Opportunity for a patent cooperation treaty model training method and system based on unsupervised learning and terminal

Abstract

This invention provides an OPC model training method, system, and terminal based on unsupervised learning. It constructs a simplified OPC training dataset by performing multi-dimensional lithographic feature extraction, feature preprocessing, clustering, classification, and hierarchical filtering on multiple lithographic mask test patterns. An OPC model is then trained based on this dataset to perform optical proximity correction on the lithographic mask patterns. This invention combines multi-dimensional lithographic characteristics to accurately cover different pattern scenarios, avoiding the omission of key samples and ensuring the accuracy of OPC model correction. It employs an unsupervised learning algorithm to lightweight sample data and classifies differentiated pattern categories through clustering, effectively improving the process coverage of the OPC model. It fully adapts to the changing lithography process requirements under advanced process nodes and requires no customized annotation rules, allowing for rapid migration to various lithography processes. It combines high precision, full coverage, and strong versatility.

Description

Technical Field

[0001] This invention relates to the field of semiconductor lithography, and in particular to an OPC model training method, system, and terminal based on unsupervised learning. Background Technology

[0002] Optical proximity correction (OPC) is a key technology in semiconductor lithography. It mainly compensates for pattern distortion caused by optical diffraction, imaging nonlinearity and process effects in exposure, development and subsequent processes by pre-introducing edge offsets and auxiliary patterns on the mask pattern. This makes the actual pattern on the wafer as close as possible to the design target, thereby improving key dimension control, process accuracy, pattern fidelity, process window and product yield.

[0003] Currently, most OPC model training methods in the industry involve training with full datasets. To ensure model process coverage, massive amounts of lithography test pattern data with different structures, sizes, and arrangements need to be collected, resulting in significant data redundancy. This massive amount of data not only significantly increases the computational cost of model training and lengthens the iteration cycle, but also easily leads to overfitting. Furthermore, traditional data selection relies on manual experience, using only a single size metric to select samples. This fails to consider multi-dimensional lithography characteristics such as pattern density, light intensity response, and pattern spacing, easily overlooking crucial samples from specific process scenarios. Consequently, the final OPC model suffers from insufficient local correction accuracy and incomplete process coverage, failing to meet the high-precision, full-scenario lithography correction requirements of advanced processes at 7nm and below.

[0004] Existing machine learning solutions for optimizing OPC mostly employ supervised learning models, relying on a large number of manually labeled samples. This results in high labeling costs, poor versatility, and an inability to adapt to the ever-changing lithography process scenarios. Therefore, there is an urgent need for a data optimization solution that eliminates the need for manual labeling and automatically classifies multi-dimensional lithography features based on images, thereby simplifying OPC training data while ensuring model accuracy and comprehensive coverage. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide an OPC model training method, system and terminal based on unsupervised learning, which solves the problems of redundant training data, high computing cost, strong subjectivity of manual selection, insufficient model accuracy and low process coverage in traditional OPC model training.

[0006] To achieve the above and other related objectives, a first aspect of the present invention provides an OPC model training method based on unsupervised learning, comprising: constructing an original OPC modeling dataset; wherein the original OPC modeling dataset includes: multiple lithographic mask patterns of different sizes, arrangements, and structures; performing multi-dimensional lithographic feature extraction on each lithographic mask pattern in the original OPC modeling dataset to obtain a multi-dimensional graphic feature vector for each lithographic mask pattern; performing feature preprocessing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain a standardized feature vector, and constructing a standardized feature dataset; performing automatic clustering on the standardized feature dataset to obtain a sample set of multiple differentiated graphic categories; performing hierarchical filtering on the sample set of each differentiated graphic category to construct a simplified OPC training dataset; and training an OPC model based on the simplified OPC training dataset to perform optical proximity correction on the lithographic mask patterns.

[0007] In some embodiments of the first aspect of the present invention, the step of extracting multi-dimensional lithographic features from each lithographic mask pattern in the original OPC modeling dataset to obtain a multi-dimensional graphic feature vector for each lithographic mask pattern includes: extracting multi-dimensional lithographic features such as pattern density, lithographic imaging light intensity, key pattern dimensions, and pattern spacing from each lithographic mask pattern in the original OPC modeling dataset, and constructing a multi-dimensional graphic feature vector for each lithographic mask pattern.

[0008] In some embodiments of the first aspect of the present invention, the step of performing feature preprocessing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain a standardized feature vector and constructing a standardized feature dataset includes: normalizing, denoising and outlier processing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain a standardized feature vector that eliminates acquisition errors, and constructing a standardized feature dataset.

[0009] In some embodiments of the first aspect of the present invention, the step of automatically clustering the standardized feature dataset to obtain a sample set of multiple differentiated graphic categories includes: using an unsupervised learning algorithm to cluster and classify the standardized feature vectors in the standardized feature dataset according to clustering parameters to obtain a sample set of multiple differentiated graphic categories.

[0010] In some embodiments of the first aspect of the present invention, the clustering parameters include at least: feature similarity and photolithographic response characteristics.

