them, G represents the generator, D represents the discriminator, represents the real sample, represents the input of the generator, represents the sample generated by the generator, and E represents the mathematical expectation; is the unknown sample, N is the total number of samples, represents belongs to the probability of the real distribution, represents belongs to the probability of the generated distribution; λ1 and λ2 are the coefficients of binary cross-entropy and Wasserstein loss.

[0022] Further, the hotspot detection network is a GoogLeNet model, which is successively connected by two 3×3 convolutional blocks, two 3×3 max-pooling blocks, two 7×7 convolutional blocks, four improved Inception modules, a 7×7 max-pooling layer, a fully connected layer, and a soft-max layer in the order shown in the figure.

[0023] Further, the improved Inception includes four main branches. The first main branch only includes a 1×1 convolution. The second main branch includes a backbone composed of a 1×1 convolution and a 3×3 convolution connected in sequence, and two sub-branches of a 1×3 convolution and a 3×1 convolution are connected after the 3×3 convolution; the third main branch includes a backbone composed of a 1×1 convolution and two sub-branches connected after the 1×1 convolution, and each sub-branch is a 1×3 convolution; the fourth main branch includes a 3×3 convolution and a 1×1 convolution connected in sequence; the data of all channels are concatenated and merged for the input of the next layer.

[0024] According to another aspect of the specification, there is also provided a deep learning-based lithography hot spot detection device, including a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, the described deep learning-based lithography hot spot detection method is implemented.

[0025] According to another aspect of the specification, there is also provided a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the described deep learning-based lithography hot spot detection method is implemented.

[0026] Advantages of the present invention:

[0027] 1. Based on the GAN network, a hybrid data augmentation method HDAM is proposed to generate more lithography hot spot layout samples, effectively solving the problem of serious imbalance in the sample quantity distribution, improving the training stability of the GAN network and the ability to generate auxiliary samples with complete features and high quality.

[0028] 2. The GoogLeNet deep learning model pre-trained with a large dataset is adopted, effectively solving the problem of overfitting easily in the case of too many model parameters and limited training datasets; reducing the computational complexity brought by the network structure, facilitating practical applications; being more easily optimized, and solving the problem that it is difficult to optimize the model due to gradient dispersion. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flowchart of a deep learning-based lithography hot spot detection method provided by an embodiment of the present invention based on HDAM and an improved GoogleNet model;

[0030] Figure 2 It is a schematic diagram of a test dataset sample provided by an embodiment of the present invention;

[0031] Figure 3 It is a text flowchart of HDAM provided by an embodiment of the present invention;

[0032] Figure 4 It is a network structure diagram of the generator of GAN provided by an embodiment of the present invention;

[0033] Figure 5 It is a schematic diagram of the GoogLeNet model structure provided by an embodiment of the present invention;

[0034] Figure 6 It is a network structure diagram of the improved Inception module provided by an embodiment of the present invention;

[0035] Figure 7 It is a schematic diagram of a deep learning-based lithography hot spot detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following further elaborates on the specific implementation manners of the present invention in conjunction with the accompanying drawings.

[0037] As Figure 1 shown, a lithography hot spot detection method based on deep learning provided by the present invention includes:

[0038] S1: Collect an integrated circuit layout image through an optical nanoscale camera, segment and obtain the image, and generate a sample data set including lithography layout hot spots and non-hot spots;

[0039] S2: Perform data augmentation on the lithography layout hot spot samples through the HDAM method based on the GAN network, and reduce the difference in the sample quantity distribution under different categories;

[0040] S3: Process the samples in S2 and divide them into a training sample set and a test sample set according to a certain proportion;

[0041] S4: Input the hot spot layout sample set trained in S3 into the improved GoogLeNet model for training. The training method is to divide the training samples into a training sample set and a test sample set according to a certain proportion, and the test metrics are recall rate, accuracy rate, and F1 score.

