Method, apparatus, and storage medium for extracting and sparsely storing parasitic parameters of a map

By combining the boundary element method and Voronoi diagram with a sparse hierarchical storage structure, the contradiction between high precision and high speed is resolved, achieving efficient parasitic parameter extraction and sparse storage, which is applicable to the field of integrated circuit electronic design automation.

CN123070048APending Publication Date: 2026-11-10ZHUHAI SILICON CORE TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611586943.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

At advanced process nodes and chip-level layout scales, existing technologies struggle to simultaneously achieve high-precision extraction and high-speed calculation of parasitic parameters, creating a contradiction between accuracy and efficiency.

Method used

The boundary element method is used to extract parasitic parameters of the self-network based on the contour of arbitrary polygons and electrical function nodes. Combined with Voronoi graph and sparse hierarchical storage structure, the coupling candidate polygon pairs are determined by Voronoi neighbor relationship, and sparse coupling edges are extracted and stored.

Benefits of technology

It achieves high-precision parasitic parameter extraction and rapid coupling calculation at the chip-level layout scale, reducing computational complexity and improving extraction efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070048A_ABST
    Figure CN123070048A_ABST
Patent Text Reader

Abstract

This application discloses a method, apparatus, and storage medium for extracting and sparsely storing parasitic parameters of a layout. The method includes acquiring layout data, identifying arbitrary polygons in the layout data, and determining electrical functional nodes on the arbitrary polygons; using each arbitrary polygon as a unified electromagnetic entity, extracting self-network parasitic parameters based on the contours and electrical functional nodes of the arbitrary polygons to obtain multi-port self-network blocks of the arbitrary polygons; constructing a Voronoi diagram based on the spatial positions of multiple arbitrary polygons, and determining Voronoi neighbor relationships between the arbitrary polygons according to the adjacency relationships in the Voronoi diagram; determining coupling candidate polygon pairs according to the Voronoi neighbor relationships, and extracting coupling parasitic parameters from the coupling candidate polygon pairs to obtain sparse coupling edges; and storing the multi-port self-network blocks and sparse coupling edges according to a sparse hierarchical adjacency storage structure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of integrated circuit electronic design automation (EDA) technology, specifically to a method, apparatus, and storage medium for extracting and sparsely storing layout parasitic parameters. Background Technology

[0002] With the continuous evolution of integrated circuit process nodes and the rapid development of 3D integrated circuit multi-chip stacking technology, the layout geometry in chip physical design is becoming increasingly complex, with non-Manhattan structures, rounded polygons, arbitrary angle traces, and planar layers containing internal vias appearing in large numbers. Parasitic parameter extraction, as a key step connecting layout design with subsequent timing analysis, signal integrity analysis, and power integrity analysis, aims to extract parasitic parameters such as resistance, capacitance, and inductance of interconnects from GDS or OASIS layout data.

[0003] However, under advanced processes, the three-dimensional effect can no longer be effectively compensated by traditional two-dimensional or 2.5-dimensional approximations. Furthermore, the high-precision three-dimensional full-wave method is difficult to apply to chip-level extraction scenarios with millions of polygons due to the superlinear increase in computational cost with the layout size. As a result, parasitic parameter extraction cannot simultaneously achieve high precision and high speed under advanced process nodes and chip-level layout scales, forming a fundamental contradiction between precision and efficiency. Summary of the Invention

[0004] This application provides a method, apparatus, and storage medium for extracting and sparsely storing parasitic parameters in a layout, which can simultaneously achieve high-precision extraction, fast coupled calculation, and sparse hierarchical storage of parasitic parameters at the chip-level layout scale.

[0005] In a first aspect, embodiments of this application provide a method for extracting and sparsely storing layout parasitic parameters, including: Acquire layout data, identify any polygon in the layout data, and determine the electrical function nodes on the arbitrary polygon; Using each of the arbitrary polygons as a unified electromagnetic entity, self-network parasitic parameters are extracted based on the contour of the arbitrary polygon and the electrical functional node to obtain the multi-port self-network block of the arbitrary polygon. A Voronoi diagram is constructed based on the spatial positions of the arbitrary polygons, and the Voronoi neighbor relationships between the arbitrary polygons are determined according to the adjacency relationships in the Voronoi diagram. Based on the Voronoi neighbor relationship, coupled candidate polygon pairs are determined, and the coupling parasitic parameters are extracted from the coupled candidate polygon pairs to obtain sparse coupled edges. The multi-port self-network block and the sparsely coupled edge are stored according to a sparse hierarchical adjacency storage structure.

[0006] In the layout parasitic parameter extraction and sparse storage method provided in this application embodiment, the step of using each arbitrary polygon as a unified electromagnetic entity, and extracting self-network parasitic parameters based on the contour of the arbitrary polygon and the electrical function node to obtain the multi-port self-network block of the arbitrary polygon includes: The contour of the arbitrary polygon is piecewise linearly discretized to obtain multiple boundary segments; Using the electrical functional nodes as ports, the boundary element method is used to construct the port impedance matrix of the arbitrary polygon based on multiple boundary segments; Based on the port impedance matrix, an arbitrary polygonal multi-port self-network block is obtained.

[0007] In the layout parasitic parameter extraction and sparse storage method provided in this application embodiment, the step of constructing the port impedance matrix of the arbitrary polygon based on multiple boundary segments using the electrical functional node as the port includes: Each of the aforementioned boundary segments is set as a constant current element; Based on the Green's function of a two-dimensional conductive plane, the impedance contribution between any two boundary segments is calculated to obtain the impedance relationship of the boundary segments. Based on the correspondence between the electrical functional nodes and each of the boundary segments, the impedance relationships of the boundary segments are assembled into a port impedance matrix with the electrical functional nodes as ports.

[0008] In the layout parasitic parameter extraction and sparse storage method provided in this application embodiment, the step of determining the coupling candidate polygon pairs based on the Voronoi neighbor relationship and extracting the coupling parasitic parameters from the coupling candidate polygon pairs to obtain sparse coupling edges includes: Two arbitrary polygons having the Voronoi neighbor relationship are identified as a pair of coupled candidate polygons. For the coupled candidate polygon pair, obtain the local geometry and relative angle of the nearest edge between the two arbitrary polygons; Based on the local geometry and the relative angle, shape-aware coupling calculation is performed on the pair of coupling candidate polygons to obtain the coupling capacitance between the pair of coupling candidate polygons. The pair of coupling candidate polygons and the coupling capacitor are used as sparse coupling edges.

[0009] In the method for extracting and sparsely storing parasitic parameters of a layout provided in this application embodiment, the sparse hierarchical adjacency storage structure includes an adjacency graph index layer, a self-network block storage layer, and a coupled edge sparse storage layer; storing the multi-port self-network block and the sparse coupled edge according to the sparse hierarchical adjacency storage structure includes: The adjacency relationships of the arbitrary polygons are stored in the adjacency graph index layer, which uses a compressed sparse row format to store polygon identifiers, neighbor polygon identifiers, and edge types. The multi-port subnet blocks are stored in the subnet block storage layer, and the subnet block storage layer stores the port impedance matrix and capacitance to ground of the multi-port subnet blocks in a compressed column storage format. The sparse coupling edges are stored in the coupling edge sparse storage layer, and the coupling edge sparse storage layer stores the coupling candidate polygon pairs and the coupling capacitance between the coupling candidate polygon pairs in a triplet format.

[0010] The method for extracting and sparsely storing layout parasitic parameters provided in this application embodiment further includes: For each metal layer of each chip layer, a two-dimensional Voronoi diagram is constructed based on the spatial positions of multiple arbitrary polygons, and the Voronoi neighbor relationship between each arbitrary polygon is determined according to the adjacency relationship in the two-dimensional Voronoi diagram. Project the geometric centroid of the arbitrary polygon in the upper metal layer onto the planar coordinate system of the lower metal layer, and determine cross-layer coupling candidate pairs in the Voronoi diagram of the lower metal layer; Calculate the three-dimensional effective spacing of the cross-layer coupling candidate pairs, and extract the coupling parasitic parameters only for cross-layer coupling candidate pairs whose three-dimensional effective spacing is less than a preset threshold to obtain cross-layer sparse coupling edges; The cross-layer sparse coupling edges are stored according to the sparse hierarchical adjacency storage structure.

[0011] The method for extracting and sparsely storing layout parasitic parameters provided in this application embodiment further includes: Arbitrary polygons in the same layer are spatially sorted according to Hilbert curves; The sorted arbitrary polygons are divided into multiple batches, and Ghost Zones are set at the batch boundaries. The Ghost Zones contain Voronoi neighbor arbitrary polygons that cross the batch boundaries. Each batch performs self-network parasitic parameter extraction and coupled parasitic parameter extraction in parallel. Any polygon within the Ghost Zone participates in coupled parasitic parameter extraction but does not participate in self-network parasitic parameter extraction, and is written to the sparse hierarchical adjacency storage structure without lock according to the pre-allocated address.

[0012] In the layout parasitic parameter extraction and sparse storage method provided in this application embodiment, after constructing the port impedance matrix of the arbitrary polygon based on multiple boundary segments using the boundary element method, the method further includes: Calculate at least one of the following: aspect ratio, compactness, concavity, and curvature factor of the arbitrary polygon; Based on the calculation results, local insertion of refinement nodes or constraint Delaunay triangulation is selected to form a hybrid solution domain including boundary element region and refinement region; Impedance matching boundary conditions are applied at the interface of the hybrid solution region, and the refined region is equivalently folded back into the port impedance matrix.