[0011] In some embodiments of the first aspect of the present invention, the step of performing hierarchical screening of the sample set for each differentiated graphic category to construct a simplified OPC training dataset includes: screening the samples in the sample set for each differentiated graphic category according to screening criteria to obtain core samples of multiple differentiated graphic categories; wherein, the screening criteria include at least: feature dispersion and process sensitivity; and integrating the core samples of all differentiated graphic categories to obtain a simplified OPC training dataset.

[0012] In some embodiments of the first aspect of the present invention, the step of training an OPC model based on the simplified OPC training dataset to perform optical proximity correction on a lithographic mask pattern includes: training an initial OPC model using the simplified OPC training dataset; verifying whether the initially obtained OPC model meets the standards, and if the OPC model does not meet the standards, adjusting the clustering parameters to obtain a new simplified OPC training dataset and training a new OPC model, until the obtained OPC model meets the standards, and obtaining the final OPC model.

[0013] In some embodiments of the first aspect of the present invention, the method for verifying whether the OPC model meets the requirements includes: verifying the OPC model from three dimensions: graphic size correction accuracy, special process scenario adaptability, and full process coverage; if the OPC model meets the required graphic size correction accuracy, special process scenario adaptability, and full process coverage, it is considered to meet the requirements; if the OPC model does not meet the required graphic size correction accuracy, special process scenario adaptability, and full process coverage, it is considered to fail to meet the requirements.

[0014] To achieve the above and other related objectives, a second aspect of the present invention provides an OPC model training system based on unsupervised learning, comprising: applying the OPC model training method based on unsupervised learning as described in any of the above claims, wherein the system comprises: a data acquisition module for acquiring multiple lithographic mask test patterns to construct an original OPC modeling dataset; a data processing module connected to the data acquisition module for performing multi-dimensional lithographic feature extraction on each lithographic mask pattern in the original OPC modeling dataset to obtain a multi-dimensional graphic feature vector for each lithographic mask pattern, and performing feature preprocessing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain a standardized feature vector, thereby constructing a standardized feature dataset; and classification and filtering. The first module, connected to the data processing module, employs an unsupervised learning algorithm to cluster and classify the standardized feature vectors in the standardized feature dataset according to clustering parameters, obtaining sample sets of multiple differentiated graphic categories. It then performs stratified filtering on the samples in each differentiated graphic category's sample set to construct a simplified OPC training dataset. The second module, connected to the classification and filtering module, trains an OPC model based on the simplified OPC training dataset and verifies whether the OPC model meets the standards. If the OPC model does not meet the standards, it returns to the classification and filtering module, adjusts the clustering parameters to obtain a new simplified OPC training dataset, and trains a new OPC model until the OPC model meets the standards.

[0015] To achieve the above and other related objectives, a third aspect of the present invention provides an electronic terminal, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the method described above.

[0016] As described above, the present invention has the following beneficial effects:

[0017] This invention provides an OPC model training method, system, and terminal based on unsupervised learning. It constructs a simplified OPC training dataset by performing multi-dimensional lithographic feature extraction, feature preprocessing, clustering, classification, and hierarchical filtering on multiple lithographic mask test patterns. An OPC model is then trained based on this dataset to perform optical proximity correction on the lithographic mask patterns. This invention combines multi-dimensional lithographic characteristics to accurately cover different pattern scenarios, avoiding the omission of key samples and ensuring the accuracy of OPC model correction. It employs an unsupervised learning algorithm to lightweight sample data and classifies differentiated pattern categories through clustering, effectively improving the process coverage of the OPC model. It fully adapts to the changing lithography process requirements under advanced process nodes and requires no customized annotation rules, allowing for rapid migration to various lithography processes. It combines high precision, full coverage, and strong versatility. Attached Figure Description

[0018] Figure 1The diagram shows a flowchart of an OPC model training method based on unsupervised learning according to an embodiment of the present invention.

[0019] Figure 2 The diagram shows a flowchart of a method for verifying the compliance of an OPC model in one embodiment of the present invention.

[0020] Figure 3 The diagram shown is a structural schematic of an OPC model training system based on unsupervised learning according to an embodiment of the present invention.

[0021] Figure 4 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of the present invention. Detailed Implementation

[0022] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0023] In embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.

[0024] It should be noted that in the embodiments of the present invention, the words "exemplary" or "for example" indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0025] In this embodiment of the invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0026] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0027] <1> Semiconductor photolithography is the most core and critical pattern transfer process in semiconductor manufacturing. Its basic principle is to use photochemical reactions to precisely copy the integrated circuit pattern pre-fabricated on a mask onto the surface of a semiconductor wafer coated with photoresist through an optical projection system. After development, the pattern is formed into a fine pattern, and finally, the pattern is permanently transferred onto the silicon wafer through subsequent processes such as etching or ion implantation.

[0028] <2> Optical Proximity Correction (OPC) is a resolution enhancement technique in semiconductor lithography. It compensates for imaging distortions caused by light diffraction, interference, and process nonlinearity by pre-correcting the mask pattern, so that the actual pattern of the silicon wafer matches the original design as closely as possible.

