A method, device and medium for determining hyperparameters of layout feature extraction

Abstract

The present invention discloses a method, device and medium for determining hyperparameters for layout feature extraction, the method comprising: sampling N sample points on a layout according to a sampling window of an initial size, and obtaining N layout slices correspondingly, and the N sample points correspond to N groups of attribute parameters; extracting features from the N layout slices according to the physical size of a single pixel, so as to convert them into N feature matrices; based on the N groups of attribute parameters, calculating the mutual information value corresponding to each single pixel position of the N feature matrices, and obtaining a mutual information map; according to the mutual information map, determining the target size of the sampling window as a hyperparameter for layout feature extraction. The method, device and medium provided by the present invention are used to solve the technical problem in the prior art that when extracting features from a layout for model training or model application, the operation speed and calculation accuracy are both poor due to inappropriately selected hyperparameters. The operation speed and calculation accuracy are effectively improved.

Description

Technical Field

[0001] The present disclosure relates to the semiconductor field, and in particular to a method, device and medium for determining hyperparameters for layout feature extraction. Background Art

[0002] With the development of integrated circuit technology, the iteration of process technology nodes is getting faster and faster. In the research and development of new process nodes, a large amount of data is usually generated for process model establishment, guiding the physical design in the research and development of integrated circuit products and the optimization of masks in integrated circuit manufacturing. On the other hand, the development of machine learning in recent years has made its algorithms more capable of processing and analyzing large-scale data and modeling. Many studies combine the two technologies and use machine learning algorithms to process the data generated in the research and development of integrated circuits, so as to more efficiently generate accurate and reliable process models, such as etching models, photoresist models, bad pixel detection models, optical proximity effect correction models, etc. developed based on machine learning algorithms.

[0003] In various model algorithms, for a sample point in the design layout, in order to extract the feature information related to the graphic structure in its surrounding environment, it is necessary to set a sampling window around each sample point, extract the layout graphic slices within a certain range, pixelate them, and convert them into a feature matrix (such as a density matrix). Each value in the matrix represents the graphic feature value (such as the graphic density value) in the corresponding single pixel. During the conversion process, the physical size of the single pixel on the layout represented by each single pixel and the size of the sampling window are two hyperparameters that need to be determined in advance.

[0004] In traditional methods, the physical size of a single pixel is usually selected as the unit distance (database unit) of the design layout or the minimum feature size (feature size) of the design graphics. The size of the sampling window needs to be selected according to the relevant process physical parameters. For example, in models related to photolithography, the optical diameter (OD) is usually selected as the side length of the sampling window when extracting features.

[0005] However, in machine learning-based modeling, the training time and computing speed of the model are positively correlated with the size of the sample feature vector. Too small a single pixel physical size and too large a sampling window size will make the extracted feature matrix too large, which will lead to a large feature vector, resulting in redundant feature information, increasing the model training time and decreasing the computing speed when the model is applied, while the accuracy of the model calculation results has not improved. Taking the 28nm process node as an example, the unit distance of the design layout is 0.001μm and the optical diameter is 2.6μm. If the single pixel physical size and the sampling window size are selected as these values, the feature matrix generated by each sampling point will contain millions of data, and such large-scale data calculation will take a lot of time. In addition, the proximity effect of many processes is very complex, and reasonable parameters cannot be obtained through physical mechanism analysis. Blindly selecting hyperparameters in feature extraction may lead to information redundancy or loss.

[0006] It can be seen that when the current feature extraction of the layout is used for model training or model application, there is a technical problem that the operation speed and calculation accuracy are poor due to the inappropriate selection of hyperparameters. Summary of the invention

[0007] The purpose of the present disclosure is at least partially to solve the technical problem in the prior art that when extracting features from a layout for model training or model application, inappropriate hyperparameters are selected, resulting in poor operation speed and calculation accuracy.

[0008] In a first aspect, the present disclosure provides a method for determining hyperparameters for layout feature extraction, comprising:

[0009] According to the sampling window of the initial size, N sample points on the layout are sampled respectively, and N layout slices are correspondingly obtained, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters;

[0010] Performing feature extraction on the N layout slices respectively according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices accordingly;

[0011] Based on the N groups of attribute parameters, the mutual information value corresponding to each single pixel position of the N feature matrices is calculated to obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value;

[0012] According to the mutual information graph, a target size of the sampling window is determined as a hyperparameter for layout feature extraction.