[0013] Secondly, embodiments of this application provide a layout parasitic parameter extraction and sparse storage apparatus, including: The layout parsing module is used to acquire layout data, identify any polygon in the layout data, and determine the electrical function nodes on the arbitrary polygon; The self-network extraction module is used to extract parasitic parameters of the self-network based on the contour of the arbitrary polygon and the electrical function node, taking each arbitrary polygon as a unified electromagnetic entity, to obtain the multi-port self-network block of the arbitrary polygon. The coupling extraction module is used to construct a Voronoi diagram based on the spatial positions of multiple arbitrary polygons, and determine the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi diagram; determine the coupling candidate polygon pair according to the Voronoi neighbor relationship, and extract the coupling parasitic parameter from the coupling candidate polygon pair to obtain sparse coupling edges. A sparse storage module is used to store the multi-port self-network block and the sparse coupling edge according to a sparse hierarchical adjacency storage structure.

[0014] Thirdly, this application provides a storage medium storing multiple instructions that are applicable to a processor for loading to execute the layout parasitic parameter extraction and sparse storage method described in any of the preceding claims.

[0015] Fourthly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the layout parasitic parameter extraction and sparse storage method described in any of the preceding claims.

[0016] In summary, the layout parasitic parameter extraction and sparse storage method provided in this application includes acquiring layout data, identifying arbitrary polygons in the layout data, and determining electrical function nodes on the arbitrary polygons; using each arbitrary polygon as a unified electromagnetic entity, performing self-network parasitic parameter extraction based on the contour of the arbitrary polygon and the electrical function nodes to obtain a multi-port self-network block of the arbitrary polygon; constructing a Voronoi graph based on the spatial positions of multiple arbitrary polygons, and determining the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi graph; determining coupling candidate polygon pairs according to the Voronoi neighbor relationship, and extracting coupling parasitic parameters from the coupling candidate polygon pairs to obtain sparse coupling edges; and storing the multi-port self-network block and the sparse coupling edges according to a sparse hierarchical adjacency storage structure. This application embodiment can simultaneously achieve high-precision extraction of parasitic parameters, fast coupling calculation, and sparse hierarchical storage at the chip-level layout scale. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram illustrating an application scenario of the layout parasitic parameter extraction and sparse storage method provided in the embodiments of this application.

[0019] Figure 2 This is a flowchart illustrating the method for extracting and sparsely storing layout parasitic parameters provided in an embodiment of this application.

[0020] Figure 3 This is a schematic diagram of coupling check comparison provided in the embodiments of this application.

[0021] Figure 4 This is a schematic diagram comparing the extraction principles of directly connected polygonal networks and traditional trapezoidal decomposition.

[0022] Figure 5 This is a schematic diagram of the layout parasitic parameter extraction and sparse storage device provided in the embodiments of this application.

[0023] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0025] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0026] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0027] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.

[0028] In the description of this application, it should be noted that the terms "upper," "lower," "left," "right," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. In addition, terms such as "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] Under advanced processes, three-dimensional effects can no longer be effectively compensated by traditional two-dimensional or 2.5-dimensional approximations. Furthermore, high-precision three-dimensional full-wave methods are difficult to apply to chip-level extraction scenarios with millions of polygons due to the superlinear increase in computational cost with the layout size. This makes it difficult to simultaneously achieve high precision and high speed in parasitic parameter extraction at advanced process nodes and chip-level layout sizes, resulting in a fundamental contradiction between accuracy and efficiency.

[0030] Based on this, embodiments of this application provide a method, apparatus, and storage medium for extracting and sparsely storing layout parasitic parameters. Specifically, the layout parasitic parameter extraction and sparse storage apparatus can be integrated into an electronic device, which can be a server or a terminal, etc. The terminal can include mobile phones, wearable smart devices, tablets, laptops, and personal computers (PCs), etc., and other computer and auxiliary devices. The server can be a single server or a server cluster composed of multiple servers, and can be a physical server or a virtual server.

[0031] For example, such as Figure 1 As shown, the electronic device can acquire layout data, identify arbitrary polygons in the layout data, and determine the electrical function nodes on the arbitrary polygons; using each arbitrary polygon as a unified electromagnetic entity, self-network parasitic parameters are extracted based on the contour of the arbitrary polygon and the electrical function nodes to obtain a multi-port self-network block of the arbitrary polygon; a Voronoi graph is constructed based on the spatial positions of multiple arbitrary polygons, and the Voronoi neighbor relationship between each arbitrary polygon is determined according to the adjacency relationship in the Voronoi graph; coupling candidate polygon pairs are determined according to the Voronoi neighbor relationship, and coupling parasitic parameters are extracted from the coupling candidate polygon pairs to obtain sparse coupling edges; the multi-port self-network block and sparse coupling edges are stored according to a sparse hierarchical adjacency storage structure.

[0032] The technical solutions shown in this application will be described in detail below through specific embodiments. It should be noted that the order of description of the following embodiments is not intended to limit the priority of the embodiments.

[0033] Please see Figure 2 , Figure 2 This is a schematic flowchart illustrating the layout parasitic parameter extraction and sparse storage method provided in an embodiment of this application. The specific flow of this layout parasitic parameter extraction and sparse storage method can be as follows: 101. Obtain layout data, identify any polygon in the layout data, and determine the electrical function nodes on the arbitrary polygon.

[0034] The layout data can be a layout file in GDSII or OASIS format. The arbitrary polygon can include polygons of any shape, such as non-Manhattan structures, rounded polygons, arbitrary angle traces, and planar layers with internal holes. It can also include polygons defined by the BOUNDARY and PATH types in GDSII or OASIS formats.

[0035] In this embodiment, an electrical functional node on an arbitrary polygon refers to a point that satisfies any of the following conditions: a point in contact with a via, a point connected to a device pin, or a junction of three or more electrical segments. Specifically, a point in contact with a via refers to the projected coordinates of the via on the layer containing the arbitrary polygon; a point connected to a device pin refers to the metal lead-out of a standard cell or macrocell; and a junction of three or more electrical segments refers to the geometric intersection of metal line bifurcations.

[0036] In the specific implementation process, it is necessary to remove redundant intermediate nodes on any polygon so that the number of electrical function nodes is less than the number of segments generated by trapezoidal decomposition of any polygon.

[0037] In some embodiments, the outline of an arbitrary polygon can be obtained by parsing a layout file in GDSII or OASIS format; by combining the network, via, and branch connection relationships in the DEF netlist, physical connection ports on the arbitrary polygon can be extracted; the physical connection ports can be classified by function into via landing points, trace branch points, and device pins, and only physical connection ports that are directly connected to the boundary or internal holes of the arbitrary polygon are retained.

[0038] Understandably, the number of electrical function nodes in an arbitrary polygon is typically 3 to 50, with a typical design value of 8 to 24. However, the number of segments generated by decomposing an arbitrary polygon using trapezoidal decomposition can reach hundreds to thousands. Therefore, determining the electrical function nodes and eliminating redundant intermediate nodes can significantly reduce the number of ports subsequently extracted from network parasitic parameters, thereby reducing computational complexity.

[0039] 102. Using each arbitrary polygon as a unified electromagnetic entity, extract the parasitic parameters of the self-network based on the contour and electrical function nodes of the arbitrary polygon to obtain the multi-port self-network block of the arbitrary polygon.

[0040] In this embodiment, each arbitrary polygon can be used as a unified electromagnetic entity. Instead of performing trapezoidal or rectangular decomposition on the arbitrary polygon, the self-network parasitic parameters are extracted directly based on the contour and electrical functional nodes of the arbitrary polygon.

[0041] Specifically, the contour of any polygon can be segmented linearly to obtain multiple boundary segments; using electrical functional nodes as ports, the boundary element method is used to construct the port impedance matrix of any polygon based on multiple boundary segments; and the multi-port self-network block of any polygon is obtained based on the port impedance matrix.

[0042] Piecewise linear discretization of the contour of an arbitrary polygon refers to dividing the contour of the arbitrary polygon into segments, each of which is approximated by a straight line segment, resulting in multiple boundary segments.

[0043] It should be noted that piecewise linear discretization is performed only on the boundaries of arbitrary polygons, without performing volume meshing inside arbitrary polygons. Therefore, the computational complexity is much lower than that of the global finite element method.

[0044] In some embodiments, the step "constructing an arbitrary polygonal port impedance matrix based on multiple boundary segments using the boundary element method with electrical functional nodes as ports" may include: setting each boundary segment as a constant current element; calculating the impedance contribution between any two boundary segments based on the Green's function of a two-dimensional conductive plane to obtain the boundary segment impedance relationship; and assembling the boundary segment impedance relationship into a port impedance matrix with electrical functional nodes as ports according to the correspondence between electrical functional nodes and each boundary segment.

[0045] In this embodiment, any polygon can be considered as a boundary value problem of a two-dimensional conductive plane, whose internal potential distribution φ(r) satisfies the Poisson equation:

[0046] The boundary conditions are: the potentials of the electrical functional nodes are known, and the normal component of the current in any polygonal boundary is zero; these are the Neumann boundary conditions. In scenarios where the number of electrical functional nodes is no more than 100, the boundary element method combined with a two-dimensional conductive plane Green's function is used to construct the port impedance matrix. The two-dimensional conductive plane Green's function is:

[0047] Where t is the thickness of the metal layer, ρ is the resistivity, and r and r' are the position coordinates within the arbitrary polygon.