[0029] <3> A mask pattern is a pattern used in micro- and nano-fabrication processes such as photolithography to define the pattern to be transferred onto a wafer or other substrate. It is the core carrier of the photolithography process in semiconductor integrated circuit manufacturing, equivalent to the "master" or "film" of chip design. Its function is to transfer the design pattern onto the photoresist on the silicon wafer through exposure.

[0030] <4> Unsupervised learning algorithms: Without human labeling, the algorithm automatically mines similar features from the data and performs clustering and feature partitioning based on the data's own distribution, distance, and similarity. Its core characteristics are: no labels, no standard answers, autonomously learning the inherent patterns of the data; grouping and classifying based on feature similarity; and it is often used for data clustering, feature dimensionality reduction, and anomaly detection.

[0031] <5> Clustering algorithms: core algorithms of unsupervised learning. Without label guidance, they automatically group similar data into the same class and those with large differences into different clusters based on sample feature similarity. Their core principle is to calculate the distance or similarity between samples; samples with high similarity are clustered together, autonomously completing data grouping.

[0032] To facilitate understanding of the embodiments of the present invention, firstly, in conjunction with Figure 1 Detailed explanation. Figure 1 This document illustrates a flowchart of an OPC model training method based on unsupervised learning, as described in an embodiment of the present invention. The OPC model training method based on unsupervised learning in this embodiment mainly includes the following steps:

[0033] Step S11: Construct the original OPC modeling dataset; wherein the original OPC modeling dataset includes: multiple lithographic mask patterns of different sizes, arrangements and structures.

[0034] Specifically, a batch of lithographic mask test pattern datasets are obtained to be modeled. These test patterns include various sizes, arrangements, and structures; for example, they include at least several mainstream structures such as dense linewidth patterns, sparse independent patterns, corner patterns, and hole patterns. These test patterns are then integrated into the original OPC modeling dataset.

[0035] Among them, dense linewidth patterns are characterized by narrowing of dense lines, increased spacing, and milder corner rounding. Narrowing of dense lines (CD contraction) causes adjacent diffracted light to overlap, reducing the central light intensity and making the lines thinner. Increased spacing causes the lines to "repel" each other, resulting in an actual spacing larger than designed. In this case, corner rounding is mild, the environment is symmetrical, and diffraction is uniform. Sparse independent patterns are characterized by wider linewidths (CD expansion), shorter line ends, and severe corner rounding. Wider linewidths (CD expansion) result in no cancellation of diffraction on one side, high edge light intensity, and thicker lines. Shortened line ends cause diffusion of light intensity at the line ends, "retraction" of the ends, and contact holes are prone to becoming elliptical. In this case, corners are severely rounded, right angles become rounded, and sharp corners are prone to disappearing. There is also a semi-dense or semi-sparse pattern, which is between dense and independent, such as array edges and local sparse regions. It is typically characterized by nonlinear CD (pattern critical dimension): the same linewidth has different dimensions at different locations, and edge distortion is between the two. All these phenomena will affect the conduction performance of the device.

[0036] Step S12: Perform multi-dimensional lithographic feature extraction on each lithographic mask pattern in the original OPC modeling dataset to obtain a multi-dimensional graphic feature vector for each lithographic mask pattern.

[0037] In one embodiment, step S12 includes: extracting multi-dimensional lithographic features such as pattern density, lithographic imaging light intensity, key pattern size, and pattern spacing for each lithographic mask pattern in the original OPC modeling dataset, and constructing a multi-dimensional pattern feature vector for each lithographic mask pattern.

[0038] Specifically, parameter analysis is performed on each lithographic mask pattern in the original OPC modeling dataset. For each lithographic mask test pattern, the core features of the multi-dimensional lithographic features are extracted using lithography simulation software: pattern density, lithographic imaging light intensity, pattern critical dimension (CD), and pattern spacing. These are combined into a four-dimensional feature vector, forming a multi-dimensional pattern feature vector for each lithographic mask test pattern.

[0039] Preferably, auxiliary features such as edge roughness, optical diffraction deviation, and process window margin are added to each lithographic mask pattern in the original OPC modeling dataset to ensure that the constructed multi-dimensional graphic feature vectors do not miss key samples of special process scenarios.

[0040] Pattern density directly determines the total exposure absorption and morphology of the photoresist after development. In semiconductor lithography, a typical problem caused by pattern density is that the critical dimension (CD) of dense and isolated areas on the same wafer can differ by 5-20 nm. This is one of the core issues that optical proximity correction (OPC) needs to address. The light intensity in lithography imaging is equivalent to the superposition of the light fields at various points on the image plane after the mask pattern is diffracted by the optical system. Light intensity directly determines CD. In positive photoresists, increased light intensity leads to increased dissolution, thus increasing CD (overexposure actually decreases CD). Furthermore, the superposition of incident light and substrate reflected light causes periodic fluctuations in light intensity within the photoresist layer, resulting in longitudinal non-uniformity of linewidth. The critical dimension (CD) refers to the minimum controllable feature size of the lithographic pattern, typically referring to linewidth, contact hole diameter, or minimum spacing. Pattern spacing refers to the minimum center-to-center distance or edge-to-edge distance between two adjacent identical patterns. Spacing determines whether the diffraction order can be captured by the objective lens. In layout design, the minimum allowable distance between two components is crucial. A distance smaller than this value will result in overlapping patterns. The minimum spacing must ensure that short circuits do not occur under edge effects. Spacing and CD (Distribution Density) together determine the pattern density, thus triggering density effects.

