Layout hot spot detection method, device and equipment and medium

Abstract

The invention relates to the technical field of chip manufacturing, in particular to a layout hot spot detection method, device and equipment and a medium, and aims to improve the accuracy of layout hot spot detection. According to the method, a plurality of graphs contained in a to-be-detected layout are converted into a text sequence, the text sequence comprises a plurality of texts recorded in a set sequence, each text is used for describing one graph, and the set sequence is used for representing arrangement positions of the graphs in the layout. Therefore, each graph in the layout is simply and accurately expressed in the form of the text sequence; inputting the text sequence into a pre-trained layout language model to enable the layout language model to perform feature extraction on each text in the text sequence, judging whether the text contains features corresponding to the hot spots or not, and obtaining a detection result that whether a layout corresponding to the text sequence and output by the layout language model contains the hot spots or not, therefore, the accuracy of layout hot spot detection can be further improved.

Description

Technical Field

[0001] This application relates to the field of chip manufacturing technology, and in particular to a method, apparatus, equipment and medium for detecting hot spots on a layout. Background Technology

[0002] Layouts are susceptible to defects caused by the photolithography process, thus requiring the detection of lithographic defects, i.e., lithographic hotspots. Traditional defect detection methods, such as photolithography simulation and pattern matching, have limitations. Photolithography simulation offers high accuracy but is time-consuming and computationally resource-intensive. Pattern matching is fast but has low accuracy for unknown layouts. Machine learning-based defect detection is currently a more effective method, but it struggles with learning long-distance layout correlations on larger layouts and often performs poorly on more challenging datasets due to design problems. To address these issues, researchers have proposed detection models based on attention mechanisms and graph neural networks. In industry, layout graphics are typically stored as binary Graphic Design System (GDS) or OASIS files. Previous data preprocessing methods converted binary files into pixel-based images for feature extraction and analysis. These image-based input methods introduce irrelevant redundant information, such as pixel values ​​that are not part of the graphic outline, into the model. Meanwhile, to meet the model's input requirements, these image inputs are often downsampled, causing irreversible distortion. In summary, image-based layout feature extraction suffers from redundant layout representations and lossy downsampling, leading to feature loss and low accuracy in layout lithography hotspot detection.

[0003] Therefore, improving the accuracy of hotspot detection on the map has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a method, apparatus, device, and medium for detecting hotspots on a map, which can improve the accuracy of hotspot detection on a map.

[0005] Firstly, this application provides a method for detecting hotspots on a map, the method comprising:

[0006] The multiple graphics contained in the layout to be detected are converted into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text is used to describe a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout;

[0007] The text sequence is input into a pre-trained map language model, and the detection result of whether the map contains hotspots is obtained from the output of the map language model.

[0008] In one possible implementation, converting multiple graphics of the layout to be detected into a text sequence includes:

[0009] For each graphic in the layout to be detected, a set vertex of the graphic is used as the starting point, and each line of the graphic is described in a clockwise or counterclockwise order to obtain the text of the graphic;

[0010] Determine the arrangement positions of multiple graphics contained in the map, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement positions.

[0011] In one possible implementation, the step of describing each line of the graphic as a starting point, in a clockwise or counterclockwise order, to obtain the text of the graphic, includes:

[0012] The coordinate information of the starting point is determined by taking a specific vertex of the graphic as the starting point.

[0013] Describe the line direction and length of each line in the figure in a clockwise or counterclockwise order;

[0014] Based on the coordinates of the starting point, as well as the direction and length of each line, the text of the graphic is obtained.

[0015] In one possible implementation, the text sequence further includes a map identifier representing the map and a graphic identifier representing the graphic type; different graphic types have different graphic identifiers.

[0016] The prefix of the map identifier is located before the text corresponding to the first image in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last image in the text sequence.

[0017] The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

[0018] In one possible implementation, the arrangement position includes arranging each graphic in a clockwise or counterclockwise direction, starting from a preset point on the map.

[0019] Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset vertex are arranged in ascending order.

[0020] In one possible implementation, the training process of the layout language model includes:

[0021] Obtain any sample text sequence from the sample set; after randomly masking the sample text sequence, input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model;

[0022] Obtain any sample text sequence from the sample set and the corresponding label information of the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain a map language model; the label information is used to indicate the presence of hotspots in the map corresponding to the sample text sequence.

[0023] In one possible implementation, the layout language model includes a multi-head self-attention layer;

[0024] The multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

[0025] Secondly, embodiments of this application provide a map hotspot detection device, the device comprising:

[0026] A conversion module is used to convert multiple graphics contained in a layout to be detected into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text describes a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout;

[0027] The detection module is used to input the text sequence into a pre-trained map language model and obtain the detection result of whether the map output by the map language model contains hotspots.

[0028] In one possible implementation, the conversion module is specifically used to, for each graphic in the layout to be detected, take a set vertex of the graphic as the starting point, describe each line of the graphic in a clockwise or counterclockwise order to obtain the text of the graphic; determine the arrangement position of multiple graphics contained in the layout, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement position.

