Optical proximity effect correction method and device, electronic equipment and storage medium

Abstract

The invention relates to an optical proximity effect correction method and device, electronic equipment and a storage medium. The correction method comprises the following steps: acquiring an original mask plate, wherein the original mask plate is a mask plate pattern designed based on a target wafer image; inputting the original mask plate into a preset probability diffusion model to obtain a corrected target mask plate; wherein the preset probability diffusion model is obtained by training original mask plates of different wafer images and corrected mask plates corresponding to the original mask plates; and outputting the target mask so as to obtain a target wafer image based on the target mask. According to the method, the probabilistic diffusion model is introduced into the correction method, end-to-end learning of the complex mapping relation between the original mask and the corrected mask is realized, the correction processing time is shortened, the correction efficiency and accuracy are improved, the throughput of semiconductor manufacturing is improved, and the calculation cost is reduced.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to an optical proximity effect correction method, device, electronic device and storage medium. Background Art

[0002] In the integrated circuit lithography process of semiconductor manufacturing, the optical proximity effect (OPE) is the core physical problem that causes silicon wafer imaging distortion. In order to improve the image fidelity, optical proximity correction (OPC) of the original mask needs to be performed during the design stage.

[0003] Existing OPC methods involve correcting mask patterns based on predefined empirical rules. This method is simple and fast, but lacks accuracy for complex patterns and smaller process nodes. Other OPC methods involve simulating the photolithography process using light propagation models, resist response models, etc., and adjusting the mask pattern through iterative optimization algorithms until the imaging on the wafer matches the design target. This method has high accuracy, but is extremely computationally complex, especially for full-chip OPC, which requires enormous computing resources and time, becoming a bottleneck in the production process. Other OPC methods involve combining rules and models, precalculating a series of correction values ​​and storing them in a lookup table, and performing table lookup corrections based on the actual pattern. However, the lookup table is expensive to construct, suffers from the curse of dimensionality, and has limited coverage. It has limited adaptability to new process conditions or complex patterns, and new patterns require fallback model calculations because they are not pre-stored.

[0004] In summary, it is difficult for existing OPC methods to simultaneously achieve the goals of high computational efficiency and high correction accuracy. Summary of the Invention

[0005] In order to solve the above technical problems, the present application provides an optical proximity effect correction method, device, electronic device and storage medium. By using a trained probability diffusion model, a complex correction process can be completed with only one forward inference calculation of the model, thereby improving the efficiency of converting the original mask into the corrected mask and reducing the computing cost.

[0006] In a first aspect of the present application, a method for correcting an optical proximity effect is provided, comprising:

[0007] Obtaining an original mask, wherein the original mask is a mask pattern designed based on a target wafer image;

[0008] Inputting the original mask into a preset probability diffusion model to obtain a corrected target mask; wherein the preset probability diffusion model is trained based on original masks of different wafer images and corrected masks corresponding to the original masks;

[0009] The target mask is output to obtain a target wafer image based on the target mask.

[0010] In some embodiments of the present application, the preset probability diffusion model is trained in the following manner:

[0011] Based on the original mask of the wafer image, determining first data corresponding to the original mask of the wafer image and second data corresponding to the corrected mask;

[0012] Noise-processing the first data according to a preset first noise to obtain third data;

[0013] performing noise reduction processing on the third data according to the predicted second noise to obtain second data;

[0014] The predicted second noise is confirmed based on an image segmentation model.

[0015] In some embodiments of the present application, the step of adding noise to the first data according to a preset first noise to obtain third data includes:

[0016] From the current time step data x t-1 Add noise to the next time step data x t The probability distribution in is determined by the following formula:

[0017]

[0018] In the formula, q(x t |x t-1 ) is the transition probability distribution function of the noise processing, β t is the noise schedule corresponding to the preset first noise at the t-th time step, Indicates the mean The variance is β t Gaussian distribution, I is the unit matrix that ensures the independence of each dimension.