[0041] Step S13: Perform feature preprocessing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain standardized feature vectors and construct a standardized feature dataset.

[0042] In one embodiment, step S13 includes: normalizing, denoising, and outlier processing the multi-dimensional graphic feature vectors of each photolithographic mask pattern to obtain multi-dimensional graphic feature vectors that eliminate acquisition errors, and constructing a standardized feature dataset.

[0043] Specifically, the feature preprocessing includes normalizing, denoising, and outlier removal for the multi-dimensional graphic feature vectors of each photolithographic mask pattern. For example, the min-max normalization method is used to map all feature values ​​to the [0,1] interval, and the 3σ (3 times the standard deviation) criterion is used to remove faulty graphic samples with abnormal parameters. After feature preprocessing, the multi-dimensional graphic feature vectors eliminate differences in the dimensions of different features and acquisition errors, thereby constructing a standardized feature dataset.

[0044] Step S14: Perform automatic clustering on the standardized feature dataset to obtain sample sets of multiple differentiated graphic categories.

[0045] In one embodiment, step S14 includes: using an unsupervised learning algorithm to cluster and classify the standardized feature vectors in the standardized feature dataset according to clustering parameters to obtain a sample set of multiple differentiated graphic categories.

[0046] In one embodiment, step S14 further includes: the clustering parameters include at least: feature similarity and photolithographic response characteristics.

[0047] Specifically, unsupervised learning algorithms, particularly clustering algorithms, are used to automatically cluster and classify the standardized feature dataset. Test images with high feature similarity and consistent lithographic response characteristics are grouped into the same category, resulting in sample sets of multiple differentiated image categories. The clustering algorithms include K-Means, DBSCAN, and hierarchical clustering; any one of these algorithms can be used when clustering the standardized feature dataset. For example, using the K-Means unsupervised clustering algorithm, the optimal number of clusters is determined to be 8 based on the elbow rule. All test images are divided into 8 categories according to their lithographic response characteristics: dense high-sensitivity images, dense low-sensitivity images, sparse large-space images, and corner irregular images. These 8 categories are then integrated into a differentiated image category sample set.

[0048] Step S15: Perform stratified filtering of the sample set for each differentiated graphic category to construct a simplified OPC training dataset.

[0049] In one embodiment, step S15 includes: filtering samples in the sample set of each differentiated graphic category according to the filtering criteria to obtain core samples of multiple differentiated graphic categories; wherein the filtering criteria include at least: feature dispersion and process sensitivity; and integrating the core samples of all differentiated graphic categories to obtain a simplified OPC training dataset.

[0050] Specifically, for each differentiated graphic category's sample set, samples are filtered based on feature dispersion and process sensitivity indicators. Redundant, duplicate, and low-value samples are removed to obtain the most representative core sample set within each differentiated graphic category's sample set. These core sample sets from all differentiated graphic categories are then integrated to form a simplified OPC training dataset. For example, for each differentiated graphic category, the feature variance of the samples within that category is calculated. Samples with variances in the middle and closely aligned with the category's feature center are selected as core samples. Each category retains 10%-15% of representative samples, ultimately resulting in a simplified OPC training dataset.

[0051] Step S16: Train an OPC model based on the simplified OPC training dataset to perform optical proximity correction on the lithographic mask pattern.

[0052] In one embodiment, step S16 includes: training a preliminary OPC model using the simplified OPC training dataset; verifying whether the preliminary OPC model meets the standards, and if the OPC model does not meet the standards, adjusting the clustering parameters to obtain a new simplified OPC training dataset and training a new OPC model, until the obtained OPC model meets the standards, and obtaining the final OPC model.

[0053] like Figure 2 As shown, step S16 specifically includes:

[0054] Step S111: Train the OPC model using the simplified OPC training dataset;

[0055] Step S112: Verify whether the OPC model meets the standards;

[0056] Step S113: If the OPC model meets the requirements, the OPC model construction is completed;

[0057] Step S114: If the OPC model does not meet the standard, adjust the clustering parameters to obtain a new simplified OPC training dataset and train a new OPC model until the OPC model meets the standard.

[0058] Repeat steps S111 to S112 until the OPC model meets the requirements.

[0059] Specifically, step S114 includes: verifying whether the OPC model meets the standards; if the OPC model does not meet the standards, adjusting the clustering parameters and re-clustering and classifying the standardized feature vectors in the standardized feature dataset to obtain a new sample set of multiple differentiated graphic categories; re-stratifying and filtering the new sample set of multiple differentiated graphic categories to construct a new simplified OPC training dataset; training a new OPC model again based on the new simplified OPC training dataset to obtain a new OPC model, and verifying whether the new OPC model meets the standards; repeating this process until the OPC model meets the standards.

