Optical proximity correction method, simulation model, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to an optical proximity correction method, a simulation model, electronic equipment and a storage medium. The method comprises the following steps: performing optical proximity correction on a first part in a layout to generate a first correction part; determining an initial offset value for performing optical proximity correction on the original layout data of the second part based on the original layout data of the first part and the correction layout data of the first correction part; optical proximity correction is performed on the second portion based on the initial offset value to generate a second corrected portion. According to the technical scheme, the calculation efficiency can be remarkably improved, the applicability is enhanced, and the training cost can be reduced.

Description

Technical Field

[0001] Embodiments of the present disclosure generally relate to integrated circuits, and more particularly, to optical proximity correction methods, simulation models, electronic devices, and storage media. Background Art

[0002] As semiconductor manufacturing processes advance to the nanometer scale, the diffraction effect of light during photolithography causes significant deviations between the mask pattern and the actual wafer pattern. Optical proximity correction (OPC) is widely used in semiconductor manufacturing to reduce the discrepancy between wafer and mask patterns. OPC compensates for distortion by adjusting the mask pattern, with bias correction being a core method. This includes rule-based bias correction, model-based bias correction, and bias correction value prediction using artificial intelligence (AI).

[0003] Rule-based deviation correction applies a fixed offset to specific patterns (such as line width and spacing) based on empirical formulas, which cannot adapt to the nonlinear effects of complex patterns and advanced processes. Model-based deviation correction iteratively optimizes mask patterns through lithography simulation models and is suitable for advanced processes. However, as accuracy increases, the computational complexity increases, and multiple rounds of iterative simulation are required to converge. While using AI to learn the mapping relationship between graphic features and deviations in historical data and directly output correction values can significantly improve the efficiency of deviation correction, it faces the problems of high training data acquisition costs, strong data dependence, and limited generalization capabilities.

[0004] Therefore, how to accelerate deviation correction has become a technical problem that needs to be solved urgently in the field of electronic design automation (EDA). Summary of the Invention

[0005] According to example embodiments of the present disclosure, an optical proximity correction scheme is provided to at least partially overcome the above or other potential drawbacks.

[0006] According to one aspect of the present disclosure, an optical proximity correction method is provided. The method includes: performing optical proximity correction on a first portion of a layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the original layout data of a second portion based on the original layout data of the first portion and the corrected layout data of the first corrected portion; and performing optical proximity correction on the second portion based on the initial offset value to generate a second corrected portion.

[0007] In a second aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor; and a memory coupled to the processor, the memory having instructions stored therein, which, when executed by the processor, cause the device to perform actions, the actions comprising: performing optical proximity correction on a first portion of a layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the original layout data of a second portion based on the original layout data of the first portion and the corrected layout data of the first corrected portion; and performing optical proximity correction on the second portion based on the initial offset value to generate a second corrected portion.

[0008] In some embodiments, determining an initial offset value for performing optical proximity correction on the original layout data of the second portion based on the original layout data of the first portion and the corrected layout data of the first correction portion includes: determining first training data based on the original layout data of the first portion and the corrected layout data of the first correction portion; and training the AI subsystem with the first training data to predict the initial offset value.

[0009] In some embodiments, determining the first training data based on the first portion of original layout data and the first corrected portion of corrected layout data includes: determining layout data and corrected layout data that meet predetermined conditions as the first training data.

[0010] In some embodiments, the layout data satisfying the predetermined condition includes layout data having an average value of at least one of the following parameters within a corresponding threshold range: critical dimension; spacing; and pitch.

[0011] In some embodiments, determining the layout data and the corrected layout data that meet predetermined conditions as first training data includes: performing a mask rule check on the corrected layout data of the first correction part; and in response to the corrected layout data meeting the constraints of the mask rules, determining the corrected layout data and the original layout data corresponding to the corrected layout data as training data.

[0012] In some embodiments, the method further includes: using the initial offset value as an initial offset value for each of the remaining portions in the layout; and performing optical proximity correction on each portion based on the initial offset value.

[0013] In some embodiments, the method further includes iteratively predicting an initial offset value of a subsequent portion based on training data from a previous portion to perform optical proximity correction on the subsequent portion until optical proximity correction of each portion of the layout is completed.

[0014] In some embodiments, the method further includes: iteratively feeding back an initial offset value based on training data from a previous portion and updating the network weights of the AI subsystem to predict an initial offset value of a subsequent portion, thereby performing optical proximity correction on the subsequent portion until optical proximity correction of each portion of the layout is completed.

[0015] In some embodiments, performing optical proximity correction on the second part based on the initial offset value to generate a second correction part includes: iteratively performing optical proximity correction starting from the initial offset value to determine corresponding simulation values and target values; and in response to the difference between the simulation value and the target value satisfying a predetermined condition, determining the corrected second part as the second correction part.

[0016] In some embodiments, the method further includes: after optical proximity correction is completed for each part except the first part, training the AI subsystem based on the corresponding training data to respectively determine new network weights; and updating the network weights of the AI subsystem with the new network weights respectively.

[0017] In some embodiments, the layout is divided into a plurality of blocks, each portion including a corresponding number of blocks, and the method further includes: allocating the plurality of blocks to corresponding nodes of the distributed architecture according to a predetermined order; and performing optical proximity correction on the blocks at the corresponding nodes at each node.