[0019] In some embodiments of the present application, the denoising process on the third data according to the predicted second noise to obtain the second data includes:

[0020] From the current time step data x t-1 Denoise the data x at the next time step t The probability distribution in is determined by the following formula:

[0021]

[0022] Where p θ (x t-1 |x t ) is the transfer probability distribution function of the noise reduction process, N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)) indicates that the mean is μ θ (x t ,t), the variance is ∑ θ (x t ,t)’s Gaussian distribution.

[0023] In some embodiments of the present application, the mean is determined based on the following formula:

[0024]

[0025] Where,∈ θ (x t ,t) is the predicted second noise corresponding to the t-th time step, α t is the attenuation coefficient of time step t, α t =1-β t , is the accumulation coefficient of forward diffusion,

[0026] In some embodiments of the present application, a method for confirming the predicted second noise based on an image segmentation model includes:

[0027] Inputting the noisy mask image and the time step into the image segmentation model to determine the minimum mean square error;

[0028] Based on the minimized mean square error, a predicted second noise is determined.

[0029] In some embodiments of the present application, the mean square error is determined based on the following formula:

[0030]

[0031] Where L is the mean square error, ∈ t is the standard Gaussian noise corresponding to the t-th time step.

[0032] In a second aspect of the present application, an optical proximity effect correction device is provided, comprising:

[0033] An acquisition module is configured to acquire an original mask, wherein the original mask is a mask pattern designed based on a target wafer image;

[0034] A correction module is configured to input the original mask into a preset probability diffusion model to obtain a corrected target mask; wherein the preset probability diffusion model is trained based on original masks of different wafer images and corrected masks corresponding to the original masks;

[0035] The output module is configured to output the target mask to obtain a target wafer image based on the target mask.

[0036] According to a third aspect of the present application, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect when executing the computer program.

[0037] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0038] The technical solutions provided by the embodiments of the present application may have the following beneficial effects:

[0039] By introducing the probability diffusion model into the OPC method, end-to-end learning of the complex mapping relationship between the original mask and the corrected mask is achieved. Only one forward inference of the model is required to complete the correction, which greatly shortens the OPC processing time, achieving a leap from minutes to seconds or even milliseconds, improving the efficiency of OPC and greatly improving the throughput of semiconductor manufacturing.

[0040] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which constitute part of this document, are intended to provide a further understanding of this document. The exemplary embodiments and descriptions herein are intended to explain this document and do not constitute an improper limitation on this document. In the accompanying drawings:

[0042] Figure 1 1 is a flow chart of an optical proximity effect correction method according to an exemplary embodiment of the present application;

[0043] Figure 2 1 is a flow chart of a probability diffusion model training method according to an exemplary embodiment of the present application;

[0044] Figure 3 is a schematic diagram of a forward diffusion process shown in an exemplary embodiment of the present application;

[0045] Figure 4 is a schematic diagram of the reverse denoising process shown in an exemplary embodiment of the present application;

[0046] Figure 5 1 is a diagram showing multiple pairs of original masks and corresponding corrected mask images according to an exemplary embodiment of the present application;

[0047] Figure 6 1 is a schematic diagram showing how the loss function changes with the step size during training, as shown in an exemplary embodiment of the present application;

[0048] Figure 7 1 is a schematic diagram showing changes in training rounds during a training process according to an exemplary embodiment of the present application;

[0049] FIG8( a ) is a noise picture shown in an exemplary embodiment of the present application;

[0050] FIG8( b ) is a mask image after the first training round of training shown in an exemplary embodiment of the present application;

[0051] FIG8( c ) is a mask image after multiple training rounds of training shown in an exemplary embodiment of the present application;

[0052] Figure 9 1 is a schematic structural diagram of an optical proximity effect correction device according to an exemplary embodiment of the present application;

[0053] Figure 10 is a schematic structural diagram of a processing device shown in an exemplary embodiment of the present application;

[0054] Figure 11 It is a schematic diagram of the hardware structure of an electronic device shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other in any way.

[0056] A mask is a transparent substrate with a specific pattern made of opaque material. During the photolithography process, ultraviolet light passes through the mask and projects the pattern onto a silicon wafer coated with photoresist. After exposure, development and other steps, a circuit pattern consistent with the mask is formed on the silicon wafer. Finally, the circuit layer is processed through etching and other processes.