[0060] In one embodiment, step S16 further includes: verifying whether the OPC model meets the requirements by: verifying the OPC model from three dimensions: graphic size correction accuracy, special process scenario adaptability, and full process coverage; if the OPC model meets the required graphic size correction accuracy, special process scenario adaptability, and full process coverage, it is considered to meet the requirements; if the OPC model does not meet the required graphic size correction accuracy, special process scenario adaptability, and full process coverage, it is considered to fail to meet the requirements.

[0061] Specifically, the simplified OPC training dataset is used to iteratively train the OPC model. The OPC model is validated from multiple dimensions, including graphic size correction accuracy, adaptability to special process scenarios, and full process coverage. If the graphic size correction accuracy, adaptability to special process scenarios, and full process coverage are met, the OPC model construction is complete. If the graphic size correction accuracy, adaptability to special process scenarios, and full process coverage are not met, the clustering parameters are fine-tuned, and new clustering, classification, and hierarchical screening are performed to construct a new simplified OPC training dataset. The new simplified OPC training dataset is used to train the new OPC model again, and the new OPC model is verified to meet the requirements. This process is repeated until the OPC model meets the requirements for graphic size correction accuracy, adaptability to special process scenarios, and full process coverage.

[0062] To better illustrate the OPC model training method based on unsupervised learning of the present invention, the present invention provides the following specific embodiments.

[0063] Example of an OPC model training method based on unsupervised learning:

[0064] Step 1: Data Collection

[0065] This step includes: collecting a dataset of lithographic mask test images, obtaining a batch of lithographic mask test images to be modeled, and constructing the original OPC modeling dataset.

[0066] This embodiment collects a total of 10,000 sets of 5nm process lithography mask test patterns as the original OPC modeling dataset. These 10,000 sets of lithography mask test patterns contain lithography mask patterns of various sizes, arrangements, and structures, such as dense linewidth patterns, sparse independent patterns, corner patterns, and hole patterns, among other mainstream structural patterns.

[0067] Step 2: Multidimensional lithography feature extraction

[0068] This step includes: performing parameter analysis on each lithographic mask test pattern in the original OPC modeling dataset, extracting the core features from the multidimensional lithographic features, and obtaining the multidimensional graphic feature vector of each lithographic mask pattern.

[0069] In this embodiment, for each photolithography mask test pattern in the original OPC modeling dataset, four core features are extracted using photolithography simulation software: pattern density, photolithography imaging light intensity, measured CD size of the pattern (key size of the pattern), and minimum spacing between adjacent patterns (pattern spacing). These features are combined into a four-dimensional feature vector to construct a multi-dimensional pattern feature vector.

[0070] Step 3: Feature Preprocessing

[0071] This step includes: performing feature preprocessing on the multi-dimensional feature vectors of each photolithographic mask pattern to obtain standardized feature vectors, and constructing a standardized feature dataset. Feature preprocessing involves normalizing, denoising, and removing outliers from the feature vectors of all patterns to eliminate differences in the dimensions of different features and acquisition errors, thereby generating a standardized feature dataset.

[0072] In this embodiment, the min-max normalization method is used to map the feature values ​​of all standardized feature vectors to the [0,1] interval. The 3σ criterion is used to remove faulty graphic samples with abnormal parameters, and finally a standardized feature dataset consisting of 9820 sets of standardized feature vectors is obtained.

[0073] Step 4: Unsupervised learning clustering and classification

[0074] This step includes: automatically clustering the standardized feature dataset to obtain sample sets of multiple differentiated graphic categories. An unsupervised learning algorithm is used to cluster the standardized feature vectors in the standardized feature dataset according to clustering parameters, obtaining sample sets of multiple differentiated graphic categories. The clustering parameters include at least high feature similarity and lithographic response characteristics. Specifically, based on the standardized feature dataset, an unsupervised machine learning algorithm (K-Means, DBSCAN, or any one of hierarchical clustering) is used to perform automatic clustering, grouping test graphics with high feature similarity and consistent lithographic response characteristics into the same category, resulting in sample sets of multiple differentiated graphic categories.

[0075] This embodiment uses K-Means, an unsupervised learning clustering algorithm, and determines the optimal number of clusters to be 8 based on the elbow rule. The test images in the 9820 sets of standardized feature datasets are divided into 8 categories according to their lithographic response characteristics, namely, dense high-sensitivity images, dense low-sensitivity images, sparse large-spacing images, and corner irregular images. These 8 categories are then integrated into sample sets of 8 differentiated image categories.