[0029] In one possible implementation, the conversion module is specifically used to determine the coordinate information of the starting point by taking a set vertex of the graphic as the starting point; to describe the line direction and line length of each line in the graphic in a clockwise or counterclockwise order; and to obtain the text of the graphic based on the coordinate information of the starting point and the line direction and line length of each line.

[0030] In one possible implementation, the device further includes:

[0031] The training module is used to acquire any sample text sequence from the sample set; randomly mask the sample text sequence and input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model; acquire any sample text sequence from the sample set and the label information corresponding to the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the landscape corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain the landscape language model; the label information is used to indicate the presence of hotspots in the landscape corresponding to the sample text sequence.

[0032] Thirdly, this application also provides an electronic device, which includes at least a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the steps of the method as described in any of the first aspects.

[0033] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in any of the first aspects.

[0034] Fifthly, this application provides a computer program product comprising: computer program code, which, when executed on a computer, causes the computer to perform the steps of any of the methods described in the first aspect.

[0035] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection. Attached Figure Description

[0036] To more clearly illustrate the implementation methods in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0037] Figure 1 A schematic diagram of a layout hotspot detection process is provided for some embodiments of this application;

[0038] Figure 2 Another schematic diagram of a layout hotspot detection process is provided for some embodiments of this application;

[0039] Figure 3 A schematic diagram of a layout and its corresponding text sequence provided for some embodiments of this application;

[0040] Figure 4 A schematic diagram of another layout and its corresponding text sequence provided for some embodiments of this application;

[0041] Figure 5 A schematic diagram of a layout language model training process is provided for some embodiments of this application;

[0042] Figure 6 A schematic diagram of a layout language model provided for some embodiments of this application;

[0043] Figure 7 A schematic diagram of the structure of a layout hotspot detection device is provided for some embodiments of this application;

[0044] Figure 8 This is a schematic diagram of the structure of an electronic device provided for some embodiments of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, a further detailed description of this application will be provided below with reference to the accompanying drawings. Obviously, the embodiments described in this application are merely some embodiments, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0047] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0048] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0049] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0051] Before introducing the hotspot detection method provided in the embodiments of this application, the background technology of the embodiments of this application will be introduced first for ease of understanding.

[0052] Layout defects are easily introduced by the photolithography process, necessitating the detection of these defects, i.e., photolithographic hotspots. Traditional defect detection methods, such as photolithography simulation and pattern matching, have limitations. Photolithography simulation offers high accuracy but is time-consuming and computationally resource-intensive. Pattern matching is fast but has low accuracy for unknown layouts. Machine learning-based defect detection is currently more effective, but it struggles with learning long-range layout correlations on larger layouts and often performs poorly on more challenging datasets due to design issues. To address these problems, researchers have proposed detection models based on attention mechanisms and graph neural networks. In industry, layout graphics are typically stored as binary GDSII or OASIS files. Previous data preprocessing methods converted these binary files into pixel-based images for feature extraction and analysis. These image-based input methods introduce irrelevant redundant information, such as pixel values ​​of non-graphical contours, into the model. Furthermore, these image inputs are often downsampled to meet model requirements, leading to irreversible distortion. In summary, layout feature extraction based on image input suffers from redundant layout representation and lossy downsampling, which leads to the loss of layout features and results in low accuracy of layout lithographic hotspot detection.

[0053] Based on this, in order to improve the accuracy of layout lithography hotspot detection, this application provides a layout hotspot detection method. In this method, multiple graphics contained in the layout to be detected are converted into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text describes a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout; the text sequence is input into a pre-trained layout language model to obtain the detection result of whether the layout contains hotspots, output by the layout language model.

[0054] Example 1:

[0055] Figure 1 This is a schematic diagram illustrating a layout hotspot detection process provided for some embodiments of this application. For example... Figure 1 As shown, the process includes the following steps:

[0056] S101: Convert multiple graphics contained in the layout to be detected into a text sequence; the text sequence contains multiple texts recorded in a set order, wherein each text is used to describe a graphic; the set order represents the arrangement position of the multiple graphics in the layout.

[0057] The hotspot detection method in this application is applied to an electronic device, which may be a PC, server, or other such device.

[0058] Due to factors such as overly close image lines in the layout, line adhesion may occur during photolithography, a defect known as layout hotspots. Figure 1 Typically represented as binary layout files, image preprocessing methods convert these binary files into images containing many pixels for feature extraction and analysis. Image-based input methods introduce irrelevant redundant information into the model, such as pixel values ​​that are not part of the graphic contour. Furthermore, to meet model input requirements, these images are often downsampled, causing irreversible distortion. In summary, image-based layout feature extraction suffers from redundant layout representation and lossy downsampling, leading to feature loss.