[0018] In a third aspect of the present disclosure, a computer-readable storage medium is provided, on which machine-executable instructions are stored. When the machine-executable instructions are executed by a processor, the method according to the first aspect of the present disclosure is implemented.

[0019] It will be understood from the following description that the technical solution of the present disclosure can significantly improve computing efficiency, enhance applicability and reduce training costs.

[0020] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram illustrating an example environment in which embodiments of the present disclosure can be implemented; Figure 2 A flowchart illustrating a method for optical proximity correction according to some embodiments of the present disclosure is shown; Figure 3 A schematic diagram showing the principle of shifting the layout when performing optical proximity correction is shown; Figure 4A schematic diagram illustrating training and prediction of an AI subsystem for optical proximity correction according to some embodiments of the present disclosure is shown; Figure 5 A block diagram is shown of a computing device capable of implementing various embodiments of the present disclosure.

[0022] In the various drawings, the same or corresponding reference numerals denote the same or corresponding parts. DETAILED DESCRIPTION

[0023] The principles of the present disclosure will be described below with reference to the various exemplary embodiments shown in the accompanying drawings. It should be understood that the description of these embodiments is only to enable those skilled in the art to better understand and further implement the present disclosure, and is not intended to limit the scope of the present disclosure in any way. It should be noted that similar or identical reference numerals can be used in the figures where possible, and similar or identical reference numerals can represent similar or identical functions. Those skilled in the art will readily recognize, from the description below, that alternative embodiments of the structures and methods described herein can be adopted without departing from the principles of the present invention described herein.

[0024] As used herein, the term "including" and its variations represent open inclusion, i.e., "including but not limited to." Unless otherwise stated, the term "or" means "and / or." The term "based on" means "based at least in part on." The terms "one example embodiment" and "an embodiment" mean "at least one example embodiment." The term "another embodiment" means "at least one additional embodiment." The terms "first," "second," etc. may refer to different or identical objects.

[0025] As mentioned earlier, bias correction is a core method for compensating for distortion. This includes rule-based bias correction, model-based bias correction, and AI-based bias correction prediction. While these technologies have applications in their respective fields, their implementation and practical application still face the following significant drawbacks: 1. Poor adaptability: Rule-based deviation correction cannot respond to process fluctuations in real time, resulting in deviations between the correction results and the actual situation. It fails in advanced processes and requires manual intervention to correct model parameters, resulting in poor adaptability.

[0026] 2. Long calculation time and high computing resource consumption: Model-based offset correction relies on high-precision lithography simulation to iteratively optimize mask patterns pixel by pixel. As the process advances to the 3nm node and below, a single chip design contains billions of polygons. Each simulation must traverse all graphics units and calculate multi-physics coupling effects such as optics and photoresist chemical reactions. As a result, a single full-chip correction can take tens of hours or even days. Furthermore, model-based offset correction places extremely high demands on hardware computing resources, requiring parallel computing on large-scale CPU / GPU clusters.

[0027] 3. High training data acquisition costs and high data dependency limit generalization capabilities: This approach relies on AI learning the mapping relationship between graphic features and deviations from historical data and directly outputting correction values. This approach relies on a large amount of historical data (the mapping relationship between graphic features and measured deviations) for training. However, data distribution varies significantly across process nodes, lithography equipment, or design types, making the model susceptible to failure in new scenarios. Furthermore, high-quality training data relies on wafer fabrication and scanning electron microscope (SEM) inspection at the wafer fab, making acquisition expensive.

[0028] In view of this, the present disclosure provides an improved solution.

[0029] Embodiments of the present disclosure provide an improved optical proximity correction method. The method includes: performing optical proximity correction on a first portion of a layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the original layout data of a second portion based on the original layout data of the first portion and the corrected layout data of the first corrected portion; and performing optical proximity correction on the second portion based on the initial offset value to generate a second corrected portion.

[0030] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0031] Figure 1 1 shows a schematic diagram of an example environment 100 in which embodiments according to the present disclosure can be implemented. Figure 1 As shown, the example environment 100 includes a computing device 110 and a client 120 .

[0032] In some embodiments, computing device 110 may interact with client 120. For example, computing device 110 may receive input messages from client 120 and output feedback messages to client 120. In some embodiments, the input messages from client 120 may be layout data. Computing device 110 may perform corresponding mathematical operations on the layout data and output the corresponding operation results to client 120.

[0033] In some embodiments, computing device 110 may include, but is not limited to, a personal computer, a server computer, a handheld or laptop device, a mobile device (such as a mobile phone, a personal digital assistant (PDA), a media player, etc.), consumer electronics, a minicomputer, a mainframe computer, cloud computing resources, etc.

[0034] It should be understood that the structure and functionality of the example environment 100 is described for illustrative purposes only and is not intended to limit the scope of the subject matter described herein. The subject matter described herein can be implemented in different structures and / or functions. This environment is merely illustrative and is not intended to limit the application environment of the embodiments of the present disclosure.

[0035] In order to explain the principle of the present disclosure more clearly, the following will refer to Figure 2 Let's describe it in more detail.

[0036] Figure 2 A flow chart of a method 200 for optical proximity correction according to some embodiments of the present disclosure is shown.