[0076] Step 5: Stratified Screening of Samples

[0077] This step includes: stratified filtering of the sample set for each differentiated graphic category to construct a simplified OPC training dataset. Samples in the sample set for each differentiated graphic category are filtered according to screening criteria to obtain core samples for multiple differentiated graphic categories. For each differentiated graphic category, the most representative core samples within the category are selected based on feature dispersion and process sensitivity indicators, redundant, duplicate, and low-value samples are removed, and the core samples of all differentiated graphic categories are integrated to obtain the simplified OPC training dataset.

[0078] In this embodiment, for each differentiated graphic category, the sample feature variance is calculated based on the feature dispersion and process sensitivity index. Samples with the central variance that fit the category feature center are selected as core samples. 10%-15% of representative samples are retained for each category. Finally, 1200 sets of simplified training samples are selected as the simplified OPC training dataset.

[0079] Step Six: Model Training and Validation

[0080] This step includes: training an OPC model based on a simplified OPC training dataset to perform optical proximity correction on lithography mask patterns. An initial OPC model is trained using the simplified OPC training dataset. The model is then verified to ensure it meets the requirements. If the OPC model fails to meet the requirements, the clustering parameters are adjusted to obtain a new simplified OPC training dataset and a new OPC model is trained. This process continues until the obtained OPC model meets the requirements, resulting in the final OPC model. Iterative training of the OPC model is then performed using the simplified OPC training dataset. The model is validated from three dimensions: pattern size correction accuracy, adaptability to special process scenarios, and full process coverage. If the model meets the requirements, the model construction is complete. If it fails, the clustering parameters are fine-tuned, samples are re-selected, and iterative optimization is performed. This process is repeated until the OPC model meets the requirements, completing the OPC model construction.

[0081] This embodiment uses 1200 sets of simplified OPC training datasets to train the OPC model. After 500 iterations of verification, an OPC model that meets the required three dimensions of indicators: graphic size correction accuracy, adaptability to special process scenarios, and full process coverage is obtained.

[0082] To better illustrate the technical effects of the OPC model training method based on unsupervised learning of the present invention, the following specific embodiments are provided to verify the difference between the method described in the present invention and training with full data.

[0083] Example of comparing and verifying the OPC model training method based on unsupervised learning with training on full data:

[0084] Step 1: Data Collection

[0085] A total of 10,000 sets of 5nm process lithography mask test patterns were collected as the original OPC modeling dataset. These 10,000 sets of lithography mask test patterns contain lithography mask patterns of various sizes, arrangements, and structures, including various mainstream structural patterns such as dense linewidth patterns, sparse independent patterns, corner patterns, and hole patterns.

[0086] Step 2: Multidimensional lithography feature extraction

[0087] For each lithography mask test pattern in the original OPC modeling dataset, four core features are extracted using lithography simulation software: pattern area proportion density (pattern density), wafer exposure imaging peak light intensity (lithography imaging light intensity), actual measured CD size of the pattern (pattern key size), and minimum spacing between adjacent patterns (pattern spacing). These features are combined into a four-dimensional feature vector to construct a multi-dimensional pattern feature vector.

[0088] Step 3: Feature Preprocessing

[0089] The min-max normalization method was used to map the feature values ​​of all standardized feature vectors to the [0,1] interval. Failed graphic samples with abnormal parameters were removed by the 3σ criterion, and finally a standardized feature dataset consisting of 9820 sets of standardized feature vectors was obtained.

[0090] Step 4: Unsupervised learning clustering and classification

[0091] Using the K-Means unsupervised learning clustering algorithm, the optimal number of clusters was determined to be 8 based on the elbow rule. The test images in the 9820 sets of standardized feature datasets were divided into 8 categories according to their lithographic response characteristics: dense high-sensitivity images, dense low-sensitivity images, sparse large-spacing images, and corner irregular images. These 8 categories were then integrated into sample sets of 8 differentiated image categories.

[0092] Step 5: Stratified Screening of Samples

[0093] For each differentiated graphic category, the sample feature variance is calculated based on the feature dispersion and process sensitivity index. Samples with the central variance that fit the category feature center are selected as core samples. 10%-15% of representative samples are retained for each category. Finally, 1200 sets of simplified training samples are selected as the simplified OPC training dataset.

[0094] Step Six: Model Training and Validation

[0095] This embodiment uses 1200 sets of simplified OPC training datasets to train the OPC model. After 500 iterations of verification, an OPC model that meets the required three dimensions of indicators: graphic size correction accuracy, adaptability to special process scenarios, and full process coverage is obtained.

[0096] Step 7: Model Training and Comparative Validation

[0097] OPC models were trained using 10,000 sets of full data and 1,200 sets of simplified OPC training data, respectively. The OPC models obtained by both datasets were then validated. The validation results showed that the CD correction error deviation between the OPC model trained on the simplified OPC training dataset and the OPC model trained on the full data was less than 0.3 nm, and the model process coverage was consistent. At the same time, the training time of the OPC model trained on the simplified OPC training dataset was reduced by 58% compared with that of the OPC model trained on the full data, and the consumption of computing resources was also significantly reduced. This validates the effectiveness of the present invention in terms of data volume, accuracy assurance, coverage adaptation, and versatility.