[0057] Because ultraviolet light has the diffraction and interference characteristics of light waves, when the light wave passes through the pattern on the mask, its wavefront will be distorted. According to the Huygens-Fresnel principle, when the spacing between adjacent patterns on the mask is close to or less than the exposure wavelength, the secondary sub-waves generated by each point light source will superimpose on each other, forming a complex intensity distribution. This effect causes the actual light intensity distribution on the wafer surface to deviate significantly from the mask design pattern.

[0058] For example, the line width in a dense or isolated environment may become narrower or wider, resulting in line width deviation; for example, the end of the line may be shorter than designed, resulting in line end shortening; for example, the sharp corners on the mask appear rounded on the wafer, resulting in corner rounding; for example, a slight change in the mask size may lead to nonlinear changes in the size on the wafer, resulting in nonlinear effects, etc.

[0059] These deviations need to be corrected and eliminated so that the mask can be used normally to obtain a preset specific pattern.

[0060] In the related art, it is difficult for existing OPC methods to simultaneously achieve the goals of high computational efficiency and high correction accuracy.

[0061] Based on this, the present application provides an optical proximity effect correction method, which utilizes the image generation and transformation capabilities of the probability diffusion model to achieve end-to-end mapping from the original mask pattern to the corrected mask pattern, effectively improving the correction quality of the original mask image and enhancing the adaptability to complex graphics. At the same time, it significantly improves the efficiency and accuracy of the OPC method and reduces the computational cost.

[0062] An exemplary embodiment of the present application provides an optical proximity effect correction method, referring to Figure 1 As shown, the method includes:

[0063] S101 , obtaining an original mask, where the original mask is a mask pattern designed based on a target wafer image.

[0064] Specifically, the target wafer image is a preset pattern, that is, the target pattern desired to be obtained after illuminating the mask. Therefore, a mask pattern can be designed based on the target wafer image as the original mask before optical proximity effect correction.

[0065] S102 , inputting the original mask into a preset probability diffusion model to obtain a corrected target mask; wherein the preset probability diffusion model is trained based on original masks of different wafer images and corrected masks corresponding to the original masks.

[0066] For example, the pre-set probability diffusion model consists of a forward diffusion process and a reverse denoising process. The forward diffusion process, during model training, gradually applies noise to a known, precisely calibrated mask until it becomes pure noise. This process allows the model to learn the noise distribution of the data. The reverse denoising process, on the other hand, is the core function of the model. During actual calibration, the model starts with random noise and, guided by the corresponding pattern of the original mask, gradually removes the noise, ultimately generating a corrected mask.

[0067] Therefore, by inputting the original mask of different wafer images and the corresponding pattern of the corrected mask corresponding to the original mask into the preset probability diffusion model, the preset probability diffusion model can be trained to improve the accuracy of the preset probability diffusion model in outputting the correction result.

[0068] S103 , outputting the target mask to obtain a target wafer image based on the target mask.

[0069] In this embodiment, unlike traditional OPC methods such as iterative optimization or physical models, a probability diffusion model is introduced into the OPC method to achieve end-to-end learning of the complex mapping relationship between the original mask and the corrected mask. Only one forward reasoning of the model is required to complete the correction, which greatly shortens the OPC processing time, achieving a leap from minutes to seconds or even milliseconds, improving the efficiency of OPC and greatly improving the throughput of semiconductor manufacturing.

[0070] This embodiment further describes the training process of the probability diffusion model in the above embodiment.

[0071] Reference Figure 2 As shown, the preset probability diffusion model can be trained in the following way:

[0072] S201 : Based on an original mask of a wafer image, determine first data corresponding to the original mask of the wafer image and second data corresponding to the corrected mask.

[0073] Specifically, the original mask of the wafer image and the corrected mask are both images. By converting the image into a data format, the data is easier to process and the data processing process is more accurate than directly processing the image, making the processing process of the same image repeatable, thereby achieving accuracy in model processing.

[0074] S202: Noise the first data according to a preset first noise to obtain third data.