[0037] At block 202, optical proximity correction is performed on a first portion of a layout to generate a first corrected portion. The layout may be divided into two or more portions, any of which may serve as the first portion.

[0038] In some embodiments, the layout can also be divided into a large number of tiles. In some embodiments, the layout can be divided into up to millions of tiles using EDA tools. Each part includes a corresponding number of tiles, and the number of tiles in each part can be the same or different. It should be understood that the present invention is not limited to this, and the layout may not be divided into multiple tiles. The advantage of dividing into a large number of tiles is that it facilitates distributed processing of each tile, such as through a distributed processing (DP) architecture or a multi-threading architecture. When divided into multiple tiles, the multiple tiles can be assigned to the corresponding DP nodes of the distributed architecture according to a predetermined order; and optical proximity correction is performed on the tiles at the corresponding nodes at each node. The DP node can be a client or a CPU core.

[0039] In some embodiments, blocks can be divided into rows and columns, with each block assigned a unique number, such as row 1, column 1, denoted as Tile[1, 1]. Blocks can be allocated in a specific order, such as row by row, i.e., in the order [0, 0], [0, 1], ..., [0, n], [1, 0], [1, 1], ..., [1, n]... The embodiments of the present disclosure are not limited to this, and allocation can be performed in various other ways as needed.

[0040] The first portion may include a predetermined number of blocks, for example, 1000 blocks. OPC calibration may be performed on these blocks first. A conventional OPC calibration scheme may be employed to generate a first calibration. In some embodiments, after the calibration is complete, the associated data may be used to generate training data, as further described below.

[0041] At block 204 , an initial offset value for performing optical proximity correction on the second portion of the original layout data is determined based on the first portion of the original layout data and the corrected layout data of the first corrected portion.

[0042] In some embodiments, conventional methods can be used to analyze the original layout data of the first portion and the corrected layout data of the first corrected portion to determine the deviation between the original layout data and the corrected layout data. Based on this deviation, an initial offset value for optical proximity correction of the second portion of the original layout data can be predicted. The principle behind this method is that, since the lithographic conditions of the same layout are consistent, the offset of the remaining portions, such as the initial offset value (or initial offset value), can be predicted based on data from the previous portion.

[0043] In some embodiments, the AI subsystem can be used to predict the initial offset value of the subsequent portion. Specifically, for example, the AI subsystem can be trained using training data to predict the initial offset value of the subsequent portion. It should be understood that the present disclosure is not limited to this. In some embodiments, the AI subsystem can be used to predict the initial offset value of the subsequent portion directly based on the data input to the AI subsystem without requiring training.

[0044] In some embodiments, after OPC correction is completed for some DP blocks, such as the first part of blocks (eg, 1000 blocks), some data are selected based on the corresponding input layout information and the final offset result to form a training data set.

[0045] The aforementioned selection of data can include input layout data (serving as input) and the corresponding final offset data (serving as output). Training an AI subsystem often relies on a set of input and output datasets. Through training, the AI subsystem can directly infer offset data based on the input layout information. The purpose of selecting data sets is to select relatively typical datasets. For example, datasets with (almost) empty input layouts should be screened out to improve training efficiency and quality. A dataset with (almost) empty input layouts can be understood as having no (or almost no) graphics in the layout. Furthermore, as mentioned earlier, some AI systems may be able to predict the initial offset values for subsequent parts directly based on the input data without training.

[0046] In some embodiments, first training data may be determined based on the original layout data of the first portion and the corrected layout data of the first corrected portion; and the AI subsystem may be trained using the first training data to predict the initial offset value.

[0047] In some embodiments, the layout data and the corrected layout data that meet predetermined conditions are determined as the first training data. The predetermined conditions can be set by the user according to actual needs or according to other relevant standards.

[0048] In some embodiments, layout data meeting predetermined conditions includes layout data having an average value of at least one of the following parameters within a corresponding threshold range: critical dimension (CD); space; and pitch. In other words, whether the layout data is typical can be determined based on whether the average critical dimension, space, or pitch of the pattern falls within the corresponding threshold range.

[0049] The following is a further explanation of these three parameters. In the EDA (Electronic Design Automation) industry, especially in the physical design, manufacturing process, and lithography stages of chips, CD, spacing, and pitch are the most basic and critical concepts that describe the size and spacing of geometric patterns on integrated circuit layouts. Their definitions are as follows: 1) Critical dimension (CD); Meaning: Refers to the most critical design dimensions such as the width of the lines on the layout (for the wiring layer) or the side length of the through hole / contact hole (for the through hole layer).

[0050] Importance: CD is often the primary basis for naming a specific manufacturing process node (e.g., the numbers in 28nm, 14nm, and 7nm refer to the typical minimum channel length or gate CD for that node). It directly reflects the minimum feature size achievable by the process and determines transistor performance and circuit density.

[0051] For example, on a photolithography mask, the width of a metal trace is its CD. The width of the metal trace formed after exposure on the wafer is called the printed CD or final CD. The difference between the design target CD and the actual printed CD (caused by optical and etching effects, etc.) is a key metric for process development and photolithography model refinement.

[0052] 2) Spacing, also known as interval; Meaning: Refers to the minimum distance between two adjacent geometric shapes of the same type on the layout.