[0098] Comparison and verification conclusions:

[0099] This invention achieves significant cost reduction and lightweighting by automatically classifying data using unsupervised learning, eliminating the need for manual sample labeling. It accurately distinguishes pattern process characteristics based on multi-dimensional lithography features, greatly eliminating redundant data, reducing the amount of OPC training data, and lowering the computational cost and iteration cycle of model training. Regarding accuracy assurance, it abandons traditional single-index screening methods, combining core lithography parameters such as density, light intensity, CD, and spacing for modeling and classification. This accurately covers pattern scenarios with different optical responses and process sensitivities, avoiding the omission of key samples and consistently ensuring the accuracy of OPC model correction. For comprehensive coverage and adaptation, it uses clustering to divide differentiated pattern categories, covering samples from all scenarios, including regular patterns, dense patterns, sparse patterns, and special structure patterns, effectively improving the process coverage of the OPC model and adapting to the diverse lithography process requirements of advanced processes. In terms of versatility, the unsupervised learning algorithm employed has high adaptability, requiring no customized labeling rules for different processes and layouts, and can be quickly transferred to various lithography processes and various mask OPC modeling scenarios.

[0100] Figure 3This is a schematic diagram of the OPC model training system based on unsupervised learning provided in an embodiment of the present invention. Figure 3 As shown, the system includes:

[0101] Data acquisition module 1 is used to acquire multiple photolithography mask test patterns to construct the original OPC modeling dataset;

[0102] Data processing module 2, connected to data acquisition module 1, is used to perform multi-dimensional lithographic feature extraction on each lithographic mask pattern in the original OPC modeling dataset, obtain multi-dimensional graphic feature vectors of each lithographic mask pattern, and perform feature preprocessing on the multi-dimensional graphic feature vectors of each lithographic mask pattern to obtain standardized feature vectors, and construct a standardized feature dataset.

[0103] The classification and filtering module 3 is connected to the data processing module 2. It uses an unsupervised learning algorithm to cluster and classify the standardized feature vectors in the standardized feature dataset according to the clustering parameters, obtain sample sets of multiple differentiated graphic categories, and perform stratified filtering on the samples in the sample sets of each differentiated graphic category to construct a simplified OPC training dataset.

[0104] The model validation module 4 is connected to the classification and filtering module 3. It is used to train an OPC model based on the simplified OPC training dataset, and to verify whether the OPC model meets the standards. If the OPC model does not meet the standards, it returns to the classification and filtering module 3 and adjusts the clustering parameters to obtain a new simplified OPC training dataset, and trains a new OPC model until the OPC model meets the standards.

[0105] It should be understood that the specific process of each module performing the above-mentioned steps has been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

[0106] It should also be understood that the module division in the embodiments of the present invention is illustrative and only represents one logical functional division; in actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of the present invention can be integrated into a single processor, exist as separate physical entities, or two or more modules can be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0107] Figure 4 This is a schematic block diagram of an electronic terminal provided in an embodiment of the present invention. Figure 4As shown, the electronic terminal 400 includes at least one processor 401, a memory 402, at least one network interface 403, and a user interface 405. The various components in the electronic terminal 400 are coupled together via a bus system 404. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4 The general will label all buses as bus systems.

[0108] The user interface 405 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0109] It is understood that memory 402 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0110] In this embodiment of the invention, the memory 402 is used to store various types of data to support the operation of the electronic terminal 400. Examples of this data include: any executable program for operation on the electronic terminal 400, such as the operating system 4021 and application programs 4022; the operating system 4021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 4022 may contain various applications, such as a media player, browser, etc., for implementing various application services. The OPC model training method based on unsupervised learning provided in this embodiment of the invention can be included in the application program 4022.

[0111] The methods disclosed in the above embodiments of the present invention can be applied to processor 401, or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 401 or by instructions in the form of software. The processor 401 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 401 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 401 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0112] In an exemplary embodiment, the electronic terminal 400 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to execute the aforementioned method.

[0113] In summary, this invention provides an OPC model training method, system, and terminal based on unsupervised learning. It constructs a simplified OPC training dataset by performing multi-dimensional lithographic feature extraction, feature preprocessing, clustering, classification, and hierarchical filtering on multiple lithographic mask test patterns. An OPC model is then trained based on this dataset to perform optical proximity correction on the lithographic mask patterns. This invention combines multi-dimensional lithographic characteristics to accurately cover different pattern scenarios, avoiding the omission of key samples and ensuring the accuracy of OPC model correction. It employs an unsupervised learning algorithm to lightweight sample data and uses clustering to classify differentiated pattern categories, effectively improving the process coverage of the OPC model. It fully adapts to the changing lithography process requirements under advanced process nodes and requires no customized annotation rules, allowing for rapid migration to various lithography processes. It combines high precision, full coverage, and strong versatility.

