Optical proximity effect correction method and device

Abstract

The invention provides an optical proximity effect correction method and an optical proximity effect correction device, which can perform optical proximity effect correction processing with high resolution and high efficiency. The optical proximity effect correction method comprises the following steps: acquiring a to-be-processed layout image in a first format; preprocessing the layout image in the first format to obtain a layout image in a second format; inputting the layout image in the second format into a trained Brownian bridge diffusion model, performing optical proximity effect correction processing to generate an optimized mask image in the second format, the Brownian bridge diffusion model being trained to learn optical proximity effect correction processing from a layout domain to a mask domain, establishing mapping from a layout domain to a mask domain based on a Brownian bridge process; and post-processing the optimized mask image in the second format to obtain an optimized mask image in the first format.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor chip manufacturing technology, and in particular to an optical proximity effect correction method and apparatus. Background Technology

[0002] Photolithography is the most important and expensive process in semiconductor chip manufacturing, accounting for approximately 30% of the cost and 50% of the time. Its precision determines the chip's process and device performance. However, when the critical dimensions of the design pattern are smaller than the exposure wavelength, the imaging on the silicon wafer becomes severely distorted, leading to significant differences between the transferred pattern and the design layout, requiring resolution enhancement techniques for compensation. Optical Proximity Correction (OPC) can pre-compensate the mask pattern, ensuring the silicon wafer pattern meets design requirements. Advanced OPC solutions based on machine learning can achieve a better balance between computational cost and fidelity than traditional OPC. At advanced nodes, large-scale integrated circuit design flows place higher demands on OPC technology in terms of speed, process stability, and result accuracy. Traditional model-based methods struggle to balance computational cost and fidelity, while emerging machine learning-based OPC methods often lack efficient model inference algorithms and high-resolution generation methods for complex patterns, thus weakening the speed advantage of machine learning-based OPC to some extent. Summary of the Invention

[0003] This disclosure was made to solve the above-mentioned problems in the prior art, and its purpose is to provide an optical proximity effect correction method and apparatus that can perform optical proximity effect correction processing with high resolution and high efficiency.

[0004] This disclosure provides an optical proximity effect correction method, which includes the following steps:

[0005] Obtain the first format layout image to be processed;

[0006] The first format layout image is preprocessed to obtain the second format layout image;

[0007] The second-format layout image is input into a trained Brownian bridge diffusion model, and optical proximity correction processing is performed to generate an optimized mask image in the second format. The Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, establishing a mapping from the layout domain to the mask domain based on the Brownian bridge process.

[0008] The optimized mask image in the second format is post-processed to obtain the optimized mask image in the first format.

[0009] Optionally, the above optical proximity correction method further includes using an existing training dataset consisting of a layout-mask data pair composed of a layout image and a mask image after optical proximity correction to train the Brownian bridge diffusion model.

[0010] Optionally, in the above optical proximity effect correction method, the training process of the Brownian bridge diffusion model includes a forward diffusion process and a reverse diffusion process. In the forward diffusion process, noise is gradually superimposed on the mask image in the training dataset until a pattern image paired with it is obtained. In the reverse diffusion process, the mask image at a certain training moment is input into the neural network, and the neural network predicts the noise superimposed at that training moment to obtain the predicted noise. The parameters of the neural network are updated based on the error between the predicted noise and the actual noise.

[0011] Optionally, in the above optical proximity effect correction method, in the step of generating the optimized mask image of the second format using the Brownian bridge diffusion model, for the input layout image of the second format, an inverse denoising inference process is performed to generate the target mask image.

[0012] Optionally, in the above optical proximity effect correction method, the reasoning process is set as a skip-sampling deMarkov process.

[0013] Optionally, in the above optical proximity effect correction method, the parameters characterizing the diversity of the output are reduced during the inference process.

[0014] Optionally, in the above optical proximity effect correction method, the first format is a GDS file format and the second format is a grayscale image format.