[0048] In the specific implementation of the boundary element method, the boundary of any polygon is linearly discretized piecewise, with each segment representing a constant current element. The boundary integral equation, after discretization, reduces to a k×k-dimensional linear algebraic system Z·I = V, where Z is the port impedance matrix, I is the current vector, V is the voltage vector, and k is the number of electrical functional nodes. The elements of the port impedance matrix are:

[0049] Where t is the metal layer thickness, ρ is the resistivity, α is the frequency-dependent attenuation coefficient, K0 is the zero-order modified Bessel function, r0 is the characteristic length normalization factor, and r_i and r_j are the position coordinates of the i-th and j-th electrical functional nodes, respectively. These process physical parameters are read layer by layer from the standard ITF interconnect process file.

[0050] It should be noted that the Green's function of a two-dimensional conductive plane physically accurately characterizes the current diffusion effect within an arbitrary polygonal plane, automatically encompassing the current congestion / sparse distribution at the corners of any polygon. Compared to the segmented rectangular method using trapezoidal decomposition, the boundary element method, which constructs the port impedance matrix based on the Green's function of a two-dimensional conductive plane, can automatically capture the current congestion distribution at corners, the diffusion resistance at the edges of arbitrary polygons, and the current flow around internal holes, reducing the resistance extraction error by more than 85% compared to the traditional trapezoidal decomposition method.

[0051] In some embodiments, the step "obtaining a multi-port self-network block of an arbitrary polygon based on the port impedance matrix" may include: calculating the ground capacitance of the arbitrary polygon based on its true contour; and using the port impedance matrix and the ground capacitance together as a multi-port self-network block of the arbitrary polygon, wherein the port impedance matrix is ​​used to characterize the impedance characteristics between electrical functional nodes, and the ground capacitance is used to characterize the capacitance characteristics of the arbitrary polygon to the reference plane.

[0052] Specifically, based on the true contour of arbitrary polygons rather than bounding boxes or piecewise approximations, the ground capacitance of arbitrary polygons can be directly calculated using an improved Sakurai-Tamaru analytical formula combined with shape factor correction:

[0053] Where A is the area of ​​the arbitrary polygon, d is the distance to the reference plane, f_shape is the shape factor correction function of the arbitrary polygon, r_corner is the corner radius, AR is the aspect ratio, and N_vertices is the number of vertices. The shape factor function can be pre-calibrated through a large number of boundary element method numerical experiments and expressed in a closed analytical form, avoiding point-by-point solution.

[0054] In some embodiments, the shape factor function f_shape can be used to comprehensively correct capacitance calculations through analytical expressions of three geometric parameters: corner radius r_corner (0.1 to 2 micrometers) – corner radius suppresses electric field concentration, increasing capacitance by 10% to 30%; aspect ratio AR = width / height (0.2 to 5) – affects the nonlinear correction of edge charge distribution; number of vertices N_vertices (3 to 100) – reflects the complexity of arbitrary polygon contours, with approximately +1.5% capacitance correction for every 10 additional vertices. The calibration method involves theoretical modeling and numerical simulation, using the boundary element method to simulate over 5000 non-Manhattan polygons (including rounded corners, bevels, and irregular holes). A closed analytical expression is constructed using polynomial fitting and orthogonal basis expansion to avoid online iterative solutions. The function form can be as follows: f_shape ≈ 1 + 0.12·(r_corner / w) + 0.08·(1 - e-3 AR) + 0.005·log 10 (N_vertices) In the embodiments of this application, the analytical expression has a fitting accuracy of better than 2% for the boundary element method reference value within the calibration range.

[0055] In some embodiments, when the number of electrical functional nodes exceeds a preset threshold, hierarchical agglomeration clustering can be performed on the electrical functional nodes to merge multiple neighboring electrical functional nodes into a superport, and the equivalent impedance of the superport can be calculated by Schur complement in order to compress the multiport self-network block.

[0056] Specifically, for super-large polygons with more than 100 electrical functional nodes, hierarchical agglomerative clustering is used to merge the electrical functional nodes: based on spatial proximity and impedance distance between electrical functional nodes, multiple neighboring electrical functional nodes are agglomerated into a superport, and the impedance error during the agglomeration process decays at an exponential rate.

[0057] Where D_merge is the cluster merging distance and λ_eff is the effective electromagnetic wavelength, determined by both the skin depth and the plane size. This exponential convergence ensures that even with a significant reduction in the number of ports, the extraction accuracy remains within the engineering requirements, with an impedance error of less than 1%.

[0058] In this embodiment, hierarchical agglomerative clustering uses the impedance distance between electrical functional nodes as a similarity measure: Simultaneously, weighting is applied using spatial Euclidean distance. During merging, spatially adjacent point pairs with strong impedance coupling are prioritized. Each merge generates a superport, whose equivalent impedance is calculated using Schur complement. The effective electromagnetic wavelength parameter in the error convergence model is defined as... , where w is the feature width of any polygon, t is the thickness of the metal layer, typically ε0=0.02, λ_eff≈10 to 50 micrometers.

[0059] In an engineering calculation example, for a power plane with k=200 electrical functional nodes, when it is compressed to k'=12 superports through hierarchical agglomerative clustering, the impedance error is still less than 0.8%, and the calculation speed is increased by 27 times, which fully verifies the exponential convergence characteristics of the error model.

[0060] 103. Construct a Voronoi diagram based on the spatial locations of multiple arbitrary polygons, and determine the Voronoi neighbor relationships between each arbitrary polygon according to the adjacency relationships in the Voronoi diagram.

[0061] Specifically, a Voronoi graph can be constructed using the geometric centroids of multiple arbitrary polygons as seed points and the scan line algorithm. Based on the adjacency relationships in the Voronoi graph, two arbitrary polygons sharing Voronoi edges in the Voronoi graph are identified as Voronoi neighbors, thus obtaining the Voronoi neighbor relationships between each arbitrary polygon.

[0062] The scanline algorithm is the Fortune scanline algorithm, which has a complexity of O(N log N) for constructing the Voronoi diagram, where N is the number of arbitrary polygons.

[0063] The properties of the Voronoi graph guarantee that if two arbitrary polygons do not share Voronoi edges (i.e., are not Voronoi neighbors), then the distance between the two arbitrary polygons is at least greater than their respective distances to other arbitrary polygons, and the coupling capacitance decreases to a negligible order of magnitude. The average number of Voronoi neighbors is constant, approximately 6 in two dimensions. Therefore, the total number of coupling checks decreases from C(N,2) to approximately 6N, and the complexity decreases from O(N^2) to O(N^2). 2 ) decreased to O(N).

[0064] Please see Figure 3 , Figure 3 The right side shows the Voronoi neighborhood filtering O(N) of this application embodiment, which checks only 30 pairs of Voronoi neighbors, averaging 6 per arbitrary polygon. Figure 3 The left side is the traditional O(N) 2 The double-loop method checks 190 pairs of data. Therefore, the Voronoi space partitioning provided in this application can reduce coupling checks from O(N) to O(N) pairs. 2 The value is reduced to O(N·k), where k is the average number of neighbors in the Voronoi neighborhood.

[0065] 104. Determine the candidate polygon pairs for coupling based on Voronoi neighbor relationships, and extract the parasitic coupling parameters from the candidate polygon pairs to obtain sparse coupling edges.

[0066] Specifically, two arbitrary polygons with Voronoi neighbor relationships can be identified as a pair of coupling candidate polygons; for a pair of coupling candidate polygons, the local geometry and relative angle of the nearest edge between the two arbitrary polygons are obtained; based on the local geometry and relative angle, shape-aware coupling calculation is performed on the pair of coupling candidate polygons to obtain the coupling capacitance between the pair of coupling candidate polygons; the pair of coupling candidate polygons and the coupling capacitance are used as sparse coupling edges.

[0067] The local geometry includes at least one of the following: the side length of the nearest side between two arbitrary polygons, the spacing, the equivalent line width, and the projected overlap length; the relative angle is the angle between the nearest sides between the two arbitrary polygons.

[0068] In some embodiments, for a Voronoi neighbor polygon pair, the local geometry of the nearest edge between the two arbitrary polygons is taken, which may include edge lengths L1 and L2, spacing S, and relative angle θ, and can be achieved using an improved parallel line coupling capacitance formula:

[0069] Where L_eff is the effective coupling length, taken as the smaller of the projected lengths of the nearest edges of the two arbitrary polygons; W_eff is the equivalent linewidth; g(·) is the analytical coupling function of the spacing-linewidth ratio; and f_overlap is the projection overlap factor, considering the coupling attenuation when the edges of the two arbitrary polygons are not perfectly parallel. For parallel edges θ<15°, full coupling calculation is used; for near-orthogonal edges θ>75°, only the capacitance at the intersection point is calculated. Orthogonal coupling is usually less than 5% of parallel coupling, and the intermediate angle is linearly interpolated.

[0070] In some embodiments, for distant polygon pairs that do not have Voronoi neighbor relationships, they can be grouped into far-field clusters according to the space-filling curve, and far-field coupling can be calculated in an inter-cluster aggregation manner, instead of calculating the coupling between distant polygon pairs one by one.

[0071] In the implementation process, for distant polygon pairs not directly connected by Voronoi neighbors, a fast multipole-inspired hierarchical aggregation scheme is adopted instead of pairwise calculation. Specifically, non-neighbor polygons can be grouped into far-field clusters according to their Hilbert space-filling curves, and each cluster's influence on the external electric field is approximated by an equivalent multipole moment. The inter-cluster capacitance is calculated in a cluster-to-cluster aggregation manner, with the accuracy loss controlled to within 0.5%.

[0072] Understandably, identifying two arbitrary polygons with Voronoi neighbor relationships as a pair of candidate coupling polygons, and extracting coupling parasitic parameters only from these candidate polygon pairs, can reduce the complexity of coupling capacitance extraction from the traditional O(N) time complexity. 2 The computation time is reduced to O(N log N), making the chip-level (N ≥ 10) more efficient. 6 The coupling extraction of ) becomes practically feasible.

[0073] 105. Store multi-port self-network blocks and sparsely coupled edges according to a sparse hierarchical adjacency storage structure.

[0074] In this embodiment, the sparse hierarchical adjacency storage structure includes an adjacency graph index layer, a self-network block storage layer, and a coupled edge sparse storage layer.

[0075] In the specific implementation process, the adjacency relationship of any polygon can be stored in the adjacency graph index layer. The adjacency graph index layer uses a compressed sparse row format to store polygon identifiers, neighbor polygon identifiers, and edge types. Multi-port self-network blocks are stored in the self-network block storage layer. The self-network block storage layer uses a compressed column storage format to store the port impedance matrix and ground capacitance of the multi-port self-network blocks. Sparse coupled edges are stored in the coupled edge sparse storage layer. The coupled edge sparse storage layer uses a triplet format to store coupled candidate polygon pairs and the coupling capacitance between coupled candidate polygon pairs.

[0076] The adjacency graph index layer uses a compressed sparse row format to store the connection graph between arbitrary polygons. `row_ptr[i]` points to the starting position of the neighbor list for arbitrary polygon `i`, `col_ind` stores the identifier of the neighboring arbitrary polygon, and `data` stores the physical type and precision level of the connecting edges. Physical types include vias, branches, and couplings. The adjacency graph index layer supports O(1) arbitrary polygon neighbor queries and occupies O(N+M) storage space, where M is the number of connecting edges.

[0077] Within the network block storage layer, the port impedance matrix of each arbitrary polygon is stored independently as a data block in a compressed columnar storage format. For non-planar shapes, the port impedance matrix is ​​simplified to scalar resistance values; for planar polygons, the port impedance matrix is ​​stored in k×k dense blocks, but k is usually no greater than 50, so the storage size of a single block is extremely small, no more than 50×50×8 bytes = 20KB.

[0078] The coupled edge sparse storage layer stores only the coupling capacitance values ​​between Voronoi neighbor polygon pairs, one value per edge, stored in coordinate format as a triple (poly_i, poly_j, C_coup). The total number of coupled edges is approximately on the order of 6N, far smaller than the tens to hundreds of millions of coupled element entries in traditional SPEF.

[0079] Understandably, in traditional SPEF, if a polygonal planar layer is decomposed into S segments, then the resistance term occupies S records, the capacitance to ground occupies S records, and the inter-segment coupling capacitance occupies O(S) records. 2 In a sparse hierarchical adjacency storage structure, the same arbitrary polygon occupies only: 1 adjacency graph index layer entry + 1 k×k self-network block + k adjacency coupling edges, totaling O(k... 2 ) rather than O(S 2 When k is approximately 20 and S is approximately 2000, the compression ratio can reach (2000). 2 ) / (20 2 It is approximately equal to 10000:1.

[0080] In some embodiments, the adjacency graph index layer supports constant-time neighbor lookups based on polygon identifiers; the self-network block storage layer supports address block retrieval based on polygon identifiers; and the coupled edge sparse storage layer supports traversing the coupled edges of the polygon based on its polygon identifier. Therefore, querying parasitic parameters of any polygon only requires O(1) indexing, O(1) block retrieval, and O(k) edge traversal, without requiring complete file parsing throughout the process.

[0081] In the embodiments of this application, based on the steps of "constructing a Voronoi graph based on the spatial location of multiple arbitrary polygons and determining the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi graph" and "determining the coupling candidate polygon pair according to the Voronoi neighbor relationship and extracting the coupling parasitic parameters of the coupling candidate polygon pair to obtain sparse coupling edges", the layout parasitic parameter extraction and sparse storage method may also include three-dimensional integrated circuit extension.

[0082] Specifically, a two-dimensional Voronoi diagram can be independently constructed for each metal layer of each chip layer to obtain the Voronoi neighbor relationships within each layer; the geometric centroid of the polygon in the upper metal layer is projected onto the planar coordinate system of the lower metal layer, and cross-layer coupling candidate pairs are determined in the Voronoi diagram of the lower metal layer; the three-dimensional effective spacing of the cross-layer coupling candidate pairs is calculated, and cross-layer coupling parasitic parameters are extracted only for cross-layer coupling candidate pairs whose three-dimensional effective spacing is less than a preset threshold; virtual ports are established for through-silicon vias, and the virtual ports include an equivalent circuit composed of longitudinal resistance, oxide layer capacitance, and depletion layer capacitance, and the relevant parameters of the virtual ports are stored in a sparse hierarchical adjacency storage structure.

[0083] In some embodiments, for each metal layer of each chip layer, a two-dimensional Voronoi diagram is independently constructed. The scan-line algorithm can be used to generate Voronoi neighbor relationships within each layer, which are then used to extract the coupling capacitance between arbitrary polygons in the same layer. For vertically adjacent chip layers, a cross-layer Voronoi overlay diagram is constructed. The geometric centroids of all arbitrary polygons in the upper metal layer are vertically projected onto the planar coordinate system of the lower metal layer, and the positions of the projection points are marked in the lower Voronoi diagram. In the lower Voronoi diagram, arbitrary polygons in the lower layer that share Voronoi edges with the projection points are marked as cross-layer coupling candidate pairs. For cross-layer coupling candidate pairs, the three-dimensional effective spacing S_eff = sqrt(S_xy) is calculated. 2 + S_z 2 ), where S_xy is the horizontal projection spacing, and S_z is the total thickness of the dielectric between adjacent chip layers, including the interlayer dielectric layer, silicon substrate, and bonding interface layer; subsequent coupling capacitance calculation is activated only for cross-layer coupling candidate pairs whose three-dimensional effective spacing S_eff is less than a preset threshold.

[0084] In this embodiment, each independent two-dimensional Voronoi diagram is an M-time O(N log N) operation, while projection and threshold filtering are O(M×N) linear scan operations. Since in actual 3D integrated circuit design, M is typically much smaller than N (M < 10) and N > 10. 8 The overall computational complexity of layered Voronoi comprehensions remains on the order of O(N log N), maintaining the complexity advantage of Voronoi comprehensions in single-layer scenes. This is significantly better than directly performing O((M×N)) computation on all M×N arbitrary polygons. 2 The full pair coupling check of the layered Voronoi diagram reduces the computational cost of the three-dimensional integrated circuit scene by about M×N / k times, where k is the average number of Voronoi neighbors of each arbitrary polygon, which is about 6 to 8. This allows the coupling extraction of chip-level three-dimensional integrated circuits to be completed within a practically feasible timeframe for the first time.

[0085] Specifically, the longitudinal parasitic resistance of a through-silicon via (TSV), which serves as a vertical conductive pillar connecting metals of different chip layers in a three-dimensional integrated circuit, can be determined by the resistivity and geometry of the metal filling material, and can be calculated analytically using the following formula:

[0086] Where ρ_Cu is the resistivity of the copper filler, H is the vertical height of the through-silicon via (TSV), and r is the radius of the TSV metal pillar. The parasitic capacitance of the TSV sidewall consists of two series components: the oxide layer capacitance and the silicon depletion layer capacitance.

[0087]

[0088] Where ε_ox is the dielectric constant of the sidewall oxide layer, ε_si is the dielectric constant of the silicon substrate, r_ox is the outer diameter of the oxide layer, and r_dep is the outer diameter of the silicon depletion layer. The typical range of values ​​for these parameters is consistent with the measured data for silicon at process nodes from 28 nm to 16 nm.

[0089] In some embodiments, the aforementioned through-silicon via (TSV) parasitic model can be injected as an independent virtual port into a sparse hierarchical adjacency storage structure. Specifically, the virtual port carries a π-type equivalent circuit, where R_TSV is connected in series between C_ox and C_dep; a cross-layer neighbor pointer type is added to the adjacency graph index layer, enabling ports of arbitrary polygons in the upper layer metal to establish connections with their corresponding ports of arbitrary polygons in the lower layer metal through this TSV virtual port; an independent storage block is allocated to each TSV virtual port from the network block storage layer to record the lumped parameter values ​​of the π-type circuit above it. This extension method requires no modification to the boundary element solver, only an extension to the port type enumeration of the adjacency graph index layer and the storage block definition of the network block storage layer.

[0090] In three-dimensional integrated circuit stacked structures, the vertical spacing between the top metal layers of the upper and lower chips is typically only a few micrometers, with multiple non-uniform dielectric layers consisting of interlayer dielectric layers, silicon substrates, and hybrid bonding interfaces. The coupling capacitance between these metal layers cannot be directly solved using the Green's function for a single-layer uniform dielectric. Therefore, the Green's function is generalized from a single-dielectric form to a layered dielectric form:

[0091] Where J0(.) is the zeroth-order Bessel function of the first kind, Let ε_eff(k_ρ) be the transverse wavenumber component, and let ε_eff(k_ρ) be the equivalent dielectric constant of the multilayer dielectric, determined by the dielectric constant ε_i and thickness d_i of each layer through Sommerfeld integration. Preferably, the Sommerfeld integration is pre-calculated using the discrete complex mirror method as a closed analytical expression independent of specific geometric parameters, thereby reducing the computational complexity of coupling capacitance extraction from O(N) of the full three-dimensional finite element method. 3 The computation time is reduced to O(N log N), and no 3D volume mesh generation is required.

[0092] In engineering implementation, the layered dielectric Green's function is only enabled for arbitrary polygon pairs across chips with vertical spacing less than a preset threshold; arbitrary polygon pairs within the same metal layer and under far-field conditions still use the single-layer Green's function G(r, r'). The boundary integral of the layered dielectric Green's function is seamlessly compatible with the boundary element discretization framework. When constructing the port impedance matrix, it is only necessary to replace the calculation of matrix elements from the single-dielectric Green's function to the layered dielectric Green's function, while the structural dimensions of the matrix and the subsequent port compression and ground capacitance calculation steps remain completely unchanged.

[0093] In some embodiments, the sparse hierarchical adjacency storage structure can be three-dimensionally extended. Specifically, chip layer identifiers, metal layer identifiers, and edge types such as cross-layer coupling or through-silicon via (TSV) port types can be added to the adjacency graph index layer; TSV virtual port storage blocks can be added to the self-network block storage layer, and the TSV virtual port storage blocks can be distinguished from polygonal self-network blocks by block type identifiers; coupling type identifiers, three-dimensional effective spacing, and source / target chip layer identifiers can be added to the coupled edge sparse storage layer.

[0094] The three-dimensional extension of the adjacency graph index layer can be as follows: In the compressed sparse row structure of the original adjacency graph index layer, a cross-layer neighbor pointer type is added to the neighbor list of each arbitrary polygon entry. The edge type enumeration of the data field in the original adjacency graph index layer is expanded from via, branch, and coupling to via, branch, coupling, cross-layer coupling, and silicon via port; the newly added cross-layer coupling type indicates that the arbitrary polygon of the neighbor is located on an adjacent chip layer, and its coupling relationship is generated by layered Voronoi; the newly added silicon via port type indicates that the neighbor is a silicon via virtual port, and its equivalent circuit parameters are stored in the silicon via block of the self-network block storage layer. At the same time, a chip layer identifier field and a metal layer identifier field are added to the index structure of each adjacency graph index layer entry, so that downstream analysis tools can directly locate any polygon of any metal layer in the stacked chip through the triplet of chip layer identifier, metal layer identifier, and polygon identifier. The extension maintains the O(1) query characteristics of the compressed sparse row and only increases the storage overhead of each entry by about 8 bytes.

[0095] The 3D extension of the self-network block storage layer can be achieved by adding a To-Silicon Via (TSV) virtual port storage block type to each arbitrary polygon self-network block in the original self-network block storage layer. The TSV storage block contains the following fields: TSV geometry parameters, including vertical height, metal radius, oxide outer diameter, and depletion layer outer diameter; lumped circuit parameters, including vertical resistance, oxide capacitance, and depletion layer capacitance; and connectivity relationships, including upper-layer port identifier and lower-layer port identifier, both of which are triplet combinations of chip layer identifier, metal layer identifier, and polygon identifier. The format of the TSV storage block is consistent with the compressed column storage format of the arbitrary polygon self-network block, distinguished only by the block type identifier field. The block type identifier can be either polygon or TSV, allowing downstream tools to identify and parse the TSV parameters on the first read without modifying the block reading logic.

[0096] The three-dimensional extension of the coupled-edge sparse memory layer can be achieved by adding the following three-dimensional extended fields to the original coupled-edge sparse memory layer triples: a coupling type identifier, with values ​​for same-layer, neighboring-layer, and cross-chip-layer; and a three-dimensional effective spacing S_eff = sqrt(S_xy). 2 + S_z 2 The data structure is defined as follows: S_xy represents the horizontal projection spacing, and S_z represents the total thickness of the interlayer medium. The source and target chip layer identifiers identify the respective chip layers to which the coupled parties belong. The extended fields are compatible with the original triplet format. For same-layer coupling edges, the source chip layer identifier equals the target chip layer identifier, and the coupling type identifier is "same layer," simplifying the three-dimensional effective spacing to the horizontal projection spacing. For cross-layer coupling edges, all the above fields are filled in completely. The coupling type identifier allows downstream analysis tools to selectively load data of specific coupling types through simple filtering conditions, avoiding blind traversal of the entire cross-layer data.

[0097] Regarding storage requirements, the additional overhead from the 3D expansion is analyzed as follows: Adjacency graph index layer expansion adds two integer fields (chip layer identifier and metal layer identifier) ​​totaling 8 bytes for each arbitrary polygon, along with a small number of cross-layer or through-silicon via (TSV) edge pointer entries. The number of these entries is proportional to the number of TSVs and cross-layer Voronoi neighbors, typically 3% to 10% of the total number of arbitrary polygons, increasing the total storage of the adjacency graph index layer by approximately 5% to 15%. From the network block storage layer expansion, each TSV virtual port adds a storage block of approximately 200 bytes, with the total number of TSVs typically being 10. 4 Up to 10 5 The increase in total storage capacity from the network block storage layer is approximately 2 to 20 megabytes, negligible at the chip-level storage scale. The expansion of the coupled edge sparse storage layer adds approximately 20 bytes to each cross-layer coupled edge, including coupling type identifier, 3D effective spacing, and source / target chip layer identifiers. The total number of cross-layer coupled edges is approximately 5% to 15% of that of same-layer coupled edges, increasing the total storage capacity of the coupled edge sparse storage layer by approximately 5% to 10%. In summary, the total storage capacity of the sparse hierarchical adjacency storage structure in a 3D integrated circuit scenario increases by approximately 10% to 25% compared to a 2D single-chip scenario, but is still far lower than the storage requirements of the traditional standard parasitic exchange format in a 3D integrated circuit scenario, where the latter suffers from O((M×N)) due to the total number of arbitrary polygons M×N. 2 ) Magnitude expansion.

[0098] In terms of compatibility, 3D extensions are implemented by appending fields rather than modifying them. For single-chip designs without 3D integrated circuit structures, the newly added fields are all empty or have default values, and the parser of the sparse hierarchical adjacency storage structure file can read the 2D single-chip data without any modification. For multi-chip designs with 3D integrated circuit structures, the parser automatically activates the parsing branch for cross-layer coupled data after reading a non-default value 3D extension field. This backward-compatible design ensures a unified data interface for the sparse hierarchical adjacency storage structure format in different design scenarios.

[0099] The layout parasitic parameter extraction and sparse storage method provided in this application embodiment also includes an adaptive polygon mesh refinement step. Specifically, based on the step of "using electrical function nodes as ports and constructing port impedance matrices of arbitrary polygons based on multiple boundary segments using the boundary element method", at least one of the aspect ratio, compactness, concavity, and curvature factor of the arbitrary polygon can be calculated; according to the calculation results, local insertion of refinement nodes or constraint Delaunay triangulation is selected to form a hybrid solution region including the boundary element region and the refinement region; impedance matching boundary conditions are applied on the interface of the hybrid solution region, and the refinement region is equivalently folded back into the port impedance matrix.

[0100] In practical implementation, for each arbitrary polygon, the following characteristics can be calculated: aspect ratio (the aspect ratio of the smallest bounding rectangle); compactness (4π·area / perimeter). 2 Circles are equal to 1, elongated shapes are approximately equal to 0; concavity is the area of ​​the convex hull / the actual area minus 1, convex polygons are equal to 0; curvature factor is the minimum radius at the corner / the equivalent radius of any polygon.

[0101] If the aspect ratio is less than 5 and the compactness is greater than 0.6, i.e., a square or circular planar layer, the boundary element method is used. No internal mesh is needed; the solution is obtained directly using the Green's function, with a computational complexity of O(k). 2 If the aspect ratio is greater than or equal to 5 or the compactness is less than 0.3, i.e., long strips or narrow neck regions, the minimum number of refined nodes are automatically inserted at the narrow neck. Based on the zero point position of the second derivative of the width variation, only the local area where the current density gradient exceeds the threshold is refined, while other areas maintain the boundary element mode. If the concavity is greater than 0.3, i.e., highly concave, the concave polygon is subjected to controlled meshing using constrained Delaunay triangulation. The node layout is driven by the central axis of the shape rather than uniform meshing. The mesh is automatically densified at concave corners and automatically sparsed at flat boundaries.

[0102] At the interface between the boundary element region and the finite element region, the solutions of the two regions are coupled by Robin boundary conditions, i.e., impedance matching conditions, to avoid iteration. Finally, the internal network of the refined node in the finite element region is folded back into the multi-port self-network block in the form of an equivalent impedance matrix, so that the self-network block still maintains its compressed form.

[0103] The 3D integrated circuit extension of the adaptive polygon mesh refinement step adds the following exclusive features in three dimensions in addition to the original four basic geometric feature calculations: 3D aspect ratio adaptation, expanding the original two-dimensional minimum bounding rectangle to a three-dimensional minimum bounding box, adding an aspect ratio dimension to identify high-depth structures such as through-silicon vias (TSVs) and micro-bumps; 3D compactness adaptation, replacing the original two-dimensional area with the side surface area of ​​the corresponding metal layer of any polygon, and replacing the perimeter with the closed perimeter of the three-dimensional contour, correcting the compactness calculation deviation caused by three-dimensional stacking; 3D concavity adaptation, adding interlayer projection concavity calculation, statistically analyzing the concavity ratio of the overlapping area of ​​the projections of adjacent metal layers, and identifying hidden concavity areas caused by interlayer offset; 3D curvature factor adaptation, adding curvature calculation for the sidewalls of TSVs and the spherical cap surfaces of micro-bumps, expanding the original two-dimensional corner curvature to three-dimensional surface curvature, capturing three-dimensional high curvature abrupt change locations. Furthermore, three new 3D integrated circuit-specific geometric features are added: interlayer overlap, proximity distance of through-silicon vias, and bonding point offset, as 3D-specific input parameters for subsequent refinement decisions.

[0104] Based on the newly added 3D-specific geometric features, the three branch decision logics can be extended in 3D. The planar layer branch targets large-area power or ground plane layers in 3D integrated circuits. It automatically loads the layered dielectric coupling Green's function in the boundary element mode, performing a one-time parallel solution for cross-layer parasitic parameters, while maintaining a computational complexity of O(k²). The narrow neck region branch targets narrow neck traces around vias. Building upon the original zero-point location of the second derivative of the width change, it adds a via current density gradient threshold judgment. It performs local refinement only on the area within a preset range around the via, while retaining the boundary element mode in other areas, thus reducing the refinement range by approximately 70% or more compared to global refinement. The highly concave polygon branch targets densely concave regions formed by via arrays and micro-bump arrays. It employs central axis-driven constrained Delaunay triangulation, automatically refining the mesh at the via sidewalls and bonding concave corners, and automatically sparses the mesh in large-area flat dielectric regions, avoiding mesh explosion in the array region. Furthermore, a new dedicated refinement branch for through-silicon vias (TSVs) has been added. When a cylindrical structure of a TSV is identified, it automatically switches to a semi-analytical refinement strategy in cylindrical coordinates, generating a radial refinement layer only for the area within a preset range of the TSV sidewall.

[0105] The hybrid boundary element method (BEM) and finite element method (FEA) for 3D integrated circuits extends the original 2D interface coupling to 3D interlayer interface coupling. Impedance-matched Robin boundary conditions are simultaneously applied to the overlapping surface of the BEM large region and the finite element refined region, performing one-time, non-iterative coupling on the multi-layer solutions, avoiding convergence errors that may be introduced by cross-layer iteration. Layered Schur complement elimination is performed on the admittance matrices of all internal nodes in the finite element refined region, eliminating all non-port nodes within each layer. The multi-layer equivalent impedance matrices are uniformly folded back into multi-port self-network blocks, and the final output self-network blocks maintain a compressed sparse format, avoiding dimensional expansion due to 3D stacking. In the folded port network, cross-layer parasitic coupling terms between vias and bonding points are automatically retained, requiring no additional post-processing steps, directly outputting a complete parasitic netlist suitable for 3D integrated circuit post-simulation. In an engineering simulation example, this extended mechanism improves overall computational efficiency by more than 6 times compared to traditional global refinement schemes, while the parasitic extraction accuracy deviation of key 3D structures such as vias and micro-bumps can be controlled within 2%.

[0106] In some embodiments, based on the steps of "taking each arbitrary polygon as a unified electromagnetic entity, extracting self-network parasitic parameters based on the contour and electrical function nodes of the arbitrary polygon to obtain a multi-port self-network block of the arbitrary polygon" and "determining the coupling candidate polygon pairs according to the Voronoi neighbor relationship, and extracting coupling parasitic parameters from the coupling candidate polygon pairs to obtain sparse coupling edges", the layout parasitic parameter extraction and sparse storage method may also include a parallel polygon batch processing step.

[0107] In the specific implementation process, arbitrary polygons in the same layer can be spatially sorted according to the Hilbert curve; the sorted arbitrary polygons are divided into multiple batches, and Ghost Zones are set at the batch boundaries. The Ghost Zones contain arbitrary polygons that are Voronoi neighbors across the batch boundaries; each batch performs self-network parasitic parameter extraction and coupled parasitic parameter extraction in parallel. The arbitrary polygons in the Ghost Zone participate in coupled parasitic parameter extraction but do not participate in self-network parasitic parameter extraction, and are written to the sparse hierarchical adjacency storage structure without lock according to the pre-allocated address.

[0108] Specifically, for all arbitrary polygons in the same layer, a linear order on the Hilbert space-filling curve can be calculated using their centroid coordinates. The Hilbert curve has optimal spatial locality preservation properties; any polygon adjacent on the Hilbert sequence is also adjacent in two-dimensional space. The sorting complexity is O(N log N), using bitwise operations and eliminating the need for floating-point comparisons; theoretically, it is about 3 to 5 times faster than quicksort.

[0109] The Hilbert-sorted list of arbitrary polygons is uniformly divided into P batches, where P is the number of threads. Each batch performs self-network parasitic parameter extraction independently and in parallel. To address the issue of Voronoi neighbor relationships at batch boundaries being segmented by the batches, each batch extends to both sides by a Ghost Zone, containing arbitrary polygons that are Voronoi neighbors crossing the batch boundary, averaging approximately 6 per boundary. Arbitrary polygons within the Ghost Zone participate in the batch's coupling capacitance calculation but not in the self-network parasitic parameter extraction, avoiding duplication and ensuring that cross-batch coupling capacitance is neither omitted nor calculated repeatedly.

[0110] After each batch of extraction is completed, the sparse hierarchical adjacency storage structure naturally supports lock-free parallel writing. The adjacency graph index layer pre-allocates index intervals by batch, the self-network block storage layer determines block addresses using arbitrary polygon identifiers, and the coupled edge sparse storage layer entries are written using Voronoi edge identifiers. The write operations of the three layers do not overlap, requiring no locks, atomic operations, or memory barriers, achieving true parallel writing. On a 32-core processor, the theoretical speedup is close to 30 times.

[0111] In some embodiments, the method for extracting and sparsely storing parasitic parameters of a graph further includes mapping a sparse hierarchical adjacency storage structure to an adjacency matrix, node feature matrix, and edge feature matrix of a graph neural network, for generating artificial intelligence training data for parasitic parameter prediction.

[0112] Specifically, the adjacency graph index layer directly corresponds to the adjacency matrix in the graph neural network, the sub-network block storage layer directly corresponds to the node feature matrix in the graph neural network, the node feature matrix includes port impedance matrix features and ground capacitance features, and the coupled edge sparse storage layer directly corresponds to the edge feature matrix in the graph neural network. The three layers of data can be input into mainstream graph neural network models such as graph convolutional networks, graph attention networks, and GraphSAGE without any format conversion or feature engineering, eliminating the intermediate step from traditional SPEF to feature engineering.

[0113] The technical solution shown in this application will be further illustrated by specific application examples below.

[0114] The first application example is a simple rectangular trace. CLK_NET is an M4 layer copper clock network trace, composed of three rectangular polygons, P1, P2, and P3, with a total length of 120 micrometers, a width of 2 micrometers, and areas of 100 square micrometers, 240 square micrometers, and 160 square micrometers respectively. The adjacent layers above and below are GND_GRID and VDD_GRID meshes, with a spacing of 0.5 micrometers and a dielectric constant of 3.2. The adjacent meshes on the same layer are SIG_A (minimum spacing of 0.8 micrometers, parallel length of 30 micrometers) and SIG_B (minimum spacing of 1.2 micrometers, parallel length of 20 micrometers).

[0115] In the extraction of directly connected polygon networks, three arbitrary polygons, P1, P2, and P3, are each identified as a connection port, constructing a 3×3 adjacency matrix of arbitrary polygons. Traditional trapezoidal decomposition requires dividing three rectangular segments into approximately 15 trapezoidal segments, each a node, forming a 15-dimensional matrix. Directly connected polygon network extraction uses arbitrary polygons as nodes, requiring only a 3-dimensional matrix for solution, reducing the number of nodes by a factor of 5. The resistance of segment P1 to P2 is 0.54 ohms, the resistance of segment P2 to P3 is 0.46 ohms, and the total resistance of P1 to P3 is 1.00 ohms, for a total extraction resistance of 1.38 ohms. This includes the resistance between nodes P1 and P2 (0.54 ohms), the resistance between nodes P2 and P3 (0.46 ohms), and the self-resistance of P3 (including corner correction of 0.38 ohms), totaling 1.38 ohms. Two corner points are identified, with a corner resistance correction of 0.092 ohms.

[0116] Figure 4 This diagram illustrates the comparison between the extraction principles of directly connected polygonal networks and traditional trapezoidal decomposition. The left side shows the traditional trapezoidal decomposition method, and the right side shows the extraction of directly connected polygonal networks. The traditional trapezoidal decomposition method decomposes the polygon into 37 nodes and 36 internal resistance branches, resulting in a 37×37 matrix with 1369 elements; the directly connected polygonal network extraction results in 5 connection points, a 5×5 matrix, and direct solution of the Green's function impedance, resulting in a 5×5 matrix with 25 elements.

[0117] In the Voronoi shape-aware ground capacitance extraction, the entire 120-micrometer length of CLK_NET was meshed using Voronoi grids, generating 120 Voronoi elements. Green's function boundary integrals calculated the total ground capacitance to be 12.5 femtofarads, with a capacitance per unit length of 0.104 femtofarads per micrometer. Voronoi neighborhood analysis identified two coupled neighbors in the same layer: a coupling capacitance of 2.8 femtofarads with SIG_A (0.8 micrometer spacing, 30 micrometer parallel); and a coupling capacitance of 1.9 femtofarads with SIG_B (1.2 micrometer spacing, 20 micrometer parallel). The total coupling capacitance was 4.7 femtofarads, and the total capacitance was 17.2 femtofarads. The self-inductance was 85 picohens, and the mutual inductances were 12 picohens and 8 picohens, respectively.

[0118] In the sparse hierarchical adjacency storage write, the adjacency graph index layer writes the adjacency index, where P1 points to P2 and SIG_A, P2 points to P1, P3, and SIG_B, and P3 points to P2; the sub-network block storage layer writes each arbitrary polygon sub-network block, with R / P1 equal to 0.54 ohms and C / P1 equal to 4.2 fems; R / P2 equal to 0.46 ohms and C / P2 equal to 5.1 fems; R / P3 equal to 0.38 ohms and C / P3 equal to 3.2 fems; the coupled edge sparse storage layer writes the coupled edges, with 2.8 fems between P1 and SIG_A and 1.9 fems between P2 and SIG_B. The sparsity ratio is 0.15.

[0119] Tests showed that Voronoi integral extraction takes 0.5 milliseconds, compared to 8,500 milliseconds for the finite element method or the boundary element method dense matrix method, representing a speedup of 17,000 times. The sparse hierarchical adjacency storage structure with three layers of storage occupies 0.30 megabytes of memory, compared to 120 megabytes for the finite element method, saving 400 times the memory.

[0120] The second application example is an irregularly shaped power plane. VDD_CORE_PLANE is an M5 layer copper power plane network composed of four arbitrary polygons: MAIN_PLANE (main plane, area 34,200 square micrometers, perimeter 1,020 micrometers, 10 vertices); HOLE1 (via opening, 20 micrometers × 20 micrometers square, area 400 square micrometers); HOLE2 (through-silicon via, 30 micrometers × 30 micrometers square, area 900 square micrometers); and TAB1 (connector, area 900 square micrometers, 4 vertices). In total, there are 30 vertices, an effective area of ​​33,800 square micrometers, and 15 edges, including 11 straight edges and 4 beveled corners. The material properties are copper conductivity 5.8 × 10⁻⁶. 7 Siemens' resistance per meter is 0.012 ohms per cubic meter for thin films.

[0121] In the extraction of directly connected polygonal networks, the irregular plane and its openings and fin structures are identified holistically, and a 15-sided Voronoi adaptive mesh is constructed. Traditional trapezoidal decomposition requires dividing such irregular polygons into hundreds of trapezoidal segments, while the extraction of directly connected polygonal networks directly utilizes the Voronoi centroid method and edge contribution integral formula. At the Voronoi element level, with 850 elements, the resistance matrix is ​​calculated as follows: equivalent resistance 0.015 ohms in the NW to SE direction, 0.012 ohms in the N to S direction, and 0.018 ohms in the E to W direction, with an effective thin-layer resistance of 0.0135 ohms per square. The opening correction factor is 1.045, taking into account the current flow effect around the two vias or through-silicon via regions, and the independent resistance of the connecting fins is 0.0032 ohms. The sparse matrix has only 48 non-zero elements, with a compression ratio of 0.031, which is far lower than the thousands of non-zero elements in a full mesh mesh.

[0122] In the Voronoi edge contribution integral capacitance extraction, the Voronoi partition generates 850 cells, and a closed-form integral of the edge contribution is applied to each Voronoi edge. The total planar capacitance to ground is 2850 femtofarads, the areal density is 0.083 femtofarads per square micrometer, the additional capacitance due to the aperture edge effect is 45 femtofarads, and the edge electric field enhancement effect of two apertures results in a total edge capacitance of 320 femtofarads. The total extraction time is 15.2 milliseconds, of which Voronoi partitioning accounts for approximately 60%, edge integration accounts for approximately 30%, and serialization of the adjacency graph index layer, self-network block storage layer, and coupled edge sparse storage layer accounts for approximately 10%.

[0123] In the sparse hierarchical adjacency storage write, the adjacency graph index layer stores the adjacency graphs where MAIN_PLANE points to HOLE1, HOLE2, and TAB1; the sub-network block storage layer stores the k×k port impedance matrix blocks and their ground capacitance values ​​according to the four identifiers MAIN_PLANE, HOLE1, HOLE2, and TAB1 respectively. The number of connection ports k in the main plane is 20, the sub-network block has 400 elements, and each of the three holes or wing structures has k equal to 4, with 16 elements each, totaling 448 elements for the four sub-network blocks; the coupling edge sparse storage layer stores the coupling capacitance triplets on the Voronoi boundaries. After compressed column storage, only 48 non-zero elements are actually stored, resulting in an equivalent dense matrix of approximately 1548 elements, with a sparse compression ratio of 0.031.

[0124] Testing showed that Voronoi integration extraction took 15.2 milliseconds, compared to 45,000 milliseconds for the finite element method or boundary element method, representing a speedup of 2,961 times; memory usage was 2.50 megabytes, compared to 520 megabytes for the finite element method, saving 208 times. This sample had the most complex geometry among the five groups, containing internal openings and fins. The number of Voronoi subdivision elements (850) was significantly more than the 120 for simple rectangular traces, but the complexity still maintained a linear growth of O(n·k).

[0125] The third application example is five sets of sample test data. To verify the effectiveness of the layout parasitic parameter extraction and sparse storage method on actual chip layouts, five representative arbitrary polygon samples were selected for testing. The differences in extraction speed and memory usage between the traditional finite element method or boundary element method dense matrix method and the Voronoi integral method were compared. The test environment was Python 3.11, Windows 10 64-bit. After three rounds of warm-up, 15 independent iterations of formal measurement were performed, and the average value was taken.

[0126] The table below shows the measured data for five groups of samples.

[0127]

[0128] The five groups of samples showed an average speedup of 10,932 times and an average memory saving of 361 times, which verified the significant technical effects of the embodiments of this application.

[0129] As shown above, the performance comparison of the five sets of samples includes the speedup ratio of extraction speed and the memory saving ratio. The speedup ratios of extraction speed are: SimpleRectangle 17000 times, L_ShapeRouting 15000 times, IrregularPowerPlane 2961 times, Non-Manhattan45 degrees 12500 times, and RoundPadRF 7200 times, with an average of 10932 times. The memory saving ratios are: SimpleRectangle 400 times, L_ShapeRouting 360 times, IrregularPowerPlane 208 times, Non-Manhattan45 degrees 525 times, and RoundPadRF 312 times, with an average of 361 times.

[0130] Among them, Simple Rectangle is a simple rectangular trace, L_Shape Routing is an L-shaped corner trace, Irregular Power Plane is an irregular power plane, Non-Manhattan 45 degrees is a non-Manhattan 45-degree differential pair trace, and Round Pad RF is a circular RF pad.

[0131] In summary, the embodiments of this application use each arbitrary polygon as a unified electromagnetic entity, extract self-network parasitic parameters based on the contour and electrical functional nodes of the arbitrary polygon, obtain a multi-port self-network block of the arbitrary polygon, construct a Voronoi graph based on the spatial positions of multiple arbitrary polygons, determine the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi graph, determine the coupling candidate polygon pairs according to the Voronoi neighbor relationship and extract sparse coupling edges, and store the multi-port self-network block and sparse coupling edges according to a sparse hierarchical adjacency storage structure. Thus, at the chip-level layout scale, high-precision extraction of parasitic parameters, fast coupling calculation and sparse hierarchical storage are realized simultaneously, and the batch generation of artificial intelligence training data and cross-layer parasitic extraction of 3D integrated circuits are effectively supported.

[0132] To facilitate better implementation of the layout parasitic parameter extraction and sparse storage method provided in this application embodiment, this application embodiment also provides a layout parasitic parameter extraction and sparse storage device. The meanings of the terms used are the same as in the layout parasitic parameter extraction and sparse storage method described above, and specific implementation details can be found in the description of the method embodiment.

[0133] Please see Figure 5 , Figure 5 This is a schematic diagram of the layout parasitic parameter extraction and sparse storage device provided in an embodiment of this application. The layout parasitic parameter extraction and sparse storage device may include a layout parsing module 201, a self-network extraction module 202, a coupling extraction module 203, and a sparse storage module 204. The layout parsing module 201 is used to acquire layout data, identify arbitrary polygons in the layout data, and determine the electrical function nodes on the arbitrary polygons. The self-network extraction module 202 is used to extract parasitic parameters of the self-network based on the contour and electrical function nodes of the arbitrary polygon, taking each arbitrary polygon as a unified electromagnetic entity, to obtain a multi-port self-network block of the arbitrary polygon. The coupling extraction module 203 is used to construct a Voronoi graph based on the spatial positions of multiple arbitrary polygons, and determine the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi graph; determine the coupling candidate polygon pairs according to the Voronoi neighbor relationship, and extract the coupling parasitic parameters from the coupling candidate polygon pairs to obtain sparse coupling edges. The sparse storage module 204 is used to store multi-port self-network blocks and sparsely coupled edges according to a sparse hierarchical adjacency storage structure.

[0134] For specific implementation methods of each of the above units, please refer to the embodiments of the above-described method for extracting and sparsely storing layout parasitic parameters, which will not be repeated here.

[0135] In summary, the layout parasitic parameter extraction and sparse storage device provided in this application embodiment can obtain layout data through the layout parsing module 201, identify arbitrary polygons in the layout data, and determine the electrical function nodes on the arbitrary polygons; the self-network extraction module 202 uses each arbitrary polygon as a unified electromagnetic entity, and performs self-network parasitic parameter extraction based on the contour of the arbitrary polygon and the electrical function nodes to obtain a multi-port self-network block of the arbitrary polygon; the coupling extraction module 203 constructs a Voronoi graph based on the spatial positions of multiple arbitrary polygons, and determines the Voronoi neighbor relationship between each arbitrary polygon according to the adjacency relationship in the Voronoi graph; it determines the coupling candidate polygon pair according to the Voronoi neighbor relationship, and performs coupling parasitic parameter extraction on the coupling candidate polygon pair to obtain sparse coupling edges; the sparse storage module 204 stores the multi-port self-network block and sparse coupling edges according to a sparse hierarchical adjacency storage structure. The embodiments of this application can simultaneously achieve high-precision extraction of parasitic parameters, fast coupled calculation, and sparse hierarchical storage at the chip-level layout scale, and effectively support the batch generation of artificial intelligence training data and cross-layer parasitic extraction of three-dimensional integrated circuits.

