Capacitance value extraction method and apparatus based on sparse data machine learning technique

Abstract

The application provides a capacitor value extraction method and device based on sparse data machine learning technology, and belongs to the technical field of artificial intelligence and the technical field of electronic circuits. Feature extraction is performed on coordinate information of all sampling points of target point cloud data by a feature learning layer of a potential prediction model to obtain conductor form features, attribute information of each sampling point is spliced with the conductor form features to obtain target point cloud features of each sampling point, physical information of the target point cloud features is encoded by a physical information layer of the potential prediction model to obtain potentials of the sampling points. According to a target sampling point type, potentials of all sampling points located on a conductor Gaussian surface are taken as target potentials, and a capacitor value of the conductor is extracted according to the target potentials, so that a large amount of training data does not need to be calculated in advance by using a field solver, calculation resources are saved while calculation efficiency is improved, and in addition, the accuracy of capacitor value extraction can be improved by using point cloud to depict a conductor structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and the technical field of electronic circuits, and particularly relates to a capacitor value extraction method and device based on sparse data machine learning technology. BACKGROUND

[0002] In related technologies, artificial intelligence technology is used to optimize the Electronic Design Automation (EDA) process to improve chip design efficiency. In the process of artificial intelligence assisted chip design, extraction of parasitic parameters such as capacitor values is involved. When performing the capacitor value extraction process, a series of parameterized interconnection conductor structures are introduced to obtain structure description data, and the capacitor values of the interconnection conductor structures are solved on the basis of the structure description data by using a field solver. However, in actual integrated circuits, the interconnection lines are densely distributed, and the number of conductors is large, so a large amount of computing resources and time are consumed by using the field solver. In addition, the interconnection conductor structure is complex, which makes it impossible to accurately model and pattern match the interconnection conductor structure, resulting in a decrease in the prediction accuracy of the capacitor values. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a capacitor value extraction method and device based on sparse data machine learning technology, which aims to improve the calculation efficiency and accuracy of capacitor value extraction.

[0004] To achieve the above-mentioned purpose, a first aspect of the embodiments of the present application provides a capacitor value extraction method based on sparse data machine learning technology, which comprises:

[0005] Obtaining target point cloud data of a target interconnection conductor structure, wherein the target interconnection conductor structure comprises a conductor, a plurality of sampling points are arranged around the conductor, and the target point cloud data comprises attribute information of each sampling point; the attribute information comprises coordinate information and a target sampling point type of the sampling point;

[0006] Performing feature extraction on the coordinate information of all sampling points in the target point cloud data through a feature learning layer of a potential prediction model to obtain conductor morphology features of the target point cloud data;

[0007] Respectively performing feature splicing on the attribute information of each sampling point in the target point cloud data and the conductor morphology features to obtain target point cloud features of each sampling point;

[0008] Performing physical information coding on the target point cloud features of all sampling points through a physical information layer of the potential prediction model to obtain potentials of the sampling points;

[0009] Determine the sampling points on the Gaussian surface of the conductor according to the target sampling point type of each sampling point in the target point cloud data, and take the electric potential of all sampling points on the Gaussian surface of the conductor as a target electric potential;

[0010] Extract the capacitance value of the conductor according to the target electric potential.

[0011] In some embodiments, the electric potential prediction model is trained according to the following steps:

[0012] Obtain sample interconnect conductor point cloud data; the sample interconnect conductor point cloud data includes sample sampling points and sample attributes of the sample sampling points; the sample attributes include sample sampling point types;

[0013] Perform electric potential prediction on the sample attributes by a preset model to obtain a predicted electric potential of the sample sampling points under the sample sampling point types;

[0014] Obtain a reference electric potential of the sample sampling points according to the sample sampling point types;

[0015] Perform loss calculation according to the predicted electric potential and the reference electric potential to obtain a target loss;

[0016] Adjust model parameters of the preset model according to the target loss to obtain the electric potential prediction model.

[0017] In some embodiments, the loss calculation according to the predicted electric potential and the reference electric potential to obtain a target loss comprises:

[0018] Perform first loss calculation according to the predicted electric potential to obtain a first sub-loss;

[0019] Perform second loss calculation according to the predicted electric potential and the reference electric potential to obtain a second sub-loss;

[0020] Sum the first sub-loss and the second sub-loss to obtain the target loss.

[0021] In some embodiments, the sample attributes further include point cloud coordinates of the sample sampling points, the point cloud coordinates include point cloud horizontal coordinates and point cloud vertical coordinates, and the first loss calculation according to the predicted electric potential to obtain a first sub-loss comprises:

[0022] Obtain a second derivative of the predicted electric potential with respect to the point cloud horizontal coordinates to obtain a first loss value;

[0023] Obtain a second derivative of the predicted electric potential with respect to the point cloud vertical coordinates to obtain a second loss value;

[0024] Determine the first sub-loss according to the first loss value and the second loss value.

[0025] In some embodiments, the sample point types include a first boundary sample point subtype, a second boundary sample point subtype, a third boundary sample point subtype, and a fourth boundary sample point subtype, and the second loss calculation according to the predicted potential and the reference potential includes:

[0026] loss calculation according to the predicted potential and the reference potential of the sample points of the first boundary sample point subtype to obtain a third loss value;

[0027] loss calculation according to the predicted potential and the reference potential of the sample points of the second boundary sample point subtype to obtain a fourth loss value;

[0028] loss calculation according to the predicted potential and the reference potential of the sample points of the third boundary sample point subtype to obtain a fifth loss value;

[0029] determining a predicted electric field intensity according to the predicted potential of the sample points of the fourth boundary sample point subtype, determining a reference electric field intensity according to the reference potential, and performing loss calculation according to the predicted electric field intensity and the reference electric field intensity to obtain a sixth loss value;

[0030] loss summation of the third loss value, the fourth loss value, the fifth loss value, and the sixth loss value to obtain the second sub-loss.

[0031] In some embodiments, the feature learning layer includes a plurality of network sub-layers connected in sequence, and the feature learning layer of the potential prediction model performs feature extraction on the coordinate information of all sample points in the target point cloud data to obtain a conductor morphology feature of the target point cloud data, including:

[0032] performing feature extraction on the coordinate information of all sample points in the target point cloud data by the feature learning layer to obtain a first point cloud feature output by each network sub-layer;

[0033] performing feature splicing on the first point cloud features output by each network sub-layer to obtain the conductor morphology feature.

[0034] In some embodiments, the physical information layer includes a first convolutional layer, a residual layer, and a second convolutional layer, and the physical information layer of the potential prediction model encodes the target point cloud feature of all the sample points to obtain the potential of each sample point, including:

[0035] performing feature extraction on the target point cloud feature of all the sample points by the first convolutional layer to obtain a second point cloud feature;

[0036] extracting features of the second point cloud feature through the residual layer to obtain a third point cloud feature;

[0037] extracting features of the third point cloud feature through the second convolution layer to obtain the electric potential of each sampling point.

[0038] To achieve the above object, a second aspect of the embodiment of the present application provides a capacitor value extraction device based on sparse data machine learning technology, the device comprising:

[0039] an acquisition module configured to acquire target point cloud data of a target interconnection conductor structure, wherein the target interconnection conductor structure comprises a conductor, a plurality of sampling points are arranged around the conductor, and the target point cloud data comprises attribute information of each sampling point; the attribute information comprises coordinate information and a target sampling point type of the sampling point;

[0040] a feature extraction module configured to extract features of the coordinate information of all sampling points in the target point cloud data through a feature learning layer of an electric potential prediction model to obtain conductor morphology features of the target point cloud data;

[0041] a feature splicing module configured to splice the attribute information of each sampling point in the target point cloud data with the conductor morphology features respectively to obtain target point cloud features of each sampling point;

[0042] a feature encoding module configured to encode physical information of each sampling point through a physical information layer of the electric potential prediction model to obtain an electric potential of each sampling point;

[0043] a determination module configured to determine sampling points located on a conductor Gaussian surface according to the target sampling point type of each sampling point in the target point cloud data, and take the electric potential of all sampling points located on the conductor Gaussian surface as a target electric potential;

[0044] a capacitor value extraction module configured to extract a capacitor value of the conductor according to the target electric potential.

[0045] To achieve the above object, a third aspect of the embodiment of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the capacitor value extraction method based on sparse data machine learning technology of the first aspect when executing the computer program.

[0046] To achieve the above object, a fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the capacitor value extraction method based on sparse data machine learning technology of the first aspect.