[0059] To ensure the integrity of the map features and discard irrelevant redundant information, in this embodiment, multiple graphics contained in the map to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes the outline of a graphic in the map, and the predetermined order represents the arrangement of the multiple graphics in the map. The graphics contained in the map are generally polygons. The predetermined order can be counter-clockwise or clockwise, etc., and this application does not limit this.

[0060] Figure 2 This is a schematic diagram illustrating another layout hotspot detection process provided for some embodiments of this application. For example... Figure 2 As shown, the layout contains five polygons, numbered 1-5. These five polygons can be converted into the text sequence "1, 2, 3, 4, 5". Text 1 describes the outline of polygon 1, text 2 describes the outline of polygon 2, text 3 describes the outline of polygon 3, and so on. Specifically, each text can describe each line and / or vertex that constitutes a polygon, fully representing the outline of each polygon. For example, text 1 describes the top-left vertex V1 of polygon 1, and the lines E1, E2, E3, and E4 connected to it. Therefore, text 1 can be represented as V1, E1, E2, E3, E4. The text representation for each of the other polygons is the same and will not be repeated here.

[0061] S102: Input the text sequence into a pre-trained map language model and obtain the detection result of whether the map contains hotspots, output by the map language model.

[0062] To improve the accuracy of hotspot detection on the map, a map language model can be pre-trained based on sample text sequences and a language model. The language model can be a Large Language Model (LLM), such as a bidirectional encoding language model like BERT, to better understand global and local feature information in the text sequence. Since this embodiment converts multiple models contained in the map into text sequences, it does not contain redundant features and does not require downsampling or other processing. The text sequence corresponding to the map to be detected is directly input into the pre-trained map language model, allowing the map language model to extract lossless feature information from each text in the text sequence and determine whether it contains feature information corresponding to hotspots. This determines whether the map corresponding to the text sequence contains hotspots, further improving the accuracy of hotspot detection.

[0063] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection.

[0064] Example 2:

[0065] To convert multiple graphics in a map into an accurate and concise text sequence, and further improve the accuracy of map hotspot detection, based on the above embodiments, in this embodiment, the conversion of multiple graphics of the map to be detected into a text sequence includes:

[0066] For each graphic in the layout to be detected, a set vertex of the graphic is used as the starting point, and each line of the graphic is described in a clockwise or counterclockwise order to obtain the text of the graphic;

[0067] Determine the arrangement positions of multiple graphics contained in the map, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement positions.

[0068] To convert multiple graphics in a layout into an accurate and concise text sequence, in this embodiment, for each graphic in the layout to be detected, a vertex of the graphic can be pre-set as the starting point, and each line of the graphic can be described in a clockwise or counterclockwise order to obtain the text of the graphic. The arrangement of the multiple graphics in the layout is determined, and the set order of the text corresponding to the multiple graphics is determined according to the arrangement position. The multiple texts corresponding to the multiple graphics are then constructed into a text sequence according to this set order.

[0069] In this embodiment, for each graphic in the layout to be detected, a set vertex of the graphic is used as the starting point, and each line of the graphic is described in a clockwise or counterclockwise order to obtain the text of the graphic. The arrangement position of multiple graphics contained in the layout is determined, and the set order of the text corresponding to each graphic is determined according to the arrangement position. The text corresponding to each graphic is arranged in the set order to construct a text sequence, thereby converting multiple graphics in the layout into an accurate and concise text sequence, further improving the accuracy of layout hotspot detection.

[0070] Example 3:

[0071] To convert multiple graphics in a map into an accurate and concise text sequence and further improve the accuracy of map hotspot detection, based on the above embodiments, in this embodiment, the step of using a set vertex of the graphic as a starting point and describing each line of the graphic in a clockwise or counterclockwise order to obtain the text of the graphic includes:

[0072] The coordinate information of the starting point is determined by taking a specific vertex of the graphic as the starting point.

[0073] Describe the line direction and length of each line in the figure in a clockwise or counterclockwise order;

[0074] Based on the coordinates of the starting point, as well as the direction and length of each line, the text of the graphic is obtained.

[0075] To convert multiple graphics in a layout into an accurate and concise text sequence, the vertices of the graphics can be set as the starting point. The coordinate information of the starting point is determined, and the direction and length of each line in the graphics are described in a clockwise or counterclockwise order. Based on the coordinate information of the starting point, as well as the direction and length of each line, the text of the graphics is obtained.