[0015] Optionally, in the above optical proximity effect correction method, the preprocessing further involves cropping the layout image in the first or second format to obtain a layout clip, the trained Brownian bridge diffusion model processes the layout clip to output an optimized mask clip, and the postprocessing further involves splicing the optimized mask clip in the first or second format to obtain an optimized mask image.

[0016] This disclosure also provides an optical proximity correction device, comprising: a layout image acquisition module for acquiring a layout image of a first format to be processed; a preprocessing module for preprocessing the layout image of the first format to obtain a layout image of a second format; a mask image generation module for inputting the layout image of the second format into a trained Brownian bridge diffusion model to perform optical proximity correction processing to generate a target mask image, wherein the Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, and establishes a mapping from the layout domain to the mask domain based on the Brownian bridge process; and a post-processing module for post-processing the optimized mask image of the second format to obtain an optimized mask image of the first format.

[0017] This disclosure also provides an electronic device, comprising: a memory storing a computer program; and a processor communicatively connected to the memory for executing the computer program stored in the memory to implement the above-described optical proximity effect correction method.

[0018] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, is used to implement the above-described optical proximity effect correction method.

[0019] According to the optical proximity correction method and apparatus disclosed herein, optical proximity correction can be performed with high resolution and high efficiency by employing a Brownian bridge diffusion model that is trained to learn optical proximity correction processing from the layout domain to the mask domain and establishes a mapping from the layout domain to the mask domain based on the Brownian bridge process. Attached Figure Description

[0020] This disclosure can be better understood by describing exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, in which:

[0021] Figure 1 A flowchart illustrating an optical proximity effect correction method according to one embodiment of this disclosure;

[0022] Figure 2 This is a schematic diagram illustrating the forward and reverse diffusion processes of the Brownian bridge diffusion model in an optical proximity effect correction method according to one embodiment of the present disclosure.

[0023] Figure 3 A flowchart illustrating an example of the training process of the Brownian bridge diffusion model in an optical proximity effect correction method according to one embodiment of this disclosure;

[0024] Figure 4To illustrate the structural block diagram of the optical proximity effect correction device according to other embodiments of this disclosure;

[0025] Figure 5 This is a structural block diagram illustrating a computer-readable storage medium involved in other embodiments of the present disclosure;

[0026] Figure 6 This is a structural block diagram illustrating an electronic device according to other embodiments of the present disclosure. Detailed Implementation

[0027] The following describes specific embodiments of this disclosure. It should be noted that, in order to maintain brevity, this specification cannot provide a detailed description of all features of the actual embodiments. It should be understood that, in the actual implementation of any embodiment, just as in any engineering or design project, various specific decisions are often made to achieve the developer's specific goals and to meet system-related or business-related constraints, and this can change from one embodiment to another. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content of this disclosure, changes in design, manufacturing, or production based on the technical content disclosed herein are merely conventional technical means and should not be construed as insufficient content of this disclosure.

[0028] Unless otherwise defined, the technical or scientific terms used in the claims and description shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this patent application description and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the element or object preceding “comprising” or “including” encompasses the element or object listed following “comprising” or “including” and its equivalents, and do not exclude other elements or objects. The terms “connected” or “linked” and similar terms are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0029] In this disclosure, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0030] Unless otherwise specified, all embodiments and preferred embodiments mentioned herein can be combined to form new technical solutions. Similarly, unless otherwise specified, all technical features and preferred features mentioned herein can be combined to form new technical solutions.

[0031] As a classic diffusion model, the Denoise Diffusion Probabilistic Model (DDPM) is a parameterized Markov chain consisting of two processes: forward diffusion and reverse diffusion.

[0032] During the forward diffusion process, a series of normally distributed noise points are superimposed on the initial data x0 sampled from the training set, ultimately forming an implicit variable x composed of normally distributed noise points. T The Markov chain is represented by the following formula (1).

[0033]

[0034] The backdiffusion process can be represented as a joint distribution, given the final data x. T The distribution of the initial data x0 is predicted by constructing another Markov chain, which is expressed as Equation (2) below.