[0047] The capacitor value extraction method based on sparse data machine learning technology, capacitor value extraction device based on sparse data machine learning technology, electronic equipment and computer readable storage medium provided by the application, by obtaining the target point cloud data of the target interconnection line conductor structure, the capacitor value of the target interconnection line conductor structure is extracted based on the target point cloud data, and the accuracy of the capacitor value extraction is improved by the point cloud description of the conductor structure. The coordinate information of all sampling points in the target point cloud data is extracted by the feature learning layer of the electric potential prediction model, so as to directly model the target interconnection line conductor structure by using the feature learning layer, obtain the geometric features of the point cloud data, and obtain the conductor shape features of the target point cloud data without introducing additional parameters to describe the interconnection line conductor structure. Without parameterizing the conductor structure, the application has better adaptability and universality to complex interconnection line conductor structures, and accurate modeling of the complex interconnection line conductor structure is realized. The electric potential of the sampling point is related to the attribute information of the sampling point and the shape features of the point cloud where the sampling point is located. In order to improve the accuracy of electric potential prediction, the attribute information of each sampling point is spliced with the conductor shape features of the target point cloud data to obtain the target point cloud features of each sampling point. A large amount of calculation resources and time resources are required to solve the capacitor value by the field solver. In order to improve the efficiency of capacitor value extraction, the physical information layer is set in the electric potential prediction model. The electric potential satisfies the electric field equation on the target point cloud data, and the target point cloud features of each sampling point are encoded by the physical information layer of the electric potential prediction model to encode the physical information of the electric field equation by using the physical information layer, and the electric potential of each sampling point is obtained. In order to extract the capacitor of the target interconnection line conductor structure, the sampling points located on the conductor Gaussian surface are determined according to the target sampling point type of each sampling point, the electric potential of all sampling points located on the Gaussian surface of the conductor is taken as the target electric potential, and the electric potential of the Gaussian surface sampling point is used to solve the capacitor without using the field solver to calculate a large amount of training data in advance, thereby saving calculation resources and improving the efficiency of capacitor extraction. The capacitor value of the conductor is extracted according to the target electric potential, instead of using the field solver and the structure description data obtained by parameterizing the conductor structure to extract the capacitor, thereby avoiding the decrease of the prediction accuracy of the capacitor value caused by the inaccurate parameterization of the conductor structure, and improving the accuracy of the capacitor value extraction. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is a flowchart of the capacitor value extraction method based on sparse data machine learning technology provided by the embodiment of the application;

[0049] Figure 2 is a schematic diagram of the target point cloud data provided by the embodiment of the application;

[0050] Figure 3 is Figure 1 is a flowchart of step S120 in

[0051] Figure 4 is Figure 1 a flowchart of step S140 in

[0052] Figure 5a is another schematic diagram of target point cloud data provided by an embodiment of the present application;

[0053] Figure 5b is another schematic diagram of target point cloud data provided by an embodiment of the present application;

[0054] Figure 5c is another schematic diagram of target point cloud data provided by an embodiment of the present application;

[0055] Figure 5d is Figure 5a a schematic diagram of an electric field of target point cloud data of

[0056] Figure 5e is Figure 5b a schematic diagram of an electric field of target point cloud data of

[0057] Figure 5f is Figure 5c a schematic diagram of an electric field of target point cloud data of

[0058] Figure 6 is a flowchart of a training process of an electric potential prediction model provided by an embodiment of the present application;

[0059] Figure 7 is Figure 6 a flowchart of step S640 in

[0060] Figure 8 is Figure 7 a flowchart of step S710 in

[0061] Figure 9 is Figure 7 a flowchart of step S720 in

[0062] Figure 10 is a structural schematic diagram of a capacitance value extraction device based on sparse data machine learning technology provided by an embodiment of the present application;

[0063] Figure 11 is a hardware structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0065] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the device or the sequence in the flowchart. The terms "first", "second", and the like in the specification and claims and the above-described drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application is for the purpose of describing embodiments of this application only, and is not intended to be limiting of this application.

[0067] In related technologies, artificial intelligence technology is used to assist chip design, such as using a machine learning method to determine the optimal circuit layout, interconnection line parasitic parameter extraction, and the like, to greatly shorten the time required for chip design and improve the design efficiency of the chip. In the timing analysis process of chip design, parasitic parameters such as capacitance extraction are involved. How to efficiently and accurately extract the parasitic parameters of the interconnection line is a basic problem of chip design. In actual integrated circuits, the interconnection line distribution is very dense, and the number of conductors is large, making the circuit layout very complex, and the parasitic parameter extraction of a large-scale layout usually involves thousands of interconnection line conductor data. Moreover, circuit design is an iterative process, and as long as the line changes, the parasitic parameters need to be re-extracted, which makes the parasitic parameter extraction a very large amount of calculation task. In addition, with the scale reduction and increasing complexity of semiconductor process technology, the accuracy requirement of parasitic parameter extraction is significantly improved, and advanced process structures such as low dielectric constant medium, non-vertical interface conductor, and bubble medium lead to increased modeling complexity of interconnection lines, making parameterization of complex conductor structures and layouts difficult. Most of the current methods for parasitic parameter extraction use data-driven neural networks, machine learning methods, and the like, and the main idea is to describe the interconnection line conductor structure through some parameters, and set a series of sampling values for these parameters to obtain a batch of interconnection line conductor structures, and solve their capacitance values through a field solver. However, these parameters cannot accurately model complex interconnection line conductor structures, and the field solver requires a large amount of time resources and computing resources to solve the capacitance, resulting in a decrease in the prediction accuracy and efficiency of the capacitance.

[0068] Based on this, the embodiments of the present application provide a capacitance value extraction method based on sparse data machine learning technology, a capacitance value extraction device based on sparse data machine learning technology, an electronic device, and a computer readable storage medium, aiming to improve the accuracy and efficiency of capacitance extraction.

[0069] The method for extracting a capacitance value based on sparse data machine learning technology provided by the embodiment of the present application, the device for extracting a capacitance value based on sparse data machine learning technology, the electronic device and the computer readable storage medium are specifically described through the following embodiment. First, the method for extracting a capacitance value based on sparse data machine learning technology in the embodiment of the present application is described.

[0070] The method for extracting a capacitance value based on sparse data machine learning technology provided by the embodiment of the present application relates to the technical field of artificial intelligence and the technical field of electronic circuits. The method for extracting a capacitance value based on sparse data machine learning technology provided by the embodiment of the present application can be applied in a terminal, can be applied in a server end, and can also be software running in a terminal or a server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc. The server end can be configured as an independent physical server, can be configured as a server cluster or a distributed system composed of multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and big data and artificial intelligence platforms, etc. The software can be an application for implementing the method for extracting a capacitance value, etc., but is not limited to the above forms.

[0071] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0072] Figure 1 is an optional flowchart of the method for extracting a capacitance value based on sparse data machine learning technology provided by the embodiment of the present application, Figure 1 The method in can include but is not limited to including steps S110 to S160.

[0073] In step S110, target point cloud data of a target interconnection conductor structure is acquired, wherein the target interconnection conductor structure comprises a conductor, a plurality of sampling points are arranged around the conductor, the target point cloud data comprises attribute information of each sampling point, and the attribute information comprises coordinate information and a target sampling point type of the sampling point.

[0074] In step S120, the coordinate information of all the sampling points in the target point cloud data is subjected to feature extraction by a feature learning layer of the electric potential prediction model, so as to obtain conductor morphology features of the target point cloud data.

[0075] In step S130, the attribute information of each sampling point in the target point cloud data is subjected to feature splicing with the conductor morphology features, so as to obtain target point cloud features of each sampling point.

[0076] In step S140, the target point cloud features of each sampling point are subjected to physical information coding by a physical information layer of the electric potential prediction model, so as to obtain an electric potential of each sampling point.

[0077] In step S150, a sampling point located on a Gaussian surface of the conductor is determined according to the target sampling point type of each sampling point in the target point cloud data, and the electric potential of all the sampling points located on the Gaussian surface of the conductor is taken as a target electric potential.

[0078] In step S160, a capacitance value of the conductor is extracted according to the target electric potential.