[0076] Specifically, for example Figure 3This diagram illustrates a layout and its corresponding text sequence for some embodiments of this application. As shown in Figure 3, the layout contains three polygons, M1, M2, and M3, arranged in a clockwise order. For each polygon, the top-left vertex is defined as the starting point. Each line of the polygon is described in a clockwise order, resulting in the polygon's text. For example, for M1, the coordinates of the top-left vertex are (x0, y0). Each line connected in a clockwise order is d1 (horizontal to the right), d2 (vertical downward), d3 (horizontal to the left), d4 (vertical upward), d5 (horizontal to the left), and d6 (vertical upward). The resulting text for M1 could be "x0, y0, d1 (horizontal to the right), d2 (vertical downward), d3 (horizontal to the left), d4 (vertical upward), d5 (horizontal to the left), d6 (vertical upward)". The direction of each line can be represented by an arrow. The same principle applies to M2 and M3, and will not be elaborated further here. The positions of the three polygons M1, M2, and M3 in the map are determined to be the first, second, and third positions, respectively. The text corresponding to each of the three polygons M1, M2, and M3 is constructed into a text sequence according to the order of their positions.

[0077] Since natural language models cannot understand non-semantic input, this embodiment encodes the two-dimensional spatial representation of the map into a lossless one-dimensional sequence input. This requires direct lossless high-level semantic encoding of the map in binary file form to preprocess the map input. In this embodiment, a set vertex of the graphic is used as the starting point, and its coordinate information is determined. The direction and length of each line in the graphic are described in a clockwise or counterclockwise order. Based on the coordinate information of the starting point and the direction and length of each line, the text of the graphic is obtained. This allows for lossless semantic encoding of each graphic in the map, representing a single graphic without loss of absolute shape and relative spatial information, further improving the accuracy of map hotspot detection.

[0078] Example 4:

[0079] To further improve the accuracy of hotspot detection on the map, based on the above embodiments, in this embodiment, the text sequence further includes a map identifier representing the map and a graphic identifier representing the graphic type; the graphic identifiers for different graphic types are different.

[0080] The prefix of the map identifier is located before the text corresponding to the first graphic in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last graphic in the text sequence.

[0081] The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

[0082] The layout includes metal layer layouts and via layer layouts. Different types of layouts contain different graphic types and may generate different hotspot areas. To distinguish the layouts, layout identifiers can be added to the text sequence. These layout identifiers can include clip heads and clip tails. The clip head precedes the text corresponding to the first graphic in the text sequence, and the clip tail follows the text corresponding to the last graphic. To more accurately encode different types of graphics, graphic identifiers representing different graphic types can be added to the text sequence. These graphic identifiers can also include clip heads and clip tails. The clip head precedes the coordinates of the starting point of the corresponding graphic, and the clip tail follows the last line of the corresponding graphic.

[0083] Specifically, to differentiate between different clipped layouts, a layout identifier, or hierarchical affix, can be added to the text sequence. This identifier distinguishes between clipped layouts. For basic layout input, the encoding process is multi-level, divided into two levels: the outermost level is the clipped single layout (Clip). The layout identifier for a single layout consists of a prefix indicating the start of the clip (Start Of Clip, SOC) and a suffix indicating the end of the clip (End Of Clip, EOC). As described above... Figure 3 The layout shown is a metal layer layout, with the inner layer consisting of individual graphics contained within a single layout, typically polygons. The graphic identifier for each individual polygon consists of a prefix indicating the start of the polygon (SOP) and a suffix indicating the end of the polygon (EOP).

[0084] Figure 4 This is a schematic diagram illustrating another layout and its corresponding text sequence provided for some embodiments of this application. For example... Figure 4The layout shown is a via layer layout. Since the main difference between different layouts lies in the types of graphics they contain, a common layout identifier format can be used, with different graphic identifiers for different graphic types. The via layer layout contains two independent graphics with different functions: via graphics (Via) and sub-resolution assist feature (SRAF). The via graphic identifier consists of a prefix indicating the start of the via (Start Of Via, SOV) and a suffix indicating the end of the via (End Of Via, EOV). The sub-resolution assist feature identifier consists of a prefix indicating the start of the sub-resolution assist feature (Start Of SRAF, SOS) and a suffix indicating the end of the sub-resolution assist feature (End Of SRAF, EOS). The prefix of the layout identifier precedes the text corresponding to the first graphic in the text sequence, and the suffix follows the text corresponding to the last graphic in the text sequence. The prefix of the graphic identifier precedes the coordinate information of the starting point of the corresponding graphic, and the suffix follows the last line of the corresponding graphic. This allows for better location of the text sequence corresponding to each map, as well as the text corresponding to each graphic within the text sequence.

[0085] In this embodiment, the layout of the binary file is processed through multi-level lossless semantic representation. During the process of converting the layout into a text sequence, a layout identifier and a graphic identifier for each graphic are added. Different graphic identifiers are used for different graphic types, which can more accurately distinguish the text sequences of different layouts and the text corresponding to each graphic in the text sequence. This achieves lossless representation of the layout and also obtains semantic input that the layout language model can understand, further improving the accuracy of layout hotspot detection.

[0086] Example 5:

[0087] To further improve the accuracy of hotspot detection on the map, based on the above embodiments, in this embodiment, the arrangement position includes a sequence in which each graphic is arranged in a clockwise or counterclockwise direction, starting from a preset point on the map.