[0035]

[0036] From a macro perspective, a well-trained DDPM can only generate images of the target domain by progressively denoising noisy images. Moreover, DDPM relies on conditional inputs during the generation process, which leads to low model efficiency and theoretical uncertainty in ensuring the final output distribution.

[0037] This disclosure considers a Brownian bridge diffusion model based on the Brownian bridge process, building upon the DDPM model. The Brownian bridge process is a continuous-time stochastic model in which the probability distribution of the intermediate process depends on the initial and final states, as expressed in the following formula (3).

[0038]

[0039] Where t represents the current time, T represents the total number of time steps in the diffusion process, x0 represents the initial data, and x T Indicates the final data, x t This represents the data at the current moment.

[0040] A Brownian bridge can be understood as a bridge connecting given information to target information. General diffusion models infer the final result from Gaussian noise. The Brownian bridge diffusion model allows inference to proceed from a specific or given set of information, rather than from Gaussian noise. In this disclosure, the layout image of the layout domain and the mask image of the mask domain are fixed at both ends of the Brownian bridge process, enabling the Brownian bridge diffusion model to learn optical proximity correction processing from the layout domain q(z) to the mask domain q(M).

[0041] In this disclosure, optical proximity correction can be performed with high resolution and high efficiency by employing a Brownian bridge diffusion model that is trained to learn optical proximity correction processing from the layout domain to the mask domain and establishes a mapping from the layout domain to the mask domain based on the Brownian bridge process.

[0042] The embodiments of this disclosure will now be described with reference to the accompanying drawings. It should be understood that the embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0043] Figure 1 A flowchart illustrating an optical proximity effect correction method according to one embodiment of this disclosure is provided. Figure 1 As shown, an embodiment of the present disclosure relates to an optical proximity effect correction method, which includes steps S11 to S14.

[0044] In step S11, a layout image in a first format to be processed is obtained. In some embodiments of this disclosure, the layout image in the first format may be a layout image in GDS file format.

[0045] In step S12, the layout image in the first format is preprocessed to obtain a layout image in the second format. In some embodiments of this disclosure, the layout image in GDS file format can be converted into a layout image in grayscale format with a resolution of, for example, 1 nm / pixel.

[0046] In step S13, the second-format layout image is input into the trained Brownian bridge diffusion model, and optical proximity correction processing is performed to generate an optimized mask image in the second format. The Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, establishing a mapping from the layout domain to the mask domain based on the Brownian bridge process.

[0047] In step S14, the optimized mask image in the second format is post-processed to obtain the optimized mask image in the first format. In some embodiments of this disclosure, the optimized mask image in grayscale format can be converted into an optimized mask image in GDS file format.

[0048] In other embodiments of this disclosure, the layout image in the first or second format may be cropped during preprocessing to obtain a layout clip. A trained Brownian bridge diffusion model processes the layout clip and outputs an optimized mask clip. Then, the optimized mask clip in the first or second format may be stitched together during post-processing to obtain an optimized mask image.

[0049] Furthermore, in some embodiments of this disclosure, the optimized mask image obtained in the first or second format can also be used for photolithography simulation verification.

[0050] In some embodiments of this disclosure, the Brownian bridge diffusion model is trained using an existing training dataset comprising layout-mask data pairs consisting of paired and associated layout images and mask images processed by optical proximity correction. This training dataset may be publicly available, for example, obtained from publicly available data of other model-based OPC methods or machine learning-based OPC methods. In the layout-mask data pairs, the mask image is the OPC-processed mask image corresponding to the layout image.

[0051] The training process of the Brownian bridge diffusion model includes a forward diffusion process and a reverse diffusion process. Figure 2 The diagram illustrates the forward and reverse diffusion processes of the Brownian bridge diffusion model. Figure 2 As shown, M0~q(M0) represents a mask image M0 sampled as a sample from the true mask image distribution q(M0) of the training dataset, Z0 / M T ~q(Z0) represents a sample layout image Z0 taken from the true layout image distribution q(Z0) of the training dataset. This layout image Z0 is also considered as the mask image M after forward diffusion. T The layout image Z0 is paired with the mask image M0 to form a layout-mask data pair.

[0052] During the forward diffusion process, given a total number of time steps T, noise is progressively superimposed on the initial mask image M0 to obtain images M1, M2, ..., M0 in sequence.T-1 and M T (That is, Z0). During the reverse diffusion process, the layout image Z0 / M... T By progressively performing reverse denoising operations, the initial mask image M0 can be obtained. During the reverse diffusion process, the mask image at a certain training time can be input into the neural network. The neural network predicts the noise superimposed at that training time, obtaining the predicted noise. The parameters of the neural network are updated based on the error between the predicted noise and the actual noise.

[0053] Therefore, based on the aforementioned forward and reverse diffusion processes, the Brownian bridge diffusion model can be trained to learn the optical proximity effect correction process from the layout domain q(Z) to the mask domain q(M), and a mapping from the layout domain to the mask domain can be established based on the Brownian bridge process.

[0054] Figure 3 This is a flowchart illustrating an example of the training process of the Brownian bridge diffusion model in the optical proximity effect correction method according to embodiments of this disclosure. Figure 3 As shown, the training process includes steps S21 to S30.

[0055] In step S21, the initial neural network structure and the total number of time steps T in the diffusion process are defined.

[0056] In step S22, paired and associated mask images and layout images are sampled from the training dataset. The mask images are obtained by progressively adding noise through a forward diffusion process to obtain the paired and associated layout images.

[0057] In step S23, a training moment is randomly selected from the time series from 1 to the total number of time steps in the diffusion process. In some embodiments of this disclosure, the total number of time steps in the diffusion process is set to T, and a training moment t is randomly selected from the time series {1,2,…T}.

[0058] In step S24, a noise source is randomly generated as the actual noise. In some embodiments of this disclosure, a Gaussian noise source can be randomly sampled from a normal distribution N(0, 1).

[0059] In step S25, the actual noise is used to perform a forward diffusion process to obtain the mask image at the training time.

[0060] In some embodiments of this disclosure, if the training time is t, the total number of time steps in the diffusion process is T, the actual noise is noise ∈ randomly sampled from a normal distribution, the mask image is M0, the layout image is Z, and the mask image at training time t is M t Then M t Calculated using the following formula (4):

[0061]

[0062] in,

[0063] In step S26, the mask image at the training time is input into the neural network. The neural network predicts the noise superimposed at that training time (i.e., the noise that should be denoised at that training time during backdiffusion), thus obtaining the predicted noise. In some embodiments of this disclosure, the mask image M at training time t is used... t The input is fed into a neural network, which fits the backdiffusion process and outputs a predicted noise ∈ θ(Mt,t), where θ is a learnable parameter of the neural network.

[0064] Furthermore, this disclosure does not limit the type of neural network, and network models such as Noise Conditional Scoring Network (NCSN)++, Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and U-Net Neural Network can be selected.

[0065] In step S27, a training loss function is constructed based on the error between the predicted noise and the actual noise, the loss function is optimized, and the parameters of the neural network are updated.

[0066] In some embodiments of this disclosure, the prediction noise of the neural network is ∈θ(Mt,t), the learnable parameter of the neural network is θ, and the loss function is optimized by gradient descent. The gradient of the loss function is calculated by the following formula (5):

[0067]

[0068] In step S28, it is determined whether the training stopping condition is met. If the training stopping condition is not met, for example, the error between the predicted noise and the actual noise does not meet the specified requirements, then the process returns to step S22 to iteratively update the learnable parameters of the neural network. After multiple iterations, if the training stopping condition is met, for example, the error between the predicted noise and the actual noise has reached the specified requirements and the model's performance no longer significantly improves, then training can be stopped.

[0069] In step S29, the updated neural network structure is saved.

[0070] In step S30, the training process ends.

[0071] In some embodiments of this disclosure, the disclosed dataset may also include a test dataset that is different from the training dataset, which can be used to verify the generalization ability (i.e., the model’s performance on new data) of the trained Brownian bridge diffusion model.

[0072] After training through the above process, the Brownian bridge diffusion model can be used to generate optimized mask images based on the layout image to be processed. By inputting the layout image to be processed into the trained Brownian bridge diffusion model and performing the inverse denoising inference process, an optimized mask image can be generated.

[0073] In some embodiments of this disclosure, the inference process can be configured as a stepwise sampling Markov process. That is, during the inference process, the preceding and following processes are coupled together, M t The state can only be determined by M t-1 The state is obtained.

[0074] Specifically, for example, suppose the total number of time steps in the back diffusion process is the same as the total number of time steps T in the forward diffusion process, and the mask image M at time T... T Let Z0 be the layout image to be processed, and M be the mask image at time t. t The mask image at time t-1 is M. t-1 If the mask image at time 0 is the optimized mask image M0, then M t-1 Calculated using the following formula (6):

[0075]

[0076] During the inference (sampling) process, the inverse denoising operation is performed step by step from time t to 1 to generate the optimized mask image M0.

[0077] Among them, c mt It is the mask intermediate state weight, c zt It is the objective condition weight, c ∈t These are the denoising weights, where ∈θ(Mt,t) represents the prediction noise of the neural network, ω represents Gaussian noise, and when t=1, ω=0. It satisfies the following formula (7):

[0078]

[0079] in,

[0080] In some embodiments, c mt c zt c ∈t They respectively satisfy the following formula (8):

[0081]

[0082]

[0083]

[0084] The above description illustrates the case where the inference process is a stepwise sampling Markov process. However, in this inference (sampling) scheme, at each step of the sampling process, a noisy image of the same size as the layout image needs to be sampled, with pixel values ​​following a normal distribution. Such a sampling process consumes significant computational resources, especially when T is sufficiently large and the image resolution is high. Therefore, in some embodiments of this disclosure, in step S13, the inference process can also be set as a skip-step sampling de-Markov process. That is, during the inference process, assuming M... t It is related to an earlier state.

[0085] Specifically, for example, suppose the total number of time steps S in the back diffusion process is less than the total number of time steps T in the forward diffusion process, and construct a sub-time series {τ1, τ2, ..., τ...} of length S. n-1 , τ n ,...,τ s}, the mask image MT at time T is the layout image Z0 to be processed, and the mask image MT at time τ is the layout image Z0 to be processed. n The mask image is Mτ n , time τ n-1 The mask image is Mτ n-1 The mask image at time 0 is the optimized mask image M0, and the target distribution of the Markov removal process of skip sampling is represented by the following formula (9):

[0086]

[0087] Where, τ n -τ n-1 ≥1, I is the identity matrix, representing the parameters of output diversity. Furthermore, S is a hyperparameter similar to the learning rate or batch size in machine learning methods, which can be obtained through parameter tuning.

[0088] As mentioned above, by employing a non-Markov sampling path and selectively skipping certain sampling steps, each node is decoupled during inference, resulting in a significant acceleration of sampling. This acceleration is particularly pronounced when T and S differ significantly. Furthermore, the introduction of the new sampling path does not alter the training method and objectives of the latent diffusion model, nor does it affect the sampling quality.

[0089] Furthermore, in some embodiments of this disclosure, the parameters representing the diversity of the output can be reduced during the inference process. For example, when the inference process is set as a skip-sampling de-Markov process, the parameters described above can be reduced.

[0090] To address the output instability issue of machine learning-based optical proximity correction methods, this disclosure reduces output diversity and improves mask output stability by lowering the parameters characterizing output diversity. This avoids the need to process the same input layout multiple times and select the best result, further improving the speed of optical proximity correction processing.

[0091] According to the aforementioned optical proximity correction method, by employing a Brownian bridge diffusion model trained to learn optical proximity correction processing from the layout domain to the mask domain and establishing a mapping from the layout domain to the mask domain based on the Brownian bridge process, optical proximity correction processing can be performed with high resolution and high efficiency.

[0092] Furthermore, by combining the trained Brownian bridge diffusion model with a de-Markovianized skip sampling method and reducing output diversity, not only is the processing speed greatly improved, but the printability of the mask is also enhanced.

[0093] Below, the optical proximity effect correction method disclosed herein is compared with the model-based OPC method and the previous machine learning-based OPC methods PGAN-OPC and Neural-ILT in terms of three indicators: mean square error (L2 loss), process variable band (PVB), and turnaround time (TAT). The verification test set consists of ten test samples from ICCAD'13, and the comparison structure is shown in Table 1 below.

[0094] Table 1 shows the comparison results with different OPC models.

[0095]

[0096] As shown in Table 1, the optical proximity effect correction method of this disclosure significantly improves the PVB and TAT metrics, and these improvements are not achieved at the expense of a substantial increase in L2 loss. When evaluated using printability metrics (L2 loss + PVB), the architecture of this disclosure outperforms all previous methods on the ICCAD'13 test set. Specifically, compared to model-based OPC methods and previous machine learning-based OPC methods PGAN-OPC and Neural-ILT, this disclosure achieves a 50x, 93x, and 3x speedup in turnaround time, while reducing process variation bands by 29.7%, 8.0%, and 10.5%, respectively, and provides competitive L2 loss results.

[0097] Furthermore, since the deMarkovimation skip sampling (inference) of the Brownian bridge diffusion model is a denoising process for the entire layout clip image, without involving reverse lithography iterative optimization, the inference speed is independent of the complexity of the layout details and only depends on the final sampling step size and image resolution. Therefore, for a fixed number of sampling steps, i.e., different test data with the same resolution, the turnaround time remains the same.

[0098] In addition, other embodiments of this disclosure also provide an optical proximity effect correction device 40. Figure 4 A block diagram illustrating the structure of the optical proximity effect correction device 40 is shown below. Figure 4 As shown, the optical proximity correction device 40 includes: a layout image acquisition module 41 for acquiring a layout image of a first format to be processed; a preprocessing module 42 for preprocessing the layout image of the first format to obtain a layout image of a second format; a mask image generation module 43 for inputting the target layout image of the second format into a trained Brownian bridge diffusion model to perform optical proximity correction processing to generate an optimized mask image of the second format, wherein the Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, and establishes a mapping from the layout domain to the mask domain based on the Brownian bridge process; and a post-processing module 44 for post-processing the optimized mask image of the second format to obtain an optimized mask image of the first format.

[0099] In some embodiments of this disclosure, the optimized mask image obtained in the first or second format can also be used for photolithography simulation verification.

[0100] In addition, other embodiments of this disclosure also provide a computer-readable storage medium 50. Figure 5 A block diagram illustrating the structure of the computer-readable storage medium 50 is shown below. Figure 5 As shown, the computer-readable storage medium 50 stores a computer program 51, which, when executed by a processor, is used to implement the aforementioned optical proximity effect correction method. The computer-readable storage medium 50 can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or semiconductor system or propagation medium. The computer-readable storage medium 50 may also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disk. Optical disks may include optical disc-read-only memory (CD-ROM), optical disc-read / write (CD-RW), and DVD.

[0101] Other embodiments of this disclosure also provide an electronic device 60. Figure 6 To illustrate the structural block diagram of the electronic device 60. (See attached diagram.) Figure 6As shown, the electronic device 60 includes a memory 61 and a processor 62. The memory 61 stores a computer program, and the processor 62 is communicatively connected to the memory 61 to execute the computer program stored in the memory 61 to implement the above-mentioned optical proximity effect correction method.

[0102] Processor 62 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0103] Memory 61 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Memory 61 may also be internal memory of the random access memory (RAM) type.

[0104] The memory 61 and processor 62 can be integrated into one or more independent circuits or hardware, such as application-specific integrated circuits (ASICs). It should be noted that when the computer program in the memory 61 is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the various embodiments of this disclosure.

[0105] It should be understood that the above description is illustrative and not restrictive. For example, the above embodiments (and / or aspects thereof) can be used in combination with each other. Furthermore, many modifications can be made to adapt particular situations or materials to the teachings of the various embodiments of this disclosure without departing from the scope of this disclosure. While the dimensions and types of materials described herein are used to define parameters of the various embodiments of this disclosure, the embodiments are not intended to be restrictive but are exemplary. Many other embodiments will become apparent to those skilled in the art upon reading the above description. Therefore, the scope of the various embodiments of this disclosure should be determined by reference to the appended claims and the full scope of their equivalents.

Claims

1. A method for correcting the optical proximity effect, characterized in that, Includes the following steps: Obtain the first format layout image to be processed; The first format layout image is preprocessed to obtain the second format layout image; The second format layout image is input into a trained Brownian bridge diffusion model, and optical proximity correction processing is performed to generate an optimized mask image in the second format. The Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, and a mapping from the layout domain to the mask domain is established based on the Brownian bridge process. and The optimized mask image in the second format is post-processed to obtain the optimized mask image in the first format.

2. The optical proximity effect correction method as described in claim 1, characterized in that, The Brownian bridge diffusion model is trained using an existing training dataset consisting of layout images and mask images corrected for optical proximity.

3. The optical proximity effect correction method as described in claim 2, characterized in that, The training process of the Brownian bridge diffusion model includes a forward diffusion process and a reverse diffusion process. During the forward diffusion process, noise is progressively superimposed on the mask images in the training dataset until a paired and associated layout image is obtained. In the backdiffusion process, a mask image at a certain training moment is input into a neural network. The neural network predicts the noise superimposed at that training moment to obtain the predicted noise. The parameters of the neural network are updated based on the error between the predicted noise and the actual noise.

4. The optical proximity effect correction method as described in claim 1, characterized in that, In the step of generating the optimized mask image of the second format using the Brownian bridge diffusion model, for the input layout image of the second format, an inverse denoising inference process is performed to generate the optimized mask image.

5. The optical proximity effect correction method as described in claim 4, characterized in that, The reasoning process is set as a skip-sampling deMarkov process.

6. The optical proximity effect correction method as described in claim 4, characterized in that, During the inference process, parameters that represent the diversity of outputs are reduced.

7. The optical proximity effect correction method as described in claim 1, characterized in that, The first format is the GDS file format. The second format is a grayscale image format.

8. The optical proximity effect correction method as described in claim 1, characterized in that, The preprocessing also includes cropping the layout image in either the first or second format to obtain a layout clip. The trained Brownian bridge diffusion model processes the layout clipping and outputs an optimized mask clipping. The post-processing also involves splicing the optimized mask clips of the first or second format to obtain an optimized mask image.

9. An optical proximity effect correction device, characterized in that, include: The layout image acquisition module is used to acquire the layout image in the first format to be processed; The preprocessing module is used to preprocess the layout image in the first format to obtain the layout image in the second format. A mask image generation module is used to input the second format layout image into a trained Brownian bridge diffusion model, perform optical proximity correction processing to generate an optimized mask image in the second format, wherein the Brownian bridge diffusion model is trained to learn optical proximity correction processing from the layout domain to the mask domain, and establishes a mapping from the layout domain to the mask domain based on the Brownian bridge process. and The post-processing module is used to post-process the optimized mask image of the second format to obtain the optimized mask image of the first format.

10. An electronic device, characterized in that, include: A memory, wherein a computer program is stored; and A processor, communicatively connected to the memory, is configured to execute a computer program stored in the memory to implement the optical proximity effect correction method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, is used to implement the optical proximity effect correction method as described in any one of claims 1 to 8.

Images