Search

CN-122021549-A - Multi-physical field coupling simulation learning method and system for three-dimensional integrated circuit

CN122021549ACN 122021549 ACN122021549 ACN 122021549ACN-122021549-A

Abstract

The application discloses a multi-physical field coupling simulation learning method and a system for a three-dimensional integrated circuit, which are characterized in that firstly, the space geometric structure of the three-dimensional integrated circuit and the structure attribute parameters and physical attribute parameters of each structure unit forming the space geometric structure are obtained; the method comprises the steps of constructing a three-dimensional coordinate system for a space geometric structure, obtaining stress calculation original data according to a mathematical structure model of each structural unit and corresponding physical attribute parameters, and finally inputting the stress calculation original data into an LBO simulation mathematical model to obtain physical field distribution of the corresponding structural units output by the LBO simulation mathematical model. The multi-physical field coupling simulation of the three-dimensional integrated circuit is carried out by introducing the LBO simulation mathematical model of physical constraint, so that the rapid estimation of the temperature and stress distribution of each device in the three-dimensional integrated circuit can be realized.

Inventors

  • CHEN YINGSHI

Assignees

  • 深圳鸿芯微纳技术有限公司
  • 上海鸿芯科纳科技有限公司

Dates

Publication Date
20260512
Application Date
20260415

Claims (10)