[0042] The recall rate represents the proportion of the number of correctly predicted hot spot layouts in the actual positive examples:

[0043] Recall =

[0044] The precision rate represents the proportion of the actual number of hot spot layouts in the predicted hot spot layouts:

[0045] Precision =

[0046] However, simply using the accuracy rate or the recall rate cannot well evaluate the performance of the algorithm model, and the F1 score is needed to measure the comprehensive performance of the model:

[0047] F1 =

[0048] Among them, TP, FP, and FN respectively represent that the hot spot is correctly identified, the non-hot spot is misidentified as a hot spot, and the hot spot is identified as a non-hot spot. Mark the training sets and test sets of the two categories as and , where TR represents the training set, TE represents the test set, the input x size is 224×224, the output is y∈[0,1], and the formula for constructing the GoogleNet mathematical model is as follows:

[0049] (1)Model input and output relationship:

[0050]

[0051] Among them is the input image, X is the feature representation of the input image or the feature map between levels, then corresponds to the entire mapping operation process from the input image to the feature map extraction. W is the convolution kernel, and b is the bias, are hyperparameters.

[0052] (2)The classifier parameters are shown in Equation 3-9. The final output category is Y = argmax{Y(0), y(1)}, that is, the output indicates the side with a higher probability that the input image belongs to a hot spot or a non-hot spot.

[0053]

[0054] (3)Loss function is given by the following formula:

[0055]

[0056] Among them, R(W) and R(θ) represent the regularization terms, which sparsify the parameters to prevent overfitting, represents the L2 norm, , , are all hyperparameter weights, , represent the weights corresponding to whether the input image belongs to a hot spot or a non-hot spot, represents the total number of data of the input-output pairs in the training set.

[0057] S5: Input the hot layout sample set tested in S3 into the GoogLeNet model trained in S4 for lithography layout hot spot detection. The test sample set is the integrated circuit layout design data set of the ICCAD 2012 competition. Its benchmark has a total of 5 GDSII format design layouts including hot spots and non-hot spots. The original layout data format consists of a continuous list of vertex coordinates. Therefore, these vertex coordinates need to be encoded into two-dimensional density pictures. The design sample is shown by Figure 2 given, Figure 2 (a) in Figure 2 is a hot layout design,

[0058] (b) in

[0059] Table 1: Detailed information of each training and test data set design benchmark

[0060]

[0061] Specifically, in step S2, since the layout containing lithography hotspots only accounts for 5% of all layouts, the sample quantity distribution is seriously unbalanced. To improve the training stability of the GAN network and its ability to generate auxiliary samples with complete features and high quality, a Hybrid data augment method (HDAM) is proposed based on the GAN network to generate more layout samples of lithography hotspots, and an index evaluation system is constructed to analyze the quality of the generated layout samples.

[0062] As Figure 3 shown, the steps of the hybrid data augmentation method in the S2 process are as follows:

[0063] S21: Process the layout samples containing lithography hotspots through geometric transformations such as rotation and translation;

[0064] S22: Construct a hybrid loss based on the Wassersteun loss and the binary cross-entropy loss to optimize the original GAN model to enhance the generated sample data.

[0065] Its specific structure is as follows:

[0066] (1) Geometric transformation

[0067] In step S21, considering that the auxiliary samples generated by using a single data augmentation method are repetitive, it is necessary to use a hybrid data augmentation method for data augmentation. In the field of image processing, common methods include rotation, brightness adjustment, adding random noise, etc. To ensure that the generated image after processing retains its original features, this paper only uses two methods, rotation and mirror transformation, to increase the number of samples. The general expression is:

[0068]

[0069] Among them, the coefficient matrix A is the rotation matrix coefficient, θ represents the rotation angle. To prevent the loss of edge features, its value is {π / 2, π, 3π / 2}. The coefficient matrix B is the mirror coefficient matrix, which rotates about the mirror line ax + by + c = 0. In the formula, RN represents random noise, P represents the original sample, and P' represents the generated sample.