[0013] Optionally, the method for determining the single-pixel physical size is: obtaining the length values ​​of all sides of the polygonal layout figure on the layout to form a side length information group; performing statistical analysis on the side length information group, and determining the single-pixel physical size based on the statistical analysis results.

[0014] Optionally, determining the single-pixel physical size according to the statistical analysis result includes: determining a target length value with the highest distribution frequency according to the statistical analysis result, and determining the single-pixel physical size according to the target length value.

[0015] Optionally, performing feature extraction on the N map slices respectively according to the single-pixel physical size to convert the N map slices into N feature matrices respectively, including: performing feature extraction based on the density map on the N map slices respectively according to the single-pixel physical size to convert the N map slices into N density matrices respectively.

[0016] Optionally, the step of calculating the mutual information value corresponding to each single pixel position of the N feature matrices based on the N groups of attribute parameters to obtain a mutual information map includes: using the formula , calculate the mutual information value corresponding to the single pixel position at the (m, n) row and column coordinates of the N feature matrices , wherein x is the eigenvalue of each feature matrix in the N feature matrices at the single pixel position of the (m, n) row and column coordinates, and X is the set of all x; y is the attribute parameter corresponding to the feature matrix, and Y is the set of all y; p(x, y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y respectively; the single pixel positions on the N feature matrices are traversed, and the mutual information values ​​calculated according to the corresponding coordinate arrangement of the single pixel positions are obtained to obtain a mutual information graph.

[0017] Optionally, determining the target size of the sampling window as a hyperparameter for layout feature extraction based on the mutual information map includes: drawing a correlation curve based on the mutual information value on the mutual information map and the distance between the mutual information value and the center of the mutual information map; and determining the target size of the sampling window as a hyperparameter for layout feature extraction based on the inflection point position of the correlation curve.

[0018] Optionally, drawing a correlation curve based on the mutual information values ​​on the mutual information graph and the distances between the mutual information values ​​and the center of the mutual information graph includes: calculating the average value of each circle of mutual information values ​​centered on the center of the mutual information graph; and drawing a correlation curve based on the average value and the distances from each circle of mutual information values ​​to the center.

[0019] Optionally, drawing a correlation curve based on the mutual information values ​​on the mutual information graph and the distances between the mutual information values ​​and the center of the mutual information graph includes: calculating the row average of the mutual information values ​​of each row of the mutual information graph and the row distance of each row of the mutual information values ​​from the center of the mutual information graph; and drawing a row correlation curve based on the row average and the row distance; calculating the column average of the mutual information values ​​of each column of the mutual information graph and the column distance of each column of the mutual information values ​​from the center of the mutual information graph; and drawing a column correlation curve based on the column average and the column distance; determining the target size of the sampling window as a hyperparameter for layout feature extraction based on the inflection point position of the correlation curve includes: determining the lateral size of the sampling window as a hyperparameter for layout feature extraction based on the inflection point position of the row correlation curve; and determining the longitudinal size of the sampling window as a hyperparameter for layout feature extraction based on the inflection point position of the column correlation curve.

[0020] In a second aspect, the present disclosure provides a hyperparameter determination device for layout feature extraction, comprising:

[0021] A sampling module, used to sample N sample points on the layout according to a sampling window of an initial size, and obtain N layout slices, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters;

[0022] A matrix conversion module, used for performing feature extraction on the N layout slices according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices;

[0023] A mutual information graph module, used to calculate the mutual information value corresponding to each single pixel position of the N feature matrices based on the N groups of attribute parameters, and obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value;

[0024] A determination module is used to determine the target size of the sampling window as a hyperparameter for layout feature extraction based on the mutual information graph.

[0025] In a third aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods provided in the first aspect.