[0114] Therefore, this invention effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0115] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for training an OPC model based on unsupervised learning, characterized in that, include: Construct the original OPC modeling dataset; wherein, the original OPC modeling dataset includes: multiple lithographic mask patterns of different sizes, arrangements, and structures; Multidimensional lithographic feature extraction is performed on each lithographic mask pattern in the original OPC modeling dataset to obtain a multidimensional graphic feature vector for each lithographic mask pattern; The multi-dimensional graphic feature vectors of each lithographic mask pattern are preprocessed to obtain standardized feature vectors, and a standardized feature dataset is constructed. The standardized feature dataset is automatically clustered to obtain sample sets of multiple differentiated image categories; The sample sets for each differentiated graphic category are stratified and filtered to construct a simplified OPC training dataset; An OPC model is trained based on the simplified OPC training dataset to perform optical proximity correction on lithographic mask patterns.

2. The OPC model training method based on unsupervised learning according to claim 1, characterized in that, The step of extracting multi-dimensional lithographic features from each lithographic mask pattern in the original OPC modeling dataset to obtain a multi-dimensional graphic feature vector for each lithographic mask pattern includes: For each lithographic mask pattern in the original OPC modeling dataset, multi-dimensional lithographic features such as pattern density, lithographic imaging light intensity, key pattern dimensions, and pattern spacing are extracted to construct a multi-dimensional pattern feature vector for each lithographic mask pattern.

3. The OPC model training method based on unsupervised learning according to claim 1, characterized in that, The step of preprocessing the multi-dimensional graphic feature vectors of each photolithographic mask pattern to obtain standardized feature vectors and constructing a standardized feature dataset includes: Normalization, noise reduction, and outlier processing are performed on the multi-dimensional graphic feature vectors of each photolithographic mask pattern to obtain standardized feature vectors that eliminate acquisition errors, and a standardized feature dataset is constructed.

4. The OPC model training method based on unsupervised learning according to claim 1, characterized in that, The step of automatically clustering the standardized feature dataset to obtain a sample set of multiple differentially categorized graphics includes: An unsupervised learning algorithm is used to cluster and classify the standardized feature vectors in the standardized feature dataset according to the clustering parameters, thereby obtaining a sample set of multiple differentiated graphic categories.

5. The OPC model training method based on unsupervised learning according to claim 4, characterized in that, The clustering parameters include at least: feature similarity and photolithographic response characteristics.

6. The OPC model training method based on unsupervised learning according to claim 1, characterized in that, The process of stratifying and filtering the sample sets for each differentiated graphic category to construct a simplified OPC training dataset includes: The samples in the sample set of each differentiated graphic category are screened according to the screening criteria to obtain the core samples of multiple differentiated graphic categories; wherein, the screening criteria include at least: feature dispersion and process sensitivity. The core samples from all the differentiating graph categories are integrated to obtain a simplified OPC training dataset.

7. The OPC model training method based on unsupervised learning according to claim 1, characterized in that, The step of training an OPC model based on the simplified OPC training dataset to perform optical proximity correction on the lithographic mask pattern includes: The simplified OPC training dataset is used to train and obtain an initial OPC model; Verify whether the initially obtained OPC model meets the standards. If the OPC model does not meet the standards, adjust the clustering parameters to obtain a new simplified OPC training dataset and train it to obtain a new OPC model. Continue until the obtained OPC model meets the standards and obtain the final OPC model.

8. The OPC model training method based on unsupervised learning according to claim 7, characterized in that, The methods for verifying whether the OPC model meets the requirements include: The OPC model was verified from three dimensions: graphic size correction accuracy, adaptability to special process scenarios, and full process coverage. If the OPC model meets the required accuracy of graphic size correction, adaptability to special process scenarios, and full process coverage, then it is considered to have met the standards. If the OPC model does not meet the required accuracy of graphic size correction, adaptability to special process scenarios, and full process coverage, it is considered substandard.

9. An OPC model training system based on unsupervised learning, characterized in that, include: The system, employing the OPC model training method based on unsupervised learning as described in any one of claims 1-8, comprises: The data acquisition module is used to acquire multiple lithography mask test patterns to construct the original OPC modeling dataset; The data processing module, connected to the data acquisition module, is used to perform multi-dimensional lithographic feature extraction on each lithographic mask pattern in the original OPC modeling dataset, obtain a multi-dimensional graphic feature vector for each lithographic mask pattern, and perform feature preprocessing on the multi-dimensional graphic feature vector of each lithographic mask pattern to obtain a standardized feature vector, thereby constructing a standardized feature dataset. The classification and filtering module is connected to the data processing module. It uses an unsupervised learning algorithm to cluster and classify the standardized feature vectors in the standardized feature dataset according to the clustering parameters, obtain sample sets of multiple differentiated graphic categories, and perform stratified filtering on the samples in the sample sets of each differentiated graphic category to construct a simplified OPC training dataset. The model validation module, connected to the classification and filtering module, is used to train an OPC model based on the simplified OPC training dataset, and to verify whether the OPC model meets the standards. If the OPC model does not meet the standards, the module returns to the classification and filtering module and adjusts the clustering parameters to obtain a new simplified OPC training dataset, and trains a new OPC model until the OPC model meets the standards.

10. An electronic terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method as described in any one of claims 1 to 8.

Images