  1. 1. A multi-physical field coupling simulation learning method for a three-dimensional integrated circuit, comprising: acquiring a space geometry structure of the three-dimensional integrated circuit, wherein the space geometry structure is formed by at least one structural unit; Obtaining a structural attribute parameter and a physical attribute parameter corresponding to each structural unit, wherein the structural attribute parameter comprises a unit boundary parameter, and the physical attribute parameter comprises a unit material parameter and a power consumption distribution parameter; Constructing a three-dimensional coordinate system for the space geometric structure, and defining a mathematical structure model of each structural unit under the three-dimensional coordinate system; Obtaining stress calculation original data according to a mathematical structure model of each structural unit and corresponding physical attribute parameters, wherein the stress calculation original data are { x, y, z, t, psi 1 ,ψ 2 ,…,ψ k-1 ,ψ k }, x, y and z are coordinates of each point in the mathematical structure model of the structural unit in a three-dimensional coordinate system, t is a time coordinate, k is a natural number, psi 1 、ψ 2 , & phi and psi k are k LBO base functions; inputting the stress calculation original data into a preset LBO simulation mathematical model, and obtaining physical field distribution mu corresponding to the structural unit output by the LBO simulation mathematical model, Μ= { τ, σ x ,σ y ,σ xy }, τ being a temperature value, σ x 、σ y and σ xy being stress values in the three-dimensional coordinate system; Outputting the physical field distribution mu corresponding to each structural unit.
  2. 2. The multi-physical field coupling simulation learning method of claim 1, further comprising: the LBO simulation mathematical model is a multi-layer deep fully-connected neural network; the multi-physical field coupling includes thermal and stress physical field coupling; Within each of the structural units, the physical attribute parameters are consistent; The mathematical structural model of the structural unit comprises node coordinates, which are geometric vertex coordinates of the structural unit, and an interpolation function, which is a regular function connecting geometric vertices of the structural unit and filling the interior of the geometric body.
  3. 3. The multi-physical field coupling simulation learning method of claim 2, wherein the obtaining method of the LBO simulation mathematical model comprises: embedding physical attribute parameters into the LBO simulation mathematical model by differential techniques, comprising: Performing an automatic differentiation on a mathematical structural model of the structural unit, and calculating respective derivatives of a physical field distribution μ of the mathematical structural model to input coordinates (x, y, z, t); Substituting the derivative obtained by executing automatic differentiation into a preset heat conduction and stress balance partial differential equation, and calculating physical deviation to measure the deviation degree of the predictive solution and a preset physical law; A loss function is calculated to obtain weight parameters of the LBO simulation mathematical model for adjusting the loss function.
  4. 4. The multi-physical field coupling simulation learning method of claim 3, wherein the training method of the LBO simulation mathematical model comprises: selecting stress calculation original data serving as sampling points from the mathematical structure model of the structural unit according to a preset rule; Inputting the stress calculation original data serving as sampling points into the LBO simulation mathematical model, and obtaining physical field distribution mu of the corresponding sampling points output by the LBO simulation mathematical model; Performing automatic differentiation on the sampling points, and calculating each derivative of the physical field distribution mu on the input coordinates (x, y, z, t); Calculating the physical deviation of the sampling points; calculating a loss function of the sampling points; And updating the weight parameters of the LBO simulation mathematical model by applying a preset gradient descent algorithm.
  5. 5. The multi-physics coupling simulation learning method of claim 4, wherein the training method of the LBO simulation mathematical model further comprises: And repeatedly executing the training process of the LBO simulation mathematical model to iterate the weight parameters of the LBO simulation mathematical model until the total loss function converges.
  6. 6. The multi-physics coupling simulation learning method of claim 4, further comprising: the gradient descent algorithm includes Adam algorithm.
  7. 7. The multi-physical field coupling simulation learning method of claim 4, wherein the loss function has a calculation formula: loss=λ physics ×l physics +λ bc ×l bc ; Wherein loss is a loss function, l physics is an average value of physical deviation at a plurality of sampling points on the mathematical structure model, l bc is a deviation value between a predicted temperature and a real temperature at a plurality of boundary points of the mathematical structure model, and lambda physics and lambda bc are preset super-parameters for adjusting weights of physical loss and boundary condition loss in total loss.
  8. 8. A computer readable storage medium having stored thereon a computer program executable by a processor to implement the multi-physical field coupling simulation learning method of any of claims 1-7.
  9. 9. A computer program product comprising a computer program and/or instructions which, when executed by a processor, implements the multi-physical field coupling simulation learning method of any of claims 1-7.
  10. 10. A multiple physical field coupling simulation learning system for a three-dimensional integrated circuit for applying the multiple physical field coupling simulation learning method of any one of claims 1-7, the multiple physical field coupling simulation learning system comprising: A structure acquisition module for acquiring a spatial geometry of a three-dimensional integrated circuit, wherein the spatial geometry is composed of at least one structural unit; The system comprises a parameter acquisition module, a power consumption distribution module and a power consumption distribution module, wherein the parameter acquisition module is used for acquiring structural attribute parameters and physical attribute parameters corresponding to each structural unit, wherein the structural attribute parameters comprise unit boundary parameters, and the physical attribute parameters comprise material parameters and power consumption distribution parameters; The system comprises a space geometrical structure, a coordinate construction module, a mathematical structure model and a filling function, wherein the space geometrical structure is provided with a three-dimensional coordinate system, and the three-dimensional coordinate system is used for constructing a mathematical structure model of each structural unit; The system comprises an input data acquisition module, a stress calculation module and a data processing module, wherein the input data acquisition module is used for acquiring stress calculation original data according to a mathematical structure model of each structural unit and corresponding physical attribute parameters, wherein the stress calculation original data are { x, y, z, t, psi 1 ,ψ 2 ,…,ψ k-1 ,ψ k }, wherein x, y and z are coordinates of each point in the mathematical structure model of the structural unit in a three-dimensional coordinate system, t is a time coordinate, k is a natural number, and psi 1 、ψ 2 , & gt and psi k are k LBO basis functions; The physical field acquisition module is used for inputting the stress calculation original data into a preset LBO simulation mathematical model and acquiring physical field distribution mu corresponding to the structural unit output by the LBO simulation mathematical model, Μ= { τ, σ x ,σ y ,σ xy }, τ being a temperature value, σ x 、σ y and σ xy being stress values in the three-dimensional coordinate system; And the output module is used for outputting the physical field distribution mu corresponding to each structural unit.