[0026] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0027] The method, device and medium for determining the hyperparameters of layout feature extraction provided in the embodiments of the present application first preset the initial size of the sampling window, and perform sampling, feature extraction and feature matrix conversion according to the initial size. Then, a mutual information map is calculated based on the attribute parameters of the feature matrix and the sample points. The mutual information value on the mutual information map represents the correlation between the graphic features and the attribute parameters at the single pixel position corresponding to the mutual information value. Therefore, the target size determined by analyzing the mutual information map is a size that takes into account the correlation between the image features and the attribute parameters, which is more in line with the actual situation of the layout. The target size is used as a hyperparameter for layout feature extraction. When used for layout feature extraction for model training or model application, it can better take into account the calculation speed and calculation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without creative work.

[0029] Figure 1 A flowchart of a method for determining hyperparameters for layout feature extraction according to one or more embodiments of the present disclosure;

[0030] Figure 2 A schematic diagram of a layout according to one or more embodiments of the present disclosure;

[0031] Figure 3 A schematic diagram of a side length information group according to one or more embodiments of the present disclosure;

[0032] Figure 4 A schematic diagram of the statistical distribution of edge length information groups according to one or more embodiments of the present disclosure;

[0033] Figure 5 A schematic diagram of sampling and extracting layout slices according to one or more embodiments of the present disclosure;

[0034] Figure 6 A schematic diagram of converting a layout slice into a feature matrix according to one or more embodiments of the present disclosure;

[0035] Figure 7 A schematic diagram of calculating a mutual information graph according to one or more embodiments of the present disclosure;

[0036] Figure 8 Schematic diagram of drawing a correlation curve according to one or more embodiments of the present disclosure Figure 1 ;

[0037] Fig. 9Schematic diagram of drawing a correlation curve according to one or more embodiments of the present disclosure Figure 2 ;

[0038] Fig.10 A schematic diagram of a hyperparameter determination device for layout feature extraction according to one or more embodiments of the present disclosure;

[0039] Fig.11 A schematic diagram of a storage medium according to one or more embodiments of the present disclosure. DETAILED DESCRIPTION

[0040] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0041] Various structural schematic diagrams according to embodiments of the present disclosure are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may further design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0042] In the context of the present disclosure, when a layer / element is referred to as being "on" another layer / element, the layer / element may be directly on the other layer / element, or there may be an intervening layer / element therebetween. Additionally, if a layer / element is "on" another layer / element in one orientation, the layer / element may be "below" the other layer / element when the orientation is reversed. In the context of the present disclosure, similar or identical components may be denoted by the same or similar reference numerals.

[0043] In order to better understand the above-mentioned technical scheme, the above-mentioned technical scheme will be described in detail below in combination with specific implementation methods. It should be understood that the embodiments of the present disclosure and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0044] According to one aspect of the present disclosure, a method for determining hyperparameters for layout feature extraction is provided, such as Figure 1 As shown, including:

[0045] Step S101, sampling N sample points on the layout according to the sampling window of the initial size, and correspondingly obtaining N layout slices, N is greater than 1, wherein the N sample points correspond to N groups of attribute parameters;

[0046] Step S102, performing feature extraction on the N layout slices respectively according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices accordingly;

[0047] Step S103, based on the N groups of attribute parameters, calculating the mutual information value corresponding to each single pixel position of the N feature matrices to obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value;

[0048] Step S104: determining a target size of the sampling window as a hyperparameter for layout feature extraction according to the mutual information graph.

[0049] The key hyperparameters for feature extraction of the layout are the size of the sampling window and the physical size of a single pixel. Among them, the size of the sampling window is the size of each slice sample collected when sampling the layout. If it is the model training stage, there are sample points preset on the sample layout, then the sampling range is the area around the sample point within the sampling window size; if it is the model application stage, the size of each sample image sampled for the layout to be detected is the same as the sampling window size. The physical size of a single pixel is the size of a pixel on the set layout. When the sampled layout slices are sampled with a sampling window and need to be pixelated and converted into a feature matrix (such as a density matrix), the graphic features within each single pixel physical size on the layout slice sample correspond to each value in the feature matrix.

[0050] If you need to determine the hyperparameters of a certain type of layout (same type of feature size, same type of process or same type of device) during feature extraction, you need to provide similar layouts in advance as training samples for model training and as calculation samples for determining the hyperparameters.

[0051] The following describes the method of determining two hyperparameters based on the provided layout as a sample:

[0052] First, let’s introduce how to determine the physical size of a single pixel.

[0053] According to the design rules, such as Figure 2As shown in the figure, each layer of graphics on the layout is often composed of many Manhattan polygon layout graphics, which are composed of two mutually perpendicular sides in the horizontal and vertical directions. First, the length values ​​of all the sides of the polygon layout graphics on the layout are obtained to form a side length information group. Then, a statistical analysis is performed on the side length information group, and the physical size of a single pixel is determined based on the statistical analysis results.

[0054] For example, Figure 2 The side length of the polygonal layout is measured and recorded on the layout shown in the figure, and the following is obtained: Figure 3 The side length information group shown. Figure 2 In the figure, p1~p7 are the numbers of each polygonal pattern, and E1_1~E7_8 are the numbers of the edges of each polygonal pattern. Then, Figure 4 As shown, a statistical analysis is performed on the side length information group, and the physical size of a single pixel is determined based on the statistical analysis results.

[0055] In the specific implementation process, there are many methods for determining the physical size of a single pixel based on the statistical analysis results. Two methods are listed below as examples:

[0056] The first method is to first determine the target length value with the highest distribution frequency, and then determine the single pixel physical size based on the target length value.

[0057] Specifically, the side length information group is first analyzed to find the length value that appears most frequently, and this is used as the side length of the single-pixel physical size, thereby avoiding both the loss of layout features caused by the single-pixel physical size being too long and the high amount of calculation caused by the single-pixel physical size being too small.

[0058] For example, suppose that a statistical analysis of the edge length information group generates Figure 4 As shown in the length value distribution diagram, it can be seen that the edge with a length of 45nm appears most frequently, so the physical size of a single pixel is set to 45nm×45nm in the layout graphic feature extraction.

[0059] The second method is to first determine a group of length values ​​whose distribution frequency is higher than a preset frequency, then use the minimum length value as the target length value, and then determine the single pixel physical size according to the target length value.

[0060] Specifically, the side length information group is first analyzed to find out one or more length values ​​whose occurrence times are different from the preset times, and then the length value with the smallest value is selected as the side length of the single-pixel physical size to avoid losing layout features as much as possible.

[0061] Of course, the method of determining the physical size of a single pixel based on the statistical analysis results is not limited to the above two. Since this embodiment combines the numerical distribution of the length of the polygon sides on the layout to specifically set the physical size of a single pixel, it can avoid the loss of layout features caused by the single pixel physical size being too long to a large extent, and can also avoid the high amount of calculation caused by the single pixel physical size being too small.

[0062] Then it introduces how to determine the size of the sampling window.

[0063] N sample points and their corresponding attribute parameters are marked on the layout used as a sample. For example, in layout bad pixel detection based on machine learning, a layout used as a sample for model training and hyperparameter determination is provided, and the sample points marked on it are the coordinates of each bad pixel and non-bad pixel, and the corresponding attribute parameters are parameters that characterize whether each sampling point is a bad pixel (for example, the attribute parameter of a bad pixel is 1, and the attribute parameter of a non-bad pixel is 0). For another example, in an etching model based on machine learning, a layout used as a sample for model training and hyperparameter determination needs to be provided, and the sample points marked on it are the coordinates of the sample points preset according to the etching characteristics, and the corresponding attribute parameters are parameters that characterize the etching deviation of the sample points.

[0064] After the above layout is provided, step S101 is first executed to sample N sample points on the layout according to a sampling window of an initial size, extract the layout graphics of each sample point within the sampling window, and obtain N layout slices correspondingly, where N is greater than 1.

[0065] It should be noted that the initial size can be preset based on an empirical value, and in order to facilitate the subsequent determination of the target size, the initial size can be set to a relatively large value. For example, if the model to be modeled or trained is a model related to the photolithography process, the initial size can be set to the optical diameter.

[0066] After the initial size is preset, Figure 5 As shown, each hollow small circle sampling point in the layout is sampled according to the sampling window shown in the dotted box, and the layout graphics around it are obtained as layout slices, and N sample points obtain corresponding N layout slices.

[0067] Then, step S102 is executed to extract features from the N layout slices according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices.

[0068] In the specific implementation process, the single-pixel physical size is used as the basic unit to extract features from the area on each layout slice to form a feature matrix. The number of single-pixel physical size areas that can be accommodated on each layout slice corresponds to the number of single-pixel positions, and the size of the area occupied by each single-pixel position is exactly equal to the single-pixel physical size. The feature value of the layout figure extracted at any single-pixel position is used as the value on the row and column corresponding to the single-pixel position in the feature matrix.

[0069] For example, a layout slice can be divided into M*M single-pixel regions, where the feature value of the layout graphic at the single-pixel position in the nth row and mth column is used as the value of the nth row and mth column of the feature matrix corresponding to the layout slice. The layout slice is extracted and converted into a feature matrix of size M*M, where n and m are both less than or equal to M.

[0070] Depending on the specific features to be extracted, the corresponding numerical meanings on the feature matrix are also different. If the feature to be extracted is density information, such as Figure 6 As shown, the N layout slices are subjected to density map-based feature extraction according to the physical size of a single pixel, so as to convert the corresponding pixels of the N layout slices into N density matrices, where each value in the density matrix represents the layout pattern density of the area occupied by the corresponding single pixel position. The density matrix is ​​used as the feature matrix.

[0071] If the feature to be extracted is spectral information, the density matrix needs to be further converted into a spectral matrix, and the spectral matrix is ​​used as the feature matrix.

[0072] Next, execute step S103, calculate the mutual information value corresponding to each single pixel position of the N feature matrices based on N groups of attribute parameters, and obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value.

[0073] Specifically, we can first use the formula: , calculate the mutual information value corresponding to the single pixel position at the (m,n) row and column coordinates of the N feature matrices , where x is the eigenvalue of each feature matrix in the N feature matrices at a single pixel position of (m, n) row and column coordinates, X is the set of all x; y is the attribute parameter corresponding to the feature matrix, Y is the set of all y; p(x,y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y, respectively.

[0074] For example, N sample points correspond to N feature matrices of size M*M, such as Figure 7As shown, there are N*M*M eigenvalues ​​in N feature matrices. Each feature matrix has N eigenvalues ​​x with the same coordinates (i.e., the same single pixel position), and these N eigenvalues ​​x correspond to N attribute parameters y one by one. Substitute the N eigenvalues ​​x and N attribute parameters y into the above formula for calculation, and add up the corresponding N eigenvalues ​​x and N attribute parameters y. , the sum of the values As the mutual information value corresponding to the same coordinate (m, n).

[0075] Then, the corresponding mutual information values ​​are calculated for the M*M single-pixel positions on the feature matrix according to the above calculation method, and the calculated M*M mutual information values ​​are arranged according to the coordinates of the corresponding single-pixel positions to obtain a mutual information map of size M*M. That is, a mutual information map is extracted from N sample points. In the mutual information map, the value of the mutual information value on the coordinate corresponding to each single-pixel position represents: the correlation between the layout graphics at the position corresponding to the coordinate around the sample point and the sample attribute parameters. The larger the value of the mutual information value, the more important the position is, and it deserves more attention in feature extraction. Each mutual information value occupies an area of ​​the physical size of a single pixel on the mutual information map.

[0076] After the mutual information map is obtained, step S104 is executed to determine the target size of the sampling window as a hyperparameter for layout feature extraction according to the mutual information map.

[0077] In the specific implementation process, there are many methods for determining the target size of the sampling window based on the mutual information graph. The following are two examples:

[0078] The first one is to plot the mutual information graph into a curve and determine the target size based on the curve.

[0079] Specifically, a correlation curve can be drawn based on the mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph. Then, the target size of the sampling window is determined as a hyperparameter for layout feature extraction based on the inflection point position of the correlation curve. Because the inflection point position represents the position where the correlation between the layout graphics and the sample attribute parameters changes from strong to weak due to the influence of distance, the area outside the inflection point can be considered to have relatively weak correlation and there is no need to include it in the sampling window range. Therefore, the target size of the sampling window determined based on the inflection point position is more targeted and more suitable for this type of layout, taking into account the comprehensiveness of feature extraction and calculation speed of this type of layout.

[0080] Preferably, if the model does not need to consider the anisotropy of the proximity effect of this type of layout, then Figure 8As shown, the average value of each circle of mutual information values ​​(the squares shown with diagonal lines in the figure are a circle of mutual information values) centered on the center of the mutual information graph (black and white cross squares) is calculated; based on the average value and the distance from each circle of mutual information values ​​to the center, a correlation curve is drawn.

[0081] It should be noted that each circle of mutual information values ​​refers to the mutual information values ​​with the minimum number of equal mutual information values ​​between the mutual information values ​​at the center of the mutual information graph (black and white cross squares), for example Figure 8 The mutual information value of the square indicated by the slash is at least 1 mutual information value away from the mutual information value at the center, so the mutual information value of the square indicated by the slash is a circle. Similarly, the mutual information value of the outermost area is at least 2 mutual information values ​​away from the mutual information value at the center, so the mutual information value of the outermost area is also a circle.

[0082] It should be noted that the distance from each circle of mutual information values ​​to the center is the distance from the center of the outermost margin mutual information map of each circle of mutual information values; or, it is the average of the distance from the center of the outermost margin mutual information map of each circle of mutual information values ​​to the center of the margin mutual information map of the inner circle.

[0083] The distance from the calculated mutual information value of each circle to the center is used as the horizontal axis, and the average value of the corresponding mutual information value of the circle is used as the vertical axis to draw the Figure 8 The correlation curve in .

[0084] It can be seen from the correlation curve that as the distance from the center increases, the mutual information value becomes smaller and smaller, that is, the correlation becomes lower and lower. The layout graphic information at a larger distance cannot provide sufficient help for the prediction of the sample attribute parameters. Therefore, the distance value on the horizontal axis corresponding to the inflection point of the correlation curve can be multiplied by 2 as the square side length (target size) of the sampling window, thereby taking into account both integrity and calculation speed.

[0085] Preferably, if the model needs to consider the anisotropy of the proximity effect of this type of layout, then Fig. 9 As shown in (a), the row average of each row of mutual information values ​​(black and white cross squares) in the mutual information graph and the row distance of each row of mutual information values ​​from the center of the mutual information graph are calculated; and the row correlation curve is drawn based on the row average and row distance. And, as Fig. 9 As shown in (b), the column average of each column of mutual information values ​​(black and white cross squares) in the mutual information graph and the column distance of each column of mutual information values ​​from the center of the mutual information graph are calculated; and the column correlation curve is drawn based on the column average and column distance.

[0086] It should be noted that the distance from each row of mutual information values ​​to the center is the distance between the edge of each row of mutual information values ​​farthest from the center and the center of the mutual information graph; or, it is the average of the distance between the edge of each circle of mutual information values ​​farthest from the center and the center of the mutual information graph and the distance between the edge closest to the center and the center of the mutual information graph. Plot the distance from each row of mutual information values ​​to the center as the horizontal axis and the row average value of the corresponding row of mutual information values ​​as the vertical axis. Fig. 9 The method of drawing the column correlation curve is the same as that of drawing the row correlation curve, which will not be repeated here.

[0087] Then, the longitudinal size of the sampling window is determined as a hyperparameter for layout feature extraction according to the inflection point position of the row correlation curve; and the lateral size of the sampling window is determined as a hyperparameter for layout feature extraction according to the inflection point position of the column correlation curve.

[0088] Specifically, the distance value on the horizontal axis corresponding to the inflection point of the row correlation curve is multiplied by 2 as the longitudinal size of the sampling window (the size along the column direction), and the distance value on the horizontal axis corresponding to the inflection point of the column correlation curve is multiplied by 2 as the transverse size of the sampling window (the size along the row direction), so as to take into account both integrity and calculation speed.

[0089] The second method is to use machine learning algorithms.

[0090] The mutual information graph is input into a machine learning model trained with a large number of mutual information graph samples and sample sizes, and the target size is output according to the machine learning model.

[0091] Of course, the method of confirming the target size of the sampling window according to the mutual information graph is not limited to the above two methods and is not limited here.

[0092] Specifically, the two hyperparameters of the target size of the sampling window and the physical size of a single pixel determined by the method provided in this embodiment are obtained by calculating and analyzing the sample layout, and no additional analysis of the physical mechanism of modeling is required, which simplifies the modeling process. In addition, by selecting an appropriate sampling window target size that takes into account the correlation, it is ensured that sufficient graphic information is collected around the sample point coordinates to express the influence of the proximity effect. By selecting an appropriate single-pixel physical size that takes into account the size of the layout graphic, it is ensured that neither information redundancy nor information loss will occur during the pixelization process. Therefore, without affecting the accuracy of the training model, the feature data is simplified to the maximum extent, and the model training and implementation speed is accelerated. Furthermore, when drawing the correlation curve, the variation characteristics of the mutual information values ​​in different directions with the distance are considered, which can take into account the influence of the anisotropic proximity effect in the process on the feature extraction, so that the extracted layout graphic features are more relevant to the model prediction and more accurate.

[0093] Based on the same inventive concept, the present disclosure provides a hyperparameter determination device for layout feature extraction, such as Fig.10 As shown, including:

[0094] The sampling module 1001 is used to sample N sample points on the layout according to the sampling window of the initial size, and obtain N layout slices, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters;

[0095] The matrix conversion module 1002 is used to extract features from the N layout slices according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices;

[0096] The mutual information map module 1003 is used to calculate the mutual information value corresponding to each single pixel position of the N feature matrices based on the N groups of attribute parameters, and obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value;

[0097] The determination module 1004 is used to determine the target size of the sampling window as a hyperparameter for layout feature extraction according to the mutual information graph.

[0098] It should be noted that the above-mentioned device for determining the hyperparameters of layout feature extraction is the device corresponding to the method for determining the hyperparameters of layout feature extraction provided in the aforementioned embodiment. The technical features introduced in the description of the method for determining the hyperparameters of layout feature extraction are all applicable to the device and will not be described again here.

[0099] Based on the same inventive concept, the present disclosure provides a computer-readable storage medium 1100, such as Fig.11 As shown, a computer program 1111 is stored thereon, and when the computer program 1111 is executed by a processor, the following steps are implemented:

[0100] According to the sampling window of the initial size, N sample points on the layout are sampled respectively, and N layout slices are correspondingly obtained, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters;

[0101] Performing feature extraction on the N layout slices respectively according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices accordingly;

[0102] Based on the N groups of attribute parameters, the mutual information value corresponding to each single pixel position of the N feature matrices is calculated to obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value;

[0103] According to the mutual information graph, a target size of the sampling window is determined as a hyperparameter for layout feature extraction.

[0104] In a specific implementation process, when the computer program 1111 is executed by a processor, any implementation method of the method of the embodiment of the present invention can be implemented.

[0105] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0106] The method, device and medium for determining the hyperparameters of layout feature extraction provided in the embodiments of the present application first preset the initial size of the sampling window, and perform sampling, feature extraction and feature matrix conversion according to the initial size. Then, a mutual information map is calculated based on the attribute parameters of the feature matrix and the sample points. The mutual information value on the mutual information map represents the correlation between the graphic features and the attribute parameters at the single pixel position corresponding to the mutual information value. Therefore, the target size determined by analyzing the mutual information map is a size that takes into account the correlation between the image features and the attribute parameters, which is more in line with the actual situation of the layout. The target size is used as a hyperparameter for layout feature extraction. When used for layout feature extraction for model training or model application, it can better take into account the calculation speed and calculation accuracy.

[0107] In the above description, the technical details of the patterning and etching of each layer are not described in detail. However, those skilled in the art should understand that various technical means can be used to form layers, regions, etc. of desired shapes. In addition, in order to form the same structure, those skilled in the art can also design methods that are not completely the same as the methods described above. In addition, although the various embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage.

[0108] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is also intended to include these modifications and variations.