[0070] (2) Structures of the generator network and the discriminator network

[0071] The structure of the generator is as Figure 4As shown in the figure, assume that the goal of training the generator G is to generate an image with a size of 224×224×3. The input of the generator is a 100-dimensional random noise that follows a Gaussian distribution with a standard deviation of 1 and a mean of 0. First, the random noise is projected and reshaped into a feature map with a size of 7×7×512. Subsequently, these feature blocks are transformed into a convolutional representation with a size of 224×224×3 through 5 transposed convolution blocks. Among them, each convolutional block includes a transposed convolution layer, a batch normalization layer (denoted as BN in the figure), and a PReLU activation function layer.

[0072] The discriminator extracts features through convolutional operations, then reduces the dimension through pooling operations, and then randomly discards a certain proportion of neurons (set to 0.1 in the present invention) through a dropout layer to prevent the model from overfitting. Finally, the fully connected layer obtains the predicted probability value by mapping the output to the sample label space.

[0073] (3) Loss function

[0074] In step S22, when the discriminator is optimal, the loss of the generator in the generative adversarial network can be equivalently transformed into minimizing the Jensen-Shannon (JS) divergence between the real samples and the generated auxiliary samples. As an improvement of the KL (Kullback–Leibler divergence), the JS divergence can overcome the limitation of the asymmetry of the KL divergence and has the ability to characterize the distance. Its calculation formula is as follows:

[0075]

[0076] where P r represents the real sample probability distribution, P g represents the generated sample probability, D KL (∙) represents the KL divergence between the two, separated by the || symbol. Usually, the overlap between P r and P g can be ignored, so the JS divergence is approximately a constant value, which will lead to the problem of gradient disappearance when training the generator model. The Wasserstein distance has approximate differentiability and good smoothness, and can better characterize the difference between two distributions. It can be expressed as the minimum "cost" of converging P g to the real distribution P r :

[0077]

[0078] Among them, inf represents the lower bound of the threshold, γ represents the joint distribution of the true sample probability and the generated sample probability, S(,) represents all possible joint distributions of the two, and γ(x, y) represents the loss required for mapping x to y so that P g and P r follow the same distribution. The above formula can be transformed into a solvable Wasserstein loss according to the Kantorovich-Rubinstein criterion:

[0079]

[0080]

[0081] Among them, G represents the generator, D represents the discriminator, represents the true sample, represents the generator input, represents the sample generated by the generator, and E represents the mathematical expectation. L represents the set of 1-Lipschitz functions. Since the Wasserstein loss is a critical function, the generator is easier to optimize and can effectively avoid the phenomenon of gradient disappearance. Therefore, in order to make the training process of the built model more stable and easy to converge, the present invention proposes a hybrid optimization objective combining binary cross-entropy loss and Wasserstein loss, that is:

[0082]

[0083] =

[0084] Among them, is the unknown sample, N is the total number of samples, represents belonging to the true distribution probability, represents belonging to the generated distribution probability. λ1 and λ2 are the coefficients of binary cross-entropy and Wasserstein loss, which are set to 1 and 0.01 respectively. That is, in the initial stage of model training, the former plays a leading role and guides the model gradient update; when gradient disappearance occurs, the latter plays an auxiliary correction role.

[0085] Specifically, in step S3, since the size of the sample set is relatively large, with a size of 1200×1200×1, the image is compressed to 224×224×1 through linear interpolation method, and then the single-channel is copied to a sample set of 224×224×3.