[0079] In step S110 of some embodiments, the target interconnect conductor structure is an interconnect conductor structure to be extracted for parasitic parameters, which refers to a conductive path in an integrated circuit design used to connect different circuit elements (such as capacitors, resistors, transistors, etc.). These conductive paths are usually made of metal or other conductive materials, which form a complex network on multiple levels of the chip to achieve electrical connections between circuit elements. The point cloud of the target interconnect conductor structure can be generated by laser scanning, 3D modeling software, computer vision SLAM algorithm, etc. to obtain target point cloud data. The target interconnect conductor structure includes at least one conductor, and each conductor is surrounded by a plurality of sampling points. The target point cloud data includes attribute information of each sampling point, and the attribute information is multi-dimensional data including a plurality of attributes, including point cloud coordinates of the sampling point, a target sampling point type, a number of conductors in the target interconnect conductor structure (data window) to which the sampling point belongs, an outer normal vector, etc. The sampling points are divided into sampling points at the boundary of the data window, sampling points inside the calculation region, sampling points on the boundary of the conductor, and sampling points on the Gaussian surface of the conductor. The Gaussian surface is a pre-set closed surface, which can be any shape, such as a plane, a spherical surface, a cylindrical surface, etc. Among them, the sampling points at the boundary of the data window include sampling points with Dirichlet boundary conditions and sampling points with Neumann boundary conditions. The calculation region inside refers to other regions in the data window of the target point cloud data except all conductor regions. The sampling points on the boundary of the conductor include sampling points on the boundary of the main conductor and sampling points on the boundary of the accompanying conductor. The main conductor refers to the conductor to be solved for capacitance, and the accompanying conductor refers to other conductors that affect the capacitance of the main conductor. The target sampling point type is used to indicate the type of the sampling point, which is one of the main conductor boundary sampling point, the accompanying conductor boundary sampling point, the Dirichlet boundary sampling point, the Neumann boundary sampling point, the calculation region inside sampling point, and the conductor Gaussian surface sampling point.

[0080] The attribute information of the jth sampling point is represented as v j = (x j , l j , …, nnet j ), where x j represents the coordinate information (point cloud coordinates) of the jth sampling point, l j represents the boundary condition label, which is used to indicate the target sampling point type of the sampling point, and nnet j represents the number of conductors in the data window. The target point cloud data of the target interconnect conductor structure is as follows Figure 2As shown, the boundary condition of the main conductor boundary sampling point is marked as 1, the boundary condition of the accompanying conductor boundary sampling point is marked as 2, the boundary condition of the Dirichlet boundary sampling point is marked as 3, the boundary condition of the Neumann boundary sampling point is marked as 4, the boundary condition of the conductor Gaussian surface sampling point is marked as 5, and the boundary condition of the internal calculation region is marked as 6.

[0081] The potential prediction model is used to map the target point cloud data to the electric potential, and the potential prediction model comprises a feature learning layer and a physical information layer. The feature learning layer is located in the upper half of the potential prediction model, and is used to perform geometric coding on the target point cloud data to extract geometric features of the point cloud data. A neural network architecture similar to PointNet is used to construct the feature learning layer, and the main idea of this kind of network is to use a multi-layer perceptron neural network with shared weights and a global symmetric pooling function to construct a low-dimensional representation of a set of point clouds. The physical information layer is located in the lower half of the potential prediction model, and is used to encode the electric field equation to obtain physical information of the electric field equation.

[0082] Referring to Figure 3 In some embodiments, the feature learning layer comprises a plurality of network sub-layers connected in sequence, and step S120 can comprise but is not limited to steps S310 to S320:

[0083] In step S310, the feature learning layer is used to extract features of coordinate information of all sampling points in the target point cloud data, to obtain first point cloud features output by each network sub-layer;

[0084] In step S320, the first point cloud features output by each network sub-layer are spliced to obtain conductor morphology features.

[0085] In step S310 of some embodiments, the feature learning layer includes a main network and a permutation invariant network layer, the main network adopts a two-dimensional convolution operator, and the permutation invariant network layer can adopt a max pooling operator, an average pooling operator, a min operator, or a summation operator, etc. The max pooling operator refers to applying a sliding window on a feature map, selecting the maximum value from each window to generate a new feature map. The average pooling operator refers to calculating the average value of all values in each window to generate a new feature map. The min operator refers to selecting the minimum value from each window to generate a new feature map. The summation operator refers to summing all values in each window to generate a new feature map. The network sub-layer includes a two-dimensional convolution layer or a permutation invariant network layer, the feature learning layer includes five two-dimensional convolution layers and one permutation invariant network layer connected in sequence, and the five two-dimensional convolution layers are denoted as conv1, conv2, conv3, conv4, and conv5 from top to bottom. The parameter configurations of the five two-dimensional convolution layers are as follows: the convolution kernel size of conv1 is 1*2, and the output channel number is 64; the convolution kernel size of conv2 and conv3 is 1*1, and the output channel number is 128; the convolution kernel size of conv4 is 1*1, and the output channel number is 512; the convolution kernel size of conv5 is 1*1, and the output channel number is 2048. The step of all two-dimensional convolution operations is [1, 1]. An activation function is used for each two-dimensional convolution layer, and batch normalization (BN) is used before the activation function of each convolution layer to solve the internal covariate shift problem.

[0086] The attribute information includes coordinate information of the sampling points, and the coordinate information is used to indicate point cloud coordinates of the sampling points, which can be represented by two-dimensional space coordinates. If the target point cloud data includes N sampling points, a point cloud coordinate set composed of point cloud coordinates of the N sampling points is denoted as X N , and the feature dimension of X N is N*2. The coordinate information, i.e., the point cloud coordinate set of all sampling points in the target point cloud data, is input into the feature learning layer, the point cloud coordinates of each sampling point are processed by two-dimensional convolution through the five two-dimensional convolution layers connected in sequence, and the first point cloud feature output by each two-dimensional convolution layer is obtained. Through the five two-dimensional convolution layers, the channel dimension is increased from 2 to 2048. In the spatial dimension, the first point cloud feature output by the last two-dimensional convolution layer is processed by permutation invariant through the permutation invariant network layer to aggregate information shared by all pixel points in the first point cloud feature, and the first point cloud feature output by the permutation invariant network layer is obtained. The feature channel number of the first point cloud feature output by the permutation invariant network layer is the same as that of the first point cloud feature output by the last two-dimensional convolution layer, and both are 2048. The first point cloud feature is a point cloud geometric feature of all sampling points in the target point cloud data for the coordinate information, including normal, curvature, distance between sampling points, etc.

[0087] In step S320 of some embodiments, the first point cloud features output by the five two-dimensional convolutional layers and the first point cloud features output by the permutation invariant network layer are spliced in the channel dimension to obtain conductor morphology features of the target point cloud data. The conductor morphology features are global features and are used to represent the point cloud geometric features of the sampling points. If the feature dimension of the conductor morphology features of a sampling point is represented as: 64+128+128+512+2048+2048=4928, 64, 128, 128, 512, 2048 are the dimensions of the feature vectors output by the convolutional layers, and 2048 is the dimension of the feature vector after the permutation invariant operation. The feature dimension of the conductor morphology features of N sampling points, i.e., the conductor morphology features of the target point cloud data, is N×4928.

[0088] The steps S310 to S320 described above directly model the interconnection line conductor structure by using the feature extraction layer to extract the geometric features of the point cloud data, without introducing an additional series of parameters to parameterize the interconnection line conductor structure, thereby achieving accurate modeling of various types of conductor structures including complex conductor structures.

[0089] In step S130 of some embodiments, the electric potential u j of the jth sampling point depends not only on the attribute information of the jth sampling point but also on the shape features of the point cloud in which the jth sampling point is located. The attribute information of the jth sampling point is represented as v j , and the conductor morphology features of all the sampling points are represented as V i is the target point cloud data of the ith sample. In order to improve the accuracy of subsequent electric potential prediction, the attribute information of the jth sampling point is spliced with the conductor morphology features of all the sampling points in the target point cloud data to obtain the target point cloud features of the jth sampling point as

[0090] Please refer to Figure 4 In some embodiments, the physical information layer includes a first convolutional layer, a residual layer, and a second convolutional layer, and step S140 can include but is not limited to steps S410 to S430:

[0091] In step S410, the target point cloud features of all the sampling points are extracted by the first convolutional layer to obtain second point cloud features;

[0092] In step S420, the second point cloud features are extracted by the residual layer to obtain third point cloud features;

[0093] In step S430, the third point cloud features are extracted by the second convolutional layer to obtain the electric potential of each sampling point.

[0094] In step S410 of some embodiments, the input of the physical information layer includes the first point cloud feature output by each network sub-layer in the feature learning layer and the conductor morphology feature of all sampling points. The physical information layer is also composed of convolution layers, which are the same as the feature learning layer, and each convolution layer includes an activation function and a batch normalization operation. The first convolution layer includes four two-dimensional convolution layers that are sequentially connected, which are conv6, conv7, conv8, and conv9, and the convolution kernel size is 1*1. The output channel number of conv1 is 256, the output channel number of conv7 and conv8 is 128, and the output channel number of con9 is 1. The target point cloud feature of all sampling points is subjected to two-dimensional convolution through the four two-dimensional convolution layers that are sequentially connected, to obtain the second point cloud feature formed by all sampling points. Through the four two-dimensional convolution layers, the channel number of the feature can be reduced from 1928 dimensions to 1 dimension.