[0136] This application also provides an electronic device that may integrate the layout parasitic parameter extraction and sparse storage device of this application embodiment, such as... Figure 6 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically: The electronic device may include components such as a processor 301 with one or more processing cores and a memory 302 with one or more computer-readable storage media. Those skilled in the art will understand that... Figure 6 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs stored in the memory 302 and / or the methods provided in this application, and by calling data stored in the memory 302, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.

[0137] The memory 302 can be used to store software programs and the methods provided in this application. The processor 301 executes various functional applications and data processing by running the software programs stored in the memory 302 and the methods provided in this application. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store applications required for operating the storage medium and at least one function; the data storage area may store data created based on the use of the electronic device. In addition, the memory 302 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0138] Although not shown, the electronic device may also include a display unit, an input unit, and a power supply, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more application programs into the memory 302 according to the following instructions, and the processor 301 runs the application programs stored in the memory 302 to realize various functions, as follows: Acquire layout data, identify arbitrary polygons in the layout data, and determine the electrical function nodes on the arbitrary polygons; Using each arbitrary polygon as a unified electromagnetic entity, the parasitic parameters of the self-network are extracted based on the contour and electrical functional nodes of the arbitrary polygon to obtain the multi-port self-network block of the arbitrary polygon. A Voronoi diagram is constructed based on the spatial positions of multiple arbitrary polygons, and the Voronoi neighbor relationships between each arbitrary polygon are determined according to the adjacency relationships in the Voronoi diagram. Based on the Voronoi neighbor relationship, candidate polygon pairs for coupling are determined, and the parasitic coupling parameters of the candidate polygon pairs are extracted to obtain sparse coupling edges. Multi-port self-network blocks and sparsely coupled edges are stored according to a sparse hierarchical adjacency storage structure.

[0139] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0140] Therefore, embodiments of this application provide a storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the methods provided in embodiments of this application. For example, the instructions can execute the following steps: Acquire layout data, identify arbitrary polygons in the layout data, and determine the electrical function nodes on the arbitrary polygons; Using each arbitrary polygon as a unified electromagnetic entity, the parasitic parameters of the self-network are extracted based on the contour and electrical functional nodes of the arbitrary polygon to obtain the multi-port self-network block of the arbitrary polygon. A Voronoi diagram is constructed based on the spatial positions of multiple arbitrary polygons, and the Voronoi neighbor relationships between each arbitrary polygon are determined according to the adjacency relationships in the Voronoi diagram. Based on the Voronoi neighbor relationship, candidate polygon pairs for coupling are determined, and the parasitic coupling parameters of the candidate polygon pairs are extracted to obtain sparse coupling edges. Multi-port self-network blocks and sparsely coupled edges are stored according to a sparse hierarchical adjacency storage structure.

[0141] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0142] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0143] Since the instructions stored in the storage medium can execute the steps of any method provided in the embodiments of this application, the beneficial effects that any method provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0144] The methods, apparatus, and storage media for extracting parasitic parameters and sparse storage of the layout provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for extracting and sparsely storing parasitic parameters of a map, characterized in that, include: Acquire layout data, identify any polygon in the layout data, and determine the electrical function nodes on the arbitrary polygon; Using each of the arbitrary polygons as a unified electromagnetic entity, self-network parasitic parameters are extracted based on the contour of the arbitrary polygon and the electrical functional node to obtain the multi-port self-network block of the arbitrary polygon. A Voronoi diagram is constructed based on the spatial positions of the arbitrary polygons, and the Voronoi neighbor relationships between the arbitrary polygons are determined according to the adjacency relationships in the Voronoi diagram. Based on the Voronoi neighbor relationship, coupled candidate polygon pairs are determined, and the coupling parasitic parameters are extracted from the coupled candidate polygon pairs to obtain sparse coupled edges. The multi-port self-network block and the sparsely coupled edge are stored according to a sparse hierarchical adjacency storage structure.

2. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 1, characterized in that, The process of using each arbitrary polygon as a unified electromagnetic entity, extracting parasitic parameters of the self-network based on the contour of the arbitrary polygon and the electrical functional nodes, and obtaining a multi-port self-network block of the arbitrary polygon includes: The contour of the arbitrary polygon is piecewise linearly discretized to obtain multiple boundary segments; Using the electrical functional nodes as ports, the boundary element method is used to construct the port impedance matrix of the arbitrary polygon based on multiple boundary segments; Based on the port impedance matrix, an arbitrary polygonal multi-port self-network block is obtained.

3. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 2, characterized in that, The step of constructing the port impedance matrix of the arbitrary polygon based on multiple boundary segments using the electrical functional nodes as ports includes: Each of the aforementioned boundary segments is set as a constant current element; Based on the Green's function of a two-dimensional conductive plane, the impedance contribution between any two boundary segments is calculated to obtain the impedance relationship of the boundary segments. Based on the correspondence between the electrical functional nodes and each of the boundary segments, the impedance relationships of the boundary segments are assembled into a port impedance matrix with the electrical functional nodes as ports.

4. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 1, characterized in that, The step of determining candidate polygon pairs for coupling based on the Voronoi neighbor relationship and extracting coupling parasitic parameters from the candidate polygon pairs to obtain sparse coupling edges includes: Two arbitrary polygons having the Voronoi neighbor relationship are identified as a pair of coupled candidate polygons. For the coupled candidate polygon pair, obtain the local geometry and relative angle of the nearest edge between the two arbitrary polygons; Based on the local geometry and the relative angle, shape-aware coupling calculation is performed on the pair of coupling candidate polygons to obtain the coupling capacitance between the pair of coupling candidate polygons. The pair of coupling candidate polygons and the coupling capacitor are used as sparse coupling edges.

5. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 1, characterized in that, The sparse hierarchical adjacency storage structure includes an adjacency graph index layer, a self-network block storage layer, and a coupled edge sparse storage layer. The step of storing the multi-port self-network block and the sparsely coupled edge according to a sparse hierarchical adjacency storage structure includes: The adjacency relationships of the arbitrary polygons are stored in the adjacency graph index layer, which uses a compressed sparse row format to store polygon identifiers, neighbor polygon identifiers, and edge types. The multi-port subnet blocks are stored in the subnet block storage layer, and the subnet block storage layer stores the port impedance matrix and capacitance to ground of the multi-port subnet blocks in a compressed column storage format. The sparse coupling edges are stored in the coupling edge sparse storage layer, and the coupling edge sparse storage layer stores the coupling candidate polygon pairs and the coupling capacitance between the coupling candidate polygon pairs in a triplet format.

6. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 1, characterized in that, Also includes: For each metal layer of each chip layer, a two-dimensional Voronoi diagram is constructed based on the spatial positions of multiple arbitrary polygons, and the Voronoi neighbor relationship between each arbitrary polygon is determined according to the adjacency relationship in the two-dimensional Voronoi diagram. Project the geometric centroid of the arbitrary polygon in the upper metal layer onto the planar coordinate system of the lower metal layer, and determine cross-layer coupling candidate pairs in the Voronoi diagram of the lower metal layer; Calculate the three-dimensional effective spacing of the cross-layer coupling candidate pairs, and extract the coupling parasitic parameters only for cross-layer coupling candidate pairs whose three-dimensional effective spacing is less than a preset threshold to obtain cross-layer sparse coupling edges; The cross-layer sparse coupling edges are stored according to the sparse hierarchical adjacency storage structure.

7. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 1, characterized in that, Also includes: Arbitrary polygons in the same layer are spatially sorted according to Hilbert curves; The sorted arbitrary polygons are divided into multiple batches, and Ghost Zones are set at the batch boundaries. The Ghost Zones contain Voronoi neighbor arbitrary polygons that cross the batch boundaries. Each batch performs self-network parasitic parameter extraction and coupled parasitic parameter extraction in parallel. Any polygon within the Ghost Zone participates in coupled parasitic parameter extraction but does not participate in self-network parasitic parameter extraction, and is written to the sparse hierarchical adjacency storage structure without lock according to the pre-allocated address.

8. The method for extracting and sparsely storing parasitic parameters of a map as described in claim 2 or 3, characterized in that, After constructing the port impedance matrix of the arbitrary polygon based on multiple boundary segments using the boundary element method, the method further includes: Calculate at least one of the following: aspect ratio, compactness, concavity, and curvature factor of the arbitrary polygon; Based on the calculation results, local insertion of refinement nodes or constraint Delaunay triangulation is selected to form a hybrid solution domain including boundary element region and refinement region; Impedance matching boundary conditions are applied at the interface of the hybrid solution region, and the refined region is equivalently folded back into the port impedance matrix.

9. A device for extracting and sparsely storing parasitic parameters of a map, characterized in that, include: The layout parsing module is used to acquire layout data, identify any polygon in the layout data, and determine the electrical function nodes on the arbitrary polygon; The self-network extraction module is used to extract self-network parasitic parameters based on the contour of the arbitrary polygon and the electrical function node, taking each arbitrary polygon as a unified electromagnetic entity, to obtain a multi-port self-network block of the arbitrary polygon. The coupling extraction module is used to construct a Voronoi diagram based on the spatial positions of the multiple arbitrary polygons, and to determine the Voronoi neighbor relationship between each of the arbitrary polygons according to the adjacency relationship in the Voronoi diagram; Based on the Voronoi neighbor relationship, coupled candidate polygon pairs are determined, and the coupling parasitic parameters are extracted from the coupled candidate polygon pairs to obtain sparse coupled edges. A sparse storage module is used to store the multi-port self-network block and the sparse coupling edge according to a sparse hierarchical adjacency storage structure.

10. A storage medium, characterized in that, The storage medium stores multiple instructions, which are applicable to a processor for loading to execute the layout parasitic parameter extraction and sparse storage method according to any one of claims 1-8.