[0088] Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset point can be arranged in ascending order.

[0089] To further improve the accuracy of hotspot detection on the map, during the process of converting each graphic on the map into a text sequence, any point on the map can be used as a preset point. Starting from this preset point, the arrangement of each graphic is determined by arranging it in a clockwise or counterclockwise direction. Alternatively, starting from the preset point on the map, the arrangement of each graphic is determined by determining the distance from its corresponding vertex to the preset point in ascending order.

[0090] Specifically, as mentioned above Figure 3 and Figure 4 For example, Figure 3 As shown, the top left corner vertex of the layout can be used as the preset point. If the three polygons M1, M2, and M3 contained in the layout are arranged in a clockwise direction, then the positions of the three polygons M1, M2, and M3 are the first, second, and third positions, respectively.

[0091] like Figure 4 As shown, the top-left vertex of the layout can be used as the preset point. The distances from the corresponding vertices (i.e., the top-left vertices) of the 13 polygons S1, S2, V1, S3, S4, V2, V3, S5, V4, S6, S7, S8, and S9 in the layout to the preset point are arranged in ascending order. Thus, the arrangement positions of each polygon S1, S2, V1, S3, S4, V2, V3, S5, V4, S6, S7, S8, and S9 are the first to the thirteenth, respectively.

[0092] In the embodiments of this application, the arrangement position includes starting from a preset point on the map and arranging each graphic in a clockwise or counterclockwise direction, or starting from a preset point on the map and arranging the distances from the corresponding vertices of each graphic to the preset point in ascending order, thereby further improving the accuracy of map hotspot detection.

[0093] Example 6:

[0094] To further improve the accuracy of map hotspot detection, based on the above embodiments, in this application embodiment, the training process of the map language model includes:

[0095] Obtain any sample text sequence from the sample set; after randomly masking the sample text sequence, input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model;

[0096] Obtain any sample text sequence from the sample set and the corresponding label information of the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain a map language model; the label information is used to indicate the presence of hotspots in the map corresponding to the sample text sequence.

[0097] Since standard language models often only allow for unidirectional training, bidirectional encoding language models such as BERT can be used to more accurately extract feature information from text sequences of semantic maps, thereby better understanding global and local information in the text sequences. In this embodiment, a map language model is trained based on BERT as the original language model to achieve defect detection of the semantic map. The specific training process includes two stages: pre-training and fine-tuning. In the pre-training stage, any sample text sequence from the sample set is obtained, and after random masking, it is input into the original language model to obtain the reconstructed text sequence output by the original language model. The reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence. The loss value between the reconstructed text sequence and the sample text sequence before random masking is determined. The parameters in the original language model are adjusted based on the loss value to obtain the adjusted language model. In the fine-tuning stage, any sample text sequence from the sample set and the corresponding label information are obtained. The label information is used to indicate the hotspots in the map corresponding to the sample text sequence. The sample text sequence and label information are input into the adjusted language model; the detection results of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots are obtained; based on the label information and detection results, the adjusted language model is fine-tuned to obtain the map language model.

[0098] Figure 5 This is a schematic diagram illustrating a layout language model training process provided for some embodiments of this application. For example... Figure 5As shown, the GDS maps are first preprocessed using semantic encoding to obtain a sample set containing text sequences and label information for each map. The GDS maps can be from the ICCAD2012 metal dataset and the ICCAD2020 via dataset. Since label information is not needed in the pre-training stage, the label information is removed after randomly masking the sample text sequences. This removed text is then fed into the original language model via an embedding layer to obtain the reconstructed text sequence output by the original language model. The reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portions of the sample text sequences. A loss function is used to determine the loss between the reconstructed text sequence and the ground truth sample text sequence before random masking. The parameters in the original language model are adjusted based on the loss value to obtain the adjusted language model. In the fine-tuning stage, the GDS maps are similarly preprocessed, and a sample set containing text sequences and label information for each map is obtained through semantic encoding. The sample text sequence and label information are input into the adjusted language model; the detection results of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots are obtained; based on the label information and detection results, the adjusted language model is fine-tuned to obtain the map language model.

[0099] This map language model directly processes binary GDS map files and performs multi-level semantic encoding on the map to achieve lossless representation of the map and semantic representation that is understandable by the linguistic model. The map language model structure is based on a linguistic model and implements a hotspot detection process through pre-training and fine-tuning stages.

[0100] In this embodiment, during the pre-training stage, to better enable the original language model to learn the structural information within the map, the map's label information is removed, and a masked language modeling unsupervised learning method is used. Masked language modeling, by randomly masking portions of the text sequence and reconstructing the corresponding parts, allows the pre-trained original language model to achieve a deeper understanding of the global and local information of the map. During the fine-tuning stage, the weights of the pre-trained language model are reused. By reconstructing the map, the language model has a better starting point for iterative adjustments during weight fine-tuning. The input layer of the map language model is adjusted to the dimension used for classification tasks to predict hotspots and non-hotspots.