Description

Multi-physical field coupling simulation learning method and system for three-dimensional integrated circuit Technical Field The application relates to the technical field of chip design, in particular to a multi-physical field coupling simulation learning method and system for a three-dimensional integrated circuit. Background The core purpose of the three-dimensional integrated circuit (3 DIC) multi-physical field coupling simulation calculation is to synchronously analyze the association phenomenon of multiple physical fields (such as thermal stress physical fields) and a complex coupling mechanism in a three-dimensional chip structure by means of a computer simulation technology, so that the problems of frequency reduction, failure and the like caused by overheating are avoided under the preset power consumption and heat dissipation conditions of the chip, and further, the support is provided for the chip reliability design. The three-dimensional integrated circuit realizes vertical stacking integration of a plurality of bare chips through key technologies such as Through Silicon Vias (TSVs), hybrid bonding and the like, and the structural design greatly improves the chip performance, reduces the packaging size and simultaneously remarkably increases the complexity of a physical field. In the chip working process, the high power consumption area can continuously generate heat, so that the temperature of the area is increased, and the corresponding area material is driven to thermally expand, while the thermal expansion coefficients of materials (such as silicon, bonding layers, packaging materials and the like) of different functional areas in the 3DIC structure are obviously different, and the thermal expansion mismatch can cause mutual constraint among the areas, so that thermal stress is generated. When thermal stress is accumulated to a certain extent, the chip can be caused to warp and deform, if the stress is continuously overlapped, serious damage such as interlayer interface stripping and bonding position fracture can be caused, and finally the reliability failure of the chip is caused, so that the thermal-stress physical field coupling simulation calculation is a key link in the design process of the three-dimensional integrated circuit. Disclosure of Invention The application mainly solves the technical problem of how to realize and optimize the multi-physical field coupling simulation calculation of the three-dimensional integrated circuit. According to a first aspect, in one embodiment, a multi-physical field coupling simulation learning method for a three-dimensional integrated circuit is provided, including: acquiring a space geometry structure of the three-dimensional integrated circuit, wherein the space geometry structure is formed by at least one structural unit; Obtaining a structural attribute parameter and a physical attribute parameter corresponding to each structural unit, wherein the structural attribute parameter comprises a unit boundary parameter, and the physical attribute parameter comprises a unit material parameter and a power consumption distribution parameter; Constructing a three-dimensional coordinate system for the space geometric structure, and defining a mathematical structure model of each structural unit under the three-dimensional coordinate system; Obtaining stress calculation original data according to a mathematical structure model of each structural unit and corresponding physical attribute parameters, wherein the stress calculation original data are { x, y, z, t, psi 1,ψ2,…,ψk-1,ψk }, x, y and z are coordinates of each point in the mathematical structure model of the structural unit in a three-dimensional coordinate system, t is a time coordinate, k is a natural number, psi 1、ψ2, & phi and psi k are k LBO base functions; inputting the stress calculation original data into a preset LBO simulation mathematical model, and obtaining physical field distribution mu corresponding to the structural unit output by the LBO simulation mathematical model, Μ= { τ, σ x,σy,σxy }, τ being a temperature value, σ x、σy and σ xy being stress values in the three-dimensional coordinate system; Outputting the physical field distribution mu corresponding to each structural unit. In one embodiment, the multi-physical field coupling simulation learning method further includes: the LBO simulation mathematical model is a multi-layer deep fully-connected neural network; the multi-physical field coupling includes thermal and stress physical field coupling; Within each of the structural units, the physical attribute parameters are consistent; The mathematical structural model of the structural unit comprises node coordinates, which are geometric vertex coordinates of the structural unit, and an interpolation function, which is a regular function connecting geometric vertices of the structural unit and filling the interior of the geometric body. In one embodiment, the method for obtaining