[0086] Specifically, in step S4, a pre-trained GoogLeNet model is used to detect lithography hotspots. The structure is as Figure 5 shown. Two 3×3 convolutional blocks, two 3×3 max-pooling blocks, two 7×7 convolutional blocks, four Inception modules, a 7×7 max-pooling layer, a fully connected layer, and a soft-max layer are sequentially connected in the order shown in the figure. Each convolutional block contains a convolutional layer, a Relu activation function layer, and a batch normalization layer. The core of GoogLeNet is the Inception module, which is used to extract rich and delicate image features. Multiple Inception modules are stacked to form a densely connected network, aiming to extract rich features at different levels as much as possible and improve the feature extraction effect. The features extracted by the Inception module are further compressed by the 7×7 max-pooling layer while retaining the main information. The fully connected layer extracts the result after pooling compression into data for judging whether it is a hotspot (outputting dual-channel data). The soft-max layer then approximately normalizes the output of the fully connected layer using the sigmoid function to prevent instability. Finally, the hotspot judgment result of the layout is output according to the normalized value. The original Inception module uses four branches, each branch is composed of a 1×1 convolution, a 3×3 convolution, a 5×5 convolution, and a 3×3 max-pooling layer, significantly expanding the width of the network layer domain and increasing the number of neuron units at each level. At the same time, the small-size convolution kernel can overcome the problem that the algorithm complexity increases due to the deepening of the traditional neural network. At the same time, by ensuring the step size, the feature dimensions output by the four branches are the same, and then they are stacked to achieve the purpose of multi-scale feature extraction. However, the 3×3 convolution and 5×5 convolution still bring a large amount of computation. Therefore, 1×1 convolution is used to reduce the dimension of the feature map, that is, a 1×1 convolution is added before the 3×3 convolution and 5×5 convolution to reduce the computational bottleneck. At the same time, the transformation increases the number of network layers and improves the expression ability of the network. To further improve the detection accuracy and compress the computational time consumption, borrowing the idea of the VGG network, the computational efficiency is improved by reducing the large-size convolution kernel. The 5×5 convolution is replaced by two sequentially connected 3×3 convolutions to reduce the number of parameters and alleviate the overfitting phenomenon. On the basis of replacing 5×5 with 3×3, 1×3 convolution and 3×1 convolution are used to replace the 3×3 convolution. Finally, the data of all channels are concatenated and merged for the input of the next layer.

[0087] The improved Inception module is as Figure 6As shown, the improved Inception includes four main branches. The first main branch only includes 1×1 convolutions. The second main branch includes a backbone composed of sequentially connected 1×1 and 3×3 convolutions, and two sub-branches, namely 1×3 and 3×1 convolutions, are connected after the 3×3 convolution. The third main branch includes a backbone composed of 1×1 convolutions and two sub-branches connected after the 1×1 convolution, and each sub-branch is a 1×3 convolution. The fourth main branch includes sequentially connected 3×3 and 1×1 convolutions. The data of all channels are concatenated and merged for the next layer input.

[0088] This splitting of the asymmetric convolution structure has a significantly better result than splitting into several identical convolution kernels, can process more and richer image features, and increase feature diversity. At the same time, it also adds a layer of non-linear mapping, making the feature information more discriminative.

[0089] Corresponding to the foregoing embodiment of a deep learning-based lithography hot spot detection method, the present invention also provides an embodiment of a deep learning-based lithography hot spot detection device.

[0090] See Figure 7 , an embodiment of a deep learning-based lithography hot spot detection device provided by the present invention includes a memory and one or more processors. Executable code is stored in the memory, and when the processor executes the executable code, it is used to implement a deep learning-based lithography hot spot detection method in the foregoing embodiment.

[0091] The embodiment of the deep learning-based lithography hot spot detection device provided by the present invention can be applied to any device with data processing capabilities, and the device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, as Figure 7 shown, it is a hardware structure diagram of any device with data processing capabilities where the deep learning-based lithography hot spot detection device provided by the present invention is located. In addition to Figure 7 the shown processor, memory, network interface, and non-volatile memory, the device with data processing capabilities where the device in the embodiment is located usually includes other hardware according to the actual functions of the device with data processing capabilities, which will not be elaborated here.

[0092] For the implementation process of the functions and roles of each unit in the above device, please refer to the implementation process of the corresponding steps in the above method for details, which will not be elaborated here.