[0095] In step S420 of some embodiments, in order to improve the approximation accuracy of the physical information layer to the physical information of the electric field equation, a residual layer can be added to the model, and the residual layer includes a plurality of same residual structures. The residual structure can be described by the following formula:

[0096] M1 = φ(ConvR1(W 1 ,x1) + b 1 ),

[0097] N1 = φ(ConvR2(W 2 ,M1) + b 2 ),

[0098] y1 = N1 + x1 = F(x1) + x1,

[0099] wherein ConvR1 and ConvR2 represent the first convolution layer and the second convolution layer of the residual structure, W 1 and b 1 represent the weight parameter and the bias parameter of the first convolution layer; W 2 and b 2 represent the weight parameter and the bias parameter of the second convolution layer; x1 represents the feature vector input to the first convolution layer; φ represents the activation function; M1 represents the output of the first convolution layer or the input of the second convolution layer; N1 is the output of the second convolution layer; F represents the mapping function of x1 to N1; and y1 represents the output of the residual structure.

[0100] Regarding the selection of the activation function φ, since the subsequent coding involves the physical information of the electric field equation, i.e., the partial differential equation, it is necessary to ensure that the potential prediction model is differentiable, and therefore a differentiable activation function is selected, such as sin, GeLU, tanh, etc.

[0101] The second point cloud feature of each sampling point is extracted by a residual structure in a residual layer to capture more abstract and complex point cloud feature information, and third point cloud features of the sampling points are obtained.

[0102] In step S430 of some embodiments, the second convolutional layer includes a convolutional layer conv10, and the third point cloud features of all sampling points are extracted by the conv10 to obtain electric potentials of the sampling points, the electric potential of the sampling point being a one-dimensional feature. wherein E represents the electric field, represents a first-order gradient, and U represents the electric potential. The point cloud coordinates include point cloud horizontal coordinates and point cloud vertical coordinates, and the first-order gradient is a vector composed of a first-order partial derivative of the electric potential with respect to the point cloud horizontal coordinates and a first-order partial derivative of the electric potential with respect to the point cloud vertical coordinates.

[0103] The activation function of the second convolutional layer can be selected based on prior knowledge. For example, if it is required to ensure that the electric potential is within the range of [0, 1], the activation function can be defined as a sigmoid function, and the definition of the sigmoid function is as follows:

[0104]

[0105] wherein γ represents a feature vector input to the activation function.

[0106] If there is no prior knowledge, the second convolutional layer conv10 can have no activation function.

[0107] It should be noted that the number of output channels of the convolutional layer in the network structure is not fixed, but can be debugged according to the actual situation. The total number of sampling points, the complexity of the conductor structure, the sample number of the conductor point cloud data, and the memory consumed by neural network calculation, etc. These factors will affect the fitting accuracy and training efficiency of the neural network, which need to be considered comprehensively according to the actual situation after testing.

[0108] Through the above steps S410 to S430, the electric potential of each sampling point can be obtained to extract the capacitance of the conductor based on the electric potential without using a field solver, thereby improving the efficiency of capacitance extraction.

[0109] The target point cloud data of the first sample, the second sample and the third sample are respectively as follows: Figure 5a 、 Figure 5b 、 Figure 5cAs shown, the feature learning layer of the electric potential prediction model extracts features from the target point cloud data of the three samples respectively, obtaining geometric coding features of the three samples. For each sample, the geometric coding features are spliced with the target point cloud data, and the spliced features are subjected to electric potential prediction by the physical information layer to obtain the electric potential of each sample. The electric field of each sample is determined according to the electric potential of each sample. The electric field of the first sample is shown in Figure 5d The electric field of the second sample is shown in Figure 5e The electric field of the third sample is shown in 5f.

[0110] In the related art, the structure description data is obtained by parameterizing the interconnection conductor structure, and the capacitance of the interconnection conductor structure is calculated in advance by a field solver, and the neural network model is trained using the structure description data and the capacitance. However, solving the capacitance by the field solver requires a large amount of computing resources and time resources, and the parameterized modeling method cannot accurately depict complex interconnection conductor structures, resulting in a decrease in the efficiency and accuracy of capacitance extraction. In order to improve the capacitance extraction performance of the model, the present application adopts an unsupervised method for model training, without using the field solver to calculate the capacitance in advance and without parameterizing the conductor structure, which can save a large amount of computing resources and computing time, and at the same time can improve the accuracy of capacitance extraction.

[0111] Please refer to Figure 6 In some embodiments, the training process of the electric potential prediction model can include, but is not limited to, steps S610 to S650:

[0112] Step S610, obtaining sample interconnection conductor point cloud data; the sample interconnection conductor point cloud data includes sample sampling points and sample attributes of the sample sampling points; the sample attributes include sample sampling point types;

[0113] Step S620, predicting the electric potential of the sample attributes by a preset model to obtain the predicted electric potential of the sample sampling points under the sample sampling point types;

[0114] Step S630, obtaining the reference electric potential of the sample sampling points according to the sample sampling point types;

[0115] Step S640, calculating the loss according to the predicted electric potential and the reference electric potential to obtain a target loss;

[0116] Step S650, adjusting the model parameters of the preset model according to the target loss to obtain an electric potential prediction model.

[0117] In step S610 of some embodiments, a point cloud dataset is generated based on the batch interconnect conductor structures. The batch interconnect conductor structures include M interconnect conductor structure samples, each interconnect conductor structure sample corresponding to a point cloud subset. The point cloud dataset is a collection of multiple point cloud subsets, each point cloud subset consisting of a fixed number (e.g., N) of sampling points. Each point cloud subset consists of multi-dimensional data characterizing the features of that point cloud subset, including the point cloud coordinates of all sampling points within the point cloud subset, the boundary condition annotations of all sampling points, the number of conductors contained within the conductor sample window, the external normal vectors of all sampling points, etc.

[0118] Each point cloud subset contains N sampling points. Labeling all sampling points within a point cloud subset results in each subset containing N×1 boundary condition labels. This application uses a configured standard channel to write the aforementioned multidimensional point cloud data into a fixed HDF5 file for unified data storage. The specific configuration of the standard channel is as follows: the first data channel is named `data`, containing M×N×2 point cloud coordinate data, with data type `float32` (32-bit floating-point number); the second data channel is named `label`, containing M×N×1 label data (boundary condition labels), with data type `uint8` (8-bit unsigned integer); the third data channel is named `nnet`, containing M×1 `uint8` data, representing the number of conductors contained in each point cloud subset; the fourth data channel is named `nnett_gp`, containing M×15 `uint8` data, representing the number of Gaussian surface sampling points in each conductor; the fifth data channel is named `normal`, containing M×N... b ×2 data points of type float32, storing the external normal vector coordinates of all boundary sampling points, N b This represents the number of boundary sampling points. It can be assumed that each conductor data window contains no more than 15 conductors. When the number of conductors n... c When it is less than 15, the nth c The number of sampling points on the Gaussian surface after that is recorded as 0.

[0119] 80% of the point cloud data can be randomly selected from the point cloud dataset as the training set, denoted as . m is the number of samples in the training set, i.e., the number of point cloud subsets. From the remaining 20%, 10% is taken from each set as the test set and the validation set, respectively. The test set is denoted as... The verification set is denoted as l and k are the number of samples in the test set and the validation set, respectively.

[0120] A point cloud subset is obtained from the training set, and the point cloud subset is depicted as a multi-dimensional data set containing a plurality of attributes, to obtain sample interconnect conductor point cloud data. The sample interconnect conductor point cloud data includes sample sampling points and sample attributes of the sample sampling points. The sample sampling points are sampling points in the point cloud subset, and the sample attributes include point cloud coordinates of the sample sampling points, sample sampling point types, a number of conductors contained in a data window, an external normal vector, and the like. Sample interconnect conductor point cloud data of an i-th sample can be expressed as:

[0121]

[0122] wherein x j is a point cloud coordinate of a j-th sample sampling point, l j is a boundary condition label of the j-th sample sampling point, nnet j is a number of conductors in the data window, and N is a number of the sampling points.