[0053] Importance: Spacing specifies the minimum safe distance between different structures (such as between conductors on a layer or between contact holes). Adhering to this minimum spacing ensures that: during manufacturing, patterns do not adhere (bridge) due to close proximity; during circuit operation, electrical isolation requirements are met to prevent short circuits and signal crosstalk; and structures (such as metal lines) have sufficient mechanical strength.

[0054] For example, the minimum distance between the edges of two parallel metal lines is the minimum spacing of the metal layer. The minimum distance between the active areas of two adjacent transistors is the minimum spacing of the active layer.

[0055] 3) Pitch, also known as spacing (note the difference from Space).

[0056] Meaning: Refers to the distance between the center of a set of regularly recurring geometric units of the same type on a layout and the center of adjacent units. More simply, it is the length of a complete structural unit (such as a line plus the space between it and adjacent lines) in the direction of repetition.

[0057] Calculation: Pitch = CD + Space (for parallel line structures of equal width and equal spacing).

[0058] For example, if the CD of a line is 50nm and the minimum space between it and the adjacent line is 50nm, then their pitch is 50nm+50nm=100nm.

[0059] Importance: Pitch directly reflects the integration density of a circuit. A smaller pitch means more components (wires, transistors, etc.) can be placed per unit area, resulting in more powerful chips and lower costs. However, reducing pitch is one of the greatest challenges in process manufacturing, as it is more difficult than simply reducing CDs (due to a combination of constraints such as lithography resolution limits, etching accuracy, and overlay precision).

[0060] For example, the array structure of memories (such as DRAM and SRAM) typically has a very small pitch (called half pitch), which is an important indicator of storage density. The gate pitch of transistors in standard logic cell libraries is also a key process parameter.

[0061] Summary and Relationships: Width (CD): The minimum dimension of a single feature.

[0062] Space: The minimum distance between two adjacent features.

[0063] Pitch: The sum of CD and adjacent spacing, representing the length of the minimum repeating unit.

[0064] Simple formula: Pitch = CD + Space (applicable to adjacent structures in the same direction with uniform line width and spacing).

[0065] The core significance of these three parameters lies in their collective definition of the geometric rules and physical limits of integrated circuit layouts. During the design phase (routing, DRC, OPC, etc.), EDA tools must strictly adhere to the process design rules provided by the manufacturer. These rules extensively specify minimum CD, minimum spacing, and minimum pitch for different layers and scenarios. Optimizing and continuously reducing CD, spacing, and pitch is the core goal driving the continuous advancement of semiconductor processes.

[0066] In addition, it should be pointed out that the layout data that meets the predetermined conditions is not limited to the above-mentioned data related to CD, spacing, and pitch, but other data can be selected as typical data according to actual needs.

[0067] The above embodiment describes a method for determining the first training data. The embodiments of the present disclosure are not limited thereto. Other methods can be used to determine the first training data as needed, as long as the selected data helps improve training efficiency and quality.

[0068] In some embodiments, a mask rule check (MRC) may be performed on the corrected layout data of the first correction portion. If the corrected layout data meets the mask rule constraints, the corrected layout data and the original layout data corresponding to the corrected layout data are determined as training data. Specifically, the final shift result may be used to determine whether it violates basic constraints (such as mask rule check constraints), and whether to use the selected data as training data is determined. For example, if an MRC check is performed on the final shift result and it is found that the final shift result does not meet the MRC constraints, the corresponding layout data is not used as training data.

[0069] After the design has undergone OPC processing (post-OPC), it is typically subjected to a mask rule check (MRC) before being sent to the mask fabricator. MRC examines the post-OPC data to confirm that all patterns are suitable for the mask fabrication process. The rules in the MRC are provided by the mask fabricator, and the OPC engineer enters these rules into the MRC software.

[0070] The MRC rules mainly include the following: (1) Regulations are set for the minimum line width (minWidth) and minimum line spacing (minSpace) of graphics; minimum values for the line width and spacing of sub-resolution assist features (SRAFs) can also be set.

[0071] (2) Specify the minimum value of the corner-to-corner spacing between graphics; you can also limit the distance between the sub-resolution auxiliary graphics (SRAF) and the main graphics.

[0072] (3) Limit the minimum area of auxiliary graphics (minimun SRAF area).

[0073] The determined training data (e.g., the first training data) can be used to train the AI subsystem to predict the initial offset values for the subsequent portion (e.g., the second portion) of the layout. As is known in the industry, the initial offset values are used as the initial values for iterations during the OPC process. In other words, the initial offset values are the values of the original layout after a preliminary bias. The lithography simulation system performs iterative simulation based on the initial values and determines convergence based on the difference between the simulation results and the target values. The layout corresponding to convergence can be determined as the OPC-corrected layout. This is further described below.

[0074] As mentioned above, in some embodiments, the AI subsystem may be trained using the first training data to predict initial offset values for performing optical proximity correction on the second portion of raw layout data. The second portion may include a second number of blocks. The second number may be the same as or different from the first number.

[0075] In computational lithography, bias is an action that can be defined as follows: Definition: During the photolithography mask design stage, the geometric deviation of the pattern is pre-compensated (such as edge offset and auxiliary pattern addition) to offset the distortion of the silicon wafer pattern caused by light diffraction and interference.