Claims

1. A method for determining hyperparameters for layout feature extraction, characterized in that: include: According to the sampling window of the initial size, N sample points on the layout are sampled respectively, and N layout slices are correspondingly obtained, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters; Performing feature extraction on the N layout slices respectively according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices accordingly; Based on the N groups of attribute parameters, the mutual information value corresponding to each single pixel position of the N feature matrices is calculated to obtain a mutual information map, wherein the mutual information value on the mutual information map represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value; According to the mutual information graph, determining a target size of the sampling window as a hyperparameter for layout feature extraction; The step of calculating the mutual information value corresponding to each single pixel position of the N feature matrices based on the N groups of attribute parameters to obtain a mutual information map includes: Using formula , calculate the mutual information value corresponding to the single pixel position at the (m,n) coordinate of the N feature matrices , where x is the eigenvalue of each feature matrix in the N feature matrices at a single pixel position at the (m, n) coordinate, X is the set of all x; y is the attribute parameter corresponding to the feature matrix, Y is the set of all y; p(x, y) is the joint probability distribution function of x and y, and p(x) and p(y) are the marginal probability distribution functions of x and y respectively; Traversing the single pixel positions on the N feature matrices, and arranging the mutual information values ​​calculated according to the coordinates of the corresponding single pixel positions to obtain a mutual information graph; Determining the target size of the sampling window as a hyperparameter for layout feature extraction according to the mutual information graph includes: Draw a correlation curve according to the mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph; Determining the target size of the sampling window as a hyperparameter for layout feature extraction according to the inflection point position of the correlation curve; The step of drawing a correlation curve according to the mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph comprises: Calculate the average value of each circle of mutual information values ​​centered on the center of the mutual information graph; Draw a correlation curve according to the average value and the distance from each circle of mutual information values ​​to the center; The step of drawing a correlation curve according to the mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph comprises: Calculate the row average of the mutual information values ​​of each row of the mutual information graph and the row distance of each row of the mutual information value from the center of the mutual information graph; and draw a row correlation curve based on the row average and the row distance; calculate the column average of the mutual information values ​​of each column of the mutual information graph and the column distance of each column of the mutual information value from the center of the mutual information graph; and draw a column correlation curve based on the column average and the column distance; Determining the target size of the sampling window as a hyperparameter for layout feature extraction according to the inflection point position of the correlation curve includes: According to the inflection point position of the row correlation curve, the lateral size of the sampling window is determined as a hyperparameter for layout feature extraction; according to the inflection point position of the column correlation curve, the longitudinal size of the sampling window is determined as a hyperparameter for layout feature extraction.

2. The method according to claim 1, characterized in that The method for determining the physical size of a single pixel is: Obtaining the length values ​​of all sides of the polygonal layout graphics on the layout to form a side length information group; Perform statistical analysis on the side length information group, and determine the single pixel physical size according to the statistical analysis result.

3. The method according to claim 2, characterized in that Determining the single pixel physical size according to the statistical analysis result includes: According to the statistical analysis result, the target length value with the highest distribution frequency is determined, and the single-pixel physical size is determined according to the target length value.

4. The method according to claim 1, characterized in that The extracting features of the N layout slices respectively according to the physical size of a single pixel to convert the N layout slices into N feature matrices respectively includes: Density map-based feature extraction is performed on the N layout slices respectively according to the physical size of a single pixel, so as to convert the N layout slices into N density matrices accordingly.

5. A device for determining hyperparameters for layout feature extraction, the device for determining hyperparameters being used to implement the method according to any one of claims 1 to 4, characterized in that: include: A sampling module, used to sample N sample points on the layout according to a sampling window of an initial size, and obtain N layout slices, where N is greater than 1, and the N sample points correspond to N groups of attribute parameters; A matrix conversion module, used for performing feature extraction on the N layout slices according to the physical size of a single pixel, so as to convert the N layout slices into N feature matrices; A mutual information graph module, used to calculate the mutual information value corresponding to each single pixel position of the N feature matrices based on the N groups of attribute parameters, and obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value; A determination module is used to determine the target size of the sampling window as a hyperparameter for layout feature extraction based on the mutual information graph.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method described in any one of claims 1 to 4 is implemented.

Images