[0123] The point cloud coordinates include point cloud horizontal coordinates and point cloud vertical coordinates. The point cloud subset contains four different boundary conditions. The sample sampling point types include conductor Gaussian surface sampling points, sampling points in a calculation region, and boundary sampling points. The boundary sampling points include four types, namely, a first boundary sampling point subtype, a second boundary sampling point subtype, a third boundary sampling point subtype, and a fourth boundary sampling point subtype. The first boundary sampling point subtype is a main conductor boundary sampling point. The second boundary sampling point subtype is an accompanying conductor boundary sampling point. The third boundary sampling point subtype is a Dirichlet boundary sampling point (a sampling point of a Dirichlet boundary condition). The fourth boundary sampling point subtype is a Neumann boundary sampling point (a sampling point of a Neumann boundary condition). The sample sampling point types are the same as a target sampling point type, and can be labeled in the same manner as the target sampling point type. Details are not described herein.

[0124] The boundary conditions of the conductors change with the change of the interconnect conductor structure. The number and coordinates of the boundary sampling points corresponding to different boundary conditions also change. In order to facilitate the representation and calculation of the boundary conditions, One-Hot encoding, also known as one-bit effective encoding, is used to encode the boundary conditions, to obtain a discrete variable l j used to represent the point cloud data label. jBoundary condition labels for the jth sample point are used to ensure the stability and accuracy of the training when using low-precision calculations. One-hot encoding is a method used to represent discrete variables, which uses a multi-bit status register to encode multiple states, each with its own independent register bit, and only one bit is valid at any time. Through one-hot encoding, discrete variables can be converted into a multi-dimensional vector, which facilitates the encoding of boundary conditions into the capacitance solving model. Since the point cloud subset of the present application includes 6 boundary conditions, a 6-bit status register is used to encode the label state of all sample points in the point cloud subset. That is, the label state of all sample points in the point cloud subset is encoded using a 6-bit status register l j,one-hoo (dimension MxNx6). j (dimension MxNxl).

[0125] In step S620 of some embodiments, the sample interconnect conductor point cloud data is input into a preset model for training an unsupervised capacitance solving model. The model structure of the preset model is the same as that of the potential prediction model, both including a feature learning layer and a physical information layer, and the preset model is used to establish a mapping from the point cloud subset V i to the potential U i . representing the potential of the point cloud subset V i (conductor data window). The feature learning layer is used to extract features from the point cloud coordinates in the sample interconnect conductor point cloud data, obtaining conductor shape features of all sample points in the point cloud data, and the sample attributes of each sample point are spliced with the conductor shape features of all sample points to obtain the spliced features of each sample point. The physical information layer is used to predict the potential of all sample points in the point cloud data, obtaining the predicted potential of each sample point. The value of the predicted potential u j of the jth sample point depends not only on the sample attributes such as the point cloud coordinates x j , boundary condition label l j , but also on the shape of the point cloud subset where the point j is located, and the mathematical expression of the predicted potential is defined as:

[0126]

[0127] wherein, represents the geometric feature encoding of V i , i.e., the conductor shape features of all sample points; is the preset model, and Θ is the model parameter of the preset model.

[0128] The point cloud data is used to represent the conductor window data of the interconnection line to reduce the memory occupation. The shape and distribution of the interconnection line conductor are variable, and it is difficult to extract the features of these geometrical shapes and distributions by using a traditional parameterization method, which cannot describe the local details of the conductor structure and results in low accuracy of the capacitance prediction. The accuracy and accuracy of the structural feature extraction determine the upper limit of the capacitance extraction accuracy. In order to extract the capacitance with higher accuracy, an accurate structural feature extraction method is needed. The geometric features of the point cloud data are extracted by using the feature learning layer, which can be used for feature processing of the conductor data with arbitrary shape and arrangement. Compared with the existing capacitance value extraction method, the conductor structure and position do not need to be parameterized, which is more convenient and efficient, and can be widely used in various conductor interconnection line data. The complex conductor medium will lead to complex electric field distribution around the conductor. The potential and derivative that satisfy the electric field equation are predicted through the physical information layer, so as to calculate the capacitance of the conductor. While maintaining high accuracy, the training data is not needed, which saves a lot of calculation time and computing resources. After the model is trained, the potential of the conductor data window can be predicted according to the interconnection line conductor structure, and then the capacitance of the conductor can be solved according to the potential. The trained model can not only predict the electric field and potential value of the conductor window in the training set, but also predict the electric field and potential value of the conductor structure that does not appear in the training set. Through the preset model, the potential of the conductor structure can be predicted with high accuracy, and then the capacitance of the conductor can be extracted with high accuracy.

[0129] In step S630 of some embodiments, the preset model maps the input V i to U i , U i satisfies the electric field equation on V i , and the constraint condition of the electric field equation is expressed as: The predicted potential of all sample sampling points satisfies the constraint condition of the electric field equation, and the constraint condition of the boundary condition is expressed as:

[0130] U=u0, at the boundary γ p of the main conductor,

[0131] U=u1, at the boundary γ a of the accompanying conductor,

[0132] U=u D , at the boundary γ D corresponding to the Dirichlet boundary condition,

[0133] at the boundary γ N corresponding to the Neumann boundary condition,

[0134] wherein, represents the first-order gradient; U represents the potential; q represents the normal electric field intensity; k represents the dielectric constant, which can be a constant or a function varying with the solution region; a first-order gradient of the electric potential, denoted as denotes the electric potential U i a first-order partial derivative of the point cloud horizontal coordinate x j , denotes the electric potential U i a first-order partial derivative of the point cloud vertical coordinate y j ; n is an external normal vector.

[0135] Based on the constraint condition of the above boundary condition, if the sample sampling point type is a first boundary sampling point subtype (main conductor boundary sampling point), the reference electric potential is a first sub-electric potential u0; if the sample sampling point type is a second boundary sampling point subtype (companion conductor boundary sampling point), the reference electric potential is a second sub-electric potential u1; if the sample sampling point type is a third boundary sampling point subtype (Dirichlet boundary sampling point), the reference electric potential is a third sub-electric potential u D ; if the sample sampling point type is a fourth boundary sampling point subtype (Neumann boundary sampling point), the reference electric potential is u N , and a first-order gradient of the reference electric potential is calculated, and a dot product of the first-order gradient and the external normal vector is taken as the reference normal electric field strength If the sample sampling point type is a conductor Gaussian surface sampling point or a sampling point in the calculation region, the reference electric potential does not need to be obtained. It should be noted that the boundary sample sampling point needs to calculate the reference electric potential, and the conductor Gaussian surface sampling point or the sampling point in the calculation region only needs to ensure that the predicted electric potential satisfies the constraint condition of the electric field equation, without the need to calculate the reference electric potential, so that the sample interconnection line conductor point cloud data is sparse data, and the efficiency of the capacitance value extraction is further improved.

[0136] In step S640 of some embodiments, in order to measure the model training performance of the preset model, loss data needs to be calculated. According to the predicted electric potential output by the preset model, the reference electric potential obtained by using the constraint condition of the boundary condition, and the electric field equation residual, the target loss is calculated by loss calculation, and the model parameters of the preset model are adjusted according to the target loss.

[0137] In step S650 of some embodiments, the preset model is trained using the Adam optimizer with a fixed learning rate, such as keeping the learning rate at 0.0005 and iterating for several rounds (epochs). When the target loss decreases to a smaller value, the training of the preset model is stopped, and the preset model at this time is taken as the electric potential prediction model.

[0138] Through the steps S610 to S650, a large number of capacitance value data required for training does not need to be calculated by a field solver, the calculation process is more concise and efficient, a large amount of computing resources and computing time can be saved, and the conductor structure does not need to be parameterized, which has better adaptability and universality to complex conductor structures and complex media. The processing on the conductor structure is relatively flexible, can be used for capacitance parameter prediction of irregular (non-Manhattan) conductors, and does not increase the calculation complexity, and is suitable for chip conductor interconnection line parameter extraction of various processes, including interconnection line conductor capacitance prediction of arbitrary shape, distribution and complex material. While ensuring the accuracy of capacitance prediction, the capacitance prediction speed is greatly improved.

[0139] Please refer to Figure 7 In some embodiments, the step S640 can include but is not limited to steps S710 to S730:

[0140] In step S710, a first loss calculation is performed according to the predicted potential, to obtain a first sub-loss;

[0141] In step S720, a second loss calculation is performed according to the predicted potential and the reference potential, to obtain a second sub-loss;

[0142] In step S730, the first sub-loss and the second sub-loss are summed to obtain the target loss.