[0076] Technical goal: Make the final silicon wafer pattern (such as chip line width) consistent with the design target and avoid defects such as "bridging" (adjacent lines sticking together) or "line end shortening".

[0077] Refer to the following Figure 3 , Figure 3 The schematic diagram shows the principle of shifting the layout when performing optical proximity correction. Specifically, Figure 3 A schematic diagram of the process of correcting the deviation of the layout is shown in FIG. Here, according to the results of the lithography simulation, the edges of the original layout are moved to generate a new layout, and then the lithography simulation is performed again, and so on.

[0078] like Figure 3 As shown, Figure 3 (A) shows the original layout, wherein the point 310 on the edge 302 of the layout pattern is a split point, which is used to indicate the split position where the edge (Edge) constituting the layout is divided into several small segments.

[0079] Figure 3 (B) shows a first offset edge 304 obtained by performing a first round of offset on the edge of the original layout; Figure 3 (C) shows the k-th offset edge 306 obtained by performing the k-th round of offset on the edge of the original layout; Figure 3 (D) shows the n-th offset edge 308 obtained by performing the n-th round of offset on the edge of the original layout.

[0080] That is, through n rounds of offset, the original layout is finally corrected to the appearance of the "nth round of bias". Such a layout will produce the most ideal effect when it is photolithographically printed on the wafer (according to the results of lithography simulation).

[0081] In some embodiments, the AI subsystem can be trained in real time using a training data set and can predict the initial offset value of subsequent blocks, such as Figure 3 The value corresponding to the first offset edge 304 shown in (B) of FIG. In other words, the initial offset value is the value of the original layout after the initial offset. This initial offset value is then used for the OPC correction of the subsequent DP blocks. After the OPC correction is completed for some subsequent DP blocks, training data is selected from them and fed back to the AI subsystem (i.e., the AI subsystem is trained), and the network weights are updated until all DP blocks have completed OPC correction. Because the lithography conditions of the same layout are consistent, the initial offset values of the remaining layouts can be predicted based on the training data from the previous part of the layout. The training of the AI subsystem can adopt the traditional training method.

[0082] As mentioned above, in some embodiments, the initially predicted initial offset value can be used for subsequent OPC correction of all blocks in each portion. This approach can eliminate the need for subsequent AI subsystem training, thereby speeding up the entire correction process.

[0083] In other embodiments, the initially predicted initial offset value can be used to correct the OPC of the blocks in the second portion. Subsequently, the AI subsystem can feed back the initial offset value based on the training data from the second portion and update the AI subsystem's network weights to predict the initial offset value for the third portion. And so on. In this case, the initial offset value for each portion is different. In this approach, by continuously selecting training data from each portion and feeding it back to the AI subsystem (i.e., training the AI subsystem), and updating the network weights, network performance can be optimized, enabling it to better fit the training data and improve prediction accuracy.

[0084] The initial offset value predicted by the AI subsystem can accelerate simulation convergence, thereby significantly reducing the number of model-based bias iterative simulations and speeding up the OPC correction process.

[0085] At block 206 , an optical proximity correction is performed on the second portion based on the initial offset value to generate a second corrected portion.

[0086] The optical proximity correction (OPC) method employed here can be conventional. In this process, the initial offset value is used as the initial value for correction in the lithography simulation model to perform optical proximity correction to determine the corresponding simulated value and the difference between the simulated value and the target value. If the difference between the simulated value and the target value satisfies a predetermined condition, the corresponding second portion of the correction is determined as the second correction portion, confirming that the OPC correction has met expectations.

[0087] After the OPC correction of the second part of the block is completed, the training data set is selected again, the AI subsystem is trained in real time, and the initial offset value of the subsequent block (the third part of the block) is predicted for the OPC correction of the third part of the block.

[0088] In some embodiments, performing optical proximity correction on the second part based on the initial offset value to generate a second correction part may include: iteratively performing optical proximity correction starting from the initial offset value to determine corresponding simulation values and target values; and in response to the difference between the simulation value and the target value satisfying a predetermined condition, determining the corrected second part as the second correction part.

[0089] In some embodiments, second training data can be determined based on the original layout data of the second portion and the corrected layout data of the second corrected portion. The AI subsystem is trained using the second training data to predict initial offset values for performing optical proximity correction on the original layout data of the third portion. Similarly, initial offset values for subsequent portions can be iteratively predicted based on the training data from the previous portion to perform optical proximity correction on the subsequent portion, until optical proximity correction is completed for each portion of the layout.

[0090] In some embodiments, the initial offset value may be used as the initial offset value for the remaining portions in the layout; and optical proximity correction may be performed on each portion based on the initial offset value.

[0091] Therefore, in some embodiments, the initial offset value can be iteratively fed back based on the training data from the previous part and the network weights of the AI subsystem can be updated to predict the initial offset value of the latter part, thereby performing optical proximity correction on the latter part until the optical proximity correction of each part of the layout is completed.

[0092] In some embodiments, after optical proximity correction is completed for each part except the first part, new network weights are determined based on the training of the AI subsystem based on the corresponding training data; and the network weights of the AI subsystem are updated with the new network weights.

[0093] The following combination Figure 4 Provide a description. Figure 4 A schematic diagram illustrating training and prediction of an AI subsystem for optical proximity correction according to some embodiments of the present disclosure is shown.

[0094] like Figure 4 As shown, the AI subsystem and the first part block, the second part block, and the Nth part block are shown. Figure 4 As shown in , the block is divided into N parts, where N can be an integer greater than or equal to 2. The output data of each part is the offset value (essentially, what each pattern is expected to be corrected to in the end), which is used to guide the initial correction of subsequent blocks.

[0095] In some embodiments of the present disclosure, the AI subsystem may employ a convolutional neural network (CNN) or graph neural network (GNN) architecture to compress complex physical models into lightweight networks through knowledge distillation, reducing inference time. The model can be packaged as a Python or C++ library (or other library formats as needed) and integrated into EDA tools. It should be understood that the embodiments of the present disclosure are not limited to this.

[0096] Knowledge distillation is a machine learning model compression method that aims to transfer knowledge from large models to smaller models to improve model performance and generalization. The core idea of knowledge distillation is to transform the knowledge of complex models into a more concise and efficient representation, thereby maintaining high performance while reducing computational complexity and resource requirements. As an advanced machine learning technique, its core concept is to refine and "infuse" the knowledge and experience of a complex, large model (often called the teacher model) into a smaller, simpler model (the student model). This process not only preserves the predictive power and accuracy of the teacher model, but also significantly improves the operational efficiency and computational performance of the student model, enabling it to perform well even in resource-constrained environments.

[0097] In some embodiments of the present disclosure, OPC correction may be performed on the first partial block to generate a corrected first partial block (which may be referred to as a first correction block). By performing OPC correction on the first partial block, an offset result may be obtained. The offset result is fed back to the AI subsystem, such as Figure 4 As shown by the arrow after the first block.

[0098] As mentioned above, data can be selected from the first part of the block to perform initial training on the AI subsystem. After the training is completed, the initial offset value of the second part can be inferred. Specifically, data can be selected as training data based on the layout information (input) of the first part of the block and the corresponding offset result. In other words, training data can be selected based on the original layout data and the corrected layout data. The AI subsystem predicts the initial offset value of the second part of the block by using the training data for training, that is, Figure 4 The first offset initial value shown in .

[0099] Perform OPC correction on the second partial block to generate an offset result of the partial block, which can be fed back to the AI subsystem, such as Figure 4 As shown in . By feeding the results of the second part of the block back to the AI subsystem, the network weights can be updated to make the AI subsystem more accurate (compared to the initial training). The third part of the block uses the more accurate AI subsystem to infer the offset initial value, and the results are fed back to the AI subsystem again. And so on, for example Figure 4 As shown in , the OPC correction of the Nth partial block can be based on the N-1th offset initial value, where N is an integer greater than 2.

[0100] In some embodiments, the Nth portion is not predicted solely based on the N-1th portion of training data, but rather based on the 1st, 2nd, ... N-1th portions of training data. Therefore, in some embodiments, initial offset values can be iteratively fed back based on the training data from the previous portion and the network weights of the AI subsystem can be updated to predict initial offset values for subsequent portions, thereby performing optical proximity correction on the subsequent portions, until optical proximity correction is completed for each portion of the layout.

[0101] Updating network weights in a neural network optimizes network performance, enabling it to better fit the training data and improve prediction accuracy. By continuously adjusting the network weights, the neural network adapts to the training data, thereby improving its predictive power on test data. Furthermore, in non-stationary environments, such as real-time streaming data, continuously updating weights enables the model to track data changes.

[0102] In general, the AI subsystem completes initial training through the first part of the block. After each subsequent part is completed, it will continuously provide feedback and update the AI subsystem, making the AI subsystem's reasoning more and more accurate.

[0103] Furthermore, in some embodiments of the present disclosure, the sizes of the blocks may be the same or different. Furthermore, the number of blocks in each portion may be the same or different.

[0104] In some embodiments of the present disclosure, the AI subsystem can be trained, predicted, and dynamically fed back in real time. In some embodiments of the present disclosure, the AI subsystem is trained in real time. AI training refers to adjusting model parameters through data sets so that it learns the inherent laws of the data, and AI reasoning refers to applying the trained model to new data to generate prediction results. The AI subsystem is trained by selecting input layouts and corresponding final offset data to enable it to acquire the ability to predict offset values based on the input layout. Then, subsequent blocks can send the input layout data to the AI subsystem, and the AI subsystem can infer the offset data, that is, predict the initial offset value of the subsequent block.

[0105] In known traditional solutions, each block has no initial offset value, or the initial offset value can be understood as 0. An initial offset value of 0 means that the iterative simulation process begins based on the original layout. In some embodiments of the present disclosure, based on the AI subsystem's prediction of the initial offset value, if the initial offset value is close to or equal to the final offset value (the final offset result, or the final corrected layout), the model-based iterative offset simulation can converge quickly, significantly reducing the number of model-based iterative offset simulations.

[0106] In some embodiments of the present disclosure, after OPC correction is completed for some blocks, a training data set is selected from the data to train the AI subsystem in real time, predict the initial offset values for subsequent blocks, and provide feedback based on the offset results of subsequent blocks. This process design mechanism is one of the key cores of the embodiments of the present disclosure. This process design mechanism improves the efficiency of model-based offset calculations, reduces training costs, and has broad applicability.

[0107] If the AI subsystem doesn't predict the initial offset value, the initial offset is assumed to be 0. At this point, multiple rounds of model-based bias iterative simulation are required to obtain the final offset value: the first round of simulation yields the offset value "Based-1." If convergence is not achieved, the second round of simulation yields the offset value "Based-2," and so on until the kth round of simulation yields the offset value "Based-k," ultimately achieving convergence. However, if the AI subsystem predicts the initial offset value, the offset value "Based-i" is directly obtained. This offset value "Based-i" is very close to the offset value "Based-k," requiring only one or two more iterative simulations to obtain the offset value "Based-k." This significantly reduces the number of model-based bias iterative simulations. Bias-k can be understood as the "result of the kth round of bias."

[0108] That is to say, if there is no AI subsystem, it is necessary to start from the original layout and go through a complete n rounds of computational simulation to obtain the final converged result; however, in the present invention, under the prediction of the AI subsystem, a result close to the "nth round offset value (bias)" can be obtained before the first lithography simulation, so convergence can be achieved after only a few lithography simulations.

[0109] In some embodiments of the present disclosure, after optical proximity correction is completed for the remaining parts except the first part, new network weights are determined based on the training of the AI subsystem based on the corresponding training data; and the new network weights are used to update the previous network weights of the AI subsystem respectively.

[0110] In some embodiments of the present disclosure, Figure 4 As shown in Figure 1, the AI subsystem and the lithography simulation model work together in the OPC correction process.

[0111] Some embodiments of the present disclosure can significantly accelerate the OPC process of the layout by combining an AI subsystem with a lithography simulation system.

[0112] The above embodiments describe in detail the accelerated design method for offset correction, which significantly improves the speed of model-based offset correction and reduces the consumption of computing resources for offset correction. Moreover, the solution of the disclosed embodiment uses real-time training, does not rely on historical data, has a wide range of applicability, and is effective for different process nodes, lithography equipment or design types.

[0113] An embodiment of the present disclosure provides a simulation model, including: a lithography model; and an artificial intelligence subsystem coupled with the lithography model; wherein the lithography model and the artificial intelligence subsystem are configured to execute the method in the above embodiment.

[0114] The disclosed embodiments propose a model-based bias correction method that significantly accelerates the bias correction process by training an AI subsystem in real time and predicting the initial offset value of a block. Specifically, this method integrates an artificial intelligence (AI) subsystem into EDA software and trains the AI subsystem in real time based on the bias results of blocks that have already been corrected during the OPC process. This allows the AI subsystem to predict the initial offset value of uncorrected blocks, reducing the number of iterations of the model-based bias correction, accelerating simulation convergence, and saving computing time and resources.

[0115] Some embodiments of the present disclosure provide methods for optical proximity correction. It should be noted that the examples given in the above embodiments are only for illustrating the solutions of the embodiments of the present disclosure and are not intended to limit the solutions of the present disclosure. The embodiments of the present disclosure can be corrected and deformed as needed. For example, the AI subsystem can be trained and predicted in real time without a feedback mechanism. The absence of a feedback mechanism means that the network weights of the AI subsystem are not updated.

[0116] Some embodiments of the present disclosure have the following advantages: Improved computing efficiency: The number of blocks in the chip layout in OPC usually reaches millions. Through real-time training of the AI subsystem and prediction of the initial offset value, the number of iterative simulations based on model deviations for most blocks can be significantly reduced, reducing calculation time and reducing computing resource consumption. Reasonable initial offset values can accelerate convergence. Even if the initial prediction is not accurate enough, it can be gradually corrected through iterative optimization to avoid redundant calculations caused by completely random initialization, so it will not affect runtime efficiency. "Completely random initialization" means that ordinary OPC uses a lithography simulation model for correction, so before the first round of lithography simulation, the initial offset value of the graphic is the completely random initialization value.

[0117] Enhanced applicability: The embodiments of the present disclosure greatly improve applicability through real-time training, and are effective for different process nodes, lithography equipment, or design types.

[0118] Reduced training costs: This disclosure avoids reliance on historical data through real-time training, thereby significantly reducing training costs.

[0119] It can be seen that the solution of the embodiment of the present disclosure has brought significant beneficial effects in terms of computing efficiency, applicability, training cost, etc.

[0120] The present disclosure also discloses an electronic device, including a processor and a memory coupled to the processor, the memory having instructions stored therein. When executed by the processor, the instructions cause the device to perform actions, including: performing optical proximity correction on a first portion of a layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the original layout data of a second portion based on the original layout data of the first portion and the corrected layout data of the first corrected portion; and performing optical proximity correction on the second portion based on the initial offset value to generate a second corrected portion.

[0121] An embodiment of the present disclosure further discloses a computer-readable storage medium having machine-executable instructions stored thereon. When the machine-executable instructions are executed by a processor, the method according to the embodiment of the present disclosure is implemented.

[0122] It should be understood that the embodiments shown in the drawings are only for schematically illustrating some embodiments of the present disclosure and are not intended to limit the present disclosure. The embodiments of the present disclosure may also have various other forms.

[0123] Figure 5 Schematic block diagrams of electronic devices according to some exemplary embodiments of the present disclosure are shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0124] like Figure 5 As shown, device 500 includes a CPU 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. CPU 501, ROM 502, and RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to bus 504.

[0125] Multiple components in device 500 are connected to I / O interface 505, including an input unit 506, such as a keyboard, mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, optical disk, etc.; and a communication unit 509, such as a network card, modem, wireless communication transceiver, etc. The communication unit 509 allows device 500 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0126] The various processes and processing described above, such as method 200, can be executed by CPU 501. For example, in some embodiments, method 200 can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by CPU 501, one or more steps in method 200 described above can be performed.

[0127] The solutions according to the embodiments of the present disclosure may be methods, devices, systems, and / or computer program products. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing various aspects of the present disclosure are loaded. The computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. The computer-readable program instructions may be downloaded from the computer-readable storage medium to various computing / processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network.

[0128] Various embodiments of the present disclosure have been described above. The above descriptions are exemplary and are only optional embodiments of the present disclosure. They are not exhaustive and are not intended to limit the present disclosure. Although the claims in this application have been formulated for specific combinations of features, it should be understood that the scope of the present disclosure also includes any novel feature or any novel combination of features disclosed herein, whether explicitly or implicitly or in any generalization thereof, regardless of whether it relates to the same scheme in any claim currently claimed. It should be understood that new claims may be formulated to these features and / or combinations of these features during the examination of this application or in any further application derived therefrom.

[0129] The terminology used herein is selected to best explain the principles of the various embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the various embodiments disclosed herein. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this disclosure are intended to be included within the scope of protection of this disclosure.

Claims

1. An optical proximity correction method, comprising: performing optical proximity correction on a first portion in the layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the second portion of the original layout data based on the first portion of the original layout data and the first corrected portion of the corrected layout data; as well as An optical proximity correction is performed on the second portion based on the initial offset value to generate a second corrected portion.

2. The method according to claim 1 , wherein determining an initial offset value for performing optical proximity correction on the second portion of the original layout data based on the first portion of the original layout data and the first corrected portion of the corrected layout data comprises: Determine first training data based on the original layout data of the first portion and the corrected layout data of the first corrected portion; as well as The AI subsystem is trained using the first training data to predict the initial offset value.

3. The method according to claim 2, wherein determining the first training data based on the original layout data of the first portion and the corrected layout data of the first corrected portion comprises: The layout data that meets a predetermined condition and the corrected layout data are determined as the first training data.

4. The method according to claim 3, wherein the layout data satisfying the predetermined condition comprises layout data having an average value of at least one of the following parameters within a corresponding threshold range: Critical dimensions; Spacing; and pitch.

5. The method according to claim 3, wherein determining the layout data satisfying a predetermined condition and the corrected layout data as the first training data comprises: performing a mask rule check on the corrected layout data of the first corrected part; as well as In response to the corrected layout data meeting the constraint condition of the mask rule, the corrected layout data and original layout data corresponding to the corrected layout data are determined as the first training data.

6. The method according to claim 1, further comprising: Using the initial offset value as the initial offset value of the remaining parts in the layout; as well as Optical proximity correction is performed on each of the parts based on the initial offset value.

7. The method according to claim 2, further comprising: An initial offset value of a subsequent portion is iteratively predicted based on training data from a previous portion to perform optical proximity correction on the subsequent portion until optical proximity correction of each portion of the layout is completed.

8. The method according to claim 2, further comprising: Iteratively feed back an initial offset value based on training data from a previous portion and update the network weights of the AI subsystem to predict an initial offset value for a subsequent portion, thereby performing optical proximity correction on the subsequent portion until optical proximity correction of each portion of the layout is completed.

9. The method of claim 1 , wherein performing optical proximity correction on the second portion based on the initial offset value to generate a second corrected portion comprises: Iteratively performing optical proximity correction starting from the initial offset value to determine corresponding simulated values and target values; as well as In response to a difference between the simulation value and the target value satisfying a predetermined condition, the second portion of the correction is determined as the second correction portion.

10. The method according to claim 2, further comprising: After completing optical proximity correction for each part except the first part, training the AI subsystem based on the corresponding training data to determine new network weights for each part; as well as The network weights of the AI subsystems are updated respectively with the new network weights.

11. The method according to any one of claims 1 to 10, wherein the layout is divided into a plurality of blocks, each portion including a corresponding number of the blocks, the method further comprising: Allocating the plurality of blocks to corresponding nodes of the distributed architecture according to a predetermined order; as well as The optical proximity correction is performed on the block at each node respectively.

12. A simulation model comprising: photolithography models; as well as an artificial intelligence subsystem coupled to the lithography model; The lithography model and the artificial intelligence subsystem are configured to perform the method of any one of claims 1 to 11.

13. An electronic device comprising: processor; as well as A memory coupled to the processor, the memory having instructions stored therein, which, when executed by the processor, cause the device to perform actions, the actions comprising: performing optical proximity correction on a first portion in the layout to generate a first corrected portion; determining an initial offset value for performing optical proximity correction on the second portion of the original layout data based on the first portion of the original layout data and the first corrected portion of the corrected layout data; and An optical proximity correction is performed on the second portion based on the initial offset value to generate a second corrected portion. 14 . A computer-readable storage medium having machine-executable instructions stored thereon, wherein when the machine-executable instructions are executed by a processor, the processor is caused to implement the method according to claim 1 .

Images