[0075] For example, the preset first noise is added stepwise according to a preset time step, and the added noise may be Gaussian noise. After the first noise is added to the first data, the third noise is obtained, and the first noise is gradually added until the third data corresponds to an image of pure Gaussian noise.

[0076] S203 , performing denoising processing on the third data according to the predicted second noise to obtain second data; wherein the predicted second noise is confirmed based on the image segmentation model.

[0077] Exemplarily, the noise reduction process starts with the third data corresponding to the image of pure Gaussian noise, and the predicted second noise is also gradually removed according to a preset time step, and the noise is gradually reduced until the second data corresponding to the corrected mask is obtained.

[0078] Optionally, the predicted second noise can be determined by a learned neural network model, and accurate prediction of the second noise can be achieved using a known corrected mask. The basic network architecture used to predict noise can be a variety of learning models. This application adopts the classic image segmentation and image-to-image conversion network U-Net. Its encoder-decoder structure and skip connections enable it to effectively capture multi-scale features. The U-Net structure ensures that high-level semantic information can be utilized while retaining low-level spatial details during the denoising process.

[0079] In this embodiment, the accuracy of model training is improved by converting images into more easily processable data. The model training process is then completed by first forward-noising the data and then denoising the noisy data based on the secondary noise predicted by the corrected reticle. Paired original and corrected reticles are used to improve model training accuracy. By training on a large number of original / corrected reticle pairs, the model can capture nonlinear effects in the lithography process and achieve high-precision correction. Once trained, the inference phase requires only a single forward propagation, significantly improving OPC efficiency and far exceeding traditional iterative methods. The model also exhibits strong robustness to input noise. By learning from a large amount of real-world data, the model has the generalization capability to effectively correct new, unseen designs.

[0080] The embodiment of the present application further describes the forward diffusion process in detail:

[0081] Exemplarily, the step of adding noise to the first data according to a preset first noise to obtain third data includes:

[0082] From the current time step data x t-1 Add noise to the next time step data x t The probability distribution in is determined by the following formula:

[0083]

[0084] In the formula, q(x t |x t-1 ) is the transition probability distribution function of the noise processing, β t is the noise schedule corresponding to the preset first noise at the t-th time step, Indicates the mean The variance is β t Gaussian distribution, I is the unit matrix that ensures the independence of each dimension.

[0085] Let a t =1-β t , but:

[0086]

[0087] Therefore, we can directly sample x from any time step t t , and we get the following formula:

[0088]

[0089] Reference Figure 3 As shown, in this embodiment, by fixing the forward diffusion process to a first-order Markov chain, Gaussian noise is gradually added to the original data x0 within T time steps. At each time step t, the preset variance β is scheduled. t One step ahead data x t-1 Add noise until the final data x T The data is completely transformed into pure Gaussian noise data. This process enables the model to learn to recover the original data from the noise.

[0090] The present application further describes the inverse denoising process in detail:

[0091] Exemplarily, the denoising the third data according to the predicted second noise to obtain the second data includes:

[0092] From the current time step data x t-1 Denoise the data x at the next time step t The probability distribution in is determined by the following formula:

[0093] p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t))

[0094] Where p θ (x t-1 |x t ) is the transfer probability distribution function of the noise reduction process, N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)) indicates that the mean is μ θ (x t ,t), the variance is ∑ θ (x t ,t)’s Gaussian distribution.

[0095] Specifically, refer to the following derivation process:

[0096]

[0097]

[0098] Reference Figure 4 As shown in the figure, in this embodiment, the probability diffusion model attempts to fit the noise addition process in the forward process in the reverse process, with the goal of making the noise distribution predicted by the reverse denoising consistent with the distribution of the noise added in the forward process. The reverse denoising process is to obtain the noise from the pure noise data x T To begin, we learn the neural network model μ θ (x t ,t) gradually denoises until the original data x0 is restored. This process is also a Markov chain, but its transition probability p θ (x t-1 |x t ) is modeled by a neural network.

[0099] In some embodiments of the present application, the mean is determined based on the following formula:

[0100]

[0101] Where,∈ θ (x t ,t) is the predicted second noise corresponding to the t-th time step, α t is the attenuation coefficient of time step t, α t =1-β t , is the accumulation coefficient of forward diffusion,

[0102] In this embodiment, the neural network of the probability diffusion model fixes the variance when predicting noise, so the model only needs to predict the mean of the noise.

[0103] In some embodiments of the present application, a method for confirming the predicted second noise based on an image segmentation model includes:

[0104] The noisy mask image and the time step are input into the image segmentation model to determine a minimized mean square error; and a predicted second noise is determined based on the minimized mean square error.

[0105] For example, taking the basic network architecture U-Net as an image segmentation model, the input of U-Net is the image data x of the mask with noise t And the embedding of the current time step t (usually a sinusoidal position code), the output is the predicted second noise ε, and the iterative process of the model is to continuously reduce the mean square error.

[0106] In this embodiment, the encoder-decoder structure and skip connections of U-Net are utilized to enable it to effectively capture multi-scale features. Its multi-scale feature extraction capability and skip connections ensure the accurate recovery of mask pattern details and the understanding of the global structure, which is crucial for the correction of patterns of different sizes in the OPC method, and is particularly suitable for processing complex lithography patterns. The U-Net structure ensures that high-level semantic information can be utilized while retaining low-level spatial details during the denoising process.

[0107] In some embodiments of the present application, the mean square error is determined based on the following formula:

[0108]

[0109] Where L is the mean square error, ∈ t is the standard Gaussian noise corresponding to the t-th time step, x t is the result of adding noise ∈ to x0 at time step t.

[0110] In this embodiment, the definition of the mean square error is further determined to determine the minimum mean square error, thereby making the noise distribution of the reverse denoising prediction consistent with the distribution of the noise added in the forward process.

[0111] The contents described above can be implemented individually or in combination in various ways, and these variations are all within the scope of protection of this application.

[0112] The optical proximity effect correction method of the present application is further described below with one or several specific embodiments.

[0113] Dataset preparation:

[0114] Collect the original mask (M orig ) and the corresponding corrected mask (M corr ) sample pairs. These masks are usually binary black and white images, with different colors representing the presence or absence of the pattern. See Figure 5 The data used in this paper is from the open-source dataset in the paper: GAN-OPC: Mask Optimization With Lithography-Guided Generative Adversarial Nets.

[0115] Convert the original and corrected masks to a unified image format, such as PNG or Numpy array, and perform normalization, such as scaling pixel values ​​to between 0 and 1, to ensure consistent image resolution, such as 256×256 or 512×512 pixels, which can be adjusted based on process node and hardware performance.

[0116] Model Architecture:

[0117] U-Net backbone network: The standard U-Net architecture is used as the noise prediction network∈ θ . U-Net consists of an encoder, a decoder, and skip connections.

[0118] Encoder: It consists of multiple layers of convolutional blocks and downsampling layers, such as max pooling or strided convolution, to extract features at different scales. For example, there can be 4 downsampling layers.

[0119] Bottleneck layer: The connection between the encoder and decoder, usually containing one or more residual blocks.

[0120] Decoder: It consists of multiple layers of upsampling layers and convolution blocks, such as transposed convolution or bilinear interpolation followed by convolution, to restore the image resolution.

[0121] Skip connections: Skip connections are added from each layer of the encoder to the corresponding layer of the decoder to help the model retain detailed information.

[0122] Time-step embedding: To make the U-Net aware of the current time-step t, the time-step t is converted into a high-dimensional vector using sinusoidal positional encoding and incorporated into each downsampling / upsampling block of the U-Net, e.g., by addition or adaptive normalization.

[0123] Training process:

[0124] The time step T is set to 1000, the batch size is set to 64, the training rounds are set to 1000, the AdamW optimizer, the initial learning rate is 0.0001, and the noise schedule β t :Use linear scheduling β t Increased from 0.0004 to 0.02.

[0125] Randomly sample a corrected mask x0 from the data set, randomly sample a time step t∈[1,1000], and calculate x based on the forward diffusion process of x0 and t. t And the actual noise ∈ added, x t and t as input and fed into the U-Net model ∈ θ Prediction noise Calculating losses And back propagation is performed to update the model parameters. Through multiple trainings, the model learns to denoise the mask back to the corrected state under any given noise level. Figure 6 and Figure 7 It shows how the loss function changes with the step size and the number of training epochs during training.

[0126] Model input preparation: the original mask M to be corrected orig Convert to the same image format and resolution as the training data.

[0127] Inverse denoising process: From a pure Gaussian noise image x T At the beginning, corresponding to the highest noise level during training, iterate from time step 1000 to 1, and change the current image x t and time step t is input to the trained U-Net noise prediction network ∈ θ , using the predicted noise Calculate x by the reverse sampling formula t-1 , iteratively, gradually removing the noise until x0 is obtained. The final x0 is the mask corrected by the probability diffusion model.

[0128] The correction results are verified through simulation tools or actual wafer exposure experiments, and the deviations between the wafer images before correction, after traditional OPC correction, and after correction using the probability diffusion model DDPM of the present invention and the design targets, such as CD uniformity, line width control, corner sharpness, etc., are compared.

[0129] Figure 8(a) shows the initial noise input. As shown in Figure 8(a), the image is entirely noise, with no reticle pattern. After the first round of model training, the model's predicted output is visualized, yielding Figure 8(b). As shown in Figure 8(b), the reticle output is faintly visible, indicating poor model predictions, much of which is obscured by noise. After a sufficient number of training rounds, the model output is visualized, yielding Figure 8(c). As shown in Figure 8(c), the model produces valid output that closely resembles the original pattern, with no visible noise. This demonstrates that the model has effectively learned to recover the corrected reticle from the noise, conditional on the original pattern. Furthermore, the figure shows that the model's predicted output contains some scattered pixels or blocks, representing noise that can be filtered out using simple post-processing methods.

[0130] The following indicators were used to compare the consistency of the images corresponding to the original mask, the mask corrected by rule-based OPC in the prior art, and the mask corrected by OPC in this application with the designed target image, and the data in Tables 1 and 2 were obtained:

[0131] Table 1

[0132]

[0133] Table 2

[0134] index Total optimization time (min) Single graph average time (s) Speedup Rules-based OPC 134 6.7 1x This application OPC 32 1.6 4.2x

[0135] Among them, the CD error mean refers to the average deviation between the critical size and the corresponding size of the target image;

[0136] CD error 3σ refers to 3 times the standard deviation of CD error, reflecting the line width stability;

[0137] Corner fidelity refers to the degree of preservation of detected sharp corner areas;

[0138] The similarity of the graphic boundary refers to the overlap between the graphic boundary and the target image;

[0139] The total optimization time refers to the time spent on overall calculation and correction.

[0140] As can be seen from Table 1, the optimization results of the embodiment of the present application are significantly improved compared to the results without OPC correction. At the same time, compared with the traditional rule-based OPC method, it also has advantages in terms of indicators. As can be seen from Table 2, the embodiment of the present application also has a great advantage in optimization efficiency, achieving a 4.2-fold acceleration compared to the traditional method, and can train once and reuse multiple images.

[0141] The DDPM-based OPC method proposed in this application embodiment offers superior CD control accuracy, boundary reconstruction capabilities, and corner fidelity. In terms of computational time, the DDPM method, due to its parallel generation capabilities, can significantly shorten correction time. It is particularly well-suited for large-scale graphics generation and pre-silicon verification processes, and has great potential for industrial application.

[0142] In this embodiment, the model is trained on a large number of diverse original / corrected mask sample pairs, and can learn a rich variety of graphic correction patterns. Although model training requires certain computing resources, once training is completed, the computational overhead of the online inference stage is much lower than that of traditional iterative optimization methods. This helps reduce hardware investment costs in the chip design and manufacturing process, has stronger adaptability and generalization capabilities for various complex and irregular layout graphics, and reduces dependence on manual rules or model parameter adjustments.

[0143] refer to Figure 9 As shown, an exemplary embodiment of the present application further provides an optical proximity effect correction device 900, comprising:

[0144] The acquisition module 901 is configured to acquire an original mask, which is a mask pattern designed based on a target wafer image.

[0145] The correction module 902 is configured to input the original mask into a preset probability diffusion model to obtain a corrected target mask; wherein the preset probability diffusion model is trained based on the original mask of different wafer images and the corrected mask corresponding to the original mask.

[0146] The output module 903 is configured to output the target mask to obtain a target wafer image based on the target mask.

[0147] Optionally, the optical proximity effect correction device further includes a training module 904, wherein the training module 904 is specifically configured to:

[0148] Based on the original mask of the wafer image, determining first data corresponding to the original mask of the wafer image and second data corresponding to the corrected mask;

[0149] Noise-processing the first data according to a preset first noise to obtain third data;

[0150] Noise reduction processing is performed on the third data according to the predicted second noise to obtain second data;

[0151] The predicted second noise is confirmed based on an image segmentation model.

[0152] Furthermore, based on the above embodiment, the training module 904 is specifically configured to:

[0153] From the current time step data x t-1 Add noise to the next time step data x t The probability distribution in is determined by the following formula:

[0154]

[0155] In the formula, q(x t |x t-1 ) is the transition probability distribution function of the noise processing, β t is the preset first noise corresponding to the t-th time step, Indicates the mean The variance is β t Gaussian distribution, I is the unit matrix that ensures the independence of each dimension.

[0156] Furthermore, the training module 904 is specifically configured to:

[0157] From the current time step data x t-1 Denoise the data x at the next time step t The probability distribution in is determined by the following formula:

[0158]

[0159] Where p θ (x t-1 |x t ) is the transfer probability distribution function of the noise reduction process, N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)) indicates that the mean is μ θ (x t ,t), the variance is ∑ θ (x t ,t)’s Gaussian distribution.

[0160] The mean is determined based on the following formula:

[0161]

[0162] Where,∈ θ (x t ,t) is the predicted second noise corresponding to the t-th time step, α t is the attenuation coefficient of time step t, α t =1-β t , is the accumulation coefficient of forward diffusion,

[0163] In this embodiment, the training module 904 is specifically used to:

[0164] Inputting the noisy mask image and the time step into the image segmentation model to determine the minimum mean square error;

[0165] Based on the minimized mean square error, a predicted second noise is determined.

[0166] The mean square error is determined based on the following formula:

[0167]

[0168] Where L is the mean square error, ∈ t is the standard Gaussian noise corresponding to the t-th time step.

[0169] The optical proximity effect correction device provided in this embodiment can execute the optical proximity effect correction method of the above embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0170] In an embodiment of the present invention, the electronic device or main control device can be divided into functional modules according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present invention is schematic and is only a logical functional division. In actual implementation, there may be other division methods.

[0171] In a specific implementation of the aforementioned optical proximity effect correction device, each module may be implemented as a processor, and the processor may execute computer-executable instructions stored in a memory, so that the processor executes the aforementioned optical proximity effect correction method.

[0172] In summary, the application of the probability diffusion model to the processing device of the optical proximity effect correction method is described. Figure 10 As shown, the processing device includes an online reasoning module and an offline training module.

[0173] In the offline training module, a data set is constructed using multiple pairs of original masks and correction masks corresponding to the original masks, and then input into the DDPM model for training. The correction masks corresponding to the original masks are input into the noise prediction network in the DDPM model to predict the noise, thereby completing the training of the DDPM model.

[0174] In the online inference module, the original mask to be corrected is corrected using the DDPM model trained in the offline training module. After inverse denoising, the corrected mask is output.

[0175] refer to Figure 11 As shown, the electronic device 1100 includes:

[0176] At least one processor 1101 and memory 1102 .

[0177] The electronic device further includes a communication component 1103 .

[0178] The processor 1101 , the memory 1102 and the communication component 1103 are connected via a bus 1104 .

[0179] In a specific implementation process, at least one processor 1101 executes the computer-executable instructions stored in the memory 1102 , so that the at least one processor 1101 executes the optical proximity effect correction method as described above.

[0180] The specific implementation process of the processor 1101 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0181] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules in the processor.

[0182] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage.