[0143] In step S710 of some embodiments, the target loss includes two parts of an electric field equation residual part and a boundary prediction error part, and the target loss is the sum of squares of equation residuals and errors of boundary condition prediction values of all sample points in the training set. The electric field equation residual part is taken as the first sub-loss, and the boundary prediction error part is taken as the second sub-loss. Since U i In V i satisfies the electric field equation, the electric field equation residual part is added to the loss function to encode the physical information of the electric field equation in the preset model. In order to encode the electric field equation residual into the preset model, the second derivative of the potential U i is calculated x j and y j respectively represent the point cloud horizontal coordinate and the point cloud vertical coordinate of the jth sample point. In order to facilitate the calculation of the capacitance, the first derivative of the potential U can also be calculated

[0144] The present application uses U θ to approximate the potential of the calculation region in the data window and defines a residual formula based on the electric field equation, which is expressed as:

[0145]

[0146] wherein, r θ is the electric field equation residual; θ is a model parameter of the preset model; U θ is a predicted potential output by the preset model when the model parameter is θ; represents a second-order derivative of the potential; x j and y j respectively represent a point cloud horizontal coordinate and a point cloud vertical coordinate of a sample sampling point.

[0147] A first sub-loss is obtained by loss calculation according to the predicted potential of each sample sampling point. The first sub-loss is a mean square of the equation residual of all sample sampling points on the training set. The calculation method of the first sub-loss is as follows:

[0148]

[0149] wherein, M represents a sample number in the training set, that is, a point cloud subset number; N is a sample sampling point number in each point cloud subset; i represents an i-th sample; j represents a j-th sample sampling point; x i,j and y i,j respectively represent a point cloud horizontal coordinate and a point cloud vertical coordinate of the j-th sample sampling point in the i-th point cloud subset.

[0150] In step S720 of some embodiments, a second sub-loss is obtained by second loss calculation according to the predicted potential and the reference potential of each sample sampling point. The second sub-loss is a sum of mean square errors of the predicted potential and the reference potential of the sample sampling points of the four boundary conditions.

[0151] In step S730 of some embodiments, the first sub-loss and the second sub-loss are added to obtain a target loss. The target loss is represented as:

[0152]

[0153] wherein, is the target loss.

[0154] The steps S710 to S730 described above can encode the physical information of the electric field equation in the physical information layer through the first sub-loss, so that the predicted potential output by the preset model satisfies the electric field equation. The second sub-loss can make the predicted potential of the boundary condition close to the reference potential. The target loss is determined by the first sub-loss and the second sub-loss, so as to optimize the preset model based on the target loss, and make the preset model output an accurate potential.

[0155] Please refer to Figure 8In some embodiments, step S710 can include but is not limited to steps S810-S830:

[0156] In step S810, a second derivative of the predicted electric potential with respect to the horizontal coordinate of the point cloud is obtained to obtain a first loss value.

[0157] In step S820, a second derivative of the predicted electric potential with respect to the vertical coordinate of the point cloud is obtained to obtain a second loss value.

[0158] In step S830, the first sub-loss is determined according to the first loss value and the second loss value.

[0159] In step S810 of some embodiments, in the ith point cloud subset, if the predicted electric potential of the jth sample sampling point is represented as U θ , the second derivative of the predicted electric potential with respect to the horizontal coordinate x i,j of the point cloud is calculated to obtain the first loss value. The first loss value is represented as

[0160] In step S820 of some embodiments, the second derivative of the predicted electric potential U θ with respect to the vertical coordinate y i,j of the point cloud is calculated to obtain the second loss value. The second loss value is represented as

[0161] In step S830 of some embodiments, the first loss value and the second loss value are summed to obtain the electric field equation residual r θ of the jth sample sampling point in the ith point cloud subset. The square of the electric field equation residual r θ of the jth sample sampling point is calculated, the squares of the electric field equation residuals r θ of all sample sampling points in all point cloud subsets are summed, and the mean of the sum is calculated as the first sub-loss. The mean can be obtained by dividing the sum by (MxN), where M is the number of point cloud subsets and N is the number of sample sampling points in the point cloud subset.

[0162] Through the above steps S810-S830, the first sub-loss can be obtained to adjust the predicted electric potential output by the preset model based on the first sub-loss, so that the predicted electric potential can satisfy the electric field equation and improve the accuracy of the predicted electric potential prediction.

[0163] Please refer to Figure 9 In some embodiments, step S720 can include but is not limited to steps S910-S950:

[0164] In step S910, the predicted electric potential and the reference electric potential of the sample sampling point of the first boundary sampling point sub-type are used for loss calculation to obtain a third loss value.

[0165] Step S920, loss calculation is performed according to the predicted electric potential and the reference electric potential of the sample sampling point of the second boundary sampling point sub-type, to obtain a fourth loss value;

[0166] Step S930, loss calculation is performed according to the predicted electric potential and the reference electric potential of the sample sampling point of the third boundary sampling point sub-type, to obtain a fifth loss value;

[0167] Step S940, the predicted electric field intensity is determined according to the predicted electric potential of the sample sampling point of the fourth boundary sampling point sub-type, the reference electric field intensity is determined according to the reference electric potential, and loss calculation is performed according to the predicted electric field intensity and the reference electric field intensity, to obtain a sixth loss value;

[0168] Step S950, loss summation is performed on the third loss value, the fourth loss value, the fifth loss value and the sixth loss value, to obtain a second sub-loss.

[0169] In step S910 of some embodiments, the first boundary sampling point sub-type is a main conductor boundary sampling point, and the predicted electric potential of the sample sampling point of the first boundary sampling point sub-type is represented as The reference electric potential of the sample sampling point of the first boundary sampling point sub-type is represented as The square of the error between the predicted electric potential and the reference electric potential of the sample sampling point is calculated, and the square of the error is represented as:

[0170]

[0171] Wherein, l j,one-hot represents the one-hot encoding of the boundary condition label of the jth sample sampling point; k c is the label value of the main conductor boundary condition, k c = 0; · represents multiplication operation.

[0172] The sum of the squares of the errors of all sample sampling points of the first boundary sampling point sub-type in all point cloud subsets is calculated, and the mean value of the sum is calculated to obtain the third loss value. The third loss value is the mean square error of the predicted electric potential and the reference electric potential of the main conductor boundary sampling point. The mean value can be obtained by dividing the sum value by the number of main conductor boundary sampling points (MxN c,i ) in the training set. The third loss value is represented as:

[0173]

[0174] Wherein, M is the number of point cloud subsets in the training set; N c,i is the number of main conductor boundary sampling points in the ith point cloud subset.

[0175] In step S920 of some embodiments, the second boundary sampling point subtype is the accompanying conductor boundary sampling point. The square of the error between the predicted potential U and the reference potential u1 of the accompanying conductor boundary sampling point is calculated, and the square of the error of the sample sampling points of all second boundary sampling point subtypes in all point cloud subsets is summed. The sum is then divided by the number of all accompanying conductor boundary sampling points in the training set (M×N). a,i The fourth loss value is obtained by calculating the mean square error between the predicted potential and the reference potential at the sampling points along the conductor boundary. The fourth loss value is expressed as:

[0176]

[0177] Where, N a,i k represents the number of sampling points associated with the conductor boundary in the i-th point cloud subset; a k is the labeled value for the conductor boundary condition. a =1.

[0178] In step S930 of some embodiments, the third boundary sampling point subtype is a Dirichlet boundary sampling point, and the predicted potential U and reference potential u of the Dirichlet boundary sampling point are calculated. D The squared errors between points are calculated, and the sum of the squared errors of all Dirichlet boundary sampling points in the point cloud subsets is calculated. The sum is then divided by the number of all Dirichlet boundary sampling points in the training set. The fifth loss value is obtained. The fifth loss value is the mean square error between the predicted potential and the reference potential at the Dirichlet boundary sampling points. The fifth loss value is expressed as:

[0179]

[0180] in, k represents the number of Dirichlet boundary sampling points in the i-th point cloud subset; d k is the labeled value for the Dirichlet boundary condition. d =2.

[0181] In step S940 of some embodiments, the fourth boundary sampling point subtype is the Newman boundary sampling point, based on the electric field strength calculation formula. The predicted electric field strength is calculated based on the predicted potential U and the outward normal vector n at the Newman boundary sampling points, and the reference electric field strength is calculated based on the reference potential and the outward normal vector. The squared error between the predicted and reference electric field strengths is calculated, and the squared errors are summed over all Newman boundary sampling points in all point cloud subsets. The sum is then divided by the number of Newman boundary sampling points in the training set. Dividing this by the square of the Newman boundary reference length yields the sixth loss value. The sixth loss value is the mean square error between the predicted electric field strength and the reference electric field strength at the Newman boundary sampling points. The sixth loss value is expressed as:

[0182]

[0183] wherein, is the number of Neumann boundary sampling points in the i-th point cloud subset; k n is the label value of the Neumann boundary condition, k n = 3; q θ is the predicted electric field intensity, and θ is the model parameter of the preset model; represents the reference electric field intensity; h n is the Neumann boundary reference length.

[0184] Since the characteristic scales of different boundary conditions are different, in order to ensure the calculation accuracy, the error term of the Neumann boundary condition needs to be multiplied by h n -2 .

[0185] The predicted electric field intensity can also be output by the preset model based on the input vector . The reference electric field intensity can also be directly set without being obtained according to the reference potential.

[0186] N c,i , N a,i , are the numbers of the primary conductor boundary sampling points, the accompanying conductor boundary sampling points, the Dirichlet boundary sampling points, and the Neumann boundary sampling points, respectively, and:

[0187] N c,i =∑l j,one-hot [:,:,k c ],

[0188] N a,i =∑l j,one-hot [:,:,k a ],

[0189]

[0190] In step S950 of some embodiments, the loss values of the third loss value, the fourth loss value, the fifth loss value, and the sixth loss value of the four boundary conditions are summed up to obtain a second sub-loss.

[0191] Through the above steps S910 to S950, the second sub-loss can be obtained to adjust the predicted potential of different boundary sampling points based on the second sub-loss, thereby improving the accuracy of the predicted potential.

[0192] In step S150 of some embodiments, the sampling point of the target sampling point type of the conductor Gaussian surface sampling point is determined as the sampling point located on the Gaussian surface of the conductor, and the potential of all sampling points located on the Gaussian surface of the conductor is determined as the target potential.

[0193] In step S160 of some embodiments, the capacitance calculation formula is represented as:

[0194]

[0195] Wherein, C represents the capacitance; Q represents the charge quantity; U represents the voltage, i.e. the electric potential.

[0196] The interconnection line capacitance is equivalent to a multi-conductor capacitor, and the conductors constituting the capacitor are the electrodes of the capacitor. To calculate the capacitance between any two electrodes (conductors) i and j, a bias U ij is applied between the two electrodes, and the charge on the electrode surface is obtained, that is:

[0197]

[0198] According to Gauss's theorem, it can be obtained that:

[0199] Q i =∫kE·dS,

[0200] Wherein, Q i represents the charge quantity of the interconnection line conductor structure i (the i-th sample); k represents the dielectric constant; E represents the electric field; dS represents the micro-face vector on the closed surface S i ; S i is the Gauss plane of the interconnection line conductor structure i; · represents the multiplication operation.

[0201] The above formula is discretely summed, and the approximate charge value Q i of the interconnection line conductor structure i is derived, which is represented as:

[0202]

[0203] Wherein, E i,j is the normal electric field of the j-th sampling point on the conductor i Gauss plane; k i is the dielectric constant of the medium around the conductor i; is the varying area of the conductor i Gauss plane; U θ,i,j is the target potential of the j-th sampling point on the conductor i Gauss plane; N g,i is the number of sampling points on the conductor i Gauss plane.

[0204] The interconnection line conductor structure i includes nnet i conductors. After predicting the target potential of the sampling point located on the conductor Gauss plane, the first derivative of the target potential of each Gauss plane sampling point to the point cloud coordinates of the sampling point is calculated to obtain the target normal electric field of the Gauss plane sampling point. The target normal electric field is substituted into the above discrete charge equation to obtain the capacitance value of the conductor. The capacitance calculation formula is represented as:

[0205]

[0206] wherein n represents the outer normal vector of the sampling point on the Gaussian surface; Eij represents the target electric field of the jth sampling point on the Gaussian surface of the conductor i; g is the sampling interval length.

[0207] It should be noted that the conductor Gaussian surface sampling point is a sampling point obtained by uniformly sampling the Gaussian surface with h g as the reference length, and n is the outer normal vector of the corresponding sampling point.

[0208] Five kinds of interconnection conductor structures of Test case1, Test case2, Test case3, Test case4 and Test case5 are given, and the capacitance extraction method, BEM boundary element method and PINN physical neural network of the application are used to extract the capacitance of the given interconnection conductor structure, and the capacitance extraction results are shown in Table 1.

[0209] Table 1

[0210]

[0211] As can be seen from Table 1, the capacitance prediction method of the application has achieved good results on a variety of capacitance extraction tasks. In terms of capacitance prediction accuracy, the application achieves results comparable to the BEM boundary element method, with an error of less than 10%. In terms of prediction speed, the application significantly speeds up compared to the BEM boundary element method (traditional method), PINN physical neural network and other methods.

[0212] Referring to Figure 10 , the application embodiment also provides a capacitance value extraction device based on sparse data machine learning technology, which can implement the above-mentioned capacitance value extraction method based on sparse data machine learning technology. The capacitance value extraction device comprises:

[0213] The acquisition module 1010 is configured to acquire target point cloud data of a target interconnection conductor structure, wherein the target interconnection conductor structure comprises a conductor, a plurality of sampling points are arranged around the conductor, and the target point cloud data comprises attribute information of each sampling point; the attribute information comprises coordinate information and a target sampling point type of the sampling point;

[0214] The feature extraction module 1020 is configured to perform feature extraction on the attribute information of all sampling points in the target point cloud data through a feature learning layer of an electric potential prediction model, to obtain conductor morphology features of the target point cloud data.

[0215] The feature splicing module 1030 is configured to splice the attribute information of each sampling point in the target point cloud data with the conductor shape feature respectively to obtain the target point cloud feature of each sampling point.

[0216] The feature encoding module 1040 is configured to perform physical information encoding on the target point cloud feature of all sampling points by using the physical information layer of the electric potential prediction model to obtain the electric potential of each sampling point.

[0217] The determination module 1050 is configured to determine the sampling points located on the Gaussian surface of the conductor according to the target sampling point type of each sampling point in the target point cloud data, and take the electric potential of all sampling points located on the Gaussian surface of the conductor as the target electric potential.

[0218] The capacitance value extraction module 1060 is configured to extract the capacitance value of the conductor according to the target electric potential.

[0219] The specific implementation of the capacitance value extraction apparatus is basically the same as the specific embodiments of the above-mentioned capacitance value extraction method, and will not be repeated here.

[0220] The embodiments of the present application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above-mentioned capacitance value extraction method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0221] Please refer to Figure 11 , Figure 11 The hardware structure of the electronic device of another embodiment is illustrated, which includes:

[0222] The processor 1110 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, etc., and is used to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0223] The memory 1120 can be implemented in the form of a ROM (ReadOnly Memory), a static storage device, a dynamic storage device, or a RAM (Random Access Memory), etc. The memory 1120 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 1120 and are called and executed by the processor 1110 to implement the capacitance value extraction method of the embodiments of the present application.

[0224] The input / output interface 1130 is configured to realize information input and output.

[0225] The communication interface 1140 is configured to realize communication interaction between the device and other devices, and the communication can be realized through a wired manner (for example, a USB, a network cable and the like) or a wireless manner (for example, a mobile network, WIFI, Bluetooth and the like).

[0226] The bus 1150 is configured to transmit information between various components (for example, the processor 1110, the memory 1120, the input / output interface 1130 and the communication interface 1140) of the device.

[0227] The processor 1110, the memory 1120, the input / output interface 1130 and the communication interface 1140 are connected to each other through the bus 1150 to realize communication connection between devices.

[0228] The computer readable storage medium provided in the embodiment of the present application stores a computer program, and the computer program is executed by the processor to realize the above-mentioned capacitor value extraction method.

[0229] The memory is a non-transitory computer readable storage medium, and can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, for example, at least one magnetic disk storage device, a flash memory device or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and the remote memory can be connected to the processor through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0230] The embodiments described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0231] Those skilled in the art can understand that the technical solutions shown in the figure do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than the figure, or combine certain steps or different steps.

[0232] The device embodiments described above are only schematic, and the units described as separate components can be or can not be physically separated, that is, can be located in one place or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to realize the purpose of the embodiments of the present application.

[0233] Those skilled in the art can understand that all or some of the steps in the method disclosed above, the function modules / units in the system and the device can be implemented as software, firmware, hardware and appropriate combinations thereof.

[0234] The terms "first", "second", "third", "fourth" and the like in the description of the application and in the claims hereof, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so termed herein is to be interpreted to only cover the embodiments of the application described herein and not a prior art. Moreover, the use of the terms "first", "second", "third", "fourth", and / or the like, to describe a variety of elements / parameters / operators, are generally not intended to convey an ordering, except to the extent provided for in the sentence in which they are used, and are generally intended to convey a "first", "second", "third", "fourth", and / or like identification in some other manner. Additionally, the terms "comprise", "comprising", "include", "including", and their conjugates, shall cover non-exclusive inclusion such that comprising, including, and the like are to be interpreted in the manner as set forth by the United States Court of Appeals for the Federal Circuit and the United States Supreme Court: "... this transition term [i.e., comprising], is naturaly understood to open and normally does permit addition of one or more steps to complete the recited operation." Phillips v. AWH Corp., 315 F.3d 1326, 1332 (Fed. Cir. 2002) (citing Home Construction Co. v. Flova, 342 U.S. 337, 340 (1952)).

[0235] It should be understood that, in the present application, "at least one" means one or more, and "multiple" means two or more. "And / or", used to describe the relationship between associated objects, means that there can be three relationships, for example, "A and / or B" can mean that there are only A, only B, and A and B at the same time, where A and B can be singular or plural. The character " / " generally represents that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can mean a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0236] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above-mentioned units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. The coupling or direct coupling or communication connection between the displayed or discussed objects can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0237] The units described as separate components above can or can not be physically separate, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0238] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0239] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical scheme of the present application or the part of the present application that essentially contributes to the prior art or the whole or part of the technical scheme can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0240] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and the scope of the rights of the embodiments of the present application is not limited thereto. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. A method for extracting a capacitance value based on a sparse data machine learning technique, characterized by, The method comprises: acquiring target point cloud data of a target interconnect conductor structure, wherein the target interconnect conductor structure comprises a conductor, a plurality of sampling points are arranged around the conductor, and the target point cloud data comprises attribute information of each sampling point; the attribute information comprises coordinate information and a target sampling point type of the sampling point; extracting features of the coordinate information of all sampling points in the target point cloud data through a feature learning layer of a potential prediction model to obtain conductor morphology features of the target point cloud data; concatenating the attribute information of each sampling point in the target point cloud data with the conductor morphology features to obtain target point cloud features of each sampling point; encoding physical information of the target point cloud features of all sampling points through a physical information layer of the potential prediction model to obtain potentials of the sampling points; determining sampling points on a Gaussian surface of the conductor according to the target sampling point type of each sampling point in the target point cloud data, and taking the potentials of all sampling points on the Gaussian surface of the conductor as target potentials; extracting a capacitance value of the conductor according to the target potentials; the feature learning layer comprises a plurality of network sublayers connected in series, and the extracting features of the coordinate information of all sampling points in the target point cloud data through the feature learning layer to obtain the conductor morphology features of the target point cloud data comprises: extracting features of the coordinate information of all sampling points in the target point cloud data through the feature learning layer to obtain first point cloud features output by each network sublayer; and concatenating the first point cloud features output by each network sublayer to obtain the conductor morphology features; the physical information layer comprises a first convolution layer, a residual layer and a second convolution layer, and the encoding physical information of the target point cloud features of all sampling points through the physical information layer of the potential prediction model to obtain the potentials of the sampling points comprises: extracting features of the target point cloud features of all sampling points through the first convolution layer to obtain second point cloud features; extracting features of the second point cloud features through the residual layer to obtain third point cloud features; and extracting features of the third point cloud features through the second convolution layer to obtain the potentials of the sampling points.

2. The capacitance value extraction method according to claim 1, wherein The potential prediction model is obtained through training according to the following steps: acquiring sample interconnect conductor point cloud data; the sample interconnect conductor point cloud data comprises sample sampling points and sample attributes of the sample sampling points; the sample attributes comprise a sample sampling point type; predicting potentials of the sample attributes through a preset model to obtain predicted potentials of the sample sampling points under the sample sampling point type; acquiring reference potentials of the sample sampling points according to the sample sampling point type; calculating a loss according to the predicted potentials and the reference potentials to obtain a target loss; adjusting model parameters of the preset model according to the target loss to obtain the potential prediction model.

3. The capacitance value extraction method according to claim 2, wherein The calculating a loss according to the predicted potentials and the reference potentials to obtain a target loss comprises: The first loss calculation is performed according to the predicted potential, and a first sub-loss is obtained; The second loss calculation is performed according to the predicted potential and the reference potential, and a second sub-loss is obtained; The first sub-loss and the second sub-loss are summed to obtain the target loss.

4. The capacitance value extraction method according to claim 3, wherein The sample attribute further includes point cloud coordinates of the sample sampling points, the point cloud coordinates including point cloud horizontal coordinates and point cloud vertical coordinates, and the first loss calculation performed according to the predicted potential to obtain the first sub-loss includes: The second derivative of the predicted potential with respect to the point cloud vertical coordinates is obtained to obtain a second loss value; The second derivative of the predicted potential with respect to the point cloud vertical coordinates is obtained to obtain a second loss value; The first sub-loss is determined according to the first loss value and the second loss value.

5. The capacitance value extraction method according to claim 3, wherein The sample sampling point type includes a first boundary sampling point sub-type, a second boundary sampling point sub-type, a third boundary sampling point sub-type, and a fourth boundary sampling point sub-type, and the second loss calculation performed according to the predicted potential and the reference potential to obtain the second sub-loss includes: The third loss value is obtained by performing loss calculation on the predicted potential and the reference potential of the sample sampling points of the first boundary sampling point sub-type; The fourth loss value is obtained by performing loss calculation on the predicted potential and the reference potential of the sample sampling points of the second boundary sampling point sub-type; The fifth loss value is obtained by performing loss calculation on the predicted potential and the reference potential of the sample sampling points of the third boundary sampling point sub-type; The sixth loss value is obtained by performing loss calculation on the predicted electric field intensity determined according to the predicted potential of the sample sampling points of the fourth boundary sampling point sub-type and the reference electric field intensity determined according to the reference potential; The third loss value, the fourth loss value, the fifth loss value, and the sixth loss value are summed to obtain the second sub-loss.

6. A device for extracting a capacitance value based on a sparse data machine learning technique, characterized by, The device includes: An acquisition module is configured to acquire target point cloud data of a target interconnection line conductor structure, wherein the target interconnection line conductor structure includes a conductor, a plurality of sampling points are arranged around the conductor, and the target point cloud data includes attribute information of each sampling point; the attribute information includes coordinate information and a target sampling point type of the sampling point; A feature extraction module is configured to perform feature extraction on the coordinate information of all sampling points in the target point cloud data through a feature learning layer of a potential prediction model, to obtain conductor morphology features of the target point cloud data; A feature concatenation module is configured to perform feature concatenation on the attribute information of each sampling point in the target point cloud data and the conductor morphology features, respectively, to obtain target point cloud features of each sampling point; A feature encoding module is configured to perform physical information encoding on the target point cloud features of all sampling points through a physical information layer of the potential prediction model, to obtain potentials of the sampling points; A determination module is configured to determine sampling points located on a conductor Gaussian surface according to the target sampling point type of each sampling point in the target point cloud data, and take the potentials of all sampling points located on the conductor Gaussian surface as target potentials. The capacitance value extraction module is configured to extract a capacitance value of the conductor according to the target electric potential; The feature learning layer comprises a plurality of network sub-layers connected in sequence, and the feature learning layer of the electric potential prediction model is configured to perform feature extraction on the coordinate information of all sampling points in the target point cloud data to obtain a conductor shape feature of the target point cloud data, and the conductor shape feature comprises: The feature learning layer is configured to perform feature extraction on the coordinate information of all sampling points in the target point cloud data to obtain a first point cloud feature output by each network sub-layer, and perform feature splicing on the first point cloud feature output by each network sub-layer to obtain the conductor shape feature. The physical information layer comprises a first convolutional layer, a residual layer and a second convolutional layer, and the physical information layer of the electric potential prediction model is configured to perform physical information coding on the target point cloud feature of all the sampling points to obtain an electric potential of each sampling point, and the physical information layer comprises: The first convolutional layer is configured to perform feature extraction on the target point cloud feature of all the sampling points to obtain a second point cloud feature, the residual layer is configured to perform feature extraction on the second point cloud feature to obtain a third point cloud feature, and the second convolutional layer is configured to perform feature extraction on the third point cloud feature to obtain the electric potential of each sampling point.

7. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the capacitance value extraction method based on the sparse data machine learning technology according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 7. The computer program is executed by the processor to implement the capacitance value extraction method based on the sparse data machine learning technology according to any one of claims 1 to 5.

Images