[0093] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0094] The embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements a method for lithography hot spot detection based on deep learning in the above embodiment.

[0095] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store the data that has been output or will be output.

[0096] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for lithography hot spot detection based on deep learning described above.

[0097] After considering the specification and practicing the content disclosed herein, those skilled in the art will easily think of other implementation schemes of this application. This application aims to cover any variations, uses, or adaptive changes of this application. These variations, uses, or adaptive changes follow the general principles of this application and include the common general knowledge or conventional technical means in the technical field not disclosed in this application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of this application are pointed out by the claims.

[0098] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit this application. This application is not limited to the exact structures that have been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is only limited by the appended claims.

Claims

1. A lithography hot spot detection method based on deep learning, characterized in that, The method comprises the following steps: S1. Collect circuit layout images and construct a sample data set including lithography layout hotspots and non - hotspots; S2. Use a generative adversarial network to perform mixed data augmentation on the hotspot samples; S3. Use the improved data set to train a hotspot detection network; the hotspot detection network is a GoogLeNet model, which is successively connected by two 7×7 convolutional blocks, a 3×3 max - pooling block, two 3×3 convolutional blocks, a 3×3 max - pooling block, four improved Inception modules, a 7×7 max - pooling layer, a fully - connected layer, and a soft - max layer; in addition, the outputs of the four improved Inception modules are merged and then input into the 7×7 max - pooling layer; The improved Inception includes four main branches. The first main branch only includes a 1×1 convolution. The second main branch includes a backbone composed of a 1×1 convolution and a 3×3 convolution connected in sequence, and two sub - branches, a 1×3 convolution and a 3×1 convolution, are connected after the 3×3 convolution. The third main branch includes a backbone composed of a 1×1 convolution and two sub - branches connected after the 1×1 convolution, and each sub - branch is a 1×3 convolution. The fourth main branch includes a 3×3 convolution and a 1×1 convolution connected in sequence. The data of all channels are spliced and merged for the input of the next layer; S4. Input the layout to be detected into the trained hotspot detection network to obtain the hotspot detection result.

2. The method for detecting lithography hotspots based on deep learning according to claim 1, wherein The specific operation of S1 is: Use an optical nano - level camera to collect integrated circuit layout images, segment the obtained images, and generate a sample data set including lithography layout hotspots and non - hotspots.

3. The lithography hot spot detection method based on deep learning according to claim 1, wherein The mixed data augmentation includes: Processing the layout samples containing lithography hotspots through geometric transformation; constructing a mixed loss based on the Wassersteun loss and the binary cross - entropy loss to optimize the generative adversarial network model to enhance the generated sample data.

4. The method for detecting lithography hotspots based on deep learning according to claim 3, wherein, The processing of the layout samples containing lithography hotspots through geometric transformation includes: Using two methods, rotation and mirror transformation, to increase the number of samples.

5. The method for detecting lithography hotspots based on deep learning according to claim 3, wherein, The construction of the mixed loss based on the Wassersteun loss and the binary cross - entropy loss includes: L(G,D) = λ1L′(G,D)+λ2W(G,D); Among them, G represents the generator, D represents the discriminator, x represents the real sample, and z ∼ P z represents the input of the generator, represents the sample generated by the generator, E represents the mathematical expectation; y i is an unknown sample, N is the total number of samples, P r (y i ) represents that y i belongs to the probability of the real distribution, P g (y i ) represents that y i belongs to the probability of the generated distribution; λ1 and λ2 are the coefficients of binary cross-entropy and Wasserstein loss.

6. A lithography hot spot detection device based on deep learning, comprising a memory and one or more processors, wherein executable code is stored in the memory, characterized in that, When the processor executes the executable code, it implements a deep - learning - based lithography hotspot detection method as described in any one of claims 1 - 5.

7. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements a deep - learning - based lithography hotspot detection method as described in any one of claims 1 - 5.

Images