[0101] By encoding binary layout files to obtain sample text sequences, in the pre-training phase, the randomly masked sample text sequences are reconstructed based on the original language model, allowing the original language model to learn the semantic features of the sample text sequences, resulting in an adjusted language model. In the fine-tuning phase, the adjusted language model is fine-tuned based on the sample text sequences and their corresponding label information to obtain a layout language model, thereby improving the accuracy of layout hotspot detection. On both metal layer and via layer datasets with auxiliary lithographic patterns, this model achieves at least a 2% improvement in hotspot accuracy and a nearly 20% reduction in non-hotspot false positive rate compared to state-of-the-art models, demonstrating significant potential for further development.

[0102] In this embodiment, a one-dimensional text sequence is obtained by directly performing multi-level semantic encoding on the binary layout file, achieving lossless representation of the layout file and transforming the layout into semantic input that can be understood by the language model. A model training technique combining pre-training and fine-tuning achieves excellent hotspot detection results. This not only significantly improves the accuracy and efficiency of hotspot detection but also reduces computational resource consumption. Furthermore, it verifies the potential capabilities of large language models in layout representation, demonstrating high development prospects.

[0103] Example 7:

[0104] To further improve the accuracy of map hotspot detection, based on the above embodiments, in this application embodiment, the map language model includes a multi-head self-attention layer;

[0105] The multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

[0106] To further improve the accuracy of hotspot detection on the map, the map language model used in this embodiment includes a multi-head self-attention layer. The multi-head self-attention layer processes the input text sequence through linearization and scaling dot product attention, thereby enabling more accurate extraction of semantic features contained in the text sequence and further improving the accuracy of hotspot detection on the map.

[0107] Figure 6 This is a schematic diagram of a layout language model provided for some embodiments of this application. For example... Figure 6As shown, the map language model comprises an embedding layer, a map language model layer, and an output layer. The embedding layer embeds the map input (text sequence) after lossless hierarchical semantic encoding preprocessing into sequential sentences containing label and structural position information. The map language model layer is the core layer of the map language model, extracting feature values ​​from the text sequence. It includes multi-head self-attention, residual regularization, and a feedback network. Multi-head self-attention maps inputs to queues (Q), keys (K), and values ​​(V) and merges them after linearization and scaled dot-product attention, enabling the map language model to learn the contextual relationships of texts that trigger hotspot maps. The scaled dot-product attention consists of a matrix multiplication module (MatMul) with a broadcast mechanism, a softmax layer, a mask layer, and a scale layer. The mask residual regularization layer maintains the data distribution within the model. The feedback network is a linear layer that deepens the network. The output layer can be adjusted according to the stage of model use to complete different training tasks.

[0108] In this embodiment, the input text sequence is processed by a multi-head self-attention layer, thereby extracting more accurate feature information and further improving the accuracy of map hotspot detection.

[0109] Example 8:

[0110] Based on the same technical concept, and building upon the above embodiments, this application provides a map hotspot detection device. Figure 7 This is a schematic diagram of a layout hotspot detection device provided for some embodiments of this application. For example... Figure 7 As shown, the device includes:

[0111] The conversion module 701 is used to convert multiple graphics contained in the layout to be detected into a text sequence; the text sequence contains multiple texts recorded in a set order, wherein each text is used to describe a graphic; the set order represents the arrangement position of the multiple graphics in the layout;

[0112] The detection module 702 is used to input the text sequence into a pre-trained map language model and obtain the detection result of whether the map output by the map language model contains hotspots.

[0113] In one possible implementation, the conversion module 701 is specifically used to, for each graphic in the layout to be detected, take a set vertex of the graphic as the starting point, describe each line of the graphic in a clockwise or counterclockwise order to obtain the text of the graphic; determine the arrangement position of multiple graphics contained in the layout, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement position.

[0114] In one possible implementation, the conversion module 701 is specifically used to determine the coordinate information of the starting point by taking a set vertex of the graphic as the starting point; to describe the line direction and line length of each line in the graphic in a clockwise or counterclockwise order; and to obtain the text of the graphic based on the coordinate information of the starting point and the line direction and line length of each line.

[0115] In one possible implementation, the text sequence further includes a map identifier representing the map and a graphic identifier representing the graphic type; different graphic types have different graphic identifiers.

[0116] The prefix of the map identifier is located before the text corresponding to the first graphic in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last graphic in the text sequence.

[0117] The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

[0118] In one possible implementation, the arrangement position includes arranging each graphic in a clockwise or counterclockwise direction, starting from a preset point on the map.

[0119] Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset vertex are arranged in ascending order.

[0120] In one possible implementation, the device further includes:

[0121] The training module is used to acquire any sample text sequence from the sample set; randomly mask the sample text sequence and input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model; acquire any sample text sequence from the sample set and the label information corresponding to the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the landscape corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain the landscape language model; the label information is used to indicate the presence of hotspots in the landscape corresponding to the sample text sequence.

[0122] In one possible implementation, the layout language model includes a multi-head self-attention layer; the multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

[0123] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection.

[0124] Example 9:

[0125] Based on the same technical concept, and building upon the above embodiments, an electronic device is provided in this application. Figure 8 This is a schematic diagram of the structure of an electronic device provided for some embodiments of this application. For example... Figure 8 As shown, the electronic device includes: a processor 801, a communication interface 802, a memory 803, and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other through the communication bus 804.

[0126] The memory 803 stores a computer program, which, when executed by the processor 801, causes the processor 801 to perform the following steps:

[0127] The multiple graphics contained in the layout to be detected are converted into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text is used to describe a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout;

[0128] The text sequence is input into a pre-trained map language model, and the detection result of whether the map contains hotspots is obtained from the output of the map language model.

[0129] In one possible implementation, the processor 801 is further configured to, for each graphic in the layout to be detected, use a set vertex of the graphic as a starting point and describe each line of the graphic in a clockwise or counterclockwise order to obtain the text of the graphic.

[0130] Determine the arrangement positions of multiple graphics contained in the map, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement positions.

[0131] In one possible implementation, the processor 801 is further configured to take a set vertex of the graphic as a starting point, determine the coordinate information of the starting point; describe the line direction and line length of each line in the graphic in a clockwise or counterclockwise order; and obtain the text of the graphic based on the coordinate information of the starting point and the line direction and line length of each line.

[0132] In one possible implementation, the text sequence further includes a map identifier representing the map and a graphic identifier representing the graphic type; different graphic types have different graphic identifiers.

[0133] The prefix of the map identifier is located before the text corresponding to the first graphic in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last graphic in the text sequence.

[0134] The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

[0135] In one possible implementation, the arrangement position includes arranging each graphic in a clockwise or counterclockwise direction, starting from a preset point on the map.

[0136] Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset vertex are arranged in ascending order.

[0137] In one possible implementation, the processor 801 is further configured to: acquire any sample text sequence from the sample set; randomly mask the sample text sequence and input it into the original language model to acquire the reconstructed text sequence output by the original language model; the reconstructed text sequence includes the predicted value corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain an adjusted language model; acquire any sample text sequence from the sample set and the label information corresponding to the sample text sequence; input the sample text sequence and the label information into the adjusted language model; acquire the detection result of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain a map language model; the label information is used to indicate the presence of hotspots in the map corresponding to the sample text sequence.

[0138] In one possible implementation, the layout language model includes a multi-head self-attention layer;

[0139] The multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

[0140] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0141] Communication interface 802 is used for communication between the above-mentioned electronic device and other devices.

[0142] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0143] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0144] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection.

[0145] Based on the same inventive concept and the above embodiments, this application provides a computer-readable storage medium storing a computer program executable by a processor. When the program runs on the processor, it causes the processor to perform the following steps:

[0146] The multiple graphics contained in the layout to be detected are converted into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text is used to describe a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout;

[0147] The text sequence is input into a pre-trained map language model, and the detection result of whether the map contains hotspots is obtained from the output of the map language model.

[0148] In one possible implementation, converting multiple graphics of the layout to be detected into a text sequence includes:

[0149] For each graphic in the layout to be detected, a set vertex of the graphic is used as the starting point, and each line of the graphic is described in a clockwise or counterclockwise order to obtain the text of the graphic;

[0150] Determine the arrangement positions of multiple graphics contained in the map, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement positions.

[0151] In one possible implementation, the step of describing each line of the graphic as a starting point, in a clockwise or counterclockwise order, to obtain the text of the graphic, includes:

[0152] The coordinate information of the starting point is determined by taking a specific vertex of the graphic as the starting point.

[0153] Describe the line direction and length of each line in the figure in a clockwise or counterclockwise order;

[0154] Based on the coordinates of the starting point, as well as the direction and length of each line, the text of the graphic is obtained.

[0155] In one possible implementation, the text sequence further includes a map identifier representing the map and a graphic identifier representing the graphic type; different graphic types have different graphic identifiers.

[0156] The prefix of the map identifier is located before the text corresponding to the first image in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last image in the text sequence.

[0157] The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

[0158] In one possible implementation, the arrangement position includes arranging each graphic in a clockwise or counterclockwise direction, starting from a preset point on the map.

[0159] Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset vertex are arranged in ascending order.

[0160] In one possible implementation, the training process of the layout language model includes:

[0161] Obtain any sample text sequence from the sample set; after randomly masking the sample text sequence, input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model;

[0162] Obtain any sample text sequence from the sample set and the corresponding label information of the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain a map language model; the label information is used to indicate the presence of hotspots in the map corresponding to the sample text sequence.

[0163] In one possible implementation, the layout language model includes a multi-head self-attention layer;

[0164] The multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

[0165] The aforementioned computer-readable storage medium can be any available medium or data storage device that can be accessed by the processor in an electronic device, including but not limited to magnetic storage such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), optical storage such as CDs, DVDs, BDs, HVDs, etc., and semiconductor storage such as ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs), etc.

[0166] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection.

[0167] Based on the same inventive concept, this application also provides a computer program product, which includes computer program code. When the computer program code is run on a computer, it causes the computer to execute any of the related-party transaction reporting methods discussed above. Since the principle by which the above-described computer program product solves the problem is similar to that of the related-party transaction reporting method, the implementation of the above-described computer program product can refer to the implementation of the method, and repeated details will not be described again.

[0168] In this embodiment, multiple graphics contained in the layout to be detected are converted into a text sequence. The text sequence contains multiple texts recorded in a predetermined order, where each text describes a graphic. The predetermined order represents the arrangement of multiple graphics in the layout, thus representing each graphic in the layout concisely and accurately in the form of a text sequence. The text sequence is then input into a pre-trained layout language model, which extracts features from each text in the text sequence, determines whether it contains features corresponding to hotspots, and obtains the detection result of whether the layout corresponding to the text sequence output by the layout language model contains hotspots. This further improves the accuracy of layout hotspot detection.

[0169] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0170] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0171] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0172] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0173] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for detecting hotspots on a map, characterized in that, The method includes: The multiple graphics contained in the layout to be detected are converted into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text is used to describe a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout; The text sequence is input into a pre-trained map language model, and the detection result of whether the map contains hotspots is obtained from the output of the map language model.

2. The method according to claim 1, characterized in that, The step of converting multiple graphics of the layout to be detected into a text sequence includes: For each graphic in the layout to be detected, a set vertex of the graphic is used as the starting point, and each line of the graphic is described in a clockwise or counterclockwise order to obtain the text of the graphic; Determine the arrangement positions of multiple graphics contained in the map, and construct a text sequence of multiple texts corresponding to the multiple graphics according to the arrangement positions.

3. The method according to claim 2, characterized in that, The step of using a designated vertex of the graphic as a starting point and describing each line of the graphic in a clockwise or counterclockwise order to obtain the text of the graphic includes: The coordinate information of the starting point is determined by taking a specific vertex of the graphic as the starting point. Describe the line direction and length of each line in the figure in a clockwise or counterclockwise order; Based on the coordinates of the starting point, as well as the direction and length of each line, the text of the graphic is obtained.

4. The method according to claim 3, characterized in that, The text sequence also includes a map identifier representing the map and a graphic identifier representing the graphic type; different graphic types have different graphic identifiers. The prefix of the map identifier is located before the text corresponding to the first graphic in the text sequence, and the suffix of the map identifier is located after the text corresponding to the last graphic in the text sequence. The prefix of the graphic identifier is located before the coordinate information of the starting point of the corresponding graphic, and the suffix of the graphic identifier is located after the last line of the corresponding graphic.

5. The method according to claim 2, characterized in that, The arrangement position includes the order in which each graphic is arranged in a clockwise or counterclockwise direction, starting from a preset point on the map. Alternatively, starting from a preset point on the map, the distances from the corresponding vertices of each graphic to the preset point can be arranged in ascending order.

6. The method according to any one of claims 1 to 5, characterized in that, The training process of the map language model includes: Obtain any sample text sequence from the sample set; after randomly masking the sample text sequence, input it into the original language model to obtain the reconstructed text sequence output by the original language model; the reconstructed text sequence contains the predicted values ​​corresponding to the randomly masked portion of the sample text sequence; determine the loss value between the reconstructed text sequence and the sample text sequence before random masking; adjust the parameters in the original language model according to the loss value to obtain the adjusted language model; Obtain any sample text sequence from the sample set and the corresponding label information of the sample text sequence; input the sample text sequence and the label information into the adjusted language model; obtain the detection result of whether the map corresponding to the sample text sequence output by the adjusted language model contains hotspots; fine-tune the adjusted language model based on the label information and the detection result to obtain a map language model; the label information is used to indicate the presence of hotspots in the map corresponding to the sample text sequence.

7. The method according to any one of claims 1 to 5, characterized in that, The map language model includes a multi-head self-attention layer; The multi-head self-attention layer processes the input text sequence through linearization and scaling of dot product attention.

8. A map hotspot detection device, characterized in that, The device includes: A conversion module is used to convert multiple graphics contained in a layout to be detected into a text sequence; the text sequence contains multiple texts recorded in a predetermined order, wherein each text describes a graphic; the predetermined order represents the arrangement position of the multiple graphics in the layout; The detection module is used to input the text sequence into a pre-trained map language model and obtain the detection result of whether the map output by the map language model contains hotspots.

9. An electronic device, characterized in that, include: Memory, used to store program instructions; A processor is configured to invoke program instructions stored in the memory and execute the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1-7.

Images