[0183] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be classified as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0184] The above describes the solutions provided by the embodiments of the present invention with respect to the functions implemented by the electronic device and the main control device.

[0185] It is understandable that, in order to realize the above functions, the electronic device or the main control device includes the corresponding hardware structure and / or software module for executing each function.

[0186] In combination with the units and algorithm steps of each example described in the embodiments disclosed in the embodiments of the present invention, the embodiments of the present invention can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in hardware or in a manner driven by computer software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiments of the present invention.

[0187] The present application also provides a computer program product, comprising a computer program, which implements an optical proximity effect correction method when executed by a processor.

[0188] The computer program product provided in this embodiment can execute the optical proximity effect correction method of the above embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0189] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above optical proximity effect correction method is implemented.

[0190] The computer-readable storage medium provided in this embodiment can execute the optical proximity effect correction method of the above embodiment. Its implementation principle and technical effects are similar and will not be described in detail in this embodiment.

[0191] The computer-readable storage medium mentioned above can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0192] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatuses, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0193] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0194] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0195] In this application, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such article or device. In the absence of further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the article or device comprising the element.

[0196] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0197] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if such changes and modifications of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such changes and modifications.

Claims

1. An optical proximity effect correction method, characterized in that: include: Obtaining an original mask, wherein the original mask is a mask pattern designed based on a target wafer image; Inputting the original mask into a preset probability diffusion model to obtain a corrected target mask; wherein the preset probability diffusion model is trained based on original masks of different wafer images and corrected masks corresponding to the original masks; The target mask is output to obtain a target wafer image based on the target mask.

2. The optical proximity effect correction method according to claim 1, wherein: The preset probability diffusion model is trained in the following way: Based on the original mask of the wafer image, determining first data corresponding to the original mask of the wafer image and second data corresponding to the corrected mask; Noise-processing the first data according to a preset first noise to obtain third data; performing noise reduction processing on the third data according to the predicted second noise to obtain second data; The predicted second noise is confirmed based on an image segmentation model.

3. The optical proximity effect correction method according to claim 2, wherein: The step of adding noise to the first data according to a preset first noise to obtain third data includes: From the current time step data x t-1 Add noise to the next time step data x t The probability distribution in is determined by the following formula: In the formula, q(x t |x t-1 ) is the transition probability distribution function of the noise processing, β t is the noise schedule corresponding to the preset first noise at the t-th time step, Indicates the mean The variance is β t Gaussian distribution, I is the unit matrix that ensures the independence of each dimension.

4. The optical proximity effect correction method according to claim 3, wherein: The denoising process on the third data according to the predicted second noise to obtain the second data includes: From the current time step data x t-1 Denoise the data x at the next time step t The probability distribution in is determined by the following formula: p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)) Where p θ (x t-1 |x t ) is the transfer probability distribution function of the noise reduction process, M(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)) indicates that the mean is μ θ (x t ,t), the variance is ∑ θ (x t ,t)’s Gaussian distribution.

5. The optical proximity effect correction method according to claim 4, wherein: The mean is determined based on the following formula: Where,∈ θ (x t ,t) is the predicted second noise corresponding to the t-th time step, α t is the attenuation coefficient of time step t, α t =1-β t , is the accumulation coefficient of forward diffusion, 6. The optical proximity effect correction method according to any one of claims 2 to 5, wherein: The method for confirming the predicted second noise based on the image segmentation model includes: Inputting the noisy mask image and the time step into the image segmentation model to determine the minimum mean square error; Based on the minimized mean square error, a predicted second noise is determined.

7. The optical proximity effect correction method according to claim 6, wherein: The mean square error is determined based on the following formula: Where L is the mean square error, ∈ t is the standard Gaussian noise corresponding to the t-th time step.

8. An optical proximity effect correction device, characterized in that: include: An acquisition module is configured to acquire an original mask, wherein the original mask is a mask pattern designed based on a target wafer image; a correction module configured to input the original mask into a preset probability diffusion model to obtain a corrected target mask; The preset probability diffusion model is obtained by training based on original masks of different wafer images and corrected masks corresponding to the original masks; The output module is configured to output the target mask to obtain a target wafer image based on the target mask.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images