CN111539177A - Method, device and medium for determining hyper-parameters of layout feature extraction - Google Patents
Method, device and medium for determining hyper-parameters of layout feature extraction Download PDFInfo
- Publication number
- CN111539177A CN111539177A CN202010324225.8A CN202010324225A CN111539177A CN 111539177 A CN111539177 A CN 111539177A CN 202010324225 A CN202010324225 A CN 202010324225A CN 111539177 A CN111539177 A CN 111539177A
- Authority
- CN
- China
- Prior art keywords
- mutual information
- layout
- single pixel
- graph
- determining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000000605 extraction Methods 0.000 title claims abstract description 44
- 238000005070 sampling Methods 0.000 claims abstract description 77
- 239000011159 matrix material Substances 0.000 claims description 33
- 238000007619 statistical method Methods 0.000 claims description 13
- 238000009826 distribution Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 abstract description 18
- 238000012549 training Methods 0.000 abstract description 14
- 230000000875 corresponding effect Effects 0.000 description 42
- 230000008569 process Effects 0.000 description 18
- 238000013461 design Methods 0.000 description 9
- 238000010801 machine learning Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 238000005530 etching Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000001459 lithography Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000000059 patterning Methods 0.000 description 1
- 229920002120 photoresistant polymer Polymers 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/39—Circuit design at the physical level
- G06F30/392—Floor-planning or layout, e.g. partitioning or placement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Mathematical Analysis (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Geometry (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Algebra (AREA)
- Architecture (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a device and a medium for determining hyper-parameters of layout feature extraction, wherein the method comprises the following steps: respectively sampling N sample points on the layout according to a sampling window with an initial size, and correspondingly obtaining N layout slices, wherein the N sample points correspond to N groups of attribute parameters; respectively extracting features of the N layout slices according to the physical size of a single pixel to convert the N layout slices into N feature matrixes; calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph; and determining the target size of the sampling window as the hyper-parameter of the layout feature extraction according to the mutual information graph. The method, the device and the medium provided by the invention are used for solving the technical problem of poor operation speed and calculation accuracy caused by the fact that the selected hyper-parameters are not appropriate when the feature extraction is carried out on the layout for model training or model application in the prior art. The operation speed and the calculation accuracy are effectively improved.
Description
Technical Field
The present disclosure relates to the field of semiconductors, and in particular, to a method, an apparatus, and a medium for determining a hyper-parameter for layout feature extraction.
Background
With the development of integrated circuit technology, the iteration of process technology nodes is faster and faster. In the development of new process nodes, a large amount of data is usually generated for process modeling, guiding physical design in the development of integrated circuit products and mask optimization in the manufacture of integrated circuits. On the other hand, the development of machine learning field in recent years makes the ability of its algorithm to analyze and model large-scale data stronger and stronger. Many studies combine both techniques and use machine learning algorithms to process data generated during integrated circuit development to more efficiently generate accurate and reliable process models, such as etch models, photoresist models, dead-spot detection models, optical proximity correction models, etc., developed based on machine learning algorithms.
In various model algorithms, for a sample point in a design layout, in order to extract feature information related to a pattern structure in its surrounding environment, a sampling window is required to be arranged around each sample point, a layout pattern slice within a certain range is extracted, pixelized and converted into a feature matrix (such as a density matrix), and each value in the matrix represents a pattern feature value (such as a pattern density value) in a corresponding single pixel. In 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 hyper-parameters which need to be determined in advance.
In conventional approaches, the single-pixel physical size is typically selected as the unit distance (databaseunit) of the design layout or the minimum feature size (feature size) of the design pattern. The size of the sampling window needs to be selected according to the relevant process physical parameters, for example, in a model related to the lithography process, the Optical Diameter (OD) is usually selected as the side length of the sampling window during feature extraction.
However, in modeling based on machine learning, the training time and the operation speed of the model are positively correlated with the size of the sample feature vector. The extracted feature matrix is too large due to the too small physical size of a single pixel and the too large size of a sampling window, so that large feature vectors are caused, redundancy of feature information is caused, the training time of the model is increased, the operation speed is reduced when the model is applied, and the accuracy of a model calculation result is not improved. Taking a 28nm process node as an example, the unit distance of the design layout is 0.001 μm, the optical diameter is 2.6 μm, and if the physical size of a single pixel and the size of a sampling window are selected as these values, the feature matrix generated by each sampling point will contain millions of data, so that such large-scale data calculation will consume a lot of time. In addition, the proximity effect of many processes is complex and does not allow reasonable parameters to be obtained by physical mechanistic analysis. Blindly selecting hyper-parameters in feature extraction may lead to information redundancy or loss.
Therefore, when the feature extraction is carried out on the layout for model training or model application at present, the technical problems of poor operation speed and poor calculation accuracy caused by the fact that the selected hyper-parameters are not appropriate exist.
Disclosure of Invention
The purpose of the disclosure is to solve at least part of the technical problems of poor operation speed and poor calculation accuracy caused by inappropriate selected hyper-parameters when the feature extraction is performed on the layout for model training or model application in the prior art.
In a first aspect, the present disclosure provides a method for determining a hyper-parameter of layout feature extraction, including:
respectively sampling N sample points on the layout according to a sampling window with an initial size, and correspondingly obtaining N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
respectively extracting features of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N feature matrixes;
calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic characteristics and the attribute parameters at the single pixel position corresponding to the mutual information value;
and determining the target size of the sampling window as a hyper-parameter for extracting the layout features according to the mutual information graph.
Optionally, the method for determining the physical size of the single pixel includes: obtaining length values of all edges of a polygon layout graph on the layout to form an edge length information group; and performing statistical analysis on the side length information group, and determining the physical size of the single pixel according to the statistical analysis result.
Optionally, the determining the physical size of the single pixel according to the statistical analysis result includes: and 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.
Optionally, the performing feature extraction on the N layout slices according to the physical size of a single pixel respectively to correspondingly convert the N layout slices into N feature matrices includes: and respectively carrying out density map-based feature extraction on the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N density matrixes.
Optionally, the calculating, based on the N groups of attribute parameters, a mutual information value corresponding to each single pixel position of the N feature matrices to obtain a mutual information graph includes: using a formula
Calculating a mutual information value MI corresponding to a single pixel position of the N characteristic matrixes at the (m, N) row-column coordinatesm,n(X, Y), where X is a feature value of each feature matrix in the N feature matrices at a single pixel position of (m, N) row-column coordinates, and X is a 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 a joint probability distribution function of x and y, and p (x) and p (y) are marginal probability distribution functions of x and y, respectively; and traversing the single pixel positions on the N characteristic matrixes, and arranging the calculated mutual information values according to the coordinates of the corresponding single pixel positions to obtain a mutual information graph.
Optionally, the determining, according to the mutual information graph, a target size of the sampling window as a hyper-parameter for layout feature extraction includes: drawing a correlation curve according to a mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph; and determining the target size of the sampling window as the hyperparameter extracted from the layout characteristics according to the inflection point position of the correlation curve.
Optionally, the 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 includes: calculating the average value of each circle of mutual information values by taking the center of the mutual information graph as the center; and drawing a correlation curve according to the average value and the distance from each circle of mutual information value to the center.
Optionally, the 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 includes: calculating the line average value of each line of mutual information values of the mutual information graph and the line distance between each line of mutual information values and the center of the mutual information graph; drawing a line correlation curve according to the line average value and the line distance; calculating a column average value of each column of mutual information values of the mutual information graph and a column distance between each column of mutual information values and the center of the mutual information graph; drawing a column correlation curve according to the column average value and the column distance; the step of determining the target size of the sampling window according to the inflection point position of the correlation curve as the hyper-parameter for extracting the layout features comprises the following steps: determining the transverse size of the sampling window as a hyper-parameter for extracting layout features according to the inflection point position of the line correlation curve; and determining the longitudinal size of the sampling window as a hyper-parameter for extracting layout features according to the inflection point position of the column correlation curve.
In a second aspect, the present disclosure provides a hyper-parameter determining apparatus for layout feature extraction, including:
the sampling module is used for respectively sampling N sample points on the layout according to a sampling window with an initial size to correspondingly obtain N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
the matrix conversion module is used for respectively extracting the characteristics of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N characteristic matrices;
the mutual information graph module is used for calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic characteristics and the attribute parameters at the single pixel position corresponding to the mutual information value;
and the determining module is used for determining the target size of the sampling window as the hyper-parameter for extracting the layout features according to the mutual information graph.
In a third aspect, the present disclosure provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements any of the methods provided by the first aspect.
One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:
according to the method, the device and the medium for determining the hyper-parameters of the layout feature extraction, the initial size of a sampling window is preset, and sampling, feature extraction and feature matrix conversion are carried out according to the initial size. And then calculating according to the feature matrix and the attribute parameters of the sample points to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graph features and the attribute parameters at the single pixel position corresponding to the mutual information value, so that the target size determined by analyzing the mutual information graph is the size considering the correlation between the image features and the attribute parameters and is more consistent with the actual situation of the layout, the target size is taken as the hyperparameter for extracting the layout features, and the calculation speed and the calculation accuracy can be better considered when the layout features are used for model training or model application.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is apparent that the drawings in the following description are only examples of the present disclosure, and it is obvious for those skilled in the art that other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a flow diagram of a method for determining hyper-parameters for layout feature extraction in accordance with one or more embodiments of the present disclosure;
FIG. 2 is a schematic illustration of a layout in accordance with one or more embodiments of the present disclosure;
FIG. 3 is a schematic diagram of a side length information set in accordance with one or more embodiments of the present disclosure;
FIG. 4 is a schematic illustration of a statistical distribution of sets of side length information according to one or more embodiments of the present disclosure;
FIG. 5 is a schematic diagram of sampling an extracted layout slice according to one or more embodiments of the present disclosure;
FIG. 6 is a schematic diagram of a layout slice being converted into a feature matrix according to one or more embodiments of the present disclosure;
FIG. 7 is a schematic diagram of a computed mutual information graph according to one or more embodiments of the present disclosure;
FIG. 8 is a first schematic diagram of plotting a correlation curve according to one or more embodiments of the present disclosure;
FIG. 9 is a second schematic diagram of plotting a correlation curve in accordance with one or more embodiments of the present disclosure;
FIG. 10 is a schematic diagram of a hyper-parameter determination apparatus for layout feature extraction according to one or more embodiments of the present disclosure;
FIG. 11 is a schematic illustration of a storage medium according to one or more embodiments of the present disclosure.
Detailed Description
Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that the description is illustrative only and is not intended to limit the scope of the present disclosure. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure.
Various structural schematics according to embodiments of the present disclosure are shown in the figures. The figures are not drawn to scale, wherein certain details are exaggerated and possibly omitted for clarity of presentation. The shapes of various regions, layers, and relative sizes and positional relationships therebetween shown in the drawings are merely exemplary, and deviations may occur in practice due to manufacturing tolerances or technical limitations, and a person skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions, as actually required.
In the context of the present disclosure, when a layer/element is referred to as being "on" another layer/element, it can be directly on the other layer/element or intervening layers/elements may be present. In addition, if a layer/element is "on" another layer/element in one orientation, then that layer/element may be "under" the other layer/element when the orientation is reversed. In the context of the present disclosure, similar or identical components may be referred to by the same or similar reference numerals.
In order to better understand the technical solutions, the technical solutions will be described in detail below with reference to specific embodiments, and it should be understood that the specific features in the examples and examples of the present disclosure are detailed descriptions of the technical solutions of the present application, but not limitations of the technical solutions of the present application, and the technical features in the examples and examples of the present application may be combined with each other without conflict.
According to an aspect of the present disclosure, there is provided a method for determining a hyper-parameter of layout feature extraction, as shown in fig. 1, including:
step S101, respectively sampling N sample points on a layout according to a sampling window with an initial size, and correspondingly obtaining N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
step S102, respectively extracting characteristics of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N characteristic matrixes;
step S103, calculating mutual information values corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information values on the mutual information graph represent the correlation between the graph characteristics and the attribute parameters at the single pixel positions corresponding to the mutual information values;
and step S104, determining the target size of the sampling window as a hyper-parameter for extracting layout features according to the mutual information graph.
The key hyper-parameters when the layout is subjected to feature extraction include the size of a sampling window and the physical size of a single pixel. And the size of the sampling window is the size of each collected slice sample when the layout is sampled. If the model training stage is adopted, the sample points are preset on the sample layout, and the sampling range is the area around the sample points within the size of the sampling window; and if the model is in the application stage, the size of each sample picture to be detected for the layout sampling is the same as the size of the sampling window. The physical size of a single pixel is the size of a pixel point on a set layout, and when a sampling window is used for sampling a slice sample of a published image, the slice sample of the published image needs to be pixilated and converted into a feature matrix (such as a density matrix), the graphic features in each physical size of the single pixel on the slice sample of the layout correspond to each value in the feature matrix.
If the hyper-parameter of a certain type (same type of feature size, same type of process or same type of device) of layout during feature extraction needs to be determined, the same type of layout needs to be provided in advance as a training sample for model training and used as a calculation sample for determining the hyper-parameter.
The following describes respectively the method of determining two hyper-parameters based on the provided layout as a sample:
first, how the physical size of a single pixel is determined will be described.
According to the Design Rules (Design Rules), as shown in fig. 2, each layer of patterns on the layout is often composed of many Manhattan polygon (Manhattan polygon) layout patterns, and the polygon layout patterns are composed of edges in two mutually perpendicular directions, namely, a transverse direction and a longitudinal direction. The length values of all edges of a polygon layout graph on the layout are obtained first, and a side length information group is formed. Then, the side length information group is subjected to statistical analysis, and the physical size of the single pixel is determined according to the result of the statistical analysis.
For example, the side length of the polygon layout graph is measured and recorded for the layout shown in fig. 2, and the side length information group shown in fig. 3 is obtained. In fig. 2, p1 to p7 indicate the numbers of the respective polygon layouts, and E1_1 to E7_8 indicate the numbers of the sides of the respective polygon layouts. Then, as shown in fig. 4, the group of edge length information is statistically analyzed, and the physical size of a single pixel is determined according to the result of the statistical analysis.
In the specific implementation process, there are various methods for determining the physical size of a single pixel according to the statistical analysis result, and two methods are listed as examples below:
firstly, determining a target length value with the highest distribution frequency, and then determining the single pixel physical size according to the target length value.
Specifically, the length information group is analyzed to find out the length value with the largest occurrence frequency, and the length value is used as the side length of the physical size of the single pixel, so that the layout characteristic loss caused by the overlong physical size of the single pixel is avoided, and the high calculation amount caused by the overlong physical size of the single pixel is avoided.
For example, assuming that a length information group is statistically analyzed to generate a length value distribution graph as shown in fig. 4, it can be seen from the graph that edges having a length of 45nm occur most frequently, and therefore, the physical size of a single pixel is set to 45nm × 45nm in the layout pattern feature extraction.
And secondly, determining a group of length values with the distribution frequency higher than the preset frequency, taking the minimum length value as a target length value, and determining the physical size of the single pixel according to the target length value.
Specifically, the length information group is analyzed to find out one or more length values of which the occurrence times are changed to preset times, and then the length value with the minimum value is selected as the side length of the physical size of a single pixel, so that the loss of the layout characteristics is avoided as much as possible.
Of course, the method for determining the physical size of a single pixel according to the statistical analysis result is not limited to the two methods, and the physical size of the single pixel is set in a targeted manner by combining the distribution of the length value of the side of the polygon on the layout, so that the layout characteristics that are lost due to the overlong physical size of the single pixel can be avoided to a greater extent, and the high calculation amount that is caused by the overlong physical size of the single pixel can be avoided.
Then, how the size of the sampling window is determined is described.
And marking N sample points and corresponding attribute parameters on the layout serving as the sample. For example, in layout dead pixel detection based on machine learning, model training and hyper-parameters are provided to determine a used layout as a sample, where the above-labeled sample points are coordinates of each dead pixel and non-dead pixel, and the corresponding attribute parameter is a parameter indicating whether each sampling point is a dead pixel (for example, the attribute parameter of a dead pixel is 1, and the attribute parameter of a non-dead pixel is 0). For another example, in an etching model based on machine learning, a layout used for model training and hyper-parameter determination as a sample needs to be provided, the sample point marked on the top is a sample point coordinate preset according to etching characteristics, and the corresponding attribute parameter is a parameter representing the etching deviation of the sample point.
After the layout is provided, step S101 is executed first, N sample points on the layout are sampled according to the sampling window of the initial size, a layout graph of each sample point within the sampling window range is extracted, and N layout slices are obtained correspondingly, where N is greater than 1.
It should be noted that the initial size may be preset according to an empirical value, and the initial size may be set to a relatively large value in order to facilitate the subsequent determination of the target size therefrom. For example, if a model related to a lithography process is to be modeled or trained, the initial dimension may be set to an optical diameter.
After the initial size is preset, as shown in fig. 5, sampling each hollow small circle-shaped sampling point in the layout according to a sampling window shown by a dotted line frame, obtaining layout graphs around the sampling point as layout slices, and obtaining N corresponding layout slices for the N sample points.
Then, step S102 is executed, and feature extraction is performed on the N layout slices according to the physical size of the single pixel, so as to correspondingly convert the N layout slices into N feature matrices.
In the specific implementation process, the region on each layout slice is subjected to feature extraction by taking the physical size of a single pixel as a basic unit to form a feature matrix. The area which can contain the physical size of each single pixel on each layout slice corresponds to the position of each single pixel, and the area size occupied by each single pixel position is just equal to the physical size of each single pixel. The characteristic value of the layout graph extracted from any single pixel position is used as a numerical value on a row and a column corresponding to the single pixel position in the characteristic matrix.
For example, a layout slice can be divided into M × M regions with a single pixel size, wherein the feature value of the layout pattern at the single pixel position in the nth row and the mth column is used as the value of the nth row and the mth column of the feature matrix corresponding to the layout slice. And correspondingly extracting and converting the layout slices into a feature matrix with the size of M x M, wherein n and M are less than or equal to M.
According to different features to be extracted, the numerical meanings on the corresponding feature matrixes are different. If the features to be extracted are density information, as shown in fig. 6, the N layout slices are respectively subjected to density map-based feature extraction according to the physical size of a single pixel, so that the N layout slices are converted into N density matrices in a corresponding pixelization manner, and each value in the density matrices represents the layout pattern density of the region occupied by the corresponding single pixel position. And taking the density matrix as a characteristic matrix.
If the feature to be extracted is spectral information, the density matrix is further converted into a spectral matrix, and the spectral matrix is used as a feature matrix.
Next, step S103 is executed to calculate a mutual information value corresponding to each single pixel position of the N feature matrices based on the N sets of attribute parameters, and obtain a mutual information graph, where the mutual information value on the mutual information graph represents a correlation between the graphic feature and the attribute parameter at the single pixel position corresponding to the mutual information value.
In particular,the formula may be used first:
calculating a mutual information value MI corresponding to a single pixel position of the N characteristic matrixes at the (m, N) row-column coordinatesm,n(X, Y), where X is a feature value of each feature matrix in the N feature matrices at a single pixel position of (m, N) row-column coordinates, and X is a 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 a joint probability distribution function of x and y, and p (x) and p (y) are marginal probability distribution functions of x and y, respectively.
For example, N sample points are mapped to N feature matrices of size M × M, and as shown in fig. 7, N × M feature values are shared by the N feature matrices. Each feature matrix has N feature values x with the same coordinates (i.e., with the same single pixel location), and the N feature values x correspond to N attribute parameters y one-to-one. N characteristic values x and N attribute parameters y are correspondingly substituted into the formula for calculation, and the attribute parameters y corresponding to the N characteristic values x and N are cumulatively added
By adding the sum
As mutual information values corresponding to this same coordinate (m, n).
Then, corresponding mutual information value calculation is carried out on M-M single pixel positions on the feature matrix according to the 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 graph with the size of M-M. Namely, a mutual information graph is extracted from the N sample points. In the mutual information graph, the numerical value of the mutual information value at the corresponding coordinate of each single pixel position represents: the correlation between the layout graph at the position corresponding to the coordinate around the sample point and the sample attribute parameters is larger, the position is more important when the value of the mutual information value is larger, and the correlation is more worthy of attention in feature extraction. Each mutual information value occupies an area of a single pixel physical size on the mutual information map.
After the mutual information graph is obtained, step S104 is executed, and according to the mutual information graph, the target size of the sampling window is determined as the hyper-parameter for layout feature extraction.
In the specific implementation process, there are various methods for determining the target size of the sampling window according to the mutual information graph, and two methods are listed as examples below:
first, the mutual information graph is plotted as a curve from which the target size is determined.
Specifically, the correlation curve may be drawn 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. And determining the target size of the sampling window as the hyperparameter of the layout characteristic extraction according to the inflection point position of the correlation curve. Because the inflection point position represents the position where the correlation between the layout graph and the sample attribute parameters is affected by the distance and is weakened, the areas outside the distance of the inflection point can be considered as weak correlation, and the area does not need to be drawn into the range of a sampling window. Therefore, the target size of the sampling window determined according to the inflection point position is stronger in pertinence, is more suitable for the type of layout, and gives consideration to the characteristic extraction comprehensiveness and the calculation speed of the type of layout.
Preferably, if the model does not need to consider that the proximity effect of the layout has anisotropy, as shown in fig. 8, an average value of each circle of mutual information values (a square shown by oblique lines in the figure is a circle of mutual information values) centered at the center (black and white cross squares) of the mutual information graph is calculated; and drawing a correlation curve according to the average value and the distance from each circle of mutual information values to the center.
It should be noted that each circle of mutual information values refers to mutual information values having the same minimum number of mutual information values spaced from the most central mutual information value (black and white cross squares) of the mutual information graph, for example, in fig. 8, at least 1 mutual information value is spaced between the mutual information value of the squares indicated by oblique lines and the most central mutual information value, so the mutual information value of the squares indicated by oblique lines is one circle. Similarly, at least 2 mutual information values are arranged between the outermost mutual information value and the most central mutual information value, so that the outermost mutual information value is also one circle.
It should be noted that the distance from each circle of mutual information values to the center is the distance from the outermost edge of each circle of mutual information values to the center of the mutual information graph; or, the distance between the outermost periphery of each circle of mutual information values and the center of the inner circle of mutual information graph is the average value of the distance.
And drawing the correlation curve in fig. 8 by taking the distance from each circle of mutual information values to the center as an abscissa and taking the average value of the corresponding circle of mutual information values as an ordinate.
It can be seen from the correlation curve that as the distance from the center increases, the mutual information value becomes smaller, that is, the correlation becomes lower, and the layout graphic information at the position with a larger distance cannot provide sufficient help for the prediction of the sample attribute parameters, so that the distance value on the abscissa corresponding to the inflection point of the correlation curve can be multiplied by 2 to serve as the square side length (target size) of the sampling window, thereby taking into account the completeness and the calculation speed.
Preferably, if the model needs to consider that the proximity effect of the layout has anisotropy, as shown in fig. 9(a), a line average value of each line of mutual information values (black and white cross squares) of the mutual information graph and a line distance between each line of mutual information values and the center of the mutual information graph are calculated; and drawing a line correlation curve according to the line average value and the line distance. Then, as shown in fig. 9(b), a column average value of each column of mutual information values (black and white cross squares) of the mutual information graph and a column distance of each column of mutual information values from the center of the mutual information graph are calculated; and a column correlation curve is drawn according to the column mean and the column distance.
It should be noted that the distance from each line of mutual information values to the center is the distance between the edge of each line of mutual information values farthest from the center and the center of the mutual information graph; or, the distance between the side farthest from the center of each circle of mutual information value and the center of the mutual information graph and the distance between the side nearest to the center and the center of the mutual information graph are the average value. The calculated distance from each line of mutual information value to the center is taken as the abscissa, and the line average value of the corresponding line of mutual information value is taken as the ordinate, so as to draw the line correlation curve in fig. 9. The method of plotting the column correlation curve is the same as the method of plotting the row correlation curve, and will not be described herein in a repeated manner.
Then, determining the longitudinal size of a sampling window as a hyper-parameter for extracting the layout characteristics according to the inflection point position of the line correlation curve; and determining the transverse size of the sampling window as the hyperparameter of the layout feature extraction according to the inflection point position of the column correlation curve.
Specifically, the distance value on the abscissa corresponding to the inflection point of the row correlation curve is multiplied by 2 to be taken as the longitudinal dimension (dimension in the direction of the column) of the sampling window, and the distance value on the abscissa corresponding to the inflection point of the column correlation curve is multiplied by 2 to be taken as the lateral dimension (dimension in the direction of the row) of the sampling window, thereby taking into consideration the integrity and the calculation speed.
Second, a machine learning algorithm is employed.
Inputting the mutual information graph into a machine learning model which is trained by adopting a large number of mutual information graph samples and sample sizes, and outputting the target size according to the machine learning model.
Of course, the method for 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 herein.
Specifically, the two superparameters of the target size and the single-pixel physical size of the sampling window determined by the method provided by the embodiment are obtained by calculating and analyzing the sample layout, and the physical mechanism of modeling does not need to be additionally analyzed, so that the modeling process is simplified. And, by selecting a suitable sampling window target size considering the correlation, it is ensured that enough graphic information is collected around the coordinates of the sample point to express the influence of the proximity effect. By selecting proper single-pixel physical size considering the size of the layout graph, the condition that information redundancy and information loss cannot be caused in the pixelation process is ensured, so that the characteristic data is simplified to the maximum extent on the premise of not influencing the precision of a training model, and the training and implementation speed of the model is increased. Furthermore, the change characteristics of mutual information values in different directions along with the distance are considered when the correlation curve is drawn, and the influence of anisotropic proximity effect in the process on characteristic extraction can be considered, so that the extracted graph characteristics of the layout are more correlated with model prediction, and the accuracy is higher.
Based on the same inventive concept, the present disclosure provides a hyper-parameter determining apparatus for layout feature extraction, as shown in fig. 10, including:
the sampling module 1001 is configured to sample N sample points on the layout according to a sampling window of an initial size, and correspondingly obtain N layout slices, where N is greater than 1, where the N sample points correspond to N sets of attribute parameters;
the matrix conversion module 1002 is configured to perform feature extraction on the N layout slices according to the physical size of a single pixel, so as to correspondingly convert the N layout slices into N feature matrices;
a mutual information graph module 1003, configured to calculate, based on the N sets of attribute parameters, a mutual information value corresponding to each single pixel position of the N feature matrices, to obtain a mutual information graph, where a mutual information value on the mutual information graph represents a correlation between a graph feature and an attribute parameter at a single pixel position corresponding to the mutual information value;
and a determining module 1004, configured to determine, according to the mutual information graph, a target size of the sampling window as a hyper-parameter for layout feature extraction.
It should be noted that the above-mentioned hyper-parameter determining device for layout feature extraction is a device corresponding to the method for determining hyper-parameter for layout feature extraction provided in the foregoing embodiment, and the technical features introduced in the description of the method for determining hyper-parameter for layout feature extraction are all applicable to this device, and will not be described here in a repeated manner.
Based on the same inventive concept, the present disclosure provides a computer-readable storage medium 1100, as shown in fig. 11, on which a computer program 1111 is stored, the computer program 1111, when executed by a processor, implementing the steps of:
respectively sampling N sample points on the layout according to a sampling window with an initial size, and correspondingly obtaining N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
respectively extracting features of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N feature matrixes;
calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic characteristics and the attribute parameters at the single pixel position corresponding to the mutual information value;
and determining the target size of the sampling window as a hyper-parameter for extracting the layout features according to the mutual information graph.
In particular, the computer program 1111, when executed by a processor, may implement any of the methods of the embodiments of the present invention.
The technical scheme in the embodiment of the application at least has the following technical effects or advantages:
according to the method, the device and the medium for determining the hyper-parameters of the layout feature extraction, the initial size of a sampling window is preset, and sampling, feature extraction and feature matrix conversion are carried out according to the initial size. And then calculating according to the feature matrix and the attribute parameters of the sample points to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graph features and the attribute parameters at the single pixel position corresponding to the mutual information value, so that the target size determined by analyzing the mutual information graph is the size considering the correlation between the image features and the attribute parameters and is more consistent with the actual situation of the layout, the target size is taken as the hyperparameter for extracting the layout features, and the calculation speed and the calculation accuracy can be better considered when the layout features are used for model training or model application.
In the above description, the technical details of patterning, etching, and the like of each layer are not described in detail. It will be appreciated by those skilled in the art that layers, regions, etc. of the desired shape may be formed by various technical means. In addition, in order to form the same structure, those skilled in the art can also design a method which is not exactly the same as the method described above. In addition, although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used in advantageous combination.
It will be apparent to those skilled in the art that various changes and modifications can be made in the present disclosure without departing from the spirit and scope of the disclosure. Thus, if such 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 encompass such modifications and variations.
Claims (10)
1. A method for determining hyper-parameters of layout feature extraction is characterized by comprising the following steps:
respectively sampling N sample points on the layout according to a sampling window with an initial size, and correspondingly obtaining N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
respectively extracting features of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N feature matrixes;
calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic characteristics and the attribute parameters at the single pixel position corresponding to the mutual information value;
and determining the target size of the sampling window as a hyper-parameter for extracting the layout features according to the mutual information graph.
2. The method of claim 1, wherein the physical size of a single pixel is determined by:
obtaining length values of all edges of a polygon layout graph on the layout to form an edge length information group;
and performing statistical analysis on the side length information group, and determining the physical size of the single pixel according to the statistical analysis result.
3. The method of claim 2, wherein said determining said single-pixel physical size from said statistical analysis comprises:
and 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.
4. The method according to claim 1, wherein said extracting features from said N layout slices according to physical dimensions of a single pixel, respectively, to correspondingly convert said N layout slices into N feature matrices, comprises:
and respectively carrying out density map-based feature extraction on the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N density matrixes.
5. The method according to claim 1, wherein said calculating a mutual information value corresponding to each single pixel position of said N feature matrices based on said N sets of attribute parameters to obtain a mutual information map comprises:
using a formula
Calculating a mutual information value MI corresponding to a single pixel position of the N feature matrices at the (m, N) coordinatesm,n(X, Y), where X is the eigenvalue of each of the N eigenmatrices at a single pixel position in (m, N) 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 a joint probability distribution function of x and y, and p (x) and p (y) are marginal probability distribution functions of x and y, respectively;
and traversing the single pixel positions on the N characteristic matrixes, and arranging the calculated mutual information values according to the coordinates of the corresponding single pixel positions to obtain a mutual information graph.
6. The method according to claim 1, wherein the determining the target size of the sampling window as the hyper-parameter for layout feature extraction according to the mutual information graph comprises:
drawing a correlation curve according to a mutual information value on the mutual information graph and the distance between the mutual information value and the center of the mutual information graph;
and determining the target size of the sampling window as the hyperparameter extracted from the layout characteristics according to the inflection point position of the correlation curve.
7. The method of claim 6, wherein the 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:
calculating the average value of each circle of mutual information values by taking the center of the mutual information graph as the center;
and drawing a correlation curve according to the average value and the distance from each circle of mutual information value to the center.
8. The method of claim 6, wherein:
the 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:
calculating the line average value of each line of mutual information values of the mutual information graph and the line distance between each line of mutual information values and the center of the mutual information graph; drawing a line correlation curve according to the line average value and the line distance; calculating a column average value of each column of mutual information values of the mutual information graph and a column distance between each column of mutual information values and the center of the mutual information graph; drawing a column correlation curve according to the column average value and the column distance;
the step of determining the target size of the sampling window according to the inflection point position of the correlation curve as the hyper-parameter for extracting the layout features comprises the following steps:
determining the transverse size of the sampling window as a hyper-parameter for extracting layout features according to the inflection point position of the line correlation curve; and determining the longitudinal size of the sampling window as a hyper-parameter for extracting layout features according to the inflection point position of the column correlation curve.
9. A hyper-parameter determining device for layout feature extraction is characterized by comprising:
the sampling module is used for respectively sampling N sample points on the layout according to a sampling window with an initial size to correspondingly obtain N layout slices, wherein N is greater than 1, and the N sample points correspond to N groups of attribute parameters;
the matrix conversion module is used for respectively extracting the characteristics of the N layout slices according to the physical size of a single pixel so as to correspondingly convert the N layout slices into N characteristic matrices;
the mutual information graph module is used for calculating a mutual information value corresponding to each single pixel position of the N characteristic matrixes based on the N groups of attribute parameters to obtain a mutual information graph, wherein the mutual information value on the mutual information graph represents the correlation between the graphic characteristics and the attribute parameters at the single pixel position corresponding to the mutual information value;
and the determining module is used for determining the target size of the sampling window as the hyper-parameter for extracting the layout features according to the mutual information graph.
10. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010324225.8A CN111539177A (en) | 2020-04-22 | 2020-04-22 | Method, device and medium for determining hyper-parameters of layout feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010324225.8A CN111539177A (en) | 2020-04-22 | 2020-04-22 | Method, device and medium for determining hyper-parameters of layout feature extraction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111539177A true CN111539177A (en) | 2020-08-14 |
Family
ID=71977080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010324225.8A Pending CN111539177A (en) | 2020-04-22 | 2020-04-22 | Method, device and medium for determining hyper-parameters of layout feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539177A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112580292A (en) * | 2020-12-14 | 2021-03-30 | 南京华大九天科技有限公司 | Method for accelerating extraction of resistance, electronic device and computer-readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537288A (en) * | 2018-04-19 | 2018-09-14 | 辽宁大学 | A kind of real-time feature extraction method based on mutual information |
| CN109360185A (en) * | 2018-08-28 | 2019-02-19 | 中国科学院微电子研究所 | Layout test pattern extraction method, device, equipment and medium |
| US20190130264A1 (en) * | 2017-11-02 | 2019-05-02 | Palo Alto Research Center Incorporated | Deep graph representation learning |
| CN110110861A (en) * | 2019-05-09 | 2019-08-09 | 北京市商汤科技开发有限公司 | Determine method and apparatus, the storage medium of model hyper parameter and model training |
| CN110569969A (en) * | 2019-09-11 | 2019-12-13 | 北京百度网讯科技有限公司 | Network model structure sampling method and device and electronic equipment |
| WO2019240281A1 (en) * | 2018-06-15 | 2019-12-19 | 日本電信電話株式会社 | Image feature learning device, image feature learning method, image feature extraction device, image feature extraction method, and program |
-
2020
- 2020-04-22 CN CN202010324225.8A patent/CN111539177A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190130264A1 (en) * | 2017-11-02 | 2019-05-02 | Palo Alto Research Center Incorporated | Deep graph representation learning |
| CN108537288A (en) * | 2018-04-19 | 2018-09-14 | 辽宁大学 | A kind of real-time feature extraction method based on mutual information |
| WO2019240281A1 (en) * | 2018-06-15 | 2019-12-19 | 日本電信電話株式会社 | Image feature learning device, image feature learning method, image feature extraction device, image feature extraction method, and program |
| CN109360185A (en) * | 2018-08-28 | 2019-02-19 | 中国科学院微电子研究所 | Layout test pattern extraction method, device, equipment and medium |
| CN110110861A (en) * | 2019-05-09 | 2019-08-09 | 北京市商汤科技开发有限公司 | Determine method and apparatus, the storage medium of model hyper parameter and model training |
| CN110569969A (en) * | 2019-09-11 | 2019-12-13 | 北京百度网讯科技有限公司 | Network model structure sampling method and device and electronic equipment |
Non-Patent Citations (3)
| Title |
|---|
| CHEN JIE 等: "Mutual information-based dropout: Learning deep relevant feature representation architectures", NEUROCOMPUTING, vol. 361, 13 July 2019 (2019-07-13), pages 1 - 12 * |
| 任越美: "高光谱图像特征提取与分类方法研究", 中国博士学位论文全文数据库 信息科技辑, 15 January 2019 (2019-01-15), pages 140 - 160 * |
| 盖天洋 等: "基于机器学习的光刻坏点检测研究进展", 微纳电子技术, vol. 56, no. 6, 6 May 2019 (2019-05-06), pages 421 - 428 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112580292A (en) * | 2020-12-14 | 2021-03-30 | 南京华大九天科技有限公司 | Method for accelerating extraction of resistance, electronic device and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109360185B (en) | Layout test pattern extraction method, device, equipment and medium | |
| US11100266B2 (en) | Generating integrated circuit floorplans using neural networks | |
| US10885259B2 (en) | Random forest model for prediction of chip layout attributes | |
| US9690898B2 (en) | Generative learning for realistic and ground rule clean hot spot synthesis | |
| US20200027021A1 (en) | Reinforcement learning for multi-domain problems | |
| US10885261B2 (en) | Systems and methods of fabricating semiconductor devices | |
| CN103376644B (en) | A kind of mask pattern correction method | |
| US11475201B2 (en) | Inclusion of stochastic behavior in source mask optimization | |
| CN109858327B (en) | Character segmentation method based on deep learning | |
| CN117313640B (en) | Training method, device, equipment and storage medium for lithography mask generation model | |
| US11061321B1 (en) | Obtaining a mask using a cost function gradient from a Jacobian matrix generated from a perturbation look-up table | |
| US20190385300A1 (en) | Image morphing to meet desired constraints in geometric patterns | |
| CN103703341B (en) | Pattern measurement device and pattern measuring method | |
| Chen et al. | Lithography hotspot detection using a double inception module architecture | |
| US11188705B2 (en) | Pin accessibility prediction engine | |
| US20130198710A1 (en) | Semiconductor device feature density gradient verification | |
| CN111539177A (en) | Method, device and medium for determining hyper-parameters of layout feature extraction | |
| JP2008258361A (en) | Semiconductor device manufacturing method | |
| US11010525B2 (en) | Fast pattern matching | |
| Zhang et al. | Litho-NeuralODE 2.0: Improving hotspot detection accuracy with advanced data augmentation, DCT-based features, and neural ordinary differential equations | |
| CN114616571A (en) | Semiconductor layout environment around points of interest | |
| US20220171913A1 (en) | Methods and devices of correcting layout for semiconductor processes using machine learning | |
| CN111611759B (en) | Mask auxiliary graph optimization method, computer readable medium and system | |
| US11244440B2 (en) | Ranking of objects with noisy measurements | |
| JPH1187197A (en) | Simulation method and simulation apparatus